Rheumatoid arthritis (RA) is characterized by inflammation and joint destruction, with the degree of damage varying greatly among patients. Prediction of disease severity using known clinical and serologic risk factors is inaccurate. This study was undertaken to identify new serologic markers for RA severity using an in silico model of the rheumatic joint.
An in silico model of a prototypical rheumatic joint was used to predict candidate markers associated with erosiveness. The following 4 markers were chosen for validation: tartrate-resistant acid phosphatase 5b (TRAP-5b), N-telopeptide of type I collagen (NTX), angiopoietin 2 (Ang-2), and CXCL13. Serum from 74 RA patients was used to study whether radiologic joint destruction (total erosion score and total Sharp/van der Heijde score [SHS]) after 4 years of disease was associated with serum levels at the time of diagnosis. Serum marker levels were determined using enzyme-linked immunosorbent assays. For confirmation, baseline serum levels were analyzed for an association with progression of joint damage over 7 years of followup in a cohort of 155 patients with early RA.
Comparison of high and low quartiles of erosion score and SHS at 4 years showed a difference in baseline serum CXCL13 level (P = 0.011 and P = 0.018, respectively). In the confirmation cohort, elevated baseline CXCL13 levels were associated with increased rates of joint destruction during 7 years of followup (P < 0.001 unadjusted and P ≤ 0.004 with adjustment for C-reactive protein level). Analyzing anti–CCP- 2–positive and anti–CCP-2–negative RA separately yielded a significant result only in the anti–CCP- 2–negative group (P ≤ 0.001).
Our findings indicate that CXCL13 is a novel serologic marker predictive of RA severity. This marker was identified with the help of an in silico model of the RA joint.
The prospects for patients with rheumatoid arthritis (RA) have improved with the development of treatment strategies for tight disease control and the availability of potent new biologic agents. In order to balance the risks of overtreatment, inducing unnecessary costs and toxicity, and undertreatment, leading to joint destruction that could have been prevented, it is of utmost importance to be able to predict the disease outcome for each patient. To date, several markers linked to a severe course of RA have been studied extensively, such as multiple swollen joints at disease onset, relatively high baseline C-reactive protein (CRP) levels, presence of anti–cyclic citrullinated peptide (anti-CCP) antibodies, rheumatoid factor (RF), and erosions at baseline (1–3). Several attempts have been made to derive adequate prediction models or prediction matrices using these risk markers, but in all studies <50% of the patients could actually be classified (4–6). This indicates that the currently used risk factors do not allow adequate prediction for individual patients. This, taken together with the fact that the severity of the disease course is highly variable among patients, highlights the importance of identifying new markers for disease outcome in RA.
The severity of RA can be measured objectively by assessing levels of joint destruction seen on radiographs of hands and feet using validated scoring methods. It reflects the cumulative burden of inflammation over time, and strongly correlates with joint functionality and subsequent disability. Several attempts to predict radiologic damage using serologic markers have been made in the last few years. Examples are osteoprotegerin (OPG), RANKL, and matrix metalloproteinase 3 (MMP-3), which are all primarily markers that reflect joint tissue remodeling. The baseline OPG:RANKL ratio was observed to be an independent predictor of the level of joint destruction later in the disease course (7). Such an association is reported for baseline MMP-3 levels as well, although its association was dependent on a correlation with traditional risk factors such as anti-CCP antibodies (8). Importantly, a recent study indicated that these factors insufficiently account for radiologic joint damage (9). This demonstrates the need for identifying new serologic markers.
The present study aimed to identify new serologic markers that are predictive of the severity of the disease course of RA. A computer model representing the biology of the rheumatic joint was used to select candidate markers for bone erosiveness. These markers were tested in baseline serum of RA patients with 4 years of radiologic data in order to determine whether an association existed in vivo. A second cohort of RA patients for whom long-term followup data were available was used in order to confirm the predictive ability of these markers for RA disease outcome, measured by the rate of joint destruction as well as the chance of sustained disease-modifying antirheumatic drug (DMARD)–free remission.
