Dynamic changes in the osteoclast cytoskeleton in response to growth factors and cell attachment are controlled by β3 integrin

Osteoclastic bone resorption is a process requiring physical intimacy between the resorptive cell and bone matrix. Thus, cell–matrix attachment molecules, particularly integrins, play a central role in the capacity of osteoclasts (OCs) to degrade bone (Carron et al., 2000; Feng et al., 2001). The integrin αvβ3 is particularly important, in this regard, as its absence prompts OC dysfunction, eventuating in subnormal bone resorption and osteosclerosis (McHugh et al., 2000). The clinical significance of this observation is underscored by the fact that rats treated with an αvβ3 antagonist are spared the bone loss attending oophorectomy (Engleman et al., 1997).

While serving as a matrix attachment molecule in OCs, αvβ3 is also a signaling receptor (Eliceiri et al., 1998; Duong et al., 2000) that induces changes in intracellular calcium (Paniccia et al., 1993; Zimolo et al., 1994), protein tyrosine phosphorylation, and cytoskeletal reorganization (Clark et al., 1998; Schlaepfer and Hunter, 1998). The signaling capacity of integrins can be regulated by extracellular matrix molecules that, interacting with the integrin external domain, stimulate “outside-in” signaling (Takagi et al., 2002). Alternatively, “inside-out” signaling is induced by trans-activated intracellular molecules, which interact with the cytoplasmic component of αvβ3 and prompt conformational changes in the integrin's ligand binding site. Outside-in and inside-out signals control the affinity state of the integrin, thereby modulating its binding capabilities and, ultimately, intracellular events (Pelletier et al., 1995; Geiger et al., 2001; Butler et al., 2003). Several cytokines and growth factors enhance integrin-dependent intracellular events, via inside-out signaling. Platelet-derived growth factor induces αvβ3-mediated adhesion in fibroblasts (Schneller et al., 1997), basic fibroblast growth factor augments migration in vascular endothelial cells (Kiosses et al., 2001), and macrophage colony–stimulating factor (M-CSF) and hepatocyte growth factor (HGF) modulate OC function, in part by increasing activated αvβ3 in the motile area of the membrane (Faccio et al., 2002).

In most cells, αvβ3 localizes with actin and other cytoskeletal proteins in focal adhesions (Ballestrem et al., 2001; Cukierman et al., 2001). OCs contain a related, but distinct, adhesive structure called the podosome, which consists of a core of F-actin bundles surrounded by a rosette-like structure containing αvβ3, vinculin, and α-actinin (Marchisio et al., 1988). As ligand activation of αvβ3 and growth factor stimulation promote podosome reorganization (Pfaff and Jurdic, 2001; Faccio et al., 2002), interest has turned to the intracellular components of the integrin and the signaling molecules linking it to cytoskeletal proteins.

c-Src is essential to OC function, as mice deleted of this tyrosine kinase develop osteopetrosis in the face of adequate numbers of dysfunctional OCs (Soriano et al., 1991). The fact that c-Src^−/− OCs fail to organize a normal cytoskeleton suggests that c-Src may be a signaling molecule that associates with, and is activated by, αvβ3 (Duong et al., 2000; Sanjay et al., 2001). In addition, c-Cbl, a substrate of c-Src in OCs, is recruited to adhesion sites where it modulates the binding of the vitronectin (VN) receptor αvβ3 (Sanjay et al., 2001). Pyk-2, a member of the FAK family of kinases, is a signaling molecule that binds c-Src and c-Cbl, and appears to be essential for bone resorption. Pyk2 is activated when OCs are plated on ligands recognized by αvβ3 and is important for cytoskeletal organization during OC adhesion, migration, and sealing zone formation (Duong et al., 1998). The above compendium of events suggests that Pyk2 activation, in osteoclastic resorption, is mediated by αvβ3.

Rho family GTP-ases control cytoskeletal organization and dynamics and integrin-mediated signaling, as their inhibition blocks αvβ3-dependent motility (Clark et al., 1998; Ridley et al., 1999; Chellaiah et al., 2000; Ory et al., 2000). Importantly, Rho and Rac regulate the OC actin ring, and their blockade blunts the resorptive activity of the cell (Razzouk et al., 1999; Ory et al., 2000).

In the present study, we establish that the β3 cytoplasmic domain, specifically S⁷⁵², is responsible for organizing the OC cytoskeleton. Furthermore, like adhesion-dependent cytoskeletal reorganization, growth factor–induced αvβ3 inside-out signaling activates Rho GTPases. Finally, although αvβ3 is essential for activation of c-Src and c-Cbl, OCs lacking the integrin fully activate Pyk2.

