Approaches for In Situ Monitoring of Matrix Development in Hydrogel-Based Engineered Cartilage

Abstract

Near infrared (NIR) spectroscopy using a fiber optic probe shows great promise for the nondestructive in situ monitoring of tissue engineered construct development; however, the NIR evaluation of matrix components in samples with high water content is challenging, as water absorbances overwhelm the spectra. In this study, we established approaches by which NIR spectroscopy can be used to select optimal individual engineered hydrogel constructs based on matrix content and mechanical properties. NIR spectroscopy of dry standard compounds allowed identification of several absorbances related to collagen and/or proteoglycan (PG), of which only two could be identified in spectra obtained from hydrated constructs, at ∼5940 and 5800 cm⁻¹. In dry sample mixtures, the ratio of these peaks correlated positively to collagen and negatively to PG. In NIR spectra from engineered cartilage hydrogels, these peaks reflected higher collagen and PG content and dynamic modulus values, permitting the differentiation of constructs with poor and good matrix development. Similarly, the increasing baseline offset in raw NIR spectra also reflected matrix development in hydrated constructs. However, weekly monitoring of NIR spectra and the peaks at ∼5940 and 5800 cm⁻¹ was not adequate to differentiate individual constructs based on matrix composition. Interestingly, changes in the baseline offset of raw spectra could be used to evaluate the growth trajectory of individual constructs. These results demonstrate an optimal approach for the use of fiber optic NIR spectroscopy for in situ monitoring of the development of engineered cartilage, which will aid in identifying individual constructs for implantation.

Impact statement

A current demand in tissue engineering is the establishment of nondestructive approaches to evaluate construct development during growth in vitro. In this article, we demonstrate original nondestructive approaches by which fiber optic NIR spectroscopy can be used to assess matrix (PG and collagen) formation and mechanical properties in hydrogel-based constructs. Our data provide a cohesive molecular-based approach for in situ longitudinal evaluation of construct development during growth in vitro. The establishment of these approaches is a valuable step toward the real-time identification and selection of constructs with optimal properties, which may lead to successful tissue integration upon in vivo implantation.

Introduction

Articular cartilage has limited capacity to self-heal when damaged by trauma or disease, due in large part to the lack of blood vessels, nerves, and a lymphatic system.¹ Tissue engineering is a potential regenerative alternative to the primary clinical interventions in use today for cartilage repair.^2–4 Cartilage tissue engineering entails the in vitro development of a “construct,” typically composed of cells, extracellular matrix (ECM), and a scaffold material, which is then surgically implanted into a defect. The intention is to integrate the construct with surrounding healthy tissue, followed by eventual tissue regeneration and healing.

Many approaches for tissue engineering cartilage have been investigated, including methods that explore biomaterials, acellular and cell-laden constructs, and the inclusion of time-released growth factors, all with the final goal of generating a construct that once implanted will restore tissue function.^5–7 Hydrogels have exhibited particular promise as a biomaterial for cartilage tissue engineering due to their viscoelastic properties that partially resemble the properties of native cartilage and their ability to support cell viability and subsequent production of cartilaginous ECM.^8,9

The matrix of native articular cartilage is primarily composed of water (60–85 wt.% of wet weight), type II collagen (10–30 wt.% of wet weight), and proteoglycans (PGs) (e.g., chondroitin sulfate [CS]) (10–15 wt.% of wet weight), with other noncollagenous proteins and glycoproteins in lesser amounts.^1,10 Evaluation of matrix development in tissue engineered cartilage constructs is a reasonable approach to verify the “readiness” of a construct before implantation, which could improve the potential for construct integration and successful regeneration of the damaged tissue.

Although the evaluation of constructs within a group of developing samples can yield information about the average properties of the group, this does not yield information about the properties of specific individual constructs, which can vary substantially.¹¹ Construct development is heterogeneous and, typically, results in a range of matrix properties, even among constructs cultured in the same conditions. Consequently, having the ability to evaluate longitudinal matrix development in individual constructs with a nondestructive method would be advantageous and would allow the selection of constructs with optimal development.

Gold standard methods for analyzing matrix development in engineered cartilage constructs are generally destructive. End point assays such as biochemical and histological methods require sacrificing a sample that cannot be used for future implantation. Furthermore, these methods rely on the assumption that the properties of the sacrificial construct are very similar to the properties of the construct for implantation, without consideration of variation among individual constructs. Thus, investigation of alternative nondestructive methods for assessment of construct development is desired.

