Distinct Osteoclast Precursors in the Bone Marrow and Extramedullary Organs Characterized by Responsiveness to Toll-Like Receptor Ligands and TNF-α ¹ Free

Osteoclasts play essential roles in bone remodeling, bone resorption, and tooth eruption (1, 2). They are large, multinucleated cells located on endosteal bone surfaces and the periosteal surface beneath the periosteum, and they are tightly attached to bone matrix (1). In contrast to mature osteoclasts, precursor cells that have the potential to differentiate into mature functional osteoclasts in culture are widely distributed throughout the body (1, 3, 4).

Two cytokines critical for osteoclast differentiation have been reported. One is M-CSF, which is produced by various types of mesenchymal cells, including stromal cells and osteoblasts (5). Another factor is receptor activator of NF-κB (RANK) ⁴ ligand (RANKL), which is produced by stromal cells, osteoblasts, and certain immunocompetent cells (6, 7). RANKL is detected in a variety of extramedullary organs, including spleen, thymus, lymph node, and lung (6). In culture, the presence of these two factors is sufficient to support the differentiation of precursor cells into mature osteoclasts. Because osteoprotegerin (OPG), a decoy receptor for RANKL is also present, osteoclastogenesis may be negatively regulated in extraskeletal tissues in vivo (8, 9). However, inhibition of osteoclast differentiation in extramedullary organs by OPG is not likely, because OPG (gene symbol: Tnfrsf11b) null mice have dramatically increased numbers of osteoclasts, but osteoclasts are detected only in the bone tissues (10, 11). Mechanisms that restrict osteoclastogenesis to the bone tissues, are not obvious.

Products of microbes such as LPS are known to accelerate bone lysis (12, 13, 14). LPS signaling via the Toll-like receptor (TLR) shares downstream pathways with RANKL/RANK signaling (12, 15, 16, 17). However, signaling via TLRs has never been reported to mimic RANKL/RANK signaling or to induce osteoclastogenesis in the absence of RANKL (12). Although mouse TNF-α has recently been reported to induce osteoclastogenesis in vitro (18, 19), neither TNF-α nor TLR ligands substitutes for RANKL/RANK function in vivo, as shown by the fact that RANKL- or RANK-deficient mice show severe osteopetrosis (20, 21).

Many studies have evaluated osteoclastogenic molecules produced by bone tissues (1, 12). In contrast, few reports have demonstrated differences between intramedullary and extramedullary osteoclast precursors (3, 4). In this study, we characterized osteoclast precursors by their responsiveness to signaling via TLRs and TNF-α. Each tissue tested in this study contained cells that had the potential to differentiate into mature osteoclasts, but their characteristics were not identical.

C57BL/6 and (C57BL/6 × DBA/2) (B6D2)F₁ mice were purchased from Japan CLEA (Yokohama, Japan) and C3H/HeN and C3H/HeJ mice were purchased from SLC (Shizuoka, Japan). B6C3FeJ-a/a-Csf1^op/Csf1^op homozygotes and their littermates (+/?) were raised at The Jackson Laboratory (Bar Harbor, ME).

C7-TY is a macrophage-like subline of the C7 cell line, which was established from 129-trp53^tm1TyJ bone marrow (BM) cells (22, 23) and maintained in α-MEM (Life Technologies, Grand Island, NY) supplemented with 10% FBS (JRH Biosciences, Lenexa, KS), 50 U/ml streptomycin, 50 μg/ml penicillin (Meiji Seika Kaisha, Tokyo, Japan), and 50 ng/ml recombinant human M-CSF (a gift from Otsuka Pharmaceutical, Tokushima, Japan) at 37°C with 5% CO₂ in a humidified incubator. C7-TY cells were harvested by treatment with 0.25% trypsin/EDTA for 10–15 min at 37°C.

The BM-derived stromal cell line ST2 was plated in a 24-well plate (Corning Costar, Corning, NY) 1 day before responder cells were added (14, 24). Cells were cultured in α-MEM containing 10% FBS, streptomycin, and penicillin with or without 10⁻⁸ M 1α,25-dihydroxyvitamin D₃ (1α,25(OH)₂D₃) (Biomol Research Laboratories, Plymouth Meeting, PA) and 10⁻⁷ M dexamethasone (Dex; Sigma-Aldrich, St. Louis, MO). Numbers of tartrate-resistant acid phosphatase-positive (TRAP⁺) multinuclear cells (MNCs) were expressed as the mean ± SD of triplicate cultures (25).

