Inflammatory lesions of the spine on magnetic resonance imaging predict the development of new syndesmophytes in ankylosing spondylitis: Evidence of a relationship between inflammation and new bone formation

Abstract

Objective

To determine whether a vertebral corner that demonstrates an active corner inflammatory lesion (CIL) on magnetic resonance imaging (MRI) in patients with ankylosing spondylitis (AS) is more likely to evolve into a de novo syndesmophyte visible on plain radiography than is a vertebral corner that demonstrates no active inflammation on MRI.

Methods

MRI scans and plain radiographs were obtained for 29 patients recruited into randomized placebo-controlled trials of anti–tumor necrosis factor α (anti-TNFα) therapy. MRI was conducted at baseline, 12 or 24 weeks (n = 29), and 2 years (n = 22), while radiography was conducted at baseline and 2 years. A persistent CIL was defined as a CIL that was found on all available scans. A resolved CIL was defined as having completely disappeared on either the second or third scan. A validation cohort consisted of 41 AS patients followed up prospectively. Anonymized MRIs were assessed independently by 3 readers who were blinded with regard to radiographic findings.

Results

New syndesmophytes developed significantly more frequently in vertebral corners with inflammation (20%) than in those without inflammation (5.1%) seen on baseline MRI (P ≤ 0.008 for all reader pairs). They also developed more frequently in vertebral corners where inflammation had resolved than in those where inflammation persisted after anti-TNF treatment. This was confirmed in the analysis of the prospective cohort, in which significantly more vertebral corners with inflammation (14.3%) compared with those without inflammation (2.9%) seen on baseline MRI developed new syndesmophytes (P ≤ 0.003 for all reader pairs).

Conclusion

Our findings indicate that a syndesmophyte is more likely to develop from a prior inflammatory lesion, supporting a relationship between inflammation and ankylosis.

Ankylosing spondylitis (AS) is a chronic inflammatory disorder that primarily targets the joints of the axial spine and entheses. A hallmark of the disease is new bone formation in the spine, which typically leads to ankylosis across disc spaces and is thought to follow the onset of inflammation (1). Bone proliferation typically starts at the vertebral rim and grows in the direction of the annulus fibrosus, perpendicular to the vertebral axis (2). Ossification may extend across the vertical length of the disc and becomes visible on radiographs as a syndesmophyte.

The hypothesis that such ankylosis is invariably preceded by inflammation has, in fact, never been proven by prospective evaluation, and it remains possible that ankylosis may develop through noninflammatory pathways, as implied in patients with diffuse idiopathic skeletal hyperostosis. Formal proof of this hypothesis has not been feasible until now, since this requires a noninvasive approach to the prospective assessment of inflammatory spinal lesions. Testing of this hypothesis has assumed increased importance since it was demonstrated that antiinflammatory interventions, such as anti–tumor necrosis factor (anti-TNF) agents, might not prevent the development of spinal ankylosis (3, 4).

Magnetic resonance imaging (MRI) is now well established as the most sensitive imaging modality for detection of active inflammatory changes in the spine and sacroiliac joints of patients with AS. Active inflammatory lesions in the vertebral bodies are evident as increased signal on STIR sequences at the vertebral corners and adjacent to vertebral end plates, reflecting the increased water signal associated with inflammatory edema (5). The former are thought to reflect Romanus lesions, and the latter are thought to reflect spondylodiscitis. The results of a study of 32 patients with spondylarthritis (SpA) suggested that the severity of MRI change is correlated with histopathologic grading of inflammation (6). Reduction in active lesions is readily visible on MRI within a few weeks following initiation of anti-TNF treatment (7, 8), and although the effects of treatment are sustained, extended followup evaluation shows that lesions may not resolve completely with treatment (9, 10). The combination of MRI and administration of anti-TNF therapy therefore offers an opportunity for noninvasive study of the effects of both persistent and resolved vertebral inflammation on the subsequent development of new bone formation.

The development of new bone in the spine in the form of syndesmophytes and ankylosis is still evaluated by plain radiography (11, 12). A minimum of 2 years is required before radiographic changes can be reliably detected, and predictors of progression are limited to baseline radiographic damage and serum levels of matrix metalloproteinase 3 (13–15). Because development of new syndesmophytes is the clinically most relevant feature and is also predictive of further radiographic progression, it has been proposed as a new gold standard for measurement of radiographic change (16). In this study, we directly tested the hypothesis that a vertebral corner that demonstrates an active inflammatory lesion on MRI using the STIR sequence is more likely to evolve into a de novo syndesmophyte than is a vertebral corner that demonstrates no active inflammation on MRI.

