2.1.

Tissues

This research was carried out following the ethical principles established by the Brazilian Health Ministry and was approved by the local ethics in research committee (certificate H149/2006/CEP—UniVap). Patients were informed concerning the subject of the research and gave their permission for the collection of tissue samples.

Rotator cuff supraspinatus tendon tissue samples were obtained from 39 patients submitted to shoulder surgery at São Gabriel Clinic, São José dos Campos, São Paulo, Brazil. Immediately after the surgical procedure, the samples were identified, snap frozen and stored in liquid nitrogen $\left(77\mathrm{K}\right)$ in cryogenic vials $\left({\mathrm{Nalgene}}^{\circledR }\right)$ prior to Fourier transform (FT)-Raman spectra recording. Each patient’s tissue was sectioned into three parts, resulting in 117 samples. Raman spectra were measured at three different points on each sample, resulting in 351 spectra. Next, the samples were fixed in 10% formaldehyde solution for further histopathological analysis. Each measured sample was histopathologically assessed by two pathologists and classified as G1, G2, G3, or G4, according to the histopathological grading proposed by Riley ¹⁹

2.2.

Reference Compounds

Reference compounds cysteine (Cys), cistine (Cis), proline (Pro), hydroxyproline (Hyp), and collagen types I and III were used to refine the vibrational band assignment. All compounds were of commercial molecular biology grade obtained from Sigma Aldrich (St. Louis, Missouri, USA).

2.3.

FT-Raman Spectroscopy

An FT-Raman spectrometer (Bruker RFS 100/S; Bruker Optics GmbH, Ettlingen, Germany) was used with an Nd:YAG laser at $1064\mathrm{nm}$ as the excitation light source. The laser power at the sample was maintained at $110\mathrm{mW}$ , while the resolution was set to $4{\mathrm{cm}}^{-1}$ . The spectra were recorded using 300 scans (nearly $10\min$ of acquisition time). For FT-Raman data collection, the samples were brought to room temperature and maintained moistened in 0.9% physiological solution to preserve their structural characteristics, then they were placed in a windowless aluminum holder for Raman spectra collection. Observation revealed that the chemical species in the physiological solution ( ${\mathrm{Ca}}^{2+}$ , ${\mathrm{Na}}^{+}$ , ${\mathrm{K}}^{+}$ , ${\mathrm{Cl}}^{-}$ , water) presented no measurable Raman signal and their presence did not affect the tissue spectral signal.

2.4.

Histopathological Analysis

The stains used on the tendon samples were read by two pathologists and compared with Riley’s scale,¹⁹ which comprises four levels of degeneration. Grade 1 [G1, Fig. 1a] corresponds to normal tendons and is characterized by a wavy outline of collagen fibers, easily discernible individual fibers [arrow 1, Fig. 1a], tenocyte nuclei that are elongated and parallel to the collagen bundles [arrow 2, Fig. 1a]. Grade 2 [G2, Fig, 2b ] corresponds to mild degeneration and it is characterized by relatively well aligned collagen fibers presenting irregular waviness [circles 3 and 4, Fig. 1b]. In this case, the individual fibers are not readily identified, tenocyte cell nuclei are shorter, and the nuclei are displayed in Indian file. Grade 3 [G3, Fig. 1c] is moderate degeneration. The onset of the process of collagen hyalinization occurs in this grade. The hyaline state is characterized by the deposition of collagen peptide fragments in the extracellular matrix. Increasing quantities of hydroxylysine, hydroxyglysilpyridine, and lysylpyridioniline due to incomplete healing, result in greater collagen solubility, thus favoring hyalinization.¹³ When stained, this state is characterized by eosinophilic pink regions [squares 1 and 2, Fig. 1c]. Loss of orientation among the fibers and tenocyte nuclei and increases in cell nuclei density are also features of G3. A relative increase in tenocyte numbers [Fig. 1c], compared to lower grades, is clearly seen. Finally, grade 4 [G4, Fig. 1d] is designated as severe degeneration. In this grade, the collagen fibers present a diffuse hyalinization of homogeneous appearance, the tenocytes present a rounded shape, complete loss fiber orientation occurs, and the number of nuclei is reduced.