PATIENTS AND METHODS
Identification of markers.
An in silico model of a prototypical articular joint in a patient with RA was created previously (RA PhysioLab platform; Entelos) (10, 11). This RA PhysioLab platform represents the biology of RA with regard to synovial tissue inflammation, cartilage destruction, and bone erosion. The model integrates relevant in vitro and clinical data into a computer-based platform to reproduce the disease characteristics of RA. When run, the model simulates disease or biologic response to treatment in a prototypical joint representative of the affected joints of RA patients. The platform models the life cycle of inflammatory cells, endothelium, synovial fibroblasts, chondrocytes, and bone cells, as well as their products and interactions. During simulation experiments with the computer model, the interplay between these cells and processes results in a reproduction of RA disease characteristics: self-perpetuating inflammation and breakdown of cartilage and bone. These characteristics are represented by numerical readouts in the model that closely resemble evaluations accepted in the clinic, such as American College of Rheumatology response criteria, Disease Activity Score in 28 joints (DAS28), and bone erosion progression rate. For example, bone erosion is computed as the net loss of bone volume due to bone synthesis and resorption, which in turn depend on the density and activation state of osteoblasts and osteoclasts.
Computed treatment responses are consistent with clinically observed responses (11). RA is a multifactorial disease with a heterogeneous manifestation. To capture this heterogeneity, the computer model uses the virtual patient concept. Different settings for selected (combinations of) model parameters are used in parallel for the simulations. This results in a range of disease activities and therapy outcomes. Every combination of such predefined parameter sets represents a virtual patient. Examples of these settings are an increased rate of production of tumor necrosis factor α (TNFα) by TNFα-producing cells or a decrease of the effect of methotrexate (MTX) on MTX-sensitive pathways in the model (for a more detailed description, see ref.12). We used RA PhysioLab version 3.2 and a set of 120 distinct virtual patients to predict candidate serologic markers associated with localized bone loss, representing erosiveness. Therefore, erosiveness was the main outcome measure used in this study. Since the erosion score is part of the Sharp/van der Heijde score (SHS) (13), the method used to score radiologic joint damage, the total SHS was assessed as well.
Patients included in the BeSt (Behandelstrategieën voor Reumatoide Artritis [Treatment strategies for RA]) cohort for whom baseline serum was available were used to test the association of serologic markers with the level of joint destruction after 4 years of followup (total erosion score and total SHS) (n = 74). The BeSt cohort included patients with recent-onset RA with a disease duration of ≤2 years who were randomly allocated to one of 4 treatment groups with different DAS-guided combinations and applications of DMARDs (14). At the time this study was initiated, 4 years was the maximum available followup duration. Treatments consisted of sequential DMARD monotherapy (group 1), step-up combination therapy (group 2), initial combination therapy with tapered high-dose prednisone (group 3), and initial combination therapy with the TNF antagonist infliximab (group 4).
At baseline, blood samples were obtained for routine diagnostic laboratory screening, and serum was stored at −70°C. There were no significant differences between the baseline characteristics of patients with and those without serum data available, apart from slightly higher numbers of swollen joints in the group with missing data (data not shown). All 4-year radiographs were scored independently by 2 trained readers who were blinded with regard to patient identity, treatment group, and the sequence of the radiographs. The mean of the scores assigned by the 2 readers was used for the analysis. The interobserver correlation coefficient was 0.96 (15).
