Gene expression profiling in murine autoimmune arthritis during the initiation and progression of joint inflammation

The use of human cartilage from joint replacement surgeries for antigen isolation was approved by the Institutional Review Board, and all animal experiments were approved by the Institutional Animal Care and Use Committee. Female BALB/c mice at the age of 24–26 weeks (National Cancer Institute, Kingston Colony, New York, USA) were injected intraperitoneally with 100 μg of cartilage PG (measured as protein) emulsified in dimethyldioctadecylammonium bromide (DDA) adjuvant (Sigma-Aldrich, St Louis, Missouri, USA). The use of adjuvant DDA allowed us to avoid the harmful effects of oil and bacterial proteins present in Freund's adjuvants [35, 36]. Booster injections of the same doses of PG with DDA were given on days 21 and 42. BALB/c mice develop swelling and redness of one or more limbs 7–10 days after the second or third injection with PG in adjuvant [25]. Arthritis was assessed daily, and inflammation was scored from grade 0 to grade 4 for each paw [13, 36, 37]. Female SCID mice of the BALB/c background (NCI/NCrC.B-17-scid/scid; henceforth BALB/c^SCID) were used for adoptive cell transfer. BALB/c^SCID mice were purchased from the National Cancer Institute and maintained under germ-free conditions.

To ensure uniformity and reproducibility of disease transfer, donor spleen cells were isolated from arthritic BALB/c mice within 1–2 weeks after the onset of inflammation. At least two paws of donor BALB/c mice were arthritic, and the cumulative inflammation score (for four paws) was in the range 5–8. Spleen cells of arthritic BALB/c mice were collected and cultured in six-well plates (2.5 × 10⁶ cells/ml) with cartilage PG (50 μg/ml) for 4 days in Dulbecco's modified Eagle's medium supplemented with 5% fetal bovine serum (HyClone Laboratories, Logan, Utah, USA). After stimulation in vitro for 4 days with cartilage PG, non-adherent cells were collected, and live cells (lymphocytes) were separated on Lympholyte-M (Cedarlane, Ontario, Canada). Finally, 2 × 10⁷ lymphocytes were injected intraperitoneally on days 0 and 7 into recipient BALB/c^SCID mice as described [32].

A standard scoring system used for primary arthritis was applied to the assessment of disease severity in BALB/c^SCID mice [24, 37]. Typically, one to four paws became inflamed simultaneously 3–5 days after the second cell transfer, and the rest of the peripheral joints became inflamed within 2–4 days after the onset of the first symptoms. BALB/c^SCID mice were scored twice daily, and were killed as soon as the inflamed paw reached an individual arthritis score of 2, but not later than 24 hours after the onset of arthritis. This paw was designated as acute arthritic (AA), and contralateral or ipsilateral paws that were not inflamed at that time were used as pre-arthritic (PA) samples. The PA joints did not show evidence of inflammation on histopathological examination, although thickening of the synovial lining in small joints was observed occasionally (data not shown). Several arthritic BALB/c^SCID mice were scored daily and were killed 8–10 days after disease onset. These joint samples represented subacute-chronically arthritic (CA) samples. In addition to PA, AA and CA experimental conditions, paws of naive non-immunized BALB/c^SCID mice were used as 'absolutely negative' (control naive; AN) samples for RNA isolation and subsequent hybridization. Each sample represented RNA pooled from four paws of two mice.

Synthesis and biotinylation of cRNA and hybridization were performed in accordance with the manufacturer's instructions (Affymetrix, Santa Clara, California, USA). In brief, total RNA was isolated from normal or inflamed paws of mice by using TRIzol reagent (Invitrogen, Gaithersburg, Maryland, USA) with additional purification on RNeasy columns (Qiagen, Valencia, California, USA). RNA quality was confirmed by spectrophotometry and electrophoresis on formaldehyde gels [38]. Double-stranded complementary DNA was synthesized with the T7-dT24 primer incorporating a T7 RNA polymerase promoter. Biotinylated cRNA was prepared with the Enzo BioArray High Yield RNA Transcript Labeling Kit (Enzo Diagnostics, Inc., Farmingdale, New York, USA) and hybridized to the murine genome Affymetrix U74v2 chip set, which included three DNA chips, MG_U74Av2, MG_U74Bv2 and MG_U74Cv2, interrogating more than 36,000 genes that represented essentially the entire mouse genome [39‐42]. Fluorescent hybridization signals were developed with phycoerythrin-conjugated streptavidin and were further enhanced with fluorescently labeled anti-streptavidin antibodies. DNA chips were scanned to obtain quantitative gene expression levels. DNA chip hybridization, Fluidics Station operations, scanning, and preliminary data management were performed in accordance with Affymetrix protocols as described previously [43, 44].

Fluorescent intensity data from Affymetrix Microarray Suite version 5 were exported as CEL files and imported into DNA-Chip Analyzer version 1.3 [45]. Data were normalized, and expression values, based on the perfect match/mismatch (PM/MM) model, were calculated for each DNA chip. All chips were examined for the image spikes, chip and gene outliers. Exported expression values for each DNA chip were combined into a single file (three chips × four experimental conditions × three to five replicates), and imported back to DNA-Chip Analyzer; the resulting data were normalized by using an array with median probe intensity.

For the pairwise comparison of experimental conditions, signals were filtered by using several criteria. Gene expression was considered above the background if it showed the signal on most chips (more than 50%; that is, for three replicates, the gene should be detectable on at least two chips; for five replicates, the gene should be present on at least three DNA chips). Fold changes for gene expression were calculated when any of three following criteria were met: (1) the gene was present in the experimental condition but absent in the basal condition; (2) the gene was present in the basal condition but absent in the experimental condition; (3) the gene was present in both basal condition and experimental conditions. Student's t-test was used to determine the statistical significance of the difference in gene expression between basal and experimental conditions (P < 0.05 was taken as significant). An additional cut-off threshold of twofold change in gene expression (either upregulation or downregulation) was used to characterize a gene as being differentially regulated (for example, a negative twofold value corresponded to a twofold downregulation). The Fisher exact test (implemented by us in Visual Basic code for MS Excel 2000) and the Mann–Whitney U-test (SPSS, Chicago, Illinois, USA) were used to verify non-paired Student's t-test calculations of the probability of gene expression differences in pairwise comparisons. Finally, the false discovery rate was established with 500 permutations for each pairwise comparison to estimate the proportion of false-positive genes.

To characterize gene expression patterns, hierarchical gene clustering was performed with a DNA-Chip Analyzer program [45, 46]. The algorithm was based on the distance between two genes defined as 1 - r, where r is the Pearson correlation coefficient between the standardized expression values of the two genes across the samples used. To characterize functional relationships between differentially expressed genes, Gene Ontology terms classification [47], incorporated in DNA-Chip Analyzer, was performed [48]. The significance level for a functional cluster was set at P < 0.05, and the minimum size of a cluster was three genes. Venn diagram calculations were performed in Visual Basic code for MS Excel 2000 to analyze overlapping of sets of genes differentially expressed in the samples at different phases of arthritis.

Being aware of the importance of data reproducibility, we determined the optimal number of arrays to be included in experimental design by monitoring the convergence of variance for gene expression signals in five replicates representing the condition AA. For each replicate, we pooled equal amounts of quality-controlled RNA samples, isolated from two inflamed paws of two BALB/c^SCID mice that had been identically treated (in terms of the number of donor cells and antigen stimulation) and had similar disease onset and severity. A total of five replicates represented 20 paws of 10 arthritic mice. We used the coefficient of variation (CV) to measure data variability. The CV for each gene on the chip and the mean CV for the entire probe set were calculated. Mean CV reached a plateau when the number of replicates increased beyond three (Fig. 1, experimental condition AA) and there was no significant change afterwards. Therefore, for all other experimental conditions, we used three replicates representing three independent hybridization experiments of three RNA samples isolated from six paws. Mean CV after sampling of the three repeats ranged between 0.21 and 0.25 for all experimental conditions.

Paws of naive BALB/c^SCID mice and still non-inflamed (PA) paws were clinically normal with no sign of inflammation, and comparison of these two experimental conditions (PA/AN) identified a relatively small number of differentially expressed genes. Only 37 of the 36,000 screened genes were differentially expressed (that is, showed greater than a ± twofold change relative to threshold level), of which 11 genes were over the ± threefold threshold, and seven genes changed beyond ± fivefold (Fig. 2). The seven genes with the most significant change in expression levels encoded chemokine CC motif receptor 5 (Ccr5), chemokine CXC motif ligand 1 (Cxcl1), interferon-γ-inducible protein (Ifi47), membrane-spanning 4-domains subfamily A member 6C (Ms4a6c), tumor necrosis factor-α-induced protein 6 (Tnfip6), T cell receptor β variable 13 (Tcrbv13), and Terf1-interacting nuclear factor2 (Tinf2) (Table 2). Although the upregulation of Tcrbv13, Tgtp and interferon-induced genes might indicate the appearance of antigen-specific T cells in the synovium (Table 2), the significant upregulation of Tnfip6 suggests the activation of an anti-inflammatory cascade [51]. Thus, gene expression related to pro-inflammatory and anti-inflammatory events can be detected even before the migration of inflammatory leukocytes into the joints.

To characterize major biological functions in context with the initiation phase of the disease, we assigned the 37 early genes (Table 2, Additional file 1) to separate groups according to the corresponding protein functions and Gene Ontology classification [47, 48]. We found that differentially expressed genes in PA joints were related to immune responses, chemokine activity (including chemotaxis), cell adhesion, proteolysis regulation, inflammation and wounding, cytokines, and cytoskeletal activity (Fig. 3, yellow circles). All clustered genes were upregulated at the pre-inflamed phase of arthritis.

To monitor the progression of disease, we analyzed genes that were differentially expressed in paws with acute and chronic joint inflammation. Both AA and CA experimental conditions were associated with the activity of a large number of genes: 256 genes were upregulated and 21 were downregulated in acute arthritis (AA/AN comparison), and 201 genes were upregulated and 217 were downregulated in chronic inflammation (CA/AN) (Fig. 2, Additional files 2 and 3). A Venn diagram summarizes the relationships between gene sets that were differentially expressed at different phases of the disease. Only 15 genes were differentially expressed in all three phases of the disease (PA, AA, and CA), 25 genes were differentially expressed both at the PA phase and during acute inflammation, 127 genes were active both in acute and chronic phases, and 17 transcripts shared a common expression pattern in pre-inflamed and chronically inflamed joints (Fig. 2).

Using Gene Ontology terms for the functional classification of genes differentially expressed in acute and chronic arthritis [47], dozens of cell signaling pathways and gene clusters were identified. By further filtering of functional clusters, and by combining clusters encoding proteins with similar functions, we found that the acute and chronic phases of the disease can be comprehensively described by the differential expression of 15 macro-clusters (Fig. 3). Six clusters were found in all three phases of inflammation; they were related to immune response, chemokine activity, cytokines, inflammation and wounding, cell adhesion, and proteolysis regulation. The most abundantly represented genes in inflamed joints were those involved in immune responses: 51 genes in AA and 25 genes in CA. These genes were upregulated as much as 31-fold (group average) in acute arthritis and 15-fold in chronic arthritis (Fig. 3). Cytokine and chemokine genes demonstrated the highest overexpression levels: about 64-fold in acute and 28-fold in chronic arthritis, where both groups included more than a dozen genes. Proteolysis-regulating genes (proteases and their inhibitors) were highly represented at the acute phase (45 genes), but were less abundant in chronic arthritis (19 genes). Extracellular matrix-related genes, mostly relevant to tissue repair and healing, were more abundant in chronic than acute disease. Some functional clusters were phase-specific, such as lysosome, antigen presentation, scavenger receptors, immunoglobulin binding, and complement cascade; these genes were preferentially expressed in acute joint inflammation. Suppression of genes related to the respiratory chain complex was specific to chronic inflammation (Fig. 3).

To identify genes whose expression might be specific for the actual phase of arthritis, and to combine transcripts by the pattern of their expression through all disease phases, we applied a hierarchical clustering technique [46]. Genes that were specific for pairwise comparisons (PA/AN, AA/AN, and CA/AN) were combined into one single file (excluding redundant genes); the merged set included 507 genes. Hierarchical clustering was performed for all experimental conditions studied (AN, PA, AA, and CA), and four major gene clusters were identified, each with a distinct expression pattern (Fig. 4, clusters I–IV). Using further classification analysis with Gene Ontology terms, to examine the functions of genes inside each cluster, we identified genes encoding proteins whose biological functions were the most relevant to arthritis development and progression.

Cluster I contained genes with major functions in collagen turnover and tissue repair; the expression of these genes reached a peak in chronically inflamed joints.

Cluster II was the largest cluster including about half of all phase-specific genes (Fig. 4). The cluster included genes with roles in immune, inflammatory and stress responses, extracellular matrix formation, cell growth, and receptor activity. The expression of cluster II genes reached a peak at the acute phase of joint inflammation.

Transcription of genes in clusters III and IV gradually decreased during disease progression (Fig. 4). These genes were mostly related to cytoskeleton remodeling, the formation of cell junctions, and the production of structural molecules such as desmin, β-3 laminin, envoplakin, and dystonin (for a detailed gene list see Additional files 1, 2, 3). Genes associated with early arthritis (Table 2) were found in clusters III and IV, further underlining the importance of cell adhesion and cytoskeleton remodeling during the initiation phase of arthritis.

Hierarchical clustering of a large number of phase-specific genes (n = 507) (Fig. 4) obscured the expression pattern of a relatively small number (n = 37) of early arthritis genes (Table 2). A separate hierarchical clustering was therefore performed for these 37 early genes, and the levels of expression were monitored at later phases of the disease. Six distinct expression patterns were identified (Fig. 5, clusters A–F) using this approach. Clusters A–D contained early arthritis genes whose transcription increased as the disease progressed, reaching a peak in the pre-inflamed joint or during inflammation. Cluster A included genes that coded for variable parts of the T cell receptor, together with genes related to cytoskeleton reorganization such as Rho interacting protein 3, myosin, and β-actin (reviewed in [52, 53]). Cluster A genes were at the peak of their expression in the PA joint. However, most early arthritis genes in clusters C and D showed an expression peak later, at the acute phase of inflammation (Fig. 5), and encoded chemokine receptors (Ccr2 and Ccr5) and chemokine ligands (Cxcl1, Ccl2, Ccl7, and Ccl9). Clusters C and D also included interferon-activating genes Ifi203, Ifi47, and Ifigtp, and cell differentiation antigens such as CD48 and CD53.

Hierarchical clusters E and F contained four genes whose expression was downregulated in the pre-inflamed joint but returned to a 'normal' level (as expressed in naive paws) during arthritis progression. Clusters E and F included genes encoding Terf1-interacting nuclear factor 2, tissue inhibitor of metalloproteinase 1, makorin, and DNA clone 4833424O15 with unknown function (Table 2 and Fig. 5).