Osteoimmunology and osteoporosis

The concept of osteoimmunology is based on growing insight into the links between the immune system and bone at the anatomical, vascular, cellular, and molecular levels. In both rheumatoid arthritis (RA) and ankylosing spondylitis (AS), bone is a target of inflammation. Activated immune cells at sites of inflammation produce a wide spectrum of cytokines in favor of increased bone resorption in RA and AS, resulting in bone erosions, osteitis, and peri-inflammatory and systemic bone loss. Peri-inflammatory bone formation is impaired in RA, resulting in non-healing of erosions, and this allows a local vicious circle of inflammation between synovitis, osteitis, and local bone loss. In contrast, peri-inflammatory bone formation is increased in AS, resulting in healing of erosions, ossifying enthesitis, and potential ankylosis of sacroiliac joints and intervertebral connections, and this changes the biomechanical competence of the spine. These changes in bone remodeling and structure contribute to the increased risk of vertebral fractures (in RA and AS) and non-vertebral fractures (in RA), and this risk is related to severity of disease and is independent of and superimposed on background fracture risk. Identifying patients who have RA and AS and are at high fracture risk and considering fracture prevention are, therefore, advocated in guidelines. Local peri-inflammatory bone loss and osteitis occur early and precede and predict erosive bone destruction in RA and AS and syndesmophytes in AS, which can occur despite clinically detectable inflammation (the so-called 'disconnection'). With the availability of new techniques to evaluate peri-inflammatory bone loss, osteitis, and erosions, peri-inflammatory bone changes are an exciting field for further research in the context of osteoimmunology.

system in isolation but should consider bone and the immune system to be an integrated whole [4,5].

Anatomical connections
Bone, by virtue of its anatomy and vascularization, is at the inside and outside and is in direct and indirect and in close and distant contact with the immune system. At the inside, bones are the host for hematopoiesis, allowing bone and immune cells to cooperate locally. At the outside, bone is in direct contact with the periost, the synovial entheses within the joints at the periost-and cartilage-free bare area [7], the fi brous tendon entheses, the calcifi ed component of cartilage and tendon insertions, and the intervertebral discs.
Until recently, it was thought, on the basis of plain radiographs of the hands, that there is only rarely a direct anatomical connection between bone marrow and joint space. Bone erosions have been found in hand joints of presumably healthy controls in less than 1% with plain radiology and in 2% with MRI [8]. However, exciting new data have shown that, with the use of high-resolution quantitative computer tomography (HRqCT), small erosions (<1.9 mm) in the metacarpophalangeal (MCP) joints can be found in 37% of healthy subjects without any signs or symptoms of RA, indicating that small erosions are not specifi c for RA [9]. Large erosions (>1.9 mm) were found to be specifi c for RA. Interestingly, 58% of erosions detected by HRqCT in healthy volunteers were not visible on plain radiographs [9]. In healthy controls, the erosions in the MCP joints were not randomly located but were located at the bare area and at high-pressure points adjacent to ligaments, which are erosion-prone sites in RA [10]. Bone erosions are also extremely common in healthy controls in the entheses [11] and in the vertebral cortices covered by periost and the intervertebral discs (in AS) [12]. Th e immune system, bone, and its internal and external surfaces not only are connected by these local anatomical connections but also are connected with the general circulation by the main bone nutrition arteries and locally with the periost (by its vasculature that perforates cortical bone) and within the bone compartment by attachments of fi brous entheses and the calcifi ed components of cartilage and fi brocartilage up to the tidemark, which separates calcifi ed from non-calcifi ed components of cartilage and tendons [11].

