Background
Breast cancer is the most common malignancy in women in the United States and the second leading cause of death by cancer. It is estimated that 235,030 new cases of breast cancer will be diagnosed in the United States in 2014 [
1]. Early diagnosis is a significant prognostic factor. The American Cancer Society is recommending annual screening mammograms starting at age 40 [
2]. Conventional mammography is known to have a sensitivity of about 66 % and specificity of about 92 % [
3]. However, recent studies show that screening with mammography does not reduce mortality, it may lead to a 30 % rate of overdiagnosis and may increase unnecessary surgical procedures and patient anxiety [
4,
5]. Furthermore, women with dense breasts, in whom mammography is of limited value and high-risk patients with suspicious mammography findings, usually require additional evaluation with ultrasound or magnetic resonance imaging [
6]. This may contribute to the diagnosis in some cases but it may increase recall examinations due to false-positive results in others [
7,
8]. Alternative methods such as thermography, transillumination, and positron emission tomography, have not been proven yet to have better sensitivity or specificity than mammography [
9].
In the last few decades, researchers have introduced the use of serum tumor markers for cancer screening. However, none of the markers tested has proved suitable for screening the entire population because of low specificity and sensitivity at the early stages of disease [
10‐
12]. To improve these results, attempts have been made to apply combinations of markers [
13,
14]. Thus, multi-molecular biochemical analysis could be useful for this purpose.
Fourier transform infrared (FTIR) spectroscopy is a simple, rapid, reagents-free biochemical tool that provides information on the total molecular composition of biological samples [
15]. Organic compounds absorb infrared light at an energy (wavenumber) corresponding to the nature of the bonds between its atoms, yielding a unique spectral “fingerprint”. Thus, spectroscopy of a biological sample generates an absorption spectrum of the compounds in that sample, reflecting their molecular structure. FTIR spectroscopy is a powerful analytical biochemical and imaging method however, in a complex samples such as blood components, it is complicated to locate a change in a specific molecule due to the overlapping bands and the plenty of vast molecules which compose biological samples. Yet, FTIR can be widely used for differentiating between two different samples and locate the bands and the possible molecules which may contribute to the spectral differences.
FTIR spectroscopy has been found to be useful for the detection and characterization of a broad variety of cancer cells and tissues [
15‐
17]. A previous study by our group in patients with leukemia identified markers of the disease by FTIR spectroscopy of peripheral blood mononuclear cells (PBMCs) which were then used to monitor the disease during chemotherapy [
18]. The method was effective even in cases in which blasts were hardly present in the peripheral blood [
18], indicating the overall biological influence of malignancy on PBMCs. In another study, our group demonstrated the potential of FTIR analysis of plasma for the detection of solid tumors, mostly breast, colorectal, and lung. Using advanced algorithms, we identified the patients with cancer out of the whole study population with 93.33 % sensitivity and 90.7 % specificity [
19].
Prompted by these findings of the systemic effect of malignancy on the FTIR spectra of PBMCs and plasma, in the present study, we sought to investigate the utility of FTIR spectroscopy for breast cancer screening in conjunction with the gold standard diagnostic methods such as mammography and ultrasound.
Methods
Patients
The study was conducted at Rabin Medical Center under local Ethics Committee approval at 2011 and 2012. The study group included 29 patients with confirmed breast cancer and 30 control patients without breast cancer as determined by biopsy and standard mammography examination. The control group included 15 patients without pathological findings and 15 patients with benign neoplasms. The patients were randomly selected from population performing routine breast cancer screening and from population prior surgery. Qualified personnel obtained informed consent from each participant. Exclusion criteria were pregnancy, lactation, or presently undergoing fertility treatment, known active inflammation or infection, past treatment for malignant of benign tumor, any type of active autoimmune disease, and current intake of medications such as steroids. Cancer diagnoses were confirmed by clinical, histological, and pathologic means. Cancers were graded according to the National Cancer Institute classification.
Blood sample collection and preparation
By preparing PBMCs and plasma samples for FTIR measurements we considered all the possible contaminations and interferences from biochemical materials involved in the sample preparation due to the nature of FTIR as highly sensitive biochemical analytical tool. Thus the samples are needed to be clean from reagents. For each participant, 2 ml of blood were collected from a peripheral vein into EDTA tubes (BD Vacutainer® Tubes, BD Vacutainer, Toronto) using standard phlebotomy procedures. Samples were processed within 2 hours of collection. Some of the patients with cancer underwent lymphoscintigraphy with Tc-99 m-labeled nanocolloidal albumin to detect the sentinel node a few minutes before blood collection, but the possibility of an effect of lymphoscintigraphy on the spectra of the blood components was ruled out using FTIR spectroscopy of pure Tc-99 and plasma spectral comparison. The blood was diluted 1:1 in isotonic saline (0.9 % NaCl solution), applied carefully to a Ficoll 1077 gradient (Sigma Chemical Co., St. Louis, MO) in 15 ml collection tubes, and centrifuged at 400 g for 30 min. To discard platelets and cell debris, we placed 1 ml of the plasma in 1.5 ml tubes which were centrifuged at 6000 g for 10 min. The supernatant was transferred to a new 1.5 ml tube, and 0.8 μl of plasma was deposited on a zinc selenide (ZnSe) slide and air-dried for 1 hour under laminar flow. The dried plasma was then subjected to FTIR microspectroscopy.
