Serum levels of 14-3-3η protein supplement C-reactive protein and rheumatoid arthritis-associated antibodies to predict clinical and radiographic outcomes in a prospective cohort of patients with recent-onset inflammatory polyarthritis

The longitudinal EUPA cohort was previously described [8‐10]. Starting in 1998, consecutive adult patients with at least three swollen joints for one to twelve months were evaluated at the Centre hospitalier universitaire de Sherbrooke (CHUS) and asked to participate. We excluded patients with bacterial or crystal-induced arthritis or a defined connective tissue disease or systemic vasculitis according to American College of Rheumatology (ACR) criteria [11]. Patients were treated by rheumatologists using early and intensive treatment with dosage individualized to achieve sustained remission defined as 0 swollen joints out of 66 joints [2, 12]. Serum samples were coded and stored at –20 °C. The Ethics Review Board of the CHUS approved the study (ClinicalTrials.gov ID: NCT00512239) and all patients provided written informed consent.

A rheumatologist completed joint counts and a trained coordinator conducted a structured interview at inclusion and at each of the follow-up visits scheduled at 18, 30, 42 and 60 months after onset. Time of onset was self-reported as the week during which symptoms/signs of inflammatory arthropathy had appeared. Variables assessed included demographics; 68 tender joint count (TJC) and 66 swollen joint count (SJC); drug use at and between each visit; modified Health Assessment Questionnaire (M-HAQ) [13]; serum CRP (upper normal limit: 8.0 mg/L); components of the Simplified Disease Activity Index (SDAI); and RA-associated antibodies (see below). Radiographs of the hands and feet were obtained at inclusion and at each scheduled assessment. Joint space narrowing and erosions were scored on these radiographs according to the Sharp score modified by van der Heijde (SvH) method, with a maximum score of 448 units [14]. Radiographs were read in known time sequence by two blinded assessors, one of which was a study investigator (GB). Under these conditions, the smallest detectable change (SDC) was 5 units [15].

IgM rheumatoid factor (RF) was measured using RapiTex RF (Dade Behring Inc, Newark DE, USA) (positive ≥40 IU/ml). Anti-CCP2 antibodies were measured using QuantaLite™, Inova Diagnostics (San Diego, CA, USA) using titer levels as suggested by the manufacturer (positive >20.0 U/ml) or, since 2009, using the EuroImmun assay (positive >5.0 U/ml). As both the EuroImmun and the Inova assays use the same enzyme linked immunosorbent assay (ELISA) plates coated with the same antigens (as do all other antibodies to cyclic citrullinated peptide, second generation (anti-CCP2) assays across the world), have similar sensitivity and specificity, and give linear results across a wide range of concentrations, their results are easily interconvertible using the following logarithmic transformation:

Inova anti-CCP2 = –26.44 + 28.86 * ln(EuroImmun anti-CCP2).

The anti-Sa/citrullinated vimentin in-house ELISA was described previously (positive threshold ≥0.20 optical density units) [16]. From 2012 on, we used a commercial anti-Sa assay (EuroImmun; positive >19 RU/ml), validated against our in-house ELISA [17].

Radiographic progression was defined by the difference between the SvH scores over time and baseline damage. Definite SvH progression was defined as an increase ≥5 in either the total score (ΔSvH) or its erosion component (ΔErosion). Remission was defined as SDAI ≤3.3 [18].

Serum 14-3-3η levels were measured using the 14-3-3η ELISA according to the manufacturer’s protocol (Augurex Life Sciences Corp, Vancouver, BC, Canada). Samples with levels below the reportable range were assigned a concentration of 0.0 ng/ml and those with levels above the upper limit were defined as having levels ≥20 ng/ml. Positivity for 14-3-3η was defined by the manufacturer at ≥0.19 ng/ml.

