Subtype-specific peripheral blood gene expression profiles in recent-onset juvenile idiopathic arthritis




To identify differences in peripheral blood gene expression between patients with different subclasses of juvenile idiopathic arthritis (JIA) and healthy controls in a multicenter study of patients with recent-onset JIA prior to treatment with disease-modifying antirheumatic drugs (DMARDs) or biologic agents.


Peripheral blood mononuclear cells (PBMCs) from 59 healthy children and 136 patients with JIA (28 with enthesitis-related arthritis [ERA], 42 with persistent oligoarthritis, 45 with rheumatoid factor [RF]–negative polyarthritis, and 21 with systemic disease) were isolated from whole blood. Poly(A) RNA was labeled using a commercial RNA amplification and labeling system (NuGEN Ovation), and gene expression profiles were obtained using commercial expression microarrays (Affymetrix HG-U133 Plus 2.0).


A total of 9,501 differentially expressed probe sets were identified among the JIA subtypes and controls (by analysis of variance; false discovery rate 5%). Specifically, 193, 1,036, 873, and 7,595 probe sets were different in PBMCs from the controls compared with those from the ERA, persistent oligoarthritis, RF-negative polyarthritis, and systemic JIA patients, respectively. In patients with persistent oligoarthritis, RF-negative polyarthritis, and systemic JIA subtypes, up-regulation of genes associated with interleukin-10 (IL-10) signaling was prominent. A hemoglobin cluster was identified that was underexpressed in ERA patients but overexpressed in systemic JIA patients. The influence of JAK/STAT, ERK/MAPK, IL-2, and B cell receptor signaling pathways was evident in patients with persistent oligoarthritis. In systemic JIA, up-regulation of innate immune pathways, including IL-6, Toll-like receptor/IL-1 receptor, and peroxisome proliferator–activated receptor signaling, were noted, along with down-regulation of gene networks related to natural killer cells and T cells. Complement and coagulation pathways were up-regulated in systemic JIA, with a subset of these genes being differentially expressed in other subtypes as well.


Expression analysis identified differentially expressed genes in PBMCs obtained early in the disease from patients with different subtypes of JIA and in healthy controls, providing evidence of immunobiologic differences between these forms of childhood arthritis.

Juvenile idiopathic arthritis (JIA) encompasses chronic childhood arthritis of unknown cause and is manifest by diverse clinical symptoms and outcomes (1). Patients with persistent oligoarthritis have cumulative involvement of fewer than 5 joints, whereas those with extended oligoarthritis have involvement of 5 or more joints, some time after 6 months of disease. Polyarthritis is defined as the involvement of 5 or more joints within the first 6 months of disease and is subdivided according to the presence or absence of rheumatoid factor (RF). Enthesitis-related arthritis (ERA) typically affects older (>6 years) male children, who frequently have HLA–B27 and may have a family history of spondylarthropathy. Systemic JIA consists of chronic arthritis and associated systemic features that may include quotidian fevers, erythematous rash, generalized lymphadenopathy, and hepatosplenomegaly.

Heterogeneity of JIA can be partly accounted for by interactions of complex genetic and environmental factors. While some genetic associations reported in other autoimmune diseases are also found in JIA, there are additional genetic factors that are unique to JIA and specific for the JIA subtypes (for review, see ref.2). These include well-documented HLA class II associations (for review, see ref.3) and subtype-specific genetic linkages (4). Understanding the interacting genetic traits may one day contribute to determining the diagnosis and prognosis of JIA.

Outcomes in JIA are variable and range from complete recovery to persistently active arthritis, with subsequent joint destruction and/or ankylosis that produce significant disability. For most patients, long-term outcome is difficult to predict early on, and identification of patients that would benefit from early aggressive treatment is uncertain. Whole-genome gene expression analysis has significant discovery potential regarding JIA classification, prognosis, and pathogenesis. The genome-wide coverage of this technology offers an unbiased view of disease processes and can generate novel hypotheses, since it does not involve investigating specific genes of interest based on any previous understanding of the disease. This comprehensive approach has been successfully applied to several rheumatic diseases, including systemic lupus erythematosus (SLE) (5) and some forms of JIA (6–10).

In the present study, we analyzed gene expression in peripheral blood mononuclear cells (PBMCs) from a large cohort of patients with JIA of recent onset, prior to treatment with disease-modifying antirheumatic drugs (DMARDs) or biologic agents. We found that for each subtype of JIA, the PBMC gene expression patterns could largely distinguish the JIA patients from the healthy control subjects. To our knowledge, this is the first comparison of PBMC gene expression profiles in multiple subtypes of recent-onset JIA.