Fourier transform infrared spectroscopy has been extensively used to investigate the chemical composition of articular cartilage in both mid infrared (MIR) (4000–400 cm⁻¹) and near infrared (NIR) (12,500–4000 cm⁻¹) regions of the electromagnetic spectrum.^12–16 Chemical bonds within molecules vibrate at specific frequencies in the infrared spectral region, resulting in spectra composed of absorbance bands arising from specific components of samples.¹⁷ In the MIR region, the absorbance bands arise from fundamental vibrations of molecular bonds and are typically strong and specific; the NIR bands are often weaker, broader, and overlapping, arising from combination (5300–4000 cm⁻¹) and overtone (first overtone: 7100–5300 cm⁻¹, second overtone: 9000–7100 cm⁻¹) from O-H, N-H, C-O, and C-H vibrations.¹⁸

Conversely, the penetration depth of MIR is limited to a few microns, which makes it a poor choice to evaluate three-dimensional samples, such as tissue engineered cartilage constructs. An important advantage of NIR is that it exhibits a deeper depth of penetration into the sample. In cartilage, NIR radiation has been shown to reach a depth of ∼1–2 mm in the 5100–4000 cm⁻¹ frequency range, up to ∼3 mm in the 7000–5100 cm⁻¹ range, and up to ∼5 mm in the 9000–7000 cm⁻¹ range.¹⁹ Considering that average human articular cartilage is about 1–3 mm thick and that engineered cartilage constructs for implantation are of similar thickness, information from the entire construct depth could be obtained with NIR spectral analysis.

NIR spectroscopy has been previously used to nondestructively monitor matrix development in tissue engineered cartilage constructs,^14,20,21 and when coupled to a fiber optic probe, this approach can provide a portable nondestructive modality to monitor construct development in situ.

However, NIR spectral analysis also has challenges. In particular, using NIR spectroscopy to evaluate matrix formation in samples with high water content is not straightforward, as the absorbances from water often overwhelm the spectra and obscure matrix absorbances, including those of collagen and PG.^22,23 Hyaluronic acid hydrogels used for cartilage tissue engineering absorb large amounts of water and can result in 1000-fold expansion of the original volume.²⁴ Thus, NIR assessment of hydrated hydrogel-based engineered cartilage can be particularly complex.

In a recent study, a NIR-based multivariate method was developed for assessment of hydrogel-based TE bovine cartilage constructs, which produced reasonable results for assessment of harvested constructs at experimental end points.¹³ However, assessment of individual constructs during growth was not performed, and thus, the sensitivity of that method to incremental changes in matrix during construct development is not known.

Accordingly, the goal of the current study was to identify features of NIR spectra obtained from hydrogel-based engineered constructs that could be used to assess individual hydrated construct development in real time. This was achieved with a three-part study design.

First, diffuse reflectance NIR analysis was performed on dried samples of articular cartilage and standards (pure and mixtures) of the primary matrix components of cartilage, collagen, and PG and a hydrogel scaffold, methacrylated hyaluronic acid (MeHA), to define spectral absorbances that were unique to matrix components; Second, diffuse reflectance NIR spectra of hydrated native cartilage and tissue engineered cartilage constructs were obtained and compared to spectral data from the dry samples, to assess which molecular absorbances from matrix components were still visible in the spectra from hydrated samples; And third, diffuse reflectance NIR spectra were obtained from hydrated samples with differences in matrix composition and analyzed based on spectral parameters defined in these initial studies.

Together, these data provide a cohesive molecular-based approach for in situ longitudinal assessment of hydrated construct development. The results provide a strong foundation for a nondestructive modality to distinguish individual constructs based on their developing matrix properties during in vitro growth.

Materials and Methods

Standard samples

Dry standards

Dry standards of pure CS (Sigma), the primary PG in collagen, collagen type I (Col) (Sigma) MeHA (prepared as previously described²⁵), and mixtures at varying concentrations CS–MeHA (10–90%, 25–75%, and 50–50%), Col–MeHA (10–90%, 25–75%, and 50–50%), and CS–Col–MeHA (20–40–40%, 33–33–34%, and 50–25–25%) were prepared as concentrated pellets under high pressure. Each component was lyophilized for 24 h before making pellets. Although type II collagen is the primary collagen in cartilage, type I collagen was used here due to its availability and significant lower cost, which were critical for the quantities required in this study.

Furthermore, we confirmed in an earlier study that the NIR spectra of type I and type II collagen were indistinguishable.²³ In addition, lyophilized immature bovine cartilage was ground and prepared as a pellet. To make pellets of uniform thickness, 100 mg of the pure and mixture proteins were evenly distributed in a 13 mm diameter pellet dye and pressed under high pressure (1000 lb. for 1 min) using a pellet press (Carver, Wabash, IN). At least three replicate pellets were prepared for each protein and mixture of proteins.

