Serum protein profile in systemic-onset juvenile idiopathic arthritis differentiates response versus nonresponse to therapy

Banked sera from 23 patients (14 boys, 9 girls) with SJIA according to the criteria established by the International League of Association for Rheumatology [13] were obtained from the Department of Pediatrics, Yokohama City University School of Medicine, Yokohama, Japan. All the patients were Asian and their mean age at the start of the study was 7.25 ± 0.92 years. Eight of them had obtained a clinical response to conventional therapy. Clinical response was defined as the absence of fever rash, hepatosplenomegaly, and arthritis for at least 3 months, accompanied by normalization of serum C-reactive protein. Briefly, conventional therapy consisted of three doses of intravenous methylprednisolone (30 mg/kg per day) or oral prednisolone (1 to 2 mg/kg), followed by nonsteroidal anti-inflammatory drugs (NSAIDs) and a tapering dose of oral prednisolone. In addition, methotrexate (2.5 to 5 mg/m² per week orally) was used in three patients and cyclosporin (5 mg/kg per day orally) in two. The mean period from acute status to complete clinical response was 27.7 ± 14.6 months. Fifteen patients who had inadequate response to the above therapy, as well as to the addition of azathioprine (five patients), mizoribine (five patients), sulfasalazine (two patients), or plasma exchange (three patients), had been administered humanized anti-IL-6 receptor antibody (MRA; Chugai Pharmaceuticals, a subsidiary of Roche Pharmaceuticals). All 15 patients had a clinical response to MRA. The mean period from acute status to clinical response was 11.2 ± 5.1 months. Pretreatment sera were collected before starting conventional treatment or giving the initial dose of MRA. Post-treatment sera were collected 2 to 3 months after patients achieved a clinical response. Ethical approval for this study was granted by Yokohama University. The study was approved by both Chugai Pharmaceuticals and Roche Pharmaceuticals.

Serum samples were thawed on ice, denatured, and processed in duplicate on IMAC-3 (immobilized metal affinity capture) copper ProteinChip^® Arrays (Ciphergen Biosystems, Fremont, CA, USA). ProteinChips were loaded, processed, and prepared for mass spectrometry using a Biomek2000 liquid handling robot (Beckman-Coulter, Fullerton, CA, USA) and optimized for reproducibility using validated protocols. ProteinChips were read in a PBSIIc mass spectrometer (Ciphergen) with mass deflection at 1 kDa and time-lag focusing. The resulting mass spectra were examined between m/z values of 2 and 100 kDa for quantitative comparison of identifiable peak features. The parameters used for spectral preprocessing and peak selection were: external calibration (seven peptide calibrants, 1 to 7 kDa, Ciphergen), baseline subtraction by 8 × expected peak width and smoothing, filtering by average using 0.2 expected peak width, noise defined over 1500 Da, normalization by total ion current (TIC) over 1500 Da, peaks detected over 2000 kDa by centroid mass. Weak spectra were excluded from analysis if the normalization factor exceeded 2 standard deviations above the mean normalization factor.

Peak clustering among sample groups was performed with the Biomarker Wizard (Ciphergen) tool, with a peak detection threshold of 5 for signal-to-noise ratio, and mass tolerance of 0.3%, for any peak appearing in at least 5% of experimental spectra being compared. The Biomarker Wizard compares the mean intensity of peak clusters, by sample group, using the nonparametric Mann–Whitney U test (two–way comparisons) and generates P values that reflect probabilities that mean peak intensities at a given m/z value differ by random chance. The intensity values of the automatically clustered peaks (averaged between technical replicates of each sample) were used in classification tree analysis (CART) using Ciphergen's Biomarker Patterns Software. This supervised learning process uses cross-validation to optimize the minimization of classification error.

A pooled sample from four sera taken before conventional treatment was incubated with either Protein A–Sepharose beads alone or Protein A–Sepharose beads bound with 100 μg of anti-SAA antibody (Anogen, Yes Biotech Laboratories, Mississauga, ON, Canada). After immunoprecipitation, the depleted serum was subjected to SELDI-TOF MS.

