MCP-1 Is Induced by Receptor Activator of Nuclear Factor-κB Ligand, Promotes Human Osteoclast Fusion, and Rescues Granulocyte Macrophage Colony-stimulating Factor Suppression of Osteoclast Formation*

Human osteoclast formation from monocyte precursors under the action of receptor activator of nuclear factor-κB ligand (RANKL) was suppressed by granulocyte macrophage colony-stimulating factor (GM-CSF), with down-regulation of critical osteoclast-related nuclear factors. GM-CSF in the presence of RANKL and macrophage colony-stimulating factor resulted in mononuclear cells that were negative for tartrate-resistant acid phosphatase (TRAP) and negative for bone resorption. CD1a, a dendritic cell marker, was expressed in GM-CSF, RANKL, and macrophage colony-stimulating factor-treated cells and absent in osteoclasts. Microarray showed that the CC chemokine, monocyte chemotactic protein 1 (MCP-1), was profoundly repressed by GM-CSF. Addition of MCP-1 reversed GM-CSF suppression of osteoclast formation, recovering the bone resorption phenotype. MCP-1 and chemokine RANTES (regulated on activation normal T cell expressed and secreted) permitted formation of TRAP-positive multinuclear cells in the absence of RANKL. However, these cells were negative for bone resorption. In the presence of RANKL, MCP-1 significantly increased the number of TRAP-positive multinuclear bone-resorbing osteoclasts (p = 0.008). When RANKL signaling through NFATc1 was blocked with cyclosporin A, both MCP-1 and RANTES expression was down-regulated. Furthermore, addition of MCP-1 and RANTES reversed the effects of cyclosporin A and recovered the TRAP-positive multinuclear cell phenotype. Our model suggests that RANKL-induced chemokines are involved in osteoclast differentiation at the stage of multinucleation of osteoclast precursors and provides a rationale for increased osteoclast activity in inflammatory conditions where chemokines are abundant.

Osteoclasts are bone-resorbing cells that differentiate from hematopoietic precursors of the monocyte/macrophage lineage (1). Osteoclasts are multinuclear giant cells that stain positive for tartrate-resistant acid phosphatase (TRAP). 1 Receptor ac-tivator of nuclear factor-B ligand (RANKL) and macrophage colony-stimulating factor (M-CSF) are necessary signals for osteoclast differentiation (2). RANKL is present on the surface of stromal cells and osteoblasts (3). RANKL interacts with receptor activator of nuclear factor-B on osteoclast precursors, resulting in a cascade of gene expression controlled by transcription factors including nuclear factor-B and NFATc1 (4). Authentic human osteoclasts that have high bone-resorbing activity are made in vitro from peripheral blood mononuclear cells (PBMCs) by culturing with M-CSF and recombinant RANKL (MϩR treatment).
Granulocyte macrophage colony-stimulating factor (GM-CSF) is a cytokine produced by T cells following activation and by most myeloid lineage cells, such as macrophages and granulocytes (5). The effect of GM-CSF on osteoclast formation is controversial: both inhibition (6 -9) and stimulation (10) are reported. Short-term treatment with GM-CSF potentiated osteoclast differentiation, whereas long-term exposure suppressed osteoclast differentiation (11). We previously showed that the GM-CSF receptor-␣ was induced during osteoclast differentiation (9), providing a basis for paracrine signaling from GM-CSF-producing cells to osteoclasts.
Chemokines are small cytokines known to be involved in immune response and in development of several cell types (12). Chemokines are classified into two main subfamilies, CC and CXC, according to the location of the first two of the four cysteine residues (13). Many ligands within the CC chemokine superfamily are capable of sharing receptors (12). Monocyte chemotactic protein 1 (MCP-1) is a CC chemokine commonly found at the site of tooth eruption, rheumatoid arthritic bone degradation, and bacterially induced bone loss (14). MCP-1 is expressed by mature osteoclasts, and its expression is regulated by nuclear factor-B (15,16). In this report we show that GM-CSF suppresses the formation of TRAP-positive multinuclear osteoclasts by RANKL and M-CSF. Gene expression studies show that GM-CSF treatment causes potent down-regulation of MCP-1. Addition of exogenous MCP-1 reversed the GM-CSF-mediated suppression of osteoclast formation, permitting the recovery of authentic multinuclear bone-resorbing osteoclasts.

