Receptor Activator of NF-κB and Osteoprotegerin Expression by Human Microvascular Endothelial Cells, Regulation by Inflammatory Cytokines, and Role in Human Osteoclastogenesis*

The receptor activator of NF-κB (RANKL) is the essential signal required for full osteoclast (OC) development, activation, and survival. RANKL is highly expressed in areas of trabecular bone remodeling and inflammatory bone loss, is increased on marrow stromal cells or osteoblasts by osteotropic hormones or cytokines, and is neutralized by osteoprotegerin (OPG), a soluble decoy receptor also crucial for preventing arterial calcification. Vascular endothelial cells (VEC) are critically involved in bone development and remodeling and influence OC recruitment, formation, and activity. Although OCs develop and function in close association with bone VEC and sinusoids, signals mediating their interactions are not well known. Here, we show for the first time that human microvascular endothelial cells (HMVEC) express transcripts for both RANKL and OPG; inflammatory cytokines tumor necrosis factor-α and interleukin-1α elevate RANKL and OPG expression 5–40-fold in HMVEC (with an early OPG peak that declines as RANKL rises), and RANKL protein increases on the surface of tumor necrosis factor-α-activated HMVEC. Cytokine-activated HMVEC promoted the formation, fusion, and bone resorption of OCs formed in co-cultures with circulating human monocytic precursors via a RANKL-mediated mechanism fully antagonized by exogenous OPG. Furthermore, paraffin sections of human osteoporotic fractured bone exhibited increased RANKL immunostaining in vivo on VEC located near resorbing OCs in regions undergoing active bone turnover. Therefore, cytokine-activated VEC may contribute to inflammatory-mediated bone loss via regulated production of RANKL and OPG. VEC-derived OPG may also serve as an autocrine signal to inhibit blood vessel calcification.

Osteoprotegerin (OPG, also known as osteoclastogenesis inhibitory factor or OCIF) is a naturally occurring soluble member of the TNF receptor superfamily that is widely expressed in multiple tissues and binds to RANKL, thereby neutralizing its function (1)(2)(3)(4). OPG therefore acts as a secreted decoy receptor to negatively regulate OC differentiation, activity, and survival both in vivo and in vitro (15,16). OPG production by OB and BMSC is regulated by calcitropic hormones and cytokines, and the balance in the ratio of RANKL to OPG critically determines net effects on OC development and bone resorption (1)(2)(3)(4)11). In vivo administration of OPG to normal rats reduces osteoclastogenesis and increases bone density, OPG prevents estrogen deficiency-associated bone loss in ovariectomized animals, and transgenic mice overexpressing OPG exhibit increased bone density and severe osteopetrosis (15). Conversely, mice deficient in OPG display increased OC development and activity, early onset osteoporosis, and arterial calcification (17,18). Thus, OPG has been proposed to regulate bone resorption both locally and systemically, as well as to serve as a physiological suppressive signal of local calcification in blood vessels (2,17).
Vascular endothelial cells (VEC) are intimately associated with pre-OC and OC during both their formation and resorption of bone (19 -22). This close physical interaction permits VEC to influence directly and convey local and systemic regulatory signals for OC development and bone remodeling under normal or pathological conditions (19 -22). Recent studies indicate that many pre-OCs reside in the peripheral circulation as well as in the bone marrow (23)(24)(25)(26). Circulating pre-OC may therefore also be exposed to and potentially activated by signaling molecules displayed on the VEC surface during and following their transmigration across the VEC layer of blood vessels in response to local stimulatory signals to reach the bone microenvironment. Although all OCs develop and function in close association with the VEC and sinusoids of bone, specific signals mediating interactions between these cells are not well known. Because VEC function as primary immune response cells that are potently activated by TNF-␣ and IL-1 (27,28), we investigated whether primary human VEC expressed RANKL and/or OPG, and if such expression was regulated by these or other pro-resorptive stimuli. The functional consequences of RANKL expression and regulation in HMVEC were assessed relative to the in vitro development and activity of multinucleated bone-resorptive OCs from precursors present in human peripheral blood mononuclear cell preparations.

EXPERIMENTAL PROCEDURES
HMVEC Culture and Treatments-Primary HMVEC of normal adult female dermal tissue origin, media (essential growth medium-microvascular, EGM-MV, and essential basal medium, EBM), and media supplements (packaged as SingleQuots containing human recombinant epidermal growth factor, hydrocortisone, gentamicin, bovine brain extract, and fetal bovine serum, FBS) were obtained from Clonetics Corp. (San Diego, CA). HMVEC were grown, subcultured by trypsin/EDTA, and used within 4 passages as recommended. HMVEC expressed all the hallmark characteristics of endothelial cells (morphology, tubule formation, acetylated low density lipoprotein uptake, factor VIII expression, PECAM-1, ICAM-1, VCAM-1, ELAM-1, P-selectin, ACE, vimentin, and no smooth muscle ␣-actin). For molecular analyses, HMVEC were cultured to near confluence in 24-well dishes in EGM-MV complete medium (with supplements and 5% FBS); modulators were administered the next day in fresh medium, and the cells were incubated for the times indicated before RNA was harvested. For withdrawal experiments, the modulator medium was removed after the induction period, the cells were rinsed twice briefly with fresh medium, and the cells were cultured in medium without modulator for various times before RNA was harvested. Cytokine release by HMVEC was evaluated in cells grown to near confluence in EGM-MV complete medium and switched to phenol red-free EBM (lacking supplements) plus 5% FBS for 16 -24 h before modulators were administered in fresh medium. The conditioned medium (briefly centrifuged) and cells were harvested after 24 h and stored at Ϫ80°C until analyzed for cytokine and protein levels, respectively. Modulators used were 1,25-dihydroxyvitamin D3 (VD 3 , a gift of Hoffmann-La Roche), dexamethasone (Dex), and human parathyroid hormone-(1-34) (PTH-(1-34); both from Sigma), and human recombinant cytokines TNF-␣, IL-1␣, and M-CSF (all from R & D Systems, Minneapolis, MN).
Primary Human Osteoblast and Bone Marrow Stromal Cell Populations-Primary human osteoblasts (HOB) were obtained as cultured outgrowth cells from trabecular bone explants according to the Robey/ Termine method as described previously (29). Human bone marrow stromal cells (HBMSC) were isolated from discarded thoracic rib surgical specimens or bone obtained from accident victims and cultured in ␣-MEM with 10% FBS and 1% antibiotic/antimycotic (Life Technologies, Inc.), with or without Dex (100 nM) and VD 3 (10 nM) to promote a more OB-like phenotype as described previously (29,30).