Hydrated samples

Hydrated samples of native cartilage were used as a reference sample for the tissue-engineered constructs. Intact pieces of cartilage with 6 mm diameter and 2 mm thickness were harvested from immature bovine stifle joints (Research 87). Before analysis, excess water was removed by contact with an absorbent paper. The NIR spectrum of pure water was obtained from a 100 μL deionized (DI) water drop. Gels with different gelatin contents were also used as standards. Hydrogels with 1–16 wt./vol.% of gelatin were prepared by mixing gelatin powder (Sigma) with DI water, followed by cooling overnight at 4°C. At least three sample replicates were analyzed for each sample type.

Methacrylated hyaluronic acid synthesis

MeHA was synthesized by methacrylation of HA as previously reported.^25,26 Briefly, 1 wt./vol.% sodium hyaluronate (65 kDa; Lifecore) solution was prepared in DI water and kept at 4°C overnight. On the next day, methacrylic anhydride (Sigma; ∼20-fold excess) was added dropwise to the HA solution, and the pH was continuously adjusted to 8–9 by adding 5 N NaOH. The methacrylation was carried out for 6 h on ice, and the solution was purified using dialysis (molecular weight cutoff 6–8 kDa) against DI water for 14 washes over 7 days. After dialysis, the final product was lyophilized and stored at −20°C.

To prepare the 2 × MeHA stock solution for cell encapsulation, the lyophilized MeHA was dissolved in phosphate buffered saline (PBS) at 2 wt./vol.%, and photoinitiator Irgacure I2959 (Sigma) was added to the solution at 0.01 wt./vol.% to allow UV-mediated polymerization into hydrogels. The solution was filter sterilized using a 0.2 μm filter (Millipore) before cell encapsulation.

Chondrocyte isolation and culture in MeHA hydrogels

Articular cartilage pieces were harvested from 6-month-old Yorkshire porcine stifle joints obtained from a slaughterhouse (Animal Biotech, Inc.). The tissues were minced using scalpel blades and digested in spinner flasks at 37°C and 5% CO₂ for 18 h. The digestion solution consisted of 275 U/mg type II collagenase (Worthington Biochemical) in 100 mL of Dulbecco's modified Eagle's medium (DMEM; Life Technologies) containing 1% Penicillin-Streptomycin-Fungizone (PSF; Life Technologies). The digest was filtered through a strainer and a 70-μm nylon filter, followed by a 10 min centrifugation at 1000g. The primary chondrocyte pellet was then washed thrice in PBS containing 1% PSF and counted using a hemocytometer.

A cell suspension was prepared in PBS and mixed with the MeHA stock solution in a 1:1 proportion to make a final solution with 60 million cells/mL in 1 wt./vol.% MeHA containing 0.005 wt./vol.% Irgacure I2959 (Sigma). The solution was then transferred into a gel caster between glass plates separated by 2 mm spacers. The solution was exposed to UV light for 15 min using a 265 nm Black-Ray UV lamp (Model no. UVL-56) to induce hydrogel polymerization. Chondrocyte-laden hydrogel constructs were created using 4 mm biopsy punches (Miltex) and placed in glass-bottomed well plates (In Vitro Scientific) to allow NIR collection (glass do not show absorbances in the NIR range).

The constructs were cultured at 37°C and 5% CO₂ in a chemically-modified medium showed to favor chondrogenesis,²⁷ consisting of high glucose DMEM supplemented with 10 ng/mL TGF-β3 (R&D systems), 0.1 M dexamethasone, 50 mg/mL ascorbic acid, 40 mg/mL L-proline, 100 mg/mL sodium pyruvate, 6.25 g/mL insulin, 6.25 g/mL transferrin, 6.25 g/mL selenious acid, 1.25 mg/mL bovine serum albumin (BSA), 5.35 g/mL linoleic acid, and 1% PSF. Cell media was changed twice a week, and constructs were harvested at weekly time points for up to 8 weeks. Three experimental rounds were carried out.

Near infrared spectral data collection

NIR spectra were collected using a Nicolet iS5N spectrometer (Thermo Fisher Scientific) with a fiber optic probe (Remspec Corp). The probe contains a fiber bundle of 3.5 mm in diameter, with 35 individual fibers of 400 μm in diameter (20 for illumination and 15 for collection).