EXPERIMENTAL PROCEDURES
Preparation and Culture of Human Monocytes-Human PBMCs were isolated by Ficoll-Paque (Amersham Biosciences) density gradient centrifugation as previously described (9). PBMCs were plated at 10 6 * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18  cells/cm 2 and non-adherent cells removed by washing in normal saline. Cells were cultured in minimal essential medium (supplemented with 10% fetal bovine serum, 1% penicillin-streptomycin (Invitrogen)), 5% CO 2 supplemented with 25 ng/ml M-CSF and 20 ng/ml soluble RANKL to induce osteoclast formation. GM-CSF was used at concentrations from 0.1 to 25 ng/ml. MCP-1 and RANTES were used at 25 ng/ml. Neutralizing antibody directed against MCP-1 was used at 4 g/ml according to the manufacturer's protocols. Control antibody was goat anti-rabbit IgG used at 10 g/ml (Serotec). GM-CSF, RANKL, M-CSF, MCP-1, RANTES, and neutralizing anti-MCP-1 antibody were purchased from Peprotech (Rocky Hill, NJ). All data are based on a minimum of three replicate experiments performed independently on different occasions, unless otherwise stated. All cultures were for 21 days. After 21 days, PBMC cultures were fixed in acetone, citrate, and formaldehyde solution and stained for TRAP using a leukocyte acid phosphatase staining kit (Sigma). TRAP-positive cells that had three or more nuclei were considered multinuclear. Bone resorption assays were performed on dentine slices in 96-well plates as previously described (11). Dentine slices were sputter-coated with gold and observed by scanning electron microscopy.
Flow Cytometry Analysis of CD1a Expression-Cells cultured on BioCoat collagen I plates (BD Biosciences) for 21 days were dissociated using cell dissociation buffer (Invitrogen), incubated with fluorescein isothiocyanate-conjugated human CD1a antibody (Chemicon, Temecula, CA) for 45 min on ice, and washed with phosphate-buffered saline prior to flow cytometry (FACS Calibur; BD Biosciences). The unstained cells were gated out and data acquisition and analysis were done using CellQuest software (BD Biosciences).
RNA Studies-At 21 days, cultures were lysed using 4 M guanidium isothiocyanate, 1% lauryl sarcosine, and total RNA pelleted through a 5.7 M cesium chloride, 100 mM EDTA cushion by ultracentrifugation in a Beckman SW41 rotor at 27,000 rpm for 16 h (17). Total RNA was converted into cDNA using ImProm-II reverse transcriptase (RT; Promega) and oligo(dT) primer. Quantitative PCRs were performed and analyzed using SYBR Green I Supermix (Bio-Rad) in a Bio-Rad i-Cycler (9). Primers and conditions for quantitative PCR assays were as described previously (9, 18) except for MCP-1 assays that used primers 5Ј-TCGCGAGCTATAGAAGAATCA-3Ј and 5Ј-TGTTCAAGTCTTCG-GAGTTTG-3. Gene arrays containing 19,000 duplicate spotted cDNA representing human genes were hybridized and analyzed according to the manufacturer's protocols (University of Ontario Cancer Center).
Statistical Analysis-Analysis of variance with Fisher's post hoc t test was used to determine significance of effects. Data are presented as mean values Ϯ S.E.

Phenotype of Cells Treated with GM-CSF in the Presence of RANKL and M-CSF-
We clarified the effect of continuous exposure of human PBMC to GM-CSF on osteoclast differentiation mediated by MϩR treatment (Fig. 1). The appearance of TRAP-positive multinuclear cells was suppressed dose dependently by GM-CSF, with osteoclast differentiation suppressed up to 97% by exposure to 25 ng/ml GM-CSF (Fig. 1A). Furthermore, delayed addition of GM-CSF had similar suppressive effects (Fig. 1B). Whereas normal osteoclasts differentiated using MϩR treatment were TRAP-positive and multinuclear with potent bone resorption (Fig. 1C), cultures treated with maximal doses of GM-CSF and MϩR treatment were TRAPnegative, mononuclear, and completely negative for bone resorption (Fig. 1C).
