Raman Spectroscopic Probe Provides Optical Biomarkers of Cartilage Composition Predictive of Tissue Function

Abstract

OBJECTIVE:

The diagnosis of early osteoarthritis when therapeutic interventions may be most effective at reversing cartilage degeneration presents a clinical challenge. We describe a Raman arthroscopic probe and spectral analysis that measures biomarkers reflective of the content of predominant cartilage ECM constituents—glycosaminoglycans (GAG), collagen, water—essential to cartilage function. We compare the capability of Raman-probe-derived biomarkers to predict functional properties of cartilage to quantitative MRI and histopathology assessments.

DESIGN:

Osteochondral blocks were sectioned from 6 bovine femoral condyles with no macroscopic injury (n=62 blocks) and 6 condyles with a focal chondral lesion (n=32 blocks), but no macroscopic degeneration of surrounding cartilage (n=34 blocks). Blocks from 10 human knees were further analyzed (age 27-75;n=235 blocks). Using a custom arthroscopic Raman spectroscopy probe, spectra of chondral layers were measured and subjected to multivariate linear decomposition to extract ECM biomarker scores, reflecting the contribution of each ECM constituent to the spectra. Blocks were further analyzed for elastic modulus, T2/T2* MRI relaxation times, and OARSI scores.

RESULTS:

For bovine tissues, Raman biomarkers revealed depleted GAG and cartilage softening peripheral to the lesion despite no macroscopic degeneration. Raman biomarkers accounted for 78% of GAG content variation and 71% of modulus variation. For human tissues, Raman biomarkers accounted for 71% of modulus variation. Raman biomarkers accounted for a greater variation of modulus (71%-72%) than OARSI (12-54%), T2* (15%-27%), or T2 (25%-30%).

CONCLUSIONS:

These data support the application of Raman probe-derived biomarkers for molecular assessment of key ECM constituents that define cartilage properties in health and disease.

Introduction

Hyaline cartilage is a viscoelastic, biphasic, composite material comprised of a type-II collagen fibril network (5-20%w/v) that provides tensile strength, complemented by a porous-permeable glycosaminoglycan (GAG) matrix (5-15%w/v) of negatively charged, sulfated polysaccharides that regulate the transport of interstitial water (70-90%w/v).¹ These constituents synergistically impart the rheological and tribological material properties essential for articular cartilage function: supporting the applied compressive and shear loads, while maintaining a nearly frictionless interface between articulating joint surfaces.²

Trauma, internal derangement, joint instability, ligamentous deficiency, skeletal malalignment, or obesity incite mechanical overloading and inflammation with subsequent injury to hyaline cartilage. These conditions induce surface wear,³ fissuring/fibrillation,⁴ tissue swelling,⁵ compositional loss of GAG,⁶ and derangement of the collagen network,⁷ which manifest anatomically as chondral lesions. Left untreated, degenerated cartilage will advance to post-traumatic osteoarthritis (PTOA), with degradation extending through the cartilage depth until bone-on-bone contact is evident, producing pain and disability. Treatment of chondral lesions aims to restore tissue composition essential for mechanical function. This often requires arthroscopic procedures whereby the degraded cartilage is debrided and neocartilage repair tissue is induced via techniques such as microfracture, autologous cell implantation [ACI], or osteochondral autograft transfer system [OATS].^8,9

Developing and implementing cartilage repair strategies requires objective assessment of changes in the composition of degraded cartilage and regenerated neocartilage that affect the tissue’s functional performance. Advanced imaging modalities, such as compositional MRI (T1ρ, T2, T2* mapping),^10,11 delayed gadolinium-enhanced MRI of cartilage (dGEMRIC),^12,13 ultrasound,^14,15 and optical coherence tomography (OCT)¹⁶ portray the composition, structure, or material properties of hyaline cartilage, but are encumbered by limitations as they are: 1) predominantly sensitive for diagnosing late stage PTOA after irreversible macroscopic changes in tissue composition has occurred, 2) only moderately correlated with direct quantitative valuations of cartilage composition and material behavior, or 3) not conducive to perioperative implementation. Arthroscopic-based cartilage grading schemes (e.g., Outerbridge) are limited by subjective estimates of articular surface topography that are poorly predictive of tissue pathoanatomy/physiology.¹⁷ Histopathological grading schemes that portray changes in the composition and morphology of cartilage (e.g., OARSI, Mankin) ^18,19 are standard for diagnosing the pathoanatomic stages of PTOA. However, these evaluations are invasive, destructive, and incompatible with sequential in-vivo clinical examinations. Accordingly, the development of a low cost, minimally invasive, diagnostic platform that achieves rapid and safe appraisal of cartilage composition and material behavior is required to evaluate and implement cartilage repair therapies.