RNA Isolation and RT-PCR Analysis-RNA was isolated from cells using RNA STAT-60 (Tel-Test, Inc., Friendswood, TX). Semi-quantitative RT-PCR amplification for RANKL was performed using forward and reverse primers (designed by Drs. N. Weitzmann and L. Rifas, Washington University, St. Louis, MO) to the extracellular region of OPGL/TRANCE/RANKL (nucleotides 4 -735) of cloned human TRANCE (nucleotides 1-738, GenBank TM accession number AF013171) and Amersham Pharmacia Biotech Ready-to-Go RT-PCR beads. Oligonucleotide primers were as follows: forward, 5Ј-GCTCTA-GAGCCATGGATCCTAATAGAAT-3Ј, and reverse 5Ј-ATCTCGAGTCA-CTATTAATGATGATGATGATGATGATCTATATCTCGAACTTTAAA-AGCC-3Ј. RT-PCR amplification for OPG was performed using forward and reverse primers to cloned human OPG (GenBank TM accession number U94332) as follows: forward, 5Ј-GGGGACCACAATGAACAAG-TTG-3Ј (nucleotides 85-106), and reverse, 5Ј-AGCTTGCACCACTCCA-AATCC-3Ј (nucleotides 473-493) (31). Parallel reactions were performed for every assay using primers designed to amplify human glyceraldehyde-3-phosphate dehydrogenase (GAPDH) as described previously (32). Initial trials were performed using RNA obtained from unstimulated cultures as well as each modulator treatment to establish optimal cycle numbers and RNA amounts for routine use so that RT-PCRs would yield RANKL, OPG, and GAPDH amplifications within an exponentially linear range over the amount of input RNA used. Thus, cycle numbers were varied from 20 to 35 (OPG), 30 to 35 (RANKL), and 15 to 35 (GAPDH), and RNA amounts from 0.15 ng to 1.5 g (OPG), 25 ng to 12 g (RANKL), and 1 pg to 4 g (GAPDH), in up to 6 replicate trials each. No products were obtained in controls lacking either RNA or first strand primers, and amplified products were not eliminated or reduced by DNase treatment of the RNA samples. Conditions were chosen that consistently provided mid-linear range amplification of each PCR product for all further studies. Therefore, PCRs were run at 94°C for 1 min, 60°C for 1 min, and 72°C for 2 min for 35 (RANKL), 26 (OPG), or 20 (GAPDH) cycles. Products were separated by agarose gel electrophoresis, visualized by ethidium bromide staining, photographed using a Polaroid camera, and quantified in a scanner (Scanjet II, Hewlett-Packard) computer linked to a Quantimet image analysis system (Leica, United Kingdom). RANKL or OPG signals were normalized to GAPDH signals determined in parallel for each sample, and data were expressed as a percentage of the RANKL/GAPDH or OPG/GAPDH ratio for untreated HMVEC measured in the same trial. The 731-bp RANKL and 408-bp OPG amplicons generated by RT-PCR from HM-VEC (as well as HOB and HBMSC) were directly sequenced using an ABI Prism Cycle Sequencing kit (PerkinElmer Life Sciences), and DNA sequences were compared with published sequences to confirm their identity using computation performed at the NCBI and the BLAST network service.
Quantification of Cytokine Secretion by HMVEC-Cytokine levels in conditioned medium were measured using specific enzyme-linked immunoassay kits (Quantikine kits, R & D Systems, Minneapolis, MN) for human IL-1␤, TNF-␣, and M-CSF as recommended. Standard curves were run with each assay, and control medium was analyzed for background levels of each cytokine (which were insignificant), and modulators were tested in at least 3 separate culture wells per trial for 2-8 independent HMVEC cultures. Results were normalized for cell protein using the bicinchoninic acid protein assay (Pierce) and bovine serum albumin (BSA) as a standard (33), and data were expressed as the mean Ϯ S.E. ng/ml of cytokine released per mg of cell protein during 24 h of culture.
Immunodetection of RANKL Protein Expression on HMVEC-HMVEC were cultured in EGM-MV complete medium on glass coverslips in 24-well tissue culture dishes to near confluency, TNF-␣ (1 nM) was administered in fresh medium for 24 h, and the cells were fixed and immunostained (34). Briefly, HMVEC were rinsed, fixed in 3% paraformaldehyde/HBSS (15 min), rinsed, blocked for 1 h with 1% BSA and 10% horse serum in phosphate-buffered saline (PBS), and reacted for 1 h with (or without) primary antibodies diluted in block. Mouse monoclonal antibodies (mAb) specific for human ICAM-1 or VCAM-1 (Serotec, Raleigh, NC, each at 1:200 dilution), mAb to the angiogenesisrelated integrin ␣ v ␤ 3 (LM 609, Chemicon International, Temecula, CA, 1:100 dilution), and goat polyclonal antibody (pAb) raised to a C-terminal extracellular peptide region of human RANKL (Santa Cruz Biotechnology, Santa Cruz, CA, 1:100 dilution) were used. Primary antibody binding was detected using a secondary goat anti-mouse fluorescein isothiocyanate conjugate (Life Technologies, Inc.; 1:200) for mAbs or a biotinylated donkey anti-goat antibody (Santa Cruz Biotechnology; 1:200) followed by a streptavidin-Texas Red conjugate (Life Technologies, Inc.; 1:1000) for RANKL pAb. In some cases, HMVEC were simultaneously immunostained for both ICAM-1 and RANKL to visualize their co-localization on the HMVEC plasma membrane. Coverslips were mounted on glass microscope slides in glycerol-buffered mounting medium (Becton Dickinson, Cockeysville, MD), and images were viewed and digitally captured using a Leica scanning laser (argon/krypton/He) confocal microscope (TCS-SP-2) equipped with a 20ϫ phase objective. Wavelengths for excitation and emission for fluorescein isothiocyanate were 494 and 518 nm, respectively, and those for Texas Red were 595 and 615 nm, respectively.
RANKL Immunodetection in Paraffin-embedded Sections of Human Osteoporotic Bone-Human osteoporotic bone was obtained from fractured femoral heads discarded during hip replacement surgery and briefly held at 4°C in ␣-MEM before fixation in 10% buffered formalin. Samples were decalcified, paraffin-embedded, and sectioned by standard procedures. Sections were prepared for immunostaining by deparaffinization in xylene, hydration through 100% EtOH, 95% EtOH, and water, and heating (97°C for 20 min) in an antigen unmasking Target Retrieval Solution (Dako, Carpinteria, CA). Cooled sections were PBSrinsed, endogenous peroxidase activity was quenched in Dako Peroxidase Blocking reagent (15 min), rinsed sections were blocked with Dako serum-free Protein Block (10 min), and sections were reacted overnight at 4°C with the pAb to human RANKL described above (diluted 1:500 to 1:1000 in PBS ϩ 1.5% Dako Protein Block). Sections were rinsed, incubated with biotinylated donkey anti-goat pAb (Santa Cruz Biotechnology, 1:100 dilution in PBS/block, 45 min) followed by Dako streptavidin-peroxidase (1:300 in PBS, 15 min), reacted with Dako diaminobenzidine solution (5 min), briefly counterstained using Dako hematoxylin solution, and mounted on glass slides with Permount. Immunostained sections were viewed by light microscopy, and images were digitally captured using a computer-linked Olympus microscope.
OC Development in Human Peripheral Blood Mononuclear Cells (PBMC) Co-cultured with HMVEC-HMVEC were cultured in 24-well dishes in EGM complete medium to near confluency, two-thirds of the wells were treated for 48 h with either 1 nM TNF-␣ or IL-1␣ to maximally induce RANKL (while allowing stimulated OPG levels to decline), and the cells were washed (3 times) to remove cytokines just prior to the addition of human PBMC for co-culture. Human PBMC were prepared from heparinized blood obtained from the American Red Cross (St. Louis, MO). Mononuclear cells were isolated by Ficoll-Paque (26,32,33), resuspended in ␣-MEM plus 10% FBS and 1% antibiotic/antimycotic, and added (1.6 ϫ 10 6 PBMC/well) to the 24-well dish containing unactivated or cytokine-pre-activated HMVEC. Some wells also received 100 ng/ml recombinant human OPG:Fc fusion peptide (Alexis Corporation, San Diego, CA). The next day (day 1) all wells were treated with 10 nM VD 3 and 25 ng/ml M-CSF, and OPG:Fc was readministered to the wells originally receiving this treatment. On day 5 the cells were refed with M-CSF, with or without OPG:Fc, and the cells were harvested on day 7, rinsed, fixed in 3% PF/HBSS, rinsed, and stained for TRAP activity (34 -36). Cells were co-stained with DAPI (Molecular Probes, Eugene, OR) to label nuclei, and the numbers of mononuclear and multinucleated TRAPϩ cells, as well as the number of nuclei per TRAPϩ cell, were counted (encompassing ϳ2000 -7000 TRAPϩ cells per condition per trial) across a constant number of sequential random fields using an Olympus light and fluorescent microscope. To evaluate simultaneously OC formation and acquisition of bone-resorbing capability, HMVEC were cultured in 24-well dishes in EGM complete medium to near confluency, half the wells were pre-activated with TNF-␣ (1 nM, 21 h), HMVEC were washed, and human PBMC were added (1.6 ϫ 10 6 PBMC/well) and initially co-cultured with VD 3 and M-CSF, with or without OPG:Fc, as above. Every 4th day, cells were refed with M-CSF with or without OPG:Fc, a small circular disc of ivory was added to each well on day 9, and the cells and ivory pieces were harvested on day 16, rinsed, fixed in 1% PF/HBSS, rinsed, and stained for TRAP activity (34 -36). Cells in the wells were co-stained with DAPI and analyzed for the numbers of TRAPϩ cells formed and nuclei per cell. Ivory was subjected to resorption pit analysis (below). Parallel human PBMC were cultured alone (without HMVEC) with VD 3 and M-CSF, in the presence or absence of human recombinant RANKL (75 ng/ml, Alexis Corp., San Diego, CA) and/or OPG:Fc fusion peptide (100 ng/ml), according to the same feeding regimen used for the co-cultures, and the cells and ivory were harvested on day 16 for analysis as in the co-cultures.
