Classical and Paradoxical Effects of TNF-α on Bone Homeostasis

Tumor necrosis factor-α (TNF-α) plays an essential role in the regulation of bone homeostasis in several chronic immune and inflammatory joint diseases, where inhibition of TNF has led to significant clinical improvement. However, TNF-activated pathways and mechanisms involved in bone remodeling remain unclear. So far, TNF-α was known as an inhibitor of osteoblast differentiation and an activator of osteoclastogenesis. Recent contradictory findings indicated that TNF-α can also activate osteoblastogenesis. The paradoxical role of TNF-α in bone homeostasis seems to depend on the concentration and the differentiation state of the cell type used as well as on the exposure time. This review aims to summarize the recent contradictory findings on the regulation of bone homeostasis by TNF-α at the isolated cell, whole bone, and whole body levels. In addition, the involvement of TNF-α in the bone remodeling imbalance is observed in inflammatory joint diseases including rheumatoid arthritis and ankylosing spondylitis, which are associated with bone destruction and ectopic calcified matrix formation, respectively. Both diseases are associated with systemic/vertebral osteoporosis.


INTRODUCTION
Tumor necrosis factor-α (TNF-α) is one of the several proinflammatory cytokines involved in bone remodeling. Classically, TNF-α plays a major role in the regulation of bone homeostasis by stimulating osteoclastogenesis and inhibiting osteoblast (OB) function. It is also involved in the pathogenesis of chronic inflammation in both mouse and human models. Based on these observations, treatments of several chronic inflammatory diseases such as Crohn's disease, ulcerative colitis, and psoriasis target TNF-α. TNF-α is also associated with the pathogenesis of inflammatory joint diseases including rheumatoid arthritis (RA), which is characterized by massive juxta-articular bone destruction, and ankylosing spondylitis (AS), which is characterized by simultaneous bone destruction and excessive formation, respectively. However, opposite findings suggest that TNF-α may also induce osteogenic differentiation. This review summarizes the classical and paradoxical effects of TNF-α on bone homeostasis by focusing on osteoblastogenesis and osteoclastogenesis.

TNF-α AND ITS SIGNALING PATHWAYS
The immune response is modulated by several inflammatory cytokines including TNF-α. TNF-α acts through several pathways including the activation of nuclear factor kappa-B (NF-κB) involved in inflammation and apoptosis (1). TNF-α is produced mainly by monocytes, but can also be secreted by many other cell types (1)(2)(3). It is produced as a transmembrane protein (mTNFα) that is subsequently cleaved by the TNF-α converting enzyme (TACE) to a soluble form (sTNF-α) (4). Both types are biologically active and bind as a trimer to either TNF receptors TNFR1 (also known as TNFRSF1A or p55) or to TNFR2 (also known as TNFRSF1B or p75).
Activation of TNFR1 death domain initiates the apoptotic signaling through the activation of several factors such as caspase and NF-κB resulting in transcription of pro-inflammatory cytokines, chemokines, and anti-apoptotic molecules (5). Unlike TNFR1, the cell expression of TNFR2 is far more limited and does not contain a death domain. Its effects include T cell activation and proliferation via signaling pathways that involve NF-κB, activator protein 1 (AP-1), and the mitogen-activated protein kinases (MAPKs) (6). The pro-inflammatory effects of this cytokine are well documented (7) and are involved in many osteo-immunological diseases including RA and AS (8).
In humans, TNF-α antagonists are clearly effective in a large number of conditions, as first shown in patients with RA (9, 10). However, incidents of lupus-like syndrome and skin lesions have been reported in patients receiving anti-TNF therapies (11). The cellular and molecular mechanisms underlying the TNF-αmediated pro-inflammatory effects and the occasionally paradoxical effects of TNF-α antagonists in chronic inflammatory diseases remain incompletely understood (10). This also applies to bone biology.