Intra-articular Raman spectroscopy can serve as a transformative diagnostic platform to assess cartilage biochemical composition during arthroscopic procedures. In Raman spectroscopy, an inelastic optical light scattering technique, an incident laser’s photons are scattered with a wavelength change that corresponds to specific Raman active vibrational modes within a molecule.²⁰ Accordingly, the Raman spectrum provides a quantitative, point-wise optical fingerprint of a tissue’s molecular building blocks (amides, sulfates, carboxylic acids, and hydroxyls), allowing identification of the molecular constituents comprising the ECM. There is growing interest in using Raman spectroscopy ^21–29 and infrared spectroscopy ^30,31 to assess cartilage degeneration. Prior Raman spectroscopy studies have utilized predominantly microscopy-based systems incompatible with in-vivo diagnostics. We have developed a Raman probe capable of interrogating cartilage during intra-articular procedures,^32,33 allowing for the non-destructive, real-time, in-vivo evaluation of cartilage composition in patients. Our Raman spectroscopy platform adopts a multivariate spectral analysis workflow to extract ECM-specific optical biomarkers that reflect the composition of cartilage tissue. Previously, we showed that Raman probe-derived biomarkers account for GAG depletion and softening of cartilage explants subjected to enzymatic degeneration ex-vivo. ³² To advance intra-articular Raman spectroscopy towards clinical applications, we now evaluate the ability of our Raman probe to predict the composition and functional mechanical properties of naturally degenerated cartilage tissue using bovine and human cadaveric knee specimens to test the hypothesis that Raman probe-derived biomarkers predict cartilage material properties better than arthroscopic visual grading schemes (Outerbridge), histopathologic grading (OARSI), and quantitative compositional MRI (T2/T2* mapping). This hypothesis is examined through ex-vivo analyses of osteochondral tissue excised from: 1) skeletally mature bovine femoral condyle specimens with naturally occurring focal chondral lesions, and 2) human cadaver knee cartilage specimens over a range of ages and degenerative states.

Methods

Tissue specimens

Bovine cartilage:

The femoral articular surface of 12 disarticulated, skeletally mature, bovine knees were examined macroscopically and divided into a healthy femoral condyle [HC] group (n=6) and a lesion-afflicted condyle [LAC] group (n=6) possessing a ~∅10-15mm focal chondral lesion within the load bearing region of one of the femoral condyles, but no apparent macroscopic evidence of degeneration in the peripheral cartilage (Fig.1A). For tissue analysis, one condyle from each knee was sectioned along a rectangular grid using a fine-tooth handsaw (Fig.1C) to produce ~6x6x6mm osteochondral tissue blocks (n=10-15/condyle). To facilitate mechanical indentation testing of the cartilage, plano parallel surfaces were created by cutting the subchondral bone surface parallel to the articular surface via a diamond saw (Buehler Isomat 4000@1100 rpm). For LAC specimens, analyses were performed on specimens excised from the chondral lesion site (LS) and surrounding cartilage (i.e., lesion peripheral sites (LPS);Fig.1A). The material properties and composition of 128 osteochondral specimens excised from 12 condyles were examined by Raman probe spectroscopy, mechanical indentation, biochemical assay, and Outerbridge visual grading (Table.S1). Raman derived ECM biomarkers were regressed to cartilage biochemical content and tissue modulus. OARSI grading was compared to tissue modulus for a random subset of 30 specimens. An additional five intact bovine condyles were evaluated by MRI, then sectioned into 54 osteochondral block specimens and subjected to Raman probe spectroscopy and mechanical testing (Table.S2). Predictions of cartilage composition and tissue modulus achieved by Raman probe spectroscopy were compared to predictions from quantitative MRI.

Figure 1: Tissue models and experimental analysis.

Skeletally mature (A) bovine healthy condyles (HC) and bovine lesion afflicted condyles (LAC). LAC analysis is performed within the lesion site (LS) and lesion-peripheral site (LPS). (B) Representative human knee tissue specimens with and without macroscopic degeneration. Anatomical orientation of knee represented as: [m] medial, [l] lateral, [a] anterior, and (p) posterior. (C) Analytical workflow consists of sectioning condyles into osteochondral blocks for anatomical site-specific analysis.