Fig. 2

Box plot representing all tendon spectral data. Each spectrum was baseline corrected and vector normalized. The solid line is the mean Raman spectra. The vertical gray lines indicate the region between the first and third quartiles.

4.1.

Biochemical Analysis

Figure 2 shows the box plot of the 351 measured spectra. The black line is the mean spectrum, while the gray vertical lines represent the region between the first and third quartiles. Major spectral variability was found in the low-frequency region, $\omega <800{\mathrm{cm}}^{-1}$ , while the high-frequency region, $\omega >1450{\mathrm{cm}}^{-1}$ , presented more reproducible spectra.

Figure 3 shows the FT-Raman spectra of Cys, Cis, Pro, Hyp, and collagens types I and III. Together with glycine (Gly), these are the major tendon components.¹³ It was possible establish complete assignment for the bands displayed in Fig. 2 with help of these spectra and Refs. 24, 25, 26, 27, 28, 29. The assignment is summarized in Table 1 . Observation of this table verified that the Raman spectrum was dominated by amino acids and collagen protein bands, as expected from the tendon compositional profile.^{8, 14, 24} Figure 4 compares the mean FT-Raman spectra for each degenerative grade.

Fig. 3

FT-Raman spectra of cysteine (Cys), cistine (Cis), proline (Pro), hydroxyproline (Hyp), and collagen types I and III.

Fig. 4

Mean FT-Raman spectra for grades G1 to G4 between 350 and $800{\mathrm{cm}}^{-1}$ (panel I), 800 and $1200{\mathrm{cm}}^{-1}$ (panel II), and 800 and $1800{\mathrm{cm}}^{-1}$ (panel III). Each vibrational band is numbered according assignment shown in Table 1.

Table 1

Vibrational assignment for the mean tendon spectra of Fig. 3 based on Refs. 24, 25, 26, 27, 28, 29.

Figure 4 (panel I) shows the mean spectra between 350 and $800{\mathrm{cm}}^{-1}$ . At first glance, an increasing spectral intensity in the 350- to $710\text{-}{\mathrm{cm}}^{-1}$ region is identifiable for the pathological groups G2, G3, and G4. This region is dominated by the stretching modes of Cys/Cis (disulphide bridges), skeletal modes of Pro, Phe, Tyr, and guanine (G), and ${\mathrm{CO}}_2$ wagging/bending mode of Gly. An unidentified band at $590{\mathrm{cm}}^{-1}$ also occurred. The increased intensity was notably greater for G3 with I(G3)/I(G1) $\sim 3$ , in comparison with I(G4)/I(G1) $\sim 1,4$ and I(G2)/I(G1) $\sim 1,3$ .

Characteristic collagen and Pro bands dominated the remaining region from 710 to $800{\mathrm{cm}}^{-1}$ . The band at $727{\mathrm{cm}}^{-1}$ is present in both type I and type III collagens and its intensity was almost grade independent. However, this fact cannot be considered conclusive, because the peak intensity itself is close to the background noise level. The collagen I band at $760{\mathrm{cm}}^{-1}$ presented a similar intensity for grades G1, G2, and G3, when considering the data noise level. G4 presented the greatest intensity.

Figure 4 (panel II) shows the mean spectra between 800 and $1200{\mathrm{cm}}^{-1}$ . The spectral region of 800 to $1030{\mathrm{cm}}^{-1}$ was dominated by $\mathrm{C}\text{—}\mathrm{O}\text{—}\mathrm{C}$ , $\mathrm{C}\text{—}\mathrm{C}$ , and $\mathrm{S}\text{—}\mathrm{S}$ stretching bands of residue backbone; collagen backbone; and amino acids Pro, Hyp, Phe, and Cis. G1 normal tendons presented a lower spectral intensity in this region. The intensity of G2 and G3 were almost the same, while G4 presented the greatest intensity.

The spectral region of 1030 to $1200{\mathrm{cm}}^{-1}$ is dominated by modes of $\mathrm{C}\text{—}\mathrm{N}$ , $\mathrm{C}\text{—}\mathrm{O}$ , $\mathrm{C}\text{—}\mathrm{C}$ , ${\mathrm{NH}}_3$ , DNA, and ${\mathrm{PO}}_2$ $\mathrm{OH}$ of lipids and nucleic acids. The greatest intensity was displayed by the G3 spectrum.

