The register protocol was reviewed and approved by the Ethics Committee for Clinical Research at the Instituto de Investigación Sanitaria La Princesa. All patients were informed about the study and signed an informed consent form before inclusion in the EA register.

The study sample comprised 91 patients enrolled on our EA register. The inclusion criteria included more than 1 swollen joint for at least 4 weeks and symptoms for less than a year. Only data from patients fulfilling the 1987 American College of Rheumatology criteria for RA [21] within the 5-year follow-up (n = 69) or with chronic undifferentiated arthritis (n = 22) were analyzed. Patients with diagnosis of spondyloarthritis, connective tissue diseases or crystal induced arthritis, were excluded from the analysis. The register protocol included 5 visits (baseline, 6, 12, 24, and 60 months), and at each one we recorded the following data in an electronic database: clinical and demographic information; disease duration at the beginning of follow-up; 28-joint Disease Activity Score (DAS28) [22]; global disease activity on a 100-mm visual analogue scale assessed by both the patient and the physician; Health Assessment Questionnaire score (HAQ; Spanish version) [23]; and the results of laboratory tests including erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), rheumatoid factor (RF, assessed by nephelometry; positive >20 IU/ml), and anti-citrullinated peptide antibody (ACPA, measured by enzyme immunoassay [EIA]: Euro-Diagnostica Immunoscan RA; positive >50 IU/ml).

Healthy donors (n = 100) were recruited from the Centro de Transfusiones de la Comunidad de Madrid. Following the Spanish Personal Data Protection Law, their demographic information was confidential.

VIP levels were determined using a competitive EIA with a commercially available kit according to the manufacturer's instructions (Phoenix Pharmaceutical, Karlsruhe, Germany). We previously optimized sample preparation procedure to obtain optimal results. First, it was tested the extraction of peptides from serum using a SEP-COLUMN containing 200 mg of C18, according to the manufacturer’s recommendations. Second, serum samples were concentrated by using a lyophilizer to assay different concentrations. Then, VIP levels were determined. The best results were obtained in reconstituted dried extract (2∶1) without the use of columns. The assay was performed as follow, briefly, the samples were freeze-dried and dissolved in EIA buffer (2∶1), added to an immunoplate pre-coated with a secondary antibody, and incubated with biotinylated VIP and a primary antibody whose Fab fragment competitively binds to the biotinylated peptide and targeted peptide in samples. After washes, the wells were incubated with streptavidin-horseradish peroxidase, which catalyzed the oxidation of the substrate solution. The reaction was stopped using the stop solution and absorbance measured at 450 nm. A standard curve of known concentration was established. The concentration in the samples was determined by extrapolation to this curve and by applying the corresponding dilution factor. Samples from each patient were assayed twice. The minimum detectable concentration was 0.12 ng/ml of sample, with an intra-assay and interassay variation of ≤5% and 15%, respectively.

Since serum samples were stored at –80°C for different periods of time until assay (range, 1 to 112 months), we first analyzed the effect of sample frozen-time on the ability of our assay to detect VIP. As shown in Figure S1, no significant correlation was observed between VIP serum level and frozen-time.

Normally distributed quantitative variables were expressed as the mean ± standard deviation, while non-normally distributed variables were expressed as the median and interquartile range (IQR). Qualitative variables were described using an estimation of the proportions. Variables with a normal distribution were analyzed using the t test, while the Mann-Whitney or Kruskal-Wallis tests were used for variables with a non-normal distribution. A χ² test or the Fisher exact test was used to compare categorical variables.

Considering that the raw data contained considerably left-shifted values (Figure S2A), we decided to censor values higher than 1000 pg/ml in this figure (Figure S2B) to avoid missing information if patients with very high VIP values were excluded. In addition, since the censored variable does not show a Gaussian distribution, data were normalized through logarithmic transformation (Figure S2C).

Additional variables were defined to further describe the role of VIP in the progress of arthritis. VIP levels were considered to be low when the concentration was below the 25^th percentile of its concentration in the healthy population (370 pg/ml). The intensity of disease-modifying anti-rheumatic drugs (DMARDs) treatment (IDT) was assessed as the number of days of treatment with each DMARD during follow-up adjusted for weighted coefficients, as described elsewhere [24].

To identify factors that influenced VIP levels during follow-up (logarithmic transformation of the censored VIP levels as the dependent variable), we used data from 340 visits in the 88 patients with all information available in at least 2 visits to fit a population-averaged model by means of generalized linear models nested by patient and visit using the xtgee command of Stata 12 for Windows (StataCorp LP, College Station, Texas, USA). The population-averaged generalized estimating equations were first modeled by adding all variables with a p value <0.15 in the bivariate analysis. The final models were constructed using a quasi-likelihood estimation based on the independence model information criterion [25] and Wald tests, after removing all variables with p>0.15.

