1. Introduction

Worldwide, colorectal cancer (CRC) is the third most common type of cancer and fourth leading cause of cancer-related death [1]. Around 85% of CRC are considered sporadic and the development of CRC is commonly believed to follow the adenoma-carcinoma pathway, where the progression from adenomas to invasive cancer usually takes around 10 years [2]. Early diagnosis is the key to long-term patient survival. Despite the high cost and intensive treatment required, 40-50% symptomatic CRC patients will die of metastatic disease eventually. However, patients treated at an early stage diagnosis through the removal of adenomatous polyps, the CRC precursors, have achieved significantly improved survival [3, 4]. Programmatic screening for the general population has shown to be cost-effective in reducing the incidence and mortality of CRC [5]. Current CRC screening strategy relies on a broad range of test techniques that could be loosely classified as early detection tools and cancer prevention tools [6]. Clinically validated fecal occult blood test (FOBT) and especially the more sensitive fecal immunochemical tests (FIT) are non-invasive, low in cost and simple to use when compared to other screen modalities [7]. Other screening techniques include flexible sigmoidoscopy, total colonoscopy, and computed tomographic colonography (CTC), etc [8].

Several studies comparing the diagnostic performance of FIT with high-sensitivity guaiac-based FOBT tests found no apparent superiority of one method over the other [9]. With specificity above 95%, sensitivities ranged from 65.8% to 81.8% and from 27.1% to 41.3% for CRC and advanced adenomas (AAs) respectively. The fact that asymptomatic CRC and 95% of adenomas do not bleed and do not ulcerate explains the relatively low sensitivities of fecal blood based tests for precursor detections [10]. Besides diagnostic accuracy of screening tests, patient compliance is the other major determinant of the effectiveness of screening program. With a compliance rate lower than 50% for FOBT and a 10% higher rate for FIT, the population-based screening efforts have been greatly compromised [11]. Recent research looks at finding molecular biomarkers in body fluids such as blood and urine, as well as in feces [6]. Among these tests, blood test is perceived as the choice of method because of patients’ familiarity, minimal invasiveness, and relative ease of sample collection during regular clinic visit. Because colorectal carcinogenesis is the result of genetic mutations, attempts have been made to identify mutated deoxyribonucleic acid (DNA), micro ribonucleic acid (microRNA), and mutated proteins related to CRC [12]. Most recently, serum metabolites have been studied to be used as potential biomarkers for early CRC detection [13]. Some important biomarkers characteristic of early CRC development such as methylated Septin 9 (SEPT9) and miRNA-21 have been reported [14, 15]. These experimental screening techniques, though promising to provide potential markers for early CRC diagnosis, still require further validation in population-based studies [16]. On the other hand, the relatively high number of false-positive results, higher cost and complexity of some of the test procedures for individual biomarkers may be some of the challenges for alternative screening tests to be implemented clinically [17].

The ideal screening test needs to be accurate, minimally invasive, convenient, safe, and affordable [18]. Surface-enhanced Raman spectroscopy (SERS) is one of the viable options as a highly sensitive analytical technique for biomolecules [19]. Raman scattering is inherently weak with a tiny portion (10⁻⁶~10⁻⁸) of the incident photons going through the inelastic scattering [20]. Conventional Raman spectroscopy relies on shining intense laser light onto target samples to cause molecules to interact with incident light, producing Raman peaks at frequencies other than the incident light. The Raman peaks are representative of the underlying molecular structures and serve as molecular “fingerprints” for the analyte under investigation [21]. One of the main problems with spontaneous Raman spectroscopy is that it does not work well with detecting biological molecules at low concentrations because of the rather small cross-sections (10⁻³⁰ cm²) per molecule [22]. In addition, biological samples generate strong auto-fluorescence background that could overwhelm the weak Raman signals, making their identification difficult or impossible [23]. Surface-enhanced Raman scattering (SERS) effect results from molecules attached to the nano-structured metallic surfaces and up to 10⁶ or even 10¹⁴ times signal enhancement could be achieved [24]. With the signal enhancement of this magnitude, detection of biological molecules at ultra low concentration level is possible. Research in using SERS for detecting serum/blood plasma biomarkers for gastric cancer, nasopharyngeal cancer, and CRC etc. has been previously reported [25, 26]. The capability of SERS to differentiate between blood samples from healthy volunteers and CRC patients based on individual blood biomarkers such as RNA, carcinoembryonic antigen (CEA), and proteins have also been demonstrated [27–29]. No SERS based serum/plasma analysis for colonic polyp detection has been reported.

