Who’s Who? Discrimination of Human Breast Cancer Cell Lines by Raman and FTIR Microspectroscopy

Simple Summary Normal-to-malignant transition in human cells is still a poorly understood process, as well as the mechanisms leading to cancer invasion and metastasis. An in-depth characterisation and understanding of the genetic and chemical variations underlying carcinogenesis and cancer progression are paramount for improving diagnosis and design better therapeutic strategies, aiming at a better prognosis for oncology patients. The present study reports the application of spectroscopic methods (micro-Raman and infrared) for identifying specific biomarkers of malignancy in several types of human breast cancer cells. The results thus obtained led to an accurate discrimination between the highly metastatic mammary tumours (triple-negative) and the less aggressive ones (non-triple negative), as well as among triple-negative subtypes. Abstract (1) Breast cancer is presently the leading cause of death in women worldwide. This study aims at identifying molecular biomarkers of cancer in human breast cancer cells, in order to differentiate highly aggressive triple-negative from non-triple-negative cancers, as well as distinct triple-negative subtypes, which is currently an unmet clinical need paramount for an improved patient care. (2) Raman and FTIR (Fourier transform infrared) microspectroscopy state-of-the-art techniques were applied, as highly sensitive, specific and non-invasive methods for probing heterogeneous biological samples such as human cells. (3) Particular biochemical features of malignancy were unveiled based on the cells’ vibrational signature, upon principal component analysis of the data. This enabled discrimination between TNBC (triple-negative breast cancer) and non-TNBC, TNBC MSL (mesenchymal stem cell-like) and TNBC BL1 (basal-like 1) and TNBC BL1 highly metastatic and low-metastatic cell lines. This specific differentiation between distinct TNBC subtypes—mesenchymal from basal-like, and basal-like 1 with high-metastatic potential from basal-like 1 with low-metastatic potential—is a pioneer result, of potential high impact in cancer diagnosis and treatment.


Introduction
Breast cancer, a highly heterogeneous disease originating from the epithelial cells lining the milk ducts, is the second most common type of cancer and the most common in women (affecting one in every four). Despite all efforts regarding screening and treatment, this type of neoplasia, with a growing incidence in the last four decades, remains a leading cause of death among women worldwide (mainly due to bone, lung, brain and liver metastasis) [1,2]. Based on their overall morphology and structural organisation, breast tumours were average biochemical information and can be reliably applied to both in vitro and in vivo conditions. Hence, there is an increasing number of reported studies on the application of infrared and Raman spectroscopies to monitor cells, both live and fixed (e.g., healthy from cancer discrimination and metabolic response to drugs) and tissues (e.g., cancer diagnosis and treatment monitoring) [4,[26][27][28][29][30][31][32][33][34][35][36][37][38][39][40][41][42][43][44]. Thanks to their distinctive characteristics and ease of automation, these techniques constitute an improved approach for cancer detection when compared to the currently used diagnostic methods such as the gold standard histopathological assessment (which lacks sensitivity and specificity at an early stage of the disease). Therefore, their application in the clinical arena may greatly accelerate point-of-care decisions and improve patient outcomes.
The aim of the present study was to take a step forward and apply complementary Raman and FTIR microspectroscopy (upon multivariate analysis) to unveil specific and sensitive molecular biomarkers of cancer in human neoplastic breast cells, assessing whether these techniques are able to detect even more subtle differences than those between malignant and non-malignant cells. Accordingly, an unambiguous discrimination was sought between: (i) triple-negative and non-triple-negative; (ii) different TNBC subtypes, namely mesenchymal vs. basal-like and (iii) TNBC basal-like 1 with high-metastatic potential vs. TNBC basal-like 1 with low-metastatic potential. The results thus obtained should provide particular features of malignancy and identify specific biochemical reporters, which will contribute to differentiate different types of human breast carcinoma. This constitutes a paramount advance in cancer diagnosis, particularly regarding a type of malignancy with high morbidity and mortality.

Materials and Methods
The list of chemicals as well as the details of the cell culture procedure, Raman and FTIR microspectroscopy measurements, spectral pre-processing and multivariate analysis are described in the Supplementary Materials.

