5.1 Neurosurgery
In 2017, there was an estimate of over 23,000 cases of brain and other nervous system cancers in the USA with a 70% mortality rate [
28]. According to the most recent data (2010–2014) from the Central Brain Tumor Registry of the United States (CBTRUS), brain and central nervous system cancers were the fifth most common cause of death for ages 15–39 [
29]. Glioblastomas, grade IV according to the World Health Organization (WHO), accounted for 14.9% of brain and CNS tumors and 47.1% of malignant tumors with a 4-year survival rate of 7.1% [
29]. Petrecca et al. analyzed 20 patients and found that in 17 patients the tumor recurred only at the resection margin; thus, complete tumor resection is crucial for patient longevity [
30]. Stummer et al. found that survival for patients with no residual tumor was, on average, 23.6 months; for patients with residual tumors < 1.5 cm survival was, on average, 16.9 months; and patients with residual tumors > 1.5 cm survival was, on average, 13.9 months [
31]. This finding underlies the importance of maximum tumor resection during surgery. One of the characteristics of glioblastomas is that it grows in a diffuse manner beyond the primary tumor location. Current image modalities used in presurgical imaging, MRI, do not capture the diffuse nature of glioblastomas. MRI imaging can suffer from brain shift between presurgical pictures and intrasurgery due to gravity, intrasurgical deformation, tumor resection, brain swelling, and cerebrospinal fluid [
32,
33]. Raman spectroscopy is a potential modality that can identify the margins of the tumor intraoperatively.
The majority of research into using RS for brain tumor assessment has been done using standard RS [
33‐
45]. Kast et al. and Kalkanis et al. [
35,
40] demonstrated RS’s ability to distinguish between white matter, gray matter, glioblastoma, and necrosis. Kast et al. created images from frozen sections of brain tissue samples using Raman peak intensities at 1004, 1300:1344, and 1660 cm
−1 which are indicative of protein and lipid content. Raman spectra were acquired on five frozen section tissues (one normal, one necrotic, one GBM, and two infiltrating glioma) with an inVia Raman microscope (Renishaw) using an excitation wavelength of 785 nm. The sections were mapped in their entirety using a 300-μm
2 step size. Smaller regions of interest were also mapped using a 25-μm step size, with each step corresponding to a discrete Raman spectrum. For each Raman image, the pixels were comprised of data from the selected Raman features. Each peak (or peak ratio) was assigned a color: red (1004 cm
−1), green (1300:1344 cm
−1), or blue (1660 cm
−1). The colored images allow interpretation of boundaries between gray matter, white matter, and diseased tissue that corresponded with the findings from adjacent hematoxylin and eosin-stained sections. Performing leave-one-out discriminant function analysis using the three Raman features provided more than 90% classification accuracy [
35]. Kalkanis et al. used discriminant functional analysis to distinguish normal tissue, necrosis, and glioblastoma. The Raman spectra from 95 regions from 40 frozen tissue sections were acquired with an inVia Raman microscope (Renishaw) using an excitation wavelength of 785 nm. The spectra were split into a test set, a validation set, and a secondary validation set of tissue with regions containing freeze artifacts. Discriminant function analysis showed 99.6, 97.8, and 77.5% accuracy in distinguishing tissue types in the training, validation, and validation with freeze artifacts datasets, respectively. Decreased classification in the freeze artifacts group was due to tissue preparation damage [
40]. Jermyn et al. demonstrated that a handheld RS probe could detect cancer cells intraoperatively that could not be detected by T1-contrast-enhanced and T2-weighted MRI [
34]. The gliomas were detected with 93% sensitivity and specificity of 91% [
34]. The handheld fiber optic probe (EmVision LLC, FL, USA) was connected to a 785-nm laser (Innovative Photonic Solutions, NH, USA) and a high-resolution charge-coupled device spectroscopic detector (ANDOR Technology, Belfast, UK). The probe was placed in direct contact with the brain at the resection cavity margin for each measurement, with a 0.2-s acquisition time. A supervised machine learning boosted-trees classification algorithm that utilizes all spectral data was used to distinguish samples containing invasive cancer cells
versus normal brain. The use of a handheld RS probe that can be used intraoperatively is a significant advance and has been used in several studies to successfully identify cancerous cells [
33,
34,
37].
