Colorectal cancer detection by gold nanoparticle based surface-enhanced Raman spectroscopy of blood serum and statistical analysis

The capabilities of using gold nanoparticle based surfaceenhanced Raman spectroscopy (SERS) to obtain blood serum biochemical information for non-invasive colorectal cancer detection were presented in this paper. SERS measurements were performed on two groups of blood serum samples: one group from patients (n = 38) with pathologically confirmed colorectal cancer and the other group from healthy volunteers (control subjects, n = 45). Tentative assignments of the Raman bands in the measured SERS spectra suggested interesting cancer specific biomolecular changes, including an increase in the relative amounts of nucleic acid, a decrease in the percentage of saccharide and proteins contents in the blood serum of colorectal cancer patients as compared to that of healthy subjects. Both empirical approach and multivariate statistical techniques, including principal components analysis (PCA) and linear discriminant analysis (LDA) were employed to develop effective diagnostic algorithms for classification of SERS spectra between normal and colorectal cancer serum. The empirical diagnostic algorithm based on the ratio of the SERS peak intensity at 725 cm 1 for adenine to the peak intensity at 638 cm 1 for tyrosine achieved a diagnostic sensitivity of 68.4% and specificity of 95.6%, whereas the diagnostic algorithms based on PCA-LDA yielded a diagnostic sensitivity of 97.4% and specificity of 100% for separating cancerous samples from normal samples. Receiver operating characteristic (ROC) curves further confirmed the effectiveness of the diagnostic algorithm based on PCA-LDA technique. The results from this exploratory study demonstrated that gold nanoparticle based SERS serum analysis combined with PCA-LDA has tremendous potential for the non-invasive detection of colorectal cancers. ©2011 Optical Society of America OCIS codes: (240.6695) Surface-enhanced Raman scattering; (170.1470) Blood or tissue constituent monitoring; (170.4580) Optical diagnostics for medicine. References and links 1. C. P. Xavier, C. F. Lima, A. Preto, R. Seruca, M. Fernandes-Ferreira, and C. Pereira-Wilson, ―Luteolin, quercetin and ursolic acid are potent inhibitors of proliferation and inducers of apoptosis in both KRAS and BRAF mutated human colorectal cancer cells,‖ Cancer Lett. 281(2), 162–170 (2009). 2. S. J. Winawer, ―Colorectal cancer screening,‖ Best Pract. Res. Clin. Gastroenterol. 21(6), 1031–1048 (2007). 3. R. Labianca, G. D. Beretta, S. Mosconi, L. Milesi, and M. A. Pessi, ―Colorectal cancer: screening,‖ Ann. Oncol. 16(Suppl 2), ii127–ii132 (2005). 4. R. M. McLoughlin and C. A. O‘Morain, ―Colorectal cancer screening,‖ World J. Gastroenterol. 12(42), 6747– 6750 (2006). 5. A. Kudelski, ―Analytical applications of Raman spectroscopy,‖ Talanta 76(1), 1–8 (2008). #146382 $15.00 USD Received 22 Apr 2011; revised 11 Jun 2011; accepted 11 Jun 2011; published 29 Jun 2011 (C) 2011 OSA 4 July 2011 / Vol. 19, No. 14 / OPTICS EXPRESS 13565 6. Z. Huang, A. McWilliams, H. Lui, D. I. McLean, S. Lam, and H. Zeng, ―Near-infrared Raman spectroscopy for optical diagnosis of lung cancer,‖ Int. J. Cancer 107(6), 1047–1052 (2003). 7. S. K. Teh, W. Zheng, K. Y. Ho, M. Teh, K. G. Yeoh, and Z. Huang, ―Diagnostic potential of near-infrared Raman spectroscopy in the stomach: differentiating dysplasia from normal tissue,‖ Br. J. Cancer 98(2), 457–465 (2008). 8. S. Wachsmann-Hogiu, T. Weeks, and T. Huser, ―Chemical analysis in vivo and in vitro by Raman spectroscopy—from single cells to humans,‖ Curr. Opin. Biotechnol. 20(1), 63–73 (2009). 9. S. Devpura, J. Thakur, F. Sarkar, W. Sakr, V. Naik, and R. Naik, ―Detection of benign epithelia, prostatic intraepithelial neoplasia, and cancer regions in radical prostatectomy tissues using Raman spectroscopy,‖ Vib. Spectrosc. 53(2), 227–232 (2010). 10. U. Utzinger, D. Heintzelman, A. Mahadevan-Jansen, A. Malpica, M. Follen, and R. Richards-Kortum, ―Nearinfrared Raman spectroscopy for in vivo detection of cervical precancers,‖ Appl. Spectrosc. 55(8), 955–959 (2001). 