Sample Preparation for Spectroscopic Analysis
Upon harvesting by trypsinisation, the cells were centrifuged and the pellet was resuspended in culture medium and seeded at a concentration of 3 × 10 4 cells cm −2 onto optical substrates suitable for either Raman [45] or FTIR data collection [46]: respectively MgF 2 (2 × 20 mm) or CaF 2 (UV-grade, 1 × 13 mm) disks, previously cleaned with 70% ethanol and sterilised.
After a 24 h incubation period (allowing the cells to attach), the growth medium was removed and the cells were washed twice with phosphate buffered saline (PBS), upon which they were fixed in 4% formalin (diluted in NaCl-0.9% from the commercial 10% neutral-buffered formalin solution) for 10 min [46]. After repeated washing with pure water (to remove any residual salt), the disks were allowed to air-dry (at room temperature) prior to spectroscopic analysis.
All samples were prepared in triplicate, in a single experiment.

Raman Microspectroscopy
The Raman spectra were recorded in a WITec confocal Raman microscope system alpha 300R (WITec GmbH, Ulm, Germany) with an automated sample stage, coupled to an ultrahigh-throughput spectrometer UHTS 300Vis-NIR, using a frequency-doubled Nd:YAG exciting laser line, at 532 nm.

FTIR Microspectroscopy
The microFTIR spectra were acquired in the mid-IR interval (400-4000 cm −1 ), in transmission mode, using a Bruker Optics Vertex 70 FTIR spectrometer coupled to a Bruker Hyperion 2000 microscope, both purged by CO 2 -free dry air.
Both FTIR and Raman data were captured as single point spectra, each spectrum recorded on a different single cell, across a cell population comprising randomly distributed cells displaying a high heterogeneity. Table 1 comprises the Raman and FTIR experimental wavenumbers obtained for the cancer cells under study, reflecting their biochemical profile (mainly signals from DNA, proteins, lipids and polysaccharides) and high heterogeneity. The FTIR and Raman fingerprint spectra of each cell line were pre-processed and normalized ( Figure S1).

All Breast Cancer Cells-Overall Characterisation
Mean spectral biochemical signatures were generated for each type of cell line, by averaging the corresponding Raman or FTIR spectra acquired from different cellular locations (Figure 1a-d, respectively).
The cells were chemically fixed with a 4% formaldehyde solution (formalin). Formalin causes crosslinking between the aldehyde and the primary and secondary amine groups of cellular proteins, thus retaining the cellular constituents in a close to in vivo relationship. It is expected that formalin fixation procedure induces some reduction of vibrational band intensities, related to the disruption of lipid assembly and conformational protein changes. Nevertheless, the fixation has shown to have a weak impact on the overall molecular content and yielding the spectral profile closest to that observed in the live state. This being a comparative study, all cell lines were virtually equally affected by the fixation process, which does not impact on the discrimination attained between the distinct types of cancer cell lines under the same conditions.

Raman Microspectroscopy
The Raman spectra were recorded in a WITec confocal Raman microscope system alpha 300R (WITec GmbH, Ulm, Germany) with an automated sample stage, coupled to an ultrahigh-throughput spectrometer UHTS 300Vis-NIR, using a frequency-doubled Nd:YAG exciting laser line, at 532 nm.

FTIR Microspectroscopy
The microFTIR spectra were acquired in the mid-IR interval (400-4000 cm −1 ), in transmission mode, using a Bruker Optics Vertex 70 FTIR spectrometer coupled to a Bruker Hyperion 2000 microscope, both purged by CO2-free dry air.
Both FTIR and Raman data were captured as single point spectra, each spectrum recorded on a different single cell, across a cell population comprising randomly distributed cells displaying a high heterogeneity. Table 1 comprises the Raman and FTIR experimental wavenumbers obtained for the cancer cells under study, reflecting their biochemical profile (mainly signals from DNA, proteins, lipids and polysaccharides) and high heterogeneity. The FTIR and Raman fingerprint spectra of each cell line were pre-processed and normalized ( Figure S1).

