Spontaneous OA in Dunkin-Hartley guinea pig
Sixty male 3-week-old Dunkin-Hartley guinea pigs, purchased from Charles River Laboratories (Paris, France) with identification by microchip, were used in the study. They were bred under pathogen-free conditions with free access to water. In experimental studies, they were housed three per solid-bottom cage and fed with a standard guinea pig chow (Special Diets Service, Essex, England) containing vitamin C (394 mg/kg) and vitamin D3 (1973 IU/kg), allowing 2 weeks for acclimatization. Polyvinyl chloride pipes were added to the cages to improve housing conditions and minimize stress. Separate groups of 12 animals were sacrificed and analyzed at age 4, 12, 20, 28 and 36 weeks. There was no repeated analysis of animals in the study. The number of animals per group was chosen according to the OARSI recommendation [11]. Animal body weight and food consumption were recorded weekly. Blood samples were collected by intracardiac puncture under general anesthesia (sodium pentobarbital 200 mg/kg intraperitoneally) immediately before animals were killed. Blood samples were centrifuged (2000 × g, 5 minutes), and serum was stored at − 80 °C until analysis. Samples were centrifuged within 1 hour of collection. All experimental procedures and protocols were reviewed and approved by the Institutional Animal Care and Use Ethics Committee of the University of Liège (Belgium) (reference 1648).
Histology
At the time animals were killed, cartilage samples were processed for histological evaluation. The right knee joint (femoral condyles and tibial plateaus) from each animal was fixed for 24 hours in 4% paraformaldehyde, followed by decalcification in hydrochloric acid (DC2 medium; Labonord, Templemars, France) for 4 hours at 4 °C before embedding in paraffin. The right kidney and a piece of the liver were fixed in 4% paraformaldehyde and embedded in paraffin.
Sections (6 μm) of the femoral condyles and tibial plateaus were cut with a microtome in the central area not covered by meniscus following the Cushin plane, as recommended by OARSI [11]. Three sections at 200-μm intervals were stained with hematoxylin, Fast Green, and Safranin-O, and one supplementary central section was stained with toluidine blue. Each compartment of the section (tibial median, tibial lateral, femoral median, and femoral lateral) was scored by two trained experts blinded from sample identity following OARSI recommendations for the guinea pig model. Briefly, the evaluation considered the cartilage surface integrity (0–8), the proteoglycan content (0–6), the cellularity (0–3), the tidemark integrity (0–1), and the osteophyte (0–3), with a maximum of 21 per compartment. The mean score of three sections was calculated for each knee compartment. To assess the global OA score, scores of each compartment were added, giving a maximal score of 84. Lateral and medial synovial membranes were also scored (synovial lining cells hyperplasia 0–2, villous hyperplasia 0–3, degree of cellular infiltration by perivascular lymphocytes and mononuclear cells 0–5), and the mean of lateral and median membrane was calculated to assess the global synovial score (maximum score of 10) [11].
Biomechanical testing by Mach-1® micromechanical tester
The left knee joint (femoral condyles and tibial plateaus) of each animal was used for testing the biomechanical properties of articular cartilage assessed using a Mach-1® micromechanical tester (Mach-1; Biomomentum Inc., Laval, QC, Canada) [12]. Prior to testing, samples were thawed at room temperature in PBS for 30 minutes to equilibrate before starting experiments. Subsequently, the femoral condyle or tibial plateau was fixed with LOCTITE® 4013 glue (Henkel, Stamford, CT, USA) in a small plastic container (Additional file 1: Figure S2). Throughout the testing, each sample was kept moist with PBS. Using top-view pictures of each sample, at least 50 positions per articular surface were tested using the automated indentation and thickness-mapping protocol. The instantaneous modulus—a measure of cartilage stiffness and cartilage thickness—was calculated using the Mach-1 analysis software (see Additional file 1).