Patients and controls.

Following informed consent, patients with recent-onset JIA were enrolled at 5 clinical sites and were followed up for as long as 2 years. The clinical sites were Cincinnati Children's Hospital Medical Center (CCHMC; 61 patients), Schneider Children's Hospital (28 patients), Children's Hospital of Philadelphia (26 patients), Children's Hospital of Wisconsin (14 patients), and Toledo Children's Hospital (Toledo, OH; 7 patients). Patients were classified according to the criteria of the International League of Associations for Rheumatology (11), using the cumulative clinical and laboratory information available from all study visits.

Patients were generally enrolled early in their disease course (median 5 months; 69% <6 months and 90% <12 months), and those with relatively long disease duration were retained because of slow disease evolution (3 with ERA) or delayed initiation of DMARD therapy (1 with persistent oligoarthritis, 9 with RF-negative polyarthritis, and 1 with systemic JIA). Patients had not received DMARDs or biologic agents (antimalarials, azathioprine, cyclosporine, tacrolimus, gold salts, leflunomide, methotrexate, penicillamine, sulfasalazine, adalimumab, etanercept, infliximab, or other biologic agent) prior to sample acquisition. Most patients were receiving nonsteroidal antiinflammatory drugs, and some of them were receiving other medications (atenolol in 1, homatropine eyedrops in 1, corticosteroid eye drops in 3, omeprazole in 2, and oral corticosteroids in 4 [dosage range 0.5–1 mg/kg/day]). Five patients received intraarticular corticosteroids within 30 days prior to sampling (within 4 days [RF-negative polyarticular JIA], 24 days [ERA], 27 days [oligoarticular JIA], 27 days [RF-negative polyarticular JIA], and 29 days [RF-negative polyarticular JIA]), and all of them had joints with active disease at the time of sampling. Of the 9 patients who had previously taken steroids, 8 clustered with their respective subtypes, with only 1 being an outlier (see discussion of Figure 2 below). None of the patients appeared to have another inflammatory disease in addition to JIA.

The 59 controls were apparently healthy children from the Cincinnati area. Because of the wide range of demographic features among the JIA subtypes, it was impossible for the controls to be perfectly matched with each subtype. To account for this difference in characteristics, a broad age range of controls was included. Most of the controls were white, since most JIA patients were also white. Sex-related probe sets were removed from the analysis, as discussed below (see Results). Selected characteristics of the JIA patients and control subjects are shown in Table 1.

Table 1. Selected characteristics of the healthy control subjects and JIA patients, by JIA subtype*
 Healthy controls (n = 59)JIA patients (n = 136)
ERA (n = 28)Persistent oligoarthritis (n = 42)RF-negative polyarthritis (n = 45)Systemic (n = 21)
  • *

    JIA = juvenile idiopathic arthritis; ERA = enthesitis-related arthritis; RF = rheumatoid factor; NA = not applicable.

% female5811718238
Age at JIA onset, median (range) yearsNA12.65 (6.4–16.9)4.35 (1.1–13.8)7.6 (1.2–16.3)4.5 (0.8–15.7)
Age at sampling, median (range) years8.7 (1.8–23.8)13.25 (6.9–17.2)4.85 (1.3–14.4)8.8 (1.5–16.8)4.5 (1–15.9)
Duration of disease at sampling, median (range) yearsNA0.4 (0–2.7)0.35 (0–1.2)0.5 (0.1–3.3)0.2 (0–2.5)
No. of joints with active disease at sampling, median (range)NA0.5 (0–17)1 (0–4)9 (0–45)8 (0–64)
Primary race     
 African American43122
No. of patients with a joint count of 0 at samplingNA14411

Sample preparation.

Peripheral blood was collected using acid citrate dextrose. PBMCs were isolated over Ficoll, and RNA was immediately stabilized in TRIzol reagent (Invitrogen, Carlsbad, CA). Processing was accomplished as quickly as possible, as measured by the time to freezing (the length of time between phlebotomy and freezing in TRIzol). Samples were stored at –80°C at the collection site prior to shipment on dry ice to the CCHMC. RNA was purified at the CCHMC using RNeasy columns (Qiagen, Valencia, CA), then stored in water at –80°C. RNA samples were randomized into groups of 11, and a universal standard was included in each group to provide a technical replicate for measuring batch-to-batch variations. The universal standard was comprised of pooled PBMC RNA from 35 healthy adult volunteer donors.


