Ex-Vivo Equine Cartilage Explant Osteoarthritis Model - A Metabolomics and Proteomics Study

Osteoarthritis (OA) is an age-related degenerative musculoskeletal disease characterised by loss of articular cartilage, synovitis, abnormal bone proliferation and subchondral bone sclerosis. Underlying OA pathogenesis is yet to be fully elucidated with no OA specific biomarkers in clinical use. Ex-vivo equine cartilage explants (n=5) were incubated in TNF-α/IL-1β supplemented culture media for 8 days, with media removed and replaced at 2, 5 and 8 days. Acetonitrile metabolite extractions of 8 day cartilage explants and media samples at all time points underwent 1D 1H nuclear magnetic resonance metabolomic analysis with media samples also undergoing mass spectrometry proteomic analysis. Within the cartilage, metabolites glucose and lysine were elevated following TNF-α/IL-1β treatment whilst adenosine, alanine, betaine, creatine, myo-inositol and uridine levels decreased. Within the culture media, four, four and six differentially abundant metabolites and 154, 138 and 72 differentially abundant proteins, with > 2 fold change, were identified for 1-2 day, 3-5 day and 6-8 day time points respectively. Nine potential novel OA neopeptides were elevated in treated media. Our innovative study has identified differentially abundant metabolites, proteins and extracellular matrix derived neopeptides, providing insightful information on OA pathogenesis, enabling potential translation for clinical markers and possible new therapeutic targets.


Introduction
Osteoarthritis (OA) is an age-related degenerative musculoskeletal disease characterised by loss of articular cartilage, synovial membrane dysfunction, abnormal bone proliferation, subchondral bone sclerosis and altered biochemical and biomechanical properties 1,2 . For horses in the UK, OA is one of the leading welfare issues, resulting in substantial morbidity and mortality 3,4 . It is estimated that OA accounts for 60% of lameness seen in horses 5 .
Within OA, extracellular matrix (ECM) degradation is driven by multiple matrix metalloproteinases (MMPs) and a disintegrin and metalloproteinases with thrombospondin motifs (ADAMTSs) 6 . However, the underlying pathogenesis of OA is yet to be fully elucidated with no disease-modifying treatments currently available 7,8 . Whilst a number of putative biomarkers have been identified for OA diagnosis in the horse, none are currently used within clinical practice 9 . Presently, equine OA is predominantly diagnosed through diagnostic imaging and clinical examination. However, due to the slow onset of the condition, this often leads to substantial pathology of the joint, particularly to articular cartilage prior to diagnosis 10 . There is therefore a need to develop diagnostic tests which are sensitive and specific to the early stages of OA, which are repeatable and reproducible, as well as gaining a greater understanding of the underlying pathogenesis 11,12 . Early detection of OA could enable timely management interventions which could potentially slow the progression of the disease.
Proteomics is the systematic, large scale study of proteins within biological systems to assess quantities, isoforms, modifications, structure and function 27 . Previous studies have undertaken mass spectrometry (MS) based proteomics using TNF-α and IL-1β OA models for secretome analysis of chondrocytes in vitro and ex-vivo cartilage explants [20][21][22]25 .
Results from these studies included increased media levels of MMPs, cartilage oligomeric matrix protein (COMP), aggrecan and collagen VI.
During OA pathology, disease-associated peptide fragments (neopeptides) are generated from cartilage breakdown due to increased enzymatic activity/abundance of MMPs, ADAMTSs, cathepsins and serine proteases [28][29][30] . MS analysis of these neopeptides can then be applied to identify potential early OA biomarkers 31 . Previously a murine 32 amino acid peptide fragment, generated through increased activity of MMP and ADAMTS-4/5 and subsequent aggrecan degradation, was found to drive OA pain via Toll-like receptor 2 32 .
Neopeptide targeting therefore has the potential to provide a localised analgesic at the site of joint degeneration 31 . Numerous equine OA studies investigating both synovial fluid (SF) and cartilage have identified potential neopeptides of interest 28,[33][34][35] . Development of antibodies targeted to OA specific neopeptides would provide the ability to monitor cartilage degeneration, assess therapeutic response and potentially provide future novel therapeutic targets 31,36 .
Metabolomics uses a systematic methodology to comprehensively identify and quantify the metabolic profiles of biological samples 37 . 1 H Nuclear magnetic resonance (NMR) metabolomics analysis provides a high level of technical reproducibility with a minimal level of sample preparation 38 . 1 H NMR analysis has previously been used to investigate OA in the SF of humans, horses, pigs and dogs 9,[39][40][41][42][43][44][45] . Synovial metabolites alanine, choline, creatine and glucose have been identified as differentially abundant in OA across multiple studies and species 9,39,[41][42][43][44] . NMR techniques have also previously been used to characterise cartilage with high resolution magical angle spinning (HRMAS) NMR utilised to assess enzymatic degradation of bovine cartilage [46][47][48] . A guinea pig OA model using HRMAS NMR identified elevations in methylene resonances associated with chondrocyte membrane lipids and an increase in mobile methyl groups of collagen 49 . Another HRMAS NMR study of human OA cartilage identified a reduction in alanine, choline, glycine, lactate methyne and N-acetyl compared to healthy control cartilage 50 . However, no NMR studies to date have investigated the metabolic profile of culture media following the incubation of ex-vivo cartilage within an OA model. This is the first study to carry out 1 H NMR metabolomic analysis of extracted cartilage metabolites and also to undertake 1 H NMR analysis of culture media using the TNF-α/IL-1β ex-vivo OA cartilage model. Additionally, this is also the first study to use a multi 'omics' approach to simultaneously investigate the metabolomic profile of ex-vivo cartilage and metabolomic/proteomic profiles of culture media using this OA model. It was hypothesised that following TNF-α/IL-1β treatment of ex-vivo equine cartilage, 1 H NMR metabolomic and MS proteomic platforms would identify a panel of cartilage metabolites which were able to differentiate control from treated cartilage and a panel of metabolites, proteins and neopeptides within the associated culture media which were differentially abundant at each tested time point of the OA model.