Protein identification by MS was carried out as previously described [14]. Briefly, serum samples were subjected to immunoprecipitation with anti- SAA or with an IgG isotype control. Immunoprecipitates were washed extensively in phosphate-buffered serum and centrifuged, and the pellet was sonicated for 10 min in 8 M urea/400 mM NH₄HCO₃. The supernatant was diluted in water to a final concentration of 2 M urea/100 mM NH₄HCO₃ and digested overnight at 37°C with 1 μg trypsin. The tryptic digest was subjected to nano-LC-ESI-MS/MS analysis that was performed on a Q-TOF-2™ (Waters, Milford, MA, USA), coupled on line to a CapLC system equipped with three separate syringe pump modules, an auto injector, a 10-port valve and a 250-μm (inner diameter) × 1-mm pre-column. Separations were performed on a 7-cm × 75-μm (inner diameter) capillary column. Both columns were packed with Microsorb C18 (Varian, Mississauga, ON, Canada) reverse-phase material. Peptides were eluted at a flow rate of 0.25 μl/min with the following linear gradient of solvent B (80% aqueous acetonitrile with 10% isopropanol and 0.2% formic acid) in solvent A: from 0 to 60% B in 40 min, to 90% B in 7 min, and to 10% B in 8 min. Spectra were acquired in auto MS/MS mode conducted using survey scans to choose up to three precursor ions. Collision energies were selected automatically as a function of m/z value and charge state. The Q-TOF mass spectrometer was calibrated by infusing a solution of either NaI containing a small amount of cesium ion dissolved in 50% aqueous isopropanol (0.2 μg/μl) or Glu-fibrinopeptide B (1 pmol/μl) dissolved in 30% aqueous acetonitrile containing 0.2% formic acid. Protein identification was performed using the MASCOT search program (Matrix Science Limited, http://​www.​matrixscience.​com) and the NCBI (National Center for Biotechnology Information) (Bethesda, MD, USA) protein database.

To determine whether SELDI-TOF MS could be a valuable tool for analyzing serum protein profiles in autoimmunity, we chose SJIA as a test model, since the disease has systemic features, in addition to arthritis, likely to be reflected in the serum. Paired sera were available from patients who had been followed up for a mean of 24.4 ± 6.9 months. The availability of paired sera from each patient allowed for longitudinal comparison and substantially reduced sample variability between the two groups. Sera from eight patients with SJIA who responded to conventional therapy (therapy and definition of clinical response are described in detail in the Materials and methods section) were analyzed before and after therapy by SELDI-TOF MS using Ciphergen IMAC-3 (immobilized metal affinity capture) copper chips (Fig. 1). All samples were run in duplicate. Table 1 shows the most significant differences between mass spectra of sera before and after conventional therapy. Only variables with nonparametric P values of <0.001 are given. Of 282 spectral peaks identified, 23 had mean signal intensities that were significantly different (P < 0.001) before and after response to treatment. These peaks represent potential biomarkers of active disease. We next performed a similar analysis on paired sera from 15 patients who had failed conventional therapy but responded to an experimental antibody to the IL-6 receptor (MRA) [15]. These sera were obtained after failure of conventional therapy. Pre- and post-MRA sera revealed similar profiles to those observed in the pre- and post-conventional therapy group. Thus, substantial consistency was observed in protein profiles, regardless of whether patients with active disease responded to conventional therapy or to MRA. Eight of the differentially expressed peaks represent prominent, visually distinct spectral features. These peaks are represented in Table 1 in bold, along with the number of paired patient sera in which each peak was differentially expressed by visual inspection of the spectra. Representative examples of these peaks are shown in Fig. 2.