Recently, Desroches et al. used a RS needle biopsy system to ensure cells are collected from an area that is dense enough with cancer cells to provide accurate biopsy information, with proof of concept demonstrated during surgery on a pig [
45]. Following pig surgery, a different system was used intraoperatively during human glioma surgery to verify that it could detect cancer tissue in biopsy locations [
45]. A 671-nm spectrum stabilized near-infrared laser (Laser Quantum, Inc) was used for Raman excitation with spectra collected at 0.5 s acquisition time. Using high wavenumber Raman spectroscopy, dense cancer with > 60% cancer cells was detected
in situ during surgery with a sensitivity and specificity of 80 and 90%, respectively. The support vector machine (SVM) technique was used for RS tissue classification using 141 features of the spectra. Leave-one-out cross-validation was used to determine the classification accuracy, sensitivity, and specificity. These studies suggest that RS can be used prior to surgery to ensure the biopsy is taken from the correct area and intraoperatively to detect cancerous cells more effectively than current modalities.
Another type of RS being investigated for use during brain surgery is surface-enhanced Raman scattering (SERS) [
46‐
50]. Much of this research is still being completed in animal models due to the requirement of nanoparticles to enhance the surface for RS. Of note, Kircher et al. used a trimodality of MRI, photoacoustic imaging, and SERS in mice to get whole-brain tumor location prior to surgery and during surgery [
47]. They measured the Raman signal with a customized Raman microscope (inVia, Renishaw) using an excitation wavelength of 785 nm. Magnetic resonance imaging–photoacoustic imaging–Raman imaging nanoparticles (MPRs) were injected intravenously into glioblastoma-bearing mice. The MPR is a gold-silica–based SERS nanoparticle coated with Gd
3+ ions. The MPRs accumulated and were retained by the tumors, with no MPR accumulation in the surrounding healthy tissue. The MPRs were detected by all three modalities with at least picomolar sensitivity both
in vitro and in living mice. Prior to surgery, nanoparticles were visible through the skin and skull of mice to a depth of about 2–5 mm [
47]. SERS was used during tumor resection [
47]. Residual blood-borne Gd
3+ was removed by renal function.
Additionally, Karabeber et al. used a handheld Raman probe to detect gold-silica SERS nanoparticles in glioblastoma tumors grown in mice [
50]. The particles were intravenously injected into the mice and allowed to circulate for 24 h to ensure that they accumulated in the tumors. Mouse brains were then harvested and fixed in 4% paraformaldehyde. Tumors were then resected with and without Raman guidance. Image guidance with a MiniRam Raman handheld scanner (B&W TEK, Inc., Newark, DE) using a 785-nm excitation laser and 1–2-s long acquisition times was cross-validated with a conventional Raman microscope. The conventional static system was a customized benchtop inVia Raman microscope (Renishaw) equipped with a 785-nm laser as the excitation source with an integration time of 2 s. Both handheld and static SERS image-guided resections were more accurate than resection using white light visualization alone. Correlation with histology showed that SERS nanoparticles accurately outlined the extent of the tumor. Although the Raman scanner cannot acquire the entire SERS images, as with the static system (which takes minutes to hours to map a sample), it has important advantages in that the form factor is conducive for operating room use, it provides near real-time scanning, and it can probe areas of the operative bed due to variable tile angles. The authors demonstrated the handheld probe was able to detect microscopic foci of cancer in the resection bed that were not seen on static SERS images [
50]. Although SERS is not as mature as standard RS, it still has considerable potential to be used to detect tumor margins.
Surface-enhanced resonant Raman spectroscopy (SERRS) is another variety of Raman being used to image brain tumors [
51‐
53]. Much like SERS, the research is currently being conducted in animal models, as it requires the use of nanoparticles. Of note, Huang et al. found that the SERRS signal was orders of magnitude higher than nonresonant SERS and is capable of imaging just a few cells [
52]. In this study, GBM-bearing mice were intravenously injected with integrin-targeted RGD SERRS nanoparticles. Raman imaging of paraffin-embedded coronal brain sections was accomplished with an inVia Raman microscope (Renishaw) using an excitation wavelength of 785 nm. Integrin targeting was shown to be highly specific to tumor but not normal tissue and enabled visualization of the extent of tumor and the diffuse margin of the main tumor. This also included areas distinct from the main tumor, tracks of migrating cells of two to three cells in diameter and isolated distant tumor cell clusters of less than five cells [
52].