Equine Ex-Vivo Cartilage Collection
Full thickness cartilage was removed from all articular surfaces within five separate metacarpophalangeal joints of five nine-year-old mares of unknown breed within 24 hr of slaughter at a commercial abattoir (F Drury and Sons, Swindon, UK). Cartilage samples were collected as a by-product of the agricultural industry. The Animals (Scientific Procedures) Act 1986, Schedule 2, does not define collection from these sources as scientific procedures and ethical approval was therefore not required. Cartilage collected from all joints was considered macroscopically normal with a score of 0 according to the OARSI histopathology initiative scoring system for horses 51  Cartilage was dissected into 3 mm 2 sections and divided into two for each donor (control and treatment wells) on a twelve well plate (Greiner Bio-One Ltd., Stonehouse, UK).
Explants were incubated for 24 hr in complete media within a humidified atmosphere of 5% (v/v) CO2 at 37°C. Culture media was removed, explants washed in phosphate buffered saline (PBS, Sigma-Aldrich) and replaced with serum free media (control) or serum free media supplemented with 10 ng/ml TNF-α (PeproTech EC Ltd., London, UK) and 10 ng/ml IL-1β (R&D Systems Inc., Minneapolis, Minnesota, USA) (treatment). After 48 hr, media was removed, centrifuged at 13,000g, 4°C for 10 min, supernatant removed and ethylenediaminetetraacetic acid (EDTA)-free protease inhibitor cocktail (Roche, Lewes, UK) added to cell-free media. Supernatant was then snap frozen in liquid nitrogen and stored at -80°C. Cartilage was washed in PBS and control/treatment culture media replaced as appropriate. Media collection was repeated at five and eight days. After day eight, cartilage was washed in PBS, weighed, snap frozen in liquid nitrogen and stored at -80°C.

Cartilage -Metabolite Extraction
Equal masses of cultured cartilage explants (in addition to three macroscopically normal equine cartilage samples, each divided into three to assess metabolite extraction reproducibility) were thawed out over ice and added to 500 µl of 50:50 (v/v) ice cold acetonitrile (ThermoFisher Scientific, Massachusetts, USA):dd 1 H2O and incubated on ice for 10 min. Samples were then sonicated using a microtip sonicator at 50 KHz and 10 nm amplitude in an ice-bath for three 30 s periods, interspersed with 30 s rests (that ensured extraction mixture temperature did not exceed 15°C). The extraction mixture was then vortexed for 1 min, centrifuged at 12,000g for 10 min at 4°C, supernatant transferred before being snap frozen in liquid nitrogen, lyophilised and stored at -80°C 37 .