Human cartilage:

The medial and lateral femoral condyles of 10 disarticulated human cadaver knees (Fig.1B) procured from the National Disease Research Interchange, ranging in age from 27 to 75 years (Table.S3), were examined macroscopically and cartilage morphology classified according to Outerbridge visual grading scheme prior to sectioning into ~6x6x6mm osteochondral specimens. Raman spectroscopy derived ECM biomarkers were regressed to GAG content (n=125 blocks) and tissue modulus (n=235 blocks). OARSI scores were compared to cartilage mechanical properties for a random subset of 71 specimens. Advanced pathoanatomy prevented analysis of one lateral condyle and biochemical assays were not performed on four lateral and three medial condyles. To compare predictions of cartilage mechanical properties derived from Raman spectroscopy to predictions from quantitative MRI, the medial/lateral femoral condyles from one donor and tibial plateaus from five donors were subjected to MRI then sectioned into 88 osteochondral block specimens and subjected to Raman spectroscopy and mechanical testing (Table.S4). Raman versus MRI comparisons were further performed for a subset of specimens with early OA pathoanatomy as defined by originating from a chondral surface with a mean Outerbridge grade <2.0 (Table S4).

Raman instrument and analysis

The Raman probe consists of an NIR diode laser (ex=785nm, 500mW, B&WTek), fiber-coupled spectrograph (QEPro, Ocean Optics), and thin 2mm probe tip (⊘2×50mm) with a mounted ⊘2mm sapphire ball lens that outputs 100mW laser power and provides ~270μm depth of penetration into the cartilage tissue (Fig.2A).³² The probe, fixed to a vertical translation micrometer stage was advanced until the ball lens was in gentle contact with the chondral surface (~0.1N tare load). To benchmark analysis using high quality Raman spectra, these were acquired with a 30-second integration time for bovine and human specimens. The spectral fingerprint range (800-1800cm⁻¹) was preprocessed using a 3^rd order polynomial baseline subtraction, followed by area-under-curve normalization. The resulting cartilage spectral profile was subjected to decomposition via curve-fit to a multivariate linear model:

$$Cartilag{e}_{\text{spectra}}={\text{GAG}}_{\text{score}}*\left(GA{G}_{REF}\right)+{\text{COL}}_{\text{score}}*\left(CO{L}_{REF}\right)+{\text{H}}_{\text{2}}{\text{O}}_{\text{score}}*(H_2{O}_{REF})$$

Figure 2: Raman spectral analysis.

(A) Depiction of Raman probe in contact with articular cartilage surface and magnified view of ⊘2mm ball lens interface. (B) Representative preprocessed fingerprint range spectra of a cartilage specimen. 2D stacked area plot showing the cumulative contribution of GAG, COL, H₂O, to composite fingerprint Raman spectra for a (C) healthy condyle and (D) lesion site specimen. (E) Representative preprocessed high wavenumber range spectra of a cartilage specimen. Area-under-curve depictions of CH₂ and OH bands for a (F) healthy condyle and (G) lesion site specimen.

where the derived regression coefficients (GAG_score, COL_score, H₂O_score) represent the relative contribution of the predominant constituents of cartilage ECM (GAG, COL, H₂O) to the composite cartilage Raman spectra, calculated using the spectra of purified reference chemicals for each ECM constituent: GAG_REF=chondroitin sulfate (shark cartilage; Sigma[C4384]); COL_REF=type-II collagen (chicken sternal cartilage; Sigma[C9301]); H₂O_REF=PBS (Fig.2B-D). For each constituent, the Raman score ranges from a value of zero to one based on the constituent’s weighted contribution to the tissue spectra. The high wavenumber range (2700-3800cm⁻¹) was preprocessed using a linear baseline subtraction, area-under-curve normalization, followed by measurement of the area under the organic-content-associated CH₂ region (CH_area) and water-content-associated OH region (OH_area) (Fig.2E-G), akin to Unal et al.²¹ To determine the sensitivity of Raman spectra to integration time, for bovine specimens, spectra were further serially acquired at 2-, 5-, and 10-second integration times. Repeatability tests were performed to assess the precision of Raman-derived biomarker scores (GAG_score, COL_score, H₂O_score). Raman spectra were further measured on bovine specimens under varying compressive tare strains to determine the effect of tissue deformation on Raman biomarker scores. Methodological details of these validation tests are described in Supplementary Materials.