PBMCs were obtained using a Histopaque 1077 gradient (Sigma, St. Louis, MO) according to the manufacturer’s protocol. The cells were aspirated from the interface, rinsed twice with isotonic saline at 250 g, and re-suspended in 5 μl fresh isotonic saline. Thereafter, 0.4 μl of washed cells were deposited on ZnSe slides to create an approximate uniform layer of cells. The cells were air-dried for 1 hour under laminar flow and analyzed by FTIR microspectroscopy. The samples need to be dried since water molecules strongly absorb infrared light which may mask the signal from the sample.
FTIR microspectroscopy
All spectroscopy studies were performed with the Nicolet Centaurus FTIR microscope equipped with a liquid-nitrogen-cooled mercury-cadmium-telluride detector coupled to Nicolet iS10 OMNIC software (Nicolet, Madison, WI). To achieve a high signal-to-noise ratio (SNR), 128 co-added scans were collected in each measurement in the 700 to 4000 cm−1 wavenumber region. At a spectral resolution of 4 cm−1 (0.482 cm−1 data spacing), each spectrum contains 6845 data points. The dimensions of the measurement site were 100 μm X 100 μm. Measurements were performed in transmission mode at least 5 times at different spots in each sample of PBMCs or plasma.
Spectral preprocessing
The FTIR spectra for PBMCs and plasma were first examined for unsuccessful measurements, such as absorption intensity above or below normal (defined as 0.5 to 1 absorption units according to Amide I band) and water vapor contamination. Next, we focused on the relevant region of 1800–700 cm
−1 which contains most of the biochemical data of PBMCs and plasma. Following standard vector normalization to obtain a unity total energy of each spectrum [
19,
20], we applied a moving average filter to increase the SNR. Finally, we sought a numerical estimation for the second derivative of the spectra to accentuate the bands, reduce the background interference, and reveal the genuine biochemical characteristics [
21]. Although the second-derivative method is known to be highly susceptible to full width at half maximum changes in the infrared bands, these changes are not relevant in biological samples in which all cells of the same type and plasma are composed of similar basic components that yield relatively broad bands [
22]. Spectrum parameters were calculated by our in-house algorithms; the code was employed using MATLAB (Version R2011B: MathWorks Inc., Natick, MA).
Feature selection
The spectra obtained contained 2282 data points or dimensions. For successful and less complex classification, the number of dimensions needed to be reduced. Our goal was to identify a subset of specific wavenumbers or intervals in the spectra that represented the different spectral patterns of the groups. To improve the model, we defined two criteria for potential feature evaluation. First, we performed a Student’s t-test analysis between the no cancer class (benign or no breast tumor) and the cancer class. A feature was considered significant at P <0.005. Next, for each potential feature, we obtained the probability distribution of each class and measured the similarity of the probability density functions. In this manner, we were able to evaluate the amount of overlap between the two populations.
Statistical analysis
Following feature selection, quadratic discriminant analysis (QDA), a multivariate data analysis method, was performed to classify the different groups under the assumption that each feature is normally distributed. The QDA classifier produces a new discriminative score for each subject that can be classified according to the cut-off point. The best cut-off point was determined by creating a receiver operating characteristics (ROC) curve and choosing the one with the best performance [
23]. Monte-Carlo cross-validation was used to determine the accuracy of classifier predictions for different cut-offs [
23].
Discussion
The present study describes a novel concept for breast cancer detection based on the immune system response to the presence of tumor in the body rather than on observation of the tumor cells themselves. Furthermore, by using infrared spectroscopy, we were able to analyze the entire biochemical signature (including proteins, lipids, nucleic acids, and carbohydrates) of the affected immune cells rather than focusing on a single specific protein as a biomarker. We also analyzed the malignancy-induced biochemical changes in plasma to obtain more information about the disease and as an auxiliary means of cancer detection.
The results provide evidence that the PBMCs and plasma of patients with breast cancer are biochemically distinct from the PBMCs and plasma of healthy subjects, including patients with benign tumors, with no significant differences in PBMC spectra between patients with benign tumors and healthy subjects. For plasma, there was a biochemical similarity between patients with benign tumors and healthy subjects for some spectral absorption bands, and between patients with benign tumors and patients with malignant tumors for other absorption bands. Further analysis of the data within the group of cancer patients revealed a correlation of the spectral changes of PBMCs and plasma with clinically relevant parameters known to influence the diagnosis and prognosis of breast cancer, such as disease stage and vascular invasion.