All Breast Cancer Cells-Overall Characterisation
Mean spectral biochemical signatures were generated for each type of cell line, by averaging the corresponding Raman or FTIR spectra acquired from different cellular locations ( Figure 1 a-d, respectively).
The cells were chemically fixed with a 4% formaldehyde solution (formalin). Formalin causes crosslinking between the aldehyde and the primary and secondary amine groups of cellular proteins, thus retaining the cellular constituents in a close to in vivo relationship. It is expected that formalin fixation procedure induces some reduction of vibrational band intensities, related to the disruption of lipid assembly and conformational protein changes. Nevertheless, the fixation has shown to have a weak impact on the overall molecular content and yielding the spectral profile closest to that observed in the live state. This being a comparative study, all cell lines were virtually equally affected by the fixation process, which does not impact on the discrimination attained between the distinct types of cancer cell lines under the same conditions.  The signals exclusively detected by either infrared or Raman are shaded in grey or in orange, respectively. a A-adenine; C-cytosine; dG-deoxyguanine; dT-deoxythymine; G-guanine; Hyp-hydroxyproline; Metmethionine; Phe-phenylalanine; Pro-proline; T-thymine; Trp-tryptophan; Tyr-tyrosine; U-uracil; Valvaline. δ-deformation; It is expected that formalin fixation procedure induces some reduction of band intensities, related to the disruption of lipid assembly and al protein changes. Nevertheless, the fixation has shown to have a weak e overall molecular content and yielding the spectral profile closest to that the live state. This being a comparative study, all cell lines were virtually ted by the fixation process, which does not impact on the discrimination een the distinct types of cancer cell lines under the same conditions. n and FTIR experimental data for the human breast cancer cell lines MDA-MB-231, , HCC-1143 and MCF-7.
The spectral regions that showed statistically significant differences across cell lines comprise contributions from the major cellular components (Table 1): nucleic acids, lipids (including membrane phospholipids), proteins and carbohydrates (876, 934, 1030, 1062 cm −1 ). Significant variations were detected in the Raman profiles, namely regarding the bands assigned to the modes ν(CC) DNA/bases (ca. 780 cm −1 ), ν(CC) RNA (ca. 1302 cm −1 ), ν(C=C) proteins/lipids (ca. 1580 cm −1 ) and ν(CH 2 ) proteins/lipids (ca. 2850 cm −1 ), the main differences having been observed for the TNBC BL1 ductal carcinoma (MDA-MB-468) relative to the other cell lines under analysis. Protein (Amide I and III, respectively at 1650 and 1272 cm −1 ) and lipid ratios were found to vary among these breast cancer subtypes. Specifically, regarding the protein spectral profile, these variations are ascribed to the different structures and conformations present within the cell (mainly α-helix or β-sheet) [47,48]. In addition, proteins are regarded as important indicators in cancer biology, since protein metabolism is closely associated to cell proliferation and division, transcription (RNA synthesis), translation (protein synthesis) and migration processes [4,48]. Some authors have reported a predominance of proteins in abnormal tissue, both benign and malignant [28,48,49]. In addition, Mohamed and co-workers [35] verified good discrimination between two TNBC cell lines (MDA-MB-231 and SUM-149) mainly based on the spectral signature of proteins, nucleic acids and amino acids. Furthermore, the characteristic CH 2 /CH 3 stretching modes (namely from lipid carbon chains), detected at high wavenumbers (Table 1), are considered relevant spectral biomarkers for the identification of distinct types of cancer [48]. These modes are known to vary considerably regarding the type (e.g., unsaturation degree) and amount of cellular lipids (e.g., membrane fatty acids). Studies by Ozek and co-workers [50] highlighted the influence of high amounts of polyunsaturated fatty acids on the transport and signalling pathways, associated to cell reshaping and adhesion processes that determine metastatic ability. The current study revealed an increased amount of lipids in the two TNBC BL1 cell lines (MDA-MB-468 and HCC-1143), which is in accordance with the infrared and Raman data formerly obtained by the authors for the TNBC MSL MDA-MB-231 cell line [33] that evidenced a high lipid content, particularly regarding glycerophospholipids known to be augmented in cancer cells [51]. In addition, other studies have reported an increase in the lipid content in breast cancer cells as compared to normal