2.1.

Synthesis of Silver Nanoparticles

The Lee and Meisel method for preparation of colloidal silver nanoparticles (AgNPs) was used.⁴¹ Eighteen mg of ${\mathrm{AgNO}}_3$ was dissolved in 100 mL of distilled water by stirring. Then the solution was heated to boiling, and 2 mL of 1% (w/v) trisodium citrate solution was dropped into this boiling solution. Finally, the solution was kept boiling until it was only half of the initial volume. The solution was concentrated by centrifugation at 5500 rpm for 30 min, and one-third of the supernatant was decanted to increase the final concentration of the AgNPs colloidal suspension to four times, which was called $4\times$. The maximum of the UV/visible spectrum from the resulting AgNP suspension was at 420 nm [Fig. 1(a)] indicating the average size of the AgNPs as 50 nm. The final density of the AgNPs in the $4\times$ concentrated suspension was calculated as $2.08\times {10}^{11}\text{particles}/\mathrm{mL}$.⁴²

Fig. 1

(a) UV/visible absorption spectra of silver nanoparticle (AgNP) colloidal suspension. (b) Surface-enhanced Raman scattering (SERS) spectrum of colloidal AgNPs. The insert shows the transmission electron microscopy micrograph of AgNPs.

2.2.

Kidney Cancer Tissue Sample

The biopsy samples of kidney tissues were obtained from the cancer patients with consent of the ethical approval from the Department of Urology at Okmeydani Education and Research Hospital. The tissue samples obtained during surgery were separated into two parts. One portion was saved for the histopathological examination, while the other half was stored at $-80°\mathrm{C}$ in plastic tubes until the sample preparation for the SERS measurements. The histopathological tumor staging was performed by a panel of pathologists. Table 1 shows the kidney tumors stages, the age, and the gender of the biopsied patients. For the SERS measurements, the sample preparation was performed using our previously reported method.²⁶ Briefly, a piece of tissue approximately in the size of $2\times 2\times 3{\mathrm{mm}}^3$ was cut from sample tissue and tumor samples and placed in a mortar with 5 mL of liquid nitrogen to be crushed with a pestle. This crushed and liquefied tissue was mixed with $200\mu \mathrm{L}$ of the AgNP suspension. Then a $5\mu \mathrm{L}$ volume of this mixture was placed onto a ${\mathrm{CaF}}_2$ slide, and each of the ${\mathrm{CaF}}_2$ slides was dried in an inverted position under sterile conditions.⁴³ The average number of spectra collected on each tissue sample was 10 with three repeated Raman measurements. A total of 800 SERS spectra were acquired from 40 normal and 40 abnormal tissues, in which 280 SERS spectra were from T1 stage tumor tissues, 120 SERS spectra were from T2–T3 stage tumor tissues, and 400 SERS spectra were from the normal tissues.

Table 1

Histopathological tumor staging, age and gender of patients that tissues are biopsied.

2.3.

Raman System and Surface-Enhanced Raman Scattering Measurements

A Renishaw InVia Reflex Raman microscopy system (Renishaw Plc., New Mills, Wotton-under-Edge, UK) equipped with an 830-nm diode was calibrated by using the silicon phonon mode at $520{\mathrm{cm}}^{-1}$. The incident laser power was 3 mW on the sample, and the spectral data acquisition time was 10 s. The SERS spectra were acquired over a spectral range of $400-1800{\mathrm{cm}}^{-1}$ with a $50\times$ microscope objective (NA: 0.50). The SERS spectra were collected from randomly selected points on the sample using the “map image acquisition method” function in WIRE 2.0 software, and the WIRE 2.0 software carried out the spectral analyses.

2.4.

