Osteoarthritis (OA) is the leading cause of disability in the US population [1]. OA is characterized in large part by the degeneration of articular cartilage, the load-bearing connective tissue of the synovial joint, leading to severe pain and disability. The load-bearing functionality of healthy cartilage is supported by the presence of a mechanically robust extracellular matrix (ECM), consisting of densely concentrated, large glycosaminoglycan (GAG) aggregates, interspersed within a network of type II collagen fibers. The collagen fibers are highly organized, exhibiting a beneficial depth-dependent zonal anisotropy in healthy cartilage, whereby they are aligned parallel to the articular surface in the topmost superficial zone, isotropic in the middle zone, and perpendicular in the deep zone [2].

While OA progression is generally characterized by the severe erosion of articular cartilage, there is a growing view that early stages of the disease may represent a critical clinical treatment window when disease-modifying therapeutics may be most efficacious at reversing or halting OA progression [3]. The early stages of OA are characterized by relatively subtle tissue changes, predominantly: (1) loss of GAG from the cartilage ECM [4] and (2) disruption/alignment changes of collagen at the cartilage surface [5,6]. However, the current standards for OA diagnosis [radiography, magnetic resonance imaging (MRI), and computed tomography (CT)] are mainly suited for the assessments of pronounced, late-stage OA tissue changes, such as cartilage erosion (joint space narrowing), osteophytes, and subchondral bone cysts. Emerging, state-of-the-art imaging platforms (e.g.,  contrast-based imaging, OCT, ultrasound) may provide improved diagnostic outcomes, but they are associated with numerous intrinsic limitations, including limited imaging resolution, dependence on intravenous delivery of contrast agents, high infrastructure cost, lengthy imaging durations, and low sensitivity to early OA [7,8]. As such, there exists an urgent need for the development of robust, low-cost, and easy-to-implement diagnostic platforms for early OA.

Raman spectroscopy is a vibrational spectroscopic technique with a unique potential for early-stage OA diagnostics [9]. Raman spectroscopy characterizations are label-free, non-ionizing, and non-destructive. Accordingly, they can be used for minimally invasive, intra-articular biochemical analysis of articular cartilage. When an incident laser excites the vibrational modes of molecules, the resulting Raman scattering provides information about the biochemical composition of the sample. In recent work, we have demonstrated that Raman spectroscopic assessments of cartilage can measure the hallmark GAG loss associated with early OA [2]. However, a Raman probe-based platform to measure the collagen alignment changes of early OA does not exist. Previously, a few studies have reported polarized Raman spectroscopic assessments of cartilage [10–12]. Using a microscopy system, we were previously able to determine the collagen fiber orientation, based on polarized Raman spectra, using multivariate analysis techniques [10]. This is because specific vibrational modes depend on the polarization of the incident laser, as well as the orientation of the molecules, enabling the acquisition of structural information. In general, studies into the spectral differences arising from the relative orientation of collagen fibrils to laser polarization have found a clear relationship between relative peak intensities and fibril orientation [13]. While polarized Raman scattering has been used in microscopy settings for thin tissue slides, it has been well accepted that polarized Raman spectroscopy of bulk tissues scrambles the polarization due to diffuse light scattering in the tissue, thereby reducing structural information content [14].

In this Letter, to the best of our knowledge, we demonstrate a novel multiplexed polarized Raman needle probe for the biostructural analysis of cartilage tissue changes that occur in early OA. By tightly focusing the laser light on the tissue surface, we can preserve polarized Raman signals from bulk tissue through a needle probe, which allows us to extract both biochemical and structural information from articular cartilage. As a demonstration of the platform’s diagnostic capacity, we examine the capability of our platform to assess early-OA-like collagen alignment changes in a cartilage tissue model system.