RNA quality was assessed in the CCHMC Affymetrix GeneChip Core facility using an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA) according to standard protocols. RNA (100 ng) was labeled using NuGEN Ovation version 1 (NuGEN, San Carlos, CA). Labeled complementary DNA (cDNA) was hybridized to Affymetrix Human Genome U133 (HG-U133) Plus 2.0 arrays (Affymetrix, Santa Clara, CA) and scanned with an Agilent G2500A GeneArray Scanner. Data from GeneChips were assessed for quality using a combination of positive and negative spike-in controls, percentage present calls, and average background.

Real-time polymerase chain reaction (PCR).

RNA from PBMCs was reverse transcribed using a blend of oligo(dT) and random hexamers provided in the iScript cDNA Synthesis kit (Bio-Rad, Hercules, CA). Real-time PCR reactions were performed in a 20-μl volume using an iCycler instrument (Bio-Rad), gene-specific primers, and TaqMan probes for haptoglobin, interleukin-10 (IL-10), membrane-spanning 4 domain subfamily A, member 4A (MS4A4A), and suppressor of cytokine signaling 3 (SOCS-3) (TaqMan Assays-on-Demand; Applied Biosystems, Foster City, CA). Raw data were normalized and expressed relative to the housekeeping gene tubulin.

Statistical analysis.

Microarray data were imported into GeneSpring GX 7.3.1 software (Agilent Technologies) and preprocessed using robust multichip analysis (RMA), followed by normalization of each probe to the median of all samples. Distance-weighted discrimination was used to align centroids of predefined groups (12–16) to control for batch-to-batch variation. GeneChip data are available through Gene Expression Omnibus at the National Center for Biotechnology Information (NCBI) (17), series accession number GSE13501 (

A supervised analysis was performed using analysis of variance (ANOVA; Benjamini-Hochberg false discovery rate 5%) followed by Tukey's post hoc testing to identify genes with differential expression between predefined groups. Hierarchical clustering of samples using genes selected by supervised analysis was performed using Pearson's correlation. Clustering using Spearman's correlation gave similar results, and the stronger gene clusters were stable when using different correlations (data not shown). Gene lists were analyzed using Ingenuity Pathways Analysis software (Ingenuity Systems, Redwood City, CA) to identify biologic pathways with differential expression.


Quality control and data management.

In preparation for a study of this magnitude and duration, with multiple investigators and multiple centers, extensive quality control measures were instituted to reduce variation and to ensure reliability of results (Figure 1). Prior to sample collection and processing, a standardized protocol was established and adopted at each center, and in-person training was provided. Emphasis was placed on minimizing the delay in sample processing. A pilot study was performed in which several identical aliquots of peripheral blood were maintained at room temperature for various periods of time, then processed and analyzed in parallel. Consistent with other studies (18,19), extended time to freezing was associated with changes in the expression of a subset of genes (Barnes MG, et al: unpublished observations). Thus, samples with a time to freezing that was >240 minutes (∼5% of samples) were excluded from analysis.

Figure 1.

Flow chart showing quality control (QC) steps. In a study of this size and duration, quality control of sample collection and processing is essential. The quality control plan consisted of 3 general categories: sample collection, sample processing (array), and data processing (data set). The percentage of samples removed at each step is indicated. Samples were collected at multiple centers, and all RNA was isolated at the Cincinnati Children's Hospital Medical Center (CCHMC). RNA samples were sent to the Affymetrix GeneChip Core facility at CCHMC, where probes were synthesized and hybridized to Affymetrix Human Genome U133 Plus 2.0 GeneChips. Data were preprocessed and quality controlled. Data from 54,459 probe sets and 195 arrays that passed all quality control measures were used in all further analyses. PBMC = peripheral blood mononuclear cells; TTF = time to freezing; US = universal standard; RMA = robust multichip analysis; DWD = distance-weighted discrimination.

Prior to labeling and hybridization, RNA quality was assessed according to standard protocols of the CCHMC Affymetrix GeneChip Core facility (Figure 1, RNA quality control). Poor-quality RNA was infrequent, but resulted in removal of ∼2% of samples. An additional 8% failed to label or hybridize properly, based on initial evaluation of microarray results (Figure 1, Core Microarray quality control), but virtually all of these samples were rerun successfully.