Figure 4 (panel III) shows the mean Raman spectra between 1200 and $1800{\mathrm{cm}}^{-1}$ . The region between 1200 and $1500{\mathrm{cm}}^{-1}$ is dominated by amide III and distortion bands of collagens; ${\mathrm{CH}}_2$ and ${\mathrm{CH}}_3$ amino acid residues, lipids, glycosaminoglycans (GAGs), and metalloproteinases (MMPs). No marked intensity differentiation occurred, besides those discussed previously, due to the fact that the variations in the relative amounts of GAGs and MMPs are very small and their bands are always superposed. Exception was the amide III band at $1300{\mathrm{cm}}^{-1}$ since it was very intense in the G1 spectra. The region between 1500 and $1700{\mathrm{cm}}^{-1}$ is dominated by $\mathrm{C}\mathrm{N}$ , $\mathrm{C}\mathrm{C}$ , $\mathrm{C}\text{—}\mathrm{N}$ , and amide I vibrational bands of DNA, RNA, Phe, Tyr, and collagens. The remaining three bands at 1746, 1767, and $1784{\mathrm{cm}}^{-1}$ were not identified because their intensity variation occurred within background noise levels. Thus, they are not discussed or interpreted further.

To check whether all these spectral differences were sufficient to determine classification and assist coherent diagnosis for all degenerative grades, a complete statistical analysis was performed based on PCA and LR models.

4.2.

Statistical Analysis for Diagnostic Purposes

To extract the essential correlation information from the data observed, PCA analysis was performed on the spectra presented in Fig. 2. The corresponding eigenvalues and contribution percentages of the first $10\mathrm{PCs}$ are shown in Table 2 . From these data, it was verified that 96.9% of the data variability is retained up to the fourth PC. It is possible that higher components account for data noise; thus, further analysis will be based on PC1, PC2, PC3, and PC4 data.

Table 2

Eigenvalues and percentage contribution of the first 10 PCs.

Figure 5 shows the PC loadings for the first four PCs. The corresponding band assignment could be seen on Table 1, as presented on previous section. Figure 5a shows the first score (PC1), which represented 89.1% of the spectral characteristics of the data set. By comparing Fig. 2 to Fig. 5a, it was possible to confirm the well-known fact that the first PC strongly resembled the overall spectral mean.

Fig. 5

PC-loadings of the first four PCs.

Figure 5b shows the second score (PC2), which accounted for 4.0% of the overall spectral characteristics. Compared to the first score, the second presented spectral negative variations in the 850- to 900-, 1050- to 1150-, 1257-, 1270-, 1300-, 1450-, 1660-, and $1751\text{-}{\mathrm{cm}}^{-1}$ regions. In general, negative or positive variation in a PC indicates that it represents these spectra with increasing or decreasing intensity compared to the mean.

Figure 5c shows the third score (PC3), responsible for 2.4% of the spectral characteristics. The behavior of the third score up to $\sim 800{\mathrm{cm}}^{-1}$ and within the 1060- to $1180\text{-}{\mathrm{cm}}^{-1}$ region was very similar to the second score. Excluding these regions, the variations observed were positive and in the same spectral regions as those observed in the second score.

Figure 5d shows the fourth score (PC4), which was close to background noise levels and was responsible for 1.4% of the data spectral variability. Comparing its behavior with those displayed by Figs. 5b and 5c, it is clear that PC4 represented a linear combination of PC2 and PC3 with some peaks resembling the behavior of PC2 and others, those of PC3. However, after a peak-to-peak comparison, it was inferred that the fourth score showed greater similarity to the third score than to the second.

Thus, it is reasonable expect that classification between tendon degenerative states could be obtained by some combination of PC2, PC3, and PC4.

Figure 6 shows the scattering plot of PC2, PC3, and PC4. Each PC was discriminated by its histopathological grade G1 to G4, following Riley’s scale. Unfortunately, direct visual inspection clearly shows that no data separation occurred.

Fig. 6

Scattering plot for PC2, PC3 and PC4.