Bone Pit Resorption Analysis-The number of TRAPϩ cells was determined for a constant number (40) of random fields per ivory piece, the cells were then removed, and bone pit resorption within these same fields was quantified using a computer-linked dark-field reflective light microscopic image analysis system (34 -36). TRAPϩ cell counts on the ivory include both MNC and mononuclear cells (since DAPI nuclear staining cannot be used on ivory to distinguish between them). The total area (m 2 ) of ivory resorbed, the number of pits formed, and the size of each excavation were assessed. Results were also normalized to TRAPϩ cell numbers in order to compare the mean area of bone resorbed per TRAPϩ cell (area/TRAPϩ cell) and number of pits formed per TRAPϩ cell (pits/TRAPϩ cell). More than 2000 TRAPϩ cells and their associated resorption pits were evaluated per co-culture trial. Data shown are presented as means Ϯ S.E. from a representative co-culture trial performed in triplicate.
Statistical Analysis-Data are presented as the mean Ϯ S.E. of 3-12 independent trials, typically having at least 3 replicates per condition and assayed at least in duplicate. Differences between treatments were analyzed using single factor analysis of variance. For simultaneous comparisons between multiple treatments, significant differences were determined using the post-analysis of variance Bonferroni test. Differences were considered significant for p Ͻ 0.05.

Primary HMVEC Express mRNA for Both RANKL and
OPG-RT-PCR using specific primers to cloned human OPGL/ TRANCE/RANKL (hereafter referred to as RANKL) yielded a single amplicon product of the expected size (731 bp) from primary HMVEC, as well as from primary HOB and HBMSC (Fig. 1A). Complete nucleotide sequencing of the products obtained from HMVEC and HBMSC demonstrated their identity to sequences reported previously for cloned human OPGL (GenBank TM accession number AF053712), RANKL (accession number AF019047), and TRANCE (accession number AF013171). RT-PCR using specific primers to cloned human OPG also yielded a single amplicon product of the expected size (408 bp) from each of these three primary human cell types (Fig. 1B), and complete nucleotide sequencing of these three products showed that they matched the sequence reported for cloned human OPG (GenBank TM accession number U94332). Therefore, HMVEC express mRNA transcripts for both RANKL and OPG. RT-PCR using total RNA isolated from human primary HMVEC, HOB, or HBMSC and specific primers to the extracellular region of cloned human RANKL (A) or cloned human OPG (B) was performed as described under "Experimental Procedures." A and B, single amplicon products of the expected sizes were obtained for both RANKL (731 bp) and OPG (408 bp) from each of the three primary human cell types, and complete nucleotide sequencing of the products generated from HMVEC and HBMSC confirmed their identity to RANKL or OPG. Amplified RANKL and OPG products were RNA-dependent since they were not obtained in the absence of input RNA or oligonucleotides for reverse transcription into cDNA, and they were not eliminated by DNase pretreatment of RNA samples before RT-PCR (not shown).

HMVEC Differ from HBMSC in Their Hormonal
Regulation of RANKL and OPG-Because RANKL and OPG mRNA levels are regulated during OB differentiation and stimulated by PTH or VD 3 (alone or in combination with Dex) in OB or stromal cells, the modulatory actions of these hormones on RANKL and OPG mRNA expression were studied in HMVEC using semi-quantitative RT-PCR. Treatment of HMVEC for 6, 24, or 72 h with PTH-(1-34) (500 nM) or VD 3 (10 nM) and Dex (100 nM) had no effect on either RANKL or OPG mRNA expression levels normalized to GAPDH in comparison with untreated HMVEC cultured for the same times (Fig. 2, A and C). By contrast, HBMSC differentiated with VD 3 and Dex over 8 days of culture exhibited a 1.5-fold increase in RANKL/GAPDH mRNA levels (Fig. 2B), and a remarkable 10-fold decrease in OPG/GAPDH mRNA levels ( Fig. 2D), in comparison with untreated HBMSC. Although RANKL/GAPDH mRNA expression was higher in HOB than in VD 3 /Dex-differentiated HBMSC (Fig. 2B), OPG/GAPDH mRNA expression was also significantly higher in HOB (Fig. 2D). Consequently, VD 3 /Dex differentiation of HBMSC raised the relative GAPDH-normalized ratio of RANKL to OPG mRNA expression (in arbitrary densitometric units) from 0.04 in untreated HBMSC to 0.63 in VD 3 / Dex-treated HBMSC, a value close to the RANKL/OPG mRNA expression ratio associated with primary HOB (0.86) and HMVEC (set at 1.0). Therefore, RANKL and OPG mRNA levels are regulated by these calcitropic hormones in HBMSC but not in HMVEC under the conditions tested.
TNF-␣ Increases Both RANKL and OPG mRNA Levels in HMVEC but with Temporally Different Kinetics-HMVEC are highly responsive to inflammatory signals such as TNF-␣ and IL-1␣, and each of these potent immune activators increases RANKL and OPG mRNA levels in HBMSC. TNF-␣ and IL-1␣ were therefore tested for their potential effects on RANKL and OPG mRNA expression in HMVEC. TNF-␣ significantly and dose-dependently increased both RANKL and OPG mRNA levels in HMVEC by 24 h as measured by semi-quantitative RT-PCR (Fig. 3, A and C). Although maximal induction of both RANKL and OPG were achieved with 1-10 nM TNF-␣, OPG mRNA levels were increased further (25-fold) than RANKL mRNA levels (5-fold) in relation to their expression levels in untreated HMVEC.
Analysis of the temporal kinetics of RANKL mRNA expression by TNF-␣ (1 nM) revealed that stimulation was first apparent at 10 h post-treatment, rose to a maximal 3-6-fold elevation over untreated HMVEC by 24 h, and was sustained over at least 48 -72 h of culture in the continuous presence of this cytokine (Fig. 3B). Following TNF-␣ (1 nM) withdrawal from 24-h stimulated HMVEC cultures, induced RANKL mRNA levels began to decline (Fig. 3B). However, even after 48 h of culture in the absence of TNF-␣ they remained elevated 2-fold over the levels originally observed in unstimulated HMVEC cells. A similar temporal pattern of RANKL mRNA induction in HMVEC was observed in response to either 0.1 or 10 nM TNF-␣ (data not shown).
In contrast to RANKL, OPG mRNA levels in HMVEC were more rapidly and transiently increased in response to TNF-␣ (Fig. 3D). Stimulated OPG mRNA levels were apparent within 1 h, reached a maximum by 10 h, declined to less than half their peak values by 24 h, and thereafter declined more modestly over 72 h of culture in the continued presence of TNF-␣. However, OPG mRNA levels at 72 h were still 10-fold elevated over those in untreated HMVEC (Fig. 3D). Following TNF-␣ withdrawal from 24-h stimulated HMVEC, OPG mRNA levels rapidly returned to the levels associated with unstimulated HMVEC (Fig. 3D). Together, these results indicate that TNF-␣ increases both RANKL and OPG mRNA levels in HMVEC and

FIG. 2. Calcitropic hormones modulate RANKL and OPG mRNA expression in HBMSC but not in HMVEC.
