HMGB1 Regulates RANKL-Induced Osteoclastogenesis in a Manner Dependent on RAGE


  • The authors state that they have no conflicts of interest

  • Published online on February 25, 2008


High-mobility group box 1 (HMGB1), a nonhistone nuclear protein, is released by macrophages into the extracellular milieu consequent to cellular activation. Extracellular HMGB1 has properties of a pro-inflammatory cytokine through its interaction with receptor for advanced glycation endproducts (RAGE) and/or toll-like receptors (TLR2 and TLR4). Although HMGB1 is highly expressed in macrophages and differentiating osteoclasts, its role in osteoclastogenesis remains largely unknown. In this report, we present evidence for a function of HMGB1 in this event. HMGB1 is released from macrophages in response to RANKL stimulation and is required for RANKL-induced osteoclastogenesis in vitro and in vivo. In addition, HMGB1, like other osteoclastogenic cytokines (e.g., TNFα), enhances RANKL-induced osteoclastogenesis in vivo and in vitro at subthreshold concentrations of RANKL, which alone would be insufficient. The role of HMGB1 in osteoclastogenesis is mediated, in large part, by its interaction with RAGE, an immunoglobin domain containing family receptor that plays an important role in osteoclast terminal differentiation and activation. HMGB1-RAGE signaling seems to be important in regulating actin cytoskeleton reorganization, thereby participating in RANKL-induced and integrin-dependent osteoclastogenesis. Taken together, these observations show a novel function of HMGB1 in osteoclastogenesis and provide a new link between inflammatory mechanisms and bone resorption.


Inflammatory osteolysis, a major complication of rheumatoid arthritis, periodontal disease, and orthopedic implant destabilization, results in large part from accelerated bone resorption triggered by pro-inflammatory cytokines. Osteoclasts (OCs), multinucleated and terminally differentiated cells derived from hematopoietic monocyte/macrophage precursors, are responsible for bone resorption. OC differentiation is regulated by multiple factors, including macrophage-colony stimulating factor (M-CSF) and RANKL (also known as ODF or TRANCE).(1–5) M-CSF and its receptor, c-Fms (a receptor tyrosine kinase expressed on the cell surface), drive early lineage development from bone marrow macrophages (BMMs) and are essential for OC survival.(2) RANKL is present on the membrane of osteoblast progenitors and also exists in a soluble form in the bone microenvironment.(6) The essential role of RANKL in OC differentiation is established by the following observations. First, RANKL-deficient mice lack OCs and develop severe osteopetrosis, in addition to immunologic defects.(1–5) Second, systemic administration of RANKL increases bone resorption.(1–5) Third, in vitro, RANKL has all the attributes of a bona fide OC differentiation factor: it favors OC differentiation in conjunction with M-CSF; it bypasses the need for stromal cells and vitamin D3 to induce OC differentiation; and it activates mature OCs to resorb mineralized bone.(1–5) RANKL binding to the RANK receptor, a member of the TNFα receptor family localized to BMMs and OCs, promotes OC differentiation and activation.(1–5)

In addition to M-CSF and RANKL, several pro-inflammatory cytokines, also called osteoclastogenic cytokines, have been implicated in OC differentiation. Among these, TNFα, interleukin (IL)-1, and lipopolysaccharide (LPS) seem to contribute to OC differentiation/activation associated with inflammatory osteolysis.(7–9) Furthermore, integrin engagement by cell adhesion to the bone matrix, TREM2 (triggering receptors expressed by myeloid cells), a receptor of the immunoglobulin super family, and DC-STAMP, a putative seven transmembrane receptor, have been implicated in regulating OC terminal differentiation and activation.(10–15) The latter observations suggest the involvement of other ligands/osteoclastogenic factors in addition to M-CSF, RANKL, and TNFα. The identity of these other ligands/osteoclastogenic factors is currently unknown.

High-mobility group box 1 (HMGB1), also called amphoterin or HMG1, is a 30-kDa abundant nonhistone nuclear protein expressed in all eukaryotic cells and subject to release by activated macrophages or injured cells.(16,17) Once in the extracellular space, HMGB1 displays properties of a pro-inflammatory cytokine.(16,17) Extracellular HMGB1 has an essential role in innate immunity, as well as dendritic cell maturation and function, and these functions are mediated, at least in part, by receptor for advanced glycation endproduct (RAGE).(18,19) RAGE, a member of the immunoglobulin super family of cell surface molecules, binds to multiple ligands, in addition to HMGB1.(20,21) It is involved in the pathogenesis of multiple inflammation-associated disorders, including diabetic complications,(22) Alzheimer's disease,(21) and inflammatory conditions.(23) In addition to RAGE, TLR2/4, receptors for LPS that are essential for innate immunity, are also implicated as receptors for HMGB1.(24,25) Recently, we showed a role of RAGE in OC terminal differentiation and activation.(26) Such a function seems to be mediated by RAGE regulation of osteoclastic actin cytoskeleton organization and integrin signaling.(26) However, which RAGE ligands are involved in osteoclastogenesis remains unclear.