Gene expression data were subjected to RMA preprocessing and were then normalized to the median of all samples. Technical variations between batches of samples were monitored using the universal standard (see Patients and Methods). Since the universal standard was pooled RNA and was identical for multiple runs, the batch-to-batch variations in labeling and hybridization were monitored. Distance-weighted discrimination, a method that aligns centroids of predefined cohorts, was applied to adjust for this variation (14). This variation may not have been identified without the universal standard, suggesting that this experimental design should be considered for other large microarray studies.

Comparing samples from males and females, 216 probe sets identified genes whose expression differed according to sex. Interestingly, 70 of these probe sets were not encoded on either the X or the Y chromosome. These 216 probe sets were removed from further analysis to reduce differences in gene expression caused by changes in sex ratios. Therefore, our study used 54,459 probe sets from the Affymetrix HG-U133 Plus 2.0 GeneChip.

Genome-wide differences in expression.

Comparison across samples.

PBMCs were collected from JIA patients (ERA, persistent oligoarthritis, RF-negative polyarthritis, and systemic subsets) early in the disease and prior to treatment with DMARDs or biologic agents. Information used to classify patients included all clinical and laboratory data that became available during the 2-year study. For example, a patient could have been classified at baseline as having probable oligoarthritis, but later, as more joints became involved, the patient would have been reclassified as having RF-negative polyarthritis.

Comparison of samples from patients and controls (by ANOVA; false discovery rate 5%) identified 9,501 differentially expressed probe sets. Tukey's post hoc test identified 193, 1,036, 873, and 7,595 probe sets that were different between controls and the ERA, persistent oligoarthritis, RF-negative polyarthritis, and systemic JIA patients, respectively (see Supplementary Table 1, available on the Arthritis & Rheumatism Web site at These probe set lists represent 5,671, 148, 703, 608, and 4,643 unique Gene Symbol annotations (total ANOVA, ERA, persistent oligoarthritis, RF-negative polyarthritis, and systemic JIA, respectively) according to NetAffx (online at These gene numbers should be interpreted with caution, since many probe sets are either annotated with >1 gene, are not annotated with any gene, refer to predicted or hypothetical genes or proteins, or may hybridize to additional genes than are annotated. The overlap between probe set lists was relatively small, with each subtype having many unique probe sets (154, 649, 479, and 6,741 for ERA, persistent oligoarthritis, RF-negative polyarthritis, and systemic arthritis, respectively), indicating subtype-specific gene expression differences. Only 7 probe sets (6 genes) were common to all 4 lists for the JIA subtypes. The β-amyloid precursor protein binding protein 2 (APPBP2), 2 zinc-finger proteins (ZNF230 and ZND451), and 2 open-reading frame proteins (C15orf17 and C14orf012) were underexpressed in patients as compared with controls. Monocyte-to-macrophage differentiation–associated protein (MMD) was overexpressed in all patient groups.

Results were confirmed by an independent method using samples from 14 patients with systemic arthritis and 16 healthy controls. Real-time PCR was performed for 4 messenger RNAs (mRNA) identified by GeneChip analysis as being overrepresented in patients with systemic arthritis. Compared with controls, the relative expression of target mRNA in systemic arthritis was increased by 3.3 (haptoglobin), 5.7 (IL-10), 2.6 (MS4A4A), and 4.4 (SOCS-3). This is consistent with increased expression as measured by microarray analysis (6.6, 1.2, 3.7, and 2.7, respectively).

Supervised hierarchical clustering was applied to each list of differentially expressed probe sets (Figure 2). This procedure assists with visualization and is expected to produce groupings of largely homogeneous samples according to the sample designations that were used for gene selection. For ERA, persistent oligoarthritis, and RF-negative polyarthritis, the gene expression patterns were similar among many JIA patients, as shown by coclustering; however, a number of JIA patients coclustered with the healthy controls (Figures 2A–C). The gene expression patterns in patients with systemic JIA were more homogeneous, with only 2 outliers (Figure 2D).

Figure 2.