Mice deleted of the β3 integrin subunit become osteosclerotic due to dysfunctional OCs, which fail to efficiently resorb bone (McHugh et al., 2000). As expected, the function of β3^−/− OCs is rescued by transduction with retrovirus expressing full-length β3 cDNA. On the other hand, β3 lacking its cytoplasmic domain, or carrying a S⁷⁵²P mutation, is incapable of rescuing OCs devoid of endogenous β3 (Feng et al., 2001). Furthermore β3^−/− OC precursors cultured with RANKL and standard doses of M-CSF fail to differentiate fully (Faccio et al., 2003). This osteoclastogenic defect can be rescued by increasing M-CSF concentrations, but the ability of these cells to resorb bone requires the presence of the integrin (Faccio et al., 2003). To further define the role that the β3 integrin cytoplasmic domain plays in organizing the OC cytoskeleton, we studied the distribution of podosomes in β3^−/− OCs transduced with different human β3 integrin mutants. Four constructs were used for this purpose: the β3 integrin lacking its cytoplasmic tail (β3-ΔC); a mutant carrying the S⁷⁵²P mutation or that bearing the OC nonsignificant double tyrosine mutation, Y⁷⁴⁷F/Y⁷⁵⁹F; and as positive control, the full-length human β3 (β3 WT). Equivalent levels of expression of all four constructs were confirmed by flow cytometry and by Western blot analysis (Fig. 1, A and B). Bone marrow macrophages (BMMs) expressing the indicated mutants were cultured on coverslips in the presence of RANKL (100 ng/ml) and a high dose of M-CSF (100 ng/ml), conditions that lead to the formation of completely spread OCs, even in nontransduced β3^−/− cultures (Faccio et al., 2003). These cells cultured in high M-CSF are equivalent to their WT counterparts in terms of morphology, osteoclastogenic markers (Faccio et al., 2003), c-Fms (the receptor for M-CSF) levels, and pattern of integrin expression (see Fig. S1).

To delineate actin organization, OCs expressing the various mutants were immunostained with the anti–human β3 mAb 1A2 and costained with FITC–phalloidin (Fig. 2 A). The β3 integrin is organized in rosette-like structures, surrounding a core of F-actin bundles, in podosomes as previously described (Faccio et al., 2002). In nonresorbing OCs on glass, podosomes accumulate at high density at the periphery, yielding a row of actin dots flanked on each side by β3 integrin bands. This organization is present in β3^−/− OCs bearing β3 WT or the β3 Y⁷⁴⁷F/Y⁷⁵⁹F mutation. Lack of functional β3, as in β3-ΔC and β3 S⁷⁵²P mutants, completely abrogates the distribution of the integrin around the peripheral ring of F-actin, despite the normal appearance of the actin cytoskeleton. Similar results were obtained with another anti–human β3 mAb, AP3 (unpublished data). Interestingly other cytoskeletal components, including talin and vinculin, remain normally distributed in the podosomes of all mutants (see Fig. S2).

To determine if the observations made on nonresorbing OCs are replicated in resorbing cells, the four species of β3-transduced OCs were generated on whale dentin and stained for the integrin and actin. Confocal microscopy reveals that the integrin, in dentin-resorbing β3 WT– and β3 Y⁷⁴⁷F/Y⁷⁵⁹F–expressing cells, localizes with the actin ring (Fig. 2 B). Once again, the β3-ΔC and β3 S⁷⁵²P mutants are diffusely distributed throughout the cell. This diffuse localization is not due to an increase in internalization, as the flow cytometric analysis shows similar surface expression levels to β3 WT and β3 Y⁷⁴⁷F/Y⁷⁵⁹F mutants (Fig. 1 A).

One possible explanation for the failure of β3-ΔC and β3 S⁷⁵²P to localize to podosomes is that the external domain of these mutants is not able to assume the activated, high-affinity conformation necessary for appropriate substrate interaction. One tool to assess the ability of β3 integrin to assume the activated conformation of the external domain is the anti–ligand-induced binding site (LIBS) antibody AP5. In low calcium buffer, this Ab binds all αvβ3 integrin on the cell surface, converting it to the active conformation. In high calcium buffer, AP5 binds only the integrin already in the activated state (Faccio et al., 2002). An increase in fluorescence intensity of αvβ3-expressing cells when AP5 binding is assessed in low calcium, relative to high calcium buffer, indicates that the external domain of the integrin can assume the activated conformation in response to the Ab. When pre-OCs transduced with β3-ΔC and β3 S⁷⁵²P, as well as with β3 WT and β3 Y⁷⁴⁷F/Y⁷⁵⁹F, are analyzed in this manner, there is an increase in AP5 binding in low calcium, indicating that each of these β3 constructs undergoes conformational change (Fig. 3 A). Thus, the failure of the β3-ΔC and β3 S⁷⁵²P mutants to properly localize in podosomes does not reflect an inability of these integrins to assume an activated conformation.