breast cells [43,52]. Several authors have shown a correlation between the rise in cytoplasmic lipid droplets and the increase in cancer aggressiveness [38,52,53]-the number of lipid droplets in non-malignant breast cells being reported as two-and four times lower than in non-TNBC and TNBC cells, respectively. In addition, a recent Raman study of human breast cancer cell lines (MDA-MB-231, MDA-MB-435s and SK-BR-3) [38] also detected a noticeable lipid expression in the MDA-MB-231 TNBC cells when compared to the non-TNBC (luminal HER2 + ) SK-BR-3, which was suggested to be due to an increased content in unsaturated fatty acids (mostly within the membrane). It should be emphasised, however, that the epithelial MDA-MB-435s cell line was wrongly included in this study since it was found to have undergone phenotypic and genotypic drift thus expressing melanoma-specific genes [54][55][56][57]. The spectral regions that showed statistically significant differences across cell lines comprise contributions from the major cellular components (Table 1): nucleic acids, lipids (including membrane phospholipids), proteins and carbohydrates (876, 934, 1030, 1062 cm −1 ). Significant variations were detected in the Raman profiles, namely regarding the bands assigned to the modes ν(CC)DNA/bases (ca. 780 cm −1 ), ν(CC)RNA (ca. 1302 cm −1 ), ν(C=C)proteins/lipids (ca. 1580 cm −1 ) and ν(CH2)proteins/lipids (ca. 2850 cm −1 ), the main differences having been observed for the TNBC BL1 ductal carcinoma (MDA-MB-468) relative to the The chemical differences among cell lines were clearly unveiled upon PCA analysis of the spectroscopic results-both FTIR and Raman (Figures 2 and 3, respectively). Overall, the FTIR data allowed to differentiate the TNBC cell lines from the non-TNBC (MDA-MB-231, MDA-MB-468 and HCC-1143 vs. MCF-7) (Figure 2a-f). In addition, there was a very good  From the dendrograms obtained from the average FTIR spectra (Figure 2c,f) a clear differentiation is observed between MDA-MB-231 and the other cell lines, MCF-7 and MDA-MB-468 showing the highest similarity (in both spectral regions). It is also evident that the cluster distance scale shown in the first dendrogram is larger, evidencing that the underlying variability of the fingerprint region is more relevant for this discrimination. From the dendrograms obtained from the average FTIR spectra (Figure 2c,f) a clear differentiation is observed between MDA-MB-231 and the other cell lines, MCF-7 and MDA-MB-468 showing the highest similarity (in both spectral regions). It is also evident that the cluster distance scale shown in the first dendrogram is larger, evidencing that the underlying variability of the fingerprint region is more relevant for this discrimination.
The Raman data (Figure 3a-f) yielded a much less significant differentiation between cell lines when compared to infrared. In the fingerprint interval the major differences were found along PC-1 followed by PC-3 (explaining 45.7% and 7.9% of the total data variance, respectively), while in the high wavenumber region discrimination was mainly along PC-2 (4.5% of the total data variance). Separation was evident only for the TNBC BL1 cells The dendrograms obtained from the average Raman spectra (Figure 3c,f) reveal that MDA-MB-468 differentiates well from the other cell lines. Considering the fingerprint region-derived dendrogram, MCF-7 and MDA-MB-231 are the cell lines that show the highest similarity, as opposed to the high-wavenumber region where the cell lines with the highest resemblance are HCC-1143 and MCF-7. Again, the distance scale shown in the dendrogram from the fingerprint region is larger, reflecting the highest variability in this spectral region.   For the Raman classification model, sensitivities of 63.6%-94.3%, specificities of 89.8%-97.2% and overall accuracies of 85.8-95.8% were obtained (Table 3), the latter being lower than those obtained with FTIR. MDA-MB-231 is the cell line with the highest number of misclassifications (with a limited sensitivity of 63.6% and specificity of 93.8%), and HCC-1143 and MDA-MB-468 present the lowest number of misclassified spectra (with sensitivities and specificities above 92% and an overall accuracy above 95%). The most incorrect classifications were found for MDA-MB-231, which was misclassified as MCF-7. This result was expected given the PCA scores plot shown in Figure 3a (MCF-7 and MDA-MB-231 not discriminated along either PC-1, PC-2 or PC-3). Although the number of spectra per class did not vary substantially in this study, and a large number of spectra were used, a careful analysis must be performed when using decision trees algorithms given their sensitivity to class imbalance.