To further characterize the phenotype of cells treated with GM-CSF in the presence of RANKL and M-CSF, the expression of a series of osteoclast-related genes was examined ( Fig. 2A). Gene expression profiles were compared between osteoclasts (MϩR treatment, 25 ng/ml M-CSF and 20 ng/ml RANKL), macrophage-like cells (M-CSF-alone treatment, 25 ng/ml), and cells derived from continuous treatment with GM-CSF (25 ng/ ml) with RANKL and M-CSF (GMR treatment) as above. The GM-CSF receptor-␣ was induced in osteoclasts relative to macrophages and was further up-regulated in GMR treatment ( Fig. 2A). The up-regulation of GM-CSF receptor-␣ by RANKL provides an explanation why delayed addition of GM-CSF is still potent at suppressing osteoclast differentiation, because cells may be sensitized to inhibition by increased receptor. We previously established a series of nuclear factors, including NFATc1, that are regulated strongly by RANKL during osteoclast differentiation (9). These nuclear factors show pronounced differential patterns of expression in the three treatment groups ( Fig. 2A). In particular, the RANKL-mediated induction of NFATc1 is suppressed by GM-CSF in GMRtreated cells relative to osteoclasts. Because NFATc1 is considered necessary for induction of osteoclast genes (16,19), inhibition of NFATc1 induction provides a rationale for the suppression of the osteoclast phenotype by GM-CSF. In contrast, the potent RANKL-mediated induction of far upstream element-binding protein was essentially unaffected by GM-CSF in GMR treatment compared with MϩR treatment. Other osteoclast nuclear factors, GABP␣ and ILF3, were potently down-regulated by GM-CSF, similar to the effect on NFATc1. Although subunits of the same factor, GABP␣ and GABP␤ were differentially down-regulated in GMR-treated cells. KOX31 is a zinc finger protein that is highly expressed in macrophage-like cells and repressed in osteoclasts (9). GM-CSF does not reverse the RANKL-mediated repression of KOX31; in fact, KOX31 is further repressed in GMR-treated cells relative to MϩR treatment ( Fig. 2A). The further repression of KOX31 by GMR eliminates any similarity between the M-CSF-treated cells and the GMR-treated cells, because potent up-regulation of KOX31 occurs in macrophage-like cells. The overall pattern of gene expression indicates that GMR-treated cells are not similar to M-CSF-derived macrophage-like cells and probably represent a dendritic cell phenotype, because the dendritic marker CD1a (20) was up-regulated in GMR treatment (Fig. 2, B and C).
GM-CSF Suppresses RANKL-mediated Induction of MCP-1-Because GMR treatment suppressed NFATc1, we hypothesized that GM-CSF in the presence of RANKL and M-CSF inhibits the induction of key factors in osteoclast differentiation. Such factors would appear suppressed by GMR treatment compared with MϩR treatment. A large number of genes demonstrated significant differential expression in microarray analysis (data not shown), although only a few are discussed. The expression of various genes was consistent with suppression of the osteoclast phenotype by GM-CSF: cathepsin D, vacuolar Hϩ ATPase proton pump (ATP6C), NFATc1, and the osteoclast protease cathepsin K were repressed (data not shown). The GM-CSF receptor-␣ was up-regulated, in keeping with the real-time PCR analysis ( Fig. 2A). Of 19,000 genes assayed, the gene most repressed by GMR treatment in comparison to MϩR treatment was MCP-1, a CC chemokine not associated previously with osteoclast differentiation. To verify the regulation of MCP-1 by GM-CSF, new cultures were established on three separate occasions, differentiated over 21 days under comparable conditions with M-CSF alone, MϩR, and GMR treatment (Fig. 2D). MCP-1 mRNA content, measured by quantitative real-time PCR, was 17.1 Ϯ 0.1 (copies/ng total RNA) in MϩR cultures compared with 0.24 Ϯ 0.06 in GMRtreated cultures, meaning that GM-CSF treatment results in a 72-fold decrease in MCP-1 mRNA (p ϭ 5 ϫ 10 Ϫ9 ). Furthermore, MCP-1 was induced by RANKL during osteoclast differentiation (15-fold) compared with macrophage cultures (M-CSFalone treatment). These data confirm that GM-CSF represses MCP-1 expression in the GMR treatment compared with MϩR and also show that MCP-1 is induced in osteoclasts relative to macrophage-like cells. Another CC chemokine, RANTES, is induced during human (9) and mouse (15,16) osteoclast differentiation. RANTES induction by RANKL was also suppressed by GM-CSF (Fig. 2D).