Mice were killed by cervical dislocation under ether anesthesia. BM cells were collected by flushing femoral shafts using a 26G sterile needle. Cells from the peritoneal cavity were obtained by injecting 4–8 ml of ice-cold α-MEM containing 10% FBS. Livers and spleens were dissociated to single cell suspensions by disruption in medium between frosted glass slides. To deplete BM stromal cells, 20 × 10⁶ freshly prepared BM cells were applied in a 10-ml volume of Sephadex G-10 column (Amersham Pharmacia Biotech, Uppsala, Sweden), incubated at 37°C for 30 min, and eluted 8 ml with prewarmed medium (4).

LPS from Escherichia coli 055 B5 (Difco, Detroit, MI) and Salmonella minnesota Re 595 (Sigma-Aldrich) were used for in vitro and in vivo experiments. Most experiments used S. minnesota R595 LPS, unless otherwise indicated. Peptidoglycan (PGN) from Staphylococcus aureus (Fluka, Buchs, Switzerland), used to stimulate TLR2, was dissolved in water, sonicated, and sterilized in a hot water bath (26). A phosphorothioated oligonucleotide (ODN), 5′-TCC ATG ACG TTC CTG ATG CT-3′ (CpG), used as an unmethylated CpG ODN to stimulate TLR9, and a control phosphorothioated ODN, 5′-GCT TGA TGA CTC AGC CGG AA-3′, were purchased from Hokkaido System Science (Hokkaido, Japan) (27).

C7-TY cells, BM cells, fetal liver or newborn spleen cells (1–2 × 10⁴/well), peritoneal exudate cells (PECs) (2–10 × 10⁴/well) or adult spleen cells (20 × 10⁴/well) were cultured in 24-well plates (Corning Costar) with 1 ml of α-MEM supplemented with 10% FBS and antibiotics in the presence of 25 or 50 ng/ml recombinant human soluble RANKL (PeproTech, London, U.K.) and/or 50 ng/ml human M-CSF for 6 days. According to the manufacturer’s data sheet, RANKL was produced by E. coli, but the endotoxin level in the material was <0.1 ng/μg RANKL. Human recombinant OPG, a decoy receptor for RANKL, was purchased from PeproTech. Cultures were fed every 2 or 3 days by replacing spent medium with fresh medium.

Rat anti-mouse Fms (AFS98) (28) and Kit (ACK2) (29) antagonistic Abs were purified from ascites. An antagonistic rat anti-mouse TNF-α mAb (XT3) was purchased from Endogen (Woburn, MA), and used at 5 or 10 μg/ml for inhibition of TNF-α activity. A nonantagonistic anti-mouse Kit Ab (ACK4) (29) was used as control.

Inhibitors of the MAPK signaling pathway, PD098059 (2′-amino-3′ methoxyflavone; Wako Pure Industry, Kyoto, Japan) for extracellular signal-regulated kinase (ERK)1/2, and SB203580 (4-(4-fluorophenyl)-2-(4-methylsulphinylphenyl)-5-(4-pyridyl)1H-imidazole; Wako Pure Industry) for p38, were dissolved to 20 mM in DMSO and used at 20 μM.

Data are presented as means ± SD. Statistical significance was assessed by Student’s t test.

In the presence of M-CSF and RANKL, multinuclear osteoclasts expressing TRAP were differentiated from cells in the adult BM, fetal liver, newborn and adult spleen, and adult peritoneal cavity, and from a macrophage-like cell line, C7-TY. Under these conditions, addition of LPS inhibited osteoclastogenesis of adult spleen cells, PECs, and C7-TY cells, but not that of adult BM, newborn spleen, or fetal liver cells (Fig. 1 A). The effect of LPS on osteoclastogenesis depended on the type of tissue and the age of the mice, suggesting that the response of osteoclast precursors to LPS might vary with tissue location and age.

We assessed the incubation time needed to generate TRAP⁺ cells from adult BM cells, spleen cells, and PECs in the presence of M-CSF and RANKL. On day 2 of culture, no TRAP⁺ mononuclear cells or TRAP⁺ MNCs were detected. On day 4, all cultures contained TRAP⁺ MNCs, and the number of these cells increased during an additional 2 days (Fig. 1 B).