Coherent anti-Stokes Raman spectroscopy (CARS) is an alternative type of Raman being investigated to make images of brain tumors. Most studies are being conducted in murine models, but these have recently been extended to human tissue [
54‐
60]. Galli et al. conducted CARS on excised human tissue samples after 5-aminolaevulinic acid (5-ALA) was preoperatively administered. The investigators found that 5-ALA did not interfere with CARS [
57]. The fluorescence of 5-ALA-induced protoporphyrin IX was used to identify tumorous tissue. Using it as a reference, CARS images were generated with the signal at a wavenumber of 2850 cm
−1, which is used to address the distribution of lipids inside tissue. By combining CARS with two-photon excited fluorescence (TPEF) and second harmonic generation (SHG), detailed images of tissue with structures such as extracellular matrix, blood vessels, and cell bodies were produced. The cell morphology in the CARS images was useful for tumor recognition, and the chemical contrast provided by CARS allowed localization of infiltrating tumor cells in fresh tissue samples [
57]. Romeike et al. also combined CARS at wavenumber 2850 cm
−1 with TPEF to produce detailed images of human brain biopsy specimens that had been cryogenically frozen [
58]. The images demonstrate cytological and architectural features that may allow tumor typing and grading [
58]. They noted that for CARS to advance, it requires miniaturization.
Finally, stimulated Raman spectroscopy (SRS) is a further category of Raman being researched to identify brain tumors [
61‐
65]. Ji et al. used biopsies from adult and pediatric patients to detect tumor infiltration with 97.5% sensitivity and 98.5% specificity with a generalized additive model (GAM) for the classifier [
62]. In this method, a Stokes beam (1064 nm) was combined with a tunable pump beam (650–1000 nm) from an optical parametric oscillator that was focused on the sample
via a laser scanning microscope. The energy difference between the pump and Stokes beams was tuned to specific molecular vibrations, which cause an intensity loss in the pump beam, that are detectable with the aid of a lock-in amplifier. Raman frequencies of 2845 (lipids) cm
−1 and 2930 (protein) cm
−1 were chosen for two-color (green, blue) SRS imaging for each 300 × 300 μm
2 field of view (FOV). Using quantitative measurements of tissue cellularity, axonal density, and protein/lipid ratio in SRS images, they derived a classifier capable of detecting tumor infiltration [
62]. Hollon et al. also used fresh tissue from pediatric patients with classification algorithm accuracy of 93.8% on cross-validated data on normal
versus lesional tissue and 89.4% accuracy on cross-validated data for low-grade
versus high-grade tumors [
63]. SRS images were generated with a clinical fiber-laser–based SRS microscope. Raman frequencies of 2845 (lipids) cm
−1 and 2930 (protein) cm
−1 were chosen for two-color (green, blue) 400 × 400-μm
2 SRS images. These images allow neuropathologists to diagnose the tissue with 92–96% accuracy. The image features were then used to develop a random forest machine learning model for automated classification [
63]. Lu et al. profiled 41 specimens resected from 12 patients with a range of brain tumors. SRS Raman imaging data were correlated with the current clinical gold standard of histopathology and were shown to capture many essential diagnostic hallmarks for glioma classification. Interestingly, in fresh tumor samples, Lu et al. detected structures that were not evident in the H&E stains, such as abundant intracellular lipid droplets within the glioma cells, collagen deposition in gliosarcoma, and irregularity in the disruption of myelinated fibers in areas infiltrated by oligodendroglioma cells [
64].
