MRI and X-ray in axial spondyloarthritis: the relationship between inflammatory and structural changes

Demonstration of an association between inflammation and spinal ankylosis has been challenging. Until the advent of MRI, prospective study was not possible due to inaccessibility of tissue. Recent studies using MRI have described an association between the presence of bone edema at vertebral corners on MRI and the subsequent development of syndesmophytes at the corresponding vertebral corners on radiography. Although reports have also highlighted the development of new syndesmophytes where the baseline MRI shows no inflammation, MRI has limited sensitivity for detection of spinal inflammation that is clearly evident on histopathology. There are also crucial methodological challenges because radiographic assessment is limited to the anterior corners of the cervical and lumbar spine while MRI lesions in the cervical spine are often small while spurious inflammatory signal is common in the lumbar spine. Follow-up MRI evaluation in two independent studies has also shown that inflammatory lesions that resolve after anti-TNF therapy are more prone to develop into syndesmophytes. It may be possible that very early inflammatory lesions resolve completely without sequelae if anti-TNF therapy is introduced before new bone formation becomes largely autonomous. For an individual patient the overall development of new bone during anti-TNF therapy may therefore depend on the balance between the number of early and more mature inflammatory lesions. Clinical trials of anti-TNF agents in early spondyloarthritis together with prospective MRI studies will allow more detailed testing of this hypothesis as a major priority for the research agenda in spondyloarthritis.


Introduction
A hallmark pathological feature of spondyloarthritis (SpA) is the development of ankylosis in axial joints. Th is process is the primary reason for the development of major disability. Its prevention therefore constitutes a key goal of management. Th ere has been a long-standing assumption that infl ammation is the process that sets in motion the chain of events that leads to ankylosis and that the two processes continue to be inextricably linked as the disease progresses over time. However, direct testing of this hypothesis has been diffi cult due to several factors, such as the lack of availability of objective clinical and laboratory measures of infl ammation, delay in diagnosis, the slow progression of disease, and the lack of access to tissue for direct histopathological analysis. Investigators have turned to indirect approaches such as animal models of SpA and the testing of various biomarkers refl ecting infl ammation. Th e signifi cant limitations of these approaches are not the subject of this review but highlight the increasing importance of recent advances in non-invasive assessment with imaging modalities. In particular, the advent of fat-suppressed mag netic resonance imaging (MRI) allows direct visualiza tion of features related to infl ammation and this has, for the fi rst time, allowed the prospective non-invasive evaluation of infl ammation in axial joints. Th e reports of these studies assessing the relationship between infl amma tion and structural changes have generated considerable interest, culminating in numerous symposia at international meetings. Th is review sets out the reasons why this topic has generated so much interest, outlines the methodology and pitfalls surrounding the use of MRI for evaluation of the axial joints, reviews the prospective studies that focused on MRI infl ammation and its relation to radiographic changes, and suggests possible avenues of further research in SpA.

Evidence for 'uncoupling' between infl ammation and structural damage in spondyloarthritis
Th e detailed histopathological examination of necropsy material and tissue biopsies outlined by John Ball in his