Molecular connections
Bone cells exert major eff ects on the immune system. Bone cells interact with immune cells and play an essential role in the development of the bone marrow space during growth [13] and during fracture healing [14]. Osteoblasts play a central role in the regulation of renewal and diff erentiation of hematopoietic stem cells (HSCs) and of B cells in niches near the endosteum [15][16][17]. Metabolic pathways of the osteoblast which are involved in bone remodeling are also involved in the regulation of HSCs by osteoblasts, such as the calcium receptor, parathyroid hormone (PTH), bone morphogenetic proteins (BMPs), the Wnt signaling, and cell-cell interactions by the NOTCH (Notch homolog, translocation-associated (Drosophila)) signaling pathway [15][16][17][18][19]. On the other hand, multiple cytokines, chemokines, and growth factors of immune cells such as T and B cells, fi broblasts, dendritic cells, and macrophages directly or indirectly regulate osteoblast and osteoclast activity by producing or infl uencing the production of the RANKL/RANK/OPG pathway, tumor necrosis factor-alpha (TNFα), interferon-gamma (IFNγ), and inter leukins (such as IL-1, IL-6, IL-15, IL-17, IL-18, and IL-23) and the Wnt signaling with involvement of Dikkoppf (DKK), sclerostin, and BMP [4,5,[19][20][21].
In AS, increased bone formation, as refl ected by syndesmophyte formation in the spine, is related to decreased serum levels of DKK [25] and sclerostin [21], both inhibitors of bone formation, and to serum levels of BMP, which is essential for enchondral bone formation [26], and of CTX-II [27], which refl ects cartilage destruction that occurs during enchondral bone formation in syndesmophytes [26][27][28]. Th ere is, thus, increasing evidence that immune cells and cytokines are critically responsible for the changes in bone resorption and forma tion and vice versa, resulting in changes in bone quality in chronic infl ammatory conditions. Th ese conditions include RA, spondylarthopathies (SpAs) (AS, psoriatic arthritis, and infl ammatory bowel disease), systemic lupus erythematosis, juvenile RA, periodontal diseases, and even postmenopausal osteoporosis [29]. We reviewed the literature on the quantifi cation of bone involvement in RA and AS. For an in-depth discussion of the underlying metabolic pathways, a topic that is beyond the scope of this review, the reader is referred to other reviews [4,5].

Bone resorption
Bone resorption is increased in RA and AS. In RA, this has been demonstrated histologically by the presence of activated osteoclasts in the pannus at the site of bone erosions [30,31], in the periarticular trabecular and cortical bone [32,33], and, in a general way, in sites distant from infl ammation [34]. In AS, osteoclastic bone resorption has been demonstrated in the sacroiliac joints [35][36][37].
Th e introduction of MRI has shed new light on the involvement of subchondral bone and bone marrow in RA and AS ( Figure 1). Periarticular MRI lesions have been described technically as bone edema (on short T inversion recovery (STIR), indicating that fatty bone marrow is replaced by fl uid) and osteitis (on T1 after IV gadolinium) [38] and histo logi cally as osteitis as infl ammation has been demon strated on histological examination of these lesions [33]. In joint specimens of patients with RA and with MRI signs of bone edema, histological correlates have been studied in specimens obtained at the time of joint replacement and have shown the presence of greater numbers of osteo clasts than in controls and in patients with osteoarthritis and the presence of T cells, B-cell follicles, plasma cells, macrophages, decreased trabecular bone density, and increased RANKL expression [33].
Osteitis is also a major component of AS [39][40][41][42]. Osteitis was described by histology of the vertebrae in 1956 [43] and occurs early in the disease and predicts the occurrence of bone erosions [39]. It has been shown that, as in RA, these lesions contain activated immune cells and osteoclasts [44,45]. In contrast to RA, these lesions diff er in their location: in the vertebrae, the entheses, the periost of vertebrae and around the joints, the discovertebral connections, the intervertebral joints and the sacroiliac joints, and, to a lesser degree, the peripheral joints, mainly hips and shoulders ( Figure 1) [46,47].

Bone formation
In spite of the presence of cells with early markers of osteoblasts in and around erosions in RA, bone formation is locally suppressed [48]. Th is uncoupling of bone resorption and bone formation contributes to the only rare occurrence of healing bone erosions [49] and results in persisting direct local connections between the joint cavity and subchondral bone and thus between synovitis and osteitis. In contrast, in AS, local peri-infl ammatory bone formation is increased, resulting in healing of erosions, ossifying enthesitis, and potential ankylosis of sacroiliac joints and of intervertebral connections. Th e ossifi cation of entheses and sacroiliac joints involves calcifi cation of the fi brocartilage, followed by enchondral bone formation; that is, calcifi ed cartilage is replaced by bone through osteoclastic resorption of calcifi ed cartilage and deposition of bone layers on the inside of the resorption cavity with a very slow evolution and with prolonged periods of arrest [50].