Semi-quantitative RT-PCR was performed as described under "Experimental Procedures" to amplify RANKL, OPG, and GAPDH in a mid-linear range, and the signals for RANKL or OPG were each normalized to the GAPDH signals determined in parallel. All data were expressed as the mean Ϯ S.E. percentage of the RANKL/GAPDH or OPG/GAPDH mRNA expression determined in control untreated HMVEC (set at 100%). A and C, primary HMVEC cultured with PTH-(1-34) (500 nM) or VD 3 (10 nM) in combination with Dex (100 nM) for 6, 24, or 72 h did not exhibit any significant change in their GAPDH-normalized steady state mRNA expression of either RANKL or OPG. The data shown in the bar graph represents 2 independent cultures of HMVEC incubated for 24 h in the presence or absence of these hormones and analyzed for RANKL (A) and OPG (C) mRNA expression. HMVEC cultured with or without these hormones for 6 or 72 h yielded identical findings (not shown). B and D, primary HBMSC were cultured in the presence or absence of VD 3 (10 nM) and Dex (100 nM) for 8 days and analyzed from 2 independent cultures by semi-quantitative RT-PCR for mRNA expression of RANKL/GAPDH (B) and OPG/GAPDH (D). Primary isolated trabecular HOB (cultured without any modulator treatments) were similarly analyzed. In each graph (A-D), the inset shows results from a representative RT-PCR trial for RANKL or OPG (top bands) and GAPDH (lower bands in each set) corresponding to the 3 conditions represented by the bars. furthermore that their regulation exhibits a nearly reciprocal temporal relationship. OPG mRNA levels rapidly increase in response to TNF-␣ and then begin to decline, whereas RANKL mRNA levels rise more slowly and are sustained at their peak over at least 48 h of culture. Moreover, within 24 h of TNF-␣ withdrawal from HMVEC cultures, OPG mRNA expression returns to basal levels, whereas RANKL mRNA expression remains 3-fold above unstimulated levels.
IL-1␣ Increases Both RANKL and OPG mRNA Levels in HMVEC in a More Complex Manner Than TNF-␣-Treatment of HMVEC for 24 or 48 h with IL-1␣ also significantly and dose-dependently increased both RANKL and OPG mRNA levels measured by semi-quantitative RT-PCR (Fig. 4, A and C). Maximal induction of RANKL or OPG at either 24 or 48 h was achieved with 1-10 nM IL-1␣. Unlike TNF-␣, IL-1␣ (1 nM, 48 h) elicited nearly equivalent ϳ10-fold increases in both OPG and RANKL mRNA expression over the corresponding levels in untreated HMVEC (Fig. 4, A and C). Dose-response curves for RANKL and OPG mRNA expression after 48 h of IL-1␣ (1 nM) treatment paralleled, but were much higher than, RANKL and OPG mRNA stimulation measured after only 24 h of exposure to this cytokine (Fig. 4, A and C).
Analysis of the temporal kinetics of RANKL mRNA expression induced by IL-1␣ (1 nM) showed an early increase first observed 3-6 h post-treatment that rose to a reproducible level 3-4-fold over untreated HMVEC by 10 h (Fig. 4B). This initial stimulation of RANKL mRNA levels was comparable to that elicited by TNF-␣ (Fig. 3B). However, it was immediately followed by a second phase of induction in response to IL-1␣, to levels 8 -10-fold over untreated HMVEC, that peaked within 48 h and was sustained over at least 72 h of culture in the continuous presence of this cytokine (Fig. 4B). Following the withdrawal of IL-1␣ (1 nM) from either 24-or 48-h-stimulated HMVEC cultures, induced RANKL mRNA levels declined within the next 24-h period to the low levels seen in unstimulated HMVEC (Fig. 4B). This contrasts with the more moderately elevated, but sustained, levels of RANKL expression maintained following TNF-␣ induction and withdrawal. The continued presence of IL-1␣ beyond the first 24 h was essential for achieving the second higher peak of RANKL mRNA expression in cultured HMVEC because this did not occur if IL-1␣ was withdrawn immediately after the initial 24-h period. A similar bimodal temporal pattern of RANKL mRNA level induction in HMVEC was also observed in response to 10 nM IL-1␣ (data not shown).
OPG mRNA levels rose more rapidly in response to IL-1␣ than did RANKL mRNA levels (Fig. 4D). However, like RANKL, OPG mRNA levels always exhibited a bimodal temporal pattern of IL-1␣-stimulated expression (Fig. 4D). Elevated OPG mRNA levels were detectable by 1 h, reached an initial maximum by 6 h, and then partially declined briefly, but reproducibly, before rising again to maximal levels 10-fold over unstimulated HMVEC by 24 h (Fig. 4D). Thereafter, OPG mRNA expression declined over 72 h of culture in the continued presence of IL-1␣, although OPG mRNA levels at 72 h were still 4-fold elevated over those in untreated HMVEC (Fig. 4D). Compared with TNF-␣, maximal levels of OPG expression elicited by IL-1␣ were 4-fold lower than peak OPG expression induced by TNF-␣ and were equivalent to the naturally declining levels of OPG expression seen after 72 h of continuous TNF-␣ exposure (Fig. 3, D and 4D). However, like TNF-␣, IL-␣ withdrawal from 24-h stimulated HMVEC led to the rapid return within the next 24 h of OPG mRNA levels to the lower levels associated with unstimulated HMVEC (Fig. 4D). If IL-1␣ withdrawal was delayed until 48 h, when OPG mRNA levels were already naturally declining, little further acceleration in the rate of decline occurred (Fig. 4D). Together, these results indicate that IL-1␣ increases both RANKL and OPG mRNA levels in HMVEC and that their regulated expression exhibits an apparent reciprocal relationship analogous to what had been observed for RANKL and OPG in response to TNF-␣. However, in contrast to TNF-␣, IL-1␣-stimulated RANKL and OPG mRNA increases were more similar in magnitude and exhibited more complex temporal expression profiles. Thus, prolonged stimulation of HMVEC with IL-1␣ caused a greater induction of RANKL and a lesser induction of OPG expression than was elicited by TNF-␣. However, following 24 -48 h of cytokine withdrawal, RANKL expression remained partially elevated in TNF-␣-treated HVMEC, whereas it declined to basal levels in IL-1␣-treated HMVEC, and OPG expression returned to basal levels in both.
When HMVEC were simultaneously treated with TNF-␣ (1 nM) and IL-1␣ (1 nM) for 24 h, a small additive stimulation of RANKL mRNA expression over that evoked by either cytokine alone was typically observed (Fig. 5A). Additive increases were also seen using 0.01 nM of each cytokine in combination (data not shown). In contrast, RANKL mRNA expression was not raised beyond the maximal levels induced by either cytokine alone after 48 h of combined TNF-␣ (1 nM) and IL-1␣ (0.01, 0.1, or 1.0 nM) treatment (Fig. 5B). Like RANKL, OPG mRNA levels were higher in HMVEC concurrently treated for 24 h with both TNF-␣ and IL-1␣ in comparison with the levels induced by either cytokine alone (Fig. 5C). However, OPG mRNA levels at 48 h were generally no higher (except with TNF-␣ in combination with 0.01 nM IL-1␣) in the co-treated cells than in those treated with TNF-␣ alone, although these levels were significantly greater than the stimulation of OPG mRNA by IL-1␣ alone (Fig. 5D).
TNF-␣ Increases RANKL Protein Expression in HMVEC-Changes in mRNA steady state levels are not always accompanied by corresponding changes in protein expression. Therefore, immunostaining was employed to learn whether RANKL protein expression increased in parallel with RANKL mRNA levels following cytokine activation of HMVEC. By using a pAb to the C-terminal extracellular region of human RANKL, a low basal level of specific immunostaining was detected in unstimulated HMVEC (Fig. 6A) that was markedly increased in HMVEC cultured with TNF-␣ (1 nM) for 24 h (Fig. 6B). RANKL immunostaining was primarily associated with the plasma membrane of TNF-␣-activated HMVEC and was intense throughout the sample when viewed by optical sectioning confocal microscopy. Similarly, unstimulated HMVEC exhibited a low level of specific basal immunostaining using a mAb to the cell surface integrin adhesion molecule ICAM-1 (Fig. 6C), and such staining was greatly enhanced and predominantly cell surface-associated in HMVEC stimulated with 1 nM TNF-␣ for 24 h (Fig. 6D). HMVEC co-stained for both RANKL and ICAM-1 exhibited strong overlapping signals for these two molecules all over the cell surface (not shown). TNF-␣-treated HMVEC also exhibited specific increases in immunostaining for the cell surface adhesion molecules VCAM-1 and ␣ v ␤ 3 (not shown). Therefore, TNF-␣ stimulation of HMVEC increases expression of RANKL mRNA in addition to RANKL protein, which is primarily located on the surface plasma membrane of activated HMVEC.

