Immunoprofiling of active and inactive systemic juvenile idiopathic arthritis reveals distinct biomarkers: a single-center study

Abstract

Background: This study aimed to perform an immunoprofiling of systemic juvenile idiopathic arthritis (sJIA) in order to define biomarkers of clinical use as well as reveal new immune mechanisms.

Methods: Immunoprofiling of plasma samples from a clinically well-described cohort consisting of 21 sJIA patients as well as 60 age and sex matched healthy controls, was performed by a highly sensitive proteomic immunoassay. Based on the biomarkers being significantly up- or down-regulated in cross-sectional and paired analysis, related canonical pathways and cellular functions were explored by Ingenuity Pathway Analysis (IPA).

Results: The well-studied sJIA biomarkers, IL6, IL18 and S100A12, were confirmed to be increased during active sJIA as compared to healthy controls. IL18 was the only factor found to be increased during inactive sJIA as compared to healthy controls. Novel factors, including CASP8, CCL23, CD6, CXCL1, CXCL11, CXCL5, EIF4EBP1, KITLG, MMP1, OSM, SIRT2, SULT1A1 and TNFSF11, were found to be differentially expressed in active and/or inactive sJIA and healthy controls. No significant pathway activation could be predicted based on the limited factor input to the IPA. High Mobility Group Box 1 (HMGB1), a damage associated molecular pattern being involved in a series of inflammatory diseases, was determined to be higher in active sJIA than inactive sJIA.

Conclusions: We could identify a novel set of biomarkers distinguishing active sJIA from inactive sJIA or healthy controls. Our findings enable a better understanding of the immune mechanisms active in sJIA and aid the development of future diagnostic and therapeutic strategies.

Keywords: Cytokines and inflammatory mediators; High mobility group Box 1; Inflammation; Ingenuity pathway analysis; Proteomics; Systemic juvenile idiopathic arthritis.

Fig. 1

Distribution of the different subgroups based on detected inflammation-associated proteins. A Principal component analysis (PCA) of the three groups based on all included 69 proteins. Confidence level of the ellipses is 0.90. B Random forest analysis resulted in a predictive accuracy of 90.6%. The factor importance plot displays the top contributing proteins with importance higher than 0.01. C Comparative analysis of the proteins with top importance in (B) revealed significant difference among at least one of the three comparisons. Bars represent mean ± standard deviation. D Hierarchical clustering analysis based on the 10 proteins in (C) showing the grouping among active sJIA, inactive sJIA and controls. Unit variance scaling was applied to rows; both rows and columns were clustered using correlation distance and average linkage. E PCA of the three groups based on the 10 proteins shown in (C). The confidence level of the ellipses is 0.90. Statistics: ordinary two-way ANOVA with correction of multiple comparison by controlling the False Discovery Rate (FDR) of 5% via two-stage step-up method of Benjamini, Krieger and Yekutieli, * p < 0.05, ** p < 0.01, *** p < 0.001

Fig. 2

Comparative analyses of protein levels in active sJIA, inactive sJIA and matched-healthy controls. Volcano plot of biomarkers between (A) active sJIA and healthy controls, and (D) inactive sJIA and healthy controls. Dots with colors (blue representing lower levels in patients and red representing higher levels in patients as compared with controls) are significantly different between the compared groups (p < 0.05). B and E Summary of the significantly different proteins in the two comparison pairs. C and F Heat maps show correlation between disease-associated protein levels among sJIA patients. Statistics: A, B, D and E Two-way ANOVA with correction of multiple comparison by controlling the False Discovery Rate (FDR) of 5% via two-stage step-up method of Benjamini, Krieger and Yekutieli. C and F) Spearman correlation, * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001

Fig. 3

Paired analysis of active and inactive sJIA. A Comparison of protein levels in paired samples from nine patients during active and inactive disease phases. Each symbol represents one sample, and the lines link the paired samples. B Hierarchical clustering analysis shows the grouping of active and inactive sJIA based on the eleven significantly different proteins. In the heat map, rows were centered and unit variance scaling was applied to rows; both rows and columns were clustered using maximum distance and average linkage. C Principle component analysis of the three groups. The confidence level of the ellipses is 0.90. D Receiver operating characteristic (ROC) curves examining the predictive performance of the eleven proteins for distinguishing active sJIA (n = 14) in reference to inactive sJIA (n = 16) in the cross-sectional analysis. E Area under the curve (AUC) and corresponding 95% CI for each measure. Statistics: A two-way repeat-measurement ANOVA with correction of multiple comparison by controlling the False Discovery Rate (FDR) of 5% via two-stage step-up method of Benjamini, Krieger and Yekutieli, * p < 0.05, ** p < 0.01, **** p < 0.0001, (E) the 95% confidence interval were calculated by hybrid Wilson/Brown method

Fig. 4

Cellular function and canonical pathways based on the proteins differentially expressed in sJIA. The top cellular functions tended to be activated (A) and suppressed (B), as well as the top canonical pathways (C) were summarized in circular graphs. Except from Erythropoietin Signaling Pathway with negative z-score (− 2 < z-score < 0), all the pathways are with positive z-score (0 < z-score < 2)

Fig. 5

HMGB1 levels in plasma samples obtained during active and inactive sJIA. A Cross sectional analysis of HMGB1 levels in active and inactive sJIA patients. B Paired analysis of HMGB1 levels during active and inactive disease phases. C Receiver operator characteristics (ROC) analysis of HMGB1 concentration in the plasma of sJIA. Statistics: A Mann-Whitney U test, (B) Wilcoxon matched-pairs signed rank test, * p < 0.05, ** p < 0.01, (C) the 95% confidence interval were calculated by hybrid Wilson/Brown method