Culture Media -NMR Sample Preparation
Culture media was thawed over ice and centrifuged for 5 min at 21,000g and 4°C. 150 µl of thawed culture media was diluted to a final volume containing 50% (v/v) culture media, 40% (v/v) dd 1 H2O, 10% 2 H2O and 0.0025% (v/v) NaN3, within an overall concentration of 8 500 mM PO4 3-pH 7.4 buffer. Samples were vortexed for 1 min, centrifuged at 13,000g for 2 min at 4°C, 250 µl supernatant removed and transferred to 3 mm outer diameter NMR tubes using a glass pipette.

NMR Acquisition
For each individual sample, 1D 1 H NMR spectra, with the application of a Carr-Purcell-Meiboom-Gill (CPMG) filter to attenuate macromolecule (e.g. proteins) signals, were acquired using the standard vendor pulse sequence cpmgpr1d on a 700 MHz NMR Bruker Avance III HD spectrometer with associated TCI cryoprobe and chilled Sample-Jet autosampler. All spectra were acquired at 25°C, with a 4 s interscan delay, 256 transients for cartilage spectra and 128 transients for media spectra, with a spectral width of 15 ppm. Topsin 3.1 and IconNMR 4.6.7 software programmes were used for acquisition and processing undertaking automated phasing, baseline correction and a standard vendor processing routine (exponential window function with 0.3 Hz line broadening). In addition to all cartilage extract and culture media samples, protease inhibitor cocktail and treatment cytokines TNF-α and IL-1β were also analysed separately to evaluate their metabolite profiles.

Metabolite Annotation and Identification
All acquired spectra were assessed to determine whether they met minimum reporting standards (as outlined by the Metabolomics Society) prior to inclusion for statistical analysis 52 . These included appropriate water suppression, flat spectral baseline and consistent linewidths. Metabolite annotations and relative abundances were carried out using Chenomx NMR Suite 8.2 (330-mammalian metabolite library). When possible, metabolite identifications were confirmed using 1D 1 H NMR in-house spectral libraries of metabolite standards. All raw 1D 1 H NMR spectra, together with annotated metabolite HMDB IDs and annotation level, are available within the EMBL-EBI MetaboLights repository (www.ebi.ac.uk/metabolights/MTBLS1495) 53 . Quantile plots of 1D 1 H NMR spectra are shown in Figure S3.

Protein Assay and StrataClean TM Resin Processing
Culture media was thawed over ice and centrifuged for 5 min at 21,000g and 4°C. Media sample concentrations were determined using a Pierce® 660 nm protein assay (Thermo Scientific, Waltham, Massachusetts, USA). 50 µg of protein for each sample was diluted with dd H2O, producing a final volume of 1 ml. StrataClean TM resin (10 µl) (Agilent, Santa Clara, California, USA) was added to each sample, rotated for 15 min, centrifuged at 400g for 1 min and the supernatant removed and discarded. Samples were then washed through the addition of 1 ml of ddH2O, vortexed for 1 min, centrifuged at 400g for 1 min and the supernatant removed and discarded. The wash step was repeated two further times.

Protein Digestion
160 µl of 25 mM ammonium bicarbonate (Fluka Chemicals Ltd., Gillingham, UK) containing 0.05% (w/v) RapiGest (Waters, Elstree, Hertfordshire, UK) was added to each sample and heated at 80°C for 10 min. DL-Dithiothreitol (Sigma-Aldrich) was added to produce a final concentration of 3 mM, incubated at 60°C for 10 min then iodoacetamide (Sigma-Aldrich) added (9 mM final concentration) and incubated at room temperature in the dark for 30 min. 2 µg of proteomics grade trypsin (Sigma-Aldrich) was added to each sample, rotated at 37°C for 16 hr and trypsin treatment then repeated for a 2 hr incubation.
Finally, digests were centrifuged at 13,000g and 4°C for 15 min and the supernatant removed and stored at 4°C.