BONE BIOLOGY
Bone structure is affected by several paracrine and endocrine factors including mechanical abrasion and serum electrolyte levels such as calcium and phosphorus. This remodeling is regulated by several cells such as osteoclasts (OCs), OBs, and osteocytes (OYs) (12). This continuous remodeling prevents and removes fatigue-related micro-damage and allows adaptation of the bone mass and structure. The number and activity of OCs and OBs are determined by a multitude of factors such as hormones, cytokines, and locally produced signaling molecules under the influence of mechanical stimuli (13)(14)(15). Among these factors, TNF-α was found to have great effects on both bone regulator cells, OBs and OCs.

BIOLOGY OF OSTEOCLASTS
Osteoclasts are hematopoietic cells derived from the monocyte lineage. They undergo a series of differentiation steps to become mature bone-resorbing cells. Intracellular signaling cascade of transcriptional regulation of osteoclast-specific genes is activated by macrophage-colony-stimulating factor (M-CSF) and receptor activator of nuclear factor kappa-B ligand (RANKL) (16) (Figure 1). Upon differentiation, OC precursor cells express strongly RANK, the receptor for RANKL.

BIOLOGY OF OSTEOBLASTS
Osteoblasts are derived from pluripotent stem cells called "mesenchymal stem cells (MSCs)" capable of differentiating into adipocytes, chondrocytes, fibroblasts, skeletal muscle cells, and tendon cells (17). A number of factors can stimulate the differentiation toward an osteoblastic direction. The first step includes the proliferation of osteoprogenitors leading to the production of proteins such as histones, fibronectin, and type I collagen. Subsequently, osteogenic matrix is formed through the expression of genes responsible for the production of alkaline phosphatase (ALP), bone sialoprotein, and type I collagen. Finally, extracellular matrix is mineralized by activation of genes involved in the release of osteocalcin (OCN), osteopontin, and collagenase. This complex cascade of gene expression and cellular differentiation is orchestrated by several transcription factors including the runt-related transcription factor-2 (Runx2) and surface expression of RANKL (Figure 2A).

BIOLOGY OF OSTEOCYTES
Osteocytes are derived from MSCs through OB differentiation. They are located within the bone matrix and represent 90-95% of all bone cells in adults (18). In addition, they have the longest life span among bone cells, up to decades, within their mineralized environment (19). OYs are involved in the dynamics of bone structure through expression of different markers that are released during OB-to-OY ontogeny. In addition, OYs affect the function of other organs such as phosphate transport in kidney (20). Furthermore, OYs are now known to be the principal sensors for mechanical loading of bone. They translate the canalicular flow resulting from bone loading into OCs and OBs recruiting signals (21). Due to the high expression of RANKL by OYs, they have a more significant contribution to osteoclastogenesis as compared to OBs and bone MSCs (22,23).

EFFECTS OF TNF-α ON OSTEOBLASTS
Previous in vitro studies have indicated an inhibitory effect of TNF-α on OB differentiation ( Table 1). This blockage of osteogenic differentiation resulted from the inhibition of insulinlike growth factor-1 (IGF-1), osterix (Osx also known as SP7), and Runx2 (24)(25)(26). The inhibition of Runx2 seemed to be mediated by the TNF-induced up-regulation of Smurf1, a negative regulator of OB differentiation that causes the degradation of Runx2 (27). Moreover, TNF-α inhibited MSC differentiation into OBs via the ubiquitin protein ligase Wwp1 in TNF transgenic mice expressing human TNF-α (28). Another human study demonstrated an inhibitory effect of TNF-α on Runx2 and collagen expression. However, an increase in alkaline phosphatase activity and matrix mineralization was noted (29).
In full contrast with these findings, other murine and human MSC studies have reported that TNF-α can also induce osteogenic differentiation ( Table 2). In three in vitro rodent models, low concentrations of TNF-α increased osteogenic differentiation via an up-regulation of Runx2, Osx, OCN, and ALP levels (38,39). Moreover, four human experimental models indicated a similar osteogenic activity for TNF-α. In these models, osteogenic differentiation was promoted via induction of BMP-2, Osx, Runx2, and OCN (40)(41)(42)(43). In addition, human models revealed that the dual Frontiers in Immunology | Inflammation role of TNF-α on osteogenic differentiation was directly dependent on the concentration of TNF-α, the cell type, and the exposure time ( Table 2). Furthermore, a recent study showed that TNF-α stimulated the expression of Wnt5a, which was directly associated with an increase in tissue non-specific alkaline phosphatase (TNAP) levels and mineralization. This suggests an autocrine stimulation of OB activity by Wnt5a in response to TNF in hMSCs (44). This paradoxical effect of TNF-α in inhibiting or activating osteoblastogenesis, lies in the differentiation stage of the responding cells. MSCs can differentiate into several lineages. At this early stage, TNF-α binds its receptors and favors osteogenic differentiation through activation of multiple signaling pathways, notably NF-kB (40) (Figure 2B) ( Table 2). On the other hand, TNF-α inhibits osteoblastogenesis in the presence of pre-OB cells, which are already on the differentiation process. This inhibition occurs at different levels, where TNF-α induces DKK-1 expression that inhibits Wnt pathway (45), or activates Smurf that inhibits BMP-2 pathway (31) (Figure 2A) ( Table 1).