Quantitative variables were presented as mean (standard deviation (SD)) or as the median and 25th–75th percentiles (interquartile range (IQR)). Categorical variables were presented with frequencies and percentages. 14-3-3η levels at each visit were compared to baseline levels using the Wilcoxon sign rank test. SDAI scores and remission status, SvH score, ΔSvH and ΔErosion were compared with baseline 14-3-3η positivity using the Mann–Whitney U test, independent samples Student t test and Pearson’s chi-square test, as appropriate. To evaluate the benefit of using positive 14-3-3η protein to support RA diagnosis among EPA patients, we combined elevated 14-3-3η levels with positive RA-associated antibodies. The incremental benefit of positive 14-3-3η over single types of antibody or combinations of antibodies was calculated as the patients identified by a positive 14-3-3η result among patients negative for the biomarker of reference, divided by the number of patients positive for the biomarker of reference. The significance of the observed incremental benefit was evaluated using the McNemar test. Receiver operator characteristic (ROC) curves were used to establish the optimal threshold of baseline 14-3-3η positivity for prediction of ΔSvH ≥5 from inclusion to 5 years. Spearman correlation was used to evaluate association between baseline variables. Generalized linear mixed (GLMM) models with repeated measures were used to evaluate effect of baseline 14-3-3η positivity on SDAI, SvH score, ΔSvH and ΔErosion over time. Generalized estimating equations (GEE) for binary outcomes with repeated measures were used to measure relative risk (RR) of attaining SDAI remission, ΔSvH ≥5, ΔErosion ≥5 and use of biologic DMARDs over time according to 14-3-3η positivity at inclusion or at the previous visit. Multivariate GLMM and GEE models were computed to evaluate which baseline variables explained progression of SvH score over time. Age, gender, 14-3-3η, antibodies and CRP were analyzed as continuous or categorical variables. In analyses with repeated measures, the covariance structure (autoregressive, compound symmetry, variance components or unstructured) with the lowest Akaike information criteria (AIC) for GLMM or quasi-AIC (QIC) for GEE was used to model the subject variation. All variables and interaction terms were included in the first model. We then deleted one by one the variables or interaction terms that were not significant; when deletion of a variable or interaction term resulted in an increase rather than a decrease in the AIC or QIC, this variable or interaction term was kept in the model. This process continued until we reached the smallest AIC or QIC. All analyses were based only on available data without imputation, as fewer than 5 % of values for each variable were missing. Statistical analysis was performed using SAS software version 9.3, SPSS software version 23.0 and GraphPad Prism Software version 6.00 for Windows. A p value <0.05 denoted statistical significance.

As of 15 May 2014, from the 688 included in the ongoing EUPA cohort, 331 patients (62 % women, mean age 60 years) had completed 5 years of follow up and were selected for this study (Table 1). Median symptom duration at baseline was 3 months, and over 92 % already fulfilled either the 1987 ACR or the 2010 ACR/EULAR classification criteria for RA. We previously reported that not meeting the 1987 ACR criteria for RA at baseline did not significantly impact subsequent outcomes in our patients [10, 16]. Disease activity was moderate to high: median (IQR) SDAI 30.1 (19.8–45.2); median M-HAQ (IQR) 0.8 (0.4-1.4). Baseline joint damage was low, with median total SvH (IQR) of 2 (0–6) and median erosion SvH (IQR) of 1 (0–3). Patients rapidly received DMARDs, usually methotrexate (MTX), alone or in combination with other DMARDs [8‐10, 16].

Baseline levels of 14-3-3η protein were ≥0.19 ng/ml (threshold suggested by the manufacturer) in 153 patients (46.2 %) and ≥0.50 ng/ml (the optimal prognostic threshold defined in our cohort; see below) in 119 patients (36.0 %). CRP was >8.0 mg/ml in 207 patients (62.5 %), and RF, anti-CCP2 and anti-Sa antibodies (Abs) were positive in 146 (44.1 %), 133 (40.2 %) and 73 (22.1 %) of patients, respectively; 170 patients (51.5 %) had at least one positive antibody.

Median (IQR) 14-3-3η levels significantly decreased between baseline (0.14 (0.03–1.86) ng/ml) and each follow-up measurement: 0.11 (0.03–1.12), p <0.0001; 0.11 (0.03–0.99), p <0.0001; 0.10 (0.02–0.93), p <0.0001; and 0.13 (0.02–1.14) ng/ml, p = 0.001 at 18, 30, 42 and 60 months, respectively (Additional file 1). The mean (+/- SD) decrease in 14-3-3η titers between baseline and subsequent visits was 0.72 (5.20), 0.62 (5.98), 0.86 (5.54) and 0.58 (6.11) ng/ml at 18, 30, 42 and 60 months, respectively. As a consequence, the proportion of patients remaining 14-3-3η-positive among those with baseline levels ≥0.50 ng/ml dropped to 70 % and 75 % at 42 and 60 months, respectively (Additional file 1). On the contrary, 90–95 % of the patients with baseline 14-3-3η <0.50 ng/ml remained negative over follow up (Additional file 1).