Lastly, progress is being made in making SRS more portable and practical for the surgical suite. Orringer et al. demonstrated SRS microscopy in the operating room using a portable fiber-laser–based microscope and unprocessed specimens from 101 neurosurgical patients [
65]. Histologic images of fresh, unstained surgical specimens were created with the clinical SRS microscope. The all fiber-based system had a 790-nm pump beam and a tunable Stokes beam over the entire tuning range from 1010 to 1040 nm. While for clinical implementation an all fiber system is desired, the relative intensity of noise intrinsic to fiber lasers can vastly degrade SRS image quality. To address this, the authors developed a noise cancelation scheme to improve the signal-to-noise ratio by 25-fold. Images were created by mapping two biologically significant Raman shifts: 2845 cm
−1, which corresponds to CH
2 bonds in lipids, and 2930 cm
−1, which corresponds to CH
3 bonds in proteins and DNA. To produce simulated Raman histology (SRH) images, FOVs are acquired at a speed of 2 s per frame in a mosaic pattern, stitched, and recolored. A subtracted CH
3–CH
2 image was assigned to a blue channel and a CH
2 image was assigned to the green channel. Using SRH images generated by this system, pathologists diagnosed lesional from nonlesional areas with 98% accuracy and glial from nonglial tumors with 100% accuracy [
65]. The authors employed a machine learning process called a multilayer perceptron (MLP) for diagnostic prediction. The diagnostic capacity for classifying individual FOVs as lesional or nonlesional was 94.1% specificity and 94.5% sensitivity, and glial from nonglial specimens were differentiated with 90% accuracy [
65]. With this advance, SRS is now a promising technology for identifying tumor margins in brain cancer. Neuronavigation techniques and brain tumor assessment can benefit from the addition of Raman spectroscopy systems during surgery.
5.4 Pancreatic cancer
Pancreatic cancer is the third leading cause of cancer-related death in the USA [
105]. It is estimated that in 2018 about 53,600 people will be diagnosed with pancreatic cancer with 44,300 estimated deaths. The 5-year survival rate of people treated for exocrine pancreatic cancer at stage IA is 12%, stage IIA 5%, stage III 3%, and stage IV 1% [
106]. The survival rate of patients with neuroendocrine pancreatic tumors that were treated with surgery at stage I is 61%, stage II 52%, stage III 41%, and stage IV 16%. About 94% of pancreatic cancers are classified as exocrine tumors with the vast majority being adenocarcinomas [
107].
Scientists at the Leibniz Institute of Photonic Technology and Friedrich Schiller University Jena collected Raman spectra from T lymphocyte Jurkat cells and pancreatic cell lines Capan1 and MiaPaca2 [
108]. Their Raman microscopy setup uses a 785-nm single-mode excitation laser and a sample holder mounted to a motorized x–y translational stage with a manual Z-positioning stage. An oil immersion objective lens focuses the excitation laser beam into the sample plane to a spot size of approximately 0.8 μm with a focal length near 1.6 μm. The spectrometer resolution is 9 cm
−1 from 300 to 4000 cm
−1 range. The Raman signal is received by a CCD with 400 × 1340 pixels. Using a support vector machine method with linear kernel, coupled with PCA, the cell classification precision for pancreatic cell lines is higher than 90%. They found that pancreatic cells have higher lipid content, which is evident from stronger lipid-related bands in the high wavenumber region at 2854 cm
−1, and higher band ratios 1440/1660 and 1320/1340 cm
−1. Also, the acquisition of integrated Raman signals of large portions of cells allowed for sampling of single cells and simpler interpretation of the cell type differences that are comparable to the acquisition of single spectra. The integrated Raman spectra approach provided better and more stable predictions for individual cells and may have a major impact on the implementation of Raman-based cell classification.
