CpG oligodeoxynucleotides modulate the osteoclastogenic activity of osteoblasts via Toll-like receptor 9.

Regulation of osteoclastogenesis by lipopolysaccharide (LPS) is mediated via its interactions with toll-like receptor 4 (TLR4) on both osteoclast- and osteoblast-lineage cells. We have recently demonstrated that CpG oligodeoxynucleotides (CpG ODNs), known to mimic bacterial DNA, modulate osteoclastogenesis via interactions with osteoclast precursors. In the present study we characterize the interactions of CpG ODNs with osteoblasts, in comparison with LPS. We find that, similar to LPS, CpG ODNs modulate osteoclastogenesis in bone marrow cell/osteoblast co-cultures, although in a somewhat different pattern. Osteoblasts express receptors for both LPS and CpG ODN (TLR4 and TLR9, respectively). The osteoblastic TLR9 transmits signals into the cell as demonstrated by NFkappaB activation as well as by extracellular-regulated kinase (ERK) and p38 phosphorylation. Similar to LPS, CpG ODN increases in osteoblasts the expression of tumor necrosis factor (TNF)-alpha and macrophage-colony stimulating factor (M-CSF). The two TLR ligands do not affect osteoprotegerin expression in osteoblasts. CpG ODN does not significantly affect receptor activator of NFkappaB ligand (RANKL) expression, in contrast to LPS, which induces the expression of this molecule. In the co-cultures CpG ODN induces RANKL expression in osteoblasts as a result of the more efficient TNF-alpha induction. CpG ODN activity (modulation of osteoclastogenesis, gene expression, ERK and p38 phosphorylation, and nuclear translocation of NFkappaB) is specific, because the control oligodeoxynucleotide, not containing CpG, is inactive. Furthermore, these effects (unlike the LPS effects) are inhibited by chloroquine, suggesting a requirement for endosomal maturation/acidification, the classic CpG ODN mode of action. We conclude that CpG ODN, upon TLR9 ligation, induces osteoblasts osteoclastogenic activity.

In addition to LPS, a variety of bacterial products such as teichoic acid and other cell wall components have been shown to stimulate osteoclastic bone resorption (11). These pathogenderived molecules are known to induce innate immunity via toll-like receptors (TLRs) (13)(14)(15)(16)(17)(18). Recent advances demonstrate that bacterial DNA is also a pathogen-derived molecule activating the innate immune system (19 -22). This activity of bacterial DNA depends on its content of unmethylated CpG dinucleotides in particular base contexts ("CpG motif ") (20 -25). Vertebrate DNA contains a lower than expected amount of CpG dinucleotides, and these are highly methylated, which prevents their immune stimulatory effects. Oligodeoxynucleotides containing CpG motifs, CpG oligodeoxynucleotides (CpG ODNs), mimic the activity of bacterial DNA. These, together with the ability of CpG ODN to activate NFB (26 -28), a critical transcription factor in osteoclast differentiation (29,30), prompted us to examine modulation of osteoclastogenesis by the oligodeoxynucleotides. We demonstrated that CpG ODNs exert dual effect on osteoclastogenesis: they inhibit RANKL-induced osteoclastogenesis of osteoclast precursors not exposed to RANKL prior to CpG ODN addition. On the other hand, the oligodeoxynucleotides are potent osteoclastogenic agents to osteoclast precursors pretreated with RANKL (31).
The present study was designed to examine if osteoblasts are targets to CpG ODNs, and thus these cells mediate, in part, the modulation of osteoclastogenesis by the ODN. We find that CpG ODN interacts with osteoblastic TLR9 and elicits intracellular events leading to the increased expression of molecules regulating osteoclastogenesis.

EXPERIMENTAL PROCEDURES
Mice-Newborn BALB/c mice and 7-to 9-week-old male BALB/c mice were obtained from Harlan Laboratories Ltd. (Jerusalem, Israel).