Quantitative MRI

T2 and T2* relaxation times were obtained on the intact bovine and human specimens using a 9.4T Bruker BioSpec MRI scanner, mapped for the full cartilage thickness and co-registered to the position of each excised osteochondral specimen (Fig.3). T2 and T2* relaxation times increase with tissue porosity and collagen integrity loss, both of which reflect cartilage degeneration.^47,48 T2 and T2* imaging settings and sequences are described in Supplementary Materials.

Figure 3: MRI analysis.

(A&F) Set-up for MRI measurement of bovine condyles and human tibia plateaus (43-year-old/F) respectively. (B, G, D, I) First echo images from multi-echo sequences used for T2* and T2 mapping for bovine and human specimens. The dashed line represents the border of the osteochondral blocks. (D&I) Angle (θ) between the normal to the cartilage surface of each block and the direction of main magnetic field (B₀) to exclude cartilage regions influenced by the magic angle (54.7°) from the T2 and T2* quantification in the evaluated cartilage block (orange rectangle) (C, H, E, J) Color-coded T2* and T2 relaxation time maps calculated pixel-by-pixel using mono-exponential least square fitting routine in Matlab. Color-coded values in T2* and T2 maps are in milliseconds.

Outerbridge score

Outerbridge classification of chondromalacia was based on direct visualization of the articular cartilage augmented by magnified macrophotography images.³⁴ Using a superimposed grid corresponding to subsequent sectioned osteochondral tissue blocks, a grade of 0 through IV was assigned to the chondral area of interest: Grade 0=normal cartilage; Grade I=no break in surface integrity but softened tissue when pushed with a probe; Grade II=partial-thickness defect with fissures ≤0.5 inches in diameter that do not reach subchondral bone; Grade III=cartilage fissuring with an area >0.5 inches that reach subchondral bone; Grade IV=cartilage erosion that exposes subchondral bone.

Mechanical property analysis

A stress relaxation test was conducted using a ⊘3mm spherical indenter displaced 100-microns into the center of the cartilage layer of each osteochondral specimen at a 2.5mm/s velocity, followed by 180s relaxation. The elastic modulus of the chondral layer was derived from the slope of the force-displacement curve during loading, assuming Hertzian contact.³⁵

Biochemical analysis

For bovine specimens, a full-thickness, central ⊘3mm chondral core was excised and diametrically halved. One half of the core was proteinase-K digested; GAG and collagen contents were derived using the 1,9-dimethylmethylene blue (DMMB) and orthohydroxyproline (OHP) assays, respectively. For human specimens, to directly compare tissue GAG content within the topmost cartilage zone interrogated by the Raman probe, owing to the tissue’s more pronounced depth-dependent heterogeneity, the DMMB GAG assay was performed exclusively on the topmost ~300um of chondral tissue, akin to our prior work.³²

OARSI score

The other half of the excised ⊘3mm cylindrical core was formalin fixed, paraffin embedded, sectioned, and stained with safranin-O/fast-green. OARSI scores were determined as the product of grade (0-6) and stage (0-4) by a board-certified pathologist for a random subset of all analyzed tissue specimens (Table.S5,S6).

Statistical analysis

All measured and derived parameters are presented as mean±standard deviation (SD). Statistical significance was established at a two-sided p-value <0.05. Comparisons among groups were examined using one-way ANOVA and reported as mean difference (MD) and 95% confidence interval (CI). Exponential regression analyses were performed to investigate the extent that the variation of individual Raman derived ECM biomarkers (GAG_score, COL_score, H₂O_score, OH_area) and multivariate linear biomarker combinations accounted for the variation of cartilage composition (GAG, collagen, water contents) and elastic modulus. For comparison, regression analyses were performed to determine the extent that MRI T2 and T2* relaxation times and pathoanatomy grades (Outerbridge, OARSI) accounted for the variation in cartilage composition and modulus. Linear regression analyses were used for MRI versus modulus correlations for consistency with prior studies.^49,50,51 Power analysis of sample sizes is described in Supplementary Material.

