The study sample comprised 201 patients from the PEARL (Princesa Early Arthritis Register Longitudinal) study. To be eligible, patients had to have had one or more swollen joints for at least four weeks and symptoms for less than a year. Patients with other definite causes of arthritis were excluded. The register protocol included four visits during a two-year follow-up period. At each visit, the data collected and entered into an electronic database were as follows: clinical and demographic data; disease duration at the beginning of follow-up; DAS28 [7]; global disease activity on a 100-mm visual analogue scale scored both by the patient and by the physician; score for the Health Assessment Questionnaire (HAQ) (Spanish version) [8]; and laboratory data, including erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), and rheumatoid factor (RF) determined using nephelometry (positive>20 IU/ml) and anti-citrullinated peptide antibody (ACPA) determined using enzyme immunoassay (Euro-Diagnostica Immunoscan RA; positive >50 IU/ml). In addition, at each visit samples are collected for obtaining DNA, serum and mRNA.

The research carried out in this study is in compliance with Helsinki Declaration. The Ethics Committee for Clinical Research at Hospital Universitario de La Princesa reviewed and approved the protocol, and all patients signed an informed consent form.

Genomic DNA was isolated from blood samples using the MagNA Pure LC DNA Isolation system (Roche Molecular Biochemicals, Penzberg, Germany). The rs7574865 polymorphism of STAT4 was genotyped using the TaqMan SNP genotyping assay (Applied Biosystems, Foster City, California, USA; Part number: C__29882391_10). The polymerase chain reaction (PCR) assay was carried out according to the manufacturer’s recommendations. After PCR, the genotype of each sample was attributed automatically by measuring allele-specific fluorescence on a StepOnePlus Real-Time PCR System (Applied Biosystems).

Total RNA was extracted from peripheral blood mononuclear cells (PBMCs) with the Ultraspect RNA isolation reagent (Biotecx, Houston, Texas). cDNA was obtained with the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems). The expression levels of STAT4 mRNA were determined by quantitative real-time PCR with specific primers in samples from patients (182 visits, 2.64 visits/patient) and from 32 healthy controls (55% female, 58 years median age). Specific mRNA expression was analyzed using SYBR Green PCR Master Mix (Applied Biosystems) in a StepOnePlus Real-Time PCR System (Applied Biosystems). Expression of STAT4 and GAPDH was measured in parallel for each sample, both in patients and in healthy controls. Data from STAT4 mRNA expression were then adjusted for relative gene expression using the 2^-ΔΔCt method [9]. The primer sequences were as follows: STAT4 (forward) 5′-TGC AAC CAA AGG AAA TGA AGT-3′, (reverse) 5′-AAT GAA GTT CTT CAG TCA CCA TGT-3′; GAPDH (forward) 5′-GTG AAG GTC GGA GTC AAC G -3′, (reverse) 5′-TGA GGT CAA TGA AGG GGT C -3′.

PBMCs from 62 samples belonging to 36 patients of PEARL study with TT or GG genotype for rs7574865 were lysed at 4°C (30 minutes) in Tris-buffered saline (50 mM Tris-Cl, pH 7.5, 150 mM NaCl) 1% NP40 with a protease inhibitor cocktail (Roche Diagnostics). Whole lysates were analyzed by sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE), transferred to nitrocellulose membranes, and probed with the anti-STAT4 antibody H.56.9 (Thermo Fisher Scientific) and the anti-beta-Actin antibody AC-15 (Sigma-Aldrich) in Tris-buffered saline–Tween 20. Bound antibodies were conjugated with horseradish peroxidase secondary antibodies, and membranes were developed by enhanced chemiluminescence with Super-Signal West Femto chemiluminescent substrate (Pierce Chemical). Densitometric analyses were performed with ImageGauge 3.46 software (Fujifilm).