Researchers from Purdue University and Indiana University School of Medicine found a link between cholesterol esterification and metastasis in pancreatic cancer. They used SRS microscopy and Raman spectroscopy to map lipid droplets (LDs) stored inside single cells. Analyses of the composition of individual LDs revealed an aberrant accumulation of cholesteryl ester (CE) in human pancreatic cancer specimens and cell lines [
109]. Their SRS imaging was conducted using a femtosecond laser source. The pump and Stokes beams are collinearly overlapped and combined with the pump beam that is tunable from 680 to 1080 nm and the Stokes beam that is tunable from 1.0 to 1.6 μm. Images were taken on a laser scanning microscope with a ×60 water immersion objective. The signals were detected by a photodiode and then sent to a fast lock-in amplifier, which has a time constant as small as 800 ns. The lateral and axial resolutions of their SRS microscope are about 0.42 and 1.01 μm, respectively. For coherent Raman scattering imaging, two synchronized 5-ps, 80-MHz laser oscillators are temporally synchronized and collinearly combined into a laser scanning inverted microscope. The CARS signals are detected by photomultiplier tube detectors. Confocal Raman microspectroscopy is realized by mounting a spectrometer to the side port of the microscope. The pump and Stokes lasers are tuned to 707 and 885 nm, respectively, to be in resonance with the CH2 symmetric stretch vibration. The spectrometer is equipped with a 300-grooves/mm 500-nm blaze angle grating and a thermoelectrically (TE) cooled back-illuminated electron-multiplying charge-coupled device. LD amount was quantified based on the SRS images using the software ImageJ. CE level in individual LDs was quantified by analyzing the height ratio of the 702-cm
−1 peak to 1442-cm
−1 peak. They found that the peak of cholesterol at 702 cm
−1 and the peak of ester bond at 1742 cm
−1 are high for cancer tissues. They also found that abrogation of cholesterol esterification, either by an ACAT-1 inhibitor or by shRNA knockdown, significantly suppressed tumor growth and metastasis in an orthotopic mouse model of pancreatic cancer. These results demonstrate a new strategy for treating metastatic pancreatic cancer by inhibiting cholesterol esterification.
About 10 years prior, researchers at Wayne State University [
110] collected Raman spectra of normal and pancreatic tissue from mouse model using a Renishaw Raman microscope equipped with a thermoelectric cooled 578 × 385-pixel CCD. A 785-nm wavelength laser line (approximately measured at 130 × 25 um) was used to excite the tissue sample with 50 mW of power. The excitation laser line covers a section of tissue encompassing multiple cells and reflects the averaged characteristic over that section. The spectral range is from 600 to 1800 cm
−1, with the resolution of 4 cm
−1. The Raman data were analyzed by PCA and discriminant function analysis (DFA). They found that Raman spectroscopy differentiated normal pancreatic tissue from tumors in a mouse model with high sensitivity (91%) and specificity (88%), and pancreatic tumors were characterized by increased collagen content and decreased DNA, RNA, and lipid components compared to normal pancreatic tissue.
Using SERRS nanoparticles, scientists at the Memorial Sloan Kettering Cancer Center demonstrated an imaging method for the precise visualization of tumor margins, microscopic tumor invasion, and multifocal locoregional tumor spread [
111]. They designed, synthesized, and tested a new SERRS nanoprobe that is resonant in the near-infrared (NIR) window, where optical penetration in tissue is maximized. Their nanoparticles feature a star-shaped gold core, a Raman reporter resonant in the near-infrared spectrum, and a primer-free silication method. Raman scans were performed on an inVia Raman microscope (Renishaw) equipped with 785 nm diode laser and a 1-in. charge-coupled device detector with a spectral resolution of 1.07 cm
−1. The Raman maps were generated and analyzed by applying a DCLS algorithm (WiRE 3.4 software, Renishaw). Counts per second represent the intensity of the 950-cm
−1 peak of SERRS nanoparticles. Statistical analysis was performed in Excel (Microsoft). In genetically engineered mouse models of pancreatic cancer, breast cancer, prostate cancer, and sarcoma, and in one human sarcoma xenograft model, this method enabled accurate detection of macroscopic malignant lesions, as well as microscopic disease, without the need for a targeting moiety, and the sensitivity (1.5 fM limit of detection) of this method allowed imaging of premalignant lesions of pancreatic and prostatic neoplasias.
5.4.1 Raman spectroscopy of pancreatic cancer serum markers
Early-stage pancreatic cancer is difficult to detect due to the lack of symptoms, which often results in diagnosis at an advanced stage of disease. CA19-9 and carcinoembryonic antigen (CEA) are tumor markers that may be detected in the blood and are tied to pancreatic cancer. These proteins may or may not be elevated in a person with pancreatic cancer. About 59% of patients with pancreatic carcinoma have high concentrations of CEA that suggest a mucinous pancreatic cyst. However, CEA testing does not reliably distinguish between begin, premalignant, or malignant mucinous cysts. Serum CA19-9 is a tumor-associated mucin glycoprotein antigen related to the Lewis blood group protein. About 5% of the population do not produce CA19-9 antigen. The sensitivity (68–93%) and specificity (76–100%) of CA19-9 is not adequate for diagnosis and precludes it as a screening tool [
112].