Bone biomarkers
In patients with RA, markers of bone resorption are increased in comparison with controls [51]. Correlations between bone markers, bone erosions, and bone loss in RA varied according to study designs (cross-sectional or longitudinal), patient selection, and study endpoints (disease activity score, radiology, and MRI) [52]. Baseline markers of bone and cartilage breakdown (CTX-I and CTX-II) and the RANKL/OPG ratio were related to  short-and long-term (up to 11 years for RANKL/OPG) progression of joint damage in RA, independently of other risk factors of bone erosions [53,54]. Increased markers of bone resorption were related to increased fracture risk [49]. Studies on markers of bone formation in RA, such as osteocalcin, are scarce and show contradictory results, except low serum values in glucocorticoid (GC) users [55,56].
In AS, markers of bone resorption were increased [27,57] and were related to infl ammation as measured by serum IL-6 [58]. Increased serum levels of RANKL have been reported [59] with decreased OPG [60,61], and RANKL expression is increased in peripheral arthritis of SpA [62]. Markers of bone formation (type I collagen Nterminal propeptide, or PINP) were related to age, disease duration, and markers of bone resorption (CTX-I) but not with low BMD in the hip or spine [63]. Markers of cartilage breakdown (CTX-II) were related to progression of the modifi ed Stoke Ankylosing Spondylitis Spine Score (mSASSS) and the appearance of syn despomphytes [27].

Imaging of bone in rheumatoid arthritis and ankylosing spondylitis
Many methods, including histomorphometry, imaging ( Figure 2), and biomarkers, have been used to study the eff ect of infl ammation on structural and functional aspects of bone in RA and AS. Conventional radiology of the peripheral joints and the spine is used for identifying erosions, joint space narrowing, enthesitis, and syn desmo phytes for diagnosis; assessment of disease progression; and standardized scoring in clinical trials, but it is estimated that bone loss of less than 20% to 40% cannot be detected on plain radiographs [64].
Methods that quantify changes in periarticular bone include radiogrammetry, digitalized radiogrammetry (DXR) [65], peripheral dual-energy x-ray absorptiometry (pDXA) [66], quantitative ultrasound (QUS) [67], highresolution digital radiography [68], high-resolution peripheral qCT [9], and MRI [8], and methods that quantify changes in the vertebrae include DXA, qCT, MRI, and morphometry by vertebral fracture assessment on x-rays or DXA images [69] (Figure 2). At other sites of the skeleton, single x-ray absorptiometry, qCT, MRI, DXA, and QUS are available; of these, DXA is considered the gold standard [70]. Semiquantitative scoring of osteitis on MRI in the vertebrae has been standardized [40,42,71]. Local peri-infl ammatory bone formation can be evaluated semiquantitatively in a standardized way on radiographs for scoring of syndesmophytes [41,42,72]. Th ese techniques diff er in regions of interest that can be measured, in the ability to measure cortical and trabecular bone separately or in combination, and in radiation dose, cost, and precision [64,73] (Table 1).