Abstract
Demonstration of an association between infl ammation and spinal ankylosis has been challenging. Until the advent of MRI, prospective study was not possible due to inaccessibility of tissue. Recent studies using MRI have described an association between the presence of bone edema at vertebral corners on MRI and the subsequent development of syndesmophytes at the corresponding vertebral corners on radiography. Although reports have also highlighted the development of new syndesmophytes where the baseline MRI shows no infl ammation, MRI has limited sensitivity for detection of spinal infl ammation that is clearly evident on histopathology. There are also crucial methodological challenges because radiographic assessment is limited to the anterior corners of the cervical and lumbar spine while MRI lesions in the cervical spine are often small while spurious infl ammatory signal is common in the lumbar spine. Follow-up MRI evaluation in two independent studies has also shown that infl ammatory lesions that resolve after anti-TNF therapy are more prone to develop into syndesmophytes. It may be possible that very early infl ammatory lesions resolve completely without sequelae if anti-TNF therapy is introduced before new bone formation becomes largely autonomous. For an individual patient the overall development of new bone during anti-TNF therapy may therefore depend on the balance between the number of early and more mature infl ammatory lesions. Clinical trials of anti-TNF agents in early spondyloarthritis together with prospective MRI studies will allow more detailed testing of this hypothesis as a major priority for the research agenda in spondyloarthritis. concept that ankylosis was an excessive reparative response to an infl ammatory and osteo-destructive process [1]. It was expected, therefore, that anti-TNFα therapies would be benefi cial in preventing ankylosis in SpA through their profound ability to suppress infl ammation. However, the results of three studies that compared radiographic progression over the minimum required 2 year time frame in patients in anti-TNFα trials with a historical cohort of ankylosing spondylitis (AS) patients did not support a benefi cial eff ect of anti-TNFα therapy [2][3][4]. Several important limitations included the lack of comparability in disease severity, inadequate sample size, insuffi cient suppression of infl ammation, and possible diff erences in intake of non-steroidal anti-infl ammatory agents (NSAIDs), which have been shown to ameliorate progression [5]. Nevertheless, various sensitivity analyses aimed at demonstrating diff erences in progression within subgroups failed to identify any impact of anti-TNFα therapy. Assessment of anti-TNFα therapy in an animal model of SpA, ankylosing enthesitis, also failed to demon strate amelioration of ankylosis by anti-TNFα therapy [6]. Th is led to an alternative hypothesis whereby an as yet unknown pathogenic trigger(s) simultaneously induces both an acute infl ammatory reaction and activation of stromal progenitor cells, which ultimately leads to endochondral new bone [7]. But the processes of infl ammation and new bone formation proceed along essentially disconnected or 'uncoupled' pathways once each has been activated.
Discerning the true relationship between infl ammation and ankylosis is pivotal to future advances in disease modifi cation because if these processes are largely uncoupled, future therapies will need to target bone formation pathways. A major challenge in understanding the link between these two processes in human SpA is the delayed recognition of disease so that most patients are assessed when both infl ammatory and reparative processes are well established. It has been proposed that initial development of infl ammation leads to erosive cartilage bone destruction followed by tissue repair and ultimately ossifi cation once infl ammation has completely resolved [8]. Consequently, this hypothesis would imply that early and eff ective suppression of infl ammation would prevent the development of ankylosis. MRI allows non-invasive assessment of infl am mation but the optimal approach to the assessment of bone is still radiography. Consequently, recent prospective studies have combined assessment of the evolution of infl ammatory lesions using MRI with the development of new bone using radiography in the spine and sacroiliac joints (SIJs) to provide further insights regarding these hypotheses.