PATIENTS AND METHODS

Patients.

The study was conducted using images obtained from 2 cohorts of patients who had AS according to the modified New York criteria (17). In the first cohort, MRI and plain radiography of the spine were conducted in 29 patients who were recruited into 3 randomized, phase III, placebo-controlled trials of anti-TNFα therapy. MRI scans were obtained at baseline, the primary end point (12 or 24 weeks), and 52 weeks or 2 years, and plain radiographs were obtained at baseline and 2 years. In the first trial (4), patients were randomized to receive active treatment (infliximab) or placebo for 24 weeks, followed by open-label active treatment from 24 weeks to 2 years, and had MRI assessment at baseline, 24 weeks, and 2 years. In the second trial (10), patients were randomized to receive active treatment (adalimumab) or placebo for 12 weeks, followed by open-label active treatment from 12 weeks to 24 weeks (early escape) or up to 2 years, and had MRI assessment at baseline, 12 weeks, and 52 weeks. In the third trial (18), patients were randomized to receive active treatment (infliximab) or placebo for 12 weeks, followed by open-label treatment from 12 weeks to 2 years, and had MRI assessment at baseline and 12 weeks. Inclusion criteria for these trials were very similar, with a Bath Ankylosing Spondylitis Disease Activity Index (BASDAI) (19) score of ≥4 being a common requirement.

A second cohort consisted of 41 patients with AS who were followed up prospectively as part of a larger longitudinal observational cohort, and for whom MRI obtained at baseline and plain radiographs obtained at baseline and 2 years were available. Of these patients, 23 had received standard therapy (such as nonsteroidal antiinflammatory drugs or physical therapy), and 18 had received anti-TNF therapy. MRI assessment was conducted using a standardized imaging protocol (available online at www.arthritisdoctor.ca) at baseline for all patients. Clinical assessments (BASDAI, Bath Ankylosing Spondylitis Functional Index [20], patient global assessment, total back pain, and nocturnal back pain) and laboratory assessments (C-reactive protein [CRP] level) were performed at baseline and repeated every 6 months.

Standardized definitions of spinal lesions on MRI.

In reading the MRIs, we adhered to the standardized definitions of acute and chronic spinal lesions observed on MRI in patients with AS developed by the Spondyloarthritis Research Consortium of Canada (SPARCC) as well as an international working group of rheumatologists and radiologists from Canada and Denmark. For the purposes of this study, we focused on the identification of acute inflammatory lesions at the vertebral corners. Consequently, we based our MRI readings on the following definitions developed by SPARCC and the Canada/Denmark International MRI Working Group (21). An inflammatory lesion in the vertebral body is defined as increased signal in the bone marrow on STIR or T2-weighted, fat-suppressed sequences, where the bone marrow signal in the center of the vertebra, if normal, constitutes the reference for designation of normal signal. An active vertebral corner inflammatory lesion (CIL) is defined as an increased STIR signal at a vertebral corner that is present in at least 1 sagittal slice, which includes the spinal canal. An example is shown in Figure 1.

Figure 1. Sagittal STIR magnetic resonance images (MRIs) (repetition time 3,150 msec, echo time 51 msec, inversion time 140 msec) of the cervical spine of a 40-year-old man with ankylosing spondylitis. A and B, Two adjacent images from the baseline MRI obtained prior to anti–tumor necrosis factor therapy, demonstrating definite anterior corner inflammatory lesions (CILs) at the C3 upper, C3 lower, C4 upper, and C4 lower levels (arrows). Results were agreed upon by all reader pairs. C and D, Followup images obtained 2 years later, demonstrating no active inflammation. The CILs had resolved. E, Baseline radiograph (obtained at the same time as the baseline MRI), demonstrating a subtle anterior syndesmophyte at the C5 upper end plate only (arrow). F, Followup radiograph obtained 2 years later (at the same time as the followup MRI), demonstrating the formation of new syndesmophytes at the C3 upper, C3 lower, C4 upper, and C4 lower end plates (arrows). The C5 upper syndesmophyte was much more conspicuous. All 4 vertebral corners with inflammation on baseline MRI (agreed upon by all reader pairs) demonstrated resolution of inflammation on followup MRI and radiographic development of new syndesmophytes. Vertebral corners without evidence of inflammation on baseline MRI did not develop new syndesmophytes during the 2-year followup period.