A second possible explanation for the aberrant localization of β3-ΔC and β3 S⁷⁵²P pre-OCs is that these mutants are not activated by signals emanating from the cell interior. One mechanism for such inside-out activation is growth factor treatment. We have previously shown that both HGF and M-CSF activate αvβ3 in this manner (Faccio et al., 2002). To determine whether the β3-ΔC and β3 S⁷⁵²P mutants are defective in growth factor–mediated activation, pre-OCs transduced with the β3 mutants were incubated with either HGF or M-CSF for 30 min before evaluation of the activated state of the integrin via AP5 staining in high calcium buffer. HGF and M-CSF activate β3 WT and β3 Y⁷⁴⁷F/Y⁷⁵⁹F, but fail to impact the integrin lacking the entire cytoplasmic domain (β3-ΔC) or bearing S⁷⁵²P (Fig. 3 B). Thus, the β3 cytoplasmic domain, and specifically S⁷⁵², is required for growth factor–induced formation of a high-affinity αvβ3 complex.

Upon growth factor activation, OC αvβ3 moves to the motile region of the cell membrane (Faccio et al., 2002). Having found that dysfunctional β3 mutants fail to properly localize in resting OCs, and do not become activated in response to growth factors, we determined if these mutants also fail to localize properly in the newly formed lamellipodia in response to growth factor stimulation. Thus, β3^−/− OCs bearing WT β3, β3-ΔC, β3 Y⁷⁴⁷F/Y⁷⁵⁹F, or β3 S⁷⁵²P constructs grown on coverslips were exposed to HGF or M-CSF for 30 min and, after fixation, stained with AP5 (in high calcium) to detect localization of the activated form of αvβ3. Unstimulated OCs carrying β3 WT or the Y⁷⁴⁷F/Y⁷⁵⁹F mutation express the activated integrin along membrane ruffles and in lamellipodia (Fig. 4 A, CTR). Treatment with either growth factor induces AP5-positive membrane extensions (lamellipodia). This phenomenon is completely abrogated in β3-ΔC– and β3 S⁷⁵²P–bearing cells, which are unable to spread and form lamellipodia in response to growth factors (Fig. 4 A). These observations are confirmed by counting the percentage of cells with multiple lamellipodia extensions (Table I). These data show that M-CSF and HGF induce lamellipodia in cells expressing β3 WT and β3 Y⁷⁴⁷F/Y⁷⁵⁹F but not in those transduced with β3-ΔC or β3 S⁷⁵²P.

We next turned to the effect of HGF or M-CSF on dentin-residing OCs, which represent resorptive cells. OCs expressing β3 WT or β3-ΔC were exposed to the individual cytokines and stained with AP5 (Fig. 4 B). In β3 WT–bearing cells, this exercise revealed that both cytokines induce membrane protrusions (arrows) that contain a predominance of activated αvβ3. Despite their ability to attach to bone and form actin rings on dentin, β3-ΔC OCs are unaffected by the growth factors, failing to spread and lacking membrane protrusions. Note that the activated integrin is only weakly localized in the actin ring, in all mutants, as previously described (Faccio et al., 2002, 2003).