To determine the usefulness of baseline VIP levels for predicting disease progress, we performed an ordered logistic model using the ologit command of Stata 12. The dependent variable was the disease activity level at the 2-year follow-up, using the cut-off points for DAS28 proposed by Prevoo et al. (<2.6, remission; 2.6 to 3.2, low disease activity; 3.2 to 5.1, moderate disease activity; >5.1 high disease activity) [22]. In this model, the dependent variable considered remission as 0 and low, moderate, and high disease activity as 1, 2, and 3. Thus, the ordered logistic analysis can also estimate cut-off points that aid the interpretation of the coefficients for each independent variable according to the levels of the dependent variable. The analysis was modeled as described above for xtgee.

The significance level was set at p<0.05.

A total of 91 patients with RA (n = 69) or UA (n = 22) were included in the study. The main difference between these two subgroups was a higher frequency of female patients in the RA group, in which more severe disease at baseline and a higher prevalence of RF and ACPA positivity were also found (Table 1). No significant differences were detected between the groups in variables expressing a subjective component such as pain, global disease assessment by patient, tender joint count, and HAQ.

Considering serum levels of VIP at baseline, no significant differences were detected between the groups (Table 1) or between patients and healthy controls (Figure 1). Nevertheless, the distribution of the VIP serum concentration was considerably heterogeneous, especially among RA patients, and ranged from 100 to >1000 pg/ml. In addition, the lowest levels of VIP tended to be more frequent in RA and UA patients than in healthy donors (Figure 1).

Despite a significant decrease in disease activity during follow-up (Figure 2A), we could not detect a parallel down-regulation of serum VIP levels (Figure 2B). Therefore, we performed a multivariate analysis to achieve a more accurate appraisal of the potential relationship between VIP levels and disease activity in patients with EA. Our data showed that elderly patients expressed slightly but significantly higher serum levels of VIP during follow-up (Table 2 and Figure S3). In addition, VIP levels tended to be significantly higher in visits at which patients were treated with TNF blockers. Interestingly, after adjustment for these confounders, we observed a significant and negative correlation between disease activity, estimated through DAS28, and VIP serum levels (Table 2, model 1).

Next, we analyzed the concentration of VIP at the different visits after clustering the results in four groups: remission, low, moderate, and high disease activity. As shown in Figure 2C, we observed lower levels of VIP in those visits with moderate or high activity compared to visits at which patients were in remission and those differences reached statistically significance after adjusting by confounders (Table 2, model 2). However, this approach did not reveal whether inflammation induced low VIP levels or those patients with low VIP showed more intense disease activity.

Next, we studied whether having low VIP serum levels at baseline could predict long-term disease activity. As Figure 3A shows, those patients who continued to have high or moderate disease activity levels after a two-year follow-up had lower baseline VIP serum levels. This behaviour was detected despite the observed trend toward higher prescription of DMARDs in patients with low VIP serum levels (Figure 3B).

In view of the various confounding factors that could influence disease activity level after two years of follow-up, we performed a multivariate ordered logistic regression to determine which variables contribute to an increased level of disease activity. Female patients showed an odds ratio (OR) of 4.3 for poorer disease outcome than male patients (Table 3). In addition, patients requiring combined therapy displayed an OR of 6.6 for remaining at a higher disease activity level than those who did not require DMARDs. After adjustment for these confounders, having low baseline VIP levels was associated with a higher level of disease activity after two years of follow-up. However, marked interference with ACPA positivity was observed (Table 3). Thus, patients with low VIP serum levels but negative ACPA had an OR of 6.1 for being at a higher disease activity level than those with negative ACPA levels and normal VIP serum levels (Table 3 and Figure 3C). By contrast, patients with positive ACPA levels showed a nonsignificant trend toward lower disease activity regardless of VIP levels (Table 3 and Figure 3C). In addition, the model revealed that the effect of gender, treatment, and baseline VIP level was more intense when the level of disease activity increased (β coefficients for cutpoints in Table 3).

In terms of treatment, ACPA positivity was associated with higher prescription of DMARDs, as occurred when low VIP levels were associated with ACPA positivity (Figure 3D). More heterogeneous regimens were found in patients with low VIP levels (treating physician blinded) and negativity for ACPA, but they undoubtedly received more intensive treatment than the population that met neither of these conditions (Figure 3B and 3D).