To assess the potential of 14-3-3η assessment to help primary care providers consider RA diagnosis in EPA patients, we combined elevated 14-3-3η levels with positive RA-associated antibodies. Using the manufacturer’s suggested positivity cutoff of ≥0.19 ng/ml increased the number of patients with at least one of the four biomarkers to 194 (58.8 %), a 14.1 % incremental benefit (Table 2). Of the 24 additional patients identified by elevated 14-3-3η, 20 already fulfilled 1987 and/or 2010 criteria for RA at baseline, despite being antibody-negative.

The higher the 14-3-3η baseline levels the stronger the association with radiographic progression between baseline and each of the follow-up evaluations (r approximately 0.19, p <0.001). Similarly, the higher the decrease in 14-3-3η titers between baseline and 18 months, the lesser the radiographic progression from 18 to 30 months (r = –0.14, p = 0.018); ROC curve analyses defined a decrease of <0.76 ng/ml as the best to predict definite radiographic progression (ΔSvH ≥5) over 5 years: 28.6 % vs 14.3 %, RR (95 % CI) = 2.00 (1.20–3.34), p = 0.01 (area under the curve (AUC) = 0.567, sensitivity = 0.844, specificity = 0.308). The 21 patients whose 14-3-3η status reverted from ≥0.19 at baseline to negative by 18 months had lower median (IQR) baseline titers: 0.3 (0.3–0.6) versus 3.6 (1.0–21.0) ng/ml in those who remained positive. As low or transiently positive titers of a biomarker are usually less specific and have weaker associations with specific outcomes, ROC curves were drawn to define the optimal level of baseline 14-3-3η to predict radiographic outcomes. Using two thresholds for ΔSvH or ΔErosion (3 or 5 units) and the shortest length of observation (3 years), the optimal level of baseline 14-3-3η was 0.490 ng/ml (ROC curves not shown). As no patient had values between 0.490 and 0.500, we therefore selected 0.50 ng/ml as the optimal prognostic threshold for 14-3-3η. The threshold of 0.50 ng/ml at baseline was associated with more erosive progression (ΔErosion ≥5) over 5 years (GEE: RR (95 % CI) = 2.01 (1.49-2.72), p <0.001) compared to levels <0.19 ng/ml (Table 3). The 31 patients with baseline 14-3-3η levels between 0.19 and 0.49 ng/ml and the 169 patients with levels <0.19 ng/ml had similar risk for total SvH and erosion progression (Table 3). Compared to those with 14-3-3η ≥0.19 but <0.50 ng/ml, patients with levels ≥0.50 ng/ml had higher risk of definite radiographic progression (ΔSvH ≥5; RR 1.82 (1.11–3.00), p <0.05) and erosive progression (ΔErosion ≥5; RR 2.15 (1.21–3.83), p <0.01). As a consequence, a threshold for positivity of ≥0.50 ng/ml was selected for presentation in this report. Results of some analyses using a threshold of ≥0.19 ng/ml are shown in Additional file 2.

At baseline, median SDAI, DAS28-CRP, and total and erosion SvH levels were not significantly different based on positive or negative 14-3-3η levels ≥0.50 ng/ml; similar results were obtained with a threshold ≥0.19 (data not shown). When patients were grouped according to 14-3-3η-positive status at baseline, GLM analysis revealed that SDAI scores were significantly higher (p = 0.002) across all visits in 14-3-3η-positive patients at baseline (data not shown). GEE analysis revealed that fewer patients achieved SDAI remission based on a baseline 14-3-3η ≥0.50, delivering a relative risk (RR) of 0.79 (95 % CI 0.64–0.98), p = 0.03 (Fig. 1a); similar results were obtained with the 0.19 ng/ml threshold (Additional file 2A). Not surprisingly, baseline 14-3-3η ≥0.50 ng/ml was also associated with an increased risk of definite radiographic progression (ΔSvH ≥5; RR =1.60 (95 % CI 1.28–2.00), p <0.001; ΔErosion ≥5; RR =2.04 (95 % CI 1.53–2.70), p <0.001 (Fig. 1b), and a trend for use of biologic DMARDs over time (RR =1.62 (0.94–2.80), p = 0.085). Baseline 14-3-3η ≥0.19 ng/ml was associated with a lower risk for definite radiographic progression (ΔSvH ≥5; RR =1.39 (95 % CI 1.11–1.74), p = 0.005 (Additional file 2B). As a consequence, patients with baseline 14-3-3η ≥0.50 ng/ml had higher mean radiographic progression at 60 months (mean ± SD for ΔSvH: 13.08 ± 17.34 vs 7.30 ± 13.38, p = 0.002 and ΔErosion: 9.23 ± 13.22 vs 4.37 ± 9.05, p = 0.001).