Herein, we present evidence for a role of extracellular HMGB1, acting as an autocrine ligand of RAGE, in RANKL-induced osteoclastogenesis in vitro and in vivo. In addition to TNFα and LPS, RANKL stimulates HMGB1 release from macrophages. Such extracellular HMGB1 seems to be needed for RANKL-induced OC differentiation. Analogous to TNFα and IL-1, extracellular HMGB1 stimulates differentiation of OC precursors in the presence of subthreshold levels of RANKL in vitro and in vivo. The latter is blocked in RAGE-deficient mice, suggesting a central role for HMGB1–RAGE interaction. Finally, we showed that extracellular HMGB1, through RAGE, regulates macrophage polarization and spreading and osteoclastic actin ring formation, suggesting a mechanism underlying its role on osteoclastogenesis. Taken together, our results are consistent with the hypothesis that extracellular HMGB1 is a physiologically/pathologically relevant osteoclastogenic cytokine, contributing to osteoclastogenesis, thereby participating in focal/inflammatory osteolysis.


Reagents and animals

Antibody specific to phospho-PYK2 (PY402) was purchased from Transduction Laboratories (Lexington, KY, USA), and antibodies to phospho-Erk1/2 (p-Erk1/2) and Erk1/2 were from Cell Signaling (Beverly, MA, USA). Monoclonal RAGE and PYK2 antibody were purchased from Chemicon (Temecula, CA, USA) and BD Transduction Laboratory (San Jose, CA, USA), respectively. The c-fos polyclonal antibody was from Santa Cruz Biotechnology (Santa Cruz, CA, USA). Polyclonal anti-HMGB1 antibody was purchased from BD Pharmingen (San Diego, CA, USA) or generated using the recombinant HMGB1 protein as antigen (Cocalico Biologicals, Reamstown, PA, USA). Antiserum to HMGB1 was purified using a protein A-based affinity column/elution buffer system according to the manufacturer's instructions (Pierce, Rockford, IL, USA) and was repurified using a polymyxin B affinity column to remove contaminating LPS. The neutralizing activity of anti-HMGB-1 antibody was confirmed in macrophage cultures exposed to recombinant HMGB-1 and assayed for TNF release.

Murine M-CSF was obtained from R&D Systems (Minneapolis, MN, USA). Nicotine was purchased from Sigma (St Louis, MO, USA). Recombinant glutathione-S-transferase (GST)-RANKL and GST-HMGB1 proteins were generated and purified as previously described.(27,28) These recombinant proteins were repurified using a polymyxin B affinity column to remove contaminating LPS. However, the LPS content in purified HMGB1 was ∼90 pg/μg of protein, measured by the chromogenic Limulus Amebocyte Lysate (LAL) assay (Cambrex). Thus, in our stimulation experiments, polymyxin B was added to cell culture medium at ∼6 units/pg of LPS, which was confirmed to be able to neutralize the maximal amount of contaminating LPS/endotoxin.

RAGE−/− mice have been crossed into a C57BL/6 genetic background, as described.(26,29) Age-matched C57BL/6 wildtype (WT) mice were used as controls. RAGE mutation was confirmed by expression of green fluorescent protein (GFP), an indicator of Cre expression in the mice,(29) RT-PCR for expression of RAGE mRNA, and Western blot analysis. TLR4−/− (C.C3-Tlr4Lps-d/J) and control mice were purchased from The Jackson Laboratory (Bar Harbor, ME, USA). All experimental procedures were approved by the Animal Subjects Committee at the Medical College of Georgia, according to U.S. NIH guidelines.

In vivo osteoclastogenesis and bone histomorphometric analysis

Three-month-old wildtype and mutant mice were subjected to daily intraperitoneal or supracalvarial administration of PBS, HMGB1 (10 μg/mouse/d), RANKL (2 μg/mouse/d) with or without anti-HMGB1 antibody (3 μg/mouse/d) for 5 days. Animals were killed, and the calvarial bones were fixed overnight in 10% buffered formalin, decalcified in 14% EDTA, embedded in paraffin, sectioned, and stained for TRACP. The percentage of osteoclastic erosion surfaces was measured.

In vitro osteoclastogenesis and resorption assay

Mouse BMMs, perfusion osteoclasts (pre-OCs), and osteoclasts (OCs) were generated as described.(26) In brief, whole bone marrow cells were flushed from long bones of 4- to 6-wk-old WT or RAGE−/− mice and plated on 100-mm tissue culture plates in αMEM containing 10% FBS and 10 ng/ml recombinant M-CSF (R&D Systems). Cells were incubated at 37°C with 5% CO2 overnight. Nonadherent cells were harvested and subjected to Ficoll-Hypaque gradient centrifugation for purification of BMMs. Isolated BMMs were cultured in α-MEM containing 10% FBS plus 10 ng/ml recombinant M-CSF. To generate pre-OCs, BMMs were cultured in αMEM containing 10% FBS in the presence of 10 ng/ml recombinant M-CSF and 100 ng/ml recombinant RANKL for 2–3 days. To generate mature OCs, 5 × 104 BMMs were plated in 1 well of a 24-well plate in αMEM containing 10% FBS in the presence of 10 ng/ml recombinant M-CSF and 100 ng/ml recombinant GST-RANKL. The HMGB1-induced osteoclasts were generated by replacing RANKL with indicated concentrations of HMGB1 at day 3 and continued to culture for over 3 days. Mature osteoclasts (multinucleated, large, spread cells) began to form at day 4 of culture. The identity of osteoclasts was confirmed by TRACP staining. RAW 264.7 cells were cultured with DMEM with 4 μM l-glutamine, 1.5 g/liter sodium bicarbonate, and 4.5 g/liter glucose, containing 10% FBS.

