Article Text
Abstract
Objective To examine associations between a panel of soluble biomarkers and progressive joint destruction assessed by magnetic resonance imaging (MRI) and conventional radiographs as well as longitudinal associations with disease activity assessed clinically and by MRI in early rheumatoid arthritis (RA) patients.
Methods 84 early RA patients were evaluated at baseline, 3, 6 and 12 months with clinical examination, serum and urine sampling, MRI scans of the dominant wrist and conventional radiographs of the hands. A panel of biomarkers (sCTX-I, uCTX-II, sOPG, sYKL-40, sCOMP and sMMP-3) was assessed by ELISA. MRI images and conventional radiographs were scored according to the RA MRI score (RAMRIS) and the van der Heijde modified Sharp score (SHS), respectively. Longitudinal associations between biomarkers and MRI inflammation and disease activity score (DAS28) and association with the progression of damage were examined with adjustments for known predictors.
Results The baseline sCTX-I level predicted progression in joint destruction assessed by MRI and conventional radiographs, whereas the uCTX-II level was a predictor of progression in SHS but not RAMRIS. Consistent associations, both with MRI inflammation (synovitis and bone marrow oedema) and DAS28 were found for sYKL-40 and sMMP-3 in addition to C-reactive protein at baseline and in longitudinal analyses. Associations remained significant in multivariate analyses.
Conclusion Levels of sCTX-I and uCTX-II were significant predictors of progressive joint destruction, whereas sMMP-3 and sYKL-40 were merely markers of joint inflammation. The clinical value of these markers for use in individual patients is limited due to a considerable overlap in levels of patients with progression and no progression.
Statistics from Altmetric.com
The disease impact of rheumatoid arthritis (RA) is determined by inflammation and subsequent joint destruction. The disease course shows a pronounced variation between patients, and it is crucial to identify the patients with rapid joint destruction who will benefit most from aggressive treatment regimens. Biomarkers measured in serum or urine are easily obtainable and have the potential to give a quantitative assessment of the disease process. They are thus potential tools to monitor different aspects of the disease course in RA. Several biomarkers have been proposed as predictors of progression in joint destruction,1,–,7 but the associations between soluble biomarkers and joint inflammation have been less well studied. Markers of synovial inflammation could serve as more joint-specific disease activity parameters than acute phase reactants.
Assays targeting important molecules in the disease process have been developed during the past few years. Destruction of cartilage leads to the degradation of collagen type II and collagen fragments, that is, crosslinked C-terminal telopeptides (CTX-II) and other matrix proteins such as cartilage oligomeric protein (COMP) can be measured in serum or urine.8 Correspondingly, collagen type I degradation releases type I collagen C-terminal telopeptides (CTX-I). The serum levels of osteoprotegerin (OPG), the decoy receptor of the receptor activator of nuclear factor κB ligand (RANKL), a cytokine responsible for osteoclast maturation and activity, can possibly also reflect the extent of ongoing joint destruction.9 The matrix metalloproteinase 3 (MMP-3) and a protein of unknown function called YKL-40 or human glycoprotein 39, are suggested to be biomarkers of cartilage degradation.4, 5, 8, 10
Magnetic resonance imaging (MRI) is a more sensitive tool for the assessment of erosions than conventional radiographs, especially in the early disease course,11, 12, 13 and an additional major strength of MRI is the demonstration of joint inflammation. In addition to assessment of synovitis, MRI depicts bone marrow oedema, lesions thought to represent inflammatory processes within the bone.14,–,16
The clinical role of RA biomarkers has not been clarified.17, 18 An ongoing OMERACT (outcome measures in rheumatology) initiative to validate soluble biomarkers in rheumatic diseases19 has initially concluded that the present evidence for a clinical role of soluble biomarkers is limited,20 and that well-designed longitudinal studies with robust endpoints are needed to investigate further whether biomarkers assessed in the serum or urine reliably reflect the disease processes in RA.21 The aim of this prospective study of early RA patients was to examine whether one or more of a panel of soluble biomarkers could predict progressive joint destruction assessed by MRI and conventional radiographs and to examine longitudinal associations with joint inflammation assessed by MRI and clinical disease activity score.
