Ankylosing spondylitis (AS) is a frequent chronic inflammatory rheumatic disease that already affects the axial skeleton at a young age 1, starting in the sacroiliac joints and later spreading to the spine 2. Active inflammatory spinal lesions as detected by magnetic resonance imaging (MRI) 3 and chronic structural changes such as syndesmophytes as demonstrated by conventional radiography 4 are characteristic of AS and contribute to both decreased spinal mobility and functional impairments of these affected patients 5. Conventional spinal x-rays are still the gold standard for assessment of structural changes in AS 67, whereas MRI techniques using either short-tau-inversion-recovery (STIR) sequences 28 or T1-post-gadolinium (T1-post-Gd) 9 are best for assessing spinal inflammation.

For quantification of structural spinal changes in conventional radiographs, the modified Stokes AS spinal score (mSASSS) 10 is the best currently available scoring method based on the OMERACT (Outcome Measures in Rheumatology) filter 11. For a sufficient sensitivity to change in depiction of structural spinal changes in AS when using conventional radiographs, a minimal observation period of 2 years is required 12. Similarly, for assessment and quantification of inflammatory spinal changes, the AS-spinal-MRI scoring system 13 has shown good discriminatory capacity, validity, and sensitivity to change in MRI examinations for periods of between 6 weeks 14 and 2 years 151617.

Tumour necrosis factor-alpha (TNF-α) plays a key proinflammatory role in AS 1819 given that spinal inflammation was shown to be associated with the presence of TNF-α mRNA 18 and protein 20 in affected joint and bone structures. Accordingly, inhibition of TNF-α was found to substantially improve signs and symptoms of AS patients 212223. Similarly, using MRI, a significant decrease of inflammatory lesions already after 6 weeks of therapy 14 and ongoing improvement of spinal inflammation for up to 2 years 1516 of continuous treatment have been reported. However, some inflammatory lesions were still present even after this period 151724.

Chronic changes in the thoracic spine cannot be reliably assessed by conventional x-rays but a valid quantification of such lesions is possible in the cervical and the lumbar spine 4. Since MRI is able to visualise the entire spine, it is now clear that the lower part of the thoracic spine is most frequently involved in AS 31725. This is one possible reason why so far it has not been possible to demonstrate major inhibition of structural damage in AS patients on anti-TNF therapy 262728. Nevertheless, a direct link between spinal inflammation and future radiographic progression has not been sufficiently proven until now. Data from animal models have even suggested that inflammation and new bone formation are uncoupled 2930. In this study, we examined the relationship of MRI-proven spinal inflammation at baseline (BL) with respect to structural deterioration depicted by conventional radiographs after 2 years in AS patients treated with anti-TNF-α agents.

Overall, conventional radiographs of 39 AS patients who were diagnosed according to the modified New York criteria for diagnosis of AS 31 were analysed. All patients had participated in clinical studies on anti-TNF therapy with infliximab (n = 26) 2132 or etanercept (n = 13) 24 for at least 2 years. All patients whose images were analysed had already signed informed consent forms for the radiographic images to be taken and analysed, according to the ethics committees of the participating centres which approved the original studies. None of these patients received antiresorptive bone therapy such as bisphosphonates or other drugs. Inclusion criteria were the availability of complete sets of MRIs with STIR and T1-post-Gd sequences and of conventional radiographs (see below) at the time point of presentation (BL) and after 2 years of follow-up (2yFU). All MRI and x-ray examinations were conducted using the same standardised protocol, as recently reported 41324.

The results of the present study suggest that spinal inflammation and new bone formation are both uncoupled and linked in AS since (a) the majority of syndesmophytes developed without MRI evidence of spinal inflammation at BL and (b) the proportion of VEs that developed syndesmophytes within 2 years was threefold higher when spinal inflammation was present at BL (compared with edges without BL inflammation). Since this was observed in patients under anti-TNF-α treatment, these data can be interpreted only on this basis. It will be important to study whether this is also true for patients just treated with nonsteroidal anti-inflammatory drugs or other agents. Nevertheless, the data may indicate that spinal inflammation was not sufficiently suppressed by TNF blockers in this 2-year time period.

In this study, the primary outcome was based on the analysis of VEs because the patterns of spinal inflammation and the development of syndesmophytes are likely to differ in and between individuals. It was no surprise, therefore, that when we did the analyses on an individual patient basis, no differences in the proportions of patients with and without spinal inflammation with respect to the development of future syndesmophytes were found. This may also be explained by the relatively small number of patients in this cohort. However, since there clearly were patients who developed syndesmophytes irrespective of BL inflammation, we do believe that it is more useful to do the calculations on the basis of VEs rather than on the patient level.

VEs that showed persistent inflammation seemed to be more prone to develop new syndesmophytes after 2 years as compared with those edges where inflammatory lesions disappeared after anti-TNF treatment. Indeed, in this study and in others, it has been shown that spondylitis may still be present after 2 years of anti-TNF therapy – even in patients with definite clinical improvement 1524. In addition, the analysis of only the edges that were inflammation-free at 2yFU showed that the tendency for the development of new syndesmophytes was stronger for those edges that showed inflammatory lesions at BL as compared with those edges that had not been inflamed in either the BL or the 2yFU. Nevertheless, our findings are in line with previous data of ours 2628 and of other groups 3527 showing that radiographic progression in AS is not or not completely inhibited by TNF blockers.

