Optical photothermal infrared spectroscopy can differentiate equine osteoarthritic plasma extracellular vesicles from healthy controls

Background Equine osteoarthritis is a chronic degenerative disease of the articular joint, characterised by cartilage degradation resulting in pain and reduced mobility and thus is a prominent equine welfare concern. Diagnosis is usually at a late stage through radiographic examination, whilst treatment is symptomatic not curative. Extracellular vesicles are small nanoparticles that are involved in intercellular communication. The objective of this study was to investigate the feasibility of Raman and optical photothermal infrared spectroscopy to detect osteoarthritis using plasma-derived extracellular vesicles. Methods Plasma samples were derived from thoroughbred racehorses. A total of 14 samples were selected (control; n= 6 and diseased; n=8). Extracellular vesicles were isolated using differential ultracentrifugation and characterised using nanoparticle tracking analysis, transmission electron microscopy, and human tetraspanin chips. Samples were then analysed using Raman and optical photothermal infrared spectroscopy. Results Infrared spectra were analysed between 950-1800 cm-1. Raman spectra had bands between the wavelengths of 900-1800 cm-1 analysed. Bands below 900 cm-1. Spectral data for both Raman and optical photothermal infrared spectroscopy was used to obtain a classification model and confusion matrices, characterising the techniques ability to distinguish diseased samples. Optical photothermal infrared spectroscopy could differentiate osteoarthritic extracellular vesicles from healthy with good classification (93.4%) whereas Raman displayed poor classification (64.3%). Plasma-derived extracellular vesicles from osteoarthritic horses contained increased signal for proteins, lipids and nucleic acids. Discussion/ conclusion For the first time we demonstrated the ability to use optical photothermal infrared spectroscopy to interrogate extracellular vesicles and osteoarthritis-related samples. Optical photothermal infrared spectroscopy was superior to Raman in this study, and could distinguish osteoarthritis samples, suggestive of its potential use diagnostically to identify osteoarthritis in equine patients. This study demonstrates the potential of Raman and optical photothermal infrared spectroscopy to be used as a diagnostic tool in clinical practice, with the capacity to detect changes in extracellular vesicles from clinically derived samples.


Introduction 53
Osteoarthritis (OA) is a common degenerative disease of the synovial joint, characterised by catabolic 54 processes observed in articular cartilage, and a notable imbalance in bone remodelling. It results in 55 pain, inflammation and reduced mobility (Mustonen and Nieminen, 2021). OA is the most prevalent 56 cause of equine lameness, with over 60% of horses developing OA within their lifetime; a significant 57 welfare concern (McIlwraith et al, 2012). It is a complex heterogeneous condition of multiple causative 58 factors, including mechanical, genetic, metabolic and inflammatory pathway involvement (Clarke et 59 al, 2021). OA pathophysiology is conserved across species, resulting in synovitis, cartilage degradation, 60 osteophyte formation, subchondral bone sclerosis, fibrosis and reduced elastoviscosity of synovial via optical photothermal infrared spectroscopy (O-PTIR), which is based on a pump-probe 95 configuration that couples a tuneable infrared quantum cascade laser (QCL) acting as pump and a 96 visible laser to probe the thermal expansion resulting from the temperature rise induced by the QCL. 97 The probe laser also acts as excitation source for acquiring Raman spectrum simultaneously with 98 infrared data at the same spatial resolution. This scheme has been used to interrogate tissue samples, 99 mammalian cells (Banas et al, 2021;Spadea et al, 2021) and bacteria (Lima et al, 2021) but this is the 100 first study, to our knowledge, to use it in EV or OA work. 101 We hypothesised that Raman and O-PTIR can be used to identify potential biomarkers of OA using 102 plasma-derived EVs. 103

Materials and Methods 104
Sample Selection 105 Plasma samples were collected in accordance with the Hong Kong Jockey Club owner consent 106 regulations (VREC561). All samples were from thoroughbred racehorses, with the donor cohort having 107 a mean age (+/-SEM) of 6.57 +/-0.45. Horses were selected based on histological scoring of OA severity 108 using a modified Mankin score (McIlwraith et al, 2010). A total of 14 samples were selected (control; 109 n= 6 (mean score +/-SEM = 1.83 +/-0.48) and diseased; n=8 (mean score +/-SEM= 16.25 +/-1.15). 110 Extracellular Vesicle Isolation -Differential Ultracentrifugation 111 Equine plasma samples underwent differential ultracentrifugation (dUC) in order to isolate EVs. 112 Samples were subjected to a 300g spin for 10 minutes, 2000g spin for 10 minutes, 10,000g spin for 30  EV presence and morphology were characterised using transmission electron microscopy. 10 µl of 129 each sample was placed onto a carbon coated glow discharged grid and incubated at room 130 temperature for 20 minutes. Samples were then subject to a negative staining protocol. EVs were fixed 131 onto the grid with 1% glutaraldehyde for 5 minutes. The sample grids were incubated on 1% aqueous 132 uranyl acetate (UA) (Thermofisher Scientific, Massachusetts, USA), for 60 seconds, followed by 133 4%UA/2% Methyl Cellulose (Sigma Aldrich, Gillingham, UK) at a 1:9 ratio on ice for 10 minutes. Grids 134 were then removed with a 5mm wire loop and dried. The prepared grids were then viewed at 120KV 135 on a FEI Tecnai G2 Spirit with Gatan RIO16 digital camera. 136