Methods
Patients and study design
An inception cohort of 84 consecutively enrolled RA22 patients with disease duration of less than 1 year was followed longitudinally for 12 months. The study population and study design has previously been described in detail.23 In brief, median (interquartile range; IQR) age was 58 years (47–67), 77.4% of the patients were female and the median baseline disease duration was 107 days (77–188). Clinical examination, including an assessment of clinical disease activity by the disease activity score in 28 joints (DAS28), imaging procedures and laboratory analyses were performed at baseline and after 3, 6 and 12 months. The patients received treatment according to clinical practice. At baseline, disease-modifying antirheumatic drugs (DMARD) were used by 77.4% of the patients included (57.1% methotrexate monotherapy, 8.3% sulphasalazine monotherapy, 7.1% hydroxychloroquine monotherapy, 3.6% DMARD combination therapy). Antitumour necrosis factor alpha treatment was used by one patient (1.2%) and 60.7% of patients received oral corticosteroids. At 1-year follow-up DMARD were used by 91.8% of patients, antitumour necrosis factor alpha drugs by two patients (2.6%) and oral corticosteroids by 49.3%. The regional ethics committee evaluated the study and all patients included gave written informed consent.
Acquisition and assessment of MRI and conventional radiographs
MRI of the dominant wrist was performed using a GE Signa 1.5 Tesla MRI scanner (General Electric, Milwaukee, Wisconsin, USA). The MRI sequences and scoring in this study have previously been described in detail.23 Images were assessed by the rheumatoid arthritis magnetic resonance imaging score (RAMRIS),24 in which separate scores are given for bone erosions (range 0–150), bone marrow oedema (0–45) and synovitis (0–9). The MRI images were read by one reader (EAH) who has a documented high inter-reader agreement with other experienced readers.25 The conventional radiographs of both hands were scored according to the van der Heijde modified Sharp score (SHS; range 0–280) by a trained observer (PB).26 The MRI and the conventional radiograph images were read in chronological order.
Biomarker analyses
Serum and urine samples were collected at baseline, 3, 6 and 12 months and stored at −70°C. IgM rheumatoid factor (RF), antibodies to cyclic citrullinated peptide (anti-CCP), high sensitivity C-reactive protein (CRP) and erythrocyte sedimentation rate were analysed as previously described.23 Seven biomarkers were measured by commercially available ELISA kits according to recommendations from the manufacturers: serum(s) MMP-3 (Bender MedSystems GmbH, Vienna, Austria); sYKL-40 (METRA YKL-40; Quidel, San Diego, California, USA); sCOMP (AnaMar Medical AB, Lund, Sweden); sCTX-I (Crosslaps; Nordic Bioscience, Herlev, Denmark); sOPG (Biomedica Medizinproducte, Vienna, Austria); urinary(u) CTX-II (Cartilaps; Nordic Bioscience); corrected for urinary creatinine concentration measured by a standard colorimetric method. All kits have inter and intra-assay coefficients of variation less than 15%.
Statistical analyses
All statistical analyses were undertaken using the statistical package for the social sciences for Windows, version 16. Group data are presented by median and IQR. The biomarker levels and the RAMRIS bone marrow oedema and erosion scores were logarithmically transformed at all time points due to a skewed distribution of the data. Receiver operating curve analyses were performed to define cutoffs for biomarkers reflecting structural damage.
Explorative analyses of the associations between the biomarker levels and inflammation and disease activity (RAMRIS synovitis, RAMRIS bone marrow oedema and DAS28) were performed on the baseline data using univariate and multivariate linear regression analyses. Separate univariate analyses were performed with each of the biomarkers as explanatory variables for all the three outcomes. Multivariate analyses with adjustments for factors known as possible confounders (age, gender, treatment (DMARD yes/no, oral corticosteroids yes/no)) and CRP were conducted if the p values in the univariate models were below 0.15.
