1.

Introduction

Lung cancer ranked first in terms of incidence and mortality, with a five-year survival rate of 15%.¹ The routine diagnosing process includes a series of examinations such as physical examination, imaging tests, and biopsies, which are time-consuming and invasive.² Raman spectroscopy provides molecular changes of tissues and biofluids in the human body, but the low sensitivity obstructs its application. Surface-enhanced Raman spectroscopy (SERS) can increase the intensity of normal Raman (NR) by an order of 4 to 15 owing to different substrates,^3,4 and it has been widely used in the low-concentration biofluid detection that NR is hardly qualified to perform.^5–7

Saliva has shown the ability to predict many diseases via several methods, including liquid chromatography, mass spectrometry,⁸ and enzyme-linked immunosorbent assay (ELISA).⁹ But most of them are time-consuming and require a large sample compared to SERS.¹⁰ Because of the mechanism of SERS, certain medications of nano-metal particles are usually used, such as biotinylated Ag powder¹¹ and DNA aptamers capped Ag-nanoparticle.¹² Nevertheless, some studies that apply naked SERS-active nanoparticles to the detection of body fluids have been performed. SERS of serum has exhibited differences between type II diabetes mellitus and diabetic complication groups at both the Raman shifts and the principal components.⁶ A concentration of 1 g/mL of 5-fluorouracil in saliva has been measured by the removal of thiocyanate and a pretreatment of SERS-active capillaries.¹³ A difference in saliva SERS between healthy people and oral-cancer patients at three wavenumbers has been detected using a gold nanoparticle assembly.¹⁴ In addition, SERS has been used in the differentiation of saliva for the AIDS virus by a support vector machine algorithm.¹⁵ Although it has not been studied very much, we have seen the potential of using SERS of saliva as a tool for the diagnosis of diseases.

In this paper, the saliva SERS of 21 lung cancer patients and 20 healthy people was measured. The peak variations between the healthy and lung cancer groups were investigated for the diagnosis of lung cancer. Oven-heated silver colloids were used as the SERS substrate, using capillaries as the holder of the mixture of silver colloid and liquid-state saliva to reserve the bioactivity. Then, peaks were assigned in accordance with the possible components of saliva. Finally, multivariate analysis of principal component analysis (PCA) combined with linear discriminant analysis (LDA) were used on the spectroscopy to observe the diagnostic accuracy.

2.

Materials and Methods

3.

Results and Discussion

3.1.

Raman Spectroscopy

The saliva SERS $\pm 1$ standard deviation (SD) of both the lung cancer group ($n=21$) and the control group ($n=21$) were compared (Fig. 1). There were 15 major peaks at the wavelengths of about 523, 622, 696, 735, 789, 822, 884, 909, 925, 1009, 1077, 1280, 1369, 1393, and $1721{\mathrm{cm}}^{-1}$.

Fig. 1

(a) Comparison of the average spectra for the lung cancer patients (red line, $n=21$) versus that of the normal group (green line, $n=20$). The broken lines represent the standard deviations of the averages. Also shown at the bottom (blue line) is the difference spectrum. (b) Plot showing the average intensity value and the corresponding standard deviations of the selected peaks. The peaks that are marked wavenumbers are those that have $p$ values lower than 0.05 of Student’s $t$ test.

Some differences between the two groups still existed, and the difference spectrum made them apparent. The major changes were at the nine wavenumbers of 822, 884, 909, 925, 1009, 1077, 1369, 1393, and $1721{\mathrm{cm}}^{-1}$. Most of the peaks decrease between the control group to the lung cancer group. At 822, 884, 925, 1009, 1280, 1369, and $1393{\mathrm{cm}}^{-1}$ (Student’s $t$ test, $p<0.05$), the peak intensities were greater for the control group than for the lung cancer group, while the bands at 909, 1077, and $1,721{\mathrm{cm}}^{-1}$ were more intense in the saliva of the lung cancer patients (Fig. 1).

Those Raman peaks were assigned to amino acids and nucleic acid bases (Table 2), and the decreases of peak intensities represented the corresponding decrease of those bio-molecules.

Table 2

Assignments of SERS peaks and the changes between healthy people and lung cancer patients. (Peaks were regarded as the same Raman peaks with a wavenumber difference of less than 10  cm−1.)

Raman shift (${\mathrm{cm}}^{-1}$)	Average intensity		Change (%)	$p$ value	Biomolecules
	Control	Cancer
523	$-1.59$	$-1.98$	25	0.639	Lysozymes,20 proteins,26 G, T27
622	2.50	3.24	29	0.736	Protein,21 Phe,20 A28
696	2.23	3.41	53	0.096	Met,20 C28
735	3.81	4.05	6	0.674	Trp,27 coenzyme A, A,5 C, T,29 G30
789	10.42	9.78	$-6$	0.196	C, U, T26
822	5.06	3.55	$-30$	$<0.001$
884	$-0.32$	3.30	$-1,138$	$<0.001$	Pro, Val, Gly,31 Trp,20 Glu,5 Hyd32
909	12.81	16.89	32	$<0.001$	Tyr33
925	13.38	11.77	$-12$	0.002	Pro, glucose26
1,009	11.49	9.80	$-14$	$<0.001$	Try,24 Lys,34 Phe27
1,077	4.31	5.34	24	0.013	Lipids, nucleic acids, proteins, carbohydrates26
1,280	6.53	7.23	11	0.094	Phospholipid, amide III21 proteins, lipids,31 T28
1,369	12.24	10.50	$-14$	$<0.001$	Trp, porphyrins, lipids, G,26 T, protein27
1,393	8.03	5.93	$-26$	$<0.001$
1,722	6.46	9.22	43	0.007	ester group35

Nucleic acid bases: A (adenine), T (thymine), C (cytosine), U (uracil), G (guanine)Amino acids: Trp (tryptophan), Phe (phenylalanine), Lys (lysine), Pro (proline), Glu (glutamate), Val (valine), Gly (glycine), Hyd (hydroxyprolin), Met (methionine).