Hierarchical clustering of differentially expressed genes, by juvenile idiopathic arthritis (JIA) subtype. Differentially expressed genes were identified using analysis of variance (false discovery rate 5%), followed by Tukey's post hoc testing. The normalized expression level for each gene (rows) in each sample (columns) is indicated by colors. Red, yellow, and blue rectangles reflect expression levels that are greater than, equal to, or less than the mean expression level in all samples, respectively. Colored lines below the clusterings indicate the diagnoses: enthesitis-related arthritis (ERA) in orange, persistent oligoarthritis (oligo) in light blue, rheumatoid factor–negative polyarthritis (poly) in dark blue, systemic JIA (syst) in yellow, and healthy controls (cont) in green. Bars and letters to the right of each heat map indicate how the gene lists were divided for the Ingenuity Pathways Analysis (see Figure 3 for details).

Biologic meaning of differentially expressed genes (canonical pathways).

Bioinformatic approaches were used to identify cohesive biologic themes in the dataset (6, 20, 21). Overrepresentation of 163 canonical pathways (defined by Ingenuity Systems) was investigated (Figure 3). These pathways were derived from the literature and from the Kyoto Encyclopedia of Genes and Genomes (KEGG) (22) and offer starting points for the investigation of themes in gene expression datasets. The gene lists for ERA, persistent oligoarthritis, and RF-negative polyarthritis were separated into 2 groups for this analysis (labeled a and b Figure 2A, c and d in Figure 2B, and e and f in Figure 2C). The gene list for those with systemic JIA was much larger and was separated into 13 groups based on expression patterns (labeled g–s in Figure 2D). To be considered overrepresented, a pathway had to be ranked in the top 5 and have a P value that was less than or equal to 0.05 by Fisher's exact test (for discovery purposes, no correction for multiple testing was performed).

Figure 3.

Global analysis of differentially expressed genes. Gene lists based on strong patterns in the heat maps (as shown in Figure 2) were evaluated via the Canonical Pathways function of Ingenuity Systems software. To be included in the list of differentially expressed pathways, the pathway had to be overrepresented (P ≤ 0.05) in the list of interest and had to be among the top 5 pathways for that list. The negative logarithm of the P value is indicated on each horizontal axis for each juvenile idiopathic arthritis subtype, and the name of the pathway is shown on the vertical axis at the far left. Pathways are separated into broad categories to simplify interpretation. Letters and numbers indicate which portion of the heat map shown in Figure 2 contributed to the respective pathway. The clustering in Figure 2 was used to identify clusters of under- or overexpressed genes. If a pathway was found in >1 cluster, only the minimal P value (maximal negative decadic logarithm) is represented. ERA = enthesitis-related arthritis; RF = rheumatoid factor; IL-2 = interleukin-2; PDGF = platelet-derived growth factor; TGFβ = transforming growth factor β; EGF = epidermal growth factor; GABA = γ-aminobutyric acid; NRF-2 = NF-κB–repressing factor 2; PPAR = peroxisome proliferator–activated receptor; RXR = retinoid X receptor.

This analysis identified 46 pathways that were overrepresented (Figure 3), primarily related to immunity and inflammation. (Gene lists for these pathways are available from the authors or from Ingenuity Systems.) As expected from the gene expression differences (Figure 2), the relative contributions of each pathway differed between JIA subtypes. In persistent oligoarthritis, pathways for IL-2, B cell receptor, and JAK/STAT signaling were overrepresented. In the persistent oligoarthritis, RF-negative polyarthritis, and systemic subtypes, all exhibited overrepresentation of the IL-10 signaling pathway and the glucocorticoid receptor signaling pathway. The number of overrepresented pathways was highest in systemic JIA, with 34 pathways relating to this subtype. Most notably, up-regulation of innate immune pathways, including IL-6, Toll-like receptor (TLR)/IL-1 receptor (IL-1R), and peroxisome proliferator–activated receptor (PPAR) signaling pathways, as well as the complement system and coagulation cascade, was identified in systemic JIA. Conversely, natural killer (NK) cell, T cell, and antigen-presentation pathways were down-regulated in systemic JIA.

Specific clusters of interest.

In addition to the Ingenuity Pathways Analysis, we selected gene groups of interest and further investigated specific genes by literature searches. The lists presented in Supplementary Table 2 (available on the Arthritis & Rheumatism Web site at are those that were previously identified by ANOVA. (Full gene lists are available upon request from the authors.) Several groups of differentially expressed genes were quite intriguing and offer hypotheses for future consideration.

Hemoglobin/erythrocyte cluster.