Results

Raman assessments of bovine cartilage

Compared to HC, both the LS and the adjacent LPS cartilage were degraded compositionally (Fig.4): GAG (p<0.001) and collagen contents (p<0.001) were reduced, while H₂O content increased (p<0.002), indicative of increased porosity (Fig.4A). Raman spectral peak wavenumbers of cartilage specimens were similar to those reported in our prior work (Fig.S1).³² Multivariate decomposition models fit to the Raman spectra, where the regression coefficients (GAG_score, COL_score, H₂O_score) represented the relative contribution of each ECM constituent to the composite cartilage Raman spectra, accounted for 91±1.5% of the variation of the fingerprint region. Raman spectroscopy identified similar changes in cartilage GAG composition: GAG_score (Fig.4B) was decreased for both LS (MD=0.096, CI:0.074,0.119, p<0.001) and LPS (MD=0.076, CI:0.054,0.098, p<0.001) cartilage compared to HC, and the GAG_score described 78% of the variation of GAG content for all specimens (R²=0.78; p<0.001; Fig.4C). These alterations in ECM composition resulted in significant softening of both LS (MD=1.331, CI: 1.07, 1.58, p<0.001) and LPS (MD=1.15, CI:0.09,1.40, p<0.001) tissue compared to HC tissue (Fig.4D). Changes in the Raman derived ECM biomarkers accounted for changes in the material properties of the degraded cartilage (Figs.4E,F): GAG_score accounted for 71% of the variation in elastic modulus (R²=0.71; p<0.001), as did a linear combination of GAG_score and OH_area describing 72% of the variation in modulus (R²= 0.72; p<0.001). Raman COL_score did not significantly correlate with collagen content or modulus (p>0.56; Table S7). While apparently contradictory to the increase in water content observed for both LS and LPS cartilage, the water-associated Raman biomarkers (H₂O_score and OH_area) decreased for both LS (H₂O_score: MD=0.024, CI:0.008,0.041 & OH_area: MD=0.103, CI:0.05,0.14, p<0.001) and LPS (H₂O_score: MD=0.019, CI:0.002,0.035 & OH_area: MD=0.07, CI: 0.02,0.12, p<0.001) specimens compared to HC specimens, and the COL_score slightly increased (Fig.4B). This paradoxical outcome reflects deformation of the softened cartilage resulting from the tare compressive load applied by the probe’s lens to the tissue during spectral acquisition: the compressive strain induced efflux of interstitial water (described below) decreases the relative contribution of the H₂O_score to the composite cartilage Raman spectra profile.

Figure 4: Raman biomarker correlations with bovine cartilage properties.

(A) Cartilage composition: Biochemical assay measured GAG content, collagen content, and water content of bovine specimens from HC, LPS, and LS cartilage tissue regions. (B) Raman biomarkers: Raman GAG_score, COL_score, H₂O_score, and OH_area of bovine specimens. (C) Bivariate linear regression between GAG content versus Raman GAG_score. (D) Elastic modulus of bovine specimens. (E) Bivariate regression between elastic modulus versus Raman GAG_score. (F) Multivariate regression between elastic modulus versus a linear combination of Raman biomarkers (GAG_score and OH_area).

Raman biomarker reproducibility and sensitivity to integration time and tissue deformation

Repeated measures analysis of bovine cartilage specimens demonstrated that Raman biomarker scores were highly reproducible as evidenced by low coefficients of variation (CV): GAG_score CV=8.2%; COL_score CV=1.8%; H₂O_score CV=19.4%; OH_area CV=1.7% (Fig.5A). To determine the sensitivity of Raman derived biomarkers to integration time, Raman spectra were acquired over 2-, 5-, 10-, and 30-second integration times (Fig.5B, Fig.S2). Biomarker scores were insensitive to spectral integration time (Fig.5C, Fig.S3,S4). For bovine specimens, there was no significant difference (p=0.71) between the coefficients of determination for regressions relating elastic modulus to a linear combination of Raman biomarkers GAG_score and OH_area (Fig.5C) acquired over a 2-second (R²=0.82) compared to a 30-second integration time (R²=0.71). Acquisition of Raman spectra within 2 seconds is requisite for clinical implementation of real-time arthroscopic Raman assessment of cartilage composition. Raman associated water scores (H₂O_score & OH_area) decreased linearly with displacement applied by the probe tip to the chondral surface (Fig.5D).

Figure 5: Raman biomarker repeatability and sensitivity to integration time and tissue deformation.