Figure 1(a) shows a schematic of the multiplexed polarized Raman needle probe system. The Raman spectroscopy system consists of a 785 nm laser (B&W Tek BRM-785-0.55-100-0.22-FC, 600 mW) and a high-throughput near-infrared (NIR) imaging spectrometer (Princeton Instruments Acton LS785, 750–1100 nm). For the Raman probe, the laser is coupled through a Glan-laser calcite polarizer (Thorlabs GL10-B, Extinction Ratio = 100,000:1) and a 785 nm MaxLine laser clean-up filter (Semrock LL01-785-25) to polarize and sharpen the laser line before it is directed through the aluminum needle (${{\phi}} = {2}\;{\rm mm}$, ${{l}} = {50}\;{\rm mm}$) and tightly focused onto the tissue using a sapphire ball lens (AWI Industries, 2 mm AR-coated sapphire lens, NA = 0.86). The Raman scattered light reflected from the tissue is passed back through the needle through an 801 nm edge BrightLine single-edge dichroic beamsplitter (Semrock FF801-Di02-${25} \times {36}$), a 785 nm EdgeBasic long-pass edge filter (Semrock BLP01-785 R-25), and into a polarizing beamsplitter (Thorlabs CM1-PBS252, 620–1000 nm, extinction ratio = 1000:1). The Raman signal is split into two separate polarized components, parallel and perpendicular to the laser polarization, and passed through a bifurcated optical fiber (105 µm Thorlabs BFY105LS02). The bifurcated fiber, which serves as the slit, is coupled directly into the spectrometer, and the separation of the fiber cores creates two distinctly separated Raman spectra on the CCD camera [see Fig. 1(b)]. The effective numerical aperture (NA) of the system is determined by the collection fiber, NA = 0.22. A comprehensive software package and graphical user interface (GUI) were developed in the MATLAB scripting environment. The software enables real-time data acquisition and analysis. The two polarized Raman spectra are acquired simultaneously by hardware binning them separately. A premeasured system background Raman spectrum is then subtracted from the tissue Raman spectra. Next, the acquired spectra are intensity calibrated using the NIST 2241 Raman spectra standard reference material, and wavelength calibrated using an atomic lamp (Ocean Optics HG-1). The autofluorescence background is removed using a custom fifth-order polynomial fit function constrained to the lower part of the Raman spectrum [15]. Spectral smoothing is performed using a third-order Savitzky–Golay filter with a window size of 7 pixels. Finally, the spectra are normalized to their integrated area. All pre-processing is performed in real-time with control from the GUI.

 

Fig. 1. (a) Schematic representation of the multiplexed polarized Raman spectroscopy needle probe; (b) snapshot of the CCD image, showing the multiplexed polarization Raman images and binned spectra; and (c) schematic representation of the Raman probe passed through a hypodermic needle. PB, polarization beam splitter; LP, long-pass filter; DB, dichroic beam splitter; BP, bandpass filter; and P, polarization filter.

Download Full Size | PDF

A polarization experiment was performed to ensure that the laser and Raman polarization are maintained throughout the system. Polarized Raman measurements were performed on a silicon wafer. Due to the high crystalline structure of silicon, the corresponding Raman peak intensity is highly dependent on the polarization of the incident laser [16]. A silicon wafer sample was placed perpendicular to the needle probe at an arbitrary azimuth angle. The laser polarizer was adjusted so that rotation of 0 deg corresponded to the parallel laser polarization, relative to the collection fibers. Raman spectra were collected after sequentially rotating the laser polarization 20 deg. Figure 2(a) shows the parallel polarized Raman spectra across a full 360 deg rotation. Figure 2(b) shows a polar plot of the intensity change of the dominant silicon peak at ${520}\;{{\rm cm}^{- 1}}$ across a full rotation. From Figs. 2(a) and 2(b), the oscillating peak intensity demonstrates that the laser polarization is maintained throughout the system.

 

Fig. 2. (a) Parallel polarized Raman spectra of silicon under full rotation of the incident laser polarization (1 s integration time); (b) polar diagram of the ${520}\;{{\rm cm}^{- 1}}$ peak intensity under a full polarization rotation; and (c) depth of penetration on a silicon wafer, in air medium.