11. S. Feng, R. Chen, J. Lin, J. Pan, G. Chen, Y. Li, M. Cheng, Z. Huang, J. Chen, and H. Zeng, ―Nasopharyngeal cancer detection based on blood plasma surface-enhanced Raman spectroscopy and multivariate analysis,‖ Biosens. Bioelectron. 25(11), 2414–2419 (2010). 12. K. Kneipp, A. Haka, H. Kneipp, K. Badizadegan, N. Yoshizawa, C. Boone, K. Shafer-Peltier, J. Motz, R. Dasari, and M. Feld, ―Surface-enhanced Raman spectroscopy in single living cells using gold nanoparticles,‖ Appl. Spectrosc. 56(2), 150–154 (2002). 13. M. Fleischmann, P. Hendra, and A. McQuillan, ―Raman spectra of pyridine adsorbed at a silver electrode,‖ Chem. Phys. Lett. 26(2), 163–166 (1974). 14. K. Kneipp and M. Moskovits, ―Surface-enhanced raman scattering,‖ Phys. Today 60(11), 40–46 (2007). 15. Y. Badr and M. A. Mahmoud, ―Effect of silver nanowires on the surface-enhanced Raman spectra (SERS) of the RNA bases,‖ Spectrochim. Acta A Mol. Biomol. Spectrosc. 63(3), 639–645 (2006). 16. Y. Liang, J. L. Gong, Y. Huang, Y. Zheng, J. H. Jiang, G. L. Shen, and R. Q. Yu, ―Biocompatible core-shell nanoparticle-based surface-enhanced Raman scattering probes for detection of DNA related to HIV gene using silica-coated magnetic nanoparticles as separation tools,‖ Talanta 72(2), 443–449 (2007). 17. Z. S. Wu, G. Z. Zhou, J. H. Jiang, G. L. Shen, and R. Q. Yu, ―Gold colloid-bienzyme conjugates for glucose detection utilizing surface-enhanced Raman scattering,‖ Talanta 70(3), 533–539 (2006). 18. J. D. Guingab, B. Lauly, B. W. Smith, N. Omenetto, and J. D. Winefordner, ―Stability of silver colloids as substrate for surface enhanced Raman spectroscopy detection of dipicolinic acid,‖ Talanta 74(2), 271–274 (2007). 19. M. Culha, D. Stokes, and T. Vo-Dinh, ―Surface-enhanced Raman scattering for cancer diagnostics: detection of the BCL2 gene,‖ Expert Rev. Mol. Diagn. 3(5), 669–675 (2003). 20. J. D. Driskell, A. G. Seto, L. P. Jones, S. Jokela, R. A. Dluhy, Y. P. Zhao, and R. A. Tripp, ―Rapid microRNA (miRNA) detection and classification via surface-enhanced Raman spectroscopy (SERS),‖ Biosens. Bioelectron. 24(4), 917–928 (2008). 21. X. Huang, I. H. El-Sayed, W. Qian, and M. A. El-Sayed, ―Cancer cells assemble and align gold nanorods conjugated to antibodies to produce highly enhanced, sharp, and polarized surface Raman spectra: a potential cancer diagnostic marker,‖ Nano Lett. 7(6), 1591–1597 (2007). 22. S. Lee, H. Chon, M. Lee, J. Choo, S. Y. Shin, Y. H. Lee, I. J. Rhyu, S. W. Son, and C. H. Oh, ―Surface-enhanced Raman scattering imaging of HER2 cancer markers overexpressed in single MCF7 cells using antibody conjugated hollow gold nanospheres,‖ Biosens. Bioelectron. 24(7), 2260–2263 (2009). 23. D. Rohleder, W. Kiefer, and W. Petrich, ―Quantitative analysis of serum and serum ultrafiltrate by means of Raman spectroscopy,‖ Analyst (Lond.) 129(10), 906–911 (2004). 24. S. Feng, R. Chen, J. Lin, J. Pan, Y. Wu, Y. Li, J. Chen, and H. Zeng, ―Gastric cancer detection based on blood plasma surface-enhanced Raman spectroscopy excited by polarized laser light,‖ Biosens. Bioelectron. 26(7), 3167–3174 (2011). 25. H. Han, X. Yan, R. Dong, G. Ban, and K. Li, ―Analysis of serum from type II diabetes mellitus and diabetic complication using surface-enhanced Raman spectra (SERS),‖ Appl. Phys. B 94(4), 667–672 (2009). 26. X. Huang and M. El-Sayed, ―Gold nanoparticles: optical properties and implementations in cancer diagnosis and photothermal therapy,‖ J. Advert. Res. 1(1), 13–28 (2010). 27. S. Feng, J. Lin, M. Cheng, Y. Z. Li, G. Chen, Z. Huang, Y. Yu, R. Chen, and H. Zeng, ―Gold nanoparticle based surface-enhanced Raman scattering spectroscopy of cancerous and normal nasopharyngeal tissues under nearinfrared laser excitation,‖ Appl. Spectrosc. 63(10), 1089–1094 (2009). 28. K. Grabar, R. Freeman, M. Hommer, and M. Natan, ―Preparation and characterization of Au colloid monolayers,‖ Anal. Chem. 67(4), 735–743 (1995). 29. R. Liu, X. Zi, Y. Kang, M. Si, and Y. Wu, ―Surface-enhanced Raman scattering study of human serum on PVA Ag nanofilm prepared by using electrostatic self-assembly,‖ J. Raman Spectrosc. 42(2), 137–144 (2011). 30. J. W. Chan, D. S. Taylor, T. Zwerdling, S. M. Lane, K. Ihara, and T. Huser, ―Micro-Raman spectroscopy detects individual neoplastic and normal hematopoietic cells,‖ Biophys. J. 90(2), 648–656 (2006). 