TNF-␣ and IL-1␣ Stimulate the Release of M-CSF, as Well as
One Another, from HMVEC-Because M-CSF is an essential permissive factor required for OC differentiation promoted by RANKL, the effects of TNF-␣ and IL-1␣ on the production of M-CSF by HMVEC were also investigated. HMVEC cultured for 24 h with either IL-1␣ or TNF-␣ (1 or 10 nM) released 2-4-fold more M-CSF into the culture medium than did unstimulated HMVEC (Table I). These pro-inflammatory cytokines also stimulated the release of one another from HMVEC, although IL-1␣ (1 or 10 nM) elicited a somewhat greater increase in TNF-␣ release (up to 8-fold) than TNF-␣ (only at 10 nM) did in IL-1␣ release (3-fold) from HMVEC (Table I).

Inflammatory Cytokine Activation of HMVEC Promotes in Vitro Osteoclastogenesis in Co-cultured Human PBMC by a RANKL-dependent Mechanism Antagonized by OPG-Whether
cytokine stimulation of RANKL mRNA and protein expression in HMVEC enhanced their ability to promote OC formation and development was evaluated in HMVEC co-cultured with M-CSF and human PBMC as a source of OC precursors. Because TNF-␣ and IL-1␣ can themselves influence OC development, bone resorption, and/or survival, these cytokines were withdrawn from activated HMVEC before their co-culture with human PBMC so that biological effects mediated by cytokineinduced RANKL and OPG could be distinguished from potential direct actions of TNF-␣ or IL-1␣. HMVEC that were preactivated with either TNF-␣ or IL-1␣ (1 nM, 48 h), and then washed prior to the addition of human PBMC, caused a significant 1.5-fold increase in the number of TRAPϩ multinucleated cells (MNC) having 3 or more nuclei that formed in the cocultures after 1 week compared with the number generated from human PBMC co-cultured with unstimulated HMVEC (Fig. 7, A and B). These increases were attributed in each case to a cytokine-mediated induction of RANKL in HMVEC, and not other soluble or cell-surface factors (eg. ICAM-1, VCAM-1), because inclusion of an inhibitory OPG fusion peptide that directly binds and neutralizes RANKL completely blocked the stimulatory effects obtained with either TNF-␣ (Fig. 7A) or IL-1␣ (Fig. 7B). These data also suggest that endogenous OPG production by TNF-␣ or IL-1␣ pre-activated HMVEC in the co-cultures was below the inhibitory levels achieved through exogenous addition of 100 ng/ml OPG fusion peptide. In contrast to its suppression of MNC formation promoted by cyto-

FIG. 5. TNF-␣ and IL-1␣ act additively to increase RANKL and OPG mRNA expression at 24 h, but not 48 h, in HMVEC. HMVEC were cultured for either 24 (A and C) or 48 h (B and D)
in the presence or absence of TNF-␣ (1 nM), IL-1␣ (0.01 to 1 nM), or a combination of these cytokines and were subsequently analyzed by semi-quantitative RT-PCR to determine the relative mRNA expression of RANKL/ GAPDH (A and B) and OPG/GAPDH (C and D) as described under "Experimental Procedures." The results shown were compiled from 2 to 5 independent trials, and data were expressed as the mean Ϯ S.E. percentage of the RANKL/GAPDH or OPG/ GAPDH mRNA expression determined in control untreated HMVEC at 24 or 48 h. Significant differences from unstimulated HMVEC are denoted by *, p Ͻ 0.05, from TNF-␣-stimulated HMVEC by ϩ, p Ͻ 0.05, and from IL-1␣-stimulated HMVEC by #, p Ͻ 0.05.

FIG. 6. TNF-␣ increases RANKL and ICAM-1 protein expression on the HMVEC cell surface. HMVEC were cultured for 24 h in
the presence or absence of 1 nM TNF-␣, fixed, immunostained using antibodies specific to human RANKL or ICAM-1, and the labeled cells viewed in an optical sectioning confocal microscope as described under "Experimental Procedures." A and B, RANKL protein expression was immunodetected using a pAb to the C terminus of human RANKL together with a biotinylated secondary antibody and a streptavidin-Texas Red conjugate. Unactivated HMVEC (A) exhibited a low basal level of RANKL protein immunostaining, whereas a markedly stronger signal was associated with HMVEC that had been exposed to TNF-␣ (B). C and D, ICAM-1 protein expression was immunodetected using a mAb to human ICAM-1 and a fluorescein isothiocyanate-conjugated secondary antibody. As for RANKL, unactivated HMVEC (C) displayed a low basal level of ICAM-1 protein immunostaining, whereas an intense signal was detected in TNF-␣-activated HMVEC (D). No staining was apparent in parallel samples (ϮTNF-␣) developed in the absence of either primary antibody (data not shown). A-D, magnification ϫ 200. kine-stimulated HMVEC, OPG did not diminish the 2-5-fold greater MNC formation that occurred in human MN co-cultured with (untreated) HMVEC (Fig. 7, A and B) compared with human MN cultured in the absence of HMVEC (data not shown). Thus, OPG restrained TRAPϩ MNC formation stimulated by either TNF-␣-or IL-1␣-activated HMVEC via a RANKLdependent mechanism but not that promoted by unactivated HMVEC in a RANKL-independent pathway.
Inflammatory Cytokine Activation of HMVEC Promotes the Fusion of OC Precursors to Form Larger, More Multinucleated TRAPϩ MNC-In vitro osteoclastogenesis was examined in further detail using TNF-␣-pre-activated HMVEC (1 nM, 24 h) co-cultured with human PBMC for 16 days. Again, TNF-␣activated HMVEC caused a significant 2-fold increase in the number of TRAPϩ MNC (with 3 or more nuclei) that formed in the co-cultures compared with co-cultures with unstimulated HMVEC, and OPG fusion peptide fully blocked such induction, thereby implicating a RANKL-dependent mechanism (Fig. 8A). The proportion of total TRAPϩ cells that became MNC was also significantly increased by TNF-␣ pre-activation of HMVEC (Fig. 8A). However, the overall number of mononuclear or total (mononuclear plus MNC) TRAPϩ cells in the co-cultured human PBMC population was not affected by TNF-␣ pretreatment of HMVEC, suggesting that the increase in TRAPϩ MNC formation in the presence of TNF-␣ pre-activated HMVEC likely involved the stimulated fusion, rather than proliferative expansion, of TRAPϩ precursor cells (Fig. 8B). Consistent with this, microscopic examination of human PBMC co-cultured with TNF-␣-activated HMVEC revealed formation of not only more numerous but also generally larger TRAPϩ MNC than were obtained in co-cultures of human PBMC with unstimulated HMVEC (Fig. 8C). Inclusion of the OPG fusion peptide prevented these increases in TRAPϩ MNC numbers and size, without affecting basal co-culture TRAPϩ MNC formation (Fig.  8C). MNC size changes were confirmed by quantifying the number of nuclei contained within individual TRAPϩ MNC (Table II). Thus, TNF-␣-activated HMVEC day 16 co-cultures contained fewer small MNC (with 3 nuclei/MNC) and more numerous large MNC (with Ն4 nuclei/MNC). These were the only cultures to contain MNC with as many as 11-16 nuclei/ MNC, and the addition of OPG fusion peptide abrogated the stimulated formation of larger MNC from human PBMC (Table  II). Similarly, TRAPϩ MNC sizes were increased in day 7 co-cultures containing TNF-␣ (Table II) or IL-1␣ (not shown) pre-activated HMVEC, and reduced in the presence of the OPG inhibitory peptide. Thus, inflammatory cytokine-activated HMVEC stimulated both the number and multinuclearity of TRAPϩ MNC formed in co-cultures with human PBMC, in each case via a RANKL-dependent pathway antagonized by OPG.