Chemokine Effects on Osteoclasts-We reasoned that exogenous MCP-1 would affect osteoclast differentiation. Adding MCP-1 to the standard MϩR treatment protocol resulted in 33% more osteoclasts (329 Ϯ 26 n ϭ 14 versus 252 Ϯ 13 n ϭ 19, respectively; p ϭ 0.008) (Fig. 3A) that were TRAP-positive and positive for bone resorption (Fig. 3B). Unexpectedly, MCP-1 treatment with M-CSF in the absence of exogenous RANKL resulted in TRAP-positive, multinucleated giant cells (Fig. 3B). Although these cells had the appearance of osteoclasts, they were unable to form resorption pits on dentine, suggesting that MCP-1 and M-CSF treatment results in an intermediate phenotype on the path to osteoclasts, permitting monocyte fusion but not further differentiation. Like MCP-1, RANTES treatment with M-CSF resulted in multinuclear TRAP-positive cells that were unable to degrade bone (Fig. 3D). Unlike MCP-1, exogenous RANTES had no effect on osteoclast number or bone resorption in the presence of RANKL (Fig. 3, C and D). MCP-1-mediated effects were sensitive to specific neutralizing anti-MCP-1 antibody, preventing the MCP-1-mediated enhancement of osteoclast formation observed in MϩR-treated cultures and inhibiting the formation of multinuclear cells in cultures treated with MCP-1 and M-CSF (Fig. 3, A and C). Neutralizing anti-MCP-1 antibody significantly reduced the number of osteoclasts in standard MϩR cultures (p ϭ 0.01) (Fig. 3C). Control antibody had no effect on osteoclast number (data not shown, p ϭ 0.59).
Treatment with MCP-1 and M-CSF resulted in multinuclear cells that had the appearance of osteoclasts but were negative for bone resorption (Fig. 3B). The levels of expression of two osteoclast-related genes, TRAP and cathepsin K, were compared by quantitative PCR in cultures treated with MCP-1 and M-CSF compared with standard MϩR treatment (Fig. 3E). The mRNA content of TRAP was virtually identical in the multinuclear TRAP-positive non-bone-resorbing cells from MCP-1 and M-CSF treatment compared with authentic osteoclasts. In contrast, cathepsin K mRNA content was substantially lower in MCP-1-and M-CSF-treated cells. This may provide a rational for the inability of these osteoclast-like TRAP-positive multinu-clear cells to degrade bone: cathepsin K and possibly other essential osteoclast-related genes are not appropriately induced in MCP-1-treated cells.