The observation detailed in Fig. 4 led us to hypothesize that the β3 integrin controls cytoskeletal changes induced by growth factors. To address this issue, we stained unstimulated and growth factor–treated β3^+/+ and β3^−/− OCs with FITC–phalloidin and analyzed actin organization by confocal microscopy. Once again, the absence of β3 integrin does not alter the peripheral podosomal distribution of F-actin in untreated OCs (Fig. 5 A, CTR). However, in the presence of HGF or M-CSF, F-actin moves from the podosomes to short filamentous protrusions, consistent with lamellipodia formation, only in β3^+/+ OCs (Fig. 5 A, low and high magnification; Table I). To confirm that this observation is an integrin-dependent consequence of podosome reorganization, we examined the distribution of α-actinin, a cytoskeletal protein involved in the formation and stability of podosomes, and a link between actin and integrins (Pavalko et al., 1991). α-Actinin distribution in β3^+/+ and β3^−/− OCs mirrors that of actin, both in the presence and absence of growth factors (Fig. 5 A). Consistent with this finding, immunoblot analysis shows an increase in the pool of α-actinin present in the Triton X-100 soluble fraction exclusively in β3^+/+ OCs treated with HGF and M-CSF (Fig. 5 B). In β3^−/− cells, α-actinin remains in the insoluble fraction. Similar results were obtained using OCs transduced with the different β3 mutants. Dramatic changes in the peripheral ring of actin are seen in response to M-CSF (Fig. 5 C), and the content of α-actinin in the Triton X-100 soluble fraction increases in β3 WT and β3 Y⁷⁴⁷F/Y⁷⁵⁹F mutants, but not in those transduced with β3-ΔC or β3 S⁷⁵²P (Fig. 5 D). The failure of β3-null cells to respond to M-CSF is not dependent on different expression levels of c-Fms (Fig. S1) or on its different localization among the various β3 mutants (Fig. 5 E). These observations suggest that, in OCs, growth factor–induced reorganization of podosomes, leading to the formation of new membrane ruffles, is a β3-dependent event.

The fact that β3-ΔC– and β3 S⁷⁵²P–expressing OCs fail to spread in response to growth factors suggests that interaction of these mutant integrins with the extracellular matrix may be defective. To address this issue, we plated β3^−/− pre-OCs bearing the various constructs on osteopontin (OPN), a substrate recognized by the αvβ3 integrin, or, as negative control, on BSA. Pre-OC adhesion to OPN is decreased threefold in cells carrying the β3-ΔC or S⁷⁵²P mutations, compared with β3 WT pre-OCs (Fig. 6 A), mirroring the defective adhesion of β3^−/− cells (unpublished data). Furthermore, pretreatment with AP5, HGF, or M-CSF, all of which activate the integrin (Fig. 3), enhances the number of adherent β3 WT by 40–50%, and to a lesser extent β3 Y⁷⁴⁷F/Y⁷⁵⁹F–expressing cells (Fig. 6 A). On the other hand, these integrin-activating agents fail to impact the adhesion of β3-ΔC– and β3 S⁷⁵²P–bearing pre-OCs. Similar results were obtained using VN as a substrate (unpublished data). In contrast, adhesion to native collagen, a β3 integrin–independent substrate, was similar in all mutants and was not enhanced by growth factors (unpublished data).

As substrate recognition initiates lamellipodia extension and directed cell movement, we next assessed the migration of pre-OCs transduced with the various β3 constructs, using a transwell assay in which the lower surface of the membrane was coated with OPN (Fig. 6 B). Pre-OC migration toward OPN is impaired in the absence of the β3 cytoplasmic domain or with the S⁷⁵²P mutation, mimicking the behavior of β3^−/− cells (unpublished data). Moreover, AP5, HGF, and M-CSF all fail to enhance the directed mobility of β3-ΔC and S⁷⁵²P mutant cells. Once again, similar results were generated using VN as substrate (unpublished data). Thus, these data mirror the defective growth factor–induced integrin activation of β3-ΔC and S⁷⁵²P mutants (Fig. 3 B).

Numerous signal transduction molecules, including Rho GTPases, associate with integrin complexes in adherent cells and regulate adhesion-dependent morphological changes (Pavalko et al., 1991; Yamada and Miyamoto, 1995; Clark et al., 1998). Furthermore, distinct aspects of adhesion and migration are controlled by different Rho family members. For example, Rho, Rac, and Cdc42, respectively, regulate the formation of stress fibers, lamellipodia, and filopodia. As β3^−/− OCs exhibit defective adhesion, spreading, and cytoskeletal rearrangement in response to growth factors, we asked if these phenomena reflect impaired activation of Rho GTPases. To this end, β3^−/− and β3^+/+ OCs were exposed to HGF or M-CSF, and Rho and Rac activation were assessed by pull-down binding assays. Mirroring the cytoskeletal defects, Rho activation in response to both growth factors is completely arrested in β3^−/− OCs, compared with the sixfold increase seen in β3^+/+ OCs (Fig. 7 A). Similarly, whereas Rac is activated in a biphasic manner in β3^+/+ OCs exposed to M-CSF, no such activation occurs in β3^−/− OCs (Fig. 7 B). To determine if integrin-mediated Rho activation is dependent on an intact β3 cytoplasmic domain, we repeated these experiments with β3^−/− OCs transduced with the various β3 constructs. Whereas β3 WT– and β3 Y⁷⁴⁷F/Y⁷⁵⁹F–expressing OCs respond to growth factor with an increase in GTP-bound Rho, no induction occurs in those expressing β3-ΔC or β3 S⁷⁵²P (Fig. 7 C).