One receptor for LPS activation is known to be TLR4, a member of the TLR family (15). To assess whether the inhibitory effect on osteoclastogenesis is specific for LPS/TLR4, or is a general characteristic of signaling via TLR family members, we examined the effects of ligands for other TLRs, i.e., PGN for TLR2 (26), and unmethylated CpG for TLR9 (27), on osteoclastogenesis induced by recombinant M-CSF and RANKL. LPS, PGN, and CpG inhibited osteoclastogenesis of PECs from C3H/HeN mice (Fig. 2,A). Osteoclastogenesis of BM cells was not inhibited by PGN or LPS, but CpG clearly inhibited it (Fig. 2,B). When C3H/HeJ PECs and BM cells expressing a defective TLR4 were used as osteoclast precursors (15), LPS had no effect on osteoclastogenesis (Fig. 2, A and B).

The dose response of each TLR ligand for inhibiting osteoclastogenesis of PECs was determined (Fig. 2 C). Osteoclastogenesis was inhibited by as little as 0.2 ng/ml LPS, 0.1 μg/ml PGN, or 0.01 μM CpG. These results indicated that signaling via TLRs inhibited the differentiation of osteoclast precursors in the peritoneal cavity. In contrast, TLR9, but not TLR2 or TLR4, signaling selectively affected osteoclast precursors in the BM.

A cloned macrophage-like cell line, C7-TY, is able to differentiate into mature TRAP⁺ MNCs in the presence of RANKL alone, and the addition of M-CSF increases the number of TRAP⁺ MNCs that are induced in culture (22, 23). Osteoclastogenesis induced by RANKL + M-CSF was inhibited to the level induced by RANKL alone by the addition of antagonistic anti-Fms, an M-CSF receptor Ab, but osteoclastogenesis induced by RANKL was not inhibited (Fig. 3,A). Under the culture conditions used, the presence of LPS, PGN, or CpG inhibited osteoclastogenesis of C7-TY cells induced by either RANKL alone or by RANKL + M-CSF (Fig. 3). This result indicates that TLR ligands might act directly on osteoclast precursors and inhibit the RANK signaling pathway.

In both cultures with RANKL only (Fig. 3,B, upper graph), and RANKL + M-CSF (Fig. 3,B, lower graph), TRAP⁺ MNCs were detected on day 4, but not on day 2 (Fig. 3 B).

It has been reported that LPS accelerates bone lysis (12, 13). Therefore, we evaluated whether LPS interacts with stromal cells and/or osteoblasts, which regulate osteoclast development. For this study, we used another culture system for inducing osteoclast development. BM cells were cocultured with ST2 stromal cells in the presence of 1α,25(OH)₂D₃ and Dex. Unstimulated ST2 cells produce M-CSF and OPG, a decoy receptor for RANKL. Addition of 1α,25(OH)₂D₃ and Dex inhibits OPG production, and induces RANKL production (6). Addition of LPS significantly increased the number of TRAP⁺ MNCs in cultures of C3H/HeN but not C3H/HeJ BM cells (Fig. 4). The effects of CpG and PGN on the stromal cell-dependent culture were comparable to those obtained with BM cells cultured with rM-CSF and RANKL in both strains of mice (Figs. 2 B and 4A).

To assess the influence of stromal cells contaminating the BM cell fraction, BM cells were passed through G-10 columns to remove stromal cells/osteoblasts and mature macrophages. Similar results were obtained when G-10-passaged BM cells were cultured with or without LPS in the presence of M-CSF and RANKL (Fig. 4 B).

To examine whether cell populations in BM rescued the osteoclast development, mixing cultures with spleen cells or PECs and BM cells were performed. Mixed cells were cultured with or without LPS in the presence of M-CSF and RANKL for 6 days. The number of TRAP⁺ MNCs generated in culture were additive, and the rescue of inhibition by LPS was not observed (Fig. 4, C and D).

In experiments involving the addition of CpG to BM cells (Fig. 5 A) (30) or LPS to PECs or C7-TY cells (data not shown) for only a portion of the culture period, the inhibitory effect on osteoclastogenesis was marked only in the early phase (0–2 days) of cultures, but not in the middle (2–4 days) or in the late phase (4–6 days).