In this study, we reported the use of label-free SERS for analyzing blood plasma obtained from CRC patients, patients diagnosed with adnomatous polyps, and healthy volunteers. The objective is to prove the efficacy of silver nano-particle based SERS for detecting not only CRCs, but also polyps. As a blood based test, the proposed methodology expects to be translated into an alternative CRC screening test that can accurately predict CRC precursor lesions, achieving better CRC prevention.

2. Materials and methods

2.1 Silver nano-particle preparation

Preparation method for the silver (Ag) nanocollodial solution proposed by Leopold and Lendl was followed [30]. The prepared Ag particles have a mean diameter of 34 nm with a standard deviation of 5 nm as measured by transmission electron microscopy (TEM) [31]. The silver colloidal solution was centrifuged at 10,000 rpm for 10 min, of which a portion of the supernatant was discarded with the remaining concentrated solution kept for mixing with plasma samples. One part blood plasma and one part silver nanoparticles solution were mixed within the pipette tip to create a homogenous mixture. The blood plasma-silver nanoparticle mixture was incubated for 2 h at 4 °C before being transferred onto a rectangle aluminum plate for SERS measurement.

2.2 Human blood plasma

Participants were recruited at the gastroenterology clinic of the Vancouver General Hospital during scheduled colonoscopy exam. Ethnical approval was granted by the University of British Columbia (Certificate #: H14-02808). Blood samples from 21 CRC patients, 23 patients with adenomatous polyps, and 25 healthy volunteers were collected. Plasma samples were obtained by centrifuging the blood samples at 2000 rpm for 15 min. Patients have been confirmed colonoscopically and histopathologically and cleared of possible cancers at other body sites. Subjects in healthy control group went through total colonoscopy with no adenomas detected at endoscope. Medical histories for the healthy controls were also scrutinized to ensure no synchronous cancers or other medical conditions were developing at the time of blood drawn. Sample size is calculated based on the average intensity difference at wavenumber 1652 cm⁻¹, given a statistical power of 90% at a significance level of 0.05.

2.3 Raman measurement and data analysis methods

A fully integrated fiber optic probe real-time Raman spectrometer (Aura, Verisante Technology Inc., Vancouver, Canada) was used to take the SERS spectra. The 785 nm excitation laser generates 150 mW output power at the fiber optic probe tip. With an integration time of 1 s, spectra were measured in the spectral range from 500 to 1800 cm⁻¹ in duplicates. Fluorescence background in the raw data was removed to generate pure SERS spectra using a modified polynomial fitting algorithm described in our previous work [32]. For classification of Raman spectra, partial least squares (PLS) was used to reduce data dimensionality first and the PLS scores were then fed into a linear discriminant analysis (LDA) classifier to predict groupings between cancer/polyp patients and the healthy controls [33]. Classification performance was estimated using Leave-one-out cross validation (LOOCV). The LDA model also produced the posterior probabilities for each sample belonging to a particular class [34]. Receiver operating characteristic (ROC) curve was generated and area under curve (AUC) was calculated accordingly [35]. All computation was carried out in MatLAB (Mathworks, Natick, MA).

3. Results and discussion

To facilitate accurate spectral shape analysis, all measured plasma SERS spectra were normalized to the integrated area under the curve in the 550-1730 cm⁻¹ wavenumber range after the fluorescence background was removed [31]. In general, Raman peak intensity variation adhered to a sequence changing from normal to polyps followed by CRC at most dominant spectral peaks. Considering the fact that adenomatous polyp is the intermediate pathological state between normal and CRC, this varying pattern may indirectly corroborate the biological difference among the three groups. The polyps are different from normal and the CRC with their difference spectra varying according to similar trends in most spectral regions. There also exist regions where CRC exhibited significant difference from the other two groups, i.e., around 632 cm⁻¹ and regions between 1600 and 1700 cm⁻¹. Figure 1(A) compares the normalized mean SERS spectra of plasma samples for all three groups. Although the mean SERS spectra for normal and polyps may have small difference, when compared to the difference between them and the spectra of CRC plasma, such spectral difference is critical when discriminating polyps from normal group. As is evident in Fig. 1, there are obvious SERS spectral differences between normal and tumor plasma samples, but they also share common spectral peaks at 632, 725, 802, 892, 1004, 1130, 1202, 1278, 1368 and 1659 cm⁻¹. The normalized intensities of SERS bands at 725, 892, 1004 and 1368 cm⁻¹ are more intense for CRC plasma than for those of normal and polyp samples, while SERS peaks at 632, 1130 and 1659cm⁻¹ are stronger in normal and polyp plasma spectra compared to CRC spectra. In addition, the spectral peaks at 892 and 1368 cm⁻¹ in normal and polyp plasma appear to have shifted to 902 and 1349 cm⁻¹ in SERS spectra of CRC plasma samples, respectively.