Periarticular bone loss and osteitis in rheumatoid arthritis
On plain radiographs of the hands, periarticular trabe cular bone loss results in diff use or spotty demineralization and blurred or glassy bone and cortical bone loss in tunneling, lamellation, or striation of cortical bone [74] ( Figure 3). Quantifi cation of bone in the hands has consis tently shown that patients with RA have lower BMD than controls and lose bone during follow-up, depending on treatment (see below) [75][76][77]. Cortical bone loss occurs early in the disease, preferentially around aff ected joints and before generalized osteoporosis can be detected [51,78]. In studies using peripheral qCT at the forearm, trabecular bone loss was more prominent than cortical bone loss in RA patients using GCs [79,80].
Hand bone loss is a sensitive outcome marker for radiological progression. Th e 1-year hand bone loss measured by DXR predicted the 5-and 10-year occurrence of erosions in RA [73,81] and was a useful predictor of the bone destruction in patients with early unclassifi ed polyarthritis [82]. Hand bone loss measured by DXR correlated with C-reactive protein (CRP), erythrocyte sedimentation rate (ESR), disease activity score using 28 joint counts (DAS28), the presence of rheumatoid factor (RF) and anti-cyclic citrullinated peptide antibody (anti-CCP), health assessment questionnaire (HAQ) score, disease duration, and Sharp score [66,83,84]. In the forearm and calcaneus, trabecular but not cortical periarticular bone loss measured by DXA in early RA correlated with ESR, CRP, RF, and HAQ score [80]. DXR correlated with hip BMD and the presence of morphometric vertebral fractures and non-vertebral fractures in RA [85]. DXR-BMD performed as well as other peripheral BMD measurements for prediction of wrist, hip, and vertebral fractures in the Study of Osteoporotic Fractures [86].
Patients with RA have a decreased BMD in the spine and hip and consequently have a higher prevalence of osteoporosis [56,[97][98][99][100][101]. However, this was not confi rmed in the Canadian Multicentre Osteoporosis Study (CaMos) [102]. In early untreated RA, BMD was related to longer symptom duration, the presence of RF [103] and anti-CCP [104], disease activity score [105], and the presence and progression of joint damage [106].
Th e interpretation of longitudinal changes in RA is complicated by the lack of untreated patients, and this limits our insights into the natural evolution of bone changes in RA to the above-mentioned studies. In one study with early untreated RA, bone loss was found in the spine and trochanter for a period of one year [107]. However, Kroot and colleagues [108] did not fi nd bone loss over the course of a 10-year follow-up in RA patients treated with disease-modifying anti rheumatic drugs, except when these patients were treated with GCs. Generalized bone loss was related to joint damage in some studies [109,110], but this relation dis appeared after multivariate adjustment [111]. No correla tion between BMD and the presence of vertebral frac tures in RA patients treated with GCs was found [112].

Fracture risk in rheumatoid arthritis
In the largest epidemiological study, patients with RA were at increased risk for fractures of osteoporotic fractures (relative risk (RR) 1.5), fractures of the hip (RR 2.0), clinical vertebral fractures (RR 2.4), and fractures of the pelvis (RR 2.2) [113]. Th e risk of morphometric vertebral fractures was also increased [114,115]. In some but not all studies, the risk of fractures of the humerus (RR 1.9), wrist (RR 1.2), and tibia/fi bula (RR 1.3) was increased [75,116,117].
Th e etiology of increased fracture risk in RA is multifactorial and superimposed on and independent of BMD and other clinical risk factors for fractures, including the use of GCs. RA is included as an independent clinical risk factor for 10-year fracture risk calculation for major and hip fractures in the fracture risk assessment tool (FRAX) case-fi nding algorithm [118]. Stress fractures have been found in 0.8% of patients with RA, can be diffi cult to diagnose, and were related to GC use but not to BMD [119].
Fracture risk in RA was related to the duration of RA [120], the severity of disease, and its musculoskeletal conse quences, such as disability, HAQ score, lack of physical activity, and impaired grip strength [120][121][122]. Vertebral fractures were related to disease duration and severity [69]. In the general population, fracture risk was related to serum levels of IL-6, TNF, and CRP [123] and parameters of bone resorption [124], all of which can be increased in RA. Extraskeletal risk factors that infl uence fracture risk include increased risk of fall rates which were related to number of swollen joints and impaired balance tests [125].