Before the induction of osteoclast development, BM cells were pulsed with LPS, CpG, or control ODN for 2 h on ice. The numbers of recovered cells after pulsing with medium, LPS, control ODN, or CpG were comparable. After thorough washing, these cells were cultured with M-CSF and RANKL for 6 days. We could not detect any inhibitory effect of pulsing with any of these reagents on osteoclastogenesis (Fig. 5,B). Osteoclastogenesis was completely inhibited by the continuous presence of CpG, but not LPS, in the cultures (Fig. 5 B).

We also performed similar experiments using PECs pulsed with LPS for 2 h on ice in the presence or absence of M-CSF, RANKL, or M-CSF + RANKL. After thorough washing, the pulsed cells were counted, and no difference was observed in the number of recovered cells. The cells were cultured with M-CSF and RANKL for 6 days, and the number of TRAP⁺ MNCs was counted. As shown in Fig. 5 C, osteoclastogenesis of the LPS-pulsed cells was completely inhibited. In addition, in the absence of LPS, cells pulsed with RANKL, but not with M-CSF or M-CSF + RANKL, had a slightly diminished number of TRAP⁺ MNCs. Because the medium for pulsing contained 10% FBS, LPS-binding proteins and unknown TLR ligands might have been present. To rule out effects of serum components, we also performed the pulsing in α-MEM supplemented with purified 1% BSA, and the same results were obtained (data not shown).

To assess the signaling pathway involved in LPS-induced inhibition, MAPK inhibitors were added during the LPS pulsing. Partial, but significant, recovery of osteoclast development from PECs was observed in the presence of PD098059, an inhibitor of ERK, but not SB203580, an inhibitor of p38 MAPK (Fig. 5,C) (31). As reported, continuous addition of PD098059 or SB203580 inhibited M-CSF and RANKL induced osteoclastogenesis from BM cells, spleen cells, and PECs (Fig. 5,D) (32). This suggests that the inhibitory effect of LPS on PEC osteoclastogenesis might be related, further with MAPK kinase, to the ERK signaling pathway, which is critical for osteoclastogenesis (Fig. 5, C and D).

These results indicate that if osteoclast precursors in the peritoneal cavity encountered TLR ligands, they lost the potential to differentiate into the osteoclast lineage in <2 h, and osteoclast precursors in the BM became sensitive to CpG within 2 days after stimulation with M-CSF and RANKL.

To analyze the in vivo effects of TLR ligands on osteoclastogenesis, LPS (0.5 μg, ∼20 ng LPS/g of body weight) or CpG (25 μmol, ∼1 μmol CpG/g of body weight) was injected i.v. into C57BL/6 mice, and 2 days later, BM cells and PECs were prepared. The number of cells recovered from LPS-treated (5.6 × 10⁶/femur) and CpG-treated (14.0 × 10⁶/femur) BM was ∼60 and 150%, respectively, of that from PBS-treated control mice (9.2 × 10⁶ cells/femur). Flow cytometric analysis showed a preferential reduction in the number of B220-positive B lineage cells in LPS-treated BM, resulting in an increase in the percentage of Mac-1-positive myeloid lineage cells. The number of TRAP⁺ MNCs obtained from BM cells of LPS- and CpG-treated mice was increased 7.0- and 1.5-fold, respectively (Fig. 6,A). The increase was observed for 4 days and the number of TRAP⁺ MNCs returned to within the normal range after 6 days (Fig. 6,B). Two days after i.v. injection of LPS or CpG, the number of cells recovered from the peritoneal cavities of LPS- (2.2 × 10⁶/mouse) and CpG- (2.9 × 10⁶/mouse) treated C57BL/6 mice were ∼73 and 97%, respectively, of those from PBS-treated C57BL/6 control mice (3.0 × 10⁶/mouse). The number of TRAP⁺ MNCs obtained from PECs of LPS- and CpG-treated mice was decreased to one-third of the control number (Fig. 6,A). The number of cells in the peritoneal cavity recovered to normal within 6 days (6.1 × 10⁶/no-treated mouse, and 4.7 × 10⁶, 4.1 × 10⁶, and 5.6 × 10⁶/mouse LPS-treated 2, 4, and 6 days before, respectively); however, the decrease of osteoclast precursors did not recover within 6 days (Fig. 6 B).