We also assessed the longitudinal associations between the biomarkers and the inflammation measures using all the data from the four assessments, both for the explanatory and the dependent variables. To control for within-patient correlation, generalised mixed linear modelling was applied. This modelling adjusts for missing data. The first-order autoregressive correlation structure was selected. The same adjustments and selection of markers to the multivariate analyses were performed as described for the linear regression.
Associations between the biomarker levels at baseline and progression in the RAMRIS erosion score were examined using linear regression analyses. We also performed multivariate regression analyses with adjustments for age, gender, treatment and other known predictors of progression (anti-CCP status, RF status, DAS28 and RAMRIS bone marrow oedema). Backward elimination of the least significant variable was performed, but age, gender and treatment were re-entered to the model as possible confounders.
Associations between biomarker levels and radiographic progression (SHS) were assessed by logistic regression analyses, as assumptions for linear regression were not fulfilled. The same model building strategy as above was performed. As in previous publications from this cohort,23, 27 a change of 1 SHS unit or greater (hands only) was defined as progression. This cutoff also corresponds to the median value of the change in SHS. Standard diagnostic tests of model assumptions and residuals were routinely performed on all the models. All tests were two-sided and conducted at the 0.05 significance level.
Results
Baseline characteristics of the cohort
The biomarker levels and the baseline values of the outcome measures are summarised in table 1. The patients had RA of short disease duration. Fifty-five per cent of the patients were anticitrullinated protein antibody positive and 44% were IgM RF positive. Ninety-two per cent of patients completed 1-year follow-up. Further description of this cohort has previously been reported in detail.23
Associations between the biomarkers and joint inflammation at baseline
In cross-sectional analyses at baseline, sYKL-40, sMMP-3 and sCOMP, in addition to CRP, were significantly associated with the RAMRIS synovitis score. The standardised β and the R2 that can be viewed as measures of the strength of the association were slightly higher for the novel biomarkers than for CRP (table 2, RAMRIS synovitis). These significant associations remained when adjustments for possible confounders (age, gender, DMARD and oral corticosteroid use) were made (table 2, multivariate analyses of RAMRIS synovitis). When also adjusting for CRP levels (data not shown), the associations remained significant for sCOMP (p=0.01) and sYKL-40 (p=0.05), but were borderline not significant for sMMP-3 (p=0.08).
sYKL-40 and sMMP-3 in addition to CRP were also significantly associated with RAMRIS bone marrow oedema (table 2, RAMRIS bone marrow oedema). All associations remained significant in multivariate analyses after adjusting for age, gender and treatment (DMARD and oral corticosteroids) (table 2, multivariate analyses of RAMRIS bone marrow oedema). When adjusting for CRP levels sMMP-3 remained significant (p=0.03).
As a validation of our findings of a consistent association between sYKL-40 and sMMP-3 and MRI inflammation, we also examined the associations with clinical disease activity assessed by the DAS28. As shown in table 2 (DAS28), sMMP-3 and sYKL-40 were also significantly associated with DAS28. However, when adjusting for CRP, only the association for sYKL-40 remained significant (p=0.02). The strength of the association with DAS28 was as strong for sYKL-40 as for CRP according to standardised β and R2. However, there was a considerable overlap of the biomarker levels in patients with high and low disease activity (supplementary figure S1, available online only).
Longitudinal analyses
We further aimed to establish whether the associations between serum levels of YKL-40 and MMP-3 and inflammation and disease activity shown in the baseline data remained during the disease course using linear mixed model analyses. As shown in table 3, sMMP-3 and sYKL-40 in addition to CRP were longitudinally associated with both synovitis and bone marrow oedema and also with the DAS28. The p values remained significant also when adjusting for age, gender and treatment (table 3, multivariate model*). The associations were independent of CRP (table 3, multivariate model†) with the exception of the association between sYKL-40 and bone marrow oedema. The significant association between sCOMP and RAMRIS synovitis at baseline was not confirmed in the longitudinal analyses. No significant longitudinal associations were found for the other measured biomarkers with any of the three outcomes (data not shown). This consistency between cross-sectional and longitudinal analyses supports the theory that sYKL-40 and sMMP-3 are associated with inflammatory activity and that the effect is independent of and thus additional to CRP.
