Objectives To study the sequential relationship between MRI vertebral corner inflammation (VCI), vertebral corner fat deposition (VCFD) and the development/growth of radiographic syndesmophytes at the same vertebral corner (VC).
Methods Baseline, 24 and 102 weeks spinal MRIs were assessed for the presence/absence of VCI and VCFD. Anterior VCs of lateral radiographs of the cervical and lumbar spine (baseline and 102 weeks) were assessed for the development of new bone (syndesmophyte formation or syndesmophyte formation/growth combined). Data from 161 to 177 patients were analysed at the VC level using two-way and multilevel analyses adjusting for within-patient correlation and MRI reader (generalised estimating equations for binomial outcomes).
Results The presence of VCI (adjusted (adj) OR 1.75 to 1.98) as well as the presence of VCFD (adjOR 1.60 to 2.32) at any time point (TP) were significantly associated with the development of new bone. The combination of VCI and VCFD at the same VC increased the strength of the association, both for the sequential or simultaneous presence of VCI and VCFD across the three TPs (adjOR 2.12 to 2.73), as well as for the development of new VCFD preceded by VCI at a previous TP (adjOR 2.12 to 3.01). The complete absence of both VCI and VCFD across the three TPs ‘protected’ against new bone formation (adjOR 0.45 to 0.62). However, 40–66% of new bone still developed in VCs without MRI inflammation or fat degeneration at any of the three TPs.
Conclusions Both VCI and VCFD contribute to new bone formation in ankylosing spondylitis (AS), especially if VCI precedes VCFD. However, VCI, VCFD and this particular sequence of events only partially explain the development of new bone in AS.
- Ankylosing Spondylitis
- Magnetic Resonance Imaging
- Outcomes research
Statistics from Altmetric.com
Structural damage in ankylosing spondylitis (AS) is characterised by the formation of new bone in the spine. Syndesmophytes and bridging syndesmophytes are the typical lesions,1 ,2 with erosions, sclerosis and squaring being additional lesions that also reflect structural damage in AS. Syndesmophytes can lead to decreased spinal mobility, reduced physical function and loss of quality of life.3–5 Therefore, understanding the mechanisms underlying new bone formation is of importance in AS.
The processes that drive the formation of new bone in AS are not completely understood, and there is debate about whether inflammation and osteoproliferation are related or uncoupled phenomena.6–8 This is a challenging topic to investigate because the progression of structural damage is typically slow, it is problematic to perform serial histopathological examinations of spinal tissue and reliable biomarkers of new bone formation in AS are lacking.
MRI provides an indirect and non-invasive method of investigating elements of the pathophysiology of new bone formation in AS. Fat deposition can be seen on T1-weighted sequences and bone marrow oedema (reflecting inflammation) can be seen on T2-weighted sequences with fat suppression, such as the short tau inversion recovery (STIR) sequence.9–12 However, conventional radiography is still the gold standard method to assess syndesmophyte formation/bridging13 because tissues with low proton density such as cortical bone and paravertebral ligaments exhibit low or no signal intensity in all pulse sequences and are difficult to differentiate on MRI scans.14
It has been shown by us in the same cohort15 and by others in independent cohorts16–18 that vertebral corners (VCs)/units/edges with inflammation are more likely to develop new syndesmophytes than VCs/units/edges without inflammation. It has also been proposed that syndesmophytes are more likely to develop at VCs in which inflammation resolves compared with those where inflammation persists.17 ,18 Resolving inflammation has also been associated with fat deposition.19 In turn, fat deposition, with or without concomitant inflammation, has been associated with the formation of new syndesmophytes.20–22
Our aim was to expand our analytical studies about the association between inflammation and bone formation by investigating the relationship between MRI inflammation and fat deposition at a VC and the subsequent development of new bone at the same site. In this analysis, the focus was on a sequence analysis, addressing the hypothesis that vertebral corner inflammation (VCI) ‘leads to’ fat deposition, which in turn ‘leads to’ bone formation.