Primary culture of human chondrocytes
Human chondrocytes were cultured in multilayers in six-well plates and treated with interleukin-1β (IL-1β) [13]. Chondrocytes were isolated from human articular cartilage taken during the installation of total knee prosthesis. Cartilage samples were obtained from four adults (two men and two women) whose mean age was 70 years (range, 51–81 years). All specimens used were obtained with informed consent. This procedure was approved by the Ethics Committee of the Catholic University of Louvain (project no. B403201214793). Full-depth articular cartilage was excised and immersed in DMEM (with phenol red and 4.5 g/L glucose) supplemented with 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES) 10 mM, penicillin 100 U/ml, and streptomycin 0.1 mg/ml (all from Lonza, Verviers, Belgium). After three washings, chondrocytes were released from cartilage by sequential enzymatic digestions with 0.5 mg/ml hyaluronidase type IV S (Sigma-Aldrich, Bornem, Belgium) for 30 minutes at 37 °C, 1 mg/ml pronase E (Merck, Leuven, Belgium) for 1 hour at 37 °C, and 0.5 mg/ml collagenase from Clostridium histolyticum type IA (Sigma-Aldrich) for 16 to 20 hours at 37 °C. The enzymatically isolated cells were then filtered through a nylon mesh (70 μm), washed three times, counted, and filled to the density of 0.25 × 106 cells/ml of DMEM (with phenol red and 4.5 g/L glucose) supplemented with 10% FBS, 10 mM HEPES, 100 U/ml penicillin, 0.1 mg/ml streptomycin, 2 mM glutamine (all from Lonza), and 20 μg/ml proline (Sigma-Aldrich). After 21 days of culture, chondrocytes were treated in triplicate with recombinant human IL-1β (1.7 ng/ml; Roche Pharmaceuticals, Brussels, Belgium). The seeding density of the chondrocytes in the six-well plates was 50,000 cells/cm2. There was no passage of the cells; the cells overlap and form an extracellular matrix. Culture medium and IL-1β treatment were replaced at 3 and 6 days, and conditioned medium was removed at 3, 6, and 10 days and stored at − 20 °C until analysis.
Patients, healthy subjects, and sampling
Patient recruitment, characteristics, and sampling were similar to those previously described [14]. Briefly, patients with early-stage OA (eOA) (n = 28), early-stage rheumatoid arthritis (eRA) (n = 35), and inflammatory joint disease other than rheumatoid arthritis (often self-resolving) (non-RA) (n = 32) were recruited. Criteria for eOA were subjects presenting with new-onset knee pain, normal radiographs of the symptomatic knee, and routine exploratory arthroscopy with macroscopic findings classified as grade I/II on the Outerbridge scale, and recruited at the Orthopaedic Clinics, University Hospital Coventry & Warwickshire (UHCW), Coventry, UK. Patients with eRA and non-RA were recruited within 5 months of the onset of symptoms of inflammatory arthritis at the Rapid Access Rheumatology Clinic, City Hospital, Birmingham, UK. Synovial fluid and peripheral venous blood samples were collected at initial presentation, and diagnostic outcomes were determined at follow-up. Diagnosis of eRA was made according to the 1987 American Rheumatism Association criteria [15]. Diagnosis of non-RA was made when alternative rheumatological diagnoses explained the inflammatory arthritis [16]. Criteria for these clinical classifications are similar to those suggested in consensus position statements and best practice statements [16, 17]. Healthy controls were recruited at participating clinical centers (n = 29) at UHCW. For healthy control subjects, inclusion criteria were no history of joint symptoms, arthritic disease, or other morbidity, and exclusion criteria were a history of injury or pain in either knee, taking medication (excepting oral contraceptives and vitamins), and any abnormality at physical examination of the knee.
Recruitment of patients with advanced OA (n = 38) immediately prior to total knee replacement (TKR) surgery (advanced osteoarthritis [aOA], pre-TKR) was done with written informed consent from patients referred for TKR to the Norfolk and Norwich University Hospitals NHS Trust (NNUH), Norwich, UK. Patients were screened for study eligibility criteria as described previously [14]. Eligible patients were males or postmenopausal females scheduled for TKR. This study was approved by the National Research Ethics Service Committee East of England, Cambridge South, UK (approval no. 2012ORTH06L [104-07-12]). All study procedures were performed in accordance with relevant laboratory guidelines and institutional regulations.