For in vitro osteoclastic resorption activity, 5 × 104 BMMs were replated on BD BioCoat Osteologic coverslips (BD Biosciences) in 24-well plates, in the presence of M-CSF (1%), RANKL (100 ng/ml), and HMGB1 of indicated concentrations (2–10 μg/ml). The BD BioCoat Osteologic coverslips were coated with inorganic hydroxyapatite matrix. After 9 days, cells were washed and stained with 5% of silver nitrate exposed to UV light for 1 h. Resorption pits were analyzed using NIH Image J software.

Immunofluorescence staining

BMMs and osteoclasts cultured on glass were starved and stimulated with 100 ng/ml RANKL for the indicated time and fixed with 4% paraformaldehyde and stained with anti-HMGB1 with FITC-conjugated second antibodies, and Alexa Fluor 488-phalloidin as described.(26) Images were captured with a Zeiss confocal microscope.

Western blot analysis

BMMs and pre-OCs were cultured and starved overnight in serum-free media. Cells were stimulated with 2.5 μg/ml HMGB1 for the indicated times, followed by lysis in RIPA buffer. For integrin signaling experiments, pre-OCs were lifted from the growth substrate, starved for 2 h, and replated on vitronectin (VN)-coated cell culture dishes with or without further stimulation with HMGB1. For HMGB1 release and translocation experiments, cultured media were collected. Cells were harvested and isolated into nuclear and cytoplasmic fractions with the CelLytic NuCLEAR Extraction Kit per manufacturer's instruction (Sigma, St Louis, MO, USA). Lysates were subjected to SDS-PAGE and immunoblotting analysis using the indicated antibodies.

RT-PCR analysis

Total RNA was prepared from BMMs, pre-OCs, and osteoclasts using Trizol (Invitrogen). These cells were generated as described (see In vitro osteoclastogenesis using purified BMMs) at day 0 (BMMs), day 2 (pre-OCs), and day 5 (osteoclasts) in culture in the presence of M-CSF, RANKL, or HMGB1. cDNAs were synthesized from 2 μg of total RNA using SuperScript III First Strand Synthesis System (Invitrogen) in a volume of 20 μl. The reaction mixture was adjusted to 100 μl with distilled water. One microliter of cDNA was amplified with the specific primers indicated by PCR. The PCR product was separated on a 1.5% agarose gel and visualized by ethidium bromide staining. The following primers were used: integrin β3, 5′-TTACCCCGTGGACATCTACTA-3′ and 5′-AGTCTTCCATCCAGGGCAATA-3′; calcitonin receptor, 5′-CATTCCTGTACTTGGTTGGC-3′ and 5′-AGCAATCGACAAGGAGTGAC-3′; MMP9, 5′-CCTGTGTGTTCCCGTTCATCT-3′ and 5′-CGCTGGAATGATCTAAGCCCA-3′; TRACP, 5′-ACAGCCCCCCACTCCCACCCT-3′ and 3′-TCAGGGTCTGGGTCTCCTTGG-5′; cathepsin K, 5′-GGAAGAAGACTCACCAGAAGC-3′ and 3′-GTCATATAGCCGCCTCCACAG-5′; and GAPDH, 5′-TGAA GGTCGGTGTGAACGGATTTGGC-3′ and 3′-CATGTAGGCCATGAGGTCCACCAC-5′.

Statistical analysis

Data were analyzed using an unpaired two-tailed Student's t-test and are expressed as the mean ± SE; p < 0.05 was considered significant.


RANKL stimulation of HMGB1 translocation and release from macrophages and OCs

Our previous work has shown impaired in vitro osteoclastogenesis in RAGE−/− BMMs, thereby suggesting the possibility that HMGB1, a known RAGE ligand highly expressed in macrophages and osteoclasts, may be involved in this event.(25,30,31) To test this concept, we examined the effects of RANKL and M-CSF, two essential cytokines for OC differentiation and function, on HMGB1 release. RAW264.7 macrophages were treated with RANKL or M-CSF for different times. Western blotting of culture media and cell lysates with antibodies against HMGB1 was performed to assess levels of extracellular/released HMGB1 and cytoplasmic and nuclear HMGB1, respectively. Whereas M-CSF had no effect on HMGB1 release (data not shown), RANKL increased levels of HMGB1 in media, but not in cell lysates, consistent with stimulation of HMGB1 release (Fig. 1A). The latter response occurred after a lag (i.e., HMGB1 in the medium increased only after 6 h of stimulation; Fig. 1A), although it was dose dependent (maximal response at 20 ng/ml; Fig. 1B). In addition to these observations in the RAW264.7 macrophage cell line, the same effect of RANKL, but not M-CSF, on HMGB1 release was observed in BMMs (Figs. 1C and 1D). We further examined the subcellular distribution of HMGB1 by Western blotting of nuclear and cytoplasmic fractions of BMMs treated with RANKL. On exposure to RANKL for >6 h, HMGB1 in BMMs seemed to translocate significantly from the nuclear to the cytoplasmic compartment (Fig. 1C). Consistent with these data, immunostaining analysis showed that, although most HMGB1 was detected in the nuclei of quiescent BMMs, treatment with RANKL for 18 h increased cytosolic staining of HMGB1, where it seemed to colocalize with F-actin (Fig. 1E). Taken together, these observations suggest that RANKL stimulates HMGB1 translocation and release from macrophages. The specificity of this event is underscored by the absence of a similar effect with M-CSF.