MCP-1 Reverses GM-CSF Repression of Osteoclast
Differentiation-MCP-1 acts as an enhancer of osteoclast differentiation and was strongly repressed by GM-CSF. We hypothesized that the lack of MCP-1 may be influential in GM-CSF-mediated suppression of osteoclast differentiation from PBMCs. This hypothesis was tested by addition of exogenous MCP-1 under conditions of maximal suppression of osteoclast differentiation by GM-CSF (25 ng/ml). The addition of MCP-1 (at 25 ng/ml) to the GMR treatment protocol dramatically increased the formation of TRAP-positive multinucleate cells (p ϭ 3.8 ϫ 10 Ϫ8 ; Fig.  4A). Although the multinuclear cells derived from MCP-1treated GMR cultures were generally smaller than osteoclasts derived from control MϩR-treated cultures (47 Ϯ 4 and 148 Ϯ 22 m, respectively; average longest axis, p ϭ 2.2 ϫ 10 Ϫ5 ), they were almost twice as abundant and were positive for bone resorption (Fig. 4B). Although RANTES overcame the block in multinucleation imposed by GM-CSF, resulting in a similar number of TRAP-positive multinucleate cells as in MϩR controls, the cells were negative for bone resorption (Fig. 4B). If MCP-1 and RANTES are involved in cell fusion events, we reasoned that these chemokines should be able to rescue multinuclearity from blockade of osteoclast formation by other chemical agents, such as cyclosporin A, a blocker of NFATc1 activation by calcineurin. We showed in human (9), and others in mouse (16,19,21), that cyclosporin A inhibits the formation of multinuclear cells in cultures treated with RANKL and M-CSF. Cell cultures exposed to MϩR treatment and maximally repressive concentrations of cyclosporin A (1 g/ml) showed virtually no multinuclear cells (Fig. 4C). In addition, the differentiation of human osteoclasts on dentine and bone resorption activity is potently inhibited by cyclosporin A at 1 g/ml (data not shown). In marked contrast, MCP-1 or RAN-TES addition showed a dramatic recovery of TRAP-positive multinuclear cells in cultures treated with cyclosporin A and MϩR (Fig. 4C). These cells have the appearance of osteoclasts but are negative for bone resorption (Fig. 4D). Under cyclosporin A blockade, the RANKL-mediated induction of MCP-1 was repressed completely, consistent with nuclear factor-B or NFATc1 being responsible for the induction of MCP-1 during osteoclast differentiation (Fig. 4E). RANTES induction was repressed similarly. MCP-1 was the most potently down-regulated gene in microarray analysis of the inhibitory effect of GM-CSF. Furthermore, exogenous MCP-1 was able to recover authentic osteoclasts from GM-CSF inhibition of osteoclast differentiation. In the presence of sufficient MCP-1, the balance of cell fate is changed back toward the RANKL signal, countervailing the GM-CSF repressive signal. We conclude that the absence of MCP-1 in GMR-treated cultures is a key deficit in osteoclast differentiation. Once this deficit is overcome, by adding exogenous MCP-1, osteoclast differentiation can proceed in the presence of RANKL. The smaller size of the osteoclasts formed under these conditions may reflect a further inhibitory effect of GM-CSF on osteoclasts, even after rescue by MCP-1, because the up-regulation of the GM-CSF receptor by RANKL was unaffected by the presence of GM-CSF.
To date, MCP-1 has not been implicated in osteoclast function, although data exist on its role in recruitment of monocytes during tooth eruption (14). Our data on the effects of MCP-1 and RANTES on osteoclast differentiation may explain why inflammatory diseases that feature increased chemokine activity, such as rheumatoid arthritis, are associated with increased osteoclast activity leading to bone degradation. We demonstrated previously that CCR2, the receptor for MCP-1, is also induced by RANKL, providing evidence for an autocrine cycle involving MCP-1. Furthermore, both MCP-1 and RANTES treatment resulted in multinucleated cells in the absence of RANKL, suggesting that chemokines are sufficient for fusion events.