Expression of several hemoglobin genes was found to be differentially regulated in JIA subtypes (Supplementary Table 2). In samples from ERA patients, adult hemoglobin α (HBA1/A2 and HBA2) and β (HBB) were down-regulated. In systemic JIA, in addition to up-regulation of adult hemoglobin, there was increased expression of several fetal hemoglobins, including γ (HBG1/G2), δ (HBD), ϵ (HBE1), μ (HBM), and θ (HBQ1) in cluster o (Figure 2D), which also included several erythrocyte structural proteins, cell surface molecules, and enzymes.

Coagulation cascade.

The coagulation cascade was an overrepresented canonical pathway (Supplementary Table 2). Expression of several genes whose products have anticoagulant properties were modulated, including tissue factor pathway inhibitor (TFPI), which was up-regulated in RF-negative polyarthritis and systemic JIA, thrombomodulin (THBD), which was increased in systemic JIA, and protein C (inactivator of coagulation factors Va and VIIIa) (PROC), which was down-regulated in RF-negative polyarthritis. Additionally, genes for a number of procoagulant proteins exhibited increased expression in JIA, with the largest effect seen in systemic JIA.

Complement cascade.

While most complement protein synthesis occurs in the liver, local production also occurs in areas of inflammation by circulating cells (macrophages, dendritic cells, and monocytes [23–29]). Many factors of the complement cascade were found to be differentially expressed in JIA patients (Supplementary Table 2). Patients with persistent oligoarthritis, RF-negative polyarthritis, and systemic JIA showed differential expression of several complement inhibitory proteins (CR1, CR2, CD55, and CD59). Interestingly, patients with ERA had decreased expression of many immunoglobulins, including the complement-fixing IgG1, while there was a slight increase in IgG1 expression in systemic JIA patients (data not shown). Additionally, several factors from the classical pathway of complement activation (C1q, C2, and C4) displayed increased expression in systemic JIA patients.


To our knowledge, this study is the first to comprehensively evaluate gene expression patterns in PBMCs from patients with several subtypes of JIA early in the disease course and prior to initiation of DMARDs or biologic agents. Extensive quality control measures were taken during sample collection and processing and data analysis to ensure validity of results. The results demonstrate that JIA subtypes can be distinguished from healthy controls according to gene expression patterns in PBMCs. The most striking differences were found in patients with systemic JIA, and a number of pathways that were differentially affected in each JIA subtype provide a framework for future investigations of disease pathogenesis.

An important consideration in the interpretation of gene expression profiles obtained from complex mixtures of cells is the influence of cellular composition. Genes that are referred to as up- or down-regulated may actually be over- or underrepresented due to differences in the abundance of the cell populations. This was nicely demonstrated by Bennett et al (5), who found a granulopoiesis signature in SLE, and is likely to be responsible for some of the striking differences seen in systemic JIA (6).

In the current study, there are several examples of disease-specific pathways that were found to be altered. For example, in persistent oligoarthritis, the IL-2, B cell receptor, JAK/STAT, and ERK/MAPK signaling pathways were overrepresented. In RF-negative polyarthritis, pathways representing G protein–coupled receptor and cAMP-mediated signaling and NF-κB–repressing factor 2 (NRF-2)–mediated oxidative stress response signaling were prominent. IL-10 signaling was overrepresented in several subtypes, including persistent oligoarthritis, RF-negative polyarthritis, and systemic JIA, perhaps reflecting activation of antiinflammatory mechanisms. In systemic JIA, the IL-6, TLR/IL-1R, and PPAR pathways were affected, consistent with results recently reported by Ogilvie et al (8). Additionally, networks related to NK cells and T cells were down-regulated in patients with systemic JIA, a pattern that is remarkably similar to that in patients with septic shock (30). These observations support the concept that innate immune activation plays a prominent role in the pathogenesis of systemic JIA and provide evidence of immunobiologic differences between persistent oligoarthritis and RF-negative polyarthritis.

While ERA has not previously been extensively examined, several studies have reported gene expression differences in systemic JIA (6, 8, 10), a limited number have reported differences in RF-negative polyarthritis (7, 9), and a single study has reported differences in persistent oligoarthritis (7), although these JIA subtypes were not generally examined in combination. In systemic JIA, Allantaz et al (10) emphasized the importance of IL-1β by noting its increased production by patient-derived PBMCs and its dramatic responses to IL-1β inhibition in 7 of 9 patients evaluated. In contrast, Ogilvie et al (8) did not see a prominent IL-1β gene expression signature, which may be consistent with recent work by Gattorno et al (31), who identified 2 subsets of systemic JIA based on differential responsiveness to IL-1 blockade (31). In the present study, we did not see prominent overexpression of IL-1–responsive genes, but we found evidence of overrepresentation of TLR/IL-1R pathway genes, raising the possibility of excessive TLR stimulation. Taken together, these observations are consistent with systemic JIA being a heterogeneous disease, and further exploration of the disease subgroups is warranted.