Previous studies of cancer cells and tissues using FTIR spectroscopy reported an abnormal biochemical profile, expressed by various changes in the phosphate region which corresponds mainly to nucleic acids and carbohydrates [
28,
29]. Others also noted a significant increase in the ratio of CH
2/CH
3 in the higher region of lipids and protein absorption [
29,
30]. These changes were consistent for most of the tumors and depended on the stage of disease [
28,
30]. They were compatible with our findings in an earlier study of PBMC biochemistry in patients with acute leukemia [
18]. However, in the present study, which included patients with solid tumors, there was no significant change in CH
2/CH
3. The major changes observed between the groups were found in proteins structure and in several functional groups of nucleic acids, carbohydrates and phospholipids, suggesting that PBMCs from patients with solid tumors have a different profile than PBMCs from patients with hematological malignancies. Thus, our results indicate cancer-type-dependent changes in the PBMC population.
The differences in PBMC biochemistry between patients with and without cancer may be related to malignancy-induced biological effects, such as changes in the composition of the mononuclear population; specifically, the relationships between B and T cells [
31,
32]. The presence of CD4 + CD25- T cells in the peripheral blood as well as in the tumor site leads to a significant increase in the number of regulatory T cells (Treg cells, CD4 + CD25+) [
33,
34]. These findings have been reported not only in breast cancer [
31,
35,
36], but also in gastrointestinal [
37], and lung cancer [
38]. Treg cells regulate effector T cells and disable them in order to prevent them from attacking the tumor [
33]. The level of Treg cells is apparently correlated with disease stage and declines with tumor dissection [
11,
37,
38]. Studies have also provided evidence of the role of natural killer cells as a prognostic parameter and therapy target [
39‐
41]. These studies support our finding of the contribution of clinical parameters (tumor size, blood vessel and lymph node involvement) to the biochemical changes in PBMCs and highlight the potential of FTIR-spectroscopy as a prognostic and treatment follow-up tool. Although the changes in the PBMC population may be correlated with stage of disease, in the present study, there were no cases of advanced-stage breast cancer, so further studies in animal and human models are needed to address this issue.
Many studies have investigated the difference between healthy and malignant tumors, but only a few addressed the biochemistry of benign tumors (45, 46). They found no or only slightly significant differences from malignant tumors [
42,
43]. On the contrary, in the present study, only small differences were observed in the PBMC spectra between patients with benign tumors and healthy subjects. However, more extensive studies are needed to verify these preliminary results.
Our previous study showed that FTIR spectroscopy of plasma is a promising mean for distinguishing patients with cancer from healthy subjects however the benign tumors were not investigated by Ostrovsky et al. [
19]. Most of the common serum biomarkers cannot be used for distinguishing between benign and malignant tumors [
42,
44], perhaps because of the immunological similarity of the tissues. Indeed, in the present study, we identified several vibrational bands in the plasma spectra that were common to both benign and malignant tumors which correspond to carbohydrates and proteins. We further identified bands which are common to healthy and benign groups in the Amide I band which correspond mainly to protein secondary structure. Thus, the significant contribution to cancer detection may be related to the structure of proteins in the plasma rather than carbohydrates. For our purposes, we can relate only to the bands that are common to benign and healthy tissues and improve the detection of malignant tumors.
The algorithm presented here makes use of the global biochemical information obtained both for PBMCs and plasma. The sensitivity was about 90 % and the specificity was about 80 %. These values are promising considering that we were able to distinguish between nonmalignant and malignant tumors and most of the patients with malignancy were at early stages of the disease. We aim to further improve our algorithm with a larger sample size.
Competing interests
U Zelig, O Bar, C Segev and F Flomen are employees of Todos Medical LTD, Israel. Patents Number US20140166884 A1 and US20130143258 A1 by U Zelig, S Mordechai, J Kapilushnik and I Nathan on the results reported in this study, including individual spectral bands of plasma and PBMCs as markers for cancer, have been filed.
Authors’ contributions
UZ Performed blood separation and FTIR measurements, contributed to development of methodology, contributed to study design, contributed to data analysis and interpretation, drafted the manuscript. EB Requited patients, collected blood samples, contributed to acquisition of data on patient outcomes, contributed to manuscript preparation. OB Performed the FTIR spectral and statistical analysis, contributed to manuscript preparation and revision. IG Contributed to acquisition of data on patient outcomes, and spectra analysis. FF Contributed to development of methodology, contributed to spectral and statistical analysis SM, JK and IN Contributed to development of methodology, and manuscript preparation and revision. KH Coordinated the study design and contributed to manuscript revision. NW and OMG Contributed to study design, data interpretation and manuscript preparation and revision. All authors read and approved the final manuscript.