We also performed LPS-pulsing experiments in vivo. Mice were injected with 20 ng of LPS i.p., and 2 h later, PECs were harvested, washed thoroughly, and cultured in the presence of M-CSF and RANKL for 6 days. The number of TRAP⁺ MNCs was significantly reduced (PBS-injected control: 90.0 ± 34.6/well; LPS-treated: 1.3 ± 1.2/well). This further supported the idea that osteoclast precursors in the peritoneal cavity lost their potential to differentiate upon LPS treatment. Intraperitoneal stimulation was reported to induce monocyte exudation from blood vessels within 16 h (33). Because the experiments described above were performed at 2 h after LPS injection, the recovered cells might have been resident cells in the peritoneal cavity. These results suggest that osteoclast precursors in the peritoneal cavity might not be simply supplied from the BM through the peripheral blood.

Recently, TNF-α was reported to mimic the function of RANKL for stimulating osteoclastogenesis in vitro (18, 19). We cultured BM cells, PECs, and C7-TY cells (Fig. 7) with mouse TNF-α and M-CSF for 6 days. Comparable numbers of TRAP⁺ MNCs were generated from BM cells treated with 50 ng/ml TNF-α or RANKL in the presence of M-CSF (Fig. 7,A). In contrast, few TRAP⁺ MNCs were induced from PECs by M-CSF + TNF-α (Fig. 7,B), and M-CSF plus RANKL-induced osteoclastogenesis was inhibited by the addition of TNF-α as well as LPS (Fig. 7,B). Osteoclast development from C7-TY cells was less efficiently induced by M-CSF + TNF-α than by M-CSF + RANKL, and M-CSF + RANKL-induced osteoclastogenesis was inhibited by TNF-α (Fig. 7 C). These results suggest that TNF-α may function like RANKL in stimulating the differentiation of osteoclast precursors in the BM, whereas TNF-α functions like a TLR ligand in the peritoneal cavity and on C7-TY cells.

Because LPS is known to induce TNF-α production (34), the inhibitory effect of LPS on osteoclastogenesis of PECs and C7-TY cells may have resulted from TNF-α induction. In fact, we observed that BM cells, PECs, and C7-TY cells cultured with M-CSF showed increased transcription of the Tnf gene following the addition of LPS. Therefore, to examine the possible role of TNF-α, an anti-TNF-α neutralizing Ab (XT3) was added to PEC cultures in which osteoclast induction with M-CSF and RANKL was inhibited by LPS. To assess the effect and specificity of the XT3 Ab, BM cells were cultured with 50 ng/ml mouse TNF-α, RANKL, or LPS in the presence of M-CSF. Osteoclastogenesis induced by TNF-α and M-CSF could be blocked by 5 μg/ml XT3 Ab, but not 200 ng/ml OPG (Fig. 8,A). Conversely, the osteoclast development induced by RANKL and M-CSF could be inhibited by OPG, but not XT3 Ab (Fig. 8,A). Under these conditions, the inhibitory effect of TNF-α, but not LPS on RANKL + M-CSF-induced osteoclastogenesis from PECs was blocked by XT3 Ab (Fig. 8 B). The induction of TNF-α production by LPS might have been insufficient in our culture system, because >2 ng/ml mouse TNF-α were needed to inhibit osteoclastogenesis of PECs or to induce that of BM cells.

Functional M-CSF-deficient Csf1^op/Csf1^op mutant mice lack mature osteoclasts and develop osteopetrosis (5). Extramedullary hemopoiesis in the spleens and livers of these mice compensates for the loss of hemopoiesis in the BM (35, 36), because the hemopoietic space in bones of Csf1^op/Csf1^op BM is insufficient. It seems that the hemopoietic microenvironments of the spleen may be similar to that of normal BM. Therefore, the responsiveness of osteoclast precursors to TLR ligands in Csf1^op/Csf1^op spleen was assessed. Because osteopetrosis in Csf1^op/Csf1^op mice can be cured spontaneously by several months of age, we used 4-wk-old mice which displayed severe osteopetrosis as well as extramedullary hemopoiesis in the spleen and liver.

We harvested hemopoietic cells from the rudimentary osteopetrotic BM, and compared the number of osteoclasts induced in the culture with that induced in cultures from wild-type littermates. In the presence of M-CSF and RANKL, Csf1^op/Csf1^op BM cells generated a slightly higher number of TRAP⁺ MNCs than BM cells from the wild-type littermates, and CpG but not LPS, inhibited osteoclastogenesis of the mutant BM cells as well as that of BM cells from the normal littermates (Fig. 9). Thus, the responsiveness of osteoclast precursors in Csf1^op/Csf1^op BM to TLR ligands was comparable to that in normal mice.