MRI evaluation: standardization of methodology and key pitfalls
As the diversity of acute and structural lesions on MRI at diff erent stages of disease has become evident, there has been recognition for the need to develop a standardized methodology for assessment of MRI scans in SpA and to develop and validate standardized defi nitions in order to facilitate clinical research. Th is need is particularly relevant for assessment of the SIJs due to the complex anatomy of the joint. An international working group from Canada, Denmark, and Switzerland (Th e MORPHO Group) has developed a standardized module (available online at [9]) for evaluation of the SIJ [10]. Th e SIJs are scanned in the tilted semi-coronal plane and assessment of consecutive slices is undertaken from anterior to posterior according to anatomical landmarks and specifi c rules [11]. It is deemed essential that assessment of each slice includes corresponding T1-weighted spin echo (T1SE) and short tau inversion recovery (STIR) MRI sequences. Th e T1SE sequence detects the signal from fat and permits detailed assessment of the anatomy of the joint. Th e STIR sequence is a fat-suppression technique that allows the visualization of the free water signal associated with infl ammation and is particularly useful for identifying acute lesions in the bone marrow that may be obscured by marrow fat.
A similar standardized methodology is recommended for assessment of the spine. Th e Canada-Denmark International MRI Working Group has developed and validated standardized defi nitions for acute lesions in the spine based on the STIR sequence that incorporate a defi nition of what constitutes the reference normal STIR signal [12]. Th ese include vertebral corner infl ammatory lesions (CILs), which are depicted in reference images ( Figure 1) that describe typical lesions as well as set the threshold for detection. Calibration of readers using these tools is essential prior to MRI reading exercises. Imaging of the entire spine for SpA is conducted in the sagittal plane and the spine is visualized in two halves, cervico-thoracic and thoraco-lumbar portions. With this large fi eld of view the cervical vertebrae appear small and CILs may not be readily detected ( Figure 1). STIR MRI is subject to physiological motion artifacts so that blood fl owing in the inferior vena cava and the abdominal aorta may cause spurious STIR signal, termed phase encoding artifacts, which may resemble anterior CILs in the lumbar spine ( Figure 1). Consequently, achieving reliability for detection of CILs is challenging even for experienced readers [13] and MRI studies focused on these lesions should include at least two readers scoring independently, with the primary analysis being focused on concordant data from independent reader assessments.
It is apparent that the appearance of a CIL may be heterogeneous and the Canada-Denmark group has defi ned two categories of CIL. In type A CILs the STIR signal is relatively homogeneous and extends to the vertebral corner. In type B CILs, the STIR signal has receded from the vertebral corner and is typically less intense than a type A CIL and more heterogeneous ( Figure 2). It is necessary to examine the corresponding T1SE sequence to determine whether there is an erosion or new bone at the vertebral corner, which then appears dark on both the T1SE and STIR sequence, or whether there is tissue metaplasia to fat, which then appears bright on the T1SE sequence. Type B CILs likely represent a more advanced stage of evolution of the infl ammatory lesion. In particular, prospective analysis of MRI scans has directly shown that CILs evolve into fat lesions at the corresponding vertebral corner and that this is more likely to occur in patients receiving anti-TNFα therapy [14,15]. Reliable detection of type B CILs and assessment of resolution following anti-TNFα therapy are often more diffi cult than for type A CILs, especially in the lumbar spine due to phase encoding artifacts. Nevertheless, it may be important to make this distinction because the two types of CIL may carry diff erent prognostic implications for the development of new bone (see below).
Th e approach to the detection of structural lesions by MRI has been the same as for acute lesions. In particular, the proposed standardized defi nitions for fat infi ltration and erosions in both the SIJs and spine have now been validated and reference images are available to set the threshold for detection, to facilitate widespread implemen tation, and to minimize discrepancies between studies [16,17]. Th e MRI exercises undertaken by the Canada-Denmark and MORPHO study groups have highlighted several key challenges in the reliable assessment of structural lesions. Detection of erosions in the SIJ is particularly challenging without extensive calibration of readers [18]. In particular, while several reports have defi ned bone erosion as full thickness loss of the dark appearance of either iliac or sacral cortical bone of the SIJ with loss of the adjacent marrow signal on T1-Weighted images, this is often diffi cult to distinguish from bone sclerosis [19]. Detection of discovertebral erosions can be reliably accomplished with minimal reader calibration but this is more diffi cult for vertebral corner erosions because these lesions are small and it is diffi cult to be certain that there is a breach of cortical bone. While the (Reference images are also available at [44].) The arrow in the left image at T5 upper shows a vertebral corner infl ammatory lesion (CIL) at the threshold of detection. The bone marrow signal in the center of the vertebral body, if normal, constitutes the reference for designation of normal signal [11]. The arrow in the central image shows a CIL at the anterior corner of C4 lower. The large fi eld of view necessary to scan the entire spine in AS impairs detailed assessment of cervical vertebrae. The right image shows increased STIR signal in a vertical pattern across the anterior portions of the lumbar vertebrae simulating infl ammation but indicative of blood fl owing through the great vessels (phase encoding artifact). STIR, short tau inversion recovery. detection of fat infi ltration is reliably accomplished with minimal calibration, fat infi ltration is observed in the SIJs and spine of normal individuals and there is considerable heterogeneity in its morphological characteristics, particularly in the SIJs. It is presently unclear which features relate specifi cally to SpA.