Associations between biomarkers and joint destruction assessed by MRI
A median (IQR) progression in the MRI erosion score of 1 (0–2.5) was seen during the 12-month follow-up. As shown in table 4, sCTX-I at baseline was significantly associated with subsequent progression in the RAMRIS erosion over 12 months. sCTX-I remained a significant predictor in multivariate analyses after adjustment for age, gender, treatment and bone marrow oedema (table 4). Anti-CCP status, RF status and DAS28 were removed from the final model as they did not reach statistical significance. These analyses showed that in addition to the bone marrow oedema score at baseline, only CRP and sCTX-I were significant predictors of progression in erosions measured by MRI.
Associations between biomarkers and joint destruction assessed by conventional radiograph
sCTX-I and sCTX-II levels at baseline were significantly associated with radiographic progression in univariate analyses and also remained significant predictors in multivariate analyses after adjustment for age, gender and treatment (DMARD and oral corticosteroids) (table 5). DAS28, RAMRIS bone marrow oedema, anti-CCP status and RF status were removed from the final model as they did not reach statistical significance. However, the levels of both markers overlap considerably between patients with progression and no progression (supplementary figure S1, available online only). In receiver operating curve analyses appropriate cutoffs for sCTX-I (0.30 ng/ml) and u-CTX-II (180 ng/mmol) were identified. The sensitivity and specificity to identify progression was 82% and 56% for u-CTX-II and 63% and 68% for sCTX-I.
Discussion
In the present study we have used longitudinal MRI assessments to examine to what extent the serum or urine levels of proposed biomarkers reflect the disease processes in RA. We showed that sCTX-I and uCTX-II levels at baseline predicted structural damage progression, and that serum levels of MMP-3 and YKL-40 were longitudinally associated with joint inflammation. These findings were consistent across several endpoints including MRI, conventional radiograph and clinical disease activity, and associations remained after adjustment for possible confounders.
Previous studies have shown an association between s/uCTX-I and uCTX-II and progression of joint destruction measured by conventional radiographs.1,–,3, 5, 28, 29 The role of sCTX-I as a marker of erosive disease is further supported by the current study because the association between sCTX-I and bone damage was demonstrated both for progression in the MRI erosion score and SHS. The cartilage marker uCTX-II predicted only progression measured by conventional radiographs, probably because the RAMRIS does not include cartilage assessment. The sCTX-I and uCTX-II levels considerably overlapped between patients with progression and no progression. Both markers showed relatively low sensitivity and specificity for the prediction of progression. Although sCTX-I and uCTX-II were significant predictors at a group level, the present study does not support that a single measure of those two biomarkers can be used to predict progression in individual patients. The combination of uCTX-II and sCTX-I did not improve the predictive capacity as the levels of the two markers were highly correlated.
Levels of sMMP-3 and sYKL-40 have previously been suggested to reflect cartilage destruction,4, 5, 8, 10 but the present study indicated that they were merely markers of synovial inflammation. This can be biologically explained by the expression of the two markers by macrophages, synovial fibroblasts and leucocytes.30, 31 This is the first study to examine the associations between soluble biomarkers and MRI-measured joint inflammation, but sYKL-40 and sMMP3 have previously been shown to be associated with clinical measures of disease activity.32,–,35 The clinical value of sMMP-3 and sYKL-40 measurements in individual patients is hampered by a substantial overlap between patients with high and low disease activity and cannot be properly addressed with the present study design. A clinically useful marker of disease activity must be sensitive to change with effective treatment, and must show an increase during a flare in disease activity—issues that need to be explored further.