Only 308/331 patients had a full series of baseline, 18-month and 30-month radiographs; the 23 without either 18-month or 30-month radiographs had similar 14-3-3η levels at baseline and at 30 months. Of the 308 patients with complete radiographs, 84 (27.3 %) were 14-3-3η-positive (≥0.50) at both baseline and 18-month visits, and 23 patients (7.5 %), who were positive at baseline, reverted to negative and 11 patients (3.6 %), who were initially negative converted to positive at 18 months. The 23 patients who sero-reverted from positive 14-3-3η to negative at 18 months had mild progression between 18 and 30 months, similar to the 190 persistently negative patients (1.77 ± 3.83 vs 2.74 ± 3.02, p = 0.243). On the contrary, patients who were 14-3-3η-positive at 18 months developed significantly higher ΔSvH over the following years, both in those with active disease as assessed by the SDAI at 18 months and to a lesser extent, in patients in SDAI remission at 18 months (Fig. 2a). Radiographic progression slowed but persisted after the 30-month visit, despite a gradual increase in use of biologic DMARDs (6.4 %, 11.6 %, 13.8 % and 14.8 % at 18, 30, 42 and 60 months, respectively) and in the prevalence of SDAI remission (Fig. 1a). Nonetheless, 14-3-3η ≥0.50 ng/ml at each visit correlated with more rapid progression over the following years, even in those who were in SDAI remission at the same visit (Fig. 2b).

Before increasing the range of available biomarkers used for outcome prediction in RA, we need to estimate the additional information contributed by 14-3-3η detection. Table 4 presents the Spearman correlation between baseline variables. As expected, antibodies correlated moderately/strongly with each other. Interestingly, positive antibodies and 14-3-3η ≥0.50 ng/ml were also moderately correlated (r = 0.679 for RF, 0.539 for anti-CCP2 and 0.437 for anti-Sa). Age was moderately correlated with SvH scores, and 14-3-3η levels negatively correlated weakly with age (r = -0.140, p = 0.01). Levels of CRP and 14-3-3η did not correlate (Spearman correlation r =0.001; p = 0.996) (Table 4).

Univariate analyses of single biomarkers and their combinations were evaluated in relation to total and erosive joint damage progression over 5 years of disease (Table 3). Definite progression of total (ΔSvH ≥5) and erosive joint damage (ΔErosion ≥5) were associated in univariate analyses with the following baseline variables: CRP >8.0 mg/L, age ≥65 years, positive anti-Sa, positive RF, positive anti-CCP2 and 14-3-3η ≥0.50, with the RR ranging from 1.51–2.22 (Table 3). Relative to erosive progression, a slightly weaker RR for total SvH progression was observed with all variables, with the exception that age ≥65 years was more strongly associated with total than erosive progression. Coexistence of both 14-3-3η ≥0.50 and CRP >8.0 increased the RR of erosive progression ≥5 to 3.48 (2.17–5.56), p <0.001 relative to the absence of both; combining 14-3-3η ≥0.50 with age ≥65 years increased the RR of erosive progression to 3.46 (2.21–5.42), p <0.001. Combining 14-3-3η ≥0.50 with either positive RF, anti-CCP2 or anti-Sa did not markedly improve the RR for erosive progression associated with antibodies (2.27, 2.32 and 2.80, respectively). The optimal univariate combined predictor for erosive progression ≥5 was obtained using a combination of CRP >8.0 mg/L and 14-3-3η ≥0.50 ng/ml and age ≥65 years, with a RR of 5.49 (2.73–11.08) compared to the absence of all three variables (Table 3).

Multivariate predictive models using continuous and dichotomous baseline biomarkers and their significant interactions were also evaluated in relation to definite total and erosive damage progression over 5 years of disease (Table 5). Multivariate GLM analysis with repeated measures of continuous variables showed that baseline age, CRP levels, positive anti-Sa status and 14-3-3η levels were the independent variables significantly associated with both ΔSvH ≥5 and ΔErosion ≥5 over time. In multivariate GEE analysis with repeated measures using dichotomous variables and their interactions, age ≥65 years, CRP >8.0 mg/L and 14-3-3η ≥0.50 ng/ml were the independent predictors of ΔSvH ≥5; age ≥65 years, CRP >8.0 mg/L and 14-3-3η ≥0.50 were again the significant predictors of erosive progression ≥5, together with anti-Sa and RF positivity, and the interaction of 14-3-3η ≥0.50 with antibody positivity (Table 5).