Label Free LC-MS/MS
All media digests were randomised and individually analysed using LC-MS/MS on an UltiMate 3000 Nano LC System (Dionex/Thermo Scientific) coupled to a Q Exactive TM Quadrupole-Orbitrap instrument (Thermo Scientific). Full LC-MS/MS instrument methods are described in the supporting information. Tryptic peptides, equivalent to 250 ng of protein, were loaded onto the column and run over a 1 hr gradient, interspersed with 30 min blanks (97% (v/v) high performance liquid chromatography grade H20 (VWR International), 2.9% acetonitrile (Thermo Scientific) and 0.1% TFA. In addition to individual time points, pooled samples for control and treatment groups were also analysed to investigate differences in the overall secretome. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD017153 and 10.6019/PXD017153 54 .
Representative ion chromatograms are shown in Figure S4.

LC-MS/MS spectra processing and protein identification
Spectral alignment, peak picking, total protein abundance normalisation and peptide/protein quantification were undertaken using Progenesis™ QI 2.0 (Nonlinear Dynamics, Waters). The exported top ten spectra for each feature were then searched against the Equus caballus database for peptide and protein identification using PEAKS® Studio 8.5 (Bioinformatics Solutions Inc., Waterloo, Ontario, Canada) software. Search parameters were: precursor mass error tolerance, 10.0 ppm; fragment mass error tolerance, 0.01 Da; precursor mass search type, monoisotopic; enzyme, trypsin; maximum missed cleavages, 1; non-specific cleavage, none; fixed modifications, carbamidomethylation; variable modifications, oxidation or hydroxylation and oxidation (methionine). A filter of a minimum of 2 unique peptides was set for protein identification and quantitation with a false discovery rate (FDR) of 1%.

1D SDS PAGE
Media samples for each donor were combined for all time points and analysed via one dimensional sodium dodecyl sulphate polyacrylamide gel electrophoresis (1D SDS PAGE).
1 μg of each sample was added to Laemmli loading buffer Novex™ (Thermo Scientific) producing a final concentration of 15% glycerine, 2.5% SDS, 2.5% Tris (hydroxymethyl) aminomethane, 2.5% HCL and 4% β-mercaptoethanol at pH 6.8 and heated at 95°C for 5 min. Samples were loaded onto a 4-12% Bis-Tris polyacrylamide electrophoresis gel (NuPAGE™ Novex™, Thermo Scientific) and protein separation carried out at 200 V for 30 min at room temperature. Protein bands were visualised via silver staining (Thermo Scientific) following manufacturer instructions. Gel images were converted to 8 bit grey scale and protein band intensities analysed using densitometry with the software Image J (NIH, Bethesda, Maryland).

Semi-Tryptic Peptide Identification
To identify potential neopeptides a 'semi-tryptic' search was undertaken. The same PEAKS® search parameters were used as for protein identification, with the exception that 'non-specific cleavage' was altered from 'none' to 'one'. The 'peptide ion measurements' file was then exported and analysed using the online neopeptide analyser software tool 36 .

Statistical Analysis
Cartilage metabolite profiles were normalised using probabilistic quotient normalisation (PQN) 55 . Media metabolites were normalised to TSP concentration and protein profiles normalised to total ion current (TIC). Prior to multivariate analysis, metabolite and protein profiles were Pareto scaled 56 . MetaboAnalyst 3.5 (http://www.metaboanalyst.ca) was used to produce principal component analysis (PCA) plots and provide PC1 (principal component 1) loadings magnitude values. t-tests were carried out using MetaboAnalyst 3.5 (protein and metabolite abundances) and the neopeptide analyser (neopeptides) with p < 0.05 (and a fold change of > 2 for proteins) considered statistically significant. The Benjamini-Hochberg false discovery rate method was applied for correction of multiple testing 57 . The SPSS 24 software package was used to produce all box plots and PC1 loadings magnitude graphs.

Pathway Analysis
Pathway analysis was conducted for differentially abundant metabolites using the online tool KEGG for Equus caballus 58 . Differentially abundant proteins (> 2 fold change) were also analysed using KEGG via the STRING database 59 . Manual cross referencing was conducted of metabolites and proteins found to be significantly different between sample groups to biological pathways. A filter of a minimum of two metabolites/proteins was set for each sample set for inclusion of the relevant biological pathway.