Figure Figure 1.

RANKL stimulation of HMGB1 translocation and release from macrophages in a time-dependent manner. Western blot analysis of HMGB1 in media and total cell lysates from cultures of RAW264.7 macrophages (A and B) and bone marrow macrophages (BMMs) (C and D) exposed to RANKL (A-C) or M-CSF (D) for the indicated times or concentrations. In C, BMMs were also harvested and isolated into nuclear and cytoplasmic fractions with the CelLytic NuCLEAR Extraction Kit (Sigma, St Louis, MO, USA). Lysates were subjected to Western blot analysis. (E) Immunostaining of HMGB1 in BMMs exposed to RANKL for the indicated times. BMMs were fixed and immunostained using antibodies against HMGB1 (rabbit polyclonal antibody, green), phalloidin to label actin filaments (red), and DAPI to label nuclei (blue). Marker bar, 10 μm.

In addition to BMMs, we also examined the subcellular distribution of HMGB1 in OCs stimulated with RANKL. In quiescent, serum-starved OCs, HMGB1 was primarily localized to nuclei (Fig. 2A). After stimulation with RANKL (∼30 min), translocation of HMGB1 to the cytoplasm occurred (Fig. 2A). Cytosolic HMGB1 seemed to colocalize with actin rings labeled by phalloidin (Fig. 2A). A similar effect was observed on stimulation with TNFα but not M-CSF (Fig. 2B). Taken together, these results suggest that RANKL increases HMGB1 release and translocation in both macrophages and OCs.

Figure Figure 2.

RANKL, but not M-CSF, regulation of HMGB1 translocation in OCs. (A) Time course of HMGB1 translocation in OCs in response to RANKL. Polyclonal rabbit anti-HMGB1 antibody (green), phalloidin (red to label F-actin), and DAPI (blue) were used for immunostaining of OCs differentiated from BMMs. (B) Immunostaining analysis of HMGB1 in OCs exposed to M-CSF, RANKL, and TNFα for 30 min. Marker bars, 25 μm.

Requirement of extracellular HMGB1 for osteoclastic actin ring formation

HMGB1 co-localization with actin rings of OCs led us to speculate a role for HMGB1 in osteoclastic actin cytoskeleton remodeling. To test this hypothesis, we first examined the effects of nicotine on osteoclastic F-actin structures. Nicotine, an inhibitor of LPS-induced release of HMGB1,(32,33) was added to the culture medium of OCs differentiated in vitro in the presence of M-CSF/RANKL. Nicotine (at concentrations of ∼10 μM) significantly reduced RANKL-induced HMGB1 release (Fig. 3A). In parallel, treatment with the same range of nicotine concentrations resulted in a significantly reduced number of actin-ring containing cells compared with control cultures (Figs. 3B and 3C), implicating HMGB1 release in osteoclastic actin ring formation. In view of previous observations indicating that nicotine also decreases release of other cytokines, such as TNFα, in addition to HMGB1,(33) it was necessary to further assess the specificity of the inhibitory effect of nicotine. To this end, neutralizing antibody to HMGB1 was used. Indeed, the number of actin-ringcontaining cells was significantly reduced in the presence of anti-HMGB1 antibody (Figs. 3B and 3C). Taken together, these results suggest that HMGB1 release, as well as extracellular HMGB1, seems to be necessary for osteoclastic actin ring formation.

Figure Figure 3.

Requirement of extracellular HMGB1 for RANKL-induced actin ring formation in OCs in vitro. (A) RANKL-stimulated HMGB1 release in BMMs, as described in Fig. 1, was inhibited in the presence of nicotine (10 μM). (B) Immunostaining analysis of talin, phalloidin, and HMGB1 in OCs exposed to RANKL in the presence or absence of nicotine or anti-HMGB1 antibodies. Marker bars, 25 μm. (C) Quantitative analysis of data from B. Data were presented as percentage of OCs with actin ring formation (mean ± SE). **p < 0.01, significant difference from control (treatment with RANKL alone; t-test).

Requirement of extracellular HMGB1 for RANKL-induced osteoclastogenesis in vitro and in vivo

We next tested the hypothesis that HMGB1 might contribute to RANKL-induced osteoclastogenesis, because actin cytoskeleton organization is important for this event. First, we examined the effects of nicotine and/or anti-HMGB1 antibodies on RANKL-dependent in vitro osteoclastogenesis. In the presence of either nicotine (at concentrations of ∼10 μM) or anti-HMGB1 antibody, significant attenuation of RANKL-induced osteoclastogenesis was observed, as indicated by TRACP staining (Figs. 4A and 4B). Coordinately, RANKL-induced osteoclastic resorptive activity was also significantly reduced in the presence of anti-HMGB1 antibody (Figs. 4C and 4D). Together, these observations suggest a requirement of HMGB1 release, as well as extracellular HMGB1, in RANKL-induced osteoclastogenesis and osteoclast activation in vitro.