Reading of images.

A unique MRI study number was assigned to each patient, thereby ensuring that investigators were blinded with regard to all patient demographic characteristics and treatment. Computer-generated random numbers were assigned by a technologist who was not connected with the study.

Assessment was performed on a 3-monitor review station by 2 readers using computer software (Merge eFilm, Milwaukee, WI) that is optimal for this type of review, and viewing conditions were standardized. Plain radiographs and MRIs were anonymized, assigned a random number, and read independently. Since most subjects had 3 MRI scans, the scans were coded in chronological order, and all 3 scans were read at the same time. This allowed readers to more accurately record change between time points. The anterior vertebral corners of the cervical (C2 lower to T1 upper) and lumbar (T12 lower to S1 upper) spine were examined for new syndesmophytes by comparing baseline and 2-year lateral radiographs of the cervical and lumbar spine. Two readers examined the radiographs, and scores were determined by consensus. The thoracic spine was not assessed in scoring radiographic changes, due to inadequate visualization (12).

The MRIs were read independently by 3 readers who recorded the presence or absence of a CIL at the anterior vertebral corners of the cervical (C2 lower to T1 upper) and lumbar (T12 lower to SI upper) spine. MRIs for all available time points per patient were read simultaneously by each reader, so that the presence or absence of a CIL on the baseline image as well as persistence, complete resolution, or the appearance of new CILs at subsequent time points could be recorded.

Statistical analysis.

All of the analyses focused on data for which there was concordance between readers in each of the 3 reader pairs. The primary analysis was a comparison of the proportion of new syndesmophytes developing at each anterior vertebral corner in those corners where a CIL was recorded as being present at the baseline examination by both MRI readers in each reader pair, versus those corners where neither of the 2 readers recorded a CIL at baseline. Vertebral corners that already demonstrated syndesmophytes or ankylosis at the baseline examination were therefore excluded from this analysis. In a secondary analysis, we also excluded vertebral corners that demonstrated any radiographic abnormality (i.e., squaring, sclerosis, or erosion) at baseline.

We conducted 2 further secondary analyses in the clinical trial cohort. First, we compared the proportion of new syndesmophytes developing at each anterior vertebral corner from a persistent CIL at the corresponding vertebral corner with the proportion of new syndesmophytes developing at each anterior vertebral corner where there was no prior CIL. Second, we compared the proportion of new syndesmophytes developing at each anterior vertebral corner from a completely resolved CIL at the corresponding vertebral corner with the proportion of new syndesmophytes developing at each anterior vertebral corner where there was no prior CIL.

A CIL was defined as being persistent if it was recorded as being present on each MRI scan (obtained at baseline, primary end point, and extended followup) by both readers in each reader pair. In 1 of the 3 clinical trials, only 2 MRI scans were conducted (at baseline and at the primary end point at 12 weeks), and for this trial persistence was defined in the anti-TNF group as a CIL that was present on both MRI scans. Data from patients taking placebo in this trial were not included, since these patients received open-label anti-TNF therapy beginning at 12 weeks, and a third MRI scan was not available for assessment of persistence.

A CIL was defined as being completely resolved if it was recorded as being present at baseline and then absent after the introduction of anti-TNF therapy by both readers in each reader pair. Since patients randomized to receive placebo received open-label anti-TNF only after the primary end point (12 or 24 weeks), a CIL was defined as being completely resolved in these patients if the first MRI, conducted at baseline, and the second MRI, conducted at the primary end point, showed a CIL that was no longer present on the third and final MRI. As mentioned above, data from patients who received placebo in 1 of the trials were not included in this analysis, due to lack of a third MRI scan.

Pearson's chi-square test was used to compare proportions. We also assessed interreader reliability for detection of CILs by calculating the percentage agreement and Cohen's unweighted kappa. Kappa values of >0.75, 0.40–0.75, and <0.40 were designated as representing excellent, moderate, and poor reliability, respectively.

RESULTS

Patient demographics.