Serum samples are obtained at each visit of PEARL study. Samples are immediately centrifugated and the cell, free supernatant, frozen at –80°C. IL-6 serum levels is routinely measured using the Human IL-6 Quantikine high sensitivity enzyme-immune assay from R&D Systems Europe Ltd. (Abingdon, UK) in samples from those patients with, at least, three visits along the two years follow-up (769 visits; 3.8 visits per patient).

PBMCs from healthy controls or patients of PEARL study with TT or GG genotype for rs7574865 were stimulated with 10 ng/ml IL-6, 50 ng/ml IL-12 or 100 U/ml IL-2 in RPMI-1640 medium (Gibco BRL) supplemented with 10% fetal bovine serum (Gibco BRL) for 15 minutes at 37°C or unstimulated. The cells were fixed and permeabilized with 1x BD PhosFlow Lyse/Fix Buffer for 15 minutes at 37°C and then with BD PhosFlow Perm Buffer III on ice for 30 minutes. Later, we stained with phycoerythrin (PE) anti-phosphoSTAT4 (pY693), Alexa Fluor 647 anti-phosphoSTAT5 (pY694) or isotype controls (all of BD Biosciences). T cells subpopulation was delineated by staining for CD3 by employing fluorescein isothiocyanate (FITC)-conjugated anti-CD3 monoclonal antibody (BioLegend, San Diego, CA, USA).

Normally distributed variables were described as mean ± standard deviation. The Student’s t test was used to determine whether differences were statistically significant. Non-normally distributed variables were described as median (interquartile range [IQR]), and the Mann-Whitney or Kruskal-Wallis tests were used to compare their distribution between two or more groups, respectively. The correlation between continuous variables was determined using the Pearson test. Qualitative variables were described using an estimation of proportions and compared using the χ² or Fisher exact test.

In order to determine which factors influenced the level of STAT4 mRNA during follow-up, we used the information from the 67 patients whose STAT4 expression levels and full clinical data were available from at least 2 visits. Thus, data from 165 visits (average visits per patient, 2.46) were used to fit a population-averaged model by generalized linear models nested by patient and visit using the xtgee command of Stata 12.1 for Windows (StataCorp LP, College Station, Texas, USA). This model allows a better adjustment for independent variables, since by having information from several visits along the follow-up period, it can provide a better estimation of how all the independent variables are associated with the dependent variable. In addition, it is able to consider the number of visits for patient in order to avoid a higher influence of those patients with more visits. Expression of STAT4 was normalized by log transformation using the option link(log) of the xtgee command (See S1 Fig for information about this transformation). The population-averaged generalized estimating equations were first modeled by including all the variables with a p value <0.15 in the bivariate analysis. The final models were constructed using quasi-likelihood estimation based on the independence model information criterion and Wald tests[10], after removing all variables with p>0.15.

To assess whether STAT4 protein expression or IL-6 serum level was modulated by rs7574865 genotype we fitted a multivariable analysis using xtgee command of Stata as described above. The variables STAT4 protein expression and IL-6 serum levels were normalized by square root transformation (See S1 Fig for information about this transformation).

At baseline, 143 of the 201 patients (71%) fulfilled the 2010 EULAR/ACR RA classification criteria [11]. The remaining patients were categorized as undifferentiated arthritis (UA). No differences were observed by diagnosis in age at disease onset or gender (median age 54 years, 80% female; Table 1). Patients fulfilling the criteria for RA showed significantly higher baseline disease activity, disability, and CRP and ESR levels. In addition, positive ACPA and RF results were significantly more frequent than in UA patients (Table 1). Median disease duration at study entry was 5.4 months, although it was slightly lower in UA patients (Table 1). No differences in STAT4 allele distribution were observed between RA and UA patients.

The subpopulation used for the study of STAT4 mRNA expression was representative of the whole population (S1 Table).