Figure Figure 4.

Requirement of HMGB1 for RANKL-induced osteoclastogenesis in vitro and in vivo. (A) HMGB1 is required for RANKL-induced osteoclastogenesis in vitro. Purified BMMs from WT mice were incubated with M-CSF (10%) for 2–3 days and cultured in the presence of RANKL (100 ng/ml) and M-CSF (1%), with or without anti-HMGB1 antibody (0.3 μg/ml) or nicotine (10 μM) for 4 days. The arrows denote a representative TRACP+ cell in each panel. (B) Quantitative analysis of TRACP+ multinuclei cells (MNCs) based on data from A. Data were presented as percentage of total number of BMMs (mean ± SE). **p < 0.01, significant difference from control (treatment with RANKL alone; t-test). (C) Representative resorption caused by RANKL-induced OCs in the absence or presence of anti-HMGB1 antibodies (0.3 μg/ml) was visualized by von Kossa staining. The resorption assay was performed by culturing OCs on coverslips coated with calcium phosphate matrix for 9 days and staining coverslips with von Kossa to display resorption pits (white). (D) Quantitative analysis of the average resorption area based on data from C. (E) HMGB1 is required for RANKL-induced osteoclastogenesis in vivo. C57/BL6 mice (3 mo old) were supracalvarially injected daily for 5 days with vehicle (PBS), RANKL (2 μg/mouse/d) + nonspecific (NS) IgG (3 μg/mouse/d), or RANKL (2 μg/mouse/d) + anti-HMGB1 IgG (3 μg/mouse/d). Animals were killed on day 6, and histological sections of calvariae were stained for TRACP and analyzed histomorphometrically for osteoclast erosion surface. Arrows denote an area displaying TRACP+ cells in each panel. (F) Histomorphometric analysis of TRACP+ OCs covered surface area based on data from E. **p < 0.01 significant difference from RANKL alone (t-test).

We examined whether there was a requirement for HMGB1 in RANKL-induced osteoclastogenesis in vivo. For these experiments, RANKL (2 μg/mouse/d) or vehicle, with or without antibodies to HMGB1 (3 μg/mouse/d), was injected supracalvarially daily for 5 days into WT mice. At the time of death, TRACP-stained sections of calvariae were prepared to determine OC activity histomorphometrically. RANKL increased osteoclastogenesis in WT mice (Figs. 4E and 4F), as previously reported.(3,4) However, this effect was substantially attenuated when RANKL was co-injected with antibodies to HMGB1 (Figs. 4E and 4F); there was an inhibition of ∼67% in the presence of anti-HMGB1 antibodies. This is similar to the degree of inhibition of osteoclastogenesis induced by RANKL in the presence of antibodies to HMGB1 in vitro (Fig. 4B). Thus, extracellular HMGB1 seems to contribute importantly to RANKL-induced osteoclastogenesis in vivo.

HMGB1 induction of osteoclastogenesis in vitro and in vivo in a manner dependent on RAGE

Having established a role for HMGB1 in RANKL-induced OC formation, it was important to determine whether HMGB1 itself was sufficient to induce osteoclastogenesis. To address this issue, vehicle (PBS) or HMGB1 (10 μg/mouse) was injected intraperitoneally daily for 5 days into WT mice. Mice were killed on day 6, and OC number was determined by histomorphometric analysis of TRACP-stained sections of calvariae (as above). HMGB1 substantially increased osteoclastogenesis in WT mice (Figs. 5A and 5D; ∼4-fold; p < 0.01), suggesting the sufficiency of HMGB1 for induction of osteoclastogenesis in vivo.

Figure Figure 5.

HMGB1 induction of osteoclastogenesis in vivo in a manner dependent on RAGE C57/BL6, RAGE−/−, and TLR4−/− mice (3 mo old) were intraperitoneally injected daily for 5 days with vehicle (PBS) and purified recombinant HMGB1 (10 μg/mouse/d). Animals were killed on day 6, and histologic sections of calvariae were stained for TRACP and analyzed histomorphometrically for osteoclast erosion surface. Representative images from WT (A), RAGE−/− (B), and TLR4−/− (C) mice. (D) Histomorphometric analysis of TRACP+ osteoclast erosion surface. (**p < 0.01 vs. PBS control, Student's t-test). Arrows in A-C denote an area of displaying TRACP+ cells.

RAGE and TLR2/4 have been proposed as HMGB1 receptors mediating HMGB1-induced cell signaling and changes in cellular behavior.(34) To determine whether RAGE and/or TLR4 were involved in mediating HMGB1 function(s) in vivo, we took advantage of mice deficient in RAGE or TLR4. HMGB1 induction of osteoclastogenesis in vivo seemed to be largely dependent on RAGE, because HMGB1 failed to have a stimulatory effect on osteoclastogenesis in RAGE−/− mice (Figs. 5B and 5D). In contrast, the role of TLR4 in this context seemed to be less, because HMGB1 induction of osteoclastogenesis was preserved, although slightly reduced, in TLR4−/− mice (Figs. 5C and 5D). Taken together, these observations suggest that RAGE has an integral role in the HMGB1 induction of osteoclastogenesis in vivo.