The data presented thus far establish that the β3 integrin plays an essential role in OC cytoskeletal function, a process mediated at least in part by Rho family GTPases. These observations prompted us to examine the more proximal events thought to mediate αvβ3 signal transduction. Pyk2, believed to be central to the mechanisms by which OCs resorb bone, is activated when these cells interact with αvβ3 ligands (Duong et al., 2000; Sanjay et al., 2001). Surprisingly, however, Pyk2 activation, as manifest by its phosphorylation, is indistinguishable in β3^+/+ and β3^−/− pre-OCs plated on VN for 30 and 60 min (Fig. 8 A, top). In contrast, c-Src phosphorylation occurs in a β3-dependent manner, and c-Cbl phosphorylation is markedly reduced in the absence of the integrin (Fig. 8 A). In keeping with this observation, Pyk2 and c-Src coprecipitate in VN-adherent β3^+/+ cells but not those lacking the β3 integrin (Fig. 8 B). Furthermore, β3^−/− pre-OCs expressing all β3 constructs show phosphorylation of Pyk2 upon adhesion, whereas c-Src and c-Cbl activation is defective in β3-ΔC and S⁷⁵²P (Fig. 8 C). Pfaff and Jurdic (2001) have recently shown direct interaction of both Pyk2 and paxillin, a structural component of podosomes, with the COOH-terminal region of the integrin β3 tail. Interestingly, we find that binding of Pyk2 to paxillin is identical in all β3 mutants, but its ability to bind the β3 receptor itself is dependent on an intact cytoplasmic tail and S⁷⁵² (Fig. 8 D).

Our data show that Pyk2 activation requires cell–matrix recognition but does not depend upon αvβ3 status. Previous studies have shown that Pyk2 phosphorylation is calcium dependent (Sanjay et al., 2001), and we find that in β3^+/+ and β3^−/− pre-OCs, the intracellular chelator, BAPTA, eliminates Pyk2 activation (Fig. 9 A). To determine if another OPN receptor or integrin might be responsible for Pyk2 phosphorylation, we assessed this parameter in cells deleted of CD44, which has been shown to mediate monocyte attachment to OPN (Weber et al., 1996), or in cells lacking the integrin α2β1, another adhesive receptor expressed on OCs (unpublished data). Attachment to OPN leads to Pyk2 phosphorylation and to an increase in its association with paxillin in all cells tested (Fig. 9 B). To find the receptor mediating Pyk2 phosphorylation in response to cell adhesion, we turned to the β2 integrin, which is expressed in pre-OCs (Fig. S1) and is known to be a plastic receptor (Hirano et al., 2002). Thus, β3^+/+, β3^−/−, and β2^−/− pre-OCs were maintained in suspension or plated on VN or uncoated plastic for 30 min (Fig. 9 C). Plastic, like VN, induces Pyk2 phosphorylation in wild-type pre-OCs and those lacking αvβ3. However, Pyk2 phosphorylation of β2^−/− pre-OCs is completely abrogated when these mutant cells are plated on plastic and is barely detectable on VN, an observation that does not reflect diminished β3 integrin expression (Fig. 9 D). Furthermore, these findings do not result from altered Pyk2 levels in β2^−/− OCs (Fig. 9 C) or differing distribution (unpublished data). Thus, β2 integrin, but not β3 integrin or CD44, mediates Pyk2 activation in pre-OCs.

Marrow macrophages derived from β3^−/− mice, placed in standard osteoclastogenic conditions, fail to become fully differentiated OCs and resorb mineralized matrix poorly (Faccio et al., 2003). On the other hand, culture of these cells in high concentrations of M-CSF, as undertaken in this work, completely rescues the differentiation of these integrin-deficient OCs as manifest by their histological appearance and expression of osteoclastogenic markers. It is surprising, therefore, that despite the normal appearance in high dose M-CSF, β3^−/− OCs remain defective resorbers. To define the mechanisms responsible for the continued failure of β3^−/− OCs, generated in high M-CSF, to resorb bone, we analyzed the differences in cytoskeletal organization and relevant intracellular signaling molecules in OCs with and without functional β3 integrins.