Prospective studies of MRI infl ammation and radiographic progression?
Spine Several studies, all using diff erent methodological approaches, have examined the relationship between infl ammation visible on baseline STIR MRI of the spine in patients with established AS and the development of syndesmophytes on radiographs after 2 years of follow up [20][21][22][23][24][25][26]. It is important to note from the outset that all radiographic assessments in each study were confi ned to the anterior cervical and lumbar spine vertebral corners because thoracic spine abnormalities cannot be reliably detected on radiography due to overlapping structures. Similarly, assessment of posterior vertebral corner abnormalities is unreliable [27]. Consequently, MRI assessments have been confi ned to the corresponding anterior vertebral corners of the cervical and lumbar spine. As discussed above, this restricts MRI assessment to regions of the spine that may pose challenges to the reliable detection of lesions on the STIR sequence. A summary of the fi ndings from these reports is presented in Table 1 and shows that the odds ratios (ORs) for the association of baseline infl ammation and the development of new syndesmophytes after 2 years were significant for each study but varied from 1.9 to 4.6. Th e percen tage of individual vertebral units with infl ammation on baseline STIR MRI also varied from 7.5 to 17.2%, the percentage of vertebral regions with baseline infl ammation that developed new syndesmophytes varied from 6.5 to 20%, and the percentage of syndesmophytes developing in the absence of baseline infl ammation varied from 61 to 76%. So while there is consistency in demonstrating a link between infl ammation and new syndesmophytes, it has also been argued that these data more strongly support a lack of coupling between infl ammation and new bone because most new syndesmophytes developed from vertebral corners that appeared normal on STIR MRI. What could account for the diff erences between studies and how does this shed light on the hypothesis that infl ammation and ankylosis are coupled and/or uncoupled?
Th e fi rst report assessed infl ammation on baseline STIR MRI in 39 patients who participated in clinical trials of anti-TNFα therapy [20]. Infl ammation was recorded at vertebral edges (VEs), and these were defi ned as positive for infl ammation 'if the infl ammatory activity was present in the anterior half of the vertebral edge only' . Th is could refer to lesions directly at the vertebral corners as well as lesions adjacent to the endplate but not extending to the vertebral corner as in a discovertebral lesion, which could have diff erent prognostic implications. Th e report does not indicate how many readers assessed either the MRI scans or the radiographs. Infl ammation was reported in 17.2% of vertebral edges at baseline, new syndesmophytes developed from 6.5% of vertebral edges with infl ammation as opposed to 2.1% without infl ammation (OR = 3.3, 95% confi dence interval (CI) 1.5 to 7.4), and 38% of all new syndesmophytes developed from vertebral edges with baseline infl ammation while 62% had no baseline infl ammation. On follow up at 2 years, 4.4% of vertebral edges where infl ammation had resolved and 5.6% of vertebral edges that had either persistent infl ammation or had newly developed infl ammatory lesions developed a new syndesmophyte as compared to 1.3% of VEs that had no infl ammation at either baseline or follow up (P = 0.07 and 0.04, respectively).
Th e second report assessed infl ammation on STIR MRI in 29 patients in anti-TNFα clinical trials and 41 patients

. Reference image of short tau inversion recovery MRI illustrating type A and type B corner infl ammatory lesions.
In the latter, the increased short tau inversion recovery MRI signal has receded from the vertebral corner.
followed in an observational cohort (23 on standard therapies, 18 receiving anti-TNFα therapy) [21]. Vertebral CILs were assessed as defi ned by the Canada-Denmark group [12] and the primary analysis focused on concordant data from three reader pairs, each reader scoring MRI scans independently. For the primary reader pair and scans from patients in anti-TNFα trials, infl ammation was reported in 7.5% of vertebral corners at baseline, new syndesmo phytes developed from 20% of CILs at the corresponding vertebral corner as opposed to 5.1% without CILs (OR = 4.6, 95% CI 1.7 to 12.6), and 24% of all new syndesmo phytes developed from CILs while 76% had no baseline infl ammation. Data were comparable for scans from patients in the observational cohort (Table 1). On follow up at 2 years, 25% of CILs where infl ammation had resolved had developed a new syndesmophyte compared to none of the CILs that had persistent infl ammation. Th e follow-up data on the observational cohort were extended to include 23 patients receiving anti-TNFα therapy and 27 receiving standard therapy and reported in a separate manuscript [24]. Consistent with the earlier report, new syndesmophytes developed from CILs that resolved by 2 years (42.9% in anti-TNFα and 20% in standard therapy groups, respectively) but not from a single CIL where infl ammation was persistent.
A third report assessed the relationship between infl ammation on baseline STIR MRI and development of new syndesmophytes but did not directly assess infl ammation at the corresponding vertebral corner [22]. Infl am ma tion was recorded in MRI scans from 182 patients in a trial of infl iximab according to its presence/ absence at each cervical and lumbar discovertebral unit (DVU), which represents the region between two imaginary lines drawn through the middle of two adjacent vertebrae and includes the four vertebral corners, the intervertebral disc, the adjacent vertebral endplates and bone marrow [28]. Data for each of two readers scoring independently were reported but not concordant data. Infl ammation was reported in 14.8% and 20% of DVUs at baseline for readers 1 and 2, respectively. For the primary reader, new syndesmophytes developed from 12.1% of DVUs with infl ammation as opposed to 6.7% without infl ammation (OR = 1.9, 95% CI 1.3 to 2.8), and 23.8% of all new syndesmophytes developed from DVUs with base line infl ammation while 76.2% had no baseline infl am mation. At the patient level, MRI activity was not associated with change in radiography score.
Th e studies that assessed the association between baseline infl ammation and new syndesmophytes at the corresponding vertebral corner [20,21] reported higher ORs than the study that assessed infl ammation at the level of a DVU [22]. Th is, as well as the absence of an association between MRI activity and change in radiography score at the patient level [22], is not surprising. As shown in Figure 3, an infl ammatory lesion can be observed in several locations within a DVU other than the vertebral corner where a new syndesmophyte develops and this will serve to reduce the strength of the association.
Th e consensus that most new syndesmophytes developed from vertebral corners that did not demonstrate infl ammation at baseline should not necessarily be Table 1