In the complex in vivo milieu, HMGB1, like most osteoclastogenic agents, may exert its effects through receptors on several cell types, for example, stromal, osteoblastic cells and/or OCs. To determine whether HMGB1 directly promotes differentiation of OC precursors, in vitro osteoclastogenesis with purified BMMs was analyzed. First, isolated BMMs were cultured with increasing amounts of the HMGB1 in the presence/absence of a subosteoclastogenic (permissive) concentration of RANKL (10 ng/ml). HMGB1 alone increased TRACP staining in BMMs. However, under these conditions, HMGB1 did not induce formation of terminally differentiated multinucleated osteoclasts (Figs. 6A and 6B). This experiment was repeated using cultured pre-OCs (i.e., perfusion OCs generated from BMMs pre-exposed to M-CSF and RANKL for 3 days). Addition of increasing amounts of HMGB1 to culture media of pre-OCs, in the presence of a permissive concentration of RANKL, resulted in a dose-dependent increase in OC formation, based on TRACP staining (Figs. 6C and 6E). These HMGB1-induced OCs appeared to be smaller in size and have reduced numbers of nuclei (∼10 nuclei/OC) compared with RANKL-induced OCs (∼20 nuclei/OC). However, HMGB1-induced OCs expressed markers including calcitonin R (CTR), cathepsin K (Cath K), MMP9, TRACP, and integrin β3 associated with differentiated OCs (Fig. 6G), contained intact actin ring structures (Fig. 7C), and exhibited comparable bone resorptive activity as OCs formed in response to RANKL (Figs. 6H and 6I). These observations suggest that HMGB1 may be sufficient to induce osteoclastogenesis (i.e., formation of multinucleated cells with intact actin ring formation and function) from pre-OCs but not BMMs. Thus, as an osteoclastogenic agent, HMGB1 has properties analogous to TNFα or IL-1, in that formation of mature OCs requires precursor cells previously exposed to RANKL, and therefore, committed to the OC pathway.(35,36)

Figure Figure 6.

HMGB1 increases TRACP+ cells from BMMs but stimulates osteoclastogenesis from pre-OCs. BMMs of WT (A) and RAGE−/− (B) were cultured in the presence of vehicle (control) or the indicated concentrations of HMGB1 or RANKL. Where indicated, a permissive concentration of RANKL (10 ng/ml) was added. TRACP staining was performed. Arrow denotes a region with TRACP+ cells (insets show this area at higher power in the first three panels). Pre-OCs of wildtype (C) and RAGE−/− (D) were cultured in the presence of M-CSF (10 ng/ml) and HMGB1 at the indicated concentrations or vehicle alone. TRACP (C and D) staining was performed on day 5. Quantitative analysis of TRACP+ BMMs (E) and TRACP+ multinuclei cells (MNCs) (F) based on data from A-D. Data were presented as percentage of total number of BMMs (mean ± SE). **p < 0.01, significant difference from the control (t-test). #p < 0.01, significant difference from wildtype. (G) RT-PCR analysis to assess expression of transcripts for cathepsin K (Cath K), calcitonin R (CTR), MMP9, TRACP, and integrin β3 in BMMs, pre-OCs, and OCs formed in response to HMGB1 at indicated concentrations for 4 days. (H) Representative resorption caused by control (CTL; pre-OCs incubated with buffer for 9 days) OCs formed in response to indicated concentrations of HMGB1 or RANKL was visualized by von Kossa staining. The resorption assay was performed as described in the legend of Fig. 4C. (I) Quantitative analysis of the average resorption area based on data from H.

To determine the role of RAGE in HMGB1-induced osteoclastogenesis in vitro, we repeated the above experiments using macrophages and pre-OCs derived from RAGE−/− mice. In BMMs from WT or RAGE−/− mice, there was no detectable difference in terms of TRACP induction in response to HMGB1, suggesting a minimal/negligible effect of RAGE in this event (Figs. 6B and 6E), but implicating RAGE-independent receptors (e.g., TLR2/4/9) to be involved in this event. However, HMGB1 induction of the formation of multinucleated actin ring containing cells was inhibited in RAGE−/− pre-OCs, indicating a role for the HMGB1–RAGE interaction (Figs. 6D and 6F). Taken together, these results showed that HMGB1 increases TRACP activity in BMMs and enhances in vitro osteoclastogenesis from pre-OCs in the presence of permissive concentrations of RANKL. However, only the latter seemed to be dependent on RAGE.

HMGB1 regulation of actin cytoskeleton reorganization and cross-talk with integrin in a manner dependent on RAGE

In light of the observations that both extracellular HMGB1 and RAGE are needed for osteoclastic actin ring formation (Fig. 3),(26) we speculated that HMGB1, through RAGE, plays an important role in actin cytoskeleton remodeling in BMMs/OCs. To test this speculation, we first re-examined RAGE expression and distribution in differentiating OCs. RAGE protein was detected in wildtype, but not RAGE−/−, BMMs, indicating a specificity of the antibody (Fig. 7A). In addition, RAGE was highly expressed in pre-OCs and mature OCs (Fig. 7A) and mainly distributed along actin ring structures of OCs (Fig. 7B), where it co-localized with HMGB1, supporting the hypothesis. We next examined F-actin structures in HMGB1-induced OCs that are derived from WT and RAGE−/− mice. HMGB1 induction of the formation of multinucleated actin ring-containing cells was inhibited in RAGE−/− pre-OCs (Fig. 7C). Together, these results are in line with a role of HMGB1-RAGE signaling in actin ring formation of OCs.