In their study, Allantaz et al (10) defined a set of 12 differentially expressed genes that distinguish systemic JIA from other systemic illnesses, such as viral or bacterial infection and SLE. In our data set, 7 of these 12 genes were differentially expressed in systemic JIA, with a direction and magnitude of change similar to those reported by Allantaz et al. Five genes exhibited increased expression (chloride intracellular channel 2 [CLIC-2], translocation protein 1 [TLOC-1], WNK-1, C18orf10, and ubiquitin B [UBB] 3′-untranslated region [3′-UTR]), while 2 were reduced (WHDC-1 and C18orf17). Differential expression of these genes was specific to systemic JIA and was not seen in the other subtypes. There are several possible explanations for the lack of differential expression of the additional 5 genes noted by Allantaz et al, including medication exposure (such as corticosteroids and methotrexate), disease duration, and disease heterogeneity. Note that comparisons between our results and those reported by Ogilvie et al (8) are limited by the study design. Those investigators examined patients with active and inactive systemic JIA who had well-established disease and were being treated with combinations of corticosteroids, methotrexate, and tumor necrosis factor α inhibitors. They found evidence of IL-6 and IL-10 overexpression, which is consistent with our Ingenuity Pathways Analysis.

Jarvis et al (9) reported peripheral blood leukocyte gene expression changes in patients classified as having polyarticular juvenile rheumatoid arthritis (JRA) according to the American College of Rheumatology criteria, as compared with healthy controls. Comparisons with this study are limited because different cell populations were studied and the microarrays were different. Notably, buffy coat–derived leukocytes contain neutrophils, which are generally absent from Ficoll-purified PBMCs as used in our study.

We have previously reported gene expression differences in PBMCs from JRA patients with a pauciarticular or a polyarticular disease course (7). The main finding was an up-regulation of proangiogenic CXCL chemokines in polyarticular JRA patients as compared with pauciarticular JRA patients or healthy controls. We did not identify these genes as being overexpressed. This is not unexpected, since the most prominent findings of our previous study were based on comparisons of PBMCs with synovial fluid mononuclear cells, which was not part of the current study. In addition, those samples were obtained from patients with longstanding disease (average 9.3 years), were not collected with rigorous attention to quality control, and in many instances, had been stored for several years. Consequently, extensive comparisons with the current study are not possible.

A traditional way to visualize microarray data is by supervised hierarchical clustering. This method returns clusters of the patient subgroups used to derive the gene lists. In the current study, visual inspection of clustering trees suggested that gene expression patterns identified distinct subgroups within each of the JIA subtypes. Thus, the current JIA subtypes may include more heterogeneity than was previously appreciated. Studies to assess this heterogeneity and its impact on classification and prognosis are ongoing.

The decreased expression of the adult hemoglobin gene in patients with ERA was unexpected. Patients with ankylosing spondylitis and other types of spondylarthritis are rarely anemic (32). In fact, in our study, patients with ERA had the highest hemoglobin levels and those with systemic JIA had the lowest (data not shown). We hypothesize that the decreased expression of hemoglobin in ERA might be a response to transforming growth factor β, which has been reported to be overexpressed in ankylosing spondylitis (33) and can act through an activator protein 1 binding site near the hemoglobin gene.

One of the most highly correlated gene groups that distinguished systemic JIA patients from healthy control subjects was the group designated as cluster o in Figure 2D. Examination of this cluster revealed a large group of genes encoding red blood cell structural proteins and enzymes, similar to the erythropoietic signature described previously (6), where it was suggested that the signature was due to the presence of increased immature erythrocyte precursors in the peripheral circulation. This cluster also contains various hemoglobins, including those normally expressed during the embryonic and fetal stages of human development (hemoglobins γ, δ, ϵ, μ, and θ). Under normal conditions, genes encoding fetal hemoglobins undergo silencing soon after birth (34), although these proteins may sometimes be present in adults (35). We hypothesize that the increased expression of adult and fetal hemoglobins in patients with systemic JIA may be explained by the presence of immature erythrocyte precursors.