The second set of RA patients was used to confirm findings between serologic marker levels and the severity of the disease course. This set comprised 155 patients with early RA who were included in the Leiden Early Arthritis Clinic (EAC) cohort between 1993 and 2006 and for whom both baseline serum as well as yearly radiographs were available. No significant differences were observed between the baseline characteristics of patients with and those without serum available (data not shown). The Leiden EAC is a large prospective cohort that has been described previously (16). Patients were referred by general practitioners when arthritis was suspected and were included in the EAC cohort if arthritis was confirmed at physical examination and symptom duration was <2 years. At enrollment, patients were asked about their joint symptoms, and a physical examination was performed. At baseline, blood samples were obtained for routine diagnostic laboratory screening and serum was stored at −20°C (at the beginning of the study) or −70°C. The EAC patients studied were not included in the BeSt cohort.
Radiographs of the hands and feet were obtained at baseline and in consecutive years and were scored chronologically by an experienced reader (MPMvdL), as previously described (17). Intraclass correlation coefficients were 0.91 for all radiographs, 0.84 for baseline radiographs, and 0.97 for the radiographic progression rate. To encompass a reliable sample size, radiographic followup data were restricted to a maximum of 7 years. As mentioned above, the total erosion score was the main outcome measure; in addition, the total SHS was assessed. The present study had a power of 96% to detect a difference of 15 SHS points (SD 25) at the 7-year time point, with an alpha level of 0.05. Disease remission was assessed as a second outcome measure in order to further substantiate the findings. Remission was defined in its most stringent form as the persistent absence of synovitis for ≥1 year after cessation of DMARD therapy and the identification of remission by the patient's rheumatologist (18). The remission status could be reliably ascertained in 152 of 155 RA patients. Most patients in whom remission was achieved were followed up for longer than the minimally required 1 year; the median time of observation after discontinuation of DMARDs was 2.5 years.
Serum measurements of biomarker levels in both independent patient cohorts were performed by enzyme-linked immunosorbent assay according to the recommendations of the manufacturers. Measured biomarkers were crosslinked N-telopeptide of type I collagen (NTX) (Osteomark Ntx; Unipath Limited) (1:5 dilution), tartrate-resistant acid phosphatase 5b (TRAP-5b) (BoneTRAP Assay SB-TR201A; Immunodiagnostic Systems), angiopoietin 2 (Ang-2; 1:3 dilution), and CXCL13 (both from R&D Systems).
RA PhysioLab platform.
RA PhysioLab version 3.2 was used for the simulations. The patient cohort consisted of 120 virtual patients with different underlying pathophysiologies. Data for all virtual patients were analyzed after simulation of 1 year of untreated disease. Erosiveness in each virtual patient was determined by the volume of bone loss during the period of simulation. The bone loss values were categorized in quartiles, and the lowest and highest quartile were compared. In total, 150 simulation variables, including the concentrations of all mediators, were investigated for association with erosiveness. For statistical analysis, Student's t-test, Wilcoxon's rank sum test, and the Kolmogorov-Smirnov test from R software, version 2.4.1, were used.
To investigate whether the in silico model accurately predicted that erosiveness during the disease course was associated with baseline serum biomarker levels, the 4-year erosion score and SHS were categorized into quartiles. The lowest and highest quartiles were compared for differences in baseline serum level using the Mann-Whitney U test. In addition, in all RA patients the correlations between erosion score and SHS on a continuous scale and the serum levels were assessed using a nonparametric Spearman's correlation test and a linear regression analysis on log-transformed radiologic data with adjustment for treatment strategy (randomization arm).
The association between baseline serum CXCL13 levels and the rate of joint destruction during 7 years of followup was assessed by repeated-measures analysis on log-transformed radiologic data, correcting for age, sex, and applied treatment strategy, as previously described (17). The repeated-measures analysis is performed using a multivariate normal regression model that, on longitudinal data, evaluates the progression rates over time and takes into account the correlation between the measurements within one subject. In order to test whether biomarker levels were associated with joint destruction independent of inflammation, adjustments for CRP level were made as well. To test for associations between baseline CXCL13 levels and baseline clinical characteristics, analyses were performed using the nonparametric Spearman's correlation test.