Figure Figure 7.

HMGB1 regulation of actin cytoskeleton reorganization in BMMs in a manner dependent on RAGE. (A) RAGE expression in BMMs and in differentiating OCs derived from wildtype and RAGE−/− mice was examined by Western blot analysis. (B) Co-immunostaining analysis of RAGE, HMGB1, and F-actin (by phalloidin) distribution in mature OCs. (C) Immunostaining analysis of F-actin structures in OCs derived from wildtype and RAGE−/− mice. DAPI was used to label nuclei, and GFP was a marker for RAGE−/− cells.(26) (D) Defective HMGB1 signaling in RAGE−/− BMMs cultured onto VN-coated dishes. Pre-OCs from WT and RAGE−/− mice were lifted and replated on VN-coated dishes in the absence of serum. Adherent cells were treated with HMGB1 at the indicated doses for the indicated times and lysed. Equal amounts of total proteins were immunoblotted with the indicated antibodies.

We assessed the possibility of cross-talk between signaling induced by HMGB1 and integrin engagement, because RAGE, acting as a regulator of integrin signaling, participates in OC maturation and activation(26) and HMGB1-induced recruitment of neutrophil.(37) To this end, pre-OCs generated from wildtype or RAGE−/− mice were lifted and replated on VN-coated dishes in the absence of serum. After treatment with HMGB1 at different times, adherent cells were lysed, and phosphorylation of PYK2 and Erk1/2 and c-fos activation/induction were examined. Under these conditions, phosphorylation of PYK2 and Erk1/2 and c-fos activation were induced by HMGB1 in WT pre-OCs (Fig. 7D), which exhibit different time courses (e.g., earlier onset) compared with pre-OCs treated with VN alone or HMGB1 alone (data not shown). However, activation of these proteins was reduced or abolished in RAGE−/− pre-OCs (Fig. 7D). These observations suggest a convergence of signaling pathways activated by integrin engagement and HMGB1 in a RAGE-dependent manner.


HMGB1, a nuclear protein released from activated macrophages or injured cells, displays pro-inflammatory cytokine-like properties once it enters the extracellular space. As a delayed mediator of inflammation, it has been implicated in lethal sepsis, innate immunity, arthritis, and tumor-associated inflammation, and recently, in endochondral ossification.(16,25,38) However, its role in osteoclastogenesis and molecular mechanisms underlying its functions remain to be elucidated. Our work provides evidence that HMGB1, through RAGE, regulates actin cytoskeleton organization and integrin signaling in BMMs and participates in osteoclastogenesis in vitro and in vivo, which leads to the working model depicted in Fig. 8. According to this model, RANKL stimulates HMGB1 release from the nuclear compartment of BMMs. Such extracellular HMGB1 is necessary for RANKL-induced and integrin activation-dependent actin ring formation and osteoclastogenesis in vitro and in vivo. Whereas TLR2/4/9 and RAGE have been implicated as receptors for HMGB1, RAGE seems to participate in HMGB1-induced actin ring formation and OC maturation in vitro and in vivo. TLR2/4/9 may also be involved in HMGB1-induced osteoclastogenesis; however, the exact role of these receptors in this event remains to be further studied. Our observations are in line with previous reports that HMGB1 is released from bone cells (both osteoblasts and osteoclasts),(30,31) provides insight into a function of HMGB1 in regulating osteoclastogenesis, and reveal a novel signaling mechanism underlying RANKL regulation of OC differentiation. Our work highlights properties of HMGB1 as a new member of the group of osteoclastogenic cytokines participating in focal osteolysis. In this context, because release of HMGB1 is elicited by lipopolysaccharide and multiple inflammatory mediators (IL-1, TNFα, interferon-α and -γ), we propose that the HMGB1–RAGE interaction could represent a final common pathway in inflammation-associated bone resorption.(9,35,39)

Figure Figure 8.

A working hypothesis for HMGB1 regulation of osteoclastogenesis. HMGB1 is released from BMMs in response to RANKL stimulation. The extracellular HMGB1, through RAGE in large, regulates osteoclastic actin cytoskeleton remodeling, differentiation, and function.

HMGB1, a nuclear nonhistone protein, contains 215 amino acids with two DNA-binding domains (HMG boxes A and B) and a negatively charged C terminus. Release of nuclear HMGB1 from activated macrophages/monocytes or injured cells undergoing necrosis is a prerequisite for its function as a pro-inflammatory cytokine and/or participation in innate immunity.(16) In this context, LPS and TNFα have been found to stimulate HMGB1 release from macrophages.(25,40) We provide evidence that RANKL also stimulates HMGB1 release from macrophages. Although the molecular mechanisms by which RANKL regulates HMGB1 release and translocation from the nuclear compartment to the cytoplasm in macrophages are unclear, there are likely to be parallels with the response of HMGB1 to these other pro-inflammatory stimuli. LPS and TNFα induce HMGB1 translocation from the nuclear compartment to the cytoplasm after acetylation of HMGB1.(16,25,40) We speculate that RANKL may increase HMGB1 translocation by enhancing acetylation of HMGB1 as well. Cytoplasmic HMGB1 may then be released to the extracellular space through unconventional vesicles or injured membranes in activated macrophages induced by long-term treatment with cytokines.