Bone Marrow Macrophages-Primary BMMs were isolated as described (36) and incubated for 3 days prior to the experiment.
In Vitro Osteoclast Formation Assay-Osteoclasts were generated using the mouse BMM/osteoblast co-culture system in the presence of 1,25(OH) 2 D 3 (10 nM) and dexamethasone (100 nM), as described previously (37). Tartrate-resistant acid phosphatase (TRAP)-positive cells containing more than two nuclei were counted on day 7, following removal of the osteoblasts by collagenase.
Methylene Blue Staining-The relative cell number was estimated by the methylene blue staining assay using a Dynatech plate reader (Vienna, VA) (38).
Western Blot Analysis-Western analysis was performed as described previously (37). Bands were quantified by densitometry.
Northern Blot Analysis-Osteoblasts or BMM/osteoblast co-cultures were grown in ␣-MEM with 10% fetal calf serum (39). Total cellular RNA was extracted using TRI reagent, fractionated by electrophoresis on 1.2% agarose formaldehyde gels (10 g/lane), and transferred to nylon membranes (Hybond-N, Amersham Biosciences, Little Chalfont, UK). 32 P-Labeled mouse TNF-␣, macrophage colony stimulating factor (M-CSF), M-CSF receptor, IL-1␤, RANKL, OPG, rat alkaline phosphatase, or mouse ribosomal protein L32 cDNA probes were used for hybridization. The hybridized membranes were then subjected to autoradiography, and the density of each of the mRNA bands was quantified.  (40). Nuclear extracts (5 g) were incubated with the labeled probe in 20 l of reaction buffer (10 mM Tris, pH 7.9, 20 mM KCl, 1 mM EDTA, 1 mM dithiothreitol, 1 g of poly(dI-dC), and 4% glycerol) for 20 min at room temperature. Where effects of antibodies were examined, 1 g of the corresponding antibody was added to nuclear extracts 20 min before addition of DNA probe. Samples were then fractionated on a 7% polyacrylamide gel and visualized by exposing dried gel to film.

CpG ODN Modulates Osteoclastogenesis in BMM/Osteoblast
Co-cultures-We have previously shown that CpG ODNs and LPS (31, 37) exert dual effect on osteoclast differentiation. They inhibit RANKL-induced osteoclastogenesis in BMMs not exposed previously to RANKL but strongly stimulate osteoclastogenesis in RANKL-primed BMMs. These studies were performed in the absence of osteoblasts/stromal cells essential for in vivo osteoclastogenesis. In the present study we used BMM/ osteoblast co-cultures to mimic the in vivo interactions (41).
The inclusion of ODN 1826 from the beginning of the cocultures resulted in blocking (almost 100%) the osteoclastogenesis (Fig. 1A); when ODN 1826 was added 24 h after the beginning of the culture ϳ50% inhibition was observed. The addition of the ODN at later time points (days 5 and 6) caused a marked enhancement of the osteoclastogenesis. The effect was specific, because the control oligodeoxynucleotide, not containing the CpG motif (ODN 1982), was inactive. In contrast, LPS stimulated osteoclastogenesis even when it was present from the beginning of the co-culture (Fig. 1B), unlike the LPS effect in BMMs in the absence of osteoblasts (37). The inclusion of either ODN 1826 or LPS resulted in increased cellular contents of the monolayer (Fig. 1, C and D). ODN 1826 exerted its inhibitory ( Fig. 2A) and stimulatory (Fig. 2B) effects on osteoclastogenesis already at 20 nM.
In Fig. 3A we see that osteoprotegerin (OPG) inhibited the stimulation of osteoclastogenesis caused by LPS when presents throughout the experiment (7 days). OPG also inhibited the basal activity (in the absence of LPS). When LPS or ODN 1826 (Fig. 3B) were added for the last 24 h of the 7-day experiment, OPG inhibited the activity of both. Chloroquine inhibited the activity of ODN 1826, confirming the classic mode of the ODN action, involving endosomal maturation and/or acidification. As expected, chloroquine did not affect the activity of LPS.