Peripheral venous blood samples were collected with ethylenediaminetetraacetic acid (EDTA) anticoagulant from healthy subjects and patients with eOA after overnight fasting. Venous blood samples for the eRA, non-RA, and aOA study groups were collected in the nonfasted state. For analytes studied, diurnal variation in plasma and serum was 13–25%, depending on the analyte, as described previously. Blood samples were centrifuged (2000 × g, 10 minutes), and the plasma and synovial fluid supernatant was removed and stored at − 80 °C until analysis. Samples were centrifuged within 1 hour of collection. Serum was available for eRA and non-RA study groups, and plasma was used for all others. Serum was comparable to plasma because nonprotein analytes were assessed. To confirm this, venous blood samples were collected with informed consent from human volunteers (n = 6; 4 female, 2 male; age 47.8 ± 15.8 years; BMI 25.9 ± 4.0 kg/m2). Ethical approval was given by East Midlands Regional Ethics Committee (reference 16/EM/0095). Serum and plasma (with EDTA anticoagulant) was prepared and assayed for the concentrations of glycated, oxidized, and nitrated amino acids as described below. There was no significant difference between analyte levels in serum and plasma by Wilcoxon signed-rank test.
Analysis of glycated, oxidized, and nitrated protein and amino acids in serum/plasma
Glycation, oxidation, and nitration adduct residues and related precursor unmodified amino acid residues in plasma/serum protein were quantified in exhaustive enzymatic digests, with correction for autohydrolysis of hydrolytic enzymes [18, 19]. The concentrations of glycated, oxidized, and nitrated amino acids (free adducts) and hydroxyproline in plasma/serum were determined similarly in 10 kDa ultrafiltrate of plasma/serum and cell culture medium. Ultrafiltrate of plasma/serum (50 μl) was collected by microspin ultrafiltration (10 kDa cutoff) at 4 °C. Retained protein was diluted with water to 500 μl and washed in four cycles of concentration to 50 μl and dilution to 500 μl with water over the microspin ultrafilter at 4 °C. The final washed protein (100 μl) was delipidated and hydrolyzed enzymatically as described previously [19, 20]. Protein hydrolysate (25 μl, 32 μg equivalent) or ultrafiltrate (5 μl) was mixed with stable isotopic standard analytes (amounts as given previously) and analyzed by LC-MS/MS. Samples were analyzed using an ACQUITY™ ultra-high-performance liquid chromatography system with a Xevo-TQS LC-MS/MS mass spectrometer (Waters, Manchester, UK). Samples are maintained at 4 °C in the autosampler during batch analysis. The columns were 2.1 × 50-mm and 2.1 × 250-mm, 5-μm particle size Hypercarb™ (Thermo Fisher Scientific, Waltham, MA, USA) in series with programmed switching at 30 °C. Chromatographic retention was necessary to resolve oxidized analytes from their amino acid precursors to avoid interference from partial oxidation of the latter in the electrospray ionization source of the mass spectrometric detector. Analytes were detected by electrospray positive ionization and mass spectrometry multiple reaction monitoring (MRM) mode, where analyte detection response was specific for mass/charge ratio of the analyte molecular ion and major fragment ion generated by collision-induced dissociation in the mass spectrometer collision cell. The ionization source and desolvation gas temperatures were 120 °C and 350 °C, respectively; cone gas and desolvation gas flow rates were 99 and 900 L/h; and the capillary voltage was 0.60 kV. Argon gas (0.5 Pa) was in the collision cell. For MRM detection, molecular ion and fragment ion masses and collision energies optimized to ± 0.1 Da and ± 1 eV, respectively, were programmed [19]. In all sample analyses, the investigator was blinded from the sample identity. Analytes determined were as follows: glycation adducts Nε-fructosyl-lysine (FL), Nε-carboxymethyl-lysine (CML), Nε-carboxyethyl-lysine (CEL), Nω-carboxymethylarginine (CMA), glyoxal-derived hydroimidazolone (G-H1), methylglyoxal-derived hydroimidazolone (MG-H1), 3-deoxyglucosone-derived hydroimidazolone isomers (3DG-H), GSP, and pentosidine; oxidation adducts dityrosine (DT), N-formylkynurenine (NFK), α-aminoadipic semialdehyde (AASA), and glutamic semialdehyde (GSA); nitration adduct 3-nitrotyrosine (3-NT); and related amino acids [19] (see Fig. 1 for structures and expansion of acronyms). The biochemical and clinical significance is described elsewhere [6]. Protein adduct residues (normalized to their amino acid residue precursors; mmol/mol amino acid modified) and serum or plasma free adduct concentrations (μM or nM) are given. In culture medium, free adduct concentrations were corrected for cell number by normalizing to cellular DNA content.