TNF-␣ Activation of HMVEC Promotes in Vitro Formed OC Bone Pit Resorption by a RANKL-dependent Mechanism-In-
creased MNC formation was also accompanied by parallel changes in bone pit resorption activity (Fig. 9). Human PBMC co-cultured with TNF-␣-preϪactivated HMVEC exhibited substantial increases in the overall area of ivory resorbed and the total number of pits formed, effects that were completely abolished in the presence of the OPG fusion peptide (Fig. 9, A and  B). Although such increased resorption activity could simply reflect the greater numbers of TRAPϩ MNC formed under these conditions, TNF-␣-stimulated HMVEC also activated individual TRAPϩ cells formed to resorb more ivory. Thus, the mean area resorbed per TRAPϩ cell (Fig. 9D), the number of pits formed per TRAPϩ cell (Fig. 9E), and the size of lacunae formed (Fig. 9F) were all increased in the co-cultures containing TNF-␣-stimulated HMVEC, and elevations in these resorption parameters were completely prevented by the presence of the OPG fusion peptide. Therefore, overall the TNF-␣ activation of HMVEC promoted the developmental formation and FIG. 7. TNF-␣-or IL-1␣-activated HMVEC promote in vitro osteoclastogenesis by day 7 in co-cultured human PBMC via a RANKL-dependent mechanism antagonized by OPG. HMVEC were pre-activated for 48 h with 1 nM TNF-␣ or IL-1␣, washed (to eliminate potential direct effects of the cytokines), and co-cultured with human PBMC (containing OC cell precursors) in the presence or absence of a RANKL-neutralizing recombinant human OPG fusion peptide (100 ng/ml) for 7 days, and the development of TRAPϩ multinucleated cells was evaluated as described under "Experimental Procedures." A and B, the number of TRAPϩ multinucleated cells (containing 3 or more nuclei) formed in co-cultures containing TNF-␣-(A) or IL-1␣ (B)-pre-activated HMVEC were counted for each well (totaling over 25,000 MNC per trial), and the data were expressed as the mean Ϯ S.E. total number of TRAPϩ MNC formed per well. Significant differences from control co-cultures of human PBMC and unactivated HMVEC in the absence of the OPG fusion peptide are denoted by *, p Ͻ 0.05, and from TNF-␣ or IL-1␣ activated HMVEC co-cultured with human PBMC in the absence of OPG by ϩ, p Ͻ 0.05.

TABLE I
Cytokine production by HMVEC HMVEC were cultured for 24 h in the presence of IL-1␣ or TNF-␣ (1 or 10 nM), after which the conditioned media were harvested and analyzed using cytokine-specific immunoassays for the levels of IL-1␤, TNF-␣, and M-CSF released as described under "Experimental Procedures." Results obtained from duplicate wells of at least three independent trials were each normalized for cell protein, and the data were expressed as the mean Ϯ S.E. ng/ml cytokine released per mg cell protein. fusion of TRAPϩ mononuclear cells into MNC and activated their bone pit-resorptive function via RANKL-dependent mechanisms that were antagonized by OPG.

RANKL Is Highly Expressed by Bone VEC Located Near
Resorbing OC in Areas of Active Human Bone Turnover in Vivo-Because developing and resorbing OCs are situated close FIG. 8. TNF-␣-activated HMVEC promote the fusion of OC precursors to form larger, more multinucleated TRAP؉ MNC via a RANKL-dependent mechanism antagonized by OPG. HMVEC were pre-activated for 24 h with TNF-␣ (1 nM), washed, co-cultured with human PBMC (containing OC cell precursors) in the presence or absence of a RANKL-neutralizing recombinant human OPG fusion peptide (100 ng/ml) for 16 days, and the development of TRAPϩ multinucleated cells was evaluated as in Fig. 7. A, the number of TRAPϩ multinucleated cells (containing 3 or more nuclei) were counted in each co-culture well (totaling over 5000 MNC), and the data were expressed as the mean Ϯ S.E. total number of TRAPϩ MNC formed per well (open bars) or proportion of TRAPϩ MNC relative to total TRAPϩ cells formed (hatched bars). Significant differences from control co-cultures of human PBMC and unactivated HMVEC in the absence of the OPG fusion peptide are denoted by *, p Ͻ 0.05, and from TNF-␣-activated HMVEC co-cultured with human PBMC in the absence of OPG by ϩϩ, p Ͻ 0.01, and ϩϩϩ, p Ͻ 0.001. B, the mean Ϯ S.E. total number of TRAPϩ mononuclear cells (open bars) or TRAPϩ mononuclear plus MNC cells (hatched bars) per well was not influenced by TNF-␣ pre-activation of HMVEC. Although a trend toward higher numbers of TRAPϩ mononuclear cells was seen in the co-culture wells containing the OPG fusion peptide, this did not reach statistical significance. C, microscopic examination of the co-cultures fixed and stained for TRAP activity on day 16. Note the formation of more numerous and generally larger TRAPϩ MNC in wells containing TNF-␣ pre-activated HMVEC, and the loss of this stimulation upon addition of the RANKL neutralizing OPG fusion peptide. Magnifications, ϫ 50 (upper panel) and 100 (lower panel). to blood vessels and capillaries in bone, and circulating OC precursors must transmigrate through a VEC barrier to reach the bone marrow microenvironment, inflammatory regulated expression of RANKL and OPG by VEC may impact on and contribute to the development, survival, and resorptive activity of OCs at sites of localized bone loss. Therefore, to validate further the physiological relevancy of our in vitro findings, we examined whether RANKL protein was detectable on VEC within human bone tissue, particularly in areas undergoing bone turnover. Immunohistochemical staining of paraffin-embedded sections of human osteoporotic bone derived from femoral head fractures was performed using a pAb raised to human RANKL and showed RANKL protein expression consistently displayed on blood vessels or capillaries located in regions associated with OC lacunar resorption and active bone remodeling (Fig. 10, A-D). RANKL signals on VEC appeared comparable in intensity to that of other RANKL-expressing cells in these regions, including OB-like cells located along the bone surface (Fig.  10, B and D). RANKL appeared to be expressed in a polarized fashion on the basolateral surface of VEC facing the bone marrow microenvironment, where it could presumably interact with circulatory cells emigrating into the bone tissue as well as with pre-OC and resorbing OC in the bone marrow via close physical interactions between VEC and such cells (Fig. 10D). By contrast, blood vessels and capillaries proximal to newly forming osteoid or quiescent bone surfaces within these same bone sections evidenced no detectable specific RANKL protein staining (Fig. 10E). Control sections incubated without primary pAb also showed no peroxidase signals in any cell type, including VEC (Fig. 10F).

DISCUSSION
This study has demonstrated for the first time that primary HMVEC express both RANKL and OPG, that the pro-inflammatory cytokines TNF-␣ and IL-1␣ induce elevated RANKL and OPG mRNA levels in HMVEC according to differing temporal expression profiles, and that TNF-␣-or IL-1␣-stimulated levels of RANKL mRNA and protein in HMVEC function to promote the in vitro development and activation of bone pit-resorptive OCs that form in co-cultures with human PBMC. In vivo, RANKL protein expression is increased on blood vessels or capillaries in the vicinity of resorbing OCs and regions of active bone remodeling within sections of human osteoporotic bone. Therefore, microvascular endothelial cells may have an important role in regulating localized bone loss through their stimulated expression of RANKL and OPG, as well as M-CSF, the key factors involved in controlling OC development, survival, and bone resorption both in vitro and in vivo (1)(2)(3)(4).