31. Z. Movasaghi, S. Rehman, and I. Rehman, ―Raman spectroscopy of biological tissues,‖ Appl. Spectrosc. Rev. 42(5), 493–541 (2007). 32. J. L. Pichardo-Molina, C. Frausto-Reyes, O. Barbosa-García, R. Huerta-Franco, J. L. González-Trujillo, C. A. Ramírez-Alvarado, G. Gutiérrez-Juárez, and C. Medina-Gutiérrez, ―Raman spectroscopy and multivariate analysis of serum samples from breast cancer patients,‖ Lasers Med. Sci. 22(4), 229–236 (2007). 33. H. Yao, Z. Tao, M. Ai, L. Peng, G. Wang, B. He, and Y. Li, ―Raman spectroscopic analysis of apoptosis of single human gastric cancer cells,‖ Vib. Spectrosc. 50(2), 193–197 (2009). #146382 $15.00 USD Received 22 Apr 2011; revised 11 Jun 2011; accepted 11 Jun 2011; published 29 Jun 2011 (C) 2011 OSA 4 July 2011 / Vol. 19, No. 14 / OPTICS EXPRESS 13566 34. L. Brancaleon, A. J. Durkin, J. H. Tu, G. Menaker, J. D. Fallon, and N. Kollias, ―In vivo fluorescence spectroscopy of nonmelanoma skin cancer,‖ Photochem. Photobiol. 73(2), 178–183 (2001). 35. E. Gormally, E. Caboux, P. Vineis, and P. Hainaut, ―Circulating free DNA in plasma or serum as biomarker of carcinogenesis: practical aspects and biological significance,‖ Mutat. Res. 635(2-3), 105–117 (2007). 36. E. Benedetti, E. Bramanti, F. Papineschi, I. Rossi, and E. Benedetti, ―Determination of the relative amount of nucleic acids and proteins in leukemic and normal lymphocytes by means of Fourier transform infrared microspectroscopy,‖ Appl. Spectrosc. 51(6), 792–797 (1997). 37. C. P. Schultz, K. Z. Liu, J. B. Johnston, and H. H. Mantsch, ―Prognosis of chronic lymphocytic leukemia from infrared spectra of lymphocytes,‖ J. Mol. 


Introduction
Colorectal cancer (CRC) is the third most common cancer worldwide with an annual incidence of approximately 1,000,000 cases and an annual mortality of more than 500,000.The absolute number of cases will increase over the next 2 decades as a result of aging and expansion of populations in both developed and developing countries, which has been a major public health burden [1,2].Screening methods are available which can reduce the incidence by removal of adenomas and can reduce deaths in diagnosed cancer cases by earlier stage detection.However, there are many disadvantages in conventional colorectal cancer screening methods.For instance, colonoscopy screening method involves greater cost (need of a skilled examiner, time consuming) and inconvenience to the patient (whole bowel cleansing), and not all examinations visualize the entire colon.Excisional biopsies currently remain the standard approach for cancer diagnosis, though this method is invasive and impractical for mass screening of high-risk patients with multiple suspicious lesions.In brief, using the conventional colorectal cancer screening tools based on pathology, which depends on physician's skill, knowledge, experience and many other factors, it is difficult to identify early neoplasia or subtle lesions [2][3][4].
Raman spectroscopy (RS) based on the inelastic light scattering can provide important biochemical information of macromolecules such as proteins, nucleic acids and lipids, because each molecule has its own pattern of vibrations that can serve as a Raman biomarker [5].Recently, Raman spectroscopy has emerged as a novel nondestructive diagnostic tool for cancer detection and identification of malignancy at different stages of the evolution of neoplasia in tissue.For example, some groups have investigated the applications of laser Raman spectroscopy in differentiating normal and malignant tissues in various body sites, such as lung, stomach, bladder, breast, parathyroid, prostate and cervix [6][7][8][9][10].However, Raman scattering suffers the disadvantage of extremely poor efficiency because of its inherently small cross-section (e.g. 10 30 cm 2 per molecule).Besides, The Raman spectra of biological samples are often superimposed on top of a strong fluorescence background that may be overwhelming and make it difficult to extract the Raman signals.These main disadvantages make it a great challenge for practical applications of conventional Raman spectroscopy in medical diagnosis [11,12].
