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

Plasma samples from twenty-one sJIA patients were collected at Astrid Lindgren’s Children Hospital in Stockholm, Sweden. Pediatric rheumatologists diagnosed all patients according to the International League of Associations for Rheumatology (ILAR) criteria [2]. The sJIA patients were 10 males and 11 females, with sampling ages ranging from 3 to 16 years. Disease activity was classified using the criteria approved by the American College of Rheumatology Board of Directors as Provisional [13]. None of the patients has MAS at the time when the samples were collected. Patients were considered as having inactive sJIA when they had no clinical signs of disease activity such as joints with active arthritis, fever, rash, serositis and splenomegaly or generalized lymphadenopathy attributable to sJIA. In addition, when available, ESR and CRP levels had to be within normal ranges and the physician’s global assessment of disease activity score had to be 0. Comprehensive clinical and laboratory data were collected at each plasma-sampling occasion.

Plasma samples from 60 healthy controls (four, eight or twelve years old) without chronic inflammatory or autoimmune diseases, were from a population-based cohort (Barnens miljö- och hälsoundersökning) in the Stockholm region [14].

Fresh blood from all subjects was collected in EDTA tubes, kept at room temperature and centrifuged within 4 h for 10–15 min at 3000–6000 g with brake. Centrifuged plasma was finally stored at − 80 °C until analysis.

Plasma samples were analyzed using a high-throughput, multiplex immunoassay (Proseek Multiplex, Proximity Extension Assay (PEA) technology, Inflammation panel, Olink Bioscience, Sweden). The biomarkers included in the inflammation panel are reported to be involved in multiple inflammatory conditions. The panel includes 92 immune-related proteins, primarily cytokines and chemokines as listed in Supplementary Table S1. The assay utilizes epitopespecific binding and hybridization of a set of paired oligonucleotide antibody probes, which is subsequently amplified using quantitative PCR, resulting in log base-2 normalized protein expression (NPX) values. Proteins with a call rate below 20% were excluded from further analysis, resulting in 69 proteins being included. Details regarding the design of the Inflammation panel, the PEA assay protocol and data pre-processing are outlined in supplementary files.

Quantification of plasma HMGB1 was performed by Enzyme-Linked ImmunoSorbent Assay (ELISA) (IBL International, Germany) according to the manufacturer’s instructions.

In the cross-sectional analysis, ordinary two-way ANOVA was performed on active sJIA (n = 14), inactive sJIA (n = 16) and healthy controls (n = 30). Multiple t-test was performed on active sJIA (n = 14) versus healthy controls (n = 28) and on inactive sJIA (n = 16) versus healthy controls (n = 32), separately. In each cross-sectional analysis, the healthy control group was age and sex matched to the patient group. In the paired analysis, two-way repeat-measurement ANOVA was performed on paired active sJIA (n = 9) and inactive sJIA (n = 9) samples from 9 patients. All the statistical analyses were corrected for multiple comparison by controlling the False Discovery Rate (FDR) via two-stage step-up method of Benjamini, Krieger and Yekutieli. Adjusted p-values less than 0.05 were regarded as significant. The tests were performed by using GraphPad Prism version 8.4.3 (San Diego, CA, USA).

Hierarchical Clustering Analysis (HCA) and Principal Component Analysis (PCA) analysis was performed by ClustVis Web Tool [15] to visualize the clustering of samples. Feature ranking process using Random Forest (RF) algorithm was executed using Python Jupyterlab 1.2.6.

Proteins with significant separation between two compared groups were analyzed for the enrichment in biological signaling pathways using Ingenuity Pathway Analysis (IPA) software (QIAGEN Inc.). Fold-change was calculated by dividing the average NPX value of sJIA (or active sJIA) by average NPX value of healthy controls (or inactive sJIA). Expression core analysis was based on fold-change of the statistically significantly separated proteins, including direct and indirect relationships, interaction and causal networks, all node types and data sources, experimentally observed confidence, restricted to human tissues and primary cells. Z-scores represent the predicted activation state of upstream regulators using the expression patterns of the downstream factors, based on relationships published in the literature. Statistical analysis was performed using Fisher’s exact test right-tailed within the IPA software.

Additional details regarding study design and statistical analysis are outlined in Supplementary Fig. S3 and Supplementary methods.