We did not establish an association with inflammation or joint destruction for sCOMP or sOPG. A previous study7 has suggested the sRANKL/sOPG ratio as a more appropriate biomarker than the single measures, but current commercially available assays do not reliably assess sRANKL levels.
The strength of the current study is consistent findings across several endpoints reflecting the same disease process. This is the first study to use MRI as a standard comparison for the validation of biomarkers. MRI is a validated tool to assess synovitis, bone marrow oedema and erosions.24 The long acquisition times and the timing of the contrast restrict the number of joints that can be assessed by MRI, and the validity of assessment of a limited number of joints might be questioned. Assessment of the wrist and/or the metacarpophalangeal joints by MRI has been shown to have superior responsiveness compared with bilateral conventional radiograph assessment of the hands and feet.36 In addition, the validity of our results regarding inflammation assessed by MRI was supported by the finding of the same associations when the DAS28 was used as an outcome.
The patients were included regularly from clinical practice and a substantial proportion of the patients were already on treatment at inclusion, which may have an impact on biomarker levels. We have adjusted for treatment regimen in the analyses, but cannot exclude the possibility that the treatment at baseline might have biased the relationship between the biomarker levels and disease activity and progression. Nevertheless, a prognostic marker should also be useful if there is ongoing joint destruction despite treatment. Confounding by indication is an important bias in prognostic studies. Patients with active, severe disease will be treated more intensively by their physicians, and the effect of baseline prognostic markers might be disguised by differences in treatment over time.
This study included a relatively low number of patients and the progression rate was quite low. Still, we were able to detect significant and consistent associations both to inflammation and joint destruction for several of the measured biomarkers. We did not have the power to detect weak associations, but biomarkers with weaker associations are not likely to be clinically relevant. The relatively low power might also explain why anti-CCP, consistently reported as the strongest predictor of radiographic progression,6, 37, 38 is not a significant predictor in this study.
Several factors might affect the biomarker levels. We have adjusted for age, gender and treatment, but there might also be other unknown confounders. Some biomarkers such as CTX-I show considerable diurnal variation.39, 40 In this study serum samples were taken at the same time of day at each visit. Fasting samples would have been an advantage, but this was not feasible within the clinical setting of this study. The effect of storage on biomarker levels is unknown.20 The biomarker levels are, however, comparable with what we have previously reported from another cohort of RA patients.28
The performance of soluble biomarkers might be improved by increased knowledge of how to standardise sample collection and by further improvement of the biomarker assays, but to date no ‘novel’ biomarker has been proved to perform considerably better than autoantibodies and acute phase reactants in the prediction of progressive joint destruction. This study shows that, in spite of baseline treatment, levels of sCTX-I and uCTX-II are associated with progressive joint destruction, whereas sMMP-3 and sYKL-40 merely reflect joint inflammation, but our findings do not support that a single assessment of sCTX-I or uCTX-II can be used to predict progression in individual patients. In a complex immune disease such as RA, a combination of several markers will probably be necessary to predict the disease course, and sCTX-I and uCTX-II might have future clinical value in combination with other biomarkers as well as clinical and demographic predictive factors.
Acknowledgments
The authors would like to thank Inge C Olsen for help with the statistical analyses, research nurse Margareth Sveinsson for collecting clinical data, and research coordinator Tone Omreng for organising the data collection.
References
Footnotes
-
Funding This study has been financed with grants from the Eastern Norway Regional Health Authority, grants from the Research Council of Norway, the Norwegian Rheumatism Association, the Norwegian Women Public Health Association and Grethe Harbitz Legacy.
-
Competing interests None. Hans Bijlsma was the Handling Editor.
-
Patient consent Obtained.
-
Ethics approval This study was conducted with the approval of the Regional Ethics Committee, Eastern Norway.
-
Provenance and peer review Not commissioned; externally peer reviewed.