M-CSF and RANKL interactions with their respective receptors on osteoclast precursors are essential for the differentiation of these cells. Therefore, we examined the hypothesis that the inhibitory effect of CpG ODN observed in the co-cultures is caused by reduction in the levels of these receptors. To this end, co-cultures were incubated in the presence or absence of ODN 1826 for 4 -6 days. The osteoblasts were removed by collagenase (see Fig. 10A below), and RNA and protein were prepared from the remaining adherent cells. Northern analysis revealed a significant reduction (ϳ67%) in M-CSF receptor mRNA abundance (Fig. 4A). A slight non-significant reduction was observed in the abundance of RANK transcript levels (not shown). Western analysis (Fig. 4B) showed that M-CSF receptor protein level was also markedly reduced (ϳ72%) in ODN 1826-treated cells. These findings are in agreement with our studies on osteoclasts precursors in the absence of osteoblasts (31).
CpG ODN Induces NFB Activation and ERK and p38 Phosphorylation in Osteoblasts-Because effects of LPS and CpG ODNs are mediated via TLR4 and TLR9, respectively, we examined the expression of these receptors in osteoblasts using reverse transcriptase-PCR (Fig. 5). TLR4 is present in comparable levels in osteoclast precursors and in osteoblasts. In contrast, TLR9 expression is lower in osteoblasts as compared with osteoclast precursors. The lack of expression in the absence of reverse transcriptase serves as a control.
LPS and CpG ODNs are known to induce activation of NFB in the target cells. In Fig. 6A we demonstrate (using EMSA) that the two modulators induce NFB activation in osteoblasts in a time-dependent manner. Controls (NE, absence of nuclear extract; cold, excess of unlabeled nucleotides) are shown. In Fig. 6B we see that anti-p65 and anti-p50 antibodies caused a supershift and a block-shift, respectively, in osteoblasts treated with TLR ligands, indicating that these members of the NFB family are mobilized into the nucleus. On the other hand, no effects were observed with anti-p52, anti-Rel B, or anti-c-rel antibodies. In Fig. 6C it is shown that NFB nuclear association induced by either LPS or ODN 1826 is not affected by OPG. Chloroquine inhibits the effect of ODN 1826 but not the LPS effect. The control oligodeoxynucleotide, ODN 1982, did not induce NFB nuclear association.
The signaling pathways mediating the cellular effects of CpG ODN were studied (42,43). To demonstrate modulation of TLR9 signaling pathway in osteoblasts by CpG ODN we chose to measure ERK (p42/44) and p38 phosphorylation. In Fig. 7 we show that indeed p38 and, to a lesser extent, ERK are phosphorylated in response to ODN 1826 but not in response to ODN 1982. Consistent with the low TLR9 expression, as compared with BMMs, ERK and p38 phosphorylation occurs to a smaller degree in osteoblasts (Fig. 7, compare A to B).