Baseline characteristics of the 2 study populations are shown in Table 1. Because inclusion criteria and baseline characteristics for the 3 clinical trials were very similar, the baseline data for the clinical trial cohort represent a summation of all patients recruited to these trials. As expected, patients recruited to clinical trials had clinically more severe disease than those in the observational cohort. Baseline clinical characteristics were comparable between the anti-TNF and placebo arms of the clinical trial cohort, with the exception of a shorter disease duration and higher CRP level in patients receiving placebo (P < 0.05 for both). In the observational cohort, the patient global assessment and CRP level were higher in patients who received anti-TNF therapy than in those who received standard therapy, but the differences were not significant.

Table 1. Baseline characteristics and data on radiographic progression over 2 years in patients with AS recruited to phase III trials of anti-TNF therapy or an observational cohort*

	Clinical trial cohort		Observational cohort
	Anti-TNF therapy (n = 18)	Placebo (n = 11)	Anti-TNF therapy (n = 18)	Standard therapy (n = 23)
* Except where indicated otherwise, values are the mean ± SD. AS = ankylosing spondylitis; anti-TNF = anti–tumor necrosis factor; BASDAI = Bath Ankylosing Spondylitis Disease Activity Index; BASFI = Bath Ankylosing Spondylitis Functional Index; CRP = C-reactive protein.
Sex, no. male/female	14/4	8/3	11/7	21/2
Age, years	44.4 ± 9.3	43.7 ± 11.6	42.5 ± 10.8	42.3 ± 15.5
Disease duration, years	23.1 ± 10.8	16.7 ± 7.4	16.9 ± 9.7	16.0 ± 13.4
Total back pain score	7.2 ± 1.2	7.9 ± 1.5	5.3 ± 3.3	4.6 ± 2.1
Nocturnal back pain score	7 ± 1.6	7.1 ± 1.8	5.3 ± 3.3	4.2 ± 2.3
BASDAI score	6.8 ± 1.0	7.2 ± 1.4	5.8 ± 2.4	6.0 ± 2.0
BASFI score	6.5 ± 1.8	6.8 ± 2.1	5.6 ± 2.2	5.3 ± 2.6
Patient global score	7.5 ± 1.2	7.8 ± 1.0	5.2 ± 3.1	4.3 ± 2.2
CRP level, mg/dl	16.2 ± 16.1	27.5 ± 21.6	16.5 ± 16.5	10.2 ± 15.6
No. of new syndesmophytes	1.3 ± 1.9	0.8 ± 1.1	0.5 ± 0.9	0.8 ± 1.5
Median	0	0	0	0
Range	0–6	0–3	0–3	0–6

Development of new syndesmophytes.

Data were unavailable for 53 and 17 vertebral corners in the clinical trial anti-TNF and placebo groups, respectively, and for 62 and 48 vertebral corners in the observational cohort anti-TNF and standard therapy groups, respectively, primarily due to lack of radiographic visualization of the lower part of the cervical spine. The number and percentage of patients who developed ≥1 new syndesmophyte after 2 years was 8 (44.4%) and 5 (45.5%) in the anti-TNF and placebo arms of the clinical trial cohort, respectively, and 5 (27.8%) and 8 (34.8%) in the anti-TNF and standard therapy arms of the observational cohort, respectively (P > 0.05 for all).

The total number and percentage of new syndesmophytes developing from vertebral corners after 2 years was 23 (7.5%) of 306 vertebral corners and 9 (4.8%) of 187 vertebral corners in the anti-TNF and placebo arms of the clinical trial cohort, respectively, and 9 (3.3%) of 269 vertebral corners and 18 (4.3%) of 419 vertebral corners in the anti-TNF and standard therapy arms of the observational cohort, respectively. The mean ± SD number of new syndesmophytes per patient was 1.3 ± 1.9 and 0.8 ± 1.1 in the anti-TNF and placebo groups in the clinical trial cohort, respectively, and 0.5 ± 0.9 and 0.8 ± 1.5 in the anti-TNF and standard therapy arms of the observational cohort, respectively. Treatment group comparisons of new syndesmophytes within each cohort did not show significant differences.

Descriptive data on vertebral corner inflammatory lesions.