Analysis of sustained DMARD-free remission was performed by comparing Kaplan-Meier curves and by Cox regression analysis, correcting for age and sex, taking into account the differences in followup times among patients. For patients in whom remission was achieved, the dependent variable was “time-to-event,” indicating the time until remission was reached. For patients in whom remission was not achieved, the time to last followup was used.
SPSS, version 17.0 was used. P values less than 0.05 (2-sided) were considered significant.
RA PhysioLab prediction results.
After simulation of 1 year of untreated disease, the virtual patient cohort was categorized on a numerical readout representing erosiveness. For the identification of candidate biomarkers, we focused on model variables related to proteins, for reasons of biomarker detection feasibility. Other variables, such as those related to cell densities, were not taken into account for the analyses. We identified a set of proteins with significantly different concentrations in the lowest versus highest erosiveness quartiles, as determined by each of 3 statistical tests (Student's t-test, Wilcoxon's rank sum test, and the Kolmogorov-Smirnov test; P < 0.0001). Four of these mediators were selected for followup: Ang-2, NTX, TRAP-5b, and CXCL13. Selection of these 4 mediators was based on pragmatic considerations: assay availability and presence (NTX, TRAP-5b) or absence (Ang-2, CXCL13) of previous studies linking the protein to bone erosion (19, 20). The ability of CXCL13, Ang-2, NTX, and TRAP-5b to differentiate between virtual patients with high and low erosion is illustrated in Figure 1. In the RA PhysioLab platform, most mediators are tracked as synovial quantities. For the 4 mediators selected, transport between synovium and serum is modeled only for NTX. For Ang-2, TRAP-5b, and CXCL13, the difference in mediator concentration between the erosiveness quartiles relates to synovial tissue concentrations; for NTX, to synovial tissue and serum concentration.
Identification and replication in two independent cohorts.
Baseline characteristics of the 2 sets of RA patients are presented in Table 1.
Table 1. Baseline characteristics of the RA patients*
Discovery cohort (n = 74)
Confirmation cohort (n = 155)
There were no significant differences between the discovery cohort and the confirmation cohort. RA = rheumatoid arthritis; CRP = C-reactive protein.
Data on anti–cyclic citrullinated peptide 2 (anti–CCP-2) status were not available for 2 patients in the confirmation cohort.
To evaluate the in silico predictions with regard to associations between markers and erosiveness in vivo, baseline serum levels of 4 biomarkers predicted in silico were determined. The mean ± SD levels were 4,658.4 ± 1,596.0 pg/ml, 51.9 ± 24.5 nM bone collagen equivalents/liter, 2.19 ± 1.14 units/liter, and 166.7 ± 86.0 pg/ml for Ang-2, NTX, TRAP-5b, and CXCL13, respectively.
To analyze whether baseline serum levels of these biomarkers accurately predicted erosion scores and the total level of joint destruction during the disease course, the lowest and highest quartiles of SHS and erosion scores after 4 years were studied in relation to the serum levels of the markers (Table 2). This revealed a significant difference for CXCL13 (P = 0.011 for the total erosion score and P = 0.018 for the total SHS). Similarly, significant correlations between CXCL13 levels and the erosion score (P = 0.022, ρ = 0.267) and between CXCL13 levels and the SHS (P = 0.014, ρ = 0.286) were observed when the analysis was performed on continuous data from all RA patients. In addition, linear regression analysis with adjustment for treatment strategy showed that baseline CXCL13 levels remained significantly associated with the 4-year erosion score and SHS (for erosion score, β = 1.002, 95% confidence interval [95% CI] 1.000–1.005, P = 0.049; for SHS, β = 1.002, 95% CI 1.000–1.005, P = 0.033). For Ang-2, NTX, and TRAP-5b, no significant associations were observed (Table 2). Taken together, these data indicate that of the 4 serum markers predicted by the in silico model, 1 marker, CXCL13, was observed to be actually associated with joint destruction in patients.