The number of PECs recovered from Csf1^op/Csf1^op mice was severely reduced to <7% of that from normal littermates, and in some mice, <10⁴ cells were recovered. However, Csf1^op/Csf1^op PECs generated mature osteoclasts normally, and this osteoclastogenesis was inhibited by CpG and LPS similarly to that of wild-type mice (Fig. 9). Although Csf1^op/Csf1^op PECs had not been exposed to M-CSF, their responsiveness to TLR ligands was similar to that of normal PECs rather than that of BM cells.

In the presence of M-CSF and RANKL, Csf1^op/Csf1^op spleen cells produced 11 times higher numbers of osteoclasts compared with spleen cells of the littermates (Fig. 9), suggesting that compensatory extramedullary hemopoiesis occurred actively in their spleens. The combination of M-CSF and TNF-α also induced osteoclastogenesis of both Csf1^op/Csf1^op and control spleen cells comparable to M-CSF and RANKL like that of the BM cells. However, M-CSF + RANKL-induced osteoclastogenesis of both strains of mice was inhibited by LPS and CpG. This implies that the osteoclast precursors in the BM and spleen were not identical, even in the osteopetrotic mice (Fig. 9).

To assess whether osteoclast precursors from BM cells become sensitive to LPS, we precultured BM cells with M-CSF for 2, 4, 6, or 8 days before the induction of osteoclast development. As shown in Fig. 10, experiments using dish-adherent BM cells precultured with M-CSF showed that LPS inhibited osteoclastogenesis, and the responsiveness to TNF-α was decreased. After 4 days of culturing with M-CSF, a majority of living cells adhered to dishes, and ∼90% of adherent cells expressed Mac-1 and Fms, like peritoneal resident macrophages.

Osteoclastogenesis from 4, 6, and 8 day-precultured BM cells following incubation with M-CSF + RANKL was significantly inhibited by TNF-α, as was osteoclastogenesis from PECs and C7-TY cells (Fig. 10,B). It was noted that these precultured BM cells differentiated into TRAP⁺ MNCs within 2 days after the addition of M-CSF and RANKL (Fig. 10 A). These results suggest that immature osteoclast precursors in the BM proliferated and differentiated into more mature osteoclast precursors following culturing with M-CSF, and became sensitive to LPS. These findings also indicate that osteoclast precursors, like cells precultured with M-CSF, scarcely contributed to the osteoclastogenesis from normal BM cells, because the continuous addition of LPS did not reduce the number of osteoclasts.

Based on the responsiveness to TLR ligands and TNF-α, we showed that cells having the potential to differentiate into osteoclasts in the BM, spleen, and peritoneal cavity were distinct. BM cells precultured with M-CSF had similar but not identical characteristics to splenic or peritoneal osteoclast precursors.

Previously, we have reported that osteoclast precursors in the BM and peritoneal cavity display different phenotypic characteristics (4). In the BM, osteoclast precursors were enriched for Kit⁺Fms⁻ immature cells. In contrast, peritoneal cavity osteoclast precursors were Fms⁺Mac-1⁺. Recent work showed that FACS-sorted Kit⁺Fms⁻ cells were precursors of Kit⁺Fms⁺ cells, and the Kit⁺Fms⁺ cells differentiated into osteoclasts more efficiently and during a shorter period than Kit⁺Fms⁻ cells in the BM (37). It could be suggested that Kit, Fms, and RANK are expressed sequentially on osteoclast precursors, and cells at each differentiation stage might be present in the BM. However, a majority of osteoclasts are derived only from LPS-resistant precursors, because the number of TRAP⁺ MNCs in culture did not differ with or without LPS addition. Therefore, it is uncertain whether osteoclast precursors may differentiate and mature in the suggested order through this pathway.

We previously demonstrated that although cells with the potential to differentiate into osteoclasts were present in M-CSF- or GM-CSF-elicited colonies, the reduction in numbers of colony-forming cells treated by an anti-Kit antagonistic Ab scarcely influenced the number of osteoclasts in culture (38). Moreover, a study using Gata2-knockout embryonic stem cells suggested that the frequency of the precursors differentiating into osteoclasts in culture was reduced to one-twentieth of that of normal control embryonic stem cells, but that CFU-M was only reduced by 50% (39). These results support the postulate that most osteoclasts might be derived from cells other than CFU-M and CFU-GM, although these cells have the potential to differentiate into osteoclasts (38).