Descriptive data on the total number per group and mean frequency of CILs per patient recorded concordantly by 1 of the reader pairs (WPM and RGWL) are shown in Table 2. At baseline, the mean frequency of CILs per patient was higher in the anti-TNF groups in each cohort, but differences between treatment groups were not significant. At the primary end point (second MRI), the reduction in the number of CILs following anti-TNF therapy (−65.6%) was significant compared with placebo treatment (0%) in the clinical trial cohort (P = 0.004). The number of CILs increased in the placebo group because 3 new CILs were recorded in 2 patients at the second MRI assessment.

Table 2. Descriptive data on the number of CILs recorded concordantly in the anterior cervical and lumbar spine by 2 experienced readers (WPM, RGWL) in 2 cohorts of AS patients at baseline and after initiation of anti-TNF therapy*

	Clinical trial cohort		Observational cohort
	Anti-TNF therapy (n = 18)	Placebo (n = 11)	Anti-TNF therapy (n = 18)	Standard therapy (n = 23)
* CILs = active corner inflammatory lesions; IQR = interquartile range; NA = not assessed (see Table 1 for other definitions). † A third magnetic resonance image (MRI) was obtained at either 52 or 104 weeks in 21 patients, of whom 12 and 9 patients were originally randomized to the anti-TNF and placebo groups, respectively, but then received anti-TNF between the second and third MRI scans.
Baseline MRI
Total number of CILs	32	10	23	15
Number of CILs per patient
Mean ± SD	1.8 ± 2.2	0.9 ± 1.6	1.3 ± 2.1	0.7 ± 0.9
Median (IQR)	0.5 (0–3)	0 (0–2)	1 (0–1)	0 (0–2)
Range	0–6	0–5	0–9	0–2
Second MRI
Total number of CILs	11	13	NA	NA
Number of CILs per patient
Mean ± SD	0.6 ± 1.6	1.2 ± 1.8	NA	NA
Median (IQR)	0 (0–0)	0 (0–2)	NA	NA
Range	0–6	0–6
Number of new CILs	0	3	NA	NA
Third MRI†
Total number of CILs	3	1	NA	NA
Number of new CILs	2	0	NA	NA

A third MRI examination was conducted in 21 patients after extended followup (52 or 104 weeks). Of these 21 patients, 1 patient developed 2 new CILs between the second and third MRI, despite receiving anti-TNF therapy for the entire study period (52 weeks). One patient had a single persistent CIL at all MRI visits despite receiving anti-TNF therapy for the entire study period (52 weeks). One patient orginally randomized to receive placebo had a persistent CIL despite receiving open-label anti-TNF therapy after 12 weeks. Otherwise, all of the CILs in these 21 patients had resolved by the third MRI. A similar baseline frequency of CILs was recorded in patients receiving anti-TNF therapy in the observational cohort, with fewer CILs in patients receiving standard therapy, although these differences were not significant.

Reliability of detection of CILs.

Interreader reliability for detection of CILs was comparable among all reader pairs and moderate for both the baseline and second MRI. (Table 3). Reliability was also moderate for recording persistent CILs between the first and second MRI but was not as good for detection of resolved CILs.

Table 3. Interreader reliability of detection of CILs*

	WPM vs. RGWL		WPM vs. PC		RGWL vs. PC
	Kappa statistic	% agreement	Kappa statistic	% agreement	Kappa statistic	% agreement
* Magnetic resonance images (MRIs) were analyzed for corner inflammatory lesions (CILs) in the anterior cervical and lumbar spine of patients recruited to placebo-controlled clinical trials of anti–tumor necrosis factor (anti-TNF) therapy or an observational cohort where they received either anti-TNF or standard therapy.
Baseline MRI
Clinical trial cohort	0.51	88.5	0.51	87.1	0.53	87.1
Observational cohort	0.59	95.2	0.55	94.3	0.42	92.5
All patients	0.55	92.4	0.53	91.3	0.48	90
Second MRI
Clinical trial cohort	0.54	93.1	0.46	89.5	0.47	90.4
Resolved CILs	0.38	89.8	0.28	86.5	0.34	86.5
Persistent CILs	0.50	95	0.46	93.8	0.45	94

Association between CILs and development of new syndesmophytes in the clinical trial cohort.