There has been renewed interest in Raman spectroscopy technique in the past two decades owing to the discovery of the surface-enhanced Raman spectroscopy (SERS).Surfaceenhanced Raman scattering was first reported by Fleischman et al. in 1974 [13].Recent reports show, with SERS technique, Raman signals can be enhanced by 13 to 15 orders of magnitude when the probed molecules are attached to nano-textured metallic surfaces, while the autofluorescence background can be greatly reduced at the same time [11,14].SERS technology greatly improves the detection sensitivity of Raman spectroscopy, and has drawn considerable attention due to its great potential in biomedicine.Many interesting reports have been published on the applications of SERS technology for detecting biological materials, such as DNA, RNA, glucose, and dipicolinic acid [15][16][17][18].Cancer diagnosis is another type of potential applications for SERS technique.Especially, SERS based immunoassay, which is relied on a specific interaction between an antigen and a complementary antibody, is developed for most current oncology applications.Studies have showed that SERS technique cannot only detect the recognized biomarker, but also be used to explore the novel and potential cancer biomarkers [19][20][21][22].
Blood samples are a preferable material for non-invasive diagnosis, which can be taken conveniently and even continuously throughout the treatment for diagnosed patient [23].Application of SERS for disease detection based on multivariate analysis at the blood level has been reported on nasopharyngeal [11], gastric plasma samples [24] and diabetes mellitus serum samples [25].All studies mentioned above used silver nanoparticles as the SERS-active nanostructures.However, gold nanoparticles are preferred over silver nanoparticles for many biomedical applications because of their favorable physical and chemical properties and biocompatibility [22,26,27].Moreover, with the near-infrared (NIR) excitation, gold colloidal clusters will have comparably good SERS enhancement factors as silver clusters [12].In this paper, we explored the use of gold nanoparticles and NIR laser excitation for SERS application in blood serum biochemical analysis and colorectal cancer detection.Both the empirical approach and the multivariate statistical techniques were employed to develop effective diagnostic algorithms for differentiations between health subjects and cancer patients.The receiver operating characteristic (ROC) curve was employed to assess and compare the accuracy of both diagnostic algorithms.To our knowledge, this is the first report on SERS serum analysis with gold nanoparticles for colorectal cancer detection.

Preparation of glod colloids
Stable gold nanocolloid solutions were prepared using the process developed by Grabar et al [28].In short, 500 ml of 1 mM HAuCL 4 was brought to a rolling boil with vigorous stirring.Rapid addition of 50 ml of 38.8 mM sodium citrate to the vortex of the solution resulted in a color change from pale yellow to burgundy.Boiling was continued for 10 min; the heating mantle was then removed, and stirring was continued for an additional 15 min.The resulting solution of colloidal particles is characterized by an absorption maximum at 527 nm (Fig. 1).The inserted picture in the figure shows a transmission electron microscopy (TEM) photograph of the prepared gold colloid.The particle sizes follow a normal distribution with a mean diameter of 43 nm and standard deviation of 6 nm.

Preparation of human serum samples
The experimental serums were obtained from 45 healthy volunteers as the control group and from 38 patients with confirmed clinical and histopathological diagnosis of colorectal cancers.The mean age for the control group was 41 years and for the cancer group was 58 years.Table 1 shows more detailed information on these patients.All patients were from the Fuzhou Tumor Hospital and had similar ethnic and socioeconomic backgrounds.After 12 hours of overnight fasting, a single 3 ml peripheral blood samples were obtained from the study subjects between 7:00-8:00 A.M. with the use of coagulant.Before SERS measurement, 30 µl serum was mixed with 30 µl gold colloidal nanoparticles.It was mixed with the pipette tip to create a mixture as homogenous as possible.The mixture was incubated for 2 h at 4 °C before measurement.Then, a drop of this mixture was transferred onto a rectangle aluminum plate for SERS analysis.