M-CSF and RANKL-induced osteoclastogenesis of BM cells from nude mice lacking T cells, or B220⁺ cell-depleted BM cells by using magnetic cell sorting was also resistant to LPS (data not shown), as well as that from stromal cell-depleted BM cells (Fig. 4 B). These results suggest that not only the microenvironment for osteoclastogenesis, but also the osteoclast precursor itself is not identical, rather than cells that rescue the inhibition by LPS are present in the BM.

BM cells cultured with M-CSF showed increased sensitivity to LPS, as did spleen cells and PECs (Fig. 10). They contained osteoclast precursors that could differentiate into mature osteoclasts within 2 days in the presence of M-CSF and RANKL. Freshly prepared BM cells, spleen cells, PECs, and even C7-TY cells needed >2 days to generate osteoclasts. Therefore, BM cells precultured with M-CSF might not be identical to splenic or peritoneal osteoclast precursors. Furthermore, no TRAP⁺ cells were generated from freshly prepared BM cells within 2 days (Figs. 1 and 10), implying that cells equivalent to the precultured cells are rarely present in the BM.

RT-PCR using cDNAs prepared from freshly prepared PECs and BM cells, and BM cells precultured with M-CSF, M-CSF + LPS, M-CSF + RANKL, and M-CSF + RANKL + LPS for 4 days indicated that all of the cultures contained Tlr2, Tlr4, Tlr9, and Md2-expressing cells (our unpublished observation). Osteoclastogenesis was first reported to be regulated by RANKL/RANK, and subsequently by TNFR-associated factor 6, Fos, ERK, Janus kinase, p38, NF-κB, NFATc1, and IFN-β (40, 41, 42, 43, 44). Kobayashi et al. (45) reported that IL-1R-associated kinase (IRAK), in association with myeloid differentiation factor 88, and IRAK-M, encoded by Irak3 regulate the signaling via TLRs. We assessed the expression of Irak3 in PECs, freshly prepared BM cells, and BM cells cultured with M-CSF, M-CSF + RANKL, M-CSF + LPS, and M-CSF + RANKL + LPS. Irak3 gene expression was increased in the presence of LPS as reported, but not in the presence of M-CSF + RANKL (data not shown). Thus, the mechanism of LPS-induced low responsiveness (LPS tolerance) via IRAK-M might not be mediated by RANK/RANKL signaling (45, 46). Recently, we demonstrated that Notch and Wnt signaling regulates osteoclastogenesis by directly affecting precursors and through the supporting microenvironment (23, 47). Therefore, other signaling pathways should be considered.

In vivo LPS injection increased BM cell osteoclastogenesis, whereas it decreased PEC osteoclastogenesis. LPS might influence BM osteoclast precursors not only directly but also via stromal cells or other hemopoietic lineage cells (48, 49). In fact, osteoclastogenesis supported by ST2 stromal cells was accelerated by the addition of LPS in culture (Fig. 4). Following LPS injection, the recovery of osteoclast precursors in the mouse peritoneal cavity to within the normal range requires a long time. Extension of the experiment to 28 days after LPS injection showed that the decrease of osteoclast precursors did not return to within the normal range during this period (our unpublished observation). Resident macrophages in the peritoneal cavity are replenished and differentiate there (50), and even following activation with thioglycolate, macrophages might proliferate in situ, although granulocytes are immediately released from blood vessels. Peritoneal osteoclast precursors might be supplied from peritoneal replenishing cells.

In some experiments, no or low levels of osteoclastogenesis were observed from PECs of nontreated wild-type mice kept under conventional conditions. These low responder mice might have received some stimuli in the past, or might harbor natural ligands for TLRs (51). Moreover, almost all lots of FBS supported the osteoclast development of BM cells, but we needed to check lots of FBS for the ability to support that of PECs. It seemed that some lots of FBS contained endotoxins or some TLR ligands.

It was reported that macrophages from BM cells cultured with M-CSF for 7 days underwent apoptosis within 3 h of exposure to 100 ng/ml LPS (52), while 1 μg/ml LPS was not cytotoxic for thioglycolate-elicited peritoneal macrophages (53). We did not observe significant cell death following treatment with LPS in our culture conditions. Thus, it is likely that when osteoclast precursors in the spleen and peritoneal cavity receive differentiation signals and are stimulated by TLR ligands at the same time, they may lose the potential to differentiate into osteoclasts.