TNBC versus Non-TNBC Cells
A PCA analysis was performed for the vibrational data obtained for TNBC (MDA-MB-231, HCC-1143 and MDA-MB-468) as compared to non-TNBC (MCF-7) cell lines (Figure 4ai), enabling to discriminate between them. For the infrared fingerprint region, separation was obtained along PC-1 (upon a PC-1 vs. PC-5 combination, representing 55.0% and 3.3% of the total data variance, respectively) based on: a higher contribution from ν as (PO 2 ) DNA (at 1238 cm −1 ) coupled to a lower input from ν(CC) RNA/bases (at 1302 cm −1 ) and Amide I (at 1650 cm −1 ) for TNBC cells. Regarding the Raman data, PC-2 explains 25.8% of the total data variance for the fingerprint interval, followed by PC-3 (7.4%), the separation between triple-negative and non-triple negative cells being more evident along PC-3, mainly according to: ν(CC) DNA/bases (at ca. 780-800 cm −1 ), Amide I (at ca. 1650 cm −1 ) and ν(PO 2 − ) DNA (at 1092 cm −1 ), which contribute more significantly for the TNBC cells, while α(CH 2 ) (at 1450 cm −1 ) is predominant for non-TNBC. For the Raman high wavenumber range, in turn, separation between triple-negative and non-triple-negative cell lines occurs along PC-2 (4.4%) based on the ν s (CH 2 /CH 3 ) modes (at ca. 2846-2970 cm −1 ), possibly due mostly to variations in the lipidic cellular content (which are known to be closely linked to invasiveness and metastatic ability, that characterise TNBC). Figure 4c,f,i shows the confusion tables of the FTIR classification models concerning the comparison between TBNC and non-TNBC cell lines. Spectra from TNBC achieved a higher percentage of correct classification than spectra from non-TNBC. Considering TNBC as the positive class, a sensitivity of 89.6% and a specificity of 81.7 % were obtained, with an overall accuracy of 87.3%. The Raman results led to higher sensitivities (94.4% and 97.0% for the fingerprint and high wavenumber regions, respectively), at the cost of poorer specificities (43.0% and 54.4% for the fingerprint and high wavenumber regions, respectively), with an overall accuracy of 80.5% and 85.4% respectively for the fingerprint and high wavenumber regions). This reflects the incomplete TBNC vs non-TNBC separation already evidenced in the corresponding PCA scores (Figure 4d,g).   Figure 4c,f,i shows the confusion tables of the FTIR classification models concerning the comparison between TBNC and non-TNBC cell lines. Spectra from TNBC achieved a higher percentage of correct classification than spectra from non-TNBC. Considering TNBC as the positive class, a sensitivity of 89.6% and a specificity of 81.7 % were obtained, with an overall accuracy of 87.3%. The Raman results led to higher sensitivities (94.4% and 97.0% for the fingerprint and high wavenumber regions, respectively), at the cost of poorer specificities (43.0% and 54.4% for the fingerprint and high wavenumber regions, respectively), with an overall accuracy of 80.5% and 85.4% respectively for the fingerprint and high wavenumber regions). This reflects the incomplete TBNC vs. non-TNBC separation already evidenced in the corresponding PCA scores (Figure 4d,g).
This discrimination between mesenchymal-and basal-like TNBC subtypes is noteworthy and suggests a significant biochemical variation among them. To the best of the authors knowledge, this is the first report of such a differentiation based on FTIR and Raman microspectroscopic techniques. Figure 5c,f,i depicts the confusion tables of the classification models concerning the MSL vs. TBNC BL1 cells. Regarding the FTIR data, sensitivities of 99.6% and 96.2%, and specificities of 99.9% and 99.4%, were achieved for the fingerprint and high-wavenumber regions, respectively (considering MSL as the positive class). The number of misclassified spectra were lower in the fingerprint region than in the high wavenumber interval (only 3 misclassified spectra in a total of 1124). The overall accuracy was 99.8% and 98.1%, for the fingerprint and high wavenumber ranges, respectively. The Raman results were reflected a slightly less clear separation (Figure 5g), with a sensitivity of 86.4%, a specificity of 93.2%, and an overall accuracy of 90.9%. A higher percentage of Raman spectra from MSL cells were misclassified when compared to TNBC BL1.

TNBC Basal-like 1 Cells-High-Metastatic versus Low-Metastatic Potential
Apart from being able to differentiate MSL from BL1 triple-negative breast cancer cells, discrimination between the TNBC BL1 cell lines under study was also attainedhigh-metastatic potential MDA-MB-468 vs. low-metastatic potential HCC-1143 ductal carcinomas (Figure 6a-l).

Discussion
Cancer cells display a high intertumour heterogeneity-between different types of tumour, distinct stages of neoplastic progression or even across geographical regions-not yet fully understood at the molecular level. This heterogeneity is due to both intrinsic and extrinsic parameters, such as genomic instability, epigