Identification of Osteoarthritis Biomarkers by Proteomic Analysis of Synovial Fluid

Abstract

Objective:

To use proteomic analysis to identify novel candidate biomarker proteins in synovial fluid for the differential diagnosis of osteoarthritis and rheumatoid arthritis.

Methods:

Synovial fluid samples were analysed using surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF-MS). Data were used to generate an artificial neural network (ANN). The identification of one protein peak was confirmed via Western blotting.

Results:

Fluid samples were analysed from 36 patients with osteoarthritis and 24 with rheumatoid arthritis. In total, three protein peaks (mass-to-charge ratio [m/z] 3893, 10576 and 14175 Da) were identified as potential biomarkers for osteoarthritis. The ANN differentiated between osteoarthritis and rheumatoid arthritis with a sensitivity of 89.4% and a specificity of 91.2%. The protein peak at m/z 10 576 was identified as S100 calcium binding protein A12 (S100A12).

Conclusions:

A combination of SELDI-TOF-MS and ANN identified osteoarthritis biomarkers. SELDI-TOF-MS may be a useful tool in the screening of synovial fluid for osteoarthritis diagnosis.

Introduction

Osteoarthritis is the most prevalent joint disease and causes pain, stiffness, reduced motion, swelling and disability. ¹ It is characterized by the progressive destruction of articular cartilage with narrowing of the joint space, osteophyte formation, subchondral sclerosis and synovitis. ² The knee is the most common site of primary osteoarthritis involvement. ³ Physiological factors such as ageing and obesity are associated with osteoarthritis, but the disease aetiology and pathogenesis remain unclear.^4,5 Plain X-radiography is the gold-standard tool for estimating the extent of disease but, despite improvements in methodology, lack of precision and poor sensitivity prevent early detection of joint degradation and monitoring of treatment effects. ³ Magnetic resonance imaging (MRI) is a more sensitive imaging method than X-radiography, but is used less frequently due to its higher cost and limited availability. ⁶

Proteomic analysis can provide information regarding changes in protein levels throughout a disease process ⁷ and has contributed substantially to clinical disease research. ⁸ Surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF-MS) can be used for the analysis of biological samples (e.g. urine, serum, plasma or cerebrospinal fluid) without the need for any pretreatment. ⁹ SELDI-TOF-MS has several advantages over other diagnostic methods and proteomic technologies, including its high sensitivity ¹⁰ and the use of an array format that allows the analysis of hundreds of complex or crude samples in a relatively short time. Studies have shown that SELDI-TOF-MS is a noninvasive, sensitive, high-throughput, reproducible tool for the diagnosis of many diseases including colon cancer, ¹¹ AIDS, ¹² coronary artery disease ¹³ and Down's syndrome. ¹⁴

The objective of this study was to apply SELDI-TOF-MS technology to identify proteins from synovial fluid as biomarkers for the diagnosis of osteoarthritis.

Patients and methods

Study Population

The study recruited sequential patients with osteoarthritis or rheumatoid arthritis of the knee who were attending the Department of Osteology, Shandong Provincial Hospital, Shandong University, Jinan, China, between October 2010 and May 2011, for total knee replacement surgery. Osteoarthritis and rheumatoid arthritis were diagnosed according to American College of Rheumatology criteria.^15,16 Patients with diabetes, cancer or chronic inflammatory disease were excluded from the study.

The study was approved by the Institutional Review Board on Human Research of Shandong Provincial Hospital, Shandong University, and was conducted in agreement with the Declaration of Helsinki. Written informed consent was obtained from all patients prior to enrolment.

Sample Collection

Immediately before surgery, synovial fluid (1.5 ml) was aspirated from the affected knee using sterile knee puncture, centrifuged at 2500 g for 10 min at 4°C to remove cells and joint debris, and stored at −80 °C until analysis (1 – 6 months later).

Seldi-TOF-MS

Samples were divided into training and test sets. The intra- and interassay reproducibility of spectra were determined using a normal serum quality control sample (obtained from a healthy volunteer), with three proteins in the 2 – 30 kDa range used to calculate the mass and intensity mean coefficients of variance. All synovial fluid samples were read over the course of a week, in order to limit variability over time.