(A) Raman GAG_score, COL_score, H₂O_score, and OH_area for healthy cartilage specimens (n=10 bovine osteochondral blocks) obtained with four independent spectral acquisitions. (B) 2D stacked area plots depicted cumulative contribution of GAG, COL, and H₂O, to composite Raman fingerprint spectra obtained with a 2 or 30 second spectral integration time for bovine cartilage specimens. (C) Bivariate regression between elastic modulus versus a linear combination of Raman biomarkers (GAG_score and OH_area) obtained with 2 or 30 second integration time for bovine cartilage specimens. (D) Effect of applied displacement of the Raman probe into the chondral surface on water-associated Raman biomarker scores (H₂O_score and OH_area) (n=10 bovine osteochondral blocks).

Raman assessments of human cartilage

Tissue composition and stiffness of human cartilage (Fig.6A&S5) depended on donor age and anatomic location (i.e., medial vs. lateral condyle). Raman probe spectroscopy revealed higher GAG_score for stiffer tissues (Fig.6A & Fig.S5). GAG_score accounted for 60% of the variation in GAG content (Fig.6B; R²=0.60; p<0.001) and 60% of the variation in elastic modulus (Fig.6C; R²=0.60; p<0.001). A linear combination of Raman-derived biomarkers GAG_score and OH_area accounted for 71% of the variation in modulus (Fig.6D; R²=0.71, p<0.001).

Figure 6: Raman biomarker correlations with human cartilage properties.

(A) Spatial maps of elastic modulus, Raman GAG_score, and OH_area over femoral condyle surfaces. (B) Bivariate linear regression between Raman GAG_score versus GAG content. (C) Bivariate regression between Raman GAG_score versus elastic modulus. (D) Multivariate regression between a linear combination of Raman biomarkers (GAG_score and OH_area) versus elastic modulus.

Raman vs macro- and micro-scopic assessment of cartilage pathoanatomy

Outerbridge grade accounted for only 5% of the variation in modulus of human specimens (Fig.7A; R²=0.05; p<0.001). Emphasizing the relative insensitivity of macroscopic assessments of cartilage health, for the subset of Outerbridge 0 human specimens, Raman ECM biomarkers accounted for 70% of the variation in GAG content (Fig.7B; R²= 0.70, p<0.001) and 74% of the variation in modulus (Fig.7C; R²= 0.74, p<0.001); a linear combination of GAG_score and OH_area accounted for similar variation in modulus (R²=0.75; p<0.001; Fig.7D). For bovine specimens, OARSI scores varied according to the injury group: 3.0±3.4 for HC, 5.7±2.0 for LPS, and 7.4±3.6 for LS (Fig.7E). For human specimens, OARSI scores ranged from 0 to 18 (Fig.7F). OARSI score accounted for 54% of the variation of bovine tissue modulus (Fig.7G; R²=0.54; p<0.001) but only accounted for 12% of the variation of human tissue modulus (Fig.7H; R²=0.12; p<0.001). OARSI score accounted for 38% of the variation in bovine tissue GAG content (Fig.S6; R²=0.38; p<0.001) but did not significantly correlate with GAG in human tissue (Fig.S6; p=0.25). In comparison, a linear combination of Raman-derived biomarkers accounted for 86% of the variation in modulus for bovine tissue (Fig.7I; R²=0.86; p<0.001) and 70% of the variation in modulus for human tissue (Fig.7J; R²=0.70; p<0.001).

Figure 7: Raman biomarkers correlate with elastic modulus better than Outerbridge and OARSI scores.

(A) Bivariate linear regression between elastic modulus versus Outerbridge grade for human tissue specimens. Regression between (B) GAG content versus Raman GAG_score , (C) elastic modulus versus Raman GAG_score, (D) elastic modulus versus a linear combination of Raman biomarkers (GAG_score and OH_area) for the subset of Outerbridge 0 grade human tissue specimens. Representative safranin-O/fast green stained sections and corresponding OARSI scores for (E) bovine (HC, LPS, LS), and (F) human tissue specimens. Bivariate regression between elastic modulus versus OARSI score for (G) bovine and (H) human specimens. Multivariate regression between elastic modulus versus a linear combination of Raman biomarkers (GAG_score and OH_area) for (I) bovine and (J) human specimens.