SERS measurements
A confocal Raman micro-spectrometer (Renishaw, Great Britain) and a 785 nm diode laser excitation was used for the measurement of SERS spectra in the range of 300-1800 cm 1 .The SERS spectra were acquired with a 10 s integration time in backscattering geometry using a microscope equipped with a Leica 20 × objective with a spectral resolution of 2 cm 1 ; the detection of Raman signal was carried out with a Peltier cooled charge-coupled device (CCD) camera.The volume illuminated by the laser was about 3 × 10 3 µl, and the number of measured particles was approximate 7.5 × 10 7 .The software package WIRE 2.0 (Renishaw) was employed for spectral acquisition and analysis.The frequency calibration was set by reference to the 520 cm 1 vibrational band of a silicon wafer.To reduce the spectral intensity variations between different spectra and enable a better comparative analysis of the spectral shapes, all measured SERS spectra were normalized to the integrated area under the curve in the 350-1700 cm 1 wavenumber range after the removal of fluorescence background from the original SERS data.The normalized mean SERS spectra obtained from 38 colorectal cancer patient serum samples and 45 normal subject serum samples with the standard deviations overlying as shaded color fill are shown in Fig. 3(A).Comparing the normalized mean SERS spectra, primary SERS peaks at 494, 589, 638, 725, 823, 881, 1004, 1074, 1206, 1322 and 1655 cm 1 can be consistently observed in both normal and cancer serum, with the strongest signals at 494, 638, 725, 1655 cm 1 .However, the significant Raman spectral differences also exist between normal and cancer serum.The normalized intensities of SERS peaks at 494, 638, 823, 1206 and 1655 cm 1 are lower for cancer samples than for normal samples, while SERS bands at 725 and 881 cm 1 are more intense in cancer samples.These normalized intensity differences can be viewed more clearly on the difference spectra between cancer and normal serum (bottom in Fig. 3(A)).In addition, the peak positions at 1365 cm 1 in normal serum appear to have shifted to 1394 cm 1 in cancer serum, which results in the bands for the regions 1326-1400 cm 1 appear broader for cancer serum compared to normal serum.Figure 3(B) shows a comparison of the mean intensities and standard deviations of the selected peaks with significant differences (Student's t-test analysis, p < 0.05) between colorectal cancer blood serum and normal blood serum.These most obvious differences can be found in the peaks at 494, 638, 725, 823, 881, 1206 and 1655 cm 1 .The difference spectrum reveals the changes of prominent SERS peaks occurring in cancer serum, confirming a potential role of serum SERS for colorectal cancer diagnosis.

Result of statistical analysis
The SERS peak intensities at 725 and 638 cm 1 appear obvious different between normal and cancer serum, which have been regarded as an important differential diagnosis of diseases [25].In this study, an empirical diagnostic algorithm based on the ratio of the SERS peak intensity at 725 cm 1 for adenine to the peak intensity at 638 cm 1 for tyrosine is employed to classify colorectal cancer and normal serum samples.Figure 4  We also employ the multivariate statistical method (e.g., PCA and LDA) by incorporating the entire SERS spectrum to determine the most diagnostically significant SERS features for improving serum analysis and classification.PCA is a multivariate technique used in spectroscopy, which defines a new dimensional space in which the major variance in the original data set can be captured and represented by only a few principal components (PCs) variables.These PCs are used to build a model with a resolution of recognition.LDA can maximize the variance between groups and minimize the variance within groups, by computing linear combinations of variables to determine directions in the spectral space.The fluorescence background of the original SERS data was firstly removed using a modified multi-polynomial fitting algorithm, then each spectrum was normalized by the integrated area under the curve, and after that the normalized whole SERS spectrum data set was fed into the SPSS software package (SPSS Inc., Chicago) for PCA-LDA analysis.
We found that three PCs (PC1, PC2 and PC3), accounting for 63% of the variance, are most diagnostically significant (p < 0.05) for discriminating normal and cancerous groups by independent-sample T test on all the PC scores comparing normal and cancerous groups.To illustrate the use of PC scores for diagnostic classification, direct comparisons between normal and cancer groups are presented in Fig. 5.The colorectal cancer data points and the normal serum data points are very well clustered into two separate groups based on different combinations of significant PCs, and the corresponding separation lines in Figs.5(A)-5(B) classify cancer from normal serum with the sensitivity of 84.2% and 92.1%; specificity of 93.3% and 95.6%, respectively.These results show that selection of different combinations of significant PCs will give different levels of accuracy for serum classification.To actually incorporate all significant spectral features, LDA was used to generate diagnostic algorithms using the PC score for the three most significant PCs (PC1, PC2 and PC3).To prevent over-training, the leave-one-out and cross-validation procedures were used.Figure 6 shows the posterior probabilities of belonging to the normal and colorectal cancer groups as calculated from empirical approach data set (I 725 /I 638 ) and multivariate statistical techniques data set (significant PCs) in the LDA model.Using a discrimination threshold of 0.5, the diagnostic sensitivity for detecting colorectal cancer was 68.4% and 97.4% for the empirical approach and multivariate statistical techniques, respectively.The corresponding diagnostic specificities for each method were 95.6% and 100%.