Synovial fluid was thawed on ice, centrifuged at 2500 g for 5 min at 4 °C, and the supernatant retained. A 10-μl aliquot of supernatant was combined with 90 μl of 5 g/l 3-[(3-Cholamidopropyl)dimethylammonio]-1-propanesulphonate (CHAPS) in 0.01 M phosphate buffered saline (pH 7.4), mixed at 4 °C for 15 min, then added to 100 μl Cibacron^® Blue 3GA (Sigma, St Louis, MO, USA; previously equilibrated with 5 g/l CHAPS three times [5 min each equilibration]) in a 96-well plate and agitated on a platform shaker at 4 °C for 60 min. Samples were removed into centrifuge tubes and centrifuged at 1000 g for 5 min at 4 °C. The supernatant was removed and 50-μl aliquots were mixed with 150 μl 20 mmol/l 2-[4-(2-hydroxyethyl)piperazin-1-yl]ethane -sulphonic acid (HEPES, pH 7.4) and added in duplicate to wells of a bioprocessor (Ciphergen^® Biosystems, Fremont, CA, USA) containing hydrophobic surface (H4) chips previously activated with HEPES. The bioprocessor was then sealed and agitated on a platform shaker for 60 min at 4°C. Excess sample was discarded and the chips were washed three times with HEPES (5 min each wash) and twice with deionized water (1 min each wash). The chips were then removed from the bioprocessor, air-dried and crystallized by the addition of 0.5 μl of 50% α-cyano-4-hydroxycinnamic acid (Ciphergen^® Biosystems).

Chips were read using the Protein Biological System II, equipped with a mass spectrometer (Ciphergen^® Biosystems). The mean of 65 laser shots (intensity 135, detector sensitivity 7, highest mass 50 000 Da, optimized range 2 – 30 kDa) was recorded. Mass accuracy was calibrated to < 0.1% using the all-in-one peptide molecular mass standard (Ciphergen^® Biosystems).

Bioinformatics Analysis

The spectral intensities of all samples were normalized to the total ion current of mass to charge (m/z) between 2000 and 30 000 Da. Noise was filtered from the spectra and peaks were detected with an automatic peak detection pass. Peak clusters were completed using second-pass peak selection (signal-to-noise ratio > 2, within 0.3% mass window), and estimated peaks were added. Biomarker Patterns Software, version 3.1 (Ciphergen^® Biosystems) was used to compute and rank the contribution of each individual peak to the discrimination between groups.

ANN Analysis

An artificial neural network (ANN) was built using data from training group samples, using a back propagation algorithm.^17,18 The peak with the lowest P-value and the highest ability to discriminate between groups was selected to build the integrated ANN, and the discriminating ability was estimated using the test set. Peaks were added in a stepwise manner in order to train the integrated ANN, until all ten peaks with the lowest P-values has been included. In this way, ten models combining different peaks were built, and those peaks with the highest accuracy were selected as potential biomarkers.

The resulting ANN had four layers including one input layer, one output layer and two hidden layers. Each hidden layer contained 20 neurons and there was one output neuron. Prediction output values for patients with osteoarthritis and rheumatoid arthritis were set at 0 and 1 respectively. The power of each peak for between-group discrimination was estimated by Student's t-test and receiver operating characteristic (ROC) curve analysis. The ANN was trained with all samples for each peak, with 1000 repetitions used to generate P-values, with a lower P-value indicating higher between-group discriminatory ability. A 10-fold cross-validation approach was applied to reduce the risk of overtraining.

Western Blotting

Molecular weight and electric charge data for potential biomarker peaks were used to identify proteins via the SWISS-PROT database (http://www.uniprot.org/). ¹⁹ Western blotting of pooled synovial fluid samples was performed as described previously,^20,21 using mouse-antihuman S100A12 primary antibody (Novus Biologicals, Littleton, CO, USA) and rabbit antimouse secondary antibody. Membranes were developed using the Super SignalWest PicoChemiluminescent Substrate kit (Thermo Fisher Scientific, Rockford, IL, USA) followed by AlphaInnotech IS-1000 digital imaging system analysis (Protein Simple, Santa Clara, CA, USA).