The probe tip was positioned at 2 mm distance from the samples for spectra collection, using a Z-axis adjustable stage (OptoSigma) to set the precise distance. The stage was covered with a reflective foil surface and used as platform for the samples. Spectra were collected in diffuse reflectance mode in the 10,000–4000 cm⁻¹ range at spectral resolution of 16 cm⁻¹. Background spectra of air were collected with 256 co-added scans and sample data, with 128 co-added scans. Typically, a greater number of background spectra are utilized such that any noise in the ratioed sample spectra is known not to arise from background.

For the dry standards, six spectra were collected from each pellet and averaged together. For the hydrated tissue samples, gelatin gels, and tissue engineered cartilage constructs, two spectra were collected for each sample and averaged before processing. To obtain NIR spectra from developing live engineered cartilage constructs, sterile data collection was done inside a biological safety cabinet (Fig. 1) during the media change process, before adding fresh medium to the constructs, so they were minimally hydrated for a maximum of 60 s per construct.

FIG. 1.

Schematic of spectral data collection inside a biosafety cabinet. Color images are available online.

Mechanical analysis of constructs

After NIR data collection, a subset of the constructs was harvested and their thickness and diameter were measured. Unconfined compression tests were performed to obtain dynamic modulus. Briefly, a Bose ElectroForce 3230 (Bose) was utilized with a 1000 g load cell to perform stress relaxation tests at 10% strain for 1000 s, followed by five cycles of sinusoidal loading at 1 Hz and 1% strain. The constructs were kept hydrated during the tests using PBS solution. Data were processed in a custom macro in Microsoft Excel.

Biochemical analysis of construct proteoglycan and collagen content

Constructs were harvested, assessed for gravimetric wet weight, flash frozen in liquid nitrogen, lyophilized for 48 h, and kept at −80°C until analysis. Samples were digested in 1 mg/mL proteinase K (Sigma) solution for 24 h. PG content was determined by the dimethylmethylene blue assay for sulfated glycosaminoglycans,²⁸ and collagen content was analyzed using the chloramine-T assay for hydroxyproline quantification. Results were calculated as wt.% of wet weight.

Spectral data processing and multivariate analysis

NIR spectral data were processed in The Unscrambler X software (CAMO). The scattering noise in the NIR spectra from dry protein pellets was corrected using an extensive multiplicative signal correction method.²⁹ The raw spectra from hydrated samples were smoothed with a 15-point Savitzky–Golay filter to improve signal-to-noise quality. For all NIR spectra, second derivative spectra were obtained followed by a Savitzky–Golay filter with 15 smoothing points and inverted to facilitate visualization of absorbance peaks. The second derivative was obtained to resolve broad absorbances and overlapping bands into sharper peaks, allowing the identification and quantification of components with greater specificity.^30,31

All spectral processing was done to achieve noise reduction while preserving subtle spectral features. For direct comparison of samples, spectra were batch processed using the same parameters. In general, spectra are presented as averages from several sample replicates, unless otherwise noted. Univariate analyses of the spectra were based on second derivative peak intensity (height) and ratios.

Multivariate analysis was done using principal component analysis (PCA), an approach in which the spectra of samples of interest are compared based on multiple frequencies of the spectra, which get reduced to a lower number of independent principal components (PCs).^32,33 The data were plotted based on their PCs and scores, with a cluster separation among spectral groups reflecting similarities and variations in the spectra of the analyzed samples.

For analysis of the NIR data from tissue engineered constructs, spectra from groups of constructs with poor (n = 16) and good (n = 16) matrix development (determined by biochemical analysis of PG content and assessment of dynamic modulus) were batch processed as described above and univariate analysis of peak heights and PCA performed. For the individual constructs that were monitored longitudinally in culture, spectra were obtained from individual constructs at 2, 4, and 8 weeks, and poor (n = 6) and good (n = 6) development was determined based on the 8-week terminal end point biochemical and mechanical properties. Univariate features of the spectra collected from the same individual constructs at week 2, 4, and 8 were quantified and compared at specific frequencies based on data from standard compounds.

Statistical analysis

All the described analyses were repeated with multiple samples (as detailed in the Materials and Methods section) to ensure reproducibility of the results. Statistical analysis to assess differences between measured parameters was done with the t-test, with differences considered significant at p < 0.05. Linear correlations among parameters were assessed using the Pearson correlation (r) with significance at p < 0.05.