CpG ODN Modulates Osteoblast Gene Expression-We next studied how the TLR ligands affect osteoblastic genes known to participate in regulation of osteoclastogenesis. Both ODN 1826 and LPS did not affect significantly the expression of the osteoclastogenesis inhibitor OPG (Fig. 8A). LPS caused a marked increase in expression of RANKL, TNF-␣, and M-CSF in osteoblasts. The increase in TNF-␣ and M-CSF expression by ODN 1826 was moderate, and no significant effect was exerted by the ODN on RANKL expression (about 50% increase in transcript abundance, as compared with 1500% increase with LPS). The CpG ODN induction of TNF-␣, but not the LPS induction, was blocked completely by chloroquine (Fig. 8B). We wondered if the fact that RANKL induction is observed with LPS, but not with ODN 1826, could be responsible to the differential effect of the two modulators on TNF-␣ expression. The failure of OPG to modulate the effect of either LPS or ODN 1826 (Fig. 8B) rules out this possibility. The inability of ODN 1826 to induce RANKL expression in osteoblasts does not correlate with the ability of OPG to inhibit the osteoclastogenic effect of the oligodeoxynucleotide in BMM/ osteoblast co-cultures (Fig. 3). We examined, therefore, the effect of ODN 1826 on gene expression in the co-culture and found that the ODN is much more efficient in increasing the expression of IL-1␤ and TNF-␣ in BMM/osteoblast co-cultures than in osteoblasts in the absence of BMMs (Fig. 9A). Furthermore, in the co-cultures a significant increase in RANKL expression by ODN 1826 is also observed. In the co-culture, similarly to what we observed with osteoblasts, chloroquine selectively inhibited the effect of ODN 1826, whereas OPG did not change the effects of both CpG ODN and LPS (Fig. 9B). To examine which of the cells in the co-culture produce TNF-␣ and RANKL, we treated the BMM/osteoblast co-cultures at the end of the experiment with collagenase (Fig. 10) and analyzed separately the cells that were removed by the collagenase and the cells that remained adherent. In Fig. 10A we see that the cells removed by collagenase are enriched for alkaline phospha-tase (an osteoblastic marker), whereas the remaining adherent cells are enriched for the M-CSF receptor (a marker for macrophages and osteoclast-lineage cells). Consistent with results presented in Fig. 9, we see that ODN 1826 increases TNF-␣ more efficiently in the co-culture than in osteoblasts alone (Fig.  10B). Furthermore, an increase in RANKL expression in response to the ODN is observed in the co-culture, but very little in osteoblasts. The collagenase treatment revealed that most of the TNF-␣ expression is derived from the osteoclast precursorsenriched fraction, whereas most of the RANKL expression is derived from the osteoblast-enriched population.
Next we attempted to understand the mechanism of RANKL modulation by the ODN. We found that TNF-␣ and IL-1␤ increase the expression of RANKL in osteoblasts (not shown), confirming previous studies (44,45). CpG ODNs increase the expression of these two cytokines in osteoclast precursors and, to a lesser extent, in osteoblasts; therefore, we hypothesize that (a) the ODN increases osteoblastic RANKL expression indirectly via TNF-␣ and/or IL-1␤ and (b) because the effect on the cytokines levels in co-cultures is stronger than in osteoblasts alone, the efficiency of ODN in inducing RANKL is more pro- and BMMs/osteoblasts were cultured for 6 days. Some of the BMMs/ osteoblasts were treated with collagenase. The adherent and non-adherent cell populations were collected separately prior to RNA isolation. Alkaline phosphatase and M-CSF receptor mRNA levels were examined by Northern analyses in osteoblasts, in BMMs/osteoblasts co-cultures, and in the two sub-populations. B, BMMs/osteoblasts were cultured for 6 days. Cells were then washed and treated with LPS (20 ng/ml) or ODN 1826 (100 nM) for 4 h. At the end of the experiment, cells were treated with collagenase as in A. Abundance of TNF-␣ and RANKL transcripts was measured in osteoblasts, in BMMs/osteoblasts, and in the two subpopulations.

FIG. 11. Anti-TNF-␣ antibody inhibits ODN 1826-induced RANKL expression and osteoclast differentiation in BMM/osteoblast co-cultures.
A, BMM/osteoblast co-cultures were grown for 6 days. Cells were then washed and treated with ODN 1826 (100 nM) for 4 h, in the presence or absence of anti-TNF-␣, IL-1ra, anti-TNF-␣ plus IL-1ra or IgG (20 g/ml each). Transcript abundance of RANKL was measured using Northern analysis. L32 was used as a loading control. B, BMM/osteoblast co-cultures were grown for 6 days. Then ODN 1826 was added for 24 h in the presence or absence of either anti-TNF-␣ (20 g/ml) or IgG (20 g/ml). Osteoclast formation was measured. nounced in the former conditions. To test these hypotheses we have examined the ability of anti-TNF-␣ neutralizing antibody and of IL-1ra to inhibit the induction of RANKL by CpG ODN in the co-culture. Using Northern analysis we show in Fig. 11A that anti-TNF-␣ neutralizing antibody blocks the induction of RANKL by ODN 1826, whereas control IgG was not effective. IL-1ra did not affect RANKL induction by ODN 1826 in the absence or presence of anti-TNF-␣-neutralizing antibody. The inhibition of RANKL induction by ODN 1826 results in inhibition of the ODN ability to stimulate osteoclastogenesis (Fig.  11B).