Statistical Analyses

Between-group comparison of relative peak intensities was made using Student's t-test. A P-value < 1 × 10^–4 was considered statistically significant. Statistical analyses were performed using SPSS^® software, version 15.0 (SPSS Inc., Chicago, IL, USA) for Windows^®.

Results

The study included synovial fluid samples from 60 patients, 36 of whom were diagnosed with osteoarthritis (27 women and nine men; mean age 70.2 ± 5.4 years; age range 55 – 80 years) and 24 with rheumatoid arthritis (20 women and four men; mean age 68.4 ± 4.9 years; age range 50 – 75 years). The training set (n = 40) comprised samples from 24 patients with osteoarthritis and 16 with rheumatoid arthritis. The test set (n = 20) comprised samples from 12 patients with osteoarthritis and eight with rheumatoid arthritis.

The intra- and interassay coefficients of variance for mass in SELDI-TOF-MS were 0.3% and 0.6% respectively, and those for normalized intensity were 10% and 16%, respectively. Peaks < 2 kDa were mainly ion noise from the matrix and were therefore excluded. ¹⁷ After noise was filtered from the data, 37 protein peaks (2 – 30 kDa) in patients with osteoarthritis were found to differ from those in patients with rheumatoid arthritis. Of these peaks, three were identified as potential biomarkers of osteoarthritis (3893, 10 576 and 14 175 Da). The peak at 10576 Da was present at high levels in osteoarthritis but low levels in rheumatoid arthritis (Fig. 1). Peaks at 3893 and 14 175 Da were present at high levels in rheumatoid arthritis and low levels in osteoarthritis (Fig. 1). Levels of all three proteins were significantly different between osteoarthritis and rheumatoid arthritis (P < 1 × 10^–4; Table 1). The estimated sensitivity, specificity and positive predictive value of the integrated ANN using these three protein peaks are shown in Table 2.

Figure 1:

Surface-enhanced laser desorption/ionization time-of-flight mass spectra of three single biomarker candidate proteins for osteoarthritis. Those shown are at mass-to-charge ratios (m/z) of: (A) 3893 Da (downregulated relative to rheumatoid arthritis); (B) 10 576 Da (upregulated relative to rheumatoid arthritis); (C) 14 175 Da (downregulated relative to rheumatoid arthritis)

Table 1:

Relative intensity of protein peaks identified by surface-enhanced laser desorption/ionization time-of-flight mass spectrometry of synovial fluid from patients with osteoarthritis or rheumatoid arthritis

m/z, Da	Osteoarthritis n = 36	Rheumatoid arthritis n = 24	Statistical significance a	AUC
3893	3.4432 ± 1.5012	5.6736 ± 2.9372	P = 1.78 × 10−6	0.847
10 576	4.6847 ± 2.5381	2.6361 ± 1.3927	P = 1.55 × 10−5	0.832
14 175	0.6197 ± 1.1621	1.8998 ± 1.2031	P = 2.04 × 10−5	0.831
6510	1.5581 ± 0.9778	3.5981 ± 2.1727	P = 2.85 × 10−5	0.829
9368	6.8928 ± 4.4772	3.0321 ± 3.2719	P = 3.22 × 10−5	0.820
17 689	0.3002 ± 0.1798	0.4567 ± 0.1945	P = 1.22 × 10−4	0.819
5811	1.9378 ± 0.8011	2.6351 ± 0.9765	P = 1.40 × 10−4	0.819
4611	0.6136 ± 0.4078	0.9084 ± 0.4378	P = 1.77 × 10−4	0.810
9143	0.2428 ± 0.1277	0.3407 ± 0.1436	P = 4.98 × 10−4	0.809
3325	0.0998 ± 0.0819	0.1619 ± 0.0958	P = 8.92 × 10−4	0.807

Data presented as mean ± SD.

a

Student's t-test.

m/z, mass-to-charge ratio; AUC, area under receiver operating characteristic curve.

Table 2:

Estimated sensitivity, specificity and positive predictive value of an integrated artificial neural network (ANN) using surface-enhanced laser desorption/ionization time-of-flight mass spectrometry for the differential diagnosis of osteoarthritis from rheumatoid arthritis. The ANN was generated using 10-fold cross validation with previously identified mass spectrometry peaks (mass-to-charge ratio 3893, 10576 and 14175)

Parameter	Training set n = 40	Test set n = 20
Sensitivity	91.0	89.4
Specificity	92.1	91.2
Positive predictive value	91.4	90.0

Data presented as %.

The molecular weight and electric charge of the peak at 10 576 Da were consistent with S100 calcium binding protein A12 (S100A12). Western blotting analysis confirmed that this protein was present at higher levels in patients with osteoarthritis than in those with rheumatoid arthritis (Fig. 2). The remaining two protein peaks were not successfully identified.

Figure 2:

Western blotting analysis for S100 calcium binding protein A12 (S100A12) in synovial fluid from patients with osteoarthritis (A, B) or rheumatoid arthritis (C, D)

Discussion

Diagnosis of osteoarthritis is often delayed because of the poor sensitivity of radiological tests. ⁶ Synovial fluid is in direct contact with the synovium and cartilage, and may therefore be an excellent source of biomarkers. Proteomic analysis of synovial fluid could reflect the pathological state of osteoarthritis and enable the development of more reliable and cost-effective diagnostic tools than those now available. Currently, proteomic techniques mainly rely on dielectrophoresis and mass spectrometry. Dielectrophoresis is well established, but is limited by complicated methodology and the requirement for large samples. ²² Mass spectrometry has higher resolution and sensitivity than dielectrophoresis, but samples must be purified and the equipment to perform such analyses expensive.^23,24 The use of proteomics is also limited by difficulties in processing and analysing large amounts of data. SELDI-TOF-MS allows the rapid analysis of multiple directly obtained samples to generate protein expression profiles. ²⁵

Artificial neural networks combine neurology, computer science, informatics and engineering. This combination results in substantial information storage, good fault tolerance, massive parallelism and powerful procedures for self-management, self-learning and self-adaptation to circumstances. ANNs have been applied to disease prediction research.^26,27 An integrated SELDI-TOF-MS and ANN approach was used in the present study to analyse synovial fluid samples from patients with osteoarthritis or rheumatoid arthritis. This approach identified three potential biomarkers (3893, 10576 and 14 175 Da) to distinguish between osteoarthritis and rheumatoid arthritis, possibly representing a new diagnostic approach for these diseases.

The protein peak at 10576 Da was identified as S100A12. Both osteoarthritis and rheumatoid arthritis result in the destruction of articular cartilage but have different aetiologies and pathogeneses. Osteoarthritis is characterized by alterations in metabolism and cell signalling, and an imbalance of redox mechanisms leading to degradation of cartilage. ²⁸ Rheumatoid arthritis is an autoimmune disease that correlates with inflammation that is responsible for cartilage destruction. ²⁹ S100 proteins have been shown to have important extracellular proinflammatory effects and cytokine-like activities in addition to their intracellular functions. ³⁰ Studies have found S100A12 to be present in the synovial fluid of patients with rheumatoid arthritis, but at lower levels than SA100A8 and SA100A9. ³¹ In contrast, S100A12 has been shown to be upregulated in human osteoarthritis cartilage. ³² Together with findings from the present study, these data suggest that S100A12 may be a key biomarker for osteoarthritis, with diagnostic and/or prognostic value.

The protein peaks at 3893 and 14 175 Da were present at high levels in rheumatoid arthritis in the current study, and may therefore be potential biomarkers for rheumatoid arthritis. The current study was unable to identify these proteins, however.

Several measures were taken in the present study to avoid possible bias in the data. Care was taken to control sources of variation in the methods (including time, temperature, humidity and equipment used). Samples were aliquotted and frozen immediately after collection, and thawed only once. Standard protocols were developed to minimize unwanted fluctuation in SELDI-TOF-MS, and coefficients of variation were calculated by using common peaks across different spectra. All samples were analysed in duplicate, and only peaks that exhibited a reproducibly high ranking both times were used for further analyses. In this way, systematic bias in the selection of biomarkers was minimized.

In conclusion, the current study indicates the feasibility of using biomarkers for the diagnosis of osteoarthritis. SELDI-TOF-MS may be a useful tool in the screening of synovial fluid for osteoarthritis biomarkers. Further research, with larger study cohorts, is required to confirm these findings.