New syndesmophytes developed significantly more frequently in those vertebral corners with active inflammation seen on baseline MRI than in those without active inflammation on baseline MRI, in both patient cohorts (Tables 4 and 5). In the primary analysis of the clinical trial cohort, using concordant data from 1 reader pair (WPM and RGWL), 6 (20%) of 30 vertebral corners with active inflammation on baseline MRI were associated with the development of a new syndesmophyte at the corresponding vertebral corner, compared with 19 (5.1%) of 370 vertebral corners with no active inflammation on baseline MRI (P = 0.001). The corresponding percentages for the other 2 reader pairs were 17.1% for vertebral corners with active inflammation versus 5.3% for vertebral corners without active inflammation (P = 0.006) and 16.7% for vertebral corners with active inflammation versus 5.3% for vertebral corners without inflammation (P = 0.008).

Table 4. Development of new syndesmophytes at 2 years in the anterior cervical and lumbar spine in a clinical trial cohort of 29 patients with AS receiving anti-TNF therapy*

	New syndesmophytes		P
	Yes	No
* Values are the number (%) of vertebral corners associated with the presence or absence of new syndesmophytes. Patients were stratified according to the presence or absence of a corner inflammatory lesion (CIL) at the corresponding vertebral corner on baseline magnetic resonance imaging (MRI) and according to persistence or resolution of the CIL on followup MRI scans. Concordant data from each reader pair are shown. NS = not significant (see Table 1 for other definitions). † Versus vertebral corners without CILs. ‡ Comparison of new syndesmophytes developing in vertebral corners with a resolved CIL on followup MRI versus those vertebral corners with no CIL on any MRI scan. § Versus no CIL.
Baseline MRI
Reader pair WPM and RGWL
Vertebral corner CIL+	6 (20)	24 (80)	0.001†
Vertebral corner CIL−	19 (5.1)	351 (94.9)
Reader pair WPM and PC
Vertebral corner CIL+	6 (17.1)	29 (82.9)	0.006†
Vertebral corner CIL−	19 (5.3)	342 (94.7)
Reader pair PC and RGWL
Vertebral corner CIL+	6 (16.7)	30 (83.3)	0.008†
Vertebral corner CIL−	19 (5.3)	338 (94.7)
Followup MRI
Reader pair WPM and RGWL
Resolved CIL	5 (25.0)	15 (75.0)	<0.0001‡
Persistent CIL	0 (0)	7 (100)	NS§
No CIL	17 (4.9)	327 (95.1)
Reader pair WPM and PC
Resolved CIL	5 (31.3)	11 (68.7)	<0.0001‡
Persistent CIL	0 (0)	12 (100)	NS§
No CIL	18 (5.3)	320 (94.7)
Reader pair PC and RGWL
Resolved CIL	5 (22.3)	17 (77.7)	0.001‡
Persistent CIL	0 (0)	7 (100)	NS§
No CIL	17 (5.1)	315 (93.9)

Table 5. Development of new syndesmophytes at 2 years in the anterior cervical and lumbar spine in an observational cohort of 41 patients with AS receiving anti-TNF therapy or standard therapy*

	Standard therapy (n = 23)†			Anti-TNF therapy (n = 18)			All patients (n = 41)
	New syndesmophytes		P‡	New syndesmophytes		P‡	New syndesmophytes		P‡
	Yes	No		Yes	No		Yes	No
* Values are the number (%) of vertebral corners associated with the presence or absence of new syndesmophytes. Patients were stratified according to the presence or absence of a corner inflammatory lesion (CIL) at the corresponding vertebral corner on baseline magnetic resonance imaging. Concordant data per reader pair are shown. See Table 1 for other definitions. † Standard therapy included nonsteroidal antiinflammatory drugs and physical therapy. ‡ Versus vertebral corners without CILs.
Reader pair WPM and RGWL
Vertebral corner CIL+	2 (12.5)	14 (87.5)	0.08	2 (16.7)	10 (83.3)	0.001	4 (14.3)	24 (85.7)	0.001
Vertebral corner CIL−	14 (3.7)	369 (96.3)		4 (1.7)	230 (98.3)		18 (2.9)	599 (97.1)
Reader pair WPM and PC
Vertebral corner CIL+	2 (11.8)	15 (88.2)	0.08	2 (14.3)	12 (85.7)	0.008	4 (12.9)	27 (87.1)	0.003
Vertebral corner CIL−	13 (3.4)	366 (96.6)		5 (2.1)	229 (97.9)		18 (2.9)	595 (97.1)
Reader pair PC and RGWL
Vertebral corner CIL+	3 (20)	12 (80)	0.002	3 (33.3)	6 (66.7)	<0.0001	6 (25)	18 (75)	<0.0001
Vertebral corner CIL−	14 (3.7)	369 (96.3)		3 (1.3)	229 (98.7)		17 (2.8)	598 (97.2)