Importance of NFB in CpG ODN Induction of Osteoclastogenesis-Finally we have tested if NFB activation in osteoblasts plays a role the osteoclastogenic activity of these cells. To this end we have examined the effects of gliotoxin, a known inhibitor of NFB activation (46), on ODN 1826 induction of NFB activation, TNF-␣, and RANKL expression and osteoclastogenesis. In Fig. 12 we see that gliotoxin inhibits ODN 1826 stimulatory activities on osteoblasts: NFB activation (A), TNF-␣ (B and C) and RANKL (C) expression, as well as osteoclastogenesis (D).

DISCUSSION
Normal bone resorption is maintained by systemic hormones and local growth factors and cytokines that regulate the differentiation and activation of osteoclasts (1,8,(47)(48)(49)(50). Accelerated osteoclastogenesis is the cause of pathological bone loss induced by bacterial products in a variety of diseases, including periodontitis, osteomyelitis, bacterial arthritis, and infected metal implants (11,(51)(52)(53)(54). Most data on involvement of bacterial products in osteoclastogenesis and resorption were obtained with LPS. For example, it has been identified as an important factor in the pathogenesis of periodontitis, characterized by gingival inflammation and alveolar bone resorption (55). LPS stimulates osteoclastic bone resorption in vivo (56,57) and in vitro in organ culture (58,59) and promotes osteoclast differentiation in whole bone marrow cell culture (60). LPS induces RANKL expression in osteoblasts (12) and stimulates these cells to secrete IL-1, prostaglandin E 2 , and TNF-␣, each of which seems to be involved in LPS-mediated bone resorption (11). We showed that CpG ODNs, known to be responsible for the ability of bacterial DNA to elicit innate immunity (19 -25), modulate osteoclastogenesis via interactions of the ODN with osteoclast lineage cells (37). These findings were confirmed recently (61).
LPS and CpG ODNs exert their activities via interactions with TLR4 and TLR9 on the target cells, respectively (16,18,62,63). It is recognized that modulation of osteoclastogenesis by LPS occurs directly, via interactions with TLR4 on osteoclast lineage cells, as well as indirectly via interactions with TLR4 on osteoblasts (12,64).