Protease inhibitor cocktail, TNF-α and IL-1β metabolite profiles
Protease inhibitor cocktail was found to have high levels of mannitol and thus this metabolite was removed from all analyses. Within the spectral profiles of TNF-α and IL-1β acquired separately, the metabolites acetate, acetone, ethanol, formate, lactate, methanol and succinate were identified. These metabolites were therefore also removed from further analyses.

Analysis of Cartilage Metabolites
Acetonitrile metabolite extraction was identified to be highly reproducible with technical replicates clustering within a PCA plot for three separate macroscopically normal cartilage samples ( Figure 1). In total 35 metabolites were identified within equine cartilage (Table 1).
Of these, following the removal of metabolites previously mentioned, eight were identified as being differentially abundant between control and treatment groups (Figure 2). Glucose and lysine levels were elevated following TNF-α/IL-1β treatment whilst adenosine, alanine, betaine, creatine, myo-inositol and uridine levels decreased. PCA identified that metabolite profiles separated into two distinct clusters, separating control and treatment groups

Analysis of Media Metabolites
Spectral quality control via metabolomics standards initiative identified two samples that failed due to salt precipitation and as such were removed from further analyses 52,61 .
Following metabolite identification and quantification, one sample was identified as an outlier and subsequently removed from statistical analyses. Isopropanol was identified within all media samples. As this was considered a likely contaminant during cartilage culture, together with metabolites previously mentioned, isopropanol was also removed from all analyses. In total, 34 metabolites were identified within the culture media (Table   1). Time points were analysed separately with four, four and six metabolites identified as being differentially abundant between control and treatment groups for 1-2 day, 3-5 day

Analysis of Media Proteins
In total, 352 proteins were identified within analysed culture media samples with an elevated number of identified proteins within treated compared to control samples ( Figure   S5 and Table S1). Of all identified proteins, 131 were identified within TNF-α/IL-1β treated    Figure S7.

Semi-Tryptic Peptides
PCA of all identified semi-tryptic peptides within combined control and combined treated samples identified far less variation within the treatment group (Figure 8). This was also identified for all time points analysed individually ( Figure S8). In total, nine potential novel OA neopeptides were identified which were elevated in treated media samples (Table 2).
These included semi-tryptic peptides of extracellular matrix proteins aggrecan, cartilage intermediate layer protein, collagen type VI α 2 chain and vimentin.