To further evaluate and compare the performance of the empirical and multivariate approaches for colorectal cancer classification using the same SERS data set, receiver operating characteristic (ROC) curves were generated (Fig. 7) from the scatter plots in Fig. 6 at different threshold levels.The line, which is called the chance diagonal, segment from 0,0 to 1,1 has an area of 0.5.Only if its ROC curve area is greater than 0.5, diagnostic tests have ability to discriminate between patients with and without cancer.A comparative evaluation of the ROC curves indicates that PCA-LDA-based diagnostic algorithm gives more effective diagnostic performances for differentiation of colorectal cancer from normal serum samples, as illustrated by the improvement in the diagnostic sensitivities and specificities.The integration areas under the ROC curves are 1 and 0.896 for PCA-LDA-based diagnostic algorithms and the nonparametric intensity ratio algorithm, respectively.These results further demonstrate that PCA-LDA-based diagnostic algorithms yield a better diagnostics accuracy than the empirical approach.Fig. 7. Comparison of receiver operating characteristic (ROC) curves of discrimination results for SERS spectra utilizing the PCA-LDA-based spectral classification with leave-one-out, cross-validation method and the empirical approach using SERS spectra intensity ratio of I725/I638.The integration areas under the ROC curves are 1 and 0.896 for PCA-LDA based diagnostic algorithm and intensity ratio algorithm, respectively.The dotted line called the chance diagonal from 0,0 to 1,1 has an area of 0.5.

SERS spectra
The results of our exploratory study demonstrated that there are specific differences in SERS spectra between cancerous and normal serum, suggesting great potential for serum SERS in colorectal cancer detection and screening applications.In the range 350-1700 cm -1 , SERS spectra of human serum are dominated by a number of vibrational modes of biomolecules, such as nucleic acids, proteins and lipids, which may be altered in quantity or confirmation associated with colorectal cancer developments.To better understand the molecular basis for the observed SERS spectra of human serum, Table 2 lists tentative assignments for the observed SERS bands, according to the literature data [6,11,25,[29][30][31][32][33].Distinctive SERS features and intensity differences for cancer and normal serum could reflect molecular changes associated with cancer formation.For instance, the prominent SERS peak at 1655 cm 1 can be attributed to the amide I band of proteins in the α-helix conformation and human serum albumin (HSA) which is a principal extracellular transport protein.This protein is a globular molecule containing 17 disulfide bridges that also produce SERS peak characteristic of S-S stretching vibration at 494 cm 1 [29].However, in cancer serum, both of these two SERS bands are much reduced in normalized intensity, indicating that colorectal cancer may be associated with a decrease in the relative amounts of protein.We have observed similar decrease in our previous SERS study of nasopharyngeal and gastric cancer blood plasma [11,24].Besides, the SERS band of tryptophan at 881 cm -1 show higher percentage signals in cancer serum, indicating an increase in the percentage of tryptophan content relative to the total SERS-active components in the serum of colorectal cancer patients.Brancaleon et al. also observed an increase of tryptophan residues in skin tumor by fluorescence spectroscopy [34].The peak at 1365 cm -1 in normal serum due to the tryptophan is shifted to 1394 cm -1 in cancer serum.It is not entirely clear what might cause such a consistent shift in the cancer serum, but it appears plausible that this could be due to a somewhat different protein composition (different amino acids right next to tryptophan) [30].
The SERS band at 725 cm 1 corresponds to the C-H bending mode of adenine, and is higher in cancer serum than in normal serum, suggesting an abnormal metabolism of DNA or RNA bases in the serum of colorectal cancer patients.This is in agreement with SERS study of nasopharyngeal cancer blood plasma [11].The reason for increased cell-free nucleic acid levels in cancer patients' blood remains largely unknown.Two main mechanisms have been proposed: apoptosis and necrosis, or release of intact cells in the bloodstream and their subsequent lysis [35].The band can also be used as an important ‗fingerprint' for disease diagnosis [25].The SERS bands of tyrosine (638 cm 1 , 823 cm 1 and 1206 cm 1 ) and larginine (494 cm 1 ) in serum of cancer patients show lower percentage signals than those of normal serum, suggesting a decrease in the percentage of certain amino acids contents relative to the total SERS-active components in serum of colorectal cancer patients.The tumor's vigorous metabolism may lead to these changes, which is in agreement with other biochemical analysis results of tumor tissues [11].

Statistical analysis
Recently, some groups have developed simple but effective diagnostic algorithms based on the empirical analysis of Raman spectra in terms of peak intensity or peak intensity ratio measurements.For example, the ratio of intensities at 1455 and 1655 cm -1 has been used to classify cancer and normal tissue in the cervix [10] and cancer and normal serum in the breast [32].The ratio of Raman peak intensities at 725cm 1 and 638 cm 1 was considered as an important ‗fingerprint' for diabetes mellitus diagnosis at the serum level by Han et al [25].In this study, the empirical diagnostic algorithm based on the ratio of the SERS peak intensity at 725 cm 1 to the peak intensity at 638 cm 1 was also explored to classify colorectal cancer and normal serum samples (Fig. 4(A)).It was found that the ratio of the SERS peak intensity at 725 cm 1 for adenine to the peak intensity at 638 cm 1 for tyrosine was significantly higher in cancer serum than in normal serum, and the classification results showed a sensitivity and specificity of 68.4% and 95.6%, respectively.