Once in the extracellular space, HMGB1 seems to be an important participant in RANKL-induced osteoclastogenesis in vitro and in vivo. In support of this view, both blocking HMGB1 antibody and nicotine inhibited RANKL induction of in vitro osteoclastogenesis. Antibody to HMGB1 is an efficient and specific inhibitor of extracellular HMGB1 that has been shown to reverse the lethality of established sepsis.(16,25) Nicotine, a cholinergic agonist, has anti-inflammatory properties because of inhibition of cytokine release, including export of HMGB1 from activated macrophages.(33) This property of nicotine may underlie its ability to enhance survival in experimental models of severe sepsis.(33) Further support for an important role of extracellular HMGB1 in RANKL induction of osteoclastogenesis is the observation that co-injection of HMGB1 antibody along with RANKL in vivo attenuated RANKL-induced osteoclastogenesis. The co-participatory roles of RANKL and HMGB1 in osteoclastogenesis is consistent with the observation that HMGB1 induction of osteoclast formation requires that precursor cells are pre-exposed to (i.e., primed by) permissive levels of RANKL. These properties of extracellular HMGB1 remarkably resemble characteristics of other inflammatory cytokines, such as TNFα and IL-1.(35,39) Whether extracellular HMGB1 is involved in the osteoclastogenic function of TNFα or LPS is unknown but is certainly a possibility.

In the complex in vivo milieu, HMGB1 may behave similarly to other osteoclastogenic cytokines (e.g., TNFα), exerting its effects through receptors on stromal cells, osteoblasts, and/or osteoclasts. We have shown that HMGB1 is sufficient to induce in vitro osteoclastogenesis from pre-OCs. These data suggest the possibility that cells of the osteoclast lineage can both release and respond to HMGB1 in vivo. However, this does not exclude the potential involvement of osteoblasts in the in vivo effects of HMGB1. Indeed, RAGE is expressed on both osteoblasts and OCs. HMGB1-induced in vivo osteoclastogenesis seems to be a more striking effect than suggested by our in vitro studies assessing its osteoclastogenic activity. The latter observation supports the potential contribution of additional cells in vivo in supporting HMGB1-induced osteoclastogenesis.

The molecular mechanisms underlying HMGB1 function are largely unknown. RAGE and TLR2/4/9 have been implicated as receptors for HMGB1. For BMMs (and cells of the innate immune response in general), it seems that TLR4 is critical for HMGB1 early signaling events (e.g., activation of Erk and NF-κB; data not shown), whereas RAGE may have a modulatory role in later events that sustain/propagate the host response (data not shown). The latter could occur independently of TLR2/4, but coordinately with integrin activation and actin cytoskeleton organization. In pre-OCs, RAGE is necessary for integrin signaling and HMGB1 cross-talk to integrin signaling. Such an effect (regulating integrin engagement-induced signaling and actin ring formation) underlies HMGB1 induction of osteoclastogenesis from pre-OCs in vitro and in vivo. However, mechanisms through which RAGE couples to integrins and regulates actin cytoskeleton remodeling will require further study to elucidate. Interestingly, HMGB1 has been recently found to induce polarized co-distribution of RAGE with β2 integrin in neutrophils.(37) RAGE associates with β2 integrins not only in trans (endothelial RAGE interacts with neutrophil β2 integrins),(41) but also in cis (neutrophil RAGE binds to neutrophil β2 integrins).(37) Such interactions seem to be needed for inflammation-associated neutrophil recruitment.(37,41) In this context, we speculate that RAGE may function as do other immunoglobulin superfamily molecules, such as platelet endothelial cell adhesion molecule 1 (PECAM-1), in modulating the activation state of integrins(42) through its interaction with β2 integrin and/or actin cytoskeleton reorganization.

In summary, our results support the concept that HMGB1 has properties of an osteoclastogenic cytokine participating in osteoclastogenesis in vitro and in vivo. The phlogogenic properties of HMGB1, along with its osteoclastogenic activity, are likely to be highly relevant to the pathogenesis of chronic inflammatory disorders, including diabetes, atherosclerosis, and rheumatoid arthritis. The pro-inflammatory nature of these disorders results in an environment in which secretion and/or passive release (i.e., from necrotic cells) of HMGB1 is likely to be enhanced. HMGB1, through RAGE, may have a synergistic effect with RANKL or other osteoclastogenic cytokines, regulating actin cytoskeleton reorganization and integrin signaling, thereby driving osteoclastogenesis in these inflammatory disorders and leading to OC-induced bone loss or inflammatory osteolysis. Thus, our results suggest that HMGB1-RAGE signaling may provide a pathophysiologically relevant pathway for the design of therapeutic approaches to treat disorders associated with inflammation-associated bone loss.


We are grateful to Drs Bernd Arnold, Peter Nawroth, Angelika Bierhaus, Huan Yang, DJ Ferguson, and David Stern for reagents. This study was supported by NIH Grant AR048120 to W-CX. ZZ designed, performed, and analyzed the experiments and wrote the paper. Y-JH, XF, C-YW, and LM provided reagents. C-XX performed some experiments. W-CX designed experiments and wrote the paper.