Results

The scatter corrected and averaged raw spectra of dry CS, collagen I, MeHA, and cartilage pellets (Fig. 2a) have similar features, with broader overlapping peaks in the region above 7000 cm⁻¹. The combination band region in the NIR (∼4000–5400 cm⁻¹) shows more subtle differences among the spectra from these components (Fig. 2b). Collagen has specific absorbances at 4396, 4596, and 4898 cm⁻¹. Similarly, CS and MeHA both have absorbances at 4312 cm⁻¹. These absorbances are similar to those found in our earlier study where hyaline cartilage-related matrix peaks were identified at these frequencies using data collection in transmittance mode.²³ Although the first overtone (5400–7000 cm⁻¹) NIR region contains broader peaks than the combination band region, matrix peaks are still evident that correspond to collagen and CS content (Fig. 2c).

FIG. 2.

Averaged and scatter corrected raw spectra of CS, cartilage, collagen I, and MeHA pellets (dried) showing matrix peaks in the spectral range of (a) 4000–10,000 cm⁻¹, (b) 4200–5000 cm⁻¹, and (c) 5400–7000 cm⁻¹. CS, chondroitin sulfate; MeHA, methacrylated hyaluronic acid. Color images are available online.

Second derivative processing resolved the broader peaks into more pronounced peaks (Fig. 3a) and further confirmed the absorbance peaks present in raw spectra from collagen in the combination region (Fig. 3b). In the first overtone region, second derivative processing revealed peaks associated with collagen and CS that were not observed in the raw spectra (Fig. 3c). The absorbance peaks at 6402, 6556, 6687 cm⁻¹ arise from collagen absorbances, whereas peaks at 5654 and 6803 cm⁻¹ are associated with MeHA and CS. The peak at 6171 cm⁻¹ arises from MeHA, while the peaks at 5778 and 5927 cm⁻¹ are present in collagen and are shifted to slightly higher wave numbers in CS and MeHA.

FIG. 3.

Second derivative averaged spectra of CS, cartilage, collagen, and MeHA (dried) showing resolved peaks in the NIR spectral range of (a) 4000–7500 cm⁻¹, (b) 4200–5000 cm⁻¹, and (c) 5400–7000 cm⁻¹. NIR, near infrared. Color images are available online.

The PCA analysis helps to understand the contribution of specific frequencies associated with unique matrix and scaffold components. PC-1 separates the collagen and CS content among the samples, with high collagen content on the left (negative) side of PC-1 and high CS content on the right (positive) side of PC-1 (Fig. 4a). Spectra from pellets with mixture of collagen, CS, and MeHA also show the same trend with high collagen containing mixtures on the negative side of the PC-1 and with higher CS containing mixtures on the right side of PC-1, indicating that differences in matrix content are apparent in these dry samples, even in the presence of the scaffold material. Similarly, the score plot indicates that the dry cartilage pellets contain more collagen than the CS as they lie on the left side of PC-1 score plot.

FIG. 4.

PCA of second derivative spectra obtained from pure CS, lyophilized bovine cartilage, collagen, and MeHA pellets and from pellets of mixture of CS and Collagen in MeHA, in the 4000–5000 cm⁻¹ spectral range. (a) Score plot shows the separation of the pure and protein mixtures along PC-1 and PC-2. The pellets containing more collagen are to the left side of the PC-1 plot, while the pellets containing more CS are on the right side of the PC-1. The cartilage pellets contain more collagen than CS as seen in the plot. (b) The loading curve of PC-1 reflects 86% of the separation among the protein spectra, while (c) the loading curve of PC-2 accounts for 8% separation among samples. PCA, principal component analysis. Color images are available online.

The loading curve of PC-1 (Fig. 4b) shows the peaks influencing the separation of pellets along PC-1 with peaks on the positive side (4312, 4474, 4960 cm⁻¹) representing the CS content and on the negative side (4396, 4605, 4898 cm⁻¹) representing the collagen content. Similarly, the PC-2 loading curve separates the MeHA and CS pellets. The variation in the spectra from MeHA pellets is explained primarily by the absorbance at 4497 cm⁻¹ and the spectral variation in the CS pellets by the 4443 cm⁻¹ peak. The 4443 cm⁻¹ peak has been previously reported as a matrix peak correlated with an increase of ECM proteins in developing bovine tissue engineered cartilage over time.¹³

Similarly, the score plot of PCA analysis of second derivative spectra from pure and mixture pellets in the first overtone region (5400–7000 cm⁻¹) shows the distribution of pellets with higher content of collagen and CS along PC-1 and PC-2 (Fig. 5a–c). The separation along PC-1 is primarily influenced by the 5908 and 5762 cm⁻¹ peaks explaining the collagen content. The MeHA peak at 6171 cm⁻¹ is the main contributor to the separation between MeHA and CS pellets as shown by PC-2 loading curve. However, the peaks at 5924 and 5777 cm⁻¹ are also present in CS (Fig. 5c), additionally separating the pure CS pellet and MeHA pellets.