Study
Yes No OR (95% CI)
inter preted as demonstrating uncoupling between infl amma tion and radiographic progression. First, MRI detects only about half of the infl ammatory lesions evident histopathologically in the spine of patients with AS [29]. Second, the baseline MRI represents a snapshot in time and new infl ammatory lesions can develop even in patients on anti-TNFα agents [20]. Th ird, prospective studies show that infl ammatory lesions evolve into fat lesions evident on T1SE MRI and this can be seen by 52 weeks after the start of treatment, especially in patients on anti-TNFα agents [14,15]. Th is is relevant because two recent reports have demonstrated an association between baseline fat lesions at vertebral corners on T1SE MRI and the development of new syndesmophytes at the corresponding corner [23,26]. Th e fi rst assessed 100 MRI scans from 2 cohorts of patients with AS, a clinical trial cohort of patients receiving anti-TNFα therapy (n = 38) and an observational cohort (n = 62, 26 receiving anti-TNFα and 36 receiving standard therapies), while the second report assessed 76 scans from a trial of adalimumab. Consequently, since fat lesions represent a more advanced stage in the evolution of an infl ammatory lesion and are themselves associated with radiographic progression, further studies should assess both STIR and T1SE MRI scans before concluding that a syndesmophyte has developed from a vertebral corner without current or prior infl ammation. Another report analyzed the association between change in MRI SIJ infl ammation score after 22 weeks of anti-TNFα therapy and the development of new syndesmophytes [25]. Patients developing new syndesmophytes had larger reductions in MRI SIJ infl ammation score and decreases in C-reactive protein (CRP) and IL6 so that normalization of CRP and IL6 was more frequently observed in patients with new syndesmophytes even after adjustment for the extent of radiographic damage at baseline. Th ese patient level data demon strating an association between resolution of infl ammation and new bone formation are consistent with the data reported at the individual vertebral corner level [21,24]. It is important to note, however, that the predictive capacity of biomarkers associated with infl ammation for new bone is presently unclear [30,31]. CRP has been inconsistently associated with development of new bone [32,33], while IL6 has not been shown to be a predictor, which may refl ect its complex role in infl ammation and bone metabolism [34,35].
A major consideration in the analysis of the association between MRI infl ammatory and other lesions such as fat and the development of new bone is the necessity to adjust for within-patient variation in the extent of radiographic damage at baseline because it has been shown previously that the likelihood of fi nding a new syndesmophyte is higher in a patient that already has syndesmophytes and/ or ankylosis at baseline [36]. One study reported that the association between infl amma tion at the level of a DVU and new bone remained statistically signifi cant after adjustment for such within-patient correlation as well as after further adjustment for other confounders [22]. A second report described the use of generalized linear latent and mixed models (GLLAMMs) to show that vertebral corners that were infl ammation-positive signifi cantly predicted new syn des mophytes after adjustment for within-patient depen dence in the total number of vertebral corners with fat, infl ammation, and syndesmophytes/ankylosis at baseline [23].
It may be possible that very early infl ammatory type A CILs resolve completely without any sequelae if anti-TNF therapy is introduced before bone formation pathways become activated [37]. On the other hand, once a lesion has become more advanced as in a type B CIL and crossed a certain 'threshold' of maturation, introduction of anti-TNF therapy may alleviate infl ammation but bone formation may even be enhanced through down regulation of Dickkopf-1, a major inhibitor of bone formation by inhibiting signaling through Wingless proteins [38]. TNFα is a major positive regulator of Dickkopf-1. For an individual patient the overall development of new bone during anti-TNF therapy may depend, therefore, on the balance between the number of early and more mature infl ammatory lesions. Th is hypothesis could explain the lack of impact of anti-TNFα therapies on radiographic progression. A recent report supports this hypothesis, which will require further testing in early SpA cohorts [26].