EFFECTS OF TNF-α ON OSTEOCYTES
Limited results are available on the effect of TNF-α on OYs, which are known to be affected by the surrounding environment. TNF-α and interleukin-1 (IL-1), which increase with estrogen deficiency, can induce OYs apoptosis (48). In a fluid shear stress study, in which the mechanical bone loading was mimicked by applying pulsating fluid flow, the TNF-α-induced apoptosis observed in OYs was inhibited by mechanical loading. However, this effect was not observed in OBs and periosteal fibroblasts. Since apoptotic OYs attract OCs, these results suggest a key role for OYs apoptosis in osteoclastic bone resorption during bone remodeling that is in part modulated by TNF-α (49). Additionally, TNF-α was shown to inhibit the increase in nitric oxide (NO) production and intracellular calcium while strongly reducing F-actin content. This resulted in a reduction of OYs stiffness and provided a possible mechanism to explain the contribution of inflammation to loss of bone mass (50). www.frontiersin.org

EFFECTS OF TNF-α ON OSTEOCLASTS
The role of TNF-α as a stimulator of osteoclastogenesis is well established (51)(52)(53)(54)(55). The expression of a number of transcription factors, including NF-κB, is critical for osteoclastogenesis (56, 57) (Figure 1). Early bone marrow progenitors embark on a path toward pre-OCs or monocytes-macrophages under the influence of M-CSF while TNF-α, IL-1, and RANKL promote a progression toward the functional OCs phenotype. Thus, both RANKL and TNF-α are not only required for, but also synergize to induce osteoclastogenesis. The requirement for both TNF-α and RANKL allows for a more precise control of OCs numbers through a dual level of regulation. First, they affect the selective signaling of the precursor cells via the two different receptors, and subsequently their synergy further increases the activation of NF-κB and AP-1 signaling. Furthermore, the effect of the proinflammatory cytokines TNF-α, IL-17, IL-6, IL-1β, and IL-23 on osteoclastogenesis in vitro demonstrated that they have specific characteristic osteoclastogenic properties. However, they exhibited osteoclastogenesis-related activity only in the presence of permissive levels of RANKL. TNF-α appears to have osteoclastogenic properties in the early stage of osteoclastogenesis when bone marrow-derived macrophages (BMMs) are at the stage of osteoclast precursor cells (58).

EFFECTS OF TNF-α ON WHOLE BONE
These results on isolated cells need to be expanded to whole bone studies in order to reproduce the cell-cell interactions between bone specific cells and immune cells in the bone marrow. Only a few studies have examined the effects of TNF alone or combined to other cytokines such as IL-1 and IL-17 in ex vivo models or explants where both osteoclast and OB cells are present. In a RA ex vivo model of bone resorption, an IL-4-mediated inhibition of TNF-α levels corroborated with an inhibition in bone resorption, manifested as a 35% increase in the mean total bone area with IL-4 (59). Another study with RA-derived bone explants showed that inhibition of TNF-α decreased inflammation as measured by levels of IL-6 and bone resorption markers. These inhibitory effects were enhanced in combination with IL-1 and IL-17 blockade (60). Histomorphometric analysis of bone samples in patients with different diseases affecting bone such as RA, Crohn's disease, and bronchial asthma showed that eroded surface over bone surface (ES/BS), a parameter of bone resorption, was significantly increased in the context of high TNF-α levels (61). On the contrary, one study showed that addition of TNF-α to a cultured fetal mouse long bone explant induced NO production resulting in suppression of osteoclast bone resorption (62). These findings suggest a mixed effect of TNF-α on bone structure, Frontiers in Immunology | Inflammation  10 ng/ml TNF-α triggers osteogenic differentiation of human dental pulp stem cells via the NF-κB signaling pathway (47) which corroborates with the paradoxical role of TNF-α observed in vitro.