Exoview Characterisation 137
The exoview platform (NanoView Biosciences, Malvern Hills Science Park, Malvern) was used to 138 determine EV concentration, surface marker identification and to perform fluorescent microscopy and 139 tetraspanin colocalization analysis. We had previously tested both the human and murine chips on 140 equine samples and demonstrated the human chips were more compatible (data not shown). ExoView 141 analyses EVs using visible light interference for size measurements and fluorescence for protein 142 profiling. Samples were analysed in triplicate using the ExoView Tetraspanin Kit (NanoView 143 Biosciences, USA) and were incubated on the human ExoView Tetraspanin Microarray Chip for 16 144 hours at room temperature. Following this sample chips were incubated with tetraspanin labelling 145 antibodies, namely anti-CD9 CF488, anti-CD81 CF555 and anti-CD63 CF647 and the MIgG negative 146 control. The antibodies were diluted 1:500 in PBST with 2% BSA. The chips were incubated with 250 147 µL of the labelling solution for 1 hour. The sample chips were washed and imaged with the ExoView 148 R100 reader ExoView Scanner v3.0. Data was analysed using ExoView Analyzer v3.0. Fluorescent cut 149 offs were set relative to the MIgG control. Total EVs were determined as the number of detected 150 particles bound to tetraspanin antibodies (CD9, CD81, CD63) and normalised to MIgG antibody. 151

Raman Spectroscopy and Infrared Spectroscopy (O-PTIR) 152
For all samples O-PTIR measurements were acquired on single-point mode using a mIRage infrared 153 microscope (Photothermal Spectroscopy Corp., Santa Barbara, USA), with the pump consisting of a 154 tuneable four-stage QCL device, while the probe beam is a continuous wave (CW) 532 nm laser. 155 Spectral data were collected in reflection mode using a 40×, 0.78 NA, and 8 mm working distance 156 Schwarzschild objective. Single-point spectral data were acquired over a spectral region of 930−1800 cm −1 , with 2cm −1 spectral resolution and 10 scans per spectrum. Raman data were acquired using a 158 Horiba Scientific iHR-320 spectrometer coupled to mIRage, using a grating of 600 l/mm, 10 s as 159 acquisition time, spectral region of 500−3400 cm −1 , with 2 cm −1 spectral resolution and 10 scans per 160 spectrum. 161

Statistical Analysis 162
Raman and O-PTIR spectroscopic data was analysed using principal component analysis (PCA) to 163 determine each techniques ability to identify diseased samples from healthy. Spectral data was also 164 used as input for partial least squares discriminant analysis (PLS-DA) in order to generate a 165 computerised model. Further analysis involved using a classification model and confusion matrices, 166 whereby bootstrapping was performed 10000 times to permute whether the EV spectrum was 167 classified as OA or control. 168 Results 169

EV Characterisation 170
Nanoparticle tracking analysis 171 Particle size and concentration characterisation was performed using NTA. NTA determined the 172 average plasma sample concentration to be 2.02x10 9 particles/ml. Analysis was suggestive of a 173 heterogeneous population of EVs, ranging from exosomes to microvesicles ( Figure 1A). 174