MCP-1 overcomes GM-CSF-mediated repression of osteoclast differentiation, permitting the cells to pass through multinucleation to authentic bone-resorbing osteoclasts. Cyclosporin A and GM-CSF similarly inhibit both osteoclast differentiation and MCP-1 expression; GM-CSF represses NFATc1 induction, whereas cyclosporin A inhibits activation of NFATc1 by blocking calcineurin. Under continuous chemical blockade of NFATc1 by cyclosporin A, a recovery of bone re- Further chemokine production leads to the final stage, where sufficient chemokines are produced to stimulate an event that we term "pre-fusion," that is, the fusion of monocyte precursors into multinucleate giant cells prior to fusion with the cell receiving the RANKL signal. Although multinuclear cells derived from pre-fusion may be TRAP-positive, they are not competent for bone resorption, possibly because of a deficit in cathepsin K expression. This model explains a role for chemokines in RANKL signaling leading to osteoclast differentiation and also suggests that pathology associated with high levels of MCP-1 will result in both multinucleate giant cells and increased osteoclast differentiation. sorption activity would not be expected even if MCP-1 recovered the multinuclear phenotype, unless a calcineurin-independent pathway of NFATc1 activation exists in osteoclasts. As predicted from the continuous blockade of NFATc1 by cyclosporin A, MCP-1 and RANTES were unable to recover the bone resorption phenotype (see Fig. 4D). MCP-1 recovered bone resorption in GM-CSF-inhibited cultures but did not recover bone resorption in cyclosporin A-inhibited cultures. RANTES induction by RANKL was similarly repressed by cyclosporin A, and RANTES treatment of cyclosporin A blockaded MϩR-treated cells changed the cell phenotype from TRAP-positive mononuclear to TRAP-positive multinuclear cells.
Both MCP-1 and RANTES recover the multinuclear phenotype in cyclosporin A blockade of osteoclast differentiation but do not recover bone resorption, providing evidence that chemokines are involved in cell fusion events during osteoclast differentiation. In our experiments, although MCP-1 and RAN-TES were associated with cell fusion events, the presence of RANKL was necessary for bone resorption, as treatment with either MCP-1 or RANTES and M-CSF leads to multinuclear cells but not bone resorption. Taken together, these data suggest that RANKL induction of MCP-1 and RANTES is an important component of osteoclast differentiation, providing an autocrine signal acting on the osteoclast and a paracrine signal that acts on osteoclast precursors to accelerate osteoclast differentiation by promoting fusion.
The expression of osteoclast marker genes in cells treated with MCP-1 and M-CSF suggests that some osteoclast characteristics can be acquired independently of RANKL. Because RANKL induces the MCP-1 receptor (CCR2), RANKL induction of MCP-1 sets up both autocrine (affecting the osteoclastproducing MCP-1) and paracrine pathways (affecting cells destined to fuse with the RANKL-stimulated osteoclast). We present a model for chemokine action during osteoclast differentiation (Fig. 5). Osteoclasts form by fusion of RANKϩ mononuclear precursors after contact with a cell expressing RANKL. Intimate cell-cell contact is necessary in vivo for RANKL signaling, a process mimicked in vitro with soluble recombinant RANKL. An osteoclast precursor cell in contact with a RANKLpresenting cell (Fig. 5) will receive the RANKL signal and initiate a cascade of gene expression that includes the production of MCP-1, RANTES, and possibly other chemokines. MCP-1 and RANTES are chemotactic signals for monocytes, resulting in migration to the source of production of the chemokine. The data show that either MCP-1 or RANTES can cause cell fusion in monocytes treated with M-CSF. We propose that cell fusion is a key event in the next step of osteoclast differentiation (Fig. 5) where monocyte-like cells, that have not yet seen RANKL are attracted by chemokines to the site of coupling of the RANKϩ precursor and the RANKL-presenting cell. Chemokine-mediated fusion increases the size of the osteoclast and also transfers the RANKL signal to the additional nuclei that are now in the multinucleated cell. We show that fusion can occur in the absence of RANKL but that bone re-sorption depends on RANKL. If the chemokine signal is strong enough, such monocyte cell fusion could occur prior to contact with the RANKL-influenced osteoclast precursor. Such TRAPpositive "prefused" cells would still require the RANKL signal to develop into an authentic osteoclast capable of bone resorption because they have a deficit in cathepsin K expression.
In conclusion, MCP-1 is induced by RANKL during osteoclast differentiation and is sensitive to cyclosporin A and repressed by GM-CSF but is able to overcome GM-CSF repression of osteoclast differentiation. Furthermore, chemokines stimulate the formation of fused cells in the absence of RANKL and overcome the block in fusion caused by cyclosporin A. These data indicate that chemokines play a crucial role in osteoclast function.