EFFECTS OF TNF-α IN VIVO
Tumor necrosis factor-α disrupts bone homeostasis by activating osteoclastic resorption, inhibiting osteoblastic proliferation and matrix synthesis, and activating TNF receptor-associated Factor-2 (TRAF-2). The latter activates NF-κB, AP-1, and MAPKs signaling pathways, which leads to decreased bone formation. In human TNF transgenic mice, TNF-α induced DKK-1, a regulatory molecule of the Wnt pathway, which led to impaired local bone formation. In addition, blockade of DKK-1 interfered with local bone resorption by reducing OCs numbers in the joints of animal models with inflammatory arthritis (45,63,64). A recent study of collagen-induced arthritis (CIA) in rats showed that TNF inhibition did not significantly affect RANKL, TNF-α, and OPG mRNA expression, which are linked to osteoclastogenesis, while it promoted bone formation via suppression of DKK-1 expression. This was confirmed by histomorphometric analysis, which demonstrated a significant increase in bone formation parameters while the changes in bone destruction parameters were not significant (63). Clinical trials in human RA confirmed the association of TNF with arthritic bone destruction as observed in several human TNF-α transgenic mouse models.
However, this observation was not clear in TNF-deficient mouse models, due to insignificant reduction in the clinical parameters of CIA and absence of major histological and morphological abnormalities in the joints. Moreover, some mice exhibited severe disease in the absence of TNF, suggesting the contribution of other cytokines such as IL-1 (65-68).

TNF-α AND INFLAMMATORY JOINT DISEASES
Tumor necrosis factor-α has been associated with an increasing number of chronic inflammatory diseases. Among these diseases, the two most common inflammatory joint diseases RA and AS will be discussed. RA causes massive destruction of juxta-articular and systemic bone while AS leads to the concomitant bone destruction and ectopic bone formation (especially at the vertebral site) because of inflammation of tendon/ligament insertions. Thus, patients with AS experience stiffness of the spine known as ankylosis, and functional disability (69)(70)(71). Inhibition of TNF-α has substantially advanced the treatment of these two inflammatory diseases.

RHEUMATOID ARTHRITIS
Rheumatoid arthritis is a chronic disease characterized by joint inflammation leading to destruction of cartilage and bone (71). In RA, focal bone loss is due to an excessive bone resorption by OCs (72). Moreover, bone formation by OBs is greatly impaired at the erosion site in RA (73). TNF-α plays an important role in the bone destruction observed in RA patients. TNF inhibitors delay appearance of bone erosion in RA patients with no progression of bone destruction in responders and possibly to some extend, in nonresponders (9,(74)(75)(76). In a murine CIA model, neutralization of www.frontiersin.org TNF-α reduced joint pathology, especially when the treatment was started early after the onset of disease (77). Moreover, simultaneous inhibition of TNF-α and IL-17A further improved protection against joint damage even at the late stages of CIA (78). Mechanistic models of human TNF transgenic mice confirmed that ongoing TNF-driven inflammation suppresses new bone formation by Wnt signaling through up-regulation of DKK-1, and then the release of this inhibition allows new bone formation (45). Furthermore, several murine autoimmune arthritis models provided evidence that TNF deficiency reduces disease incidence (79)(80)(81).
The pro-inflammatory influence of TNF-α in RA acts at several levels. It induces other pro-inflammatory cytokines such as IL-1 and IL-17 and regulates osteoclastogenesis via several mechanisms. TNF-α accelerates cartilage destruction through inducing the production of matrix metalloproteinases (MMPs) and aggrecanases (ADAMTSs), which are enzymes that degrade components of the cartilage extracellular matrix. As an example, in human RA patients and in the human TNF-α transgenic mouse model, TNFα was shown to regulate MMP3 and 13 by the small ubiquitin-like modifier-1 (SUMO-1) through NF-κB, which led to bone destruction (82). In addition, TNF-α stimulates OCs maturation both directly and indirectly by inducing RANKL on synovial fibroblasts, the synovial lining cells, and T-cells in the synovium (52,(83)(84)(85). This increased OCs maturation, and in turn, osteoclastic bone resorption. Moreover, TNF-α possibly promotes OCs activity via enhancing oxidative stress. The production of reactive oxygen species (ROS) by OCs helps accelerate the destruction of calcified tissue (86). This was confirmed in RA patients using an anti-TNFα antibody therapy. In these patients, reduced levels of oxidative stress markers were associated with the control of joint and bone damage (87). Thus, TNF-α in RA has a destructive function as net effect.