We examined in the present study the hypothesis that CpG ODNs, in addition to their interactions with osteoclast lineage cells, are also capable of modulating the osteoclastogenic activity of osteoblasts. We demonstrate here that osteoblasts express TLR9, the receptors for CpG ODNs. Furthermore, the interaction of the oligonucleotides with osteoblasts elicits phosphorylation of ERK and p38, activation of NFB, and modulation of genes participating in the regulation of osteoclastogenesis. Gliotoxin blocks the activation of NFB by ODN 1826, the induction of TNF-␣ and RANKL, and osteoclast differentiation. In fact, also the basal osteoclastogenesis (in the absence of ODN 1826) is inhibited, consistent with the role of NFB, TNF-␣, and RANKL in basal osteoclastogenesis. Methylene blue uptake was not affected (not shown), ruling out the possibility that the reduction in osteoclast number is caused by toxicity of the gliotoxin. Modulation of TNF-␣, M-CSF, and IL-1 expression in osteoblasts by CpG ODN was less efficient than the activity exerted by LPS. Moreover, CpG ODN did not markedly affect RANKL expression in osteoblasts, in contrast to LPS. It is of note that TLR9 abundance in osteoblasts is significantly lower than in osteoclast precursors, whereas TLR4 levels are comparable in the two cell lineages. This might be the reason that in osteoblasts the effects mediated by TLR4 are stronger than those mediated by TLR9. All effects of CpG ODN were stronger in the co-culture than in osteoblasts alone. Moreover, in the co-culture CpG ODN was able to up-regulate RANKL mRNA abundance. Using collagenase treatment of the co-cultures, we analyzed separately the osteoblasts and the osteoclast lineage cells and showed that the CpG ODN-induced increase in RANKL expression is in the osteoblasts. Using anti-TNF-␣ antibody and IL-1ra, we found that CpG ODN induces RANKL via TNF-␣ (but not IL-1␤) that was shown FIG. 12. Modulation of ODN 1826 activities by gliotoxin. A, osteoblasts were cultured for 6 days and then treated with ODN 1826 (100 nM) for 90 min, with or without gliotoxin (1 g/ml, added 30 min before the ODN). Nuclear extracts were prepared and subjected to EMSA. Osteoblasts (B) or BMM/osteoblast co-cultures (C) were grown for 6 days and then treated with ODN 1826 (100 nM) for 4 h with or without gliotoxin. RNA was then prepared and examined using Northern blot analysis for transcript abundance of TNF-␣ (B and C) or RANKL (C). L32 was used as a loading control. D, BMM/osteoblast co-cultures were grown for 6 days in the presence of 1,25(OH) 2 D 3 (10 nM) and dexamethasone (100 nM). Then ODN 1826 was added for 24 h with or without of gliotoxin. Osteoclast formation was measured.
previously to increase RANKL (44,45). In osteoblasts alone, the increase in RANKL by the ODN is moderate, due to the relatively low levels of TNF-␣. Consistent with this, when a higher density of osteoblasts is examined the level of RANKL induction by the ODN is more pronounced (not shown). Our results indicate that CpG ODN-induced RANKL expression is mediated via autocrine and paracrine mechanisms, due to TNF-␣ produced by osteoblasts and osteoclast lineage cells, respectively. Under the experimental conditions that we use, the paracrine mechanism is dominant.
Our findings show that CpG ODN interacts with osteoblasts and modulates their osteoclastogenic activity. The expression of TLR9 by osteoblasts, and the ability of chloroquine to inhibit the ODN effects indicate the involvement of TLR9 in mediating the CpG ODN interaction with osteoblasts. The comparison with the effects of LPS shows that ligation of either TLR4 or TLR9 has an impact on osteoclastogenesis via interactions with both osteoclast precursors and osteoblasts. It is of note, however, that there are differences between the activities of the TLR4 ligand, LPS, and of the TLR9 ligand, CpG ODN. Thus, the mechanisms by which TLR ligands lead to osteolysis in diseases such as periodontitis and rheumatoid arthritis could include interactions of the bacteria-derived products (LPS and CpG ODN representing bacterial DNA) with osteoblasts.
CpG ODNs interactions with bone cells result in "pro" and "anti" osteoclastogenic signals (increase in TNF-␣ and RANKL expression and reduction in osteoclastic M-CSF receptors, respectively). In vivo studies are underway to examine if the net result of the ODNs administration will be increased or decreased osteoclastogenesis.
Clinical implications of CpG ODN-mediated effects are directed toward modulation of immune functions. Several therapeutic concepts have been developed, including the use of CpG ODN as adjuvant for vaccine therapy in infectious disease and cancer (65). It is important to know whether CpG ODN therapy has an impact on bone resorption. It is possible that this therapy is associated with increased bone loss, due to increased circulating RANKL levels. This is highly relevant for CpG ODN-mediated immunotherapy in general. Furthermore, a better understanding of the interaction of CpG ODN with bone cells, osteoclasts and osteoblasts, might also potentially lead to new therapies to treat bone disease.