Osteoclast precursors in the peritoneal cavity express Fms and Mac-1 (4). A significant decrease in numbers of cells expressing Fms was observed in PECs from mice treated with LPS for 2 days compared with PECs from mice treated with PBS (PBS-injected control: 33.3% vs LPS-treated: 1.3%) (54). In contrast, 7 days after LPS injection, Fms and Mac-1 expression on PECs was increased, and thereafter returned to normal on day 28. Therefore, there was no obvious relationship between Fms expression on the surface of PECs and the ability to differentiate into osteoclasts.

Interestingly, splenic osteoclast precursors even in osteopetrotic mice are distinct from BM precursors. Immature cells, including hemopoietic stem cells, flow in blood vessels and their numbers are increased in osteopetrotic mice (35). However, the phenotypes of splenic osteoclast precursors from osteopetrotic mice were similar to the precursors from normal spleen, but the BM osteoclast precursors expressed different phenotypes. The difference between the peritoneal and BM osteoclast precursors might be due to differences of their derivation, but differences between these precursors and splenic precursors might result from differences of the microenvironment for their maintenance.

Based on these results, we propose the following model of the cells maintained in each tissue with the potential to differentiate into osteoclasts (Fig. 11). Development from osteoclast precursors in the BM is inhibited by only CpG, but not LPS or PGN, and is induced by TNF-α as well as RANKL in the presence of M-CSF (type I). Splenic osteoclast precursors differentiate into mature osteoclasts in response to TNF-α as well as RANKL. Splenic osteoclastogenesis is inhibited by LPS and PGN (type II). M-CSF and RANKL-induced osteoclastogenesis from peritoneal precursors is inhibited by TNF-α (type III). The characteristics of BM cells precultured with M-CSF seem to be of an intermediate type between types II and III. Except in the case of C7-TY cells, we have not been able to stimulate osteoclastogenesis without M-CSF in cultures, even using BM cells precultured with M-CSF. The function of M-CSF in osteoclastogenesis is understood to involve supporting cell survival, because Csf1^op/Csf1^op mice carrying the Bcl2 transgene produce mature osteoclasts (55). Because C7-TY cells are derived from p53^−/− BM cells, osteoclastogenesis might be induced in these cells even in the absence of M-CSF.

It is still not clear whether osteoclastogenesis from fetal liver and newborn spleen cells is resistant to LPS like that from the BM (Fig. 1), although bone tissues and osteoclasts are detected before embryonic day 15 (56). Bony fishes and aquatic amphibians lack BM, and therefore their hemopoiesis occurs totally in extramedullary organs. However, they have osteoclasts that are involved in trimming the bone surface and the eruption of teeth (57). These findings suggest that there is a transition of osteoclast precursors from fetal type to adult type, or possibly that some alternative regulation of osteoclast development might occur depending on the cell type (58).

Recently, interesting studies demonstrated the possibility that B cell and osteoclast lineages shared their precursors, and osteoclastogenesis was regulated by stimuli for induction of B lineage cells (59, 60, 61, 62). Because it was reported that osteoclasts regulated the B lymphopoiesis (36), the relationship of B cell and osteoclast lineages should be considered.

The ability of several reagents such as TLR ligands, TNF-α, and IL-1 to support osteoclastogenesis was not always consistent in several studies (13, 19, 30, 48, 49). Osteoclast development has been studied by using freshly prepared BM cells, BM cells precultured with M-CSF, spleen cells, and cloned cell lines as a source of osteoclast precursors. However, as shown in this study, these precursor cells are not all identical. More controlled studies are needed to examine the mechanism of osteoclast development.

We thank Drs. K. Miyake (University of Tokyo, Tokyo, Japan), H. Hemmi, S. Ono (Osaka University, Osaka, Japan), S. Niida (National Institute for Longevity Science, Aichi, Japan), and M. Yoshino (Tottori University, Tottori, Japan) for helpful suggestions, T. Taki and K. Yamanishi (Otsuka Pharmaceutical, Tokyo, Japan) for M-CSF, and S. Nishikawa (Riken Center for Developmental Biology, Hyogo, Japan) for Abs. We also acknowledge Dr. T. Kurosaki (Kansai Medical University, Osaka, Japan) for his warm encouragement, Dr. T. Shibahara and N. Kubo for the maintenance of mice, and T. Shinohara for her secretarial assistance.