Since the key feature of AS, much unlike rheumatoid arthritis 36, is new bone formation rather than osteodestruction, there is reason to consider different mechanisms for structural change which appear on radiographs in these diseases. In AS, uncoupling of spinal inflammation and new bone formation has recently been suggested 37. The data of our study show that about 60% of all syndesmophytes that developed did not show inflammation as detected by MRI. Since the sensitivity of MRI to demonstrate spinal inflammation in AS is not precisely known 9, the question of whether it was possible to really detect all cases of spondylitis has to remain open and should be studied further. Furthermore, it cannot be excluded that inflammation has occurred at some point before and/or during the study. In this study, new spondylitis lesions developed in 8% of the VEs investigated. Recent immunohistological data showed low-grade spinal inflammation in biopsy specimens obtained at spinal surgery of AS patients who had undergone MRI before surgery, and no active inflammatory lesions had been detected by appropriate MRI sequences 38. Thus, since we did not perform MRIs in between, we do not know whether or for how long spinal inflammation may have occurred in the patients included in this study.

This is the first study based on patient data on this issue – even though we cannot exclude that the treatment of the patients had an impact on the results. Indeed, there is some evidence that blocking TNF-α may reverse the physiologic inhibition of osteoblast function and stimulate osteoclast resorption 39. TNF and other proinflammatory cytokines are known to promote bone formation by upregulating the expression of Dickkopf-1, a key target gene of TNF and an inhibitor of osteophyte regulators 29. Thus, by inhibiting TNF and Dickkopf-1, TNF blockers may even block negative influences on syndesmophyte formation after sufficient suppression of inflammation 29. The hypothesis that new bone formation in AS is uncoupled from inflammation has been supported by animal models showing that TNF inhibition did not affect joint ankylosis 30.

Recent biomarker data generated from ASSERT (Ankylosing Spondylitis Study for the Evaluation of Recombinant Infliximab Therapy) 2240 showed that previously low levels of osteocalcin and bone alkaline phosphatase were significantly increased under infliximab therapy 41. Furthermore, anti-TNF therapy was shown to decrease osteoclast precursor cells 42 and to increase bone mineral density 43 in AS patients. Thus, there is evidence from patient-derived data that anti-TNF agents increase bone mass. On the other hand, clinical experience may suggest that syndesmophytes grow especially at locations where spondylitic lesions had occurred. One example is the radiologic appearance of spondylitis anterior, the well-known shiny corners or Romanus lesions 44. Furthermore, it was already described some decades ago in histological studies that inflammatory spinal lesions precede new bone formation in AS patients 45. Our study shows that the likelihood that syndesmophytes developed was much higher for VEs with MRI evidence of inflammation than for those without (OR > 3). This suggests that there is some link between inflammation and new bone formation, even though that may not be a mandatory prerequisite for syndesmophyte development. Furthermore, as indicated by the analysis of T1-post-Gd sequences, which are more specific but not as sensitive as STIR in the depiction of spinal inflammation in AS 9, formation of new syndesmophytes occurred in VEs with persistent inflammation after 2 years (4.3%), whereas this was not the case in edges without persistent inflammation.

The nature and the length of the time interval between inflammation and new bone formation are unclear. Animal models imply that new bone formation in AS is mainly due to 'response to an inflammation-based bone-resorptive phase which serves as a stress factor' and is followed by enchondral new bone formation leading to bony bridges and vertebral fusion 37. Although it is unclear whether and how such findings are relevant for human disease, it is conceivable that there may be a disease stage at which new bone formation occurs without much actual inflammation; this, however, remains to be shown. Thus, it seems possible that both hypotheses are true; this implies that inflammation and new bone formation in AS are not completely uncoupled in AS, as recently proposed 37, but are at least partially linked.

While osteodestructive lesions in rheumatoid arthritis can already be inhibited by anti-TNF-α therapy after 1 year 36, inhibition of the osteoproliferation in patients with AS may need longer treatment 28. However, since it was shown that the spinal inflammation is not completely inhibited by anti-TNF therapy in this and other studies after 2 years 3527, there are also other factors 4 to be considered to explain this major difference to response to therapy between these two diseases. This includes the fact that only historical cohorts are currently available for comparison in relevant studies 3527.

In summary, in patients treated with anti-TNF-α, new bone formation occurred almost threefold more often in regions with MRI-proven spinal inflammation at BL, and, in the same cohort, most of the newly developed syndesmophytes occurred in VEs without evidence of inflammation at BL. These findings suggest both a link and some dissociation of inflammation and radiographic damage. There is no evidence for a major uncoupling of these characteristic features in AS. Thus, it seems still possible that more effective suppression of spinal inflammation may lead to a stronger inhibition of structural damage in AS.

In patients treated with anti-TNF-α, new bone formation seems to occur almost threefold more often in regions with MRI-proven spinal inflammation at BL. But, similarly, some of the newly developed syndesmophytes may also occur in VEs without evidence of inflammation at BL. These findings suggest both a link and some dissociation of inflammation and radiographic damage. There is no evidence for a major uncoupling of these characteristic features in AS. It seems still possible that more effective suppression of spinal inflammation may lead to a stronger inhibition of structural damage in AS.

2yFU: 2-year follow-up; AS: ankylosing spondylitis; BL: baseline; CI: confidence interval; MRI: magnetic resonance imaging; OR: odds ratio; STIR: short-tau-inversion-recovery; T1-post-Gd: T1-post-gadolinium; TNF: tumour necrosis factor; VE: vertebral edge.

The authors declare that they have no competing interests.

XB helped to conceive of the idea for the study, prepared the data and performed data analysis, and helped to write the manuscript. JL analysed the data, performed the statistical evaluation, and helped to write the manuscript. MR and JS helped to write the manuscript. JB helped to conceive of the idea for the study and to write the manuscript. All authors read and approved the final manuscript.