To more clearly discriminate between spectral data groupings, clustering of the PC2, PC3, and PC4 variables was performed considering a threshold of 95% similarity. The results are shown in a dendogram (Fig. 7 , panel I). As can be observed, the data grouped into two large clusters ( $A$ and $B$ ), with cluster $B$ further divided into two subclusters ( $B1$ and $B2$ ). Cluster $A$ was composed of 95 spectra: 40% from samples diagnosed as G1, 47% from G2 samples, 9% from G3, and 4% from G4. Cluster $B1$ was larger, composed of 150 spectra and of these, 16, 13, 47, and 24% were G1, G2, G3, and G4 samples, respectively. Cluster $B2$ was composed of spectra from 106 samples, 2, 47, 27, and 24% from G1, G2, G3, and G4, respectively. Considering clusters $B1$ and $B2$ together (cluster $B$ ), the percentages of samples were 11, 26, 39, and 24% of G1, G2, G3, and G4, respectively. Summarizing, it was possible to infer that G1 and G2 samples were predominant in cluster $A$ (87%), while those of grades G3 and G4 predominated in cluster $B$ (63%). According to the main histological and biochemical characteristics of the degenerative grades, note that the presence of hyalinization is a shared characteristic of grades G3 and G4. In contrast, hyaline regions are absent in both grades G1 and G2. Thus, cluster $A$ is, to a large extent, composed of nonhyaline tendons from G1 and G2, while cluster $B$ is mainly composed of those tendons presenting hyalinization from G3 and G4 grades. Thus, we labeled clusters $A$ and $B$ as nonhyalinized and hyalinized clusters, respectively.

Fig. 7

Panel I, dendogram based on the PC2, PC3, and PC4, showing the separation of the spectra in two clusters ( $A$ and $B$ ). Group $B$ presents two important subgroups ( $B1$ and $B2$ ). Spectra from nonhyaline samples (G1 and G2) predominate in cluster $A$ (87%), while the hyaline samples (G3 and G4) predominate in cluster $B$ (63%). Panel II, box plot for hyalinized (upper) and nonhyalinized (bottom) clusters.

Figure 7 (panel II) shows the box plot for the hyalinized (upper plot) and nonhyalinized (bottom plot) groups. Nearly all spectral regions in the hyalinized groups were more intense than in the nonhyalinized group. This fact is in agreement with data previously presented in Fig. 4, which showed that the mean spectra of G3 and G4 were generally more intense than those of G1 and G2.

Figure 8a shows the scattering plot PC4 versus PC2, considering these two classes of tendons. Points arising from the hyalinized tissues presented a relatively clear separation from the nonhyalinized points. It is easy to discern a nonhyalinized region to the left and a hyalinized region to the right. The LR model was applied to this case and the best fit was found to be

Fig. 8

Scattering plots of (a) PC4 versus (b) PC2, PC3 versus PC2, (C) PC4 versus PC3 for the nonhyalinized (solid circle symbols) and hyalinized (open circle symbols) clusters. The dotted line is a diagnosis line given by the LR model [Eq. 1] with $p=0.50$ .

Figure 8c shows the scattering plot PC4 versus PC3. In this case, the best fit for the LR model was found to be

The LR model was also tested for PC2, PC3, and PC4 combined, and the best fit was found to be

Figure 9 shows the ROC curves for the combinations (PC2, PC3); (PC2, PC4); (PC3, PC4); and (PC2, PC3, PC4). The straight diagonal line is the expected curve for a completely random diagnostic test. The AUCs were computed and are shown in Table 3 . The most accurate test was found using the combination (PC2, PC3, PC4), which resulted in an AUC of 0.81, classified as a good level of accuracy. The combinations (PC2, PC4) and (PC3, PC4) resulted in fair accuracy levels and the worst test was found using the combination (PC2, PC3) (poor accuracy). These results are in agreement with the predictive ability estimation of the LR presented previously, which provided the highest percentage of concordant pairs for the combinations (PC2, PC3, PC4) and (PC2, PC3).

Fig. 9

ROC curves for combinations of PC 2 and PC 3 (solid black line); PC 2 and PC 4 (solid gray line); PC 3 and PC 4 (dotted line); PC 2, PC 3, and PC 4 (dashed line); and the worthless test (thin solid line).

Table 3

Area under curve (AUC) of the ROC curves presented in Fig. 9.