Using SERS, researchers from Iowa State University, University of Nebraska Medical Center, University of Pittsburgh Medical Center, and University of Utah demonstrate the first ever detection of the potential pancreatic cancer marker MUC4 in cancer patient serum samples [
113]. Their SERS-based immunoassay chip design includes (a) a capture substrate to specifically extract and concentrate antigens from solution, (b) surface-functionalized gold nanoparticles (extrinsic Raman labels or ERLs) to bind to captured antigens selectively and generate intense SERS signals, and (c) sandwich immunoassay with SERS readout. The Raman spectra were collected with a NanoRaman I fiber-optic–based Raman system, a portable, field-deployable instrument. The light source was 632.8 nm He–Ne laser. The spectrograph consisted of an imaging spectrometer (6–8 cm
−1 resolution) and a CCD imaging array. The incident laser light was focused to a 25-μm spot on the substrate. The analyte concentration was quantified using the peak intensity of the symmetric nitro stretch at 1336 cm
−1. The amount of human mucin MUC4 was measured in CD18/HPAF lysate (positive control) by sandwich enzyme-linked immunosorbent assay (ELISA). SERS measurements showed that sera from patients with pancreatic cancer produced a significantly higher SERS response for MUC4 compared to sera from healthy individuals and from patients with benign diseases. And SERS measurement can also detect CA19-9 concentration.
Recently, scientists at the University of Massachusetts [
114] demonstrated a novel system for multiplex detection of pancreatic biomarkers CA19-9, MMP7, and MUC4 in serum samples with high sensitivity using surface-enhanced Raman spectroscopy. Their SERS-based immunoassay for biomarker quantification includes (I) functionalizing gold substrate with thiol and antibody, (II) capturing desired antigens from the serum, and (III) loading antibody-conjugated extrinsic Raman labels (ERL), and gold nanoparticles were modified with antibody and Raman reporter. Raman spectra collection was performed with a portable BWS415 i-Raman at an excitation wavelength of 785 nm. The antigen concentration was quantified using intensity at the 1336-cm
−1 position which corresponds to a symmetric stretch of the NO
2 group whose intensity of this band depends proportionally on the concentration of MUC4 in a sample. They found that immobilization of functionalized gold nanoshells with resonance wavelength of 660 nm on the gold-coated silicon substrate led to a significant improvement of SERS signals, and successfully detected three pancreatic biomarkers, CA19-9, MMP7, and MUC4, in spiked serum samples at concentrations as low as 2 ng per ml. Measuring the levels of these biomarkers in pancreatic cancer patients, pancreatitis patients, and healthy individuals revealed the unique expression pattern of these markers in pancreatic cancer patients, suggesting the great potential of using this approach for early diagnostics of pancreatic cancers.
5.6 Circulating tumor cell
Although the majority of cancer deaths result from cancer metastasis in a localized area (tumor), there is another important focus in current cancer research. Once a tumor reaches a stable size and growth, some cells separate and enter the bloodstream of the patient. These cells are referred to as circulating tumor cells (CTCs) [
155‐
157]. Initial research in 1869 into cancer discovered not only the existence of CTCs but a possible relationship between CTCs and metastasized cancer [
157]. Through research, CTCs have indicated information about cancer type, cancer progression, and patient response before, during, and after treatment [
156,
158,
159]. During diagnosis, CTCs can assist in locating a cancer tumor, by indicating cancer type. This is done by growing a new tumor in a xenograft and identifying the cancer [
156]. While the concentration of CTCs does not appear to reflect the actual size of an existing tumor, the presence of CTCs is a viable independent prognostic indicator for several cancers, including breast, prostate, and colon [
157,
160,
161]. Additionally, regardless of initial levels, changes in concentration of CTCs in the patient do correspond to changes in the cancer tumor throughout treatment including re-occurrence after treatment is concluded [
156].