Spectral Data Processing and Analysis

Each of randomly selected 10 SERS spectra obtained from each tissue sample was normalized to reduce the variations in the Raman intensity and to permit comparison of the spectral shapes. The spectral dataset was processed with the statistical package for the social science (SPSS) package (SPSS Inc., Chicago, Illinois), which contains PCA and LDA algorithms for statistical analysis to clarify the significant spectral characteristics of each tissue type and differentiate tissue types from each other for tumor stage identification. PCA was applied before LDA to reduce the number of dimensions ($d=2009$) in the original high-dimensional dataset. To apply PCA, the spectral data observations having 2009 predictors were placed into a data matrix using 10 SERS spectra for each tissue sample. The eigenvalue decomposition was performed to the covariance matrix of the spectral data matrix. The resultant eigenvectors were obtained, and the original spectral dataset was projected into the new coordinate system defined by the principal directions of variance, called the principal components (PCs), which are the linear combination of the original data variables.⁴⁴ One way analysis of variance (ANOVA)^45,46 is used to determine the most significant PCs ($p<0.05$), then PCs were used in LDA to generate diagnostic algorithms for the classification of the tumors with different stages. LOO-CV⁴⁷ methodology was applied to demonstrate the accuracy of the classification.^48,49 The significant differences in the Raman peak intensities for each tissue class were obtained by applying one-way ANOVA with a Tukey post-hoc test (significance level $p<0.05$).^45,50,51

3.1.

Spectral Characteristics of Normal and Cancerous Kidney Tissues

SERS spectra were recorded from each type of normal and cancerous tissues. The mapping method of WIRE 2.0 software was applied to collect at least 10 spectra from randomly selected points from the dried sample composed of homogenized tissue and the colloidal AgNPs. The histopathologic stages of the cancer tissues according to TNM (tumor, node, and metastasis) classification were T1a, T1b, T2a, T2b, and T3a for RCC and T3 for TCC. The mean spectra of RCC tissue from the different stages T1a (120), T1b (160), T2a (30), T2b (10), and T3a (60) with the mean spectra of TCC tissue from the T3 (20) stage tumor and the mean spectra of normal tissues (400) were normalized to the integrated area under the curve in the range of $400-1800{\mathrm{cm}}^{-1}$ to enable a better comparison of the spectral shapes and the relative peak intensities among on the spectra obtained from the different tissue samples. Figure 2 shows the comparison of the normalized average SERS spectra acquired from the normal tissues and cancerous tissues at T1 (T1a–T1b), T2 (T2a–T2b), and T3 (T3a) stages. The major SERS peaks, which can be attributed to biochemical constituents such as nucleic acids (478, 560, 723, 1086, and $1334{\mathrm{cm}}^{-1}$), proteins (524, 657, 804, 1031, 1050, 1214, 1300, 1395, 1443, 1585, and $1704{\mathrm{cm}}^{-1}$), carbohydrates ($905{\mathrm{cm}}^{-1}$), and lipids (961, 1128, and $1443{\mathrm{cm}}^{-1}$), were obtained from normal and abnormal tissue subjects. The tentative assignments for the observed SERS bands are listed in Table 2 in order to understand the possible molecular basis of the changes in the tissue samples based on previously reported studies in the literature.^40,52–63 Figure 3 shows that the normalized intensities of SERS peaks at 478, 560, 723, 961, 1031, 1086, 1128, 1214, 1334, 1443, 1585, and $1704{\mathrm{cm}}^{-1}$ are higher for the tumor tissues than those of the normal tissues, while the intensities of the SERS bands at 657, 1050, and $1395{\mathrm{cm}}^{-1}$ are higher on the spectra obtained from the normal tissues. The statistical significance of differences in the peak intensities between the different pathology groups and normal group was identified using one-way ANOVA with a Tukey post-hoc test (significance level $p<0.05$).

Fig. 2

Mean SERS spectra obtained from normal tissues and cancerous tissues at T1 (T1a-T1b), T2 (T2a-T2b), T3 (T3a) stages.

