Blood autoantibody and cytokine profiles predict response to anti-tumor necrosis factor therapy in rheumatoid arthritis

Pretreatment sera from three cohorts of patients with the diagnosis of RA based on the ACR classification criteria [15], who were initiated on therapy with the anti-TNF therapy etanercept (Enbrel^®, Amgen, Thousand Oaks, CA, USA), were analyzed using synovial antigen microarrays (except for the third cohort), ELISAs, and a multiplex 12-cytokine bead assay. The cytokines assayed were selected based on previous screening studies using a 22-cytokine assay [12].

The three cohorts included 29 Caucasian patients from the ABCoN cohort of the North American Rheumatoid Arthritis Consortium (US cohort), 43 Caucasian patients seen at Swedish tertiary care centers and collected through the Karolinska-lead EiRA initiative (Swedish cohort), and 21 Japanese patients (Japanese cohort). The patients' demographic, clinical and serologic characteristics are summarized in Table 1 and Additional data file 1. All patients signed informed consent and all sera were collected under and in accordance with Institutional Review Board-approved protocols at each institution.

Blood samples were obtained at baseline (pretreatment sample) and at least 3 months after initiation of therapy with etanercept. Analysis of the pretreatment samples was performed for the present study. Response to therapy with etanercept was assessed at least 3 months after etanercept was started, based on the ACR criteria for improvement [2]. All samples were immediately aliquoted upon receipt at the Stanford research laboratory, and separate aliquots were used for each assay to minimize the effects of additional freeze- thaw cycles.

All cytokine measurements were performed using the Luminex ×200 platform, following a previously described optimized assay protocol [12]. Briefly, to minimize potential false-positive elevations of cytokine measurements due to rheumatoid factor and other heterophilic antibodies that can cross-link the capture and detection antibodies, HeteroBlock^® was added to achieve a final concentration of 3 μg/ml, as previously described in detail [12]. For the studies performed herein, we utilized a custom 12-plex human cytokine FLEX^® kit (Upstate, Millipore, Billerica, MA, USA) that included beads specific for TNFα, IL-1α, IL-1β, IL-6, IL12p40, IL-12p70, IL-15, granulocyte- macrophage colony-stimulating factor, fibroblast growth factor-2 (FGF-2), monocyte chemoattractant protein-1 (MCP-1), eotaxin, and IFNγ-inducible protein 10.

The production of RA antigen microarrays was previously described in detail [13, 14]. Briefly, more than 500 peptides and proteins representing candidate autoantigens in RA were printed at 0.2 μg/μl onto derivatized Epoxy ArrayIt^® microscope slides (ArrayIt^®; TeleChem International Inc., Sunnyvale, CA, USA) using a robotic arrayer. The arrays were blocked and probed with sera at 1:200 dilutions, and bound serum autoantibodies detected using Cy3-conjugated goat anti-human IgG secondary antibodies (Jackson ImmunoResearch Laboratories, Inc. West Grove, PA, USA). Probed arrays were scanned with a GenePix 4000B scanner (MDS Analytical Technologies, Sunnyvale, CA, USA), and antibody reactivities quantified using GenePix Pro 5.0 software.

Peptides cfc(48–65)cit1 (shown in Figure 1) and peptides cfc(48–65)cit2 (listed in Table 2) are identical except for a difference in the degree of citrullination: cfc(48–65)cit1 has one arginine residue citrullinated, and cfc(48–65)cit2 has two arginine residues citrullinated.

Peptides and fibrinogen protein from human plasma (Calbiochem, Gibbstown, NJ, USA) were coated onto medium-binding 96-well flat-bottom polystyrene plates (Costar, Corning Inc, Corning, NY, USA) at 1 μg peptide/ml at 4°C overnight. Plates were then washed and blocked for 1 hour at room temperature using 5% dry milk in PBS, and incubated for 90 minutes using serum at 1:200 dilution in PBS. Horseradish peroxidase-conjugated anti-human IgG secondary antibodies (horseradish peroxidase-conjugated goat anti-human IgG, Fcγ-fragment specific; Jackson Immuno Research Laboratories Inc., West Grove, PA, USA) were used at 1:20,000 dilutions, and bound autoantibodies were detected by chemoluminescence (Onestep^® TMB ELISA; Pierce, Rockford, IL, USA).

The data analysis was performed using significance analysis of microarrays (SAM) (version 1.21) and prediction analysis of microarrays (PAM) (version 1.23), and the hierarchical clustering software Cluster^® and TreeView^® [16], as described previously [14, 17]. PAM uses internal cross-validation by which 90% of the training samples are randomly selected 10 times, followed by one-by-one class prediction of the remaining 10% of samples, thus identifying classification errors and overfitting [18]. PAM was used in the training, cross-validation and prediction analyses described. The general-purpose statistical package R has also been used for the analysis [19].

To further develop previously-described RA autoantigen microarrays [14], we expanded the number of peptide and protein autoantigens on the arrays. The arrays used in the experiments described herein were 2,180-feature arrays that include > 540 proteins and peptides, representing the following candidate RA antigens: biglycan; decorin; fibromodulin; clusterin; osteoglycin; fibrinogen; type I, type II, and type V collagens; vimentin; filaggrin; serine protease 11; apolipoprotein E; calpastatin; glucose-6-phosphate isomerase; heat shock proteins HSP60, HSP70, HSP90, and BiP; hnRNP-A2/B1; histones H2A and H2B; and cartilage oligomeric matrix protein (COMP), vitronectin and fibronectin. Candidate antigens were selected based on literature searches and screening experiments. Briefly, for the antigen screening experiments, immune complexes were isolated from RA cartilage and synovial tissues, and the protein antigens contained in the synovial immune complexes were identified by mass spectrometry (PM, manuscript in preparation). Identified proteins were, when available, purchased from commercial sources and peptides representing the candidate antigens synthesized for printing on arrays. Overlapping 20-amino acid peptides, both native and containing citrulline substitutions, were synthesized on a commercial custom peptide synthesis platform using Fmoc chemistry (PepScreen^®; Sigma Genosys, St Louis, MO, USA).

In line with and expanding earlier results using smaller 225-antigen arrays, we observed on the > 540-antigen arrays specific and differential serum autoantibody reactivities to COMP, clusterin, osteoglycin, apolipoprotein E, histones H2A and H2B, serine protease 11, and other candidate antigens. Representative array images of antibody reactivities are shown for sera from two different patients with RA; one patient that exhibited low serum autoantibody reactivity (Figure 2a) and another that exhibited high serum autoantibody reactivity (Figure 2b). A selection of antigen targets that were differentially recognized by serum antibodies in patients RA1 and RA2 are highlighted in the figures in colored boxes (Figure 2a, b). Selected array reactivities include citrullinated peptides (cfc1, reactive with anti-cyclic citrullinated peptide antibody-positive RA serum) and native control peptides (cfc0, no reactivity with RA sera). Features were normalized to IgG/M, and quantification of the highlighted features is summarized in Figure 2c.

To determine the correlation of antigen array and ELISA results from a subset of peptide antigens, several native and citrulline-substituted peptides were tested in ELISA experiments. We observed moderate to strong correlations for these peptide antigens, with correlation coefficients R² ranging from 0.49 for clusterin(386–405)cit to > 0.92 for hFibA(211–230)cit (see Additional data file 2).

In the next series of experiments, we utilized pretreatment samples from three independent cohorts of etanercept new-start RA patients. These cohorts included the US-based ABCoN cohort, a Swedish cohort, and a Japanese cohort.