αvβ3, in OCs, exists in two conformational states, which are differentially distributed on the cell surface (Faccio et al., 2002). In its basal condition, the receptor localizes in the sealing zone and podosomes, while the activated integrin is principally associated with motile areas of the membrane. The extracellular components of αvβ3 modulating its activation state are in hand (Beglova et al., 2002), and those in the cytoplasmic domain, which respond to growth factor stimulation and thus mediate inside-out signaling, have been partially identified (Takagi et al., 2002; Vinogradova et al., 2002). To further address this issue, we turned to a system of retroviral transduction, which previously permitted us to express various human β3 integrin mutants in OCs and their precursors (Feng et al., 2001).

Our first exercise established that the β3 cytoplasmic domain is essential for appropriate distribution of the integrin to cytoskeletal structures such as podosomes and lamellipodia. Mirroring their effect on OC spreading and matrix resorption, β3 mutants, lacking the cytoplasmic domain or bearing S⁷⁵²P, distribute abnormally in OCs residing on glass and dentin, failing to colocalize with F-actin. This observation prompted us to ask if the same components of the β3 integrin are involved in modulating the conformational state of the intact heterodimer.

Antibodies recognizing the activated conformation bind the LIBS in the NH₂ terminus of integrin heterodimers (Bodeau et al., 2001). In high calcium buffer, the anti-LIBS mAb AP5 recognizes activated, high-affinity ligand-binding αvβ3 integrin. Importantly, in low calcium buffer, AP5 binds to all αvβ3 and forces all receptors into the activated conformation. This AP5-induced conformational change of αvβ3 involves a direct effect on the external domain of the integrin and does not require the β3 cytoplasmic domain. On the other hand, the heterodimer's biological activity requires the cytoplasmic tail. Thus, β3^−/− OCs bearing β3-ΔC or β3 S⁷⁵²P have decreased adhesive and migratory capabilities, suggesting defective outside-in signaling. In contrast to the direct effect of AP5 in changing the external conformation of β3, HGF and M-CSF modulate adhesion and spreading of mature OCs by activating the αvβ3 integrin via the β3 cytoplasmic domain by inside-out signaling (Insogna et al., 1997; Teti et al., 1998; Faccio et al., 2002).

Active OCs form a stable ring of actin, which delineates the ruffled membrane where the resorptive process takes place. Our previous observations, showing abnormal actin rings in β3-null OCs generated in low dose M-CSF, indicate that β3 contributes to the formation of this structure. We have also found that αvβ3 and M-CSF cooperate during osteoclastogenesis (Faccio et al., 2003). In this study, we find that β3^−/− OCs, or those expressing the human mutation S⁷⁵²P, exhibit normal actin distribution when generated in high dose M-CSF, indicating that M-CSF can compensate for lack of αvβ3 in actin ring formation.

Podosomes, found in adherent OCs, are rosette-like structures containing αvβ3 around an actin core (Marchisio et al., 1988). In contrast to the relatively static actin ring, podosomes are dynamic, rapidly redistributing under the influence of extracellular stimuli such as HGF and M-CSF (Insogna et al., 1997; Teti et al., 1998; Faccio et al., 2002). Here we show that β3 integrin is absolutely required for dynamic changes in the actin cytoskeleton in response to growth factors or cell attachment. β3-ΔC or β3 S⁷⁵²P mutants fail to form lamellipodia when plated on glass (Fig. 4 A) or on dentin (Fig. 4 B). In agreement with these observations, α-actinin, which links actin filaments directly to integrin receptors (Pavalko et al., 1991; Otey et al., 1993), fails to enter the Triton X-100 soluble fraction of β3^−/− pre-Ocs, or β3-ΔC and β3 S⁷⁵²P mutants, in response to growth factors. In other cell types, redistribution of α-actinin, from focal adhesions to the Triton X-100 soluble fraction, is associated with loss of close apposition of cell membrane to the extracellular matrix, consistent with enhanced motility (Greenwood et al., 2000).

Modulation of the OC cytoskeleton is controlled by Rho GTPases. For example, dominant negative Rho arrests podosome organization, OC motility, and bone resorption (Chellaiah et al., 2000), and Rac inhibition decreases the resorptive activity of OCs (Razzouk et al., 1999). The mechanisms of Rho and Rac activation in OCs are poorly defined. We find that the defect in migration and lamellipodia formation in β3 S⁷⁵²P OCs, in response to growth factors, is associated with failure to activate Rho and Rac.