Further investigation shows that other different ratios such as I 725 /I 494 and I 725 /I 1655 had also been found to be effective diagnostic algorithms for colorectal cancer differentiation.Comparison of classification results based on different SERS peak intensity ratios, the ratio of I 725 /I 638 gave the best diagnostic sensitivity for colorectal cancer.According to the tentative assignments of the serum SERS bands, the significant difference of these intensity ratios between normal and colorectal cancer serum may reflect the changes in the relative amount of potential biological markers.It is notable that all the intensity ratios selected above were significantly higher in cancer serum than in normal serum, indicating increases in the nucleic acid content relative to proteins in cancerous samples, which is in consistent with biochemical analysis results of tumor cells.Earlier studies investigating the biochemical differences between normal and chronic lymphocytic leukemic cells using Fourier transform infrared micro-spectroscopy also observed increases in the DNA content relative to proteins in neoplastic cells [36,37].In addition, the intensity ratio of I 725 vs.I 638 exhibited greater fluctuation in cancer group, which is in agreement with Han's study [25].This is explainable.For healthy people, the proportions of different components in serum are relatively stable.However, in pathological conditions, the cancer related blood serum constituents could be quite variable from patient to patient due to the heterogeneity of cancer development.Hence, the above nonparametric intensity ratios may potentially be employed as effective diagnostic algorithms for colorectal cancer detection.
Note that the simplistic empirical analysis employed above uses only limited SERS peaks for group classification and most of the information contained in the SERS spectra has not been utilized.Since human serum is very complex, it is likely that there are many biochemical species influencing tumor concurrently.Moreover, as cancer belongs to part of a widely accepted multistep, continuum progression cascade from normal to carcinoma, it implies subtle and vague molecular distinction that may render characterization and discrimination tougher for SERS analysis [7].It is highly desirable to develop robust diagnostic approaches to extract all possible diagnostic information contained in serum SERS spectra for well correlation with serum changes associated with neoplastic transformation.Therefore, a multivariate statistical analysis (e.g., PCA and LDA) [11,32,38] which utilizes the entire spectrum and automatically determines the most diagnostically significant features, may improve the efficiency of the method for serum analysis and classification.PCA was performed to reduce the large amount of data contained in the measured SERS spectra into a few important principal components.Figure 5(A) shows that the scores of PC1 and PC2 for the normal and colorectal cancer groups form distinct and separate clusters.The normal group forms one cluster and the cancer group forms another cluster.If we used the PC1 and the PC3 for the two axes, an analogous comparison for the normal and cancer group is shown in Fig. 5(B).We can clearly see that they were distributed in separate areas, which means that we are able to discriminate between the SERS spectra of the colorectal cancer group and the healthy control group.The diagnostic sensitivity and specificity of 97.4% and 100%, respectively, can be achieved for identifying cancer from normal serum using the PCA-LDA-based spectral classification with the leave-one-out, cross-validation method, which had a significant improvement in diagnostic accuracy compared with the empirical method.
The diagnostic accuracy differences between the empirical approach and the multivariate statistical techniques can be viewed more clearly in Fig. 6.Using a discrimination threshold of 0.5 in the LDA model, the diagnostic efficiency for detecting colorectal cancer is obviously higher in multivariate statistical techniques than in empirical approach.Receiver operating characteristic (ROC) curves were employed to further compare the performance of the empirical and multivariate approaches for colorectal cancer classification.One of the most popular measures of the accuracy of a diagnostic test is the area under the ROC curve.The ROC curve area can take on values between 0.0 and 1.0.And the closer the ROC curve area is to 1.0, the better the diagnostic test [39,40].The integration areas under the ROC curves are 1 and 0.896 for PCA-LDA-based diagnostic algorithms and the nonparametric intensity ratio algorithm, respectively.These results demonstrate that both diagnostic algorithms have ability to discriminate colorectal cancer group from normal group, moreover, the PCA-LDA-based diagnostic algorithm employing the entire spectral features of SERS spectra gives more effective diagnostic performance for colorectal cancer detection than the empirical diagnostic algorithm.

Conclusions
Gold nanoparticle based surface-enhanced Raman spectroscopy was applied to analyze the blood serum from colorectal cancer patients and healthy volunteers.Using empirical diagnostic algorithm and PCA-LDA multivariate analysis, we were able to differentiate colorectal cancer from normal with high diagnostic sensitivity and specificity.Tentative assignments of the Raman bands in the measured SERS spectra demonstrated interesting cancer specific biomolecular changes, including an increase in the relative amounts of nucleic acid, a decrease in the percentage of saccharide and proteins contents in the blood serum of colorectal cancer patients as compared to that of healthy subjects.The empirical diagnostic algorithm based on the ratio of the SERS peak intensities achieved a diagnostic sensitivity around 60%, whereas the diagnostic algorithms based on PCA-LDA yielded a diagnostic sensitivity of 97.4% for differentiating cancerous samples from normal samples.The results from this exploratory study demonstrated great promise for developing gold nanoparticle based SERS serum analysis combined with PCA-LDA-based diagnostic algorithms into a clinical tool for non-invasive detection and screening of colorectal cancers.Our next step will be to conduct more detailed prospective studies to verify the reliability of this potential cancer detection method.