Based on the 69 proteins included in the analysis, the groups of active sJIA, inactive sJIA and healthy controls could be identified but not completely separated by HCA (Supplementary Fig. S4). PCA also showed an overlap, with the active sJIA group more separated by principal components 1 and 2 (PC1 and PC2) from the inactive sJIA group and the healthy control group (Fig. 1A). Random forest analysis identified the proteins that contributed most strongly to the separation of the three groups (Fig. 1B, full data set of the cross-sectional analysis are outlined in Supplementary Table S3), with the top contributing proteins being IL6, KITLG, IL18, TNFB, CXCL1, CCL19, CCL23, S100A12, MMP1, PLAU, CCL2, CXCL5, CST5, OSM, CXCL11 and FGF23, each with importance larger than 0.01. Comparative analysis of the top contributing proteins demonstrated that, 10 of the 16 top proteins were significantly differentially expressed in at least one of the comparisons between the three groups (Fig. 1C). Active sJIA, inactive sJIA and healthy controls were further separated in HCA based on these 10 proteins than by using all 69 recorded proteins (Fig. 1D and Supplementary Fig. S4). Similarly, PCA based on the 10 differentially expressed proteins also resulted in better separation; in particular, the active sJIA group could be separated from inactive sJIA and healthy controls (Fig. 1E and A).

To identify the proteins that could separate the inactive sJIA or the active sJIA group from the healthy control group, cross-sectional analyses were performed. Ten proteins were significantly different between active sJIA and healthy controls, in which seven protein were upregulated (IL6, OSM, IL18, MMP1, S100A12, CXCL11 and EIF4EBP1) while three were downregulated (KITLG, CD6 and TNFSF11) (Fig. 2A and B). Among the ten proteins, only OSM and TNFSF11 were negatively correlated with statistical significance (r = − 0.64, p = 0.028) (Fig. 2C). Levels of six proteins out of the 69 analyzed were significantly different between inactive sJIA and healthy controls (Fig. 2D and E): CASP8, CXCL1, CXCL5, SIRT2 and SULT1A1 showed lower levels in inactive sJIA, and these five proteins were positively correlated with statistical significance, especially CXCL1 and CXCL5 were highly positively correlated (r = 0.86, p < 0.0001). Only one protein showed higher level in inactive sJIA compared to healthy controls, namely IL18. Full data set of the cross-sectional analysis is presented in Supplementary Table S4.

To further analyze whether any proteins could distinguish active and inactive sJIA, paired samples from nine patients obtained during active and inactive disease phases were investigated. 11 proteins had significantly different levels, with IL6, MMP1 and S100A12 having p-values lower than 0.0001 (Fig. 3A). These data were in agreement with the results in the cross-sectional analysis. Full data sets are presented in Supplementary Table S5. Based on the 11 differently expressed proteins rather than including all 69 proteins, active sJIA and inactive sJIA were better separated by HCA (Fig. 3B) and PCA (Fig. 3C).

The disease activity predictive performance of the 11 proteins was examined by enlarging the subject cohort from the nine-patient paired analysis to the 21-patient cross-sectional analysis. Receiver Operating Characteristic (ROC) curves for the 11 proteins in the classification of active and inactive sJIA are shown in Fig. 3D. Only IL6, MMP1, S100A12, OSM, SIRT2, KITLG and TNFSF11 showed statistically significant efficiency and were comparable to CRP (Fig. 3D and E).

To identify activated cell signaling pathways of importance for sJIA, proteins that were differentially expressed in the three comparison pairs, (i) active sJIA versus controls, (ii) inactive sJIA versus controls, and (iii) active sJIA versus inactive sJIA, were investigated using IPA to identify gene ontology groups and relevant canonical signaling pathways. However, no significant predication could be made by IPA, most likely due to the limited numbers of input factors (full data sets are presented in Supplementary Table S6-S11). Top cellular functions which tended to be activated or suppressed, as well as the top canonical pathways are summarized in circular graphs (Fig. 4).

Cell migration, cell invasion, cells homeostasis and inflammatory response appeared to be activated (Fig. 4A), while monocytopoiesis and development of connective tissues tended to be suppressed (Fig. 4B). IL-17 Signaling, HMGB1 Signaling, Granulocyte Adhesion and Diapedesis, Role of Macrophages, Fibroblasts and Endothelial Cells in Rheumatoid Arthritis, Inflammasome pathway were predicted with positive z-score (0 < z-score < 2), while Erythropoietin Signaling Pathway were predicted with negative z-score (− 2 < z-score < 0) (Fig. 4C).

Compared with healthy controls, HMGB1 signaling tended to be upregulated in active sJIA (Fig. 4C). HMGB1 is a damage associated molecular pattern (DAMP), which has been studied in a series of inflammatory diseases, including JIA. Highly increased levels of HMGB1 have also been reported during MAS [16]. Since HMGB1 was not included in the PEA inflammation panel, HMGB1 levels were determined by ELISA. The active sJIA group had significantly higher HMGB1 levels than the inactive sJIA group (Fig. 5A). In paired analysis, eight of the nine patients had higher HMGB1 levels when in active disease phase, one patient had a slightly higher HMGB1 level in the inactive phase (0.84 ng/mL in inactive and 0.76 ng/mL in active) (Fig. 5B). The ROC analysis (AUC = 0.839, 95% confidence interval 0.688–0.989) confirmed the ability of HMGB1 to distinguish active and inactive sJIA (Fig. 5C).