Interestingly, OCs bearing β3 Y⁷⁴⁷F/Y⁷⁵⁹F are indistinguishable from wild-type cells in their appearance and resorptive capabilities, but adhesiveness to OPN is moderately decreased. Recently, phosphorylation levels of the β3 Y747 and 759 have been correlated with strength of binding during adhesion (Boettiger et al., 2001). It is possible that in OCs, these tyrosines are required for stable and strong adhesion, but not for efficient migration. As motility is requisite for bone resorption, the Y⁷⁴⁷F/Y⁷⁵⁹F mutation does not compromise the physiological function of these cells.

The ability of M-CSF to induce OC spreading and actin reorganization depends on c-Src expression (Insogna et al., 1997). Sanjay et al. (2001) have shown that upon adhesion, αvβ3 forms a complex with Pyk2 and c-Src that, in turn, recruits c-Cbl, resulting in podosome assembly. This model holds that c-Cbl binds to Tyr ⁴¹⁶ in the c-Src kinase domain, which down-regulates both Src kinase activity and integrin-mediated adhesion, prompting podosome detachment and subsequent disassembly. Consistent with this hypothesis, we find that activation of c-Src and c-Cbl, in response to VN adherence, is abrogated in β3^−/− OCs and in β3-ΔC and β3 S⁷⁵²P mutants, thereby decreasing podosome turnover and, consequently, OC adhesion and migration.

Our data, however, stand in contrast to the conclusions of Sanjay et al. (2001) and Nakamura et al. (2001), who claim that αvβ3 is essential for Pyk2 phosphorylation. We believe this discrepancy may reflect the fact that we directly assessed Pyk2 phosphorylation in authentic pre-OCs, deleted of the αvβ3 receptor. Two possibilities present themselves as to how OCs lacking αvβ3 phosphorylate Pyk-2. First, Pyk2 activation is calcium dependent (Sanjay et al., 2001). In this regard, the intracellular calcium chelator, BAPTA, blunts Pyk2 autophosphorylation in β3^+/+ and β3^−/− pre-OCs. Second, other adhesive receptors could compensate for the lack of β3 and mediate Pyk2 phosphorylation. Pyk2 activation occurs equally in cells lacking α2β1 integrin or CD44, another receptor for OPN, but not in β2^−/− pre-OCs. Thus, although it is formally possible that α2β1, CD44, and β3 compensate for each other in signaling to Pyk2, the weak Pyk2 phosphorylation detected in β2^−/− cells suggests that the β2 integrin is dominant in this process. In support of this posture, uncommitted macrophages, which have yet to express αvβ3, activate Pyk2 by β2 integrin ligation (Duong and Rodan, 2000), and we find that the same is true in β3^−/− pre-OCs. Despite having the β2 integrin, β3^−/− pre-OCs generated in high dose M-CSF do not retain a macrophage phenotype, as they express markers of committed OCs. As the β2 integrin is not present in fully mature resorptive OCs (Athanasou and Quinn, 1990), in this circumstance, αvβ3 may be the Pyk2 activating receptor.

Pyk2, independent of its phosphorylation status, is associated with paxillin, and this association is increased in adherent cells. Pyk2, however, fails to be recruited to the β3 complex in adherent cells carrying β3-ΔC or S⁷⁵²P. It is possible that the failure of c-Src and c-Cbl to be activated in these mutants reflects the inability of Pyk2 to bind the integrin.

We propose, therefore, that in OCs, cytokines stimulate the formation of new membrane extensions that contain activated αvβ3 (Fig. 10 A). These cytoskeletal rearrangements are under the control of Rho family GTPases and require functional αvβ3 (Fig. 10 B). Upon αvβ3 occupancy, phosphorylated Pyk2, an event independent of the integrin, forms a complex at the β3 cytoplasmic domain with phosphorylated c-Src and c-Cbl (Fig. 10 B). In the absence of functional αvβ3, Pyk2 may be activated by other means, such as αMβ2 or increased calcium. These alternative means of activating Pyk2 permit its association with paxillin, but the Pyk2–c-Src–c-Cbl adhesive complex fails to form, resulting in poorly resorptive OCs.

BMMs were isolated from long bones of 4- to 8-wk-old mice by culturing whole marrow for 3 d in α-MEM containing 10% heat-inactivated FBS and 1:10 CMG supernatant as a source of M-CSF (Faccio et al., 2003).

BMMs were transduced with virus containing vectors that encode for several β3 integrin mutants (Feng et al., 2001), in the presence of 1:10 CMG supernatant and 8 μg/ml polybrene (Sigma-Aldrich), without antibiotic selection. Cells were cultured for an additional 2–3 d before analysis of integrin expression or osteoclastogenesis.