Raman vs MRI assessments of cartilage properties

MRI relaxation times moderately correlated with cartilage tissue properties. For bovine specimens, T2* & T2 relaxation times (Fig.8A-B) described 27% and 30% of the modulus variation (p<0.001), respectively. In comparison, Raman-derived biomarkers better correlated with tissue properties: GAG_score accounted for 70% of the modulus variation (R²=0.70; p<0.001) for bovine specimens (Fig.8C); a linear combination of GAG_score and OH_area accounted for similar modulus variation (R²=0.72; p<0.001; Fig.8D). For human specimens, T2* & T2 relaxation times (Fig.8E,F) described 15% and 25% of modulus variation, respectively (p<0.001). For human specimens, Raman GAG_score accounted for 67% of the modulus variation (R²=0.67; p<0.001) (Fig.8G); a linear combination of GAG_score and OH_area accounted for similar modulus variation (R²=0.71; p<0.001; Fig.8H). For early-OA-associated human specimens, T2* & T2 described 15% and 45% of the modulus variation (p<0.001; Fig.8I,J). For these specimens, GAG_score accounted for 82% of the modulus variation (R²=0.82; p<0.001; Fig.8K); a linear combination of GAG_score and OH_area accounted for similar modulus variation (R²=0.84; p<0.001; Fig.8L).

Figure 8: Raman biomarkers correlate with elastic modulus better than compositional MRI.

Bivariate linear regression between elastic modulus versus (A) T2* relaxation time, and (B) T2 relaxation time for bovine tissue specimens. (C) Bivariate linear regression between elastic modulus versus Raman GAG_score for bovine specimens. (D) Multivariate regression between elastic modulus versus a linear combination of Raman biomarkers (GAG_score and OH_area) for bovine specimens. Bivariate linear regression between elastic modulus versus (E) T2* relaxation time, and (F) T2 relaxation time for all human tissue specimens. (G) Bivariate linear regression between elastic modulus versus Raman GAG_score for all human tissue specimens. (H) Multivariate regression between elastic modulus versus a linear combination of Raman biomarkers (GAG_score and OH_area) for all human tissue specimens. Bivariate linear regression between elastic modulus versus (I) T2* relaxation time, and (J) T2 relaxation time for human specimens from early OA joints (mean Outerbridge < 2). (K) Bivariate linear regression between elastic modulus versus Raman GAG_score for human specimens from early OA joints. (L) Multivariate regression between elastic modulus versus a linear combination of Raman biomarkers (GAG_score and OH_area) for human tissue specimens from early OA joints.

Discussion

This work demonstrates the potential of our Raman probe as an arthroscopic diagnostic platform for assessing articular cartilage composition and material properties. Multivariate decomposition models fit to the Raman spectra, where the regression coefficients (GAG_score, COL_score, H₂O_score) represented the relative contribution of each ECM constituent to the composite cartilage Raman spectra, account for 78% of the variation in assay-measured GAG content and 71% of the variation in elastic modulus for bovine cartilage specimens afflicted with focal chondral lesions that emulate osteochondral defects observed clinically in human femoral condyles. For human specimens, Raman probe spectroscopy reveals higher GAG_score for stiffer tissues, where the GAG_score accounted for 60% of the variation in GAG content and 60% of the variation in elastic modulus. A linear combination of Raman-derived biomarkers GAG_score and OH_area accounted for 71% of the variation in elastic modulus. In the future, advanced spectral analysis can be employed that reflect additional attributes of hyaline cartilage ECM composition and structure (e.g., GAG subtypes, non-collagen proteins, collagen crosslinks^36,37 and fiber alignment^33,38,39), which further reflect essential attributes of cartilage tissue function.

Contradictory to the increase in water content observed for degenerated cartilage^40,41 the water-associated Raman biomarkers (H₂O_score and OH_area) decrease for both LS and LPS bovine specimens, and the COL_score slightly increases. We attribute this paradoxical outcome to spectral artifacts that result from tissue deformation as the spherical probe lens contacted the cartilage. Probe-induced compression of the cartilage surface induces interstitial fluid efflux, leading to a localized decrease in tissue water content immediately below the spherical probe lens. The water biomarker score depends on a combination of the intrinsic (strain-free) interstitial water content and the probe-induced loss of free water, which are contingent on the magnitude of the compressive tare load applied by the Raman ball lens and tissue stiffness. In effect, the Raman probe acts as a strain transducer where the change in water biomarker scores is proportional to the applied tissue strain (Fig.5D). In response to a nominal tare load, the softer, more degenerated specimens experience higher strain, contributing to a pronounced decrease in water biomarker scores. In the current analysis, the mechanically sensitive OH_area score contributes to improved predictions of tissue elastic modulus. In the future, probe contact pressure will be calibrated for standardization during clinical measurements.