Transmission electron microscopy 175
To confirm the particles isolated from plasma samples were indeed EVs we negatively stained and 176 visualised them using transmission electron microscopy. Spherical structures within EV size ranges 177 (30nm-100nm (exosomes) and 100nm-1000nm (microvesicles)) were identified with a clearly defined 178 peripheral membrane as shown in Figure 1B. 179 180 Exoview 181 Exoview was used on a representative pool of plasma samples. The EVs extracted from plasma had 182 the highest particle counts on the CD9 capture spots, equating to a concentration of around 1.4x10 8 183 CD9 positive particles/ml. CD81 (7x10 7 particles/ml) and CD63 (6x10 7 particles/ml) positive particles 184 were also detectable ( Figure 2A). It was observed that most EVs detected were less than 100nm 185 ( Figure 2B). It was also found that with plasma EV samples, the greater the expression of CD9 the 186 greater the expression of CD81, whereby a distinct positive correlation can be observed ( Figure 2C). 187 Equine SF and Plasma OA EVs membranes, Raman Spectroscopy, Draft Manuscript Co-localisation analysis was also performed and identifying that 91% of plasma-derived EVs were 188 positive for the CD9 surface tetraspanin, followed by 5% expressing CD81, and 3% expressing both 189 CD81 and CD9. CD63 expression was lowest at 0.8% ( Figure 2D). Finally, plasma EVs were visualised 190 using fluorescent microscopy, as shown in Figure 2E. phosphodiester bonds in nucleic acids (Paraskevaidi et al, 2021). The peak at 1451 cm −1 corresponded 208 to bending vibration (scissoring) of acyl CH2 groups in lipids (Mihaly et al, 2017;Barth et al, 2007;209 Paolini et al, 2020), whereas the band peaking at 1395 cm −1 aroses from COO − symmetric stretching 210 from amino acid side chains and fatty acids (Mihaly et al, 2017;Barth, 2007;Paolini et al, 2020). 211 Spectral signatures from minerals were also observed in Raman spectrum acquired from healthy and 212 diseased samples in the low wavenumber region (below 900 cm -1 ), therefore, only bands peaking 213 between 900-1800 cm -1 were analysed. In Raman spectra, peaks associated to amide I, II, and III from 214 by Palviainen et al (2020) 2.46x10 9 -1.10x10 10 particles per ml. However, a study analysing plasma EVs 249 across the duration of equine endurance racing found that baseline plasma EV concentration was 250 5.6x10 12 particle per ml (de Oliveira Jr et al, 2021). Exoview analysis identified EVs that were positive 251 for the tetraspanins CD81 and CD9, however a low percentage of EVs were positive for the surface 252 Equine SF and Plasma OA EVs membranes, Raman Spectroscopy, Draft Manuscript marker CD63. This may a result of poor protein homology between equine and humans. Although, a 253 recent paper using intracellular trafficking demonstrated that CD63 expression in HeLa cells was 254 specific to exosomes, and often a lack of CD63 expression may be due to small microvesicle 255 production, referred to as ectosomes, and that this type of EV is far more prolific than CD63 positive 256 exosomes (Mathieu et al, 2021). 257 258 Infrared spectroscopy techniques have been used to probe EVs previously in order to identify 259 structural components (Kim et al, 2019) as well as proteins, lipid and nucleic acid components, as 260 found in our study (Kim et al, 2019). This is the first paper to our knowledge to probe OA EVs using 261 Raman spectroscopy and O-PTIR spectroscopy. In other work, Zhai et al (2019) found that in bone, 262 mineral and carbonate content varied significantly with OA stage, with carbonate increasing with OA. 263 They also identified using Fourier transform infrared spectroscopy that acid phosphate, collagen 264 maturity and crystallinity varied with OA. In addition, the use of infra-red spectroscopy is compounded 265 by the findings of Afra et al (2017) in a study utilising an experimental model of OA in rats. Here the 266 spectral differences between control and OA samples could be correlated to Mankin score and 267 glycosaminoglycan content. A previous study by our group used attenuated total reflection Fourier-268 transform infrared (ATR-FTIR) spectroscopy on OA equine serum. This infrared spectroscopy study 269 found separation between groups with 100% sensitivity and specificity, with the six most significant 270 peaks between groups being attributed to proteins and lipids. Similarly, this was observed within our 271 study, with increased abundance found within our OA group. The stated study postulated these 272 observations may be associated to increased lipid and protein expression including increased 273 expression of type 1 collagen, and decreased expression of type 2 collagen characteristic of OA 274 (Paraskevaidi et al, 2020). 275 276 Previously, changes in plasma and lipid concentration in plasma and serum derived from OA patients 277 has been described. One study utilised serum samples from horses to discriminate proteomic changes 278 due to exercise or the development of early OA (Frisbie et al, 2008). Researchers identified six 279 biomarkers with the ability to discriminate OA from exercise groups. For example, the concentration 280 of serum C1,2C (reflective of type 1 and 2 collagen degradation fragments) and collagen 1 was found 281 to significantly increase in OA groups compared with exercise alone (Frisbie et al, 2008). In addition, a 282 importance of Raman spectroscopy in identifying pathologically associated crystals such as 308 monosodium urate and calcium pyrophosphate dihydrate in rheumatoid diseases. 309 We recognise a number of limitations in our study. We were restrained by the number of samples 310 available to us, and our findings need validation in a larger cohort. In addition, large sample volumes 311 were necessary to have an adequate number of EVs for analysis, providing an appropriate signal to 312 noise ratio. In our future studies minimum sample volume required will be optimised. Additionally, 313 we used a single time point 'snap shot' of disease. Further work is needed to determine if O-PTIR is 314 sensitive enough to determine differences in a range of OA phenotypes and severities, and correlate 315 differences to specific biological functions of EVs. 316 Overall this study demonstrates the potential of Raman and OPTIR spectroscopy to be used as a 317 diagnostic tool in clinical practice. Further work is required to identify if OA-related changes in plasma-318 derived EVs is related to the pathogenesis in the joint. We are currently quantifying the EV cargo using NMR metabolomics, mass spectrometry proteomics and sequencing platforms in order to provide 320 complete characterisation of EVs in OA and determine their contribution to disease propagation. 321

Conclusion 322
In conclusion, EVs derived from equine plasma in OA were probed using Raman and O-PTIR 323 spectroscopy. O-PTIR spectroscopic data was found to be superior in classifying samples from OA 324 patients compared to Raman spectroscopy. O-PTIR spectroscopy is an exciting platform for bedside 325 analysis of plasma to diagnose OA.