Pre-OCs expressing the different mutants were lifted with Trypsin/EDTA (Sigma-Aldrich) and washed in a calcium-free buffer based on HBSS. Pretreated cells were incubated with HGF or M-CSF in α-MEM supplemented with 0.5% BSA for 30 min at 37°C; control cells were incubated with medium alone. After incubation, cells were washed twice and incubated with the mAb AP5 (50 μg/ml) in high calcium buffer, which recognizes the activated β3 integrin subunit. Binding of AP5 in HBSS calcium-free buffer served as positive control, identifying all β3 on the cell surface. Cells were then incubated with FITC-conjugated secondary Ab, as previously described (Faccio et al., 2002).

β3^+/+ or β3^−/− BMMs, transduced with the indicated mutants, were plated on dentin slices or glass coverslips under osteoclastogenic conditions for 4 d. For some experiments (Figs. 3 and 4), after 4 d in culture, cells were treated with HGF (50 ng/ml) or M-CSF (100 ng/ml), or media + 0.5% BSA as control, for 30 min at 37°C, and then fixed and stained as previously described (Faccio et al., 2002) and observed with a confocal microscope.

Adhesion and haptotactic migration assays were performed using pre-OCs expressing the different β3 mutants plated respectively onto 96-well plates or transwell filters, 8-μm pore size (Costar), coated with 10 μg/ml human OPN.

For both experiments, cells were preactivated with growth factors or AP5 for 30 min in suspension and then plated, and adherent cells were stained with crystal violet. For migration assay, cells that migrated to the lower side were viewed at 300× magnification and counted. Results represent the averages from 15 fields ± SEM of a representative experiment.

BMMs were cultured for 3 d in the presence of 100 ng/ml RANKL and 100 ng/ml M-CSF and starved overnight in the presence of 2% serum. Cells were lifted and replated onto the indicated matrix proteins for 30 or 60 min. In some experiments, adherent OCs were starved and restimulated with 100 ng/ml M-CSF or 50 ng/ml HGF. Cells were lysed in RIPA (Faccio et al., 2003), or in TNE (Lakkakorpi et al., 1999) for coimmunoprecipitation of c-Src and Pyk2. Precleared lysates were immunoprecipitated with 2 μg anti-Pyk2 polyclonal antibodies (Biosource International, CA), 2 μg anti-β3 mAb (clone 7G2), or 2 μg antiphosphotyrosines (PY99) for 1 h at 4°C followed by overnight incubation with protein A/G–Sepharose beads at 4°C (Santa Cruz Biotechnology, Inc.) and then analysis by SDS-PAGE and immunoblotting.

Pre-OCs were lysed in a buffer containing 50 mM Tris/HCl, pH 7.5, 1 mM EDTA, 500 mM NaCl, 10 mM MgCl₂, 1% (vol/vol) Triton X-100, and protease inhibitors (4 μg/ml leupeptin and 30 μg/ml PMSF). Lysates were incubated with glutathione–agarose beads (Sigma-Aldrich) coupled with bacterially expressed GST–RBD fusion protein for Rho pull down or GST–PAK1 for Rac pull down (Ren et al., 1999) at 4°C for 45 min. Bound proteins were analyzed by SDS-PAGE followed by immunoblotting against RhoA (Santa Cruz Biotechnology, Inc.) or Rac1 (Upstate Biotechnology).

The supplemental material (Figs. S1 and S2) is available. Fig. S1 shows the expression levels of c-Fms, α2β1, and β2 in BMMs and pre-OCs. Fig. S2 shows the localization of viniculin, talin, and β3 integrin in the podosomes.

We are thankful to Dr. T. Kunicky (Scripps Research Institute, La Jolla, CA) for providing the mAb AP5 and Dr. S. Blystone (Upstate Medical University, Syracuse, NY) for the mAb 1A2. We also thank Dr. E. Brown (University of California, San Francisco, CA) for the mAb 7G2 and the β2^−/− mice and Dr. P. Noble (Yale University, New Haven, CT) for the CD44^−/− mice.

This work was supported by National Institutes of Health grants AR48812 and AR46852 to F.P. Ross and AR48853, AR46523, AR32788, and DK-56341 (Clinical Nutrition Research Unit) to S.L. Teitelbaum, a grant from the Italian Foundation for Cancer Research (AIRC), and a grant from the Italian Space Agency (ASI) (A. Zallone).