Download Full Size | PDF

To determine the depth of field of the system in air, a depth of penetration test was performed on the same silicon sample. The needle was placed on the surface of the sample and displaced with a resolution of 25 µm. We obtained a depth of penetration in air of 325 µm [Fig. 2(c)]. We expect the depth of penetration to be slightly different in tissue and still well suited for targeting early-OA-associated collagen organizational changes at the topmost cartilage layers.

We next implemented a cartilage tissue model system to investigate the capability of the polarized Raman needle probe to detect collagen alignment changes in early OA. A hallmark of early OA is a disruption to the surface-aligned collagen at the cartilage surface [5,6]. As a model, we examined the ability of our probe to differentiate between collagen alignment in the superficial, middle, and deep zones of healthy cartilage explant specimens, which exhibit variation in collagen alignment akin to the surface alignment changes associated with early OA. The superficial zone consists of a mesh of collagen, aligned parallel to the cartilage surface (perpendicular to the optical axis), the middle zone exhibits a mixed collagen orientation, and the deep zone exhibits collagen aligned perpendicular to the cartilage surface (parallel to the optical axis). The polarized Raman signal is expected to exhibit subtle differences in the three zones, due to the changes in the average collagen orientation relative to the optical axis. Here, articular cartilage explants (${{\varnothing} 5} \times {3}\;{\rm mm}$) with the articular surface initially intact were procured from the femoral condyles of randomly selected 2-month-old bovines (${n} = {4}$). Explants were axially trimmed from the cartilage surface via a custom cutting device to expose different tissue zones: (1) the superficial zone [articular surface intact (0 µm excised)]; (2) middle zone [superficial zone removed (topmost 300 µm excised)]; and (3) deep zone [superficial/middle zones removed (topmost 600 µm excised)]. After trimming, explants were subjected to an enzymatic GAG depletion treatment (5 mg/mL hyaluronidase for 24 h) to further mimic the surface GAG loss associated with early OA [4]. Explants were fixed in 3.7% paraformaldehyde and washed in phosphate-buffered saline (PBS) prior to Raman analysis. The Raman needle probe was placed in gentle contact with the tissue, while oriented normal to the surface, and sets of polarized Raman spectra were measured. Figure 3(a) shows polarized Raman spectra of the bovine cartilage explants [mean $\pm \;{1}$ standard deviation (SD)]. A total of 27 explants [superficial zone (${ n} = {9}$), mid zone (${ n} = {9}$), and deep zone (${n} = {9}$)] were measured (${ n} = {5}$ spectra per sample) using a 10 s acquisition time to obtain a very high signal-to-noise ratio (SNR). The power on the sample was 140 mW with a spot size of ${\sim}{300}\;\unicode{x00B5}{\rm m}$, and we observed no degradation of the highly hydrated tissue.

 

Fig. 3. (a) Orientation of fibrils showing the collagen alignment in articular cartilage and polarized Raman spectra $\pm {1}$ standard deviation (SD) of superficial (${ n} = {45}$ spectra), mid- (${ n} = {45}$ spectra), and deep-zone (${ n} = {45}$ spectra) articular cartilage, using the Raman needle probe; and (b) polarization difference (parallel–perpendicular) spectra $\pm {1}$ SD for the three tissue zones.

Download Full Size | PDF

Intense Raman peaks were observed for both polarizations at 861, 930, 1257, 1448, and ${1654}\;{{\rm cm}^{- 1}}$. The peaks tentatively correspond to hydroxyproline, C-C stretch, Amide III, ${{\rm CH}^2}$ bending, and Amide I, respectively. These peaks are due to the collagen that is abundant in cartilage and has been reported in multiple studies [10,11,17]. We found consistent differences between the parallel and perpendicular polarization in all three tissue types, in particular at 861, 930, 1251, and ${1448}\;{{\rm cm}^{- 1}}$. This can be observed in the polarization difference spectra, Fig. 3(b). Most importantly, we also found that there were subtle, but consistent differences between the three zones’ spectra. The peaks observed in the polarization difference spectra have all previously been reported to be related to collagen orientation [10,11].