Risk predictors of bone changes in rheumatoid arthritis
Currently, the most widely used case-fi nding algorithm for calculating the 10-year fracture risk for major and hip fractures is the FRAX tool [118]. FRAX includes RA as a risk for fractures, independently of and superimposed on other risk factors, including BMD and use of GCs [118]. No fracture risk calculator that also includes other risk factors that are related to RA, such as disease duration and disease severity, is available. Th e Garvan fracture risk calculator (GFRC) can be used to calculate the 5-and 10-year fracture risk which includes the number of recent falls and the number of previous fractures but lacks RA as a risk factor [126]. Fracture risk is higher with GFRC than with FRAX in patients with recent falls [126]. In view of the increased fracture risk in patients with RA, systematic evaluation of fracture risk should be considered using FRAX, disease severity, and duration, and GFRC is helpful when patients report recent falls. Risk of low BMD is diffi cult to estimate in RA [90], and this suggests that bone densito metry should also be considered in fracture risk calculation in patients with active RA [127]. Many risk factors, including baseline disease severity, RF, anti-CCP, baseline bone destruction, the RANKL/OPG ratio, and CTX-I and CTX-II, have been identifi ed for the prediction of bone erosions in RA. Th is pallet of predictors can now be extended with measurement of changes in periarticular bone (by DXR) and osteitis (on MRI) early in the disease [73,81,82]. Additional studies will be necessary to study the relation between osteitis and bone loss.

Eff ect of treatment on bone changes in rheumatoid arthritis
As the pathophysiology of bone loss in RA is taken into account (Figure 4), therapy should be directed at suppressing infl ammation and bone resorption and restoring bone formation. No randomized placebo-controlled trials (RCTs) on the eff ect of treatment on fracture risk in RA are available. However, the available data suggest that control of infl ammation (TNF blockade and appropriate dose of GCs), specifi c inhibition of bone resorption (bisphosphonates and denosumab), strontium ranelate, and restoration of the balance between bone resorption and formation (teriparatide and PTH) are candidates for such studies. Bone loss early in the disease continued despite clinical improvement and suffi cient control of infl ammation through treatment, indicating a disconnect between clinical infl ammation and intramedullary bone loss [128]. However, these studies did not include TNF blockers, and, at that time, remission was not a realistic tool of therapy. Suppression of infl ammation with TNF blockers such as infl iximab and adalimumab decreased markers of bone resorption and the RANKL/OPG ratio [129], decreased osteitis, and reduced or arrested generalized (in spine and hip) bone loss [75]. Infl iximab, however, did not arrest periarticular bone loss [129]. In the Behandelstrategieën voor Reumatoide Artritis (BEST) study, both bone loss at the metacarpals and radiographic joint damage were lower in patients adequately treated with combination therapy of methotrexate plus highdose prednisone or infl iximab than in patients with suboptimal treatment [130].
Several pilot studies on the eff ect of antiresorptive drugs on bone in RA have been performed. Pamidronate reduced bone turnover in RA [131]. Zoledronate decreased the number of hand and wrist bones with erosions [132]. Denosumab strongly suppressed bone turnover and, in higher dosages than advocated for the treatment of postmenopausal ostepororotic women, prevented the occurrence of new erosions and increased BMD in the spine, hip, and hand, without an eff ect on joint space narrowing and without suppressing infl ammation, indicating an eff ect on bone metabolism but not on cartilage metabolism [133][134][135][136].
Th e eff ects of GCs on bone loss and fracture risk in RA should be interpreted with caution as GCs have a dual eff ect on bone in RA. On the one hand, controlling infl am mation with GCs strongly reduces bone loss, whereas, on the other hand, GCs enhance bone resorption, suppress bone formation, and induce osteocyte apoptosis.
Studies in glucocorticoid-induced osteoporosis (GIOP) included patients with RA. None of these studies had fracture prevention as a primary endpoint, and no data on the GIOP studies on fracture prevention in RA separately are available (see [137] for a recent review). RCTs in GIOP showed that bisphosphonate treatment (alendronate, risedronate, and zoledronate) and teripara tide prevented bone loss and increased BMD. Alendro nate and risedronate decreased the risk of vertebral fractures versus placebo and teriparatide versus alendro nate. No convincing evidence on fracture risk in GIOP for calcium and vitamin D supplements (calcitriol or alfacalcidol) is available. However, most RCTs in GIOP provided calcium and vitamin D supplements. Most guide lines, therefore, advocate calcium and vitamin D supple ments, bisphosphonates, and eventually teripara tide as a second choice because of its higher cost price in the prevention of GIOP in patients at high risk, such as those with persistent disease activity, high dose of GCs, or high background risk such as menopause, age, low BMD, and the presence of clinical risk factors [138,139].
Taken together, these data indicate that control of infl am mation is able to halt bone loss and suppress osteitis in RA. Bisphosphonates are the front-line choice for fracture prevention in GIOP, but in patients with a very high fracture risk, teriparatide might be an attractive alternative. Th e eff ect of denosumab indicates that osteoclasts are the fi nal pathway in bone erosions and local and generalized bone loss and that the bone destruction component of RA can be disconnected from infl am mation by targeting RANKL.