Sacroiliac joint
Several small studies have assessed infl ammatory changes in the SIJ as detected by increased STIR signal refl ecting bone marrow edema in subchondral bone and the subsequent development of radiographic sacroiliitis. One study evaluated 17 patients with infl ammatory back pain of 3 to 14 months duration but normal pelvic X-ray who were followed for 18 to 30 months [39]. Virtually all patients had abnormalities on baseline MRI that persisted on follow-up MRI 2 to 30 months later and 11 developed plain radiographic features of sacroiliitis on follow up, suggesting that infl ammation observed on MRI is of prognostic signifi cance. In a second prospective study of 25 consecutive HLA-B27-positive patients with infl amma tory low back pain and ≤grade 2 unilateral sacroiliitis, the positive predictive value of ≥grade 2 sacroiliitis on baseline MRI for the development of ≥grade 2 sacroiliitis on plain radiograph after 3 years was 60% (sensitivity 85%, specifi city 47%) [40]. In contrast to the previous study, subchondral marrow edema was found in only 20 SIJs and median duration of symptoms was much longer (4 years), raising questions as to the reliability of the clinical diagnosis. In a study of 40 consecutive patients presenting with infl ammatory back pain according to the Calin criteria and followed for a mean of 7.7 years [41], the combination of severe bone marrow edema with HLA-B27 positivity was a predictor of radiographic sacroiliitis (likelihood ratio 8.0, specifi city 92%) while minor bone marrow edema, which can be found in about 25% of healthy age and sex-matched healthy controls [10], had no predictive validity. However, a detailed quantitative analysis of bone marrow edema in this cohort has recently reported that about a third of patients with substantial bone marrow edema still had minimal radiographic changes after 7.7 years of follow up [42]. Although this implies additional un identifi ed prognostic factors, follow-up MRI examination was not available so it is unclear to what extent these infl ammatory features were persistent.

Future directions
It is likely that attention will now focus on patients with early disease to further understand the prognostic capacity of MRI. In particular, the development of the new Assessment of SpondyloArthritis international Society (ASAS) classifi cation criteria has broadened the spectrum of disease to include patients with nonradiographic disease [43]. While MRI is now accepted as a classifi cation criterion, its prognostic role is still uncertain. Preliminary evidence suggests that bone marrow edema alone may have insuffi cient prognostic capacity. Increasing attention is therefore also turning towards other features on MRI, such as fatty lesions and erosions. Although these are often termed 'chronic' lesions, they may be seen in a substantial proportion of patients with non-radiographic SpA [10] but little is known regarding their prognostic capacity. Th ere is keen interest in the assessment of spinal infl ammatory lesions in early disease and comprehensive follow up of suffi cient duration that will allow greater understanding of the associations between acute and structural lesions as well as between diff erent structural lesions. Th ese longitudinal data will be an essential prerequisite to the development of imaging surrogates that may supplant radiographic progression and facilitate the development of novel disease-modifying therapies.

Competing interests
The author has received honoraria and/or unrestricted grants from Pfi zer, Merck, Abbott, Amgen, Eli-Lilly, Janssen.