For this study, we have made use of the same 80% random sample of the AS Study for the Evaluation of Recombinant Infliximab Therapy (ASSERT) that we used in our previous analysis.15 Details of the ASSERT study design and population have been previously reported.23 In brief, ASSERT was a 24-week double-blind placebo-controlled clinical trial with infliximab that included patients with AS (according to the modified New York criteria), with a Bath AS Disease Activity Index24 (BASDAI) ≥4 (range 0–10) and a spinal pain score ≥4 (range 0–10), with an open extension until 102 weeks with all patients treated with infliximab.
Radiographs were scored by two readers at baseline and 102 weeks according to the modified Stoke AS Spine Score (mSASSS).25 In this study, we used the mSASSS scores from the original ASSERT trial.23 Lateral views of the cervical and lumbar spine were assessed and anterior VCs from C2-T1 and from T12-S1 (total 24 VCs) were scored for the presence of an erosion, sclerosis or squaring (score 1), syndesmophyte (score 2) or bridging syndesmophyte (score 3). Change from a score of 0 or 1 to 2 or 3 was defined as syndesmophyte formation. Change from a score of 2 to 3 was defined as syndesmophyte growth. The thoracic spine and posterior corners of the cervical and lumbar spine were not assessed because abnormalities at these sites cannot be reliably detected on radiographs.
MRIs were scored by two readers at baseline, 24 and 102 weeks using a VC approach.9–11 T1-weighted and STIR sequences were assessed and the same 24 VCs scored with the mSASSS were also scored for the presence/absence of VCI and vertebral corner fat deposition (VCFD). The level of agreement between MRI readers regarding the presence/absence of VCI and VCFD was assessed using the kappa statistic. The two MRI readers were different readers than the two X-ray readers and all readers were unaware of the patients’ identity, their treatment, the scores of the other imaging modality and the true time-order of the images (fully unbiased scores). This MRI evaluation was a completely new reading, never used in previous ASSERT publications.15 ,21 ,26 Such detailed MRI description of lesions was neither available in the original infliximab efficacy study26 nor in our previous publication looking at inflammation only (but not fat deposition) at the vertebral unit level (rather than VC level).15 This new MRI reading was also different from a previous single-reader publication that included the smaller subset of ASSERT patients that were followed up in an investigator-initiated extension study—the European AS Infliximab Cohort.21
Imaging longitudinal case definitions
Five case definitions were used to combine the information about the presence/absence of VCI at the three available time points (TPs): (1) VCI at baseline, irrespective of inflammation status at other TPs; (2) VCI at baseline only; (3) VCI at baseline and another TP; (4) VCI at any TP; and (5) VCI at all three TPs. Similar case definitions were applied to the presence/absence of VCFD at the three available TPs (figure 1A).
Six subsequent case definitions (figure 1B) were used to integrate the information about the presence/absence of both VCI and VCFD at the three available TPs: (1) sequential or simultaneous presence of VCI and VCFD across the three TPs (ie, presence of VCI and VCFD at the same or different TPs), (2) presence of VCI but not VCFD across the three TPs, (3) presence of VCFD but not VCI across the three TPs, (4) absence of VCI or VCFD across the three TPs, (5) new VCFD preceded by VCI (ie, the sequence VCI→VCFD) and (6) coexistence of VCI and VCFD at the same TP.
Radiographic data were analysed regarding syndesmophyte formation and regarding syndesmophyte formation/growth combined. Two radiographic case definitions were used in the multilevel approach: (1) one definition aiming at sensitivity: a case was defined as positive if at least one of the readers reported progression of structural damage; and (2) a definition aiming at specificity: a case was defined as positive only if both readers reported progression of structural damage (absolute agreement).