Discussion
In this study, TNF-α/IL-1β treatment of ex-vivo equine cartilage explants was used to model OA to gain a greater understanding of OA pathogenesis and identify potential OA markers. 1 H NMR metabolomic and LC-MS/MS proteomic analysis of culture media at 0-2, 3-5 and 6-8 days was undertaken. In addition, 1 H NMR metabolomic analysis of acetonitrile extracted cartilage metabolites (following 8 days incubation) was also undertaken.
Within culture media, following TNF-α/IL-1β treatment, elevations in endopeptidases MMP-1 and MMP-3 at 0-2 days, with a similar trend at both other time points, were identified as expected 62 . Elevated MMP-1 activity has previously been identified within equine OA SF with general MMP activity also found to be correlated to severity of cartilage damage 63,64 .
Also, as previously reported, elevations in the non-collagenous ECM protein COMP were also identified within the TNF-α/IL-1β equine OA model, with COMP considered a marker of cartilage breakdown 65 Following TNF-α/IL-1β treatment, elevations of glucose within the cartilage were identified. This is supported by a previous study which demonstrated that TNF-α and IL-1β upregulate glucose transport in chondrocytes through upregulation of glucose transporter (GLUT)1 and GLUT9 mRNA synthesis with increased levels of glycosylated GLUT1 incorporated into the plasma membrane 73 . This influx in glucose is likely, at least in part, to be due to the increased energy requirement following cytokine stimulation in the production of MMPs and secretion of Il-6, Il-8, hematopoietic colony-stimulating factor and prostaglandin E2 74 .
Alanine, arginine, choline and citrate were all identified to be differentially abundant within culture media following TNF-α/IL-1β treatment at the earliest time point, 0-2 days.
Within this study, arginine levels were initially identified as decreased following treatment at the earliest time point. A recent study of human plasma also identified arginine to be depleted in knee OA 75 . The authors proposed this is due to an increased activity of the conversion of arginine to ornithine resulting in an imbalance between cartilage repair and degradation. This is supported by a recent learning and network approach of OA associated metabolites in which arginine and ornithine appeared in about 30% and 25% of the generated models studied 76 . In addition to this, a reduction in arginine may be reflective of an increased production of nitric oxide (L-arginine being converted to NOHarginine and subsequently L-citrulline and nitric oxide) as identified in human OA cartilage 77,78 .
A 1 H NMR metabolomics study of equine SF also identified elevated levels of choline in OA 41 . However, elevated levels of alanine and citrate were also identified whilst these were found to be decreased within our study.
Across the whole study, alanine was found to be a central component in discriminating control and treatment groups. Alanine levels were depleted in treated cartilage extracts compared to controls, with PLS-DA identifying alanine as the fourth most important component in discriminating control and treated cartilage samples. Reduced alanine levels were also identified in human OA cartilage using HRMAS NMR spectroscopy 50 . Within culture media, at all three time points alanine was depleted in treated samples and KEGG pathway analysis revealed the 'alanine, aspartate and glutamate metabolism' pathway to be well represented by both metabolites and proteins. Alanine has previously been identified as a key component of the metabolic urinary OA profile of guinea pigs 79 .
Isoleucine was elevated within the media during the latter stages of the model (6-8 days However, within this study, although a higher abundance was recorded for isoleucine in treated compared to control cartilage, this did not reach statistical significance.
Furthermore, elevations in glutamate were identified within culture media at 3-5 days, which may be resultant of the catabolism of collagenous proline through proline oxidase 81 .
Reduced levels of collagen type VI α 2 chain and collagen type X α 1 chain were identified at 3-5 and 6-8 days following cytokine treatment. This may reflect a reduction in collagen synthesis which has previously been identified within other collagen types following TNFα/IL-1β stimulation 22 . Therefore these results provide evidence of a disruption in collagen homeostasis and suggest that collagens are being degraded within the model sooner than the 14-28 days previously reported within other ex-vivo cartilage OA models 82,83 . Fibromodulin is a small leucine-rich repeat proteoglycan which interacts with collagen fibrils and influences fibrillogenesis rate and fibril structure 92 . Experimental mice which lack biglycan and fibromodulin have been shown to develop OA in multiple joints 93,94 .
Neopeptides generated from fibromodulin degradation have also been identified as potential markers of equine articular cartilage degradation 28

Study Limitations
The cytokine preparations used within this study contained various metabolites which, following their removal from subsequent statistical analysis, prevented the analysis of some various metabolites within the experiment. Additionally, culture media was supplemented with a protease inhibitor cocktail at collection in order to inhibit general protein degradation prior to MS proteomic analysis, with results therefore representing the peptide/protein composition during experimentation. However, the high mannitol content prevented analysis of this metabolite within the samples/spectral region. Therefore, when in future using cytokines/supplements for NMR metabolomics, analysing the spectra of different manufacturers/preparations prior to experimentation may be beneficial to identify their associated metabolite profiles, selecting the most appropriate products in order to maximise downstream interpretation of results.

Conclusion
In conclusion, this is the first study to use a multi 'omics' approach to simultaneously investigate the metabolomic profile of ex-vivo cartilage and metabolomic/proteomic profiles of culture media using the TNF-α/IL-1β ex-vivo OA cartilage model. We have identified a panel of metabolites and proteins which are differentially abundant within an early phase of the OA model, 0-2 days, which may provide further information on the underlying disease pathogenesis and well as potential to translate to clinical markers. This study has also identified a panel of potential, ECM derived, neopeptides which have potential to help enable OA stratification as well as provide potential novel therapeutic targets.

Supporting Information
Liquid Chromatography Tandem Mass Spectrometry -Detailed Methods Figure S1. Five post mortem equine metacarpophalangeal joints used for ex-vivo cartilage culture.       Table S1. Proteins identified within culture media of control and TNF-α/IL-1β treated exvivo equine cartilage time points.

Ethics
Cartilage samples were collected as a by-product of the agricultural industry. The Animals (Scientific Procedures) Act 1986, Schedule 2, does not define collection from these sources as scientific procedures and ethical approval was therefore not required.

Corresponding author
James R Anderson janders@liverpool.ac.uk Tel: 01517949287

Author Contributions
Wrote

Notes
The authors declare no competing financial interest.