Although it has long been known that the vasculature of bone plays an essential role in the development, dynamic remodeling, and repair of bone, it has only recently become clear that the vascular endothelium functions as more than a permeability barrier and passive conduit of cells and endocrine signals, but also as a vital secretory and immunologic organ (19,(22)(23)(24)(25)(26)(27). OC hematopoietic precursor cells are present within both the peripheral circulation and the bone marrow, and in all cases they develop into mature functional OC within the bone tissue while in close spatial proximity to the micro- FIG. 9. TNF-␣-activated HMVEC stimulate the bone pit resorptive activity of OCs formed in co-cultures with human PBMC via a RANKL-dependent mechanism antagonized by OPG. Ivory chips harvested from the co-culture wells of Fig. 8 were stained for TRAP activity and analyzed over a constant number of random fields (40) for the overall amount of bone pit resorption activity (A and B), number of TRAPϩ cells (C), and resorptive activity per TRAPϩ cell (D-F) as described under "Experimental Procedures." Data were obtained from 3 independent wells per condition of a representative co-culture resorption trial (totaling over 2000 TRAPϩ cells and their associated resorption pits) and were expressed as the mean Ϯ S.E. for each resorption parameter. A and B, TNF-␣-pre-activated HMVEC significantly increased the mean overall area (m 2 ) of ivory resorbed (A) and the mean total number of pits formed (B) by TRAPϩ MNC formed in the co-cultures. The OPG fusion peptide completely prevented such stimulation. C, TNF-␣-pre-activated HMVEC did not significantly increase the total number of TRAPϩ cells attached to the ivory chips (consistent with the data in Fig. 8 demonstrating that the number of TRAPϩ MNC, but not total TRAPϩ cells, is increased by this co-culture treatment). D-F, TNF-␣-pre-activated HMVEC significantly increased the mean area (m 2 ) of ivory resorbed per TRAPϩ cell formed (D), the mean number of pits formed per TRAPϩ cell (E), and the mean size (m 2 ) of pits excavated (F). The OPG fusion peptide completely prevented this activation of the resorption activity of TRAPϩ cells formed in the TNF-␣-stimulated HMVEC co-cultures. Significant differences from control co-cultures of human PBMC and unactivated HMVEC in the absence of the OPG fusion peptide are denoted by *, p Ͻ 0.05, and from TNF-␣-activated HMVEC co-cultured with human PBMC in the absence of OPG by ϩ, p Ͻ 0.05, and ϩϩ, p Ͻ 0.01. vasculature and sinusoids (19 -23, 37). Thereafter, OCs remain intimately associated with microvascular endothelial cells during their active resorption of bone (19 -23, 38, 39). Whereas OB and BMSC are well documented to regulate OC formation and function through both soluble and cell contact-mediated mechanisms, much less is known regarding how microvascular endothelial cells may interact with developing and mature OC. Recently, RANKL mRNA was detected in the metaphyseal vessels of bone (5), RANKL protein in the small blood vessels of the skin (40), and OPG, RANK, and RANKL mRNA in the calcified arteries of OPG-deficient mice but only OPG mRNA in normal adult mouse arteries (41). Here, we have shown that primary HMVEC express mRNA transcripts for both RANKL and OPG in a regulated manner. Unlike BMSC or OB, HMVEC did not respond to PTH or VD 3 /Dex with any changes in RANKL or OPG mRNA levels, although VEC have been reported to respond in other ways to such signals (42)(43)(44). This contrasts with numerous studies in which human or mouse BMSC or OB have responded to these and other osteotropic hormonal signals by up-regulating RANKL mRNA and reduc-ing (or not altering) OPG mRNA levels (1-6, 10, 11, 45-48). A reciprocal pattern of RANKL and OPG mRNA expression also occurs during OB-like development, with consequent effects on osteoclastogenesis (1-6, 11, 46 -50). Our findings therefore suggest that VEC may not directly contribute to the increased OC formation and resorptive activity associated with the complex actions of these osteotropic hormones.
In contrast, VEC may have a prominent role in regulating the recruitment and development of bone-resorptive OCs at localized sites of inflammation, thereby contributing to the osteopenia associated with rheumatoid arthritis, periodontal disease, and other inflammatory disorders. Because the proinflammatory cytokines TNF-␣ and IL-1␣ potently activate VEC, stimulate OC-mediated bone resorption in vivo and in vitro, and are elevated and known to play an important role in pathological conditions associated with bone loss, their potential regulation of RANKL and OPG in HMVEC was investigated. TNF-␣ or IL-1␣ activation of HMVEC led to the doseand time-dependent stimulation of both RANKL and OPG mRNA levels, with OPG mRNA levels rising rapidly and then FIG. 10. Blood vessels or capillaries of human osteoporotic bone exhibit RANKL immunostaining in areas associated with OC resorption and bone remodeling. Paraffin-embedded sections of human osteoporotic bone derived from fractured femoral heads were immunostained using a pAb to human RANKL, developed using a peroxidase/diaminobenzidine protocol, counterstained with hematoxylin, and viewed and photographed as detailed under "Experimental Procedures." RANKL protein was consistently immunodetected on VEC situated close to resorbing OCs and regions of bone remodeling; conversely, no RANKL immunostaining was exhibited by VEC that were not located in such areas. A, immunodetection of RANKL protein on VEC (arrows) adjacent to OCs (arrowhead) engaged in the resorption and remodeling of bone. Note that other cells within the bone marrow and along the bone surface also exhibit positive signals for RANKL protein. OCs were RANKLnegative. B, immunodetection of RANKL protein on VEC (arrows) and OB-related cells (indicated by asterisks below the cells) in a region of active bone remodeling. C, RANKL immunostaining associated with a bone capillary in which a portion has been sectioned parallel to the length of the capillary (arrow). Note the relatively homogeneous distribution of the RANKL signal along the capillary surface. D, RANKL immunostaining in VEC-(arrows) and OB-related cells (asterisks) in a region undergoing bone remodeling. Note that RANKL signals detected on the bone capillary appear to be generally polarized to the outer VEC surface. E, RANKL protein is not immunodetected on VEC (arrows) associated with a region of newly forming bone osteoid (star) in a different area of the bone section shown in B. F, control section demonstrating no peroxidase signals associated with VEC (arrow) near resorbing OCs (arrowhead) in the absence of the primary pAb to human RANKL. Magnification, ϫ 600 (A) and 400 (B-F).