3.2.

PCA-LDA

PCA was used on our preprocessed SERS of saliva. The contribution rate to the total variation of spectra of the first two PCs (PC1 and PC2) was 83%; hence, the first two principal components can account for 83% of the total variability. Therefore, only the first two PCs were used for further analysis. The scatterplot of the two groups was obtained using the first two PCs, the spots representing different groups distributed separately, and they can be discriminated well. LDA was used on the first two PCs to examine the diagnostic ability of our method. The Fisher’s discriminant function (diagnostic line) of LDA is as follows:

$$0.25\times \mathrm{PC}1-0.07\times \mathrm{PC}2=0.32.$$

Most of the spots belonged to different groups distributed at the two sides of the diagnostic line (Fig. 2), which means that our SERS can be discriminated well using PCA-LDA. An accuracy of 86%, a sensitivity of 94%, and a specificity of 81% were obtained (Table 3).

Table 3

Diagnostic results of PCA-LDA on the SERS of saliva.

Diagnosis	Cases		Total	Accuracy	Sensitivity	Specificity
	Lung Cancer	Healthy
Lung cancer	18	3	21	80%	78%	83%
Healthy	5	15	20

Fig. 2

The scatterplot of the first two PCs (PC1 and PC2) of PCA and the diagnostic line from LDA. The spots of the two groups distribute separately, which means that they can be discriminated well. The Fisher’s discriminant function (diagnostic line) is $0.25\times \mathrm{PC}1-0.07\times \mathrm{PC}2=0.32$.

4.

Discussion

Although some of the peaks were assigned to certain unit structures of proteins or nucleic acids (that is, certain amino acids or nucleic acid bases), the bonds existed in most of the units. For example, ${\mathrm{COO}}^{-}$ existed in all 20 common amino acids, all 5 kinds of nucleic acids have the structure of $O-P-O$, and 3 kinds of amino acids in the aromatic $R$ group have aromatic rings. So the assignment may be expanded to those series of molecules. Such as the peak at $622{\mathrm{cm}}^{-1}$, it was assigned to both phenylalanine²⁰ and other proteins.²¹

Our results showed an overall decrease of protein and nucleic acids in the lung cancer group than in the control group. This decrease trend may have the same cause as the Raman spectroscopy of human lung epithelial tumor cells, which is induced by the breakdown of several kinds of molecular bonds.²² As blood is one of the means for disease diagnosis, and saliva is the ultra-filtration of blood, saliva can reflect the condition of tissues to a certain degree. Hence, the decrease of intensity of certain Raman bands of saliva might be a reflection of lung tissue. Further research is needed to determine if there is a distinct corresponding relationship.

SERS is not the pure enhancement of NR. There are some wavelength shifts in SERS and new peaks will appear compared to NR.²³ As the fingerprint property of Raman spectroscopy, the same bond will have different Raman shifts in different groups. For example, $C-C$ stretch vibration has a Raman shift as $1084{\mathrm{cm}}^{-1}$ in lauric acid, but it moves to $1092{\mathrm{cm}}^{-1}$ in myristic acid, and $1099{\mathrm{cm}}^{-1}$ in palmitic acid.²⁴ Therefore, certain errors in the peak assignments based on the literature might exist due to totally different measuring conditions.

Some high-abundance proteins, like amylase, lysozyme, and mucin²⁵ in saliva, have a very small diagnosis significance and might immerse the Raman peaks of low-abundance biomarkers. Removal of those substances can exclude the disturbance and improve the expression of those low-abundance biomarkers. Silver colloid in SERS is more uniform than that made by traditional water-heating, but the Ag particles are smaller; thus, there is a consequent decrease in the enhancement level. A highly ordered substrate can create a higher enhancement and more reproducible results. SERS effect is influenced by many factors, like laser power, temperature, solvent of samples, SERS substrate, the state of the analyte, and exposure time. The uniformity of SERS spectrometers and measuring conditions can eliminate the above fluctuations greatly and is the future orientation of our research.

5.

Conclusion

Saliva has shown the potential to diagnose many diseases, and recently, SERS has been used in the detection of saliva because of the high enhancement effect. The peak changes between the lung cancer group and the control group and an 80% diagnosis accuracy in our study showed the potential of saliva SERS for diagnosing lung cancer. SERS of saliva has several advantages, such as fast, noninvasive, and low-analyte requirements, and it has prospective clinic applications. High-abundance proteins will be depleted, and better-ordered SERS substrate will be introduced in our future experiments to eliminate the disturbance of those proteins and increase the enhancement and the reproducibility of SERS measurements. Possible saliva biomarkers will be selected and detected for more exact peak assignments.