Table 2. Baseline serum marker levels in patients with RA in the discovery cohort with low and high erosion after 4 years of disease duration*
Values are the mean ± SD. The Sharp/van der Heijde score (SHS) is composed of erosion and joint space narrowing scores. RA = rheumatoid arthritis; Ang-2 = angiopoietin 2; NTX = N-telopeptide of type I collagen; BCE = bone collagen equivalents; TRAP-5b = tartrate-resistant acid phosphatase 5b. Differences in levels between the first quartile (low) and the fourth quartile (high) were compared using the nonparametric Mann-Whitney U test.
4,361.2 ± 1,401.9
48.47 ± 16.41
2.45 ± 0.94
137.7 ± 80.4
5,286.8 ± 1,611.8
62.89 ± 40.66
2.61 ± 1.41
189.8 ± 66.1
4,444.0 ± 1,606.7
51.61 ± 15.64
2.41 ± 0.99
139.6 ± 80.6
5,279.2 ± 1,634.5
55.83 ± 23.23
2.08 ± 1.32
186.9 ± 64.9
For confirmation, baseline serum CXCL13 levels were measured in a second set of patients, yielding a mean ± SD concentration of 155.5 ± 98.9 pg/ml. Baseline serum levels were categorized as low or high based on quartile distribution and were studied in association with the rate of progression in joint destruction over 7 years. Repeated-measures analysis showed that higher CXCL13 levels were associated with significantly higher progression rates in erosion score (P < 0.001) as well as in SHS (P < 0.001) (analyses performed on log-transformed data) (Figure 2). For the erosion score, the increase per year in the original scale (not log-transformed) was 1.12, 1.08, and 1.18 times greater for CXCL13 quartiles 2, 3, and 4, respectively, than for quartile 1. For the total SHS, a 1.11, 1.09, and 1.18 times greater increase per year was observed compared to the lowest quartile. Over a period of 7 years, this resulted in 2.2 (95% CI 1.5–3.3), 1.8 (95% CI 1.2–2.6), and 3.1 (95% CI 2.1–4.7) times larger progression rates for the erosion score and 2.1 (95% CI 1.4–3.3), 1.9 (95% CI 1.2–2.8), and 3.2 (95% CI 2.1–4.9) times larger progression rates for the SHS. Since categorical analysis generally results in less discriminative ability, the CXCL13 level was also included in the repeated-measures analysis as a continuous variable. This analysis also showed that higher CXCL13 levels were significantly associated with higher rates of progression of the erosion score and SHS (both P < 0.001). This analysis was also used to determine the variance explained by CXCL13. This showed that 7% of the total variance in progression in erosion score was explained by CXCL13.
Associations of clinical features with CXCL13.
To study clinical factors that possibly influenced the observed association of CXCL13 with the rate of joint destruction in the confirmation cohort, baseline patient characteristics were analyzed in relation to baseline serum CXCL13 levels. Significant correlations were found between CXCL13 level and CRP level (P < 0.001, ρ = 0.429), erythrocyte sedimentation rate (P ≤ 0.001, ρ = 0.300), and number of swollen joints (P = 0.023, ρ = 0.255). In addition, mean CXCL13 levels were significantly higher in serum samples from anti–CCP-2–positive patients than anti–CCP-2–negative patients (mean ± SD 172.0 ± 104.7 pg/ml versus 134.8 ± 87.3 pg/ml; P = 0.008). Similar results were observed for IgM-RF, yielding levels of 172.9 ± 104.8 pg/ml for IgM-RF–positive patients versus 120.2 ± 77.7 for IgM-RF–negative patients (P < 0.001).
CXCL13 in relation to other serologic markers.