Cross-tabulation statistics and measures of association (OR and 95% CI) were first computed using two-way tables to test the association between the various MRI case definitions and radiographic progression after 102 weeks of follow-up. Cross-tabulation statistics were done for every possible pair of imaging readers. The total number of cases in each analysis depended on the imaging case definition and pair of imaging readers used in the analysis (eg, different readers sometimes scored different VCs as not evaluable). Furthermore, VCs with syndesmophytes/ankylosis at baseline (for the outcome syndesmophyte formation) or with ankylosis at baseline (for the outcome syndesmophyte formation/growth) were excluded from the analyses, resulting in another source of variation between readers.
Associations were retested using a multilevel approach to adjust for within-patient correlation and MRI reader (generalised estimating equations (GEEs) for binomial outcomes).27 The following variables were considered covariates and adjusted for when statistically significant in univariate analysis: gender, age, human leucocyte antigen-B27 status, body mass index, disease duration, presence of syndesmophytes/ankylosis at baseline (at the patient level) and baseline and time-averaged C-reactive protein, BASDAI24 and AS Disease Activity Score.28 The treatment variable was forced into all models.
Statistical analyses were performed with IBM SPSS Statistics V.22. Graphics were plotted using GraphPad Prism V.6.
Images belonging to 182 patients with baseline and 102-week radiographic assessments (total of 8736 VCs) and 191 patients with at least one baseline, 24 or 102 weeks MRI assessment (6 patients with one TP, 35 patients with two TPs and 150 patients with three TPs; total of 12 624 VCs) were evaluated by the imaging readers. The kappa score for MRI VCI was 0.46, and the kappa score for MRI VCFD was 0.49. After applying the predefined imaging case definitions and excluding non-evaluable VCs and VCs with syndesmophytes/ankylosis at baseline, data from 3070 to 3389 paired (MRI and radiographic) case definitions belonging to 161–177 patients were analysed. The baseline characteristics of the study population are presented in table 1.
Relationship between VCI and new bone formation
Overall, results showed that the presence of VCI increased the probability of developing new bone at the same VC after 102 weeks of follow-up, irrespective of the MRI case definition, reader pair and radiographic outcome (syndesmophyte formation alone or syndesmophyte formation/growth combined) (online supplementary table S1). OR ranged from 1.33 to 3.87 for VCI at baseline, irrespective of inflammation status at other TPs (statistically significant in 7/8 scenarios), 1.46 to 3.86 for VCI at baseline only (statistically significant in 6/8 scenarios), 0.79 to 3.15 for VCI at baseline and another TP (statistically significant in 5/8 scenarios) and 1.19 to 4.10 for VCI at any TP (statistically significant in 7/8 scenarios, figure 2A, B). The analyses for VCI at all TPs were uninterpretable due to the very low number of VCs with persistent inflammation in this tumour necrosis factor (TNF)-blocker treated population.
Relationship between VCFD and new bone formation
Overall, results showed that the presence of VCFD also increased the probability of developing new bone at the same VC after 102 weeks of follow-up. Data were consistent for all reader pairs and for the two definitions of new bone formation (online supplementary table S2). The only exception was for the presence of VCFD at baseline only where this trend was not observed. However, the very low number of VCs with VCFD at baseline only makes the interpretation of results difficult. Regarding the other VCFD case definitions, OR ranged from 2.43 to 3.27 for VCFD at baseline, irrespective of fat deposition status at other TPs, 2.62 to 3.37 for VCFD at baseline and another TP, 2.21 to 3.33 for VCFD at any TP (figure 2C, D), and 2.05 to 3.36 for VCFD at all TPs. These associations were statistically significant in all studied scenarios.
Relationship between the various combinations of MRI VCI/VCFD and new bone formation
Online supplementary table S3 shows the results for the possible combinations of VCI and VCFD across the three available TPs. The first four MRI case definitions listed in the table are mutually exclusive. The sequential or simultaneous presence of VCI and VCFD across the three TPs was consistently associated with new bone formation (OR 1.77 to 5.80, statistically significant in 7/8 scenarios, figure 3A, B). Associations were weaker for the presence of VCFD but not VCI across the three TPs (OR 1.20 to 2.34, statistically significant in 7/8 scenarios) and even weaker for the presence of VCI but not VCFD across the three TPs (OR 0.40 to 2.32, statistically significant in 2/8 scenarios). The absence of VCI or VCFD across the three TPs was negatively associated with new bone formation (OR 0.33 to 0.49, always statistically significant, figure 3E, F), in agreement with the above positive associations.