declining at about the time that RANKL mRNA levels were achieving maximal sustained levels. Compared with TNF-␣, IL-1␣ evoked a similar initial but greater maximal rise in RANKL mRNA expression (to levels comparable to HBMSC), a lesser stimulation of OPG mRNA levels, and more complex biphasic kinetics of RANKL and OPG mRNA expression. Recently, TNF-␣ and IL-1␤ were also reported to increase RANKL and/or OPG mRNA levels in primary HBMSC, HOB, and human osteosarcoma MG-63 cells (13,48,50,51). As in HMVEC, RANKL mRNA levels were stimulated 2-4-fold in HBMSC by nanomolar concentrations of the cytokines; peak RANKL mRNA levels were reached by 12 h and maintained for at least 24 h, and IL-1␤ elicited a greater increase than did TNF-␣ in RANKL mRNA levels, whereas the opposite was true for OPG mRNA induction (13). However, no temporal analysis of OPG mRNA induction was reported (13). In MG-63 and HOB, OPG mRNA levels increased in response to TNF-␣ or IL-1␣ within 2 h, peaked by 4 -8 (MG-63) or 16 h (HOB), and thereafter declined (MG-63) by 24 h (50, 51). However, parallel changes in RANKL mRNA levels were not examined. To our knowledge, the present study is the only direct demonstration that a co-stimulation of RANKL and OPG mRNA levels can involve a reciprocal temporal expression pattern, with an early rise in OPG followed by its decline in parallel with a delayed and sustained rise in RANKL. Because the net effects on OC formation, survival, and bone resorption activity are critically determined by the ratio of RANKL to OPG, the temporal nature of their mRNA regulation and the actual levels of RANKL and OPG protein produced are key parameters that govern their physiological effects (1)(2)(3)(4)41). RANKL protein levels assessed by immunostaining were increased along with RANKL mRNA levels in TNF-␣-activated HMVEC. Enhanced RANKL expression on the surface of HMVEC proved physiologically capable of inducing the in vitro formation of OCs when HMVEC were pre-activated with either TNF-␣ or IL-1␣ and then directly co-cultured with human PBMC containing OC precursors. Whereas M-CSF stimulates the early stages of OC recruitment and development (including RANK and TRAP expression), RANKL affects the later stages of OC cell fusion and differentiation into mature functional OCs (1)(2)(3)(4). Consistent with this, all of the co-cultures that received M-CSF contained TRAPϩ cells, and the total number of TRAPϩ cells was not influenced by TNF-␣ pre-activation of HMVEC. However, either TNF-␣ or IL-1␣ pre-activation of HMVEC caused a remarkable 1.5-3-fold greater number of TRAPϩ MNC to form in the co-cultures, as well as an increase in the size and multinuclearity of such TRAPϩ MNC. Addition of an anti-RANKL neutralizing inhibitory OPG fusion peptide completely abolished this stimulation of TRAPϩ MNC number, size, and multinuclearity, thereby establishing that the mechanism by which TNF-␣ or IL-1␣ activated HVMEC increased OC formation was via a RANKL-dependent pathway. The fact that TNF-␣ or IL-1␣ stimulated HMVEC each caused a similar level of RANKL-dependent human OC development by day 7 in the co-cultures, even though these cytokines induced RANKL/ GAPDH and OPG/GAPDH mRNA levels in HMVEC according to somewhat different temporal profiles, may potentially be explained by the kinetics of their effects on these molecules (Figs. 3 and 4). Thus, following TNF-␣ or IL-1␣ induction and withdrawal from stimulated HMVEC (as was performed before the addition of human PBMC to avoid potential confounding effects due to direct actions of these cytokines on human OC formation or survival), RANKL/GAPDH mRNA levels remained moderately elevated in TNF-␣-stimulated HMVEC but declined from highly elevated to basal levels in IL-1␣-stimulated HMVEC, whereas OPG/GAPDH levels returned to basal levels within 24 h in both TNF-␣and IL-1␣-stimulated HMVEC. Therefore, human PBMC might be exposed to higher initial, but briefer, elevated RANKL levels in the co-cultures containing IL-1␣ pre-activated HVMEC versus more moderate, but sustained, elevated RANKL levels in the co-cultures containing TNF-␣ pre-activated HMVEC. Only low basal OPG levels should be present in these co-cultures after 1 or 2 days due to the rapid fall in OPG expression following cytokine withdrawal, the labile nature of OPG, and its removal through medium changes. These effects might balance out overall to produce similar levels of in vitro human OC formation. Although additional experiments would be required to confirm and dissect this further, the key message derived from the current studies is that both TNF-␣ and IL-1␣ prove capable of activating HMVEC to promote human OC formation in vitro via a RANKL-dependent mechanism. Greater OC formation was accompanied by enhanced bone pit resorption activity, both overall as a result of increased OC numbers and on a per cell basis reflecting activation of these OCs for resorption. Thus, the total area resorbed, the number of pits formed, the area resorbed per cell, the resorption sites initiated per cell, and the mean lacunar pit size excavated by OCs formed in the TNF-␣-activated HMVEC co-cultures were all significantly higher than in the other co-culture conditions. A RANKL-mediated pathway was again implicated since the OPG inhibitory peptide fully suppressed this activation of OCs for resorption. All of these actions elicited by TNF-␣-or IL-1␣-activated HMVEC to promote pre-OC cell fusion, multinucleation, differentiation, and activation of OCs for bone pit resorption match those reported previously to be caused by RANKL and antagonized by OPG (1)(2)(3)(4).
In vivo, TNF-␣ and IL-1␣ tend to be co-produced at sites of localized inflammation, elicit many similar biological responses, may function in concert with one another, and often remain elevated if the inflammatory condition does not resolve and progresses into a chronic disorder. Many inflammatory cytokines, including TNF-␣ and IL-1, that act to initiate and maintain inflammation and to promote bone resorption also stimulate angiogenesis, a hallmark characteristic and a vital component of the pathology of inflammation, tumor-associated osteolysis, osteoporosis, and other skeletal disorders (19,27,28,36,(52)(53)(54). Increased angiogenesis enables greater recruitment of circulating OC precursors to localized sites of inflammation, and the close contact that VEC share with transmigrating cells and OC precursors already residing within the bone marrow allows for their direct exposure to RANKL, M-CSF, and other regulatory molecules expressed by activated VEC that could initiate their development into bone-resorptive OCs. Consistent with this, we found that sections of human osteoporotic bone prepared from femoral head fractures exhibited RANKL protein expression on VEC that were located proximal to resorbing OCs and regions of active bone remodeling in vivo. In contrast, negligible RANKL signals were detected on VEC associated with newly forming osteoid or quiescent regions of these same human bone samples. Osteolysis in such fractured femoral heads is known to be associated with increased OC numbers, trabecular scalloping, and locally elevated levels of pro-inflammatory cytokines including TNF-␣ and IL-1. Therefore, the in vivo immunohistochemical findings strongly support the in vitro HMVEC studies and suggest that activated VEC may contribute to promoting the development, activity, and survival of bone-resorptive OCs at inflammatory sites through their regulated expression of RANKL and OPG. Prolonged exposure of VEC in vivo to TNF-␣ or IL-1␣ may provoke a relatively persistent increase in RANKL expression, together with a rise and subsequent fall in OPG expression, based on our in vitro findings with continuously stimulated HMVEC. Furthermore, in contrast to our in vitro co-culture studies wherein cytokines were withdrawn after HMVEC activation, osteoclastogenic effects by VEC in vivo might be more pronounced in response to IL-1␣ versus TNF-␣ since continuous IL-1␣ stimulation caused greater RANKL induction and lesser OPG induction than did TNF-␣ in HMVEC. Moreover, the effects we have measured in vitro may be further magnified in vivo since OPG is produced as a secreted factor whose local concentrations might be dissipated systemically through circulatory flow, thereby effectively increasing the potency of transmembrane RANKL locally expressed on activated VEC. These issues will require further investigation. Such VEC-related mechanisms likely interface with similar and additional important soluble and contact-dependent regulatory signals received from BMSC, OB, and other cells present in the bone marrow microenvironment.
The regulated production of RANKL and OPG by VEC also has broader implications that extend beyond control of OCmediated bone development and remodeling. Thus, RANKL expression by activated VEC potentially allows the vasculature to participate directly, in a spatial and temporal manner, in various important RANKL-mediated developmental and immune-related processes, such as lymph node organogenesis, lymphocyte development, and T cell/dendritic cell interactions (1)(2)(3)(4)14). Conversely, VEC-derived OPG could provide a counterbalancing signal for such processes, in addition to its role in calcium homeostasis and as a local and systemic inhibitor of OC formation and bone resorption (1)(2)(3)(4). Due to the enormous surface area represented by the endothelium throughout the body, VEC could represent a major source of the circulating OPG found in serum and contribute to the increased serum OPG levels reported for aging healthy men and women and in postmenopausal osteoporotic women (a condition associated with increased vascularity and levels of IL-1 and TNF-␣) (55). Because OPG functions as an important physiological suppressor of vascular calcification, mice deficient in OPG exhibit arterial calcification in addition to early onset osteoporosis (which are both prevented by transgenic OPG expression), and OPG acts as an ␣ v ␤ 3 -induced survival factor for VEC OPG production by VEC may also serve as a key autocrine signal to inhibit blood vessel calcification (1-4, 17, 41, 56).
In conclusion, inflammatory cytokine activation of HMVEC caused a sustained up-regulation of RANKL and a more transient expression of OPG, the key molecules involved together with M-CSF in controlling OC formation, survival, and bone resorption in vivo and in vitro. Functionally, this led to an increased ability of HMVEC to stimulate the in vitro co-culture development and activity of mature bone pit-resorptive OCs from circulating human monocytic precursors. In vivo, RANKL expression was up-regulated on VEC only in the vicinity of resorbing OCs and areas of active bone remodeling. Therefore, we surmise that inflammatory-activated VEC may help promote localized bone loss via their RANKL-mediated effects on pre-OC and OC to increase the number of sites and/or rates of bone remodeling. Therapeutic intervention aimed either at preventing an increase in RANKL or further enhancing OPG production by activated VEC may therefore help to alleviate the osteopenia seen in various inflammatory diseases, metabolic bone disorders, malignancy-related osteolyses, or certain immune disorders.