1.

Introduction

According to Global Cancer Statistics, cancer is one of the most common causes of death, with about 14.1 million people being diagnosed and 8.2 million people dying of cancer in 2012 worldwide.¹ Nasopharyngeal carcinoma (NPC), a head and neck malignancy, is an endemic disease with high mortality rates in certain regions of southeastern Asia, including Hong Kong, Guangdong province of China, Fujian province of China, and some other South Asian countries.^2–5 The five-year overall survival rate is about 90% for stage I NPC patients; however, the stage IV group has a poor five-year overall survival rate of 30.3%.^6,7 Hence, early screening and effective treatment are essential for preventing the advancement of NPC and improving the five-year overall survival rate. Unfortunately, early detection remains a great challenge due to the insidious nature of NPC and the relative anatomical inaccessibility of the nasopharynx. At present, the common detection and screening methods for NPC mainly include electronic nasopharyngoscope examination, nasopharyngeal and neck magnetic resonance imaging, body positron emission tomography, bone emission CT, and biopsy, which suffer the disadvantages of being time-consuming, high-cost, complex procedures, and based on subjective judgment of clinicians with varying levels of experience.^8–10 Therefore, development of an alternative diagnostic method is of significant clinical value for NPC screening.

Raman spectroscopy (RS) is a powerful optical analytical tool with many advantages over conventional optical detection techniques. For example, RS has low water interference, high signal-to-noise ratio, and high sensitivity compared with Fourier infrared spectroscopy.¹¹ Additionally, RS has a narrow peak width and multiplexing detection ability, leading to better analytical efficiency than fluorescence spectroscopy.^12,13 RS has attracted significant attentions for its material analysis and biomedical applications, especially for cancer diagnosis, due to its ability to provide structural- and biochemical-specific information of macromolecules, such as proteins, nucleic acids, lipids, and so on.^14,15 However, the conventional RS technique has some deficiencies, including its inherent weak Raman scattering efficiency and strong fluorescence background. For instance, the Raman signal of a biological molecule is 1/10,000 of its fluorescence signal. To overcome these problems, surface-enhanced Raman spectroscopy (SERS) technology exploits the interaction between the biomolecules and metal nanoparticles (NPs) surface to dramatically increase the Raman signal (enhancement factors of up to ${10}^{13}$ to ${10}^{15}$), providing a more powerful and sensitive detection approach for cancer detection.^2,16,17 Currently, SERS technology has been widely used to detect various cancers in human samples, such as DNA, RNA, cell, blood, and tissue.¹⁸ Compared with traditional polymerase chain reactions or immunoassays, biomarker detection based on SERS avoids use of expensive reagents and complex sample preparation steps. In addition, SERS has many advantages over traditional fluorescent method, such as better multiplex capability and less photobleaching.^19,20

Human saliva, as a body fluid, contains a large number of serum components, abundant protein, and metabolites, which may change when associated with neoplastic transformation.²¹ Comparing the above-mentioned human samples, saliva is regarded as an ideal medium for cancer diagnostics due to its rapid, noninvasive, and convenient collection procedure.^22–24 Recently, application of SERS in saliva analysis was investigated extensively to detect several types of cancer.^19,25–27 These preliminary results demonstrated the potential of the saliva SERS method for cancer detection.

To further verify the efficiency of saliva SERS in NPC detection, in this work, a large number of NPC saliva samples were collected, and a saliva sample analysis based on membrane protein purification was developed for saliva SERS detection. Moreover, principal component analysis combined with a linear discriminant analysis (PCA-LDA) diagnostic algorithm was employed to classify the saliva protein SERS spectra from the NPC group and the healthy group.

2.

Materials and Methods

2.2.

Preparation of Silver Colloids

Silver (Ag) colloids were synthesized using hydroxylamine hydrochloride and Ag nitrate according to the deoxidizing method reported by Li et al.²⁸ At first, 4.5 ml sodium hydroxide solution was mixed with 5 ml of 0.06 M hydroxylamine hydrochloride solution. After that, the mixture was added rapidly to 90 ml of 0.0011 M silver nitrate aqueous solution. The resulting mixture was stirred to obtain uniformly gray Ag colloids. Meanwhile, the maximum absorption peak of Ag colloids was 418 nm, and the Ag colloids NPs sizes were represented with a mean diameter of 35 nm by transmission electron microscopy (TEM) [Fig. 1(a)]. Finally, the final Ag colloids were obtained by centrifuging the solution at 10,000 rpm for 10 min and removing the supernatant as the SERS substrate.

Fig. 1

(a) UV–visible–NIR absorption spectra of Ag colloids and TEM micrograph of the Ag colloidal surface and (b) the schematic diagram of saliva protein purification and SERS detection.

3.

Results and Discussions

Figures 2(a) and 2(b) show the SERS spectra of untreated saliva samples of 10 patients with NPC and saliva protein samples after membrane protein purification from the same samples, respectively. Great variations in spectral features, including spectral intensity, position, and width, could be found in the untreated saliva SERS within the cancer group, making it difficult to achieve efficient diagnostics using SERS spectral analysis. This is explainable. A saliva sample contains a variety of native constituents (proteins, peptides, polynucleotides, and electrolytes) and exogenous substances (nonadherent oral bacteria, food remainders, traces of medications, or chemical products).^31,32 Although preprocessing such as washing the mouth for saliva collection was implemented, there remained some exogenous substances in the saliva samples that generated prominent SERS signals. Thus, great SERS spectral variation from untreated saliva samples was probably attributed to the different exogenous substances among subjects. Interestingly, the reproducibility of SERS spectra [Fig. 2(b)] was dramatically improved by the saliva protein purification method developed. This method not only circumvented the limitations of raw saliva SERS detection but also provided a unique opportunity to use SERS to explore the changes in saliva proteins associated with cancer transformation.