Citrullinated protein and hydroxyproline
Serum citrullinated protein (CP) and Hyp were analyzed by stable isotopic dilution analysis LC-MS/MS, as previously described [20].
Machine learning
We developed algorithms using the clinical analyte data to distinguish the following four groups of subjects and patients: healthy control, eOA, eRA, and non-RA. The diagnostic algorithms were trained on the dataset using support vector machines [21]. The algorithm was validated by twofold cross-validation using five randomized repeat trials for improved robustness. A two-stage approach was taken: (1) to distinguish between disease and healthy control and (2) to distinguish between eOA, eRA, and non-RA. We used accuracy of case and control classification to optimize algorithm features. Diagnostic characteristics, including area under the ROC (AUROC), are given with 95% CI determined via bootstrap analysis. The contribution of each feature in the algorithms to classification accuracy was assessed by determining the change in AUROC when a feature was omitted from the algorithm and retrained; a negative change represents a valuable feature, and a positive change an adverse feature, for classification accuracy. Data were analyzed using MATLAB version R2017A software (MathWorks, Natick, MA, USA).
Statistical analysis
Results are expressed as mean ± SEM unless otherwise stated. Following a normality test, one-way analysis of variance (ANOVA) with Tukey’s posttest was performed for histology, MACH-1, and amino acid analytes. Pearson’s correlations were performed between global OA score, parameters of MACH-1, and amino acid biomarkers. Given the asymmetric distribution of biomarkers, a logarithmic transformation was considered to satisfy the hypothesis of normality. ANOVA was applied to compare each biomarker between age groups. The same analysis was used to compare the different parameters between age groups. The association between the log-transformed biomarkers and the parameters was assessed by Pearson’s correlation. A multiple regression model (including as independent variables age, the parameter of interest, and an interaction term between these two factors) was constructed in order to investigate the influence of this parameter in the biomarker-age relationship (potential confounding factor). The results were considered to be significant at the 5% critical level (p < 0.05). There was no repeated analysis of the guinea pigs or human subjects, so repeated measures analysis is not applicable. For longitudinal analysis of multilayer cultures, to investigate a possible difference between the two groups, a mixed model with an undefined covariance matrix was applied to the data. The independent variables considered in this model were time, IL-1β treatment, and interaction between them. This statistical approach allowed us to compare biomarker production curves between the two groups while taking into account the presence of correlated data. For significance tests and correlation analysis of 14 glycated, oxidized, and nitrated amino acids and hydroxyproline analyzed in serum filtrate and 14 glycation, oxidation, and nitration adduct residues and CP in serum protein (15 analytes in each sample type), analyzed without preconceived hypothesis, a Bonferroni correction of 15 was applied. The predictive ability of these analytes for development of OA was studied by developing a partial least squares (PLS) regression model. The model was trained to learn to predict OA histological score from concentrations of serum glycated, oxidized, and nitrated amino acids (FL, CML, CEL, MG-H1, G-H1, 3DG-H, CMA, AASA, GSP, GSA, NFK, DT, 3-NT, pyrraline, Hyp, and CP) with the 4–36 weeks guinea pig study groups. Subsequent to training, the model was used to predict OA histological score for each guinea pig. The residual error between model predictions and the actual OA histological score was estimated as root mean squares error. Error at each individual stage and the overall error at all stages were estimated. Data analysis was performed with SAS version 9.4 for Windows statistical software (SAS Institute, Cary, NC, USA).