The last two MRI case definitions in online supplementary table S3 explore two additional settings: new VCFD preceded by VCI (sequence analysis, VCI→VCFD) and the coexistence of VCFD and VCI at the same TP. VCs in which fat deposition developed de novo and was preceded by VCI showed the highest probability of developing new bone formation (OR 2.38 to 5.62, always statistically significant, figure 3C, D). This relationship was slightly weaker for the coexistence of VCFD and VCI at the same TP (OR 0.85 to 6.00, statistically significant in 5/8 scenarios).
Despite these associations, a large proportion of new bone still developed in VCs without visible MRI inflammation or fat deposition at any of the three assessed TPs (40–66%, depending on the combination of MRI and X-ray reader) (online supplementary table S3, percentages can be obtained by using syndesmophyte formation/growth rather than MRI lesions as denominator).
Multilevel GEE analyses
Overall, results of the GEE analyses (online supplementary tables S4 and S5) confirmed that both VCI and VCFD are associated with radiographic progression after 102 weeks of follow-up. Results were similar for the outcome syndesmophyte formation and for the outcome syndesmophyte formation/growth combined. The strength and significance of the associations varied depending on the MRI and radiographic case definition. Furthermore, GEE results confirmed that the sequential or simultaneous presence of VCI and VCFD further increases the probability of developing new bone formation, particularly when new VCFD is preceded by VCI (sequence analysis, VCI→VCFD→new bone formation). Some of the most consistent results were observed for VCI at any TP (adjOR 1.75 to 1.98), for VCFD at any TP (adjOR 1.60 to 2.32), for the sequential or simultaneous presence of VCI and VCFD across the three TP (adjOR 2.12 to 2.73) and for new VCFD preceded by VCI (adjOR 2.12 to 3.01). GEE analyses also confirmed that the absence of VCI or VCFD across all TPs ‘protects’ against new bone formation (adjOR 0.45 to 0.62). These associations were always statistically significant.
The other variables significantly associated with radiographic progression in the GEE multivariable analyses were gender and the presence of syndesmophytes/ankylosis at baseline (at the patient level). Table 2 shows the adjOR for the MRI case definitions most consistently associated with new bone formation as well as the adjOR of the adjustment factors (treatment, gender and the presence of syndesmophytes/ankylosis at baseline). The adjOR for the presence of syndesmophytes/ankylosis at baseline ranged from 2.81 to 5.65, always statistically significant. The adjOR for male gender ranged from 2.36 to 3.39, statistically significant in the majority of cases.
In the present study, we have confirmed that MRI VCI is associated with radiographic progression in AS, and we have shown that VCFD is also associated with radiographic progression. In addition, we have shown that the combination of fat and inflammation either at the same TP or sequentially further increases the probability of radiographic progression. Furthermore, VCFD that develops de novo can be preceded by VCI, and this sequence of events is also associated with progression of structural damage. However, VCI, VCFD and this particular sequence only partially explain the development of new bone in AS, as a large proportion of new syndesmophytes/bridging still occurred in VCs without either inflammation or fat deposition across three TPs.