To take advantage of the full range of polarized Raman peaks, we employed partial least squares discriminant analysis (PLS-DA) on the depolarization spectra [perpendicular/(parallel+n)]; here, $n$ is an arbitrary number used to avoid near-infinite values in the spectra. The size of $n$ is arbitrary and does not affect the comparison and analysis of the zones. We utilized cross-validation (leave one tissue sample out) to determine model complexity of two latent variables (LVs) and then applied the trained model (derived from ${ n} = {6}$ tissue samples from each zone, ${n} = {5}$ polarization spectra from each tissue sample) to an independent test set (${ n} = {3}$ tissue samples from each zone, ${n} = {5}$ polarization spectra from each tissue sample). Figure 4(a) shows the loading for the two LVs from the PLS-DA. LV1 accounted for a total of 42.54% of the variance in the X-Block and 20.34% in the Y-Block. LV2 accounted for a total of 2.79% of the variance in the X-Block and 20.34% in the Y-Block. Several collagen peaks (including hydroxyproline, C-C stretch, Amide III, ${{\rm CH}^2}$ bending, and Amide I) were prominent across the two LVs, strongly suggesting that the two LVs are related to collagen orientation [10–12]. Figure 4(b) shows a scatter plot of the PLS scores for the three cartilage tissue zones.

 

Fig. 4. (a) PLS-DA latent variable (LV) loading of the depolarization ratio Raman spectra showing distinct peaks associated with collagen and (b) PLS-DA score plot showing the separation of superficial and deep-zone cartilage.

Download Full Size | PDF

We found clear discrimination between the superficial versus mid-zone/deep-zone cartilage for both training and test set. Our PLS-DA model could classify the test tissues with sensitivity/specificity of the superficial zone (80.0%/93.3%), mid zone (86.7%/56.7%), and deep zone (66.7%/40%). Not surprisingly, we found that the model is best at discriminating between the superficial zone and the deeper zones. This correlates well with the structure of cartilage, suggesting the system has strong potential for the detection of collagen alignment loss in early OA.

There are limitations of this work to note. For one, collagen structural assessments were performed on a simplified model of the zones of healthy articular cartilage specimens. These assessments serve as a strong foundational model given that collagen alignment variations between the different zones are similar to surface alignment changes observed during early OA [5,6]. However, future investigations into polarization assessments of collagen alignment in early-OA-degenerated human specimens will be needed to confirm the platform’s diagnostic capacity. Second, for this proof of principle, we use a relatively long integration time of 10 s to obtain a sufficient SNR. This can possibly be reduced with further optimization (e.g., a single-mode optical fiber laser, a smaller sapphire ball lens with higher NA, and a more state-of-the-art NIR optimized CCD) so that we could significantly reduce it for clinical applications down to sub-seconds. No indication of polarization scrambling caused by internal reflections in the needle was found.

As mentioned, young bovine samples were used for this study. There are intrinsic differences between human and bovine cartilage, which will have to be taken into consideration for developing a clinical technique. Our previous studies have shown that Raman spectroscopy can detect the GAG loss, which is a hallmark of early OA. Here we demonstrate that a hypodermic needle probe can detect a second major characteristic of early OA: loss of the superficial collagen alignment. In the future, we aim to apply this technology and combine GAG and collagen analysis to offer complementary information for early OA diagnosis in humans. This technology can serve as a research tool in clinical and pre-clinical models, allowing for novel assessments of the efficacy of OA-modifying drug candidates when administered in the early stages of disease progression. Given the relatively low resolution of conventional MRI and CT diagnostic platforms, our Raman needle probe can prove to be transformative to the identification and clinical implementation of OA treatment therapeutics. Further, polarized Raman diagnostics may be applicable to a broad range of diseases, such as cancer (e.g., breast, lymph, oral cancers), where disease progression is marked by collagen structure loss.

In summary, we have built, to the best of our knowledge, a novel, multiplexed Raman needle probe system. The Raman needle probe was tested on a model for OA in articular cartilage tissue. We report a good sensitivity and specificity for detecting subtle structural changes in the superficial zone of an OA tissue model. We have shown that, through a combination of the multiplexed polarization needle probe and the diagnostic model, we can distinguish collagen alignment, which shows great potential for future early OA diagnosis in general.