In one set of secondary analyses, we focused on vertebral corners that were radiographically normal at baseline, i.e., that showed no squaring, sclerosis, or erosions. For concordant data from 1 reader pair (WPM and RGWL), 6 (20%) of 30 vertebral corners with a CIL at baseline developed a new syndesmophyte, compared with 14 (5.0%) of 279 vertebral corners that were normal on MRI and radiography (P = 0.001). In another set of secondary analyses, we focused on vertebral corners where the CIL either persisted on all MRI scans or resolved following initiation of anti-TNF therapy. For the reader pair WPM and RGWL, the percentages of vertebral corners associated with the development of new syndesmophytes were 25%, 0%, and 4.9% for those vertebral corners with resolved CILs, persistent CILs, and no prior CILs, respectively (Table 4). Percentages were similar for the 2 other reader pairs. The differences between those vertebral corners with resolved CILs and those with no prior CILs were highly significant for all 3 reader pairs (P ≤ 0.001).

Association between CILs and development of new syndesmophytes in the observational cohort.

Similar to findings in the clinical trial cohort, new syndesmophytes developed significantly more frequently in those vertebral corners with active inflammation on baseline MRI than in those without active inflammation on baseline MRI in the observational cohort (Table 5). This difference was somewhat more evident in patients who received anti-TNF therapy, where the difference between vertebral corners with and without inflammation was significant for all 3 reader pairs, while this was evident for only 1 reader pair in patients who received standard therapy.

DISCUSSION

Three main conclusions can be drawn from the results of this analysis. First, the presence of an active inflammatory lesion at the vertebral corner on the MRI STIR sequence is associated with the subsequent development of new syndesmophytes at the corresponding vertebral corner, which are visible on plain radiography. Second, new syndesmophytes will develop from a prior active inflammatory lesion even when there is agreement between readers that the lesion has completely resolved on followup MRI following the initiation of anti-TNF therapy. Third, a new syndesmophyte may also develop where both the baseline plain radiograph and the baseline MRI show a completely normal vertebra.

The finding that vertebral corners with active inflammatory lesions may ultimately develop new syndesmophytes even when complete resolution of the inflammatory lesion is evident on MRI after anti-TNF therapy contradicts, at first sight, the hypothesis that inflammation and ankylosis are related. However, several considerations should be addressed in the interpretation of this finding. First, even though MRI shows what appears to be complete resolution of inflammation, it is possible that inflammation is still present at a histopathologic level, and this may be associated with the development of syndesmophytes. The sensitivity of STIR MRI for the detection of inflammation has been addressed in a previous study, in which the spine was scanned prior to spinal surgery, and the facet joints were removed and analyzed histopathologically (22). Inflammation was evident on MRI in only 3 of 8 patients with histopathologic features of inflammation. Second, several studies have shown that conventional doses of anti-TNF agents do not completely suppress the inflammation seen on MRI in patients with SpA (9, 10, 23). This implies that higher doses and/or more frequent administration of anti-TNF agents might be warranted in patients with AS with radiographic progression. Alternatively, the residual inflammatory process may reflect non–TNF-regulated pathways of inflammation (24–26).