Generalized bone loss in ankylosing spondylitis
Bone loss in the vertebrae occurs early in the disease, as shown by DXA [140] and qCT [141]. In advanced disease, the occurrence of syndesmophytes and periosteal and discal bone apposition does not allow intravertebral bone changes with DXA to be measured accurately. Combined analyses of DXA and QCT in patients with early and long-standing disease indicate that bone loss in the vertebrae occurs early in the disease and can be measured by DXA and QCT but that, in long-standing disease, DXA of the spine can be normal, in spite of further intravertebral bone loss as shown with qCT [142,143]. As a result, in early disease, osteoporosis was found more frequently in the spine than in the hip, whereas in patients with long-standing disease, osteoporosis was more frequent in the hip [75]. Hip BMD was related to the presence of syndesmophytes and vertebral fractures, to disease duration and activity [142,144], and to CRP [145]. Osteitis in the vertebrae precedes the development of erosions and syndesmophytes [41,42].

Fracture risk in ankylosing spondylitis
Morphometric vertebral fractures (with a deformation of 15% or 20%) have been reported to be 10% to 30% in groups of patients with AS [146]. Th e odds ratios of clinical vertebral fractures were 7.7 in a retrospective population-based study [147] and 3.3 in a primary carebased nested case control study [148]. In both studies, the risk of non-vertebral fractures was not increased.
Th e risk of vertebral fractures is multifactorial and inde pendent of and superimposed on other clinical risk factors [118].
Vertebral fracture risk in AS was higher in men than in women and was associated with low BMD, disease activity, and the extent of syndesmophytes [144,149]. Vertebral fractures contributed to irreversible hyperkyphosis, which is characteristic in some patients with advanced disease with extensive syndesmophytes (bamboo spine) [150,151].
Apart from presenting with these 'classical' vertebral fractures, patients with AS can present with vertebral fractures that are specifi cally reported in AS. First, erosions at the anterior corners and at the endplates of vertebrae (Andersson and Romanus lesions) result in vertebral deformities if erosions are extensive and the results of such measurements should not be considered a classical vertebral fracture ( Figure 5) [75,152]. Second, in a survey of 15,000 patients with AS, 0.4% reported clinical vertebral fractures with major neuro logical complications [153]. Th ird, owing to the stiff ening of the spine by syndesmophytes, transvertebral fractures have been described [153]. Fourth, fractures can occur in the ossifi ed connections between the vertebrae [153]. In all of these cases, CT, MRI, and eventually bone scintigraphy are helpful to identify these lesions and the extent of neurological consequences (Figure 6) [154].