FIG. 5.

PCA of second derivative spectra obtained from pure CS, lyophilized bovine cartilage, collagen, and MeHA pellets and from pellets of mixture of CS and collagen in MeHA in the 5400–7000 cm⁻¹ spectral range. The mixture pellets are named as MeHAMIX: % of collagen–% of CS–% of MeHA (for e.g., MeHAMIX 20–40–40 has 20% collagen, 40% CS, and 40% MeHA). (a) Score plot shows the separation of the pure and protein mixtures along PC-1 and PC-2. (b) The loading curve of PC-1 represents 70% separation among the protein spectra, while (c) the loading curve of PC-2 accounts for 18% separation among samples. Color images are available online.

Even though the 5924 and 5777 cm⁻¹ peaks are present in both collagen and CS, the ratio of these peaks differs in collagen and CS. The second derivative peak height ratio of 5924–5777 cm⁻¹ in the pure and mixture pellets is inversely proportional to the CS content (r = 0.98, Fig. 6a) and directly proportional to the collagen content (r = 0.98, Fig. 6b).

FIG. 6.

Correlation of second derivative peak height ratio 5924/5777 cm⁻¹ to (a) CS gravimetric weight content (mg/mg) (b) collagen gravimetric weight content (mg/mg), p < 0.05. Color images are available online.

In hydrated native cartilage, absorbances from water overwhelm the NIR spectra, obscuring the main absorbance peaks present in the dehydrated samples (Fig. 7a). Based on the second derivative of the spectra, it was also clear that the peaks observed in hydrated cartilage arise primarily from water (Fig. 7b). Interestingly, two minor peaks of collagen and/or PG could still be identified in the hydrated cartilage spectra, at ∼5940 and ∼5800 cm⁻¹, in a region where there is no water absorbance (Fig. 7c).

FIG. 7.

Identification of collagen and PG absorbance peaks in the NIR spectra of hydrated native cartilage. Raw (a) and second derivative (b) spectra. Water absorbances overwhelm the spectra of native cartilage, hiding the main collagen and PG peaks observed in dehydrated cartilage. The peaks from native cartilage arise primarily from water, with minor contributions of collagen and PG. Orange arrows: collagen and PG peaks in dehydrated cartilage. Blue arrows: water peaks in native cartilage. Asterisks: minor peaks observed in both native and dehydrated cartilage. (c) Second derivative spectra. Only two minor peaks of collagen and/or PG can be identified in the spectra of native hydrated cartilage at ∼5940 and 5800 cm⁻¹. PG, proteoglycan. Color images are available online.

To assess whether these peaks could be used for evaluation of the amount of matrix produced in hydrogel constructs, the peaks were initially analyzed in gel standards with varying amounts of protein (gelatin). It was evident that the intensity of the peaks in the second derivative increased with greater protein content (Fig. 8a). In fact, there was a significant positive correlation between protein content and the intensity of the peaks at ∼5940 and ∼5800 cm⁻¹ (Fig. 8b). These results support the possible use of the spectral features to monitor matrix development in hydrated engineered cartilage.

FIG. 8.

The intensity of the NIR peaks at ∼5940 and 5800 cm⁻¹ reflects protein content in hydrated gelatin gels. (a) Second derivative spectra of gels with increasing amounts of protein. (b) There was a very strong correlation (p < 0.05) between the intensity of the peaks at 5940 and 5800 cm⁻¹ and the protein content in hydrated gels. Color images are available online.

Average differences in NIR spectral features in hydrated constructs with different properties

To evaluate the usefulness of this potential NIR approach, constructs were characterized as two groups, “poor” or “good” matrix production, based on quantified PG and collagen content and mechanical properties (Fig. 9a–c). Interestingly, analysis of the second derivative of the NIR spectra showed that the average intensity of the peaks at ∼5940 and ∼5800 cm⁻¹ was elevated in good constructs (Fig. 9d), with PCA illustrating a clear trend to distinguish poor and good constructs based on the 6100–5700 cm⁻¹ spectral range (Fig. 9e). In addition, a marked difference in the overall absorbance offset of the raw spectra was obvious between the poor and good constructs (Fig. 9f), which were further illustrated by PCA (Fig. 9g).

FIG. 9.