We investigated whether serum CXCL13 levels were associated with erosion score and SHS independently of the known serologic markers CRP level and anti–CCP-2. First, CRP level was entered as an adjustment variable in the repeated-measures analysis. With adjustment for CRP level, CXCL13 was still significantly associated with the erosion score (P = 0.001) and the SHS (P = 0.004). In addition, since anti–CCP-2–positive and anti–CCP-2–negative RA are considered to be separate subsets of the disease with differences in underlying pathogenic mechanisms, the analyses were repeated in the anti–CCP-2–positive and anti–CCP-2–negative subsets. In anti–CCP-2–negative patients, high CXCL13 levels were also significantly associated with higher erosion score and SHS progression rates (P < 0.001 and P = 0.001, respectively). After adjustment for CRP level, the association remained significant for the erosion score (P = 0.002), but not for SHS (P = 0.10). In anti–CCP-2–positive patients, no associations were observed with either the erosion score or SHS (data not shown).
Sustained DMARD-free remission.
To further substantiate the findings with regard to CXCL13, we investigated a different outcome measure for RA severity, the achievement of DMARD-free remission. In addition to an observed association between CXCL13 levels and joint damage, higher CXCL13 levels were associated with significantly lower chances of achieving remission. Compared to the first quartile of CXCL13 levels, hazard ratios of 3.3 (95% CI 1.0–11.1) (P = 0.049), 4.1 (95% CI 1.1–14.8) (P = 0.034), and 6.3 (95% CI 1.4–29.2) (P = 0.019) for not achieving remission were observed for the second, third, and fourth quartiles of CXCL13 levels, respectively (Figure 3). With an additional adjustment for CRP level, the hazard ratios were 4.1 (95% CI 0.8–21.2), 2.6 (95% CI 0.7–10.2), and 2.6 (95% CI 0.7–9.0), respectively. Because remission was achieved in only 3 patients in the anti–CCP-2–positive RA group, no stratified analysis was performed.
The heterogeneous nature of RA severity is presently incompletely understood and hampers accurate disease prognosis for individual patients. This highlights the need for new disease markers with improved prognostic potential in order to guide adequate treatment regimens. Also, exploration of new markers will enhance our understanding of the pathophysiologic processes involved. Exploration of suitable new markers was initiated by candidate marker prediction using the RA PhysioLab computer model of the rheumatic joint. The value of PhysioLab simulation approaches has been shown previously in RA (11) as well as in other diseases (21–23). The PhysioLab simulation platform allows question-based simulation experiments, the results of which contribute to hypothesis-driven, focused followup experimental research. For this approach it is assumed that the scope of the computer model resembles real disease and patient population behavior as closely as possible.
In this study we identified a new biomarker, CXCL13, with an association between its baseline serum level and the level of joint destruction after 4 years of disease, thereby validating the association predicted by the in silico model. Moreover, in an independent cohort, relatively high serum CXCL13 levels were also associated with enhanced progression of the rate of joint destruction over 7 years and a decreased chance of achieving DMARD-free remission, thereby con- firming the association between CXCL13 and disease severity.
The cytokine CXCL13 is also known as B lymphocyte chemoattractant or B cell–attracting chemokine 1 and is part of the CXC chemokine family. One of the main effects of CXCL13 implemented in the RA PhysioLab platform is on B cell recruitment, thus supporting the notion of a mechanism that is directly dependent on B cells. CXCL13 levels were found to be significantly higher in the serum of RA patients than healthy controls (24, 25). Serum CXCL13 levels were also reported to respond to therapeutic intervention with anti-TNFα therapy (12). Evidence of joint localization of CXCL13 has been demonstrated, both by the detection of messenger RNA in inflamed synovial tissue (26) and by the presence of ectopic lymphoid follicles expressing CXCL13 in the synovium of patients with chronic RA (27). CXCL13 has been reported to attract B lymphocytes and to interact with the receptor CXCR5, which is expressed by B cells as well as follicular B helper T cells (28–31). High levels of CXCR5 were also found on human osteoblasts, and activation by its ligand CXCL13 induced the release of extracellular matrix–degrading enzymes. As such, CXCL13 may play an important role in the process of bone remodeling (32). These data suggest that CXCL13 may affect joint damage in RA both independently of B cells and by interactions that promote B cells.