 

Fig. 1 (A) Comparison of normalized mean SERS spectra for 25 normal, 23 adenomatous polyps, and 21 CRC plasma samples. (B) Difference spectra calculated from the mean SERS spectra among the three plasma types.

Download Full Size | PDF

These changes in mean SERS spectra are even more evident in the corresponding difference spectra among the three groups in Fig. 1(B). It can be seen that the intensity of difference spectrum for CRC versus normal and CRC versus polyps are greater than that of normal versus polyps. In addition, the two difference spectra for CRC versus normal and CRC versus polyps are very similar. The close resemblance between mean SERS spectra for normal and polyp plasma result in the relatively smaller difference between spectra for normal versus polyp groups.

The significant SERS spectra changes, including SERS peak intensities, peak positions and spectral bandwidths, in the spectral ranges of 600-780, 840-1162, 1230-1420, and 1475-1700 cm⁻¹, contain signals primarily related to nucleic acid, proteins, phospholipids, lipids and amino acids etc. In particular, eight SERS peaks located at 632, 725, 902, 1004, 1128, 1275, 1350 and 1664 cm⁻¹ were identified as important spectral bands. Comparison of the SERS peak intensities and their ± 1 standard deviation (SD) at the eight identified spectral peaks is shown in Fig. 2. CRC plasma showed relatively higher intensities at 725, 902, 1004, 1275 and 1350 cm⁻¹ but exhibited decreased SERS peak intensities at 632, 1128 and 1664 cm⁻¹ when compared to SERS peaks for normal and polyp plasma samples. In most Raman bands, the polyp group mean intensity falls between the normal group and the cancer group (725, 1128, 1275, 1350, 1664 cm⁻¹). However, there is one band where the polyp group has higher mean intensity than both the normal group and the cancer group (632 cm⁻¹). And there are two bands where the polyp group has lower intensity than both the normal group and the cancer group (902 cm⁻¹, 1004 cm⁻¹). These spectral variations indicate that there exist significant differences in the percentage of specific biomolecules relative to the total SERS-active constituents in plasma belonging to the three different pathological groups. The biomolecular differences suggest that blood plasma based SERS spectroscopy possesses good diagnostic potential for noninvasive screening of adenomatous polyps and colorectal cancers.

 

Fig. 2 Histogram displaying the eight SERS peak intensities (mean ± 1 SD) for normal (gray), adenomatous polyps (blue), and CRC (red) plasma samples. The differences are statistically significant (p<0.01) between normal and CRC for all SERS peaks, between adenomatous polyps and CRC for all SERS peaks, and between normal and adenomatous polyps for the 1664 cm⁻¹ peak.

Download Full Size | PDF

To better understand the molecular basis underlying SERS spectra, Table 1 lists tentative assignments for the observed SERS bands according to literatures [31,36]. The SERS spectral features and intensity differences among normal, polyps and CRC groups could reflect molecular and cellular changes associated with malignant transformation. For example, the plasma SERS peak intensities at 632, 725, 902, 1004, 1128, 1275, 1350 and 1664 cm⁻¹ experienced similar alterations (increase/decrease) in both normal and polyps samples as compared to CRC plasma samples. For the three SERS peaks at 1004, 1275 and 1350 cm⁻¹ (Fig. 2), the Raman peak intensities were almost identical for normal and polyps plasma. This indicates that normal and polyps plasma samples may still contain some similar constituents or conformation in plasma biomolecules.