To determine the usefulness of the profiles in classifying active versus controlled SJIA, the data were subjected to CART (Biomarker Patterns Software, Ciphergen) analysis. This sample classification method is designed, through multivariate analysis, to construct classification trees recognizing a complex pattern of multiple peak intensities. The method is ideally suited for sample sets large enough to permit cross-validation internal to the 'training' data, but also the segregation of additional unused data as a validation or 'testing' set. On these relatively small sample sets, CART was used in training mode primarily as a data exploration tool. Whether using the training set as the MRA group, or as the conventional treatment group, the CART analysis returned simple classification trees consisting of one primary splitter, either 11.4 kDa or 11.6 kDa (m/z). The primary splitter at 11493 kDa correctly identified 13 of 14 pretreatment and 14 of 15 post-MRA treatment samples when conventional treatment was used as the training set. When MRA treatment was used as the training set, all of the pre- and post-conventional treatment samples were correctly identified as either pretreatment or post-treatment. The distinction between these samples by CART registered at the most extreme level of significance the program is capable of indicating. Even when forced to ignore the mass spectrum peaks at 11493 or 11650 Da, the CART program was able to effectively discriminate, using secondary peaks derived from them (at half these m/z values; attributed to doubly protonated species). This robust classification surpasses the performance of any other sample set being profiled and analyzed by this and several other statistical methods at this institution (data not shown).

A prominent group of peaks within the range 11.4 to 11.7 kDa m/z strongly distinguished the pre- and post-treatment samples (Fig. 3). The post-treatment groups showed an apparently single m/z peak at 11.75 kDa, which was also routinely observed in pooled reference sera from healthy adults (data not shown). A previous study in nasopharyngeal cancer, using the same SELDI-TOF technique and IMAC3 copper chip, identified two biomarkers, of 11.6 and 11.8 kDa, as serum amyloid A (SAA) [16]. Since SAA is a known biomarker of active SJIA [17], we compared the intensity of the 11.6-kDa peak with SAA levels in the sera, as determined by latex agglutination. As shown in Fig. 4, a strong correlation was observed (R² = 0.74), suggesting that the 11.6-kDa peak might represent SAA. To further investigate the biochemical identity of this peak, serum containing high levels of the 11.6-kDa peak was subjected to immunoprecipitation using anti-SSA antibody bound to Protein A–Sepharose beads. As shown in Fig. 5, after immunoprecipitation and SELDI analysis, the 11.4- and 11.6-kDa peaks were markedly diminished. To confirm the identity of the immunoprecipitated protein, anti-SAA precipitates were digested with trypsin and subjected to η-scale liquid chromatography electrospray ionization tandem mass spectrometry time-of-flight (LC-ESI-MC/MS-TOF) analysis. Over 90 peptide ions were examined and only 2 proteins were identified, including immunoglobulin and SAA. The SAA peptide ions represented approximately 51% of the SAA sequence.

The above data, using paired sera, demonstrate the ability of SELDI-TOF MS to identify biomarkers of active disease, as exemplified by the identification of SAA. A long-term goal is to predict clinical outcome, based on protein profiles present in the serum early in the disease course. To begin to approach this challenge, we compared the pretreatment serum profiles of the 8 patients who responded to conventional therapy with those from the 15 patients who responded poorly to conventional therapy. Similar to the preceding analysis, the latter samples were obtained after failure of conventional therapy and before MRA treatment, when the patients still had active disease. In this initial exploratory study, the number of available samples was too small for definitive conclusions; however, several interesting trends were apparent. Several highly significant differences were observed in the mass spectra of these sera, as shown in Table 2 and Fig. 6. These peaks may represent a profile predictive of response to conventional therapy. Alternatively, they could represent the effects of conventional therapy or differences between early versus long-standing disease. A number of peaks overlap with regions observed in Table 1, including the region of SAA (11.6 kDa) as well as 4504 kDa and 28 kDa.

We were fortunate to have pre-conventional treatment sera available from 3 of the 15 patients who responded poorly to conventional therapy and went on to receive MRA. These three pretreatment sera were compared with the pretreatment sera of the eight patients who responded to conventional therapy. Three peaks of interest were observed. As shown in Fig. 7, all the nonresponders had lower values for the 4825-Da feature and higher values for the 3276-Da and 3293-Da peaks than did the responders, with the exception of a single outlier sample. However the sample size is small and this observation needs further validation in a larger clinical cases series; this putative signature of nonresponse may be susceptible to statistical overfitting, even at this level of analysis.