ANKYLOSING SPONDYLITIS
Ankylosing spondylitis is the second most common form of chronic inflammatory arthritis. The unique hallmark of the disease is pathologic new bone formation (88). AS and related spondyloarthritis are characterized by a paradoxical and simultaneous bone destruction and formation occurring in proximal anatomical sites. The strongest evidence for a key role of TNF-α in AS pathophysiology comes from in vivo inhibition of TNF in AS patients. TNF blockers reduce signs and symptoms of AS, which translates into better physical function and quality of life. However, little or no effect on structural remodeling is achieved (89,90).
In AS, relevant human samples are not readily available even so TNF expression has been shown at the site of sacroiliitis. Thus, most observations are only made in rodent models. Recently, mice that specifically overexpress the transmembrane but not the soluble form of TNF-α, so called tmTNF transgenic mice, developed clinically spinal abnormalities similar to those in AS (91). These mice developed pronounced local inflammatory infiltration of the axial and peripheral joints (88). In sharp contrast with the other TNF overexpression models, these mice do not develop systemic inflammation, yet display pronounced axial and peripheral new bone formation leading to ankylosis (92). This model strongly argues for a role of tmTNF in the pathogenesis of AS but the exact molecular mechanisms still need to be elucidated. AS patients are particularly prone to develop syndesmophytes, a bony growth originating inside a ligament, and enthesitis, an inflammation at the sites where tendons or ligaments insert to bone. Therefore, in these settings, OCs may not be in direct contact with tendonderived OBs and TNF-α may exhibit an osteogenic effect there. This is different from the effect found in whole bone where massive bone loss is seen. Thus in AS, even with TNF-α inhibition, ectopic bone formation is still observed, meaning that additional factors are governing this effect.

OSTEOPOROSIS
Osteoporosis is characterized by a decrease in bone mass and density leading to bone weakness and prone to fracture. In RA patients, several studies have shown an increased incidence of osteoporosis. This could be due to the highly systemic inflammatory environment observed in these patients and the use of glucocorticoid treatment, which can affect bone remodeling (93,94). Despite ectopic bone formation, AS patients show massive systemic osteoporosis (95). Among other cytokines, TNF-α seems to be an important player implicating in osteoporosis via TNFR1 (96). In RA patients treated with TNF-α inhibitors, control of bone loss was observed in parallel to that seen on arthritis levels (76,97).

CONCLUSION
Tumor necrosis factor-α has an important role in the regulation of bone homeostasis via activation of complex signaling pathways leading to gene transcription of several regulators of bone homeostasis. The classical view remains that TNF-α is an established inducer of osteoclastogenesis and inhibitor of osteoblastogenesis. To add further complexity, it was recently discovered that TNFα also possessed osteogenic differentiation effects. Most of these paradoxical effects were only described in in vitro models and were concentration-and time-dependent, and cell-type specific. Despite the fact that TNF-α levels are high in both RA and AS, the bone dynamics are different. In RA, the dynamics are shifted toward bone destruction, combined with inhibition of specific osteogenic genes. In AS, systemic bone destruction is accompanied by ectopic bone formation at sites where OCs and OBs are disconnected. Future research should focus on the delineation of the molecular pathways used by this mediator and understand the conditions affecting and the purpose of this dual role in bone remodeling.