Differences in the NIR spectra of engineered cartilage constructs with poor and good matrix development. Constructs with poor and good development were identified based on the production of (a) PG and (b) collagen and (c) mechanical properties. (d) The average matrix peaks at ∼5940 and 5800 cm⁻¹ were more intense in good constructs. (e) PCA shows separation of poor and good construct groups based on the 6100–5700 cm⁻¹ range of the second derivative spectra. (f) Raw spectra. The average absorbance offset was higher in good constructs. (g) PCA shows separation of poor and good construct groups based on the raw spectra. Color images are available online.

Furthermore, quantification of the peak intensity at ∼5800 cm⁻¹ in the second derivative spectra (Fig. 10a) was significantly different between the poor and good groups. However, the difference in the peak intensity at ∼5940 cm⁻¹ was not statistically significant (p = 0.08) (Fig. 10b). In addition, the second derivative 5940/5800 peak ratio was significantly higher in good constructs (Fig. 10c), as was the absorbance offset at 10,000 cm⁻¹ in the raw spectra (Fig. 10d). Together, these results show that NIR spectroscopic approaches can be used to distinguish hydrated constructs with poor and good matrix development, based not only on the second derivative peak intensity at ∼5800 cm⁻¹ but also on the second derivative 5940/5800 ratio and on the absorbance offset of the raw spectra.

FIG. 10.

Quantification of differences in the NIR spectra of engineered cartilage constructs with poor and good matrix development. (a) The intensity of the peak at ∼5800 cm⁻¹ in the second derivative spectra was significantly higher in good constructs. (b) Differences in the peak at ∼5940 cm⁻¹ were not significant (p = 0.08). (c) The second derivative 5940/5800 peak ratio was significantly higher in good constructs, as well as (d) the absorbance offset at 10,000 cm⁻¹ in the raw spectra. Color images are available online.

Differences in NIR spectral features in longitudinal assessment of developing individual constructs in situ

To verify if these approaches could be used to monitor the longitudinal matrix development of individual constructs, NIR spectra from constructs with poor and good matrix development (based on PG and collagen content and mechanical properties at harvest end points of 8 weeks) were obtained and analyzed during development (Fig. 11a–c). Although peak intensity at ∼5800 cm⁻¹ was higher in the good constructs, this peak did not show consistent trends in the same individual constructs monitored over time (Fig. 11d). The same trend was observed with the second derivative 5940/5800 peak ratio (Fig. 11e).

FIG. 11.

NIR spectroscopic approaches to distinguish engineered cartilage constructs undergoing poor and good matrix development. Individual constructs were monitored by NIR spectroscopy longitudinally during development in cell culture. Individual constructs with poor (n = 6) and good (n = 6) development were identified based on the production of (a) PG, (b) collagen, and (c) mechanical properties at week 8. (d) The intensity of the peak at ∼5800 cm⁻¹ was assessed for individual constructs (connected with solid line) and was always higher in the second derivative spectra of all the good constructs, but no clear trend was observed for individual constructs over time in culture. (e) The same was observed for the second derivative 5940/5800 peak ratio. (f) The raw spectra absorbance offset at 10,000 cm⁻¹ showed a marked increase in all the good constructs over time, which was not observed in the constructs with poor matrix development. Color images are available online.

Interestingly, we noticed that the absorbance offset at 10,000 cm⁻¹ in the raw spectra showed a clear progressive increase in constructs undergoing good matrix development, which was far less pronounced in poorly developing constructs (Fig. 11f). Together, these data demonstrate that although, on average, the matrix absorbances at ∼5800 and 5900 cm⁻¹ reflect matrix development, in these constructs, those parameters are not sensitive enough to be useful for monitoring of the development of an individual construct. Alternatively, assessment of increases in the offset of the NIR spectra can be used to assess development of an individual construct.

Discussion

In this study, we demonstrate NIR spectroscopic approaches to nondestructively monitor the longitudinal matrix development of hydrogel-based tissue engineered cartilage constructs in situ. Previous studies from our laboratory have also used NIR spectroscopy to nondestructively evaluate cartilage constructs and have shown correlations with compositional and mechanical properties of developing constructs.^{13–15,21,22} However, the current studies significantly advance this approach by demonstrating the ability to assess matrix development in individual hydrated constructs over time. The results presented here show that NIR spectral features that are less affected by water can be used to distinguish constructs with poor and good matrix development.