MRI is the diagnostic standard for clinical assessments of cartilage composition. T2 and T2* relaxation times portray the water content, collagen content, and collagen fiber alignment of cartilage. However, similar to others,^10,42 we observed that T2 and T2* relaxation times account for only 15%-30% of the variation in elastic modulus. The Raman-derived biomarkers significantly outperform MRI, accounting for more than 70% of the variation in elastic modulus for human and bovine specimens. A standing benefit of MRI is its non-invasive nature and ability to provide 3D compositional images of cartilage. However, Raman probes can be interfaced with arthroscopic image guidance, allowing for cartilage assessments at discrete anatomic sites for the generation of 2D maps of tissue composition. Further, depth selectivity of Raman measures in cartilage can be achieved through modification of the probe lens configuration.³²

A notable outcome of this work is the capability of Raman biomarkers to reveal degeneration in tissue regions where macroscopic degeneration is not visually evident. Macroscopic visual assessment of tissue pathology using Outerbridge classification of chondromalacia poorly predicts cartilage material properties (Fig.7A). Raman-derived biomarkers reveal depleted GAG content and cartilage softening in the seemingly healthy tissue peripheral to a discrete chondral lesion, indicating that tissue pathology extends beyond the margins of the apparent lesion. For the subset of Outerbridge 0 specimens, Raman biomarkers account for 70% of the variation in GAG content and 74% of the variation in modulus. Exposed to catabolic cytokines post-injury, the ostensibly normal tissue peripheral to the lesion has undergone GAG depletion, but breakdown of the collagen matrix has yet to manifest. Excision and treatment of only the discrete lesion using OATS or ACI may result in suboptimal clinical outcomes, since the normal appearing peripheral tissue is also pathologic.

OARSI scores are the ‘gold standard’ for ex-vivo characterization of cartilage pathology, predicated on the assumption that the depth of cartilage degeneration reveals the biological progression of OA; involvement of deeper zones represents more severe disease phases. While a variety of tissue characteristics are assessed, proteoglycan content and cartilage structural morphology are features indicative of cartilage integrity. Parameters assessed and weighted include cell morphology, GAG and collagen staining; structural integrity, surface regularity, morphology of osteochondral junction and subchondral bone. While these features provide a comprehensive assessment of tissue pathology, they are poorly predictive of cartilage material properties.^18,43 OARSI scores accounted for only 12% and 54% of the variation in elastic modulus for ex-vivo human and bovine specimens, respectively, consistent with ranges established in prior studies.^44,45,46 Raman ECM biomarkers outperformed OARSI histopathology, accounting for 70% and 86% of the modulus of ex-vivo human and bovine specimens, respectively. While OARSI scoring was performed by a board-certified pathologist with extensive expertise in musculoskeletal pathology, the lack of multiple independent scoring assessments constitutes a limitation of the study.

This work supports the future use of probe-based Raman spectroscopy as a transformative diagnostic platform for assessing hyaline cartilage biochemical composition. A limitation of the study is that spectral analysis was performed ex-vivo without additional complications that may occur in the surgical setting. However, cartilage Raman spectra are reliably acquired over a 2-second integration time without any sensitivity loss, suggesting that the technique may provide a real-time, non-destructive, assessment of tissue composition during arthroscopy. Recently, we utilized our Raman probe for in-vivo serial monitoring of neocartilage repair tissue in response to microfracture in an equine trochlea.⁵² In the future, our probe can be modified with a longer tip and 30°/70° angled lenses to provide greater access to joint tissues during arthroscopy. Looking towards future clinical translation, Raman probe evaluations may further be implemented as an office-based procedure, akin to conventional in-office needle arthroscopy.⁵³

In the future, Raman probe assessments can further be applied across the evolving hierarchy of model systems—from in-vitro studies on explanted tissues to in-vivo preclinical animal studies to randomized controlled clinical trials. For preclinical models, as Raman assessments are non-destructive, cartilage tissue can be sampled repeatedly over time allowing for longitudinal monitoring of the evolution of tissue changes, thus minimizing inter-specimen biologic variability and allowing for reduced sample sizes for hypothesis testing. For clinical trials, Raman can serve as a primary endpoint measure to evaluate the efficacy of a drug therapy or cartilage regenerative platform to restore tissue composition. In the future, Raman probe analysis can provide standardized objective compositional biomarkers that can be applied across tissue models for standardization of tissue assessments for cartilage research and treatment development.