The association between spinal MRI inflammation and radiographic progression after 2 years of follow-up has been reported in four previous studies, including ours.15–18 OR ranged from 1.7 to 8.6, and differences could be related to methodological aspects such as sample size, type of population (trial/observational cohort, TNF-blocker/standard treatment) and anatomical approach (VC/vertebral unit/vertebral edge). One study has indicated that new syndesmophytes are more likely to develop at VCs where inflammation has completely resolved than at VCs without inflammation at baseline or follow-up.18 It has also been suggested that VCs with persistent inflammation are less likely to develop new syndesmophytes.17 ,18
The relationship between spinal MRI inflammation, fat deposition and radiographic progression has been assessed in three previous AS studies.20–22 Chiowchanwisawakit et al20 showed that VCs that were simultaneously positive for inflammation and fat at baseline had an OR of 7.6 for the development of new syndesmophytes after 2 years of follow-up. Baseline fat and inflammation were both associated with radiographic progression in univariate analysis. However, in multivariable analysis, only the presence of VCI was associated with syndesmophyte formation (OR 5.8). Maksymowych et al22 studied ‘advanced VCI’ (defined by the presence of inflammation and concomitant fat, erosion or sclerosis), ‘early VCI’ (defined by the presence of inflammation only) and VCFD in relation to syndesmophyte formation. When adjusted for the baseline level of damage (at the patient level), only ‘advanced VCI’ (OR 3.9) and VCFD (OR 4.8) were associated with the development of new syndesmophytes after 2 years of follow-up. However, when both variables were tested in the same model, this association was statistically significant only for VCFD (OR 4.0). Finally, Baraliakos et al21 found that vertebral edges with both inflammation and fat deposition at baseline had the highest OR (3.7) for syndesmophyte formation after 5 years of follow-up (the reference being vertebral edges without either inflammation or fat at baseline and 2 years). Interestingly, in an axial spondyloarthritis population, Song et al19 described a significant relationship between the disappearance of inflammation (at the vertebral unit level and using whole-body MRI) and the development of fat deposition.
It is interesting to discuss our results in relation to the question whether TNF-blockers are capable of inhibiting the progression of structural damage in AS or not. The unexpected lack of inhibition of structural damage by TNF-blockers has fuelled the discussion about the relationship between inflammation and new bone formation.29 ,30 These initial trial data have recently been challenged by observational studies suggesting a protective effect of TNF-blockers on radiographic progression.31 ,32 However, these observational data have important methodological limitations, and this is still an unsolved question.8 Our observation that the sequence VCI→VCFD is valid and contributes to new bone formation in AS could be supportive of the hypothesis that TNF-blocker treatment in AS may only be effective in protecting from structural damage once newly developed VCI is prevented (after long-term treatment), while the immediate effects of TNF-blocker treatment could paradoxically contribute to new bone formation following the abrupt suppression of VCI and the development of VCFD at the same vertebral corner. However, this equation is more complex because the biological effects of TNF-blockers are not limited to the suppression of inflammation and TNF-blockers have also been associated with osteoproliferation in animal models.33
The occurrence of VCFD at baseline only was an infrequent event in our study, and the presence of VCFD at baseline only was not associated with new bone formation. It is possible that in the minority of cases where resolution of VCFD occurs the risk of developing new bone becomes similar to the risk in VCs that never had VCFD. We also observed that VCFD increased at follow-up compared with baseline. Since we analysed a TNF-blocker treated population, this finding would be consistent with the hypothesis that VCFD is more likely to develop at VCs where inflammation resolves compared with VCs with persistent or no inflammation at baseline/follow-up; alternatively, this finding could also simply reflect the natural history of disease, with VCFD being prone to increase over time, irrespective of inflammation.
Consistent with previous studies, we have shown that a significant part of new bone formation occurs in VCs without either traceable inflammation or fat deposition. However, this does not necessarily mean that these VCs do not have inflammation/fat deposition at the microscopic level because MRI may not be sensitive enough to capture all areas of inflammation/fat deposition34 and because the time between MRI assessments may not be short enough to capture the potential fluctuation of these lesions, particularly inflammation. Conversely, these results could suggest that the mechanisms of new bone formation in AS are still largely unknown and that the triggering of osteoproliferation may be completely or partially independent of inflammation (and fat deposition).7
Our study has limitations. First, we analysed a clinical trial cohort of patients with long-standing disease treated with TNF-blockers. Therefore, results cannot be generalised to patients in earlier disease stages or treated with first line treatments only. Second, two factors that have been shown to influence radiographic progression in AS—non-steroidal anti-inflammatory drug consumption35–37 and smoking status38—could not be adjusted for in our analyses because this information was not available in sufficient detail in the database. Finally, although we have performed MRI assessments at three TPs, additional assessments at shorter intervals may be needed to completely elucidate the dynamics of inflammation and fat deposition over time.