We believe this approach can mitigate interference of water absorbances from the hydrogel scaffolds, providing an evaluation more closely associated to the matrix components produced by the cells. We show that the intensity of the peak at ∼5800 cm⁻¹ in the second derivative spectra was higher in constructs with good matrix development, as was the absorbance offset of the raw spectra. Both these spectral features may be useful to indicate matrix development in engineered cartilage constructs. In particular, analysis of changes in the absorbance offset of the raw spectra at 10,000 cm⁻¹, which does not require any spectral processing, may provide a simple and straightforward approach for the longitudinal monitoring of individual constructs during growth in cell culture.

Several earlier studies from our group focused on evaluation of NIR spectra from cartilage and the primary matrix components that comprise that tissue, in particular in the spectral region up to ∼5200 cm⁻¹, contributing increased capabilities for spectral interpretation.^{13,14,22,23,34,35} However, the NIR region of 7000–5400 cm⁻¹ had not been extensively discussed in previous research for describing the matrix components of articular cartilage. This region consists of broader overtone absorbances of OH, NH, CH, and SH vibrations. The depth of penetration of NIR radiation in the 5400–7000 cm⁻¹ is greater than in the lower frequency region, up to ∼3 mm, and has been described as an ideal region for cartilage composition assessment.¹⁹

The matrix peaks (5664, 5778, 5924, 6556, 6688 cm⁻¹) described based on the spectra of dry cartilage and its components provide a foundation for understanding and evaluation of native hydrated articular cartilage. This spectral region has also been explained as first overtone of CH and SH stretching of PGs as it correlated with swelling of cartilage.¹⁶ Unfortunately, the primary collagen and PG peaks observed in dehydrated cartilage could not be clearly identified in the NIR spectra of hydrated native cartilage.

As water can account for up to 80–90 wt.% of the wet weight of cartilage,¹ it is reasonable that the water absorbances in the spectra overwhelmed the matrix absorbances, obscuring underlying collagen and PG peaks. In particular, the presence of a broad water band at ∼5200 cm⁻¹ made it impractical to achieve a reliable resolution of matrix peaks in the 5000–4000 cm⁻¹ range, where the primary peaks associated with collagen and PG have been described.^14,15,23 However, we found that in the 6100–5700 cm⁻¹ range, a region absent of water absorbances, it was possible to identify two minor collagen and/or PG peaks in the spectra of native cartilage at ∼5940 and ∼5800 cm⁻¹. These absorbances arise from C-H stretch first overtone vibrations¹⁸ and can reflect the presence of proteins and/or PGs in the cartilage matrix.

It is also important to discuss a previous study in which a fiber-optic Raman spectroscopy approach was proposed for longitudinal monitoring of matrix development in developing tissue engineered cartilage constructs.³⁶ There the authors described the application of an in-house built Raman system for the nondestructive monitoring of construct growth during cell culture. They show that the Raman spectra collected from the constructs had strong correlations to the biochemical quantification of collagen and PG, highlighting the potential of this approach to the evaluation of matrix development in live constructs.

However, the use of Raman spectroscopy has some limitations compared to NIR. For instance, the Raman instrumentation used in the previous study was in-house built and is not easily accessible for most research groups. In contrast, the NIR spectrometer and fiber optics used here are commercially accessible and can be readily acquired. In addition, the spectral data collection using the previous Raman system requires culture of constructs in phenol-red free DMEM to avoid interference from fluorescence of the dye present in media. This change in media composition was not required for NIR analysis, which allows the use of regular media in which phenol-red is useful for visual monitoring changes in media pH during cell culture.

In addition, the fiber optic probe used for the Raman study has to be in contact with the engineered construct for the spectra collection, which can increase risk of contamination introduced by the probe tip and potentially damage the construct surface. With NIR spectra collection, the spectra can be collected without direct contact of the fiber optic probe with the growing engineered constructs,¹³ resulting in an optimized approach for data collection.

In conclusion, the data presented here demonstrate that features of the NIR spectra can be used to evaluate matrix development in hydrated engineered cartilage constructs. In particular, peaks seen at ∼5940 and 5800 cm⁻¹ in the second derivative of the spectra were found to represent the formation of collagen and PG in the constructs, as well as the improvement of mechanical performance. Interestingly, we noticed that monitoring the baseline offset of the raw spectra at ∼10,000 cm⁻¹ can enable real time assessment of matrix development during construct growth in culture, with no spectral processing required.

The establishment of these nondestructive approaches using fiber-optic NIR spectroscopy is a valuable step toward the in situ identification of individual constructs with optimal matrix properties, allowing selection of specific constructs that may lead to successful tissue integration upon in vivo implantation.

Disclosure Statement

No competing financial interests exist.

Funding Information

Funded by NIH R01 AR056145.