Table 1. SERS peak positions and vibrational mode assignments

View Table

The SERS peak intensities at 632 cm⁻¹ due to the C-S stretching vibration of tyrosine was weaker for CRC plasma samples than those for normal and polyps subjects. Such decreased intensity indicated that cancer blood was associated with a decrease in the relative amount of amino acid. Similar change was also observed in nasopharyngeal [31] and gastric cancer [37] blood samples. An increased SERS signal at 725cm⁻¹ was found for CRC cancer samples as compared to normal and polyp samples, suggesting that CRC plasma contained relatively increased amount of nucleic acid bases compared to all SERS-active constituents in the plasma. A similar increase pattern at 725cm⁻¹ was reported in our previous SERS study of nasopharyngeal [31], gastric [37] and cervical cancer [23]. Compared to healthy control subjects, circulating plasma DNA has been reported in considerably elevated concentrations in multiple solid malignancies such as nasopharyngeal, esophageal, breast, lung, liver and prostate cancers [38]. These studies suggested that plasma DNA alterations were detectable using SERS. The peak of 725 cm⁻¹ may have the potential to be used as a diagnostic biomarker for early cancer detection [39]. In addition, an increase in SERS peak intensities at 902 cm⁻¹(C-C of skeletal mode), 1004 cm⁻¹(νs(C-C) of phenylalanine), 1275 cm⁻¹(Amide III of α-helix) and 1350 cm⁻¹(Guanine) and a decrease in SERS bands of 1126 cm⁻¹(ν(C-N) of D-mannos) and 1664 cm⁻¹(ν(C = O) of α-helix and collagen) were found in plasma of CRC patients, suggesting a special biochemical changes in either quantity or structure in the CRC plasma samples. The abnormal metabolism during cancer progression may have caused these changes, which are in good agreement with biochemical analysis results of our previous CRC blood serum detection [25].

Qualitative analysis of SERS peak intensities only provided limited information regarding Raman peak variation patterns, and there are many other significant variations of SERS spectra. To develop diagnostic algorithms for differentiation between normal, polyps and CRC plasma samples, multivariate statistical analysis based on PLS-DA was used to incorporate the entire SERS spectrum and determine the most significant features to improve plasma analysis efficiency. Leave-one-out cross validation was subsequently employed to generate diagnostic algorithms. To avoid over fitting, only the first 7 latent variables (LVs) were employed for discriminate analysis. This number is determined by taking the sample size of the smallest group (CRC plasma, n = 21) divided by 3 according to the rule of thumb to ensure model robustness [40]. Figure 3(A) shows the posterior probabilities of the samples belonging to the polyps plus CRC group (versus the normal group) as calculated by the PLS-DA method. Using a discrimination threshold of 0.5, a diagnostic sensitivity (86.4%) and specificity (80%) for differentiating polyps and CRC samples from normal plasma samples were achieved. The SERS data set containing polyps and CRC samples were subsequently subject to PLS-DA analysis. Figure 3(B) shows the posterior probabilities of plasma samples belonging to the CRC group (versus the polyp group) calculated by the PLS-DA method. The corresponding diagnostic sensitivity and specificity for discriminating polyps samples from CRC samples were 71.4% and 95.6%, respectively. Figure 3(C) indicates the posterior probabilities of plasma samples belonging to the polyp group (versus the normal group) calculated by the PLS-DA method. The diagnostic sensitivity and specificity for differentiating polyp samples from normal samples were 91.3% and 80%, respectively.

 

Fig. 3 Scatter plots of the posterior probability according to the normal, polyps and colorectal cancer categories calculated from the SERS data sets using different grouping methods. (A) Normal Vs Polyps plus Cancer, (B) Polyps Vs Cancer. (C) Normal Vs Polyps.

Download Full Size | PDF

To evaluate and compare the performance of the PLS-DA based diagnostic algorithm for polyp and CRC diagnosis, the ROCs (Fig. 4) were generated from the scatter plots in Fig. 3 by varying the discrimination threshold. The areas under the ROC curve were 0.938, 0.869 and 0.945 for the three data sets for polyps plus CRC samples versus normal, polyps versus CRC samples, and polyps versus normal samples, respectively. The result demonstrated that the plasma SERS spectra hold great promise for polyps and CRC detection with high sensitivity and specificity.

 

Fig. 4 Receiver operating characteristic (ROC) curves of classification results for different groupings of plasma samples generated using PLS-DA analysis. The integration areas under the ROC curves (AUC) are 0.938, 0.869 and 0.945, respectively.

Download Full Size | PDF

4. Conclusion

Early detection of precancerous lesions is crucial to successful treatment of colorectal cancer (CRC). Population-based screening program can effectively decrease both incidence and mortality rates. However, existing screening techniques such as fecal immunochemical tests (FITs) falls short in both detecting adenomas polyps and patient compliance rate to achieve effective CRC prevention. This paper explored the use of surface-enhanced Raman spectroscopy based blood plasma analysis as a non-invasive, accurate, and convenient alternative screening test for detecting both CRC and polyps for the first time. Partial least squares discriminant analysis (PLS-DA) with leave-one-out cross validation was used to generate classification models. For CRC and polyps detection, diagnostic sensitivity of 86.4% and specificities of 80.0% were achieved.