Risk predictors of bone changes in ankylosing spondylitis
Th e diagnosis of vertebral fractures is hampered by the fi nding that only one out of three morphometric vertebral fractures is accompanied by clinical signs and symptoms of an acute fracture. Th is is probably even less in patients with AS as fractures of the vertebrae and their annexes can be easily overlooked when a fl are of back pain is considered to be of infl ammatory origin without taking into account the possibility of a fracture. In case of a fl are of back pain, special attention, therefore, is necessary to diagnose vertebral fractures in AS, even after minimal trauma. Additional imaging (CT, MRI, and bone scintigraphy) might be necessary in patients in whom a fracture is suspected in the absence of abnormalities on conventional radiographs. On the basis of the limited data on fracture risk in AS, vertebral fractures especially should be considered in patients with a fl are of back pain, persistent infl ammation, long disease duration, hyperkyphosis with increased occiput-wall distance, bamboo spine, and persistent pain after trauma, even low-energy trauma. Th e FRAX algorithm can be used to calculate the 10-year fracture risk but cannot be used to separately calculate the risk of clinical vertebral fractures [118]. Risk factors to predict erosive sacroiliitis have been identifi ed. Th ese include male gender, CRP, B27, clinical symptoms, family history [155][156][157], and the occurrence of syndesmpophytes (such as B27, uveitis, no peripheral arthritis, prevalent syndesmophytes, and disease duration) [72,158,159]. Also, CTX-II has been shown to predict syndesmophytes, which could refl ect cartilage destruc tion during enchondral new bone formation in enthesitis, including syndesmophytes [27]. Th ese risk factors can now be extended with subchondral bone involvement (as defi ned by osteitis on MRI) that has been shown to predict erosive sacroiliitis [39] and the occurrence of syndesmophytes [160,161]. To predict radiographic erosive sacro iliitis, the Assessment of Spondylo-Arthritis international Society recently developed and validated criteria that included active signs of infl ammation on MRI, which are defi ned as active infl ammatory lesions of sacroiliac joints with defi nite bone marrow edema/osteitis [156,157].

Eff ect of treatment on bone changes in ankylosing spondylitis
As the pathophysiology of vertebral fractures in AS is taken into account (Figure 7), therapy should be directed at suppressing infl ammation, bone resorption, and bone formation. No RCTs on the eff ect of treatment on the risk  of vertebral fractures in AS are available. In the General Practice Research Database, the use of non-steroidal anti-infl ammatory drugs (NSAIDs) is associated with a 30% decrease in the risk of clinical vertebral fractures, but this has not been studied prospectively [75,148]. In general, continuous use of NSAIDs, in com parison with intermittent use, and celecoxib decreased the formation of syndesmophytes [148,162]. Th e mechanisms of these eff ects are unclear. NSAIDs inhibit bone formation, as shown in fracture healing, which is also an infl ammationdriven model of increased bone formation [163,164]. One other explanation is that pain relief can ameliorate function and decrease immobility [75]. Limited studies with bisphosphonates indicated inhibi tion of infl am mation in AS [165]. Zoledronate did not prevent the occurrence of syndesmophytes in rats [166]. Bisphosphonates, however, can be considered in the treatment of osteoporosis in high-risk patients [167]. TNF blockade decreased osteitis, prevented bone loss, and decreased CRP and IL-6 [145,168] but had no eff ect on the occurrence of syndesmophytes [169]. Taken together, these data indicate that control of infl ammation is able to halt bone loss and suppress osteitis in AS but not the occurrence of syndesmophytes. Further research is needed to understand why NSAIDs could decrease fracture risk and syndesmophyte formation, why TNF blockade prevents bone loss but not syndesmophyte formation, and new ways to prevent syndesmophyte formation.