The finding that resolved CILs but not persistent CILs are associated with new syndesmophytes supports an alternative hypothesis regarding the link between inflammation and ankylosis (27). We have termed this the “TNF brake hypothesis”. Histopathologic studies in AS show that once inflammation is well established, chondrogenesis and osteogenesis often occur simultaneously with inflammatory changes at any particular site in the axial skeleton. Proinflammatory cytokines such as TNFα have been shown to stimulate expression of specific bone morphogenetic proteins (BMPs), which have been shown to induce new bone formation (28). However, TNFα also stimulates expression of dickkopf-1, a major regulator of joint remodeling through its suppression of signaling by Wnt proteins, which are key mediators of osteoblastogenesis (27). Once inflammation resolves, either spontaneously or through pharmacologic suppression of TNFα, this may allow signaling through Wnt, and potentially other growth factors, such as BMPs, to induce new bone formation. Consequently, although TNFα triggers pathways leading to new bone formation, it may act primarily as a brake on new bone formation through dickkopf-1 in an established inflammatory lesion. Resolution of the CIL by anti-TNF therapy may allow tissue repair to become manifest as new bone formation, while persistence of the CIL may preclude new bone formation. This hypothesis would be consistent with observations in the ankylosing enthesitis model of SpA in DBA/1 mice, in which etanercept did not prevent ankylosis (29). This notion is also supported by the lack of impact of anti-TNF agents on radiographic progression in established AS over 2 years (3, 4). However, if these agents prevent the development of new inflammatory lesions, it is possible that radiographic progression will be suppressed over longer time frames. Studies of the natural history of disease with MRI followup examination would therefore provide important information on the fate of the resolved CIL, but since this is not ethically feasible, additional long-term studies using different antiinflammatory strategies, particularly in early AS prior to the appearance of extensive inflammatory lesions in the spine, are warranted to address this question.

The development of new syndesmophytes in vertebrae that appear normal on MRI and plain radiography at baseline suggests either the presence of underlying inflammation that is undetectable by MRI or a role for a noninflammatory pathway. Moreover, it remains possible that the association between inflammation and new syndesmophytes is not evident in all patients, and that noninflammatory mechanisms are important in some patients due to different pathophysiologic mechanisms. The total number of patients developing new syndesmophytes in the present study was low (13 each in the clinical and observational cohorts), so the study was not sufficiently powered to address this question. Idiopathic skeletal hyperostosis is often cited as an example of a noninflammatory condition that is associated with new cartilage and bone formation in the spine. However, this has not been studied systematically in a prospective manner using MRI. Biomechanical factors may also be relevant. Previous work has shown that facet joint fusion precedes the development of vertebral ankylosis, suggesting that immobilization of a spinal unit may be a factor in the development of vertebral ankylosis (30). The facet joints were not examined in the present study.

In this study, we focused on the assessment of inflammatory lesions that were concordantly detected by 2 readers, although inflammatory lesions were predictive of the development of new syndesmophytes even when the data from single readers were analyzed (data not shown). The importance of this approach is highlighted in our assessment of reliability, which showed only moderate interreader reliability for the detection of inflammatory lesions, even between experienced readers (WPM and RGWL). This reflects the fact that some lesions are small and/or demonstrate only a slight increase in signal intensity on STIR sequences. Reader detection of normal and abnormal signals can be improved if normal signal intensity is defined and the definition systematically applied, and reference images depicting typical inflammatory lesions as well as lesions considered to be at the threshold of detection are used. In addition, MRI is subject to physiologic motion artifact, such that flowing blood in the inferior vena cava and the abdominal aorta may cause a spurious signal that mimics anterior vertebral corner inflammatory lesions in the lumbar spine (5).

In conclusion, we tested the hypothesis that a vertebral corner that demonstrates an active inflammatory lesion that is visible on MRI using the STIR sequence is more likely to evolve into a de novo syndesmophyte visible on plain cervical and lumbar spine radiographs after 2 years than is a vertebral corner that demonstrates no abnormality on MRI. Data acquired from assessments by 3 readers from both a clinical and an observational cohort of patients with AS exposed to either anti-TNF or standard therapies consistently support this hypothesis. However, we also showed that de novo syndesmophytes are more likely to develop at vertebral corners that demonstrate complete resolution of inflammation on followup MRI following the initiation of anti-TNF therapy as compared with vertebral corners that demonstrate persistent inflammation. Our data therefore support the concept of a relationship between inflammation and ankylosis but also indicate a model in which TNF not only triggers signaling pathways for bone formation, but also activates a dominant regulator of joint remodeling that suppresses new bone formation.

AUTHOR CONTRIBUTIONS

Dr. Maksymowych had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study design. Maksymowych, Pedersen, Østergaard, Lambert.

Acquisition of data. Maksymowych, Chiowchanwisawakit, Clare, Pedersen, Lambert.

Analysis and interpretation of data. Maksymowych, Chiowchanwisawakit, Pedersen, Østergaard, Lambert.

Manuscript preparation. Maksymowych, Pedersen, Østergaard, Lambert.

Statistical analysis. Chiowchanwisawakit, Pedersen.