Table 2

Peak positions and tentative assignment of Raman bands.40,52–63

Fig. 3

Column plots of 19 significant SERS peak intensities for the four tissue types (normal, T1, T2, and T3 stage cancers) (a) 478, (b) 524, (c) 560, (d) 657, (e) 723, (f) 804, (g) 905, (h) 961, (i) 1031, (j) 1050, (k) 1086, (l) 1128, (m) 1214, (n) 1300, (o) 1334, (p) 1395, (r) 1443, (s) 1585, (t) $1704{\mathrm{cm}}^{-1}$. Error bars represent the standard deviation for SERS spectra obtained from each type of tissue. *Significantly different from each other [$P<0.05$, one-way analysis of variance (ANOVA) with Tukey’s HSD post-hoc test].

The selected spectral intensities, where the standard deviations do not overlap, were displayed in Fig. 3, and the differences are thus significant and reproducible. The significant decrease and increase for biomolecules is relative to the total Raman active components in the different tissue groups. These spectral intensity differences for the different pathological tissues and normal tissues could be evaluated for a better understanding of molecular changes between malignant and normal tissue types. The peak intensities at 478, 723, and $1334{\mathrm{cm}}^{-1}$, primarily related to nucleic acids, were found to be increased in the cancer groups, indicating the uncontrolled fast replication of DNA in cancer cells. This is associated with the increased nucleic acid content in cancer cells. The band intensities at 560 and $1086{\mathrm{cm}}^{-1}$ were higher on the spectra obtained from tissues at the T3 stage than the normal and T1–T2 stages.⁶⁴ The band at $961{\mathrm{cm}}^{-1}$, attributed to cholesterol, is more intense on the spectra obtained from the tumors, especially at the T3 stage. This increased intensity may be attributed to an increased cholesterol synthesis in cancer tissues.^65,66 Phenylalanine-related bands at 1031, 1214, 1585, and $1704{\mathrm{cm}}^{-1}$ in cancer groups are associated with increased phenylalanine contents relative to the total Raman-active components in cancer tissues. ^58,63 In a study, it was found that the uptake rate of amino acids in cancer was higher.⁶⁷ The increased band intensity at $1128{\mathrm{cm}}^{-1}$ in the cancer groups may be attributed to the increased lipid concentration in the tumors. The studies comprising the high levels of fatty acid synthesis related to the tumor aggressiveness are consistent with our study results.^68–70 The band at $1443{\mathrm{cm}}^{-1}$, which is probably characteristic of collagen and phospholipids, shows a higher intense signal in malignant tissues and indicates that the collagen synthesis significantly increased in the cancerous tissues.⁷¹ The bands at 657, 1050, and $1395{\mathrm{cm}}^{-1}$, more intense in the normal tissue, are assigned to proteins, and the bands at 524, 804, and $1300{\mathrm{cm}}^{-1}$, probably originating from proteins or lipids, were more intense in the normal tissues than the tumors at the T1 and T3 stages, even though it was the most intense on the spectrum from the cancerous tissue at the T2 stage. Note that there is a possibility that more than one band’s vibrations might contribute to the band observed on the spectra. In addition, the intensity of the band at $905{\mathrm{cm}}^{-1}$, associated with protein or glycogen, was higher on the spectra obtained from the normal tissues than that of the tumor at the T3 stage. However, the highest band intensity at $905{\mathrm{cm}}^{-1}$ was obtained from tumors at the T1 and T2 stages.

3.2.