Strengths of our study are the uniquely large population of patients with AS, the large number of imaging readers, the fully unbiased nature of the imaging scoring, the fact that three MRI TPs were analysed (as opposed to one or two TPs as in the majority of previous studies), the fact that we have adjusted for multiple potential confounders using a statistical approach that adjusts for the dependence of observations in the same patient and the comprehensive list of scenarios (case definitions) that have been tested. These strengths make our study the most robust and comprehensive study investigating the relationship between VCI, VCFD and new bone formation to date.
In summary, we have shown that fat deposition in VCs (with or without concomitant inflammation) is associated with radiographic progression and that this association is stronger than for the presence of inflammation alone. Furthermore, inflammatory lesions can precede fat lesions, suggesting the possibility of a window of opportunity to prevent new bone formation. While the longitudinal absence of inflammation and fat deposition was negatively associated with radiographic progression, a significant proportion of new bone still developed at these sites. If this is indeed true and not a consequence of missing inflammation as described above, this suggests that inflammation/fat deposition and new bone formation may reflect independent pathways of the same disease, implying that new therapies specifically targeting osteoproliferation may need to be developed in order to prevent radiographic progression. Interesting future questions are how to incorporate MRI in future clinical trials and long-term observational studies, whether MRI criteria should be incorporated in future treat-to-target treatment strategies and whether new drugs with different mechanisms of action, such as drugs targeting the interleukin (IL)-23/IL-17 axis, will have a different effect in inflammation, fat deposition and structural damage.
We thank Janssen Biotech for providing the images and data necessary to perform these analyses. Janssen Biotech had no influence on data analyses to interpretation of results or writing of the article and did not provide any funding for this post hoc analysis of the ASSERT cohort.
Review history and Supplementary material
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
- Data supplement 1 - Online tables
Handling editor Tore K Kvien
Contributors PMM performed the statistical analysis and drafted the first version of the manuscript. PMM and XB scored the MRI scans. All authors participated in the design of the study to interpreted the results and contributed to drafting the final version of the manuscript. All authors read and approved the final manuscript.
Funding PMM was supported by the Fundação para a Ciência e a Tecnologia (FCT) grant SFRH/BD/62329/2009.
Competing interests PMM: consultancy/speaker fees from AbbVie to Centocor to Janssen to MSD to Novartis and Pfizer. XB: consultancy/speaker fees from AbbVie to BMS to Boehringer Ingelheim to Celgene to Centocor to Chugai to Janssen Biologics to Novartis to Pfizer to UCB. DvH: consultancy/speaker fees from AbbVie to Amgen to AstraZeneca to Augurex to BMS to Boehringer Ingelheim to Celgene to Centocor to Chugai to Covagen to Daiichi to Eli-Lilly to Galapagos to GSK to Janssen Biologics to Merck to Novartis to Novo-Nordisk to Otsuka to Pfizer to Roche to Sanofi-Aventis to UCB to Vertex. JB: consultancy/speaker fees from Abbott to Bristol Myers Squibb to Celgene to Celltrion to Chugai to Johnson & Johnson to MSD to Novartis to Pfizer to Roche to UCB Pharma. RL: consultancy/speaker fees from AbbVie to Amgen to AstraZeneca to BMS to Celgene to Centocor to Galapagos to GSK to Janssen Biologics to Merck to Novartis to Novo-Nordisk to Pfizer to Roche to TI-Genix to UCB. All authors have been involved in performing imaging reading services for clinical studies sponsored by different pharmaceutical industries.
Ethics approval Ethics committee approval was secured for the ASSERT trial. This study is an investigator-performed subanalysis of the ASSERT cohort.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.