The coagulation system pathway was identified as being overrepresented in systemic JIA patients, although many of the genes were modulated in the other JIA subtypes. These findings are consistent with the clinical observations of mildly increased levels of D-dimers in a majority of JIA patients, with the largest increase seen in systemic JIA, which suggests a coagulopathy that may correlate with disease activity (36, 37). In innate inflammatory responses, activation of coagulation and fibrin deposition is an important mechanism that helps contain inflammatory activity to the site of injury or infection; however, when not localized, coagulation can have a deleterious effect in patients with systemic inflammatory disorders, such as septic shock or macrophage activation syndrome, a well-known complication of systemic JIA. We hypothesize that the overrepresentation of several anticoagulant proteins, such as TFPI in RF-negative polyarthritis and systemic JIA, indicates an attempt to down-regulate systemic inflammation–induced coagulation and fibrin deposition.

Many factors from the complement cascade were overexpressed in JIA patients (Supplementary Table 2, available on the Arthritis & Rheumatism Web site at While a large proportion of complement synthesis occurs in the liver, production also occurs locally in areas of inflammation by circulating cells (macrophages, dendritic cells, and monocytes [23–29]). Three of the JIA subtypes had increased expression of some complement inhibitory proteins (CR1, CR2, CD55, CD59). CR1 is present on blood cells, including erythrocytes, neutrophils, monocytes, eosinophils, and T and B cells, and is involved in immune clearance and inhibition of the complement cascade by acting as a cofactor for the cleavage of C3b. CR2 is found mainly on B cells and is involved in B cell activation and immune clearance. CD55 and CD59 are markedly up-regulated on activated macrophages and less so on CD4-positive and CD8-positive lymphocytes. This up-regulation could indicate the release of complement factors during poorly controlled inflammation in the joint.

High-throughput technologies such as Affymetrix GeneChip arrays produce vast amounts of data that can be analyzed and interpreted in many different ways. Making large datasets publicly available allows investigators to apply alternative methods of analysis and increase utilization of the data. High-throughput technologies are by their nature hypothesis-generating; thus, the hypotheses presented in this study must be validated in other cohorts, and the biologic relevance of gene expression differences identified here remain to be determined at the protein and organism levels.

In conclusion, this study has identified differences in the expression of genes related to immunity and inflammation in PBMCs from patients with different JIA subtypes as compared with healthy controls. Several differentially expressed genes replicated findings in other studies, providing further confidence in our results, which have the potential to greatly expand our understanding of JIA. Expression patterns indicate that defined JIA subtypes have strong internal similarities, although clustering also indicates some degree of heterogeneity within subtypes. These findings will likely provide important mechanistic information with respect to JIA subtypes and lead to an improved molecular definition of JIA.


All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Barnes had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Barnes, Grom, Thompson, Griffin, Pavlidis, Aronow, Glass, Colbert.

Acquisition of data. Barnes, Thompson, Griffin, Itert, Aronow, Luyrink, Srivastava, Ilowite, Gottlieb, Olson, Sherry, Glass, Colbert.

Analysis and interpretation of data. Barnes, Grom, Thompson, Griffin, Pavlidis, Itert, Fall, Sowders, Hinze, Aronow, Glass, Colbert.


We wish to acknowledge the following individuals for their efforts: Wendy Bommer (clinical research coordinator, Cincinnati Children's Hospital Medical Center), Jane Boyd (study coordinator, Arthritis Foundation, Peoria, IL), Joseph Couri, MD (Methodist Medical Center of Illinois, Peoria, IL), Mark Getz, MD (Order of Saint Francis Medical Center, Peoria, IL), Jesse Gillis, PhD (data preprocessing, University of British Columbia), Anne Johnson (clinical research coordinator, Cincinnati Children's Hospital Medical Center), Marsha Malloy, RN, MBA (study coordinator, Medical Center of Wisconsin, Milwaukee, WI), Beth Martin (study coordinator, Toledo Children's Hospital), Marilyn Orlando (study coordinator, Schneider Children's Hospital), David Wilson (study coordinator, Children's Hospital of Pittsburgh, Pittsburgh, PA), Sara Jane Wilson (study coordinator, Children's Hospital of Philadelphia), and Jeremy Zimmermann (data collection, Medical College of Wisconsin, and Children's Research Institute, Milwaukee, WI).