PCA-LDA as Diagnostic Algorithms

Renal cell carcinoma staging has been used as a significant prognostic factor for kidney cancer patients because the survival rate of patients with renal cancer is reduced from the early stage to the late stage. The PCA-LDA model was built to predict the classification of tissue types and to improve the diagnostic utility of kidney cancer patients. The SERS spectra acquired from normal and abnormal tissues were processed in SPSS software package (SPSS Inc., Chicago, Illinois) for PCA-LDA analysis after the intensity of the spectra was scaled within a similar range using the min–max normalization method to compare the relative peak intensities among the normal tissue and tumor stages in a more precise manner. The significant PCs obtained using one-way ANOVA comparison test ($p<0.05$) was used in LDA to generate a diagnostic assay. The scatter plot of the posterior probabilities based on the linear discriminant scores of the normal and the cancerous tissues using the PCA-LDA diagnostic algorithm is provided in Fig. 4. Each dot on the plot is associated with the SERS spectra acquired from each type of tissue. The LDA scatter plot of the classification model developed to differentiate the cancerous and the normal tissue samples shows a good discrimination among the normal and abnormal tissues at the different tumor stages. The discrimination results based on SERS spectra using a leave one out-cross validation method to evaluate the performance of the PCA-LDA models for the classification of different tumor stages in terms of sensitivity, specificity and 95% confidence interval of accuracy was displayed in Fig. 5. The RCC tumors at T1a stage related to PCs, which were subtracted from SERS spectra, were well differentiated from RCC tumors at T1b, T2a, T2b, and T3a stages, TCC tumors at T3 stage and the normal tissues with the diagnostic sensitivities of 100%, 100%, 65%, 100%, 89%, and 65%, the specificities of 100% from all types of tumor stages and 70% from the normal subjects, and the accuracy of 100%, 100%, 88%, 93%, 100%, and 69%, respectively. The discrimination results of the diagnostic combinations of RCC tumors at T1b stage versus T2a, T2b, and T3a stages, TCC tumors at T3 stage and normal tissues, and RCC tumors at T2a stage versus RCC tumors at T2b, T3a stages, TCC tumors at T3 stage and the normal tissues were achieved with a sensitivity, specificity, and accuracy of 100% while the posterior probabilities of RCC tumors at T2b stage versus RCC tumors at T3a stage, TCC tumors at T3 stage and normal tissues were obtained with sensitivities of 88%, 100%, and 80%, specificities of 43%, 100%, and 84%, and accuracies of 57%, 100%, and 83%, respectively. The RCC tumors at the T3a stage versus normal tissues were diagnosed with a sensitivity, specificity, and accuracy of 100%, and TCC tumors at the T3 stage were classified with a sensitivity of 100% and 70%, a specificity of 98% and 60%, and an accuracy of 98% and 62%, respectively.

Fig. 4

Scatter plot of posterior probabilities for classification of normal tissues and tumors.

Fig. 5

Leave-one-out cross-validation (LOO-CV) classification results of normal tissues and tumor tissues (T1a, T1b, T2a, T2b, and T3a stages of RCC; T3 stage of TCC).

The tumors at the T1, T2, and T3 stages with normal tissues were also differentiated using PCA-LDA models of the spectral data obtained from the normal and the abnormal tissue samples. As compared to the classification results in Fig. 5, Fig. 6 shows a better discrimination among the different tissue classes with a diagnostic sensitivity of 89%, 96%, 94%, 70%, 97%, and 98%, specificity of 100%, 100%, 96%, 83%, 94%, and 77%, and accuracy of 99%, 97%, 95%, 81%, 95%, and 85%, respectively, for the classification between T1 and T2 stage cancers; T1 and T3 stage cancers; T1 stage cancer and normal tissues; T2 and T3 stage cancers; T2 stage cancer and normal groups; T3 stage cancer and normal tissue, respectively. The classification of tumors at the advanced stages (T2–T3), at the early stage (T1), and for normal tissues was obtained with sensitivities of 93% and 100%, specificities of 98% and 86%, and accuracies of 91%, and 90%, respectively, indicating that the classification of normal tissues, early stage tumors, and advanced stage tumors is possible with a high diagnostic efficacy.

Fig. 6

LOO-CV classification results of normal tissues and tumors at T1, T2, and T3 stages.

The three-dimensional scatter plot of the diagnostic probabilities of LD1, LD2, and LD3 discriminants were shown in Fig. 7, illustrating a good classification among the different tumor stages and the normal tissue groups. The diagnostic performance of PCA-LDA models on the classification of tissue types has been acquired with an improved accuracy by the selection of significant PCs and different Raman bands ($p<0.05$).

Fig. 7

Three-dimensional (3-D) scatter plot of diagnostic probabilities of normal tissues and tumors at T1, T2, and T3 stages.