There are several types and categories in CTC research; however, two categories have received attention beyond general CTC research. Cancer stem cells (CSCs) are a specific type of CTC with high metastatic activity, motility, and resistance to apoptosis. While CTCs can originate from benign tumors and are thus not necessarily pathogenic, CSCs are considered more likely to metastatic [
156,
162]. Another topic of research is circulating tumor microemboli (CTM). CTM are multicellular aggregates of epithelial-like tumor cells and may also contain information about their tumor of origin [
163].
There are currently three main detection research paths for CTCs: antibody capture, using cancer-derived DNA, and cytopathology. Each of these approaches has limitations that interfere with using the wealth of information CTCs may be able to provide. The most common current method in use with patients uses antibody capture based on epithelial marker epithelial cell adhesion molecule (EpCAM) on the CTCs. This method has an underlying assumption that is currently debated: CTCs have not undergone epithelial–mesenchymal transition [
156,
157]. If this assumption is false, then not only will the result significantly underestimate the population of CTCs but miss a subpopulation of CTCs completely. Furthermore, certain cancers like carcinoma show partial mesenchymal properties. These properties appear to increase a cell’s metastatic potential, suggesting a greater correlation with the aspects of the tumor more relevant to treating the patient [
156].
Another current method of detection involves isolating cancer-driven DNA in the plasma of the patient. This has the advantage of not requiring whole cells which would include CSCs and CTM. Various gene families including cytokeratins, prostate-specific antigens, and others studied through PCR showed correlation to metastasized cancer. While this method has its advantages, there are still too many issues with specificity and sensitivity of the results to use routinely [
157].
The third method is not currently in use, but a proposed method. This approach relies on cytopathology, which is already used in screenings for other cancers such as PAP smears. Although this method innately has a higher specificity, due to the low concentration of CTCs in samples, the lack of sensitivity makes this approach impractical. Enrichment methods such as density gradient separation and filtration were unsuccessful in increasing the sensitivity of the tests because the multiple steps damaged or degraded the cells resulting in a loss of sensitivity and specificity [
163].
Raman spectroscopy could provide increased specificity and sensitivity compared to the techniques described above. There have been three studies on applying Raman spectroscopy, mostly SERS, to CTC research. In 2008, SERS successfully detected CTC resulting from breast cancer using the same epithelial markers commonly used in CTC detection. Due to the specificity of SERS, the detection limit was 10 cells/ml with 99.7% confidence in buffer solution. This process had the advantage of needing very little sample preparation. Although there was no follow-up done with patients, this experiment provided proof of concept for Raman spectroscopy and CTCs [
164,
165].
The next Raman experiment involved spiked blood serum. Microscopic Raman spectroscopy identified MCF-7 (breast cancer), BT-20 (breast cancer), OCI-AML3 (acute myeloid leukemia), leukocytes, and erythrocytes in suspension to mimic a clinical test. The results showed a prediction accuracy of 92.4% with a false positive less than 0.5%. This compares to the false positive of the cytopathology above of 1–3% [
166].
Wang et al., using epidermal growth factor (EGF), detected CTCs at a concentration 50 tumor cells/ml of blood by detecting the expression of epidermal growth factor expression (EGFR) using SERS. Positive results for Tu212 SCCHN cells and H292 lung cancer cells (high EGFR expression), DA-MB-231 breast cancer cells (moderate EGFR expression), and H460 lung cancer cells (low EGFR expression) showed that even with a variation in expression, the system detected the cell’s existence. Further testing on 19 cancer patients with confirmed SCCHN showed that 17 out of 19 patients had CTCs (confirmed by filtration). Later, it was confirmed that the remaining two patients had localized instead of metastatic disease. Three control cancer-free patients showed no CTCs in the SERS. One research participant with a confirmed tumor was tested prior and following treatment. In this case, SERS appropriately indicated the presence of CTCs before treatment and their absence following treatment [
165,
167]. Although this is a small sample size, the research indicates that SERS is a viable detection method for CTCs. Furthermore, this process has the additional advantage of not depending on epithelial cell markers. In addition to its increased accuracy, this approach may be used to detect CSCs or other CTCs that have undergone epithelial–mesenchymal transition.