Since CXCL13 attracts B cells and CXCL13 levels are reported to be higher in autoantibody-positive RA (24), which was also observed in the present study, and since anti–CCP-2–positive and anti–CCP-2–negative RA are subsets of RA with possible differences in pathogenesis, the effect of CXCL13 was evaluated for anti–CCP-2–positive and anti–CCP-2–negative RA separately. In the anti–CCP-2–negative group, CXCL13 was significantly associated with the progression of joint destruction. Adjustment for baseline CRP levels did not alter this association for the erosion score, supporting the notion of an effect of CXCL13 on joint destruction that is independent of the level of inflammation indicated by the CRP level. In anti–CCP-2–positive RA, CXCL13 was not independently associated with progression in joint damage. Thus, from a clinical perspective, information on baseline CXCL13 levels seems most valuable in the anti–CCP-2–negative subpopulation of RA patients.
Analysis of the lowest and highest quartiles of erosiveness in the discovery cohort did not reveal a significant difference for Ang-2. However, when comparing the upper 2 quartiles with the lower 2 quartiles, a significant difference was found for Ang-2 (data not shown). This indicates that a second candidate marker from the 4 markers tested might be of interest for further exploration. In the RA PhysioLab platform, Ang-2 and CXCL13 both have an effect on endothelial life cycle, which affects all recruitment processes.
To the best of our knowledge, this is the first study to identify a new serum marker for RA that was initially predicted by a computer simulation of the rheumatic joint. This finding illustrates that the present computer model has predictive potential and may be applied to other disease outcomes.
Although the results of this study provide a solid foundation for our conclusions, the study also has limitations. Both the discovery and the confirmation cohorts consisted of a limited number of patients, which may result in limited power to accurately reach proper conclusions. As a result, only relatively large differences in effect sizes may be detected, and smaller effects may be missed. A strong association between CXCL13 levels and radiologic progression was observed in 2 separate cohorts. For Ang-2 the evidence is less conclusive than for CXCL13. Failure to detect an in vivo association for the other candidate biomarkers that were tested (NTX and TRAP-5b) despite their known association with bone biology could indicate that their behavior in a heterogeneous clinical population is not predictive.
Recently, draft validation criteria for a soluble biomarker to be regarded as a valid biomarker reflecting structural damage in RA have been established (9, 33, 34). These criteria provide guidance for the types of studies needed to demonstrate the value of CXCL13 as a marker in clinical practice. The findings of the present study indicate that it is a serologic marker with potent CRP-independent predictive value for long-term outcome in RA, thereby providing a rationale for further exploration. In conclusion, the RA PhysioLab simulation platform has helped in the identification of CXCL13 as a new serologic marker for severity of RA as measured by long-term joint destruction and the achievement of DMARD-free remission in 2 independent cohorts of RA patients.
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. van der Helm-van Mil 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 conception and design. Meeuwisse, van der Linden, Rullmann, Allaart, Nelissen, Huizinga, Toes, van Schaik, van der Helm-van Mil.
Acquisition of data. Meeuwisse, van der Linden, Rullmann, Allaart, van der Helm-van Mil.
Analysis and interpretation of data. Meeuwisse, van der Linden, Rullmann, Allaart, Nelissen, Huizinga, Garritsen, Toes, van Schaik, van der Helm-van Mil.
ROLE OF THE STUDY SPONSOR
Merck, Sharp, & Dohme played a role in the study design and the analyses concerning the PhysioLab model data. In addition, authors from Merck, Sharp, & Dohme (Meeuwisse, Rullmann, Nelissen, Garritsen, and van Schaik) were involved in writing the manuscript. Merck, Sharp & Dohme had no role in the collection, analyses, or interpretation of data from the BeSt and Leiden EAC cohorts, or in the decision to publish.