Fig. 1 .
Fig. 1.The UV/visible absorption spectrum of the Au colloid.The absorption maximum is located at 527 nm.The inserted picture shows the TEM micrograph of Au nanoparticles.

Fig. 2 .
Fig. 2. (A) SERS spectrum of the blood serum sample from a patient with colorectal cancer obtained by mixing the serum with Au colloid at a 1:1 proportion, (B) the regular Raman spectrum of the same serum sample without the Au colloid and (C) the background Raman signal of the coagulant agent mixed with Au colloid.To study the gold colloid enhancement effects on the human serum Raman scattering, we have recorded the Raman spectra and SERS spectra of serum samples from healthy group and colorectal cancer group.Figures 2(A)-2(C) show the SERS spectra of serum with added Au sol, the regular Raman spectrum of serum without Au sol and the background Raman signal of the coagulant with added Au sol.The three spectra were measured under the same instrumentation set-up of 5 mW incident laser power and 10 s spectral data acquisition time.A comparison of Fig. 2(A) and Fig. 2(B) shows that the intensity of the many dominant vibration bands increases dramatically, indicating that there is a strong interaction between the gold colloids and the serum.Because of this interaction, biochemical substances of serum are closely attached to nano-textured gold colloid surfaces, thus leading to an extraordinary enhancement in the intensity of the Raman scattering.Only a few Raman peaks could be observed in the native serum without the addition of gold solution because most of the Raman signals are masked by the large fluorescence background.An impressive decrease in the intensity of the fluorescence background and clearly resolved sharp Raman bands were observed in the SERS spectra.Moreover, Fig. 2(C) shows there is no interference signal from the coagulant with added Au sol in the interested spectral range.

Fig. 3 .
Fig. 3. (A) Comparison of the mean spectrum for the colorectal cancer serum (blue curve, n = 38) versus that of the normal serum (red curve, n = 45) samples.Each spectrum was normalized to the integrated area under the curve to correct for variations in absolute spectral intensity.The shaded areas represent the standard deviations of the means.Also shown at the bottom is the difference spectrum.(B) Comparison of the mean intensities and standard deviations of the selected peaks with the most distinguishable differences between colorectal cancer serum (blue pillar) and normal serum (red pillar).
(A) shows a scatter plot for the intensity ratio of I 725 vs.I 638 for each serum sample.The mean value (mean ± SD) of this ratio for cancer serum samples (1.54 ± 0.70,n = 38) is significantly different from the mean ratio for normal serum samples (0.64 ± 0.23,n = 45) with p < 0.05 by Student's t-test.The decision line (I 725 /I 638 = 1.11) separates cancer group from normal group with a sensitivity and specificity of 68.4% and 95.6%, respectively.Further investigation also shows that other intensity ratios including the SERS peak intensity at 725 cm 1 (adenine) with respect to the SERS peak intensities at 494 cm 1 (l-arginine) and 1655 cm 1 (amide I band of proteins) respectively, are also statistically significantly different (p < 0.05) between colorectal cancer and normal serum samples (Figs.4(B)-4(C)).The mean values (mean ± SD) of the ratios of I 725 /I 494 and I 725 /I 1655 for cancer serum samples are 2.86 ± 1.85 and 1.21 ± 0.62; the mean values for normal serum samples are 1.02 ± 0.84 and 0.63 ± 0.67, respectively.The dotted lines (I 725 /I 494 = 1.95;I 725 /I 1655 = 0.92) as diagnostic threshold classify cancer from normal with sensitivity of 57.9% and 60.5%; specificity of 97.8% and 91.1%, respectively.

Fig. 5 .
Fig. 5. (A) Plots of the first principal component (PC1) versus the second principal component (PC2) for normal group versus colorectal cancer group.The dotted line (PC2 = 1.68PC1 + 0.13) as diagnostic algorithm separates the two groups with sensitivity of 84.2% and specificity of 93.3%.(B) Plot of the first principal component (PC1) versus the third principal component (PC3) for normal group versus colorectal cancer group.The dotted line (PC3 = 1.14PC1 + 0.11) as diagnostic algorithm separates the two groups with sensitivity of 92.1% and specificity of 95.6%.

Fig. 6 .
Fig. 6.Scatter plots of the posterior probability of belonging to the normal and colorectal cancer categories calculated from the data sets with (A) empirical approach (I725/I638), (B) multivariate statistical techniques (significant PCs) in the LDA model.The posterior probability corresponding to the dashed separation line is 0.5.