The expression of STAT4 mRNA in EA patients decreased during follow-up, probably because of an improvement in their disease activity (Fig 1A). This finding was observed both in patients who were homozygous for the common allele (G) of rs7574865 and in those who carried the risk allele (T) of this SNP (Fig 1B). Although a trend toward higher expression of STAT4 mRNA was observed in those patients with at least one T allele, the differences did not reach statistical significance when compared at each visit.

Therefore, since the expression of STAT4 could be modified by several confounders, we further analyzed the effect of rs7574865 on the expression of STAT4 mRNA using multivariate analysis (Table 2). Disease activity affected expression of STAT4, with significantly higher levels of mRNA in visits in which patients were at moderate or high levels of disease activity than in visits in which patients were in remission (Fig 2 and Table 2). Furthermore, the use of corticosteroids was significantly associated with lower levels of STAT4 mRNA. In addition, our data revealed that patients older than 65 years expressed significantly higher levels of STAT4 mRNA than those aged less than 45 years. After adjustment for confounders, the data revealed an independent and significant association between the presence of the minor allele of rs7574865 and higher levels of expression of STAT4 mRNA (Table 2 and Fig 2).

The protocol of PEARL study does not include storing samples for intracellular proteins analysis. Therefore, we requested permission for using samples obtained to perform routine laboratory controls in patients with rs7574865 TT and GG genotypes in order to measure STAT4 protein expression in PBMCs through western-blot (See S2 Table for information about those patients). Unfortunately, we could not collect information about disease activity at this moment. Those patients in which samples were obtained previous to treatment prescription showed higher STAT4 levels than those treated either with mono or combined therapy of DMARDs (Table 3 and Fig 3C), likely reflecting that disease activity was higher in untreated patients. The use of prednisone was associated with lower levels of STAT4 protein as it happened with mRNA expression (Tables 2 and 3). After adjusting for age, gender, treatment and prednisone dose, patients carrying the TT genotype displayed significantly higher levels of STAT4 expression compared to GG patients (p = 0.009; Table 3 and Fig 3).

Although the signaling of IL-6 is described to be mediated by JAK1 and 2 through phosphorylation of STAT1, 3 and 5, considering that JAK/STAT pathway is quite redundant and in PEARL study we routinely measure IL-6 serum levels, we decided to check whether IL-6 induces STAT4 phosphorylation in PBLs. As it is shown in Fig 3, IL-12 stimulation induced mainly a mild STAT4 phosphorylation, probably because a low number of peripheral blood lymphocytes correspond to active Th1 lymphocytes. By contrast, IL-2 caused high STAT5 phosphorylation and IL-6 was able to induce both STAT4 and 5 phosphorylation (Fig 4). We found no differences between patients with GG or TT genotype for rs7574865 in IL-6 mediated-STAT4 phosphorylation (Fig 4B).

As it has been described that SLE patients carrying the T allele of rs7574865 show increased IFN-ɣ signaling despite lower levels of this cytokine, we comparatively analyzed the levels of IL-6 and CRP depending on rs7574865 genotype. We performed this analysis in view of the data described above suggesting that IL-6 may induce STAT4 phosphorylation. As it is shown in Fig 5A, IL-6 levels increased as it did the disease activity level (p<0.001 for either GG or GT/TT populations; Cuzic’s nonparametric test for trend across ordered groups). When patients were not in remission, those carrying the T allele of this SNP tended to show lower levels of this cytokine. When we analyzed raw data, differences were only statistically significant at low and moderate disease activity levels (Fig 5A). However, adjusting for confounders (disease activity, age, diagnosis, …), we found that carrying the minor allele of rs7574865 was clearly and significantly associated with lower levels of IL6 (Table 4).

By contrast, there were no differences by rs7574865 genotype in the level of CRP clustered by level of disease activity (Fig 5B). Interestingly, the pooled fitted linear prediction clustered by rs7574865 genotype showed a clearly higher production of CRP with lower levels of IL-6 in those patients with, at least, one T allele compared with those homozygous for the major allele G (Fig 5C).