Discussion and summary
Th ese data indicate that bone is a major target for infl ammation and that bone loss and osteoporosis are common features that contribute to the increased fracture risk in RA and AS. However, the problem of bone involvement in RA and AS is more complex than in primary osteoporosis alone. Th e consistent fi nding of peri-infl ammatory bone loss and osteitis in both RA and AS raises questions, besides fracture risk, about the clinical signifi cance of bone loss.
Periarticular bone loss and osteitis coincide early in RA and AS and not only precede but also predict the occurrence of visible erosions [76]. Th is raises the question of the mechanism by which these anatomical coincident changes in the joints, entheses, and bone marrow occur. As described above, no direct anatomical or vascular connection between the joint cavity and bone marrow is present, but some healthy subjects can have small erosions in the MCP joints without having RA and have erosions at the entheses and vertebral cortices. In subjects with small erosions before RA or AS becomes apparent clinically, it can be assumed that, when they develop arthritis or enthesitis, the erosions allow immediate contact with bone marrow, resulting in coincident joint, enthesis, and bone marrow infl ammation. Healthy subjects without such erosions could develop small erosions, resulting in measurable peri-infl ammatory bone loss, before they can be identifi ed on radiographs or MRI because of the spatial resolution of radiology and MRI and the single-plane images of radiographs. Another hypothesis is that RA and AS are primarily bone marrow diseases [170,171], with secondary invasion of the joint via erosions created by intramedullary activated osteoclasts or via pre-existing erosions. Indeed, CD34 + bone marrow stem cells have been shown to be abnormally  sensitive to TNFα to produce fi broblast-like cells [172], suggesting an underlying bone marrow stem cell abnormality in RA. In AS, the fi nding of early osteitis is even more intriguing as osteitis is occurring in the vertebrae, where no synovium but periost is present at the anterior sites and discs between vertebrae. Local communication with the periost is possible by the local vascular connections or pre-existing erosions, leaving open the possibility that periost is the primary location of infl ammation in AS. Th e same applies for the intervertebral disc, which has no direct vascular contact but can have pre-existing erosions. Whether RA and AS are initialized in the joints, enthesis, or the bone marrow is a growing fi eld of debate [170], and such hypotheses will need much more study.
Regardless of these anatomical considerations, when the size of bone edema that can be found by MRI and the extent of early periarticular bone loss are taken into account, it seems that infl ammation is as intense and extensive inside bone marrow as in the synovial joint in RA and AS and in the enthesis in AS. As bone loss and bone edema occur early in the disease, these fi ndings indicate that bone marrow infl ammation -and not just joint or enthesis infl ammation -is a classical feature of early RA and AS. To what degree impaired osteoblast function is associated with loss of control of HSC and Bcell diff erentiation in their subendosteal niches in RA is unknown and needs further study as B-cell proliferation is a feature of RA but not of AS [173][174][175].
Th e fi nding that bone involvement can be disconnected from clinically detectable infl ammation is quite intriguing. In RA, bone erosions can progress even when the infl ammatory process is adequately controlled (that is, in clinical remission) [176], and progress of bone erosions can be halted by denosumab in spite of persistent infl ammation [133][134][135][136]. In AS, the occurrence of syndesmophytes can progress in spite of suppression of infl ammation by TNF blockade [160]. Th ese fi ndings have been described as a disconnection between infl ammation and bone destruction and repair.
Th e correlation and eventual disconnection between osteitis and bone loss, parameters of disease activity, and erosions suggest a dual time-dependent role for the occurrence of erosions. Early in the disease process, the primary negative eff ect of pre-existing or newly formed erosions is the connection they create between the bone marrow and the joints, periost, and entheses. In this way, erosions contribute to local amplifi cation of infl ammation by allowing bone marrow cells to have direct local connection with extraosseous structures and creating a vicious circle of infl ammation between joints, periost, entheses, and bone marrow [177]. Only in a later stage do erosions contribute to loss of function [178]. In this hypothesis, the attack of infl ammation on bone by stimulating osteoclasts has far-reaching consequences. First, it would indicate that timely disease suppression and the prevention of the development of a fi rst erosion rather than halting erosion progression should be considered a primary objective, both in RA and AS [179]. Second, periarticular bone loss and osteitis should be considered, at least theoretically, an indication for the presence of erosions, even when erosions cannot be visualized on radiographs or MRI, and periarticular bone loss and osteitis should be considered an indication for early aggressive therapy [180]. Of course, the eff ectiveness of antirheumatic treatment based on osteitis should be demonstrated. Th ird, the fi nding of disconnection between infl ammation and bone involvement indicates that, even when infl ammation is clinically under control, the degree to which bone-directed therapy is indicated should be studied in order to prevent (further) progression of erosions and syndesmophytes. In conclusion, the involve ment of bone as a major target of infl ammation in RA and AS raises many questions [10,[181][182][183][184], opening perspectives for further research in the understanding and treatment of the complex bone disease component of RA and AS.