Fig. 2

(a) Untreated saliva SERS spectra from 10 NPC saliva samples and (b) saliva protein SERS spectra using membrane protein purification method from the same samples.

Figure 3(a) shows the mean normalized SERS spectrum of purified whole proteins obtained from normal (blue line, $n=71$) and NPC (red line, $n=170$) saliva samples. The prominent SERS peaks located at around 621, 642, 760, 854, 878, 935, 959, 1004, 1031, 1049, 1123, 1175, 1208, 1265, 1337, 1445, 1552, and $1684{\mathrm{cm}}^{-1}$ were clearly observed in both normal and NPC saliva protein samples. These prominent SERS peaks could be tentatively assigned to explain the changes in biological constituents in saliva protein samples as shown in Table 2.^33–36 The strongest peaks at 760, 1004, 1265, 1445, and $1684{\mathrm{cm}}^{-1}$ existed in the measured saliva protein SERS spectra. Compared with the SERS spectra of NPC saliva proteins, the SERS spectra of normal saliva proteins had higher intensities at 621, 935, and $1049{\mathrm{cm}}^{-1}$ and lower intensities at 1004, 1031, 1208, 1123, and $1684{\mathrm{cm}}^{-1}$. The comparisons of saliva proteins SERS spectral intensities between the NPC and normal groups could be viewed more clearly in Figs. 3(b) and 3(c). These differences in spectral intensities demonstrated that saliva proteins SERS has a potential role for NPC detection. For instance, the SERS bands of phenylalanine (1004 and $1031{\mathrm{cm}}^{-1}$) were all related to the molecular stretching and bending mode of proteins, and they showed a lower SERS signal in the normal saliva protein samples than in the NPC saliva protein samples, indicating that an increase in the percentage of these proteins content in the total SERS-active in the NPC patients. Our group also observed that the content of phenylalanine was increased when associated with malignant transformation in cervical and NPC blood plasma by the SERS technique.^2,18 However, the intensity of the saliva protein SERS spectrum exhibited decreased peaks at $935{\mathrm{cm}}^{-1}$ (praline and valine) and $1049{\mathrm{cm}}^{-1}$ (proteins) in NPC subjects, indicating that cancer patient saliva may be associated with a decreased concentration of these proteins. These results revealed that specific-protein changes between the NPC and normal groups could be detected by SERS, suggesting promising potential of saliva protein SERS for NPC screening.

Fig. 3

(a) Comparison of the mean normalized SERS spectrum of purified proteins obtained from normal saliva (blue line, $n=71$) and cancer saliva (red line, $n=170$), (b) difference spectrum calculated from the mean SERS spectra between the NPC and normal groups, and (c) comparison of the mean intensities of the selected peaks between cancer (red pillar) and normal (black pillar) samples.

Table 2

SERS peak positions and tentative vibrational mode assignments.

It should be noted that the simplistic peak intensities analysis above only uses limited information of SERS peaks, and more valuable diagnostic information contained in the SERS spectra has not been employed for spectral classification. Therefore, the PCA-LDA diagnostic algorithm was used to improve the diagnostic efficiency of the SERS technique by analyzing and classifying the saliva protein SERS spectra from NPC and normal subjects.

This sophisticated and robust diagnostic model based on PCA-LDA has been widely used for Raman spectral analysis in tissue, cell, saliva, and blood detection.^27,37,38 In this study, PCA was used first to reduce intensity variables within the raw Raman spectrum of saliva protein into a few PCs. Then, three PCs (PC5, PC6, and PC7) were selected to be the most diagnostically significant ($p<0.0001$) as defined for discriminating normal and NPC groups by an independent sample $t$-test.

To further analyze saliva protein SERS data, all three diagnostically significant PCs were loaded into the LDA with the leave-one-out cross-validation method for generating an effective diagnostic model for saliva protein sample classification. Figure 4(a) shows the posterior probability plot of the healthy group (circles) and NPC group (triangles) based on the results of the PCA-LDA diagnostic model. As we can see, there are some plots overlapping in this figure. A discrimination threshold of 0.54 was employed for accurate discriminate between the healthy volunteers and NPC patients, yielding the diagnostic sensitivity, specificity, and accuracy of 70.7% (304/430), 70.3% (161/229), and 70.5% (465/659), respectively. To further evaluate the performance of saliva protein SERS for NPC diagnosis, an ROC curve was generated from the PCA-LDA data as shown in Fig. 4(b). The integration area under the ROC curve was 0.795. These results further indicated that saliva protein SERS technology combining the PCA-LDA diagnostic algorithm has the potential as a rapid and noninvasive diagnosis method for NPC detection.

Fig. 4

(a) Scatter plots of the posterior probability of NPC (red triangles) and normal (blue circles) saliva samples using the PCA-LDA algorithm. The separation threshold was 0.54. (b) The ROC curve of discrimination results for saliva protein SERS spectra. The integration area under the ROC curve was 0.795.

4.

Conclusions

A rapid and convenient saliva SERS analysis method was developed for NPC detection. Using the membrane protein purification method, high-quality and reproducibility SERS spectra of saliva proteins were obtained, making it possible to reveal specific changes in proteins associated with cancer transformation. Furthermore, a diagnostic sensitivity of 70.7% and specificity of 70.3% could be achieved by the PCA-LDA diagnostic algorithm for NPC identification, demonstrating promising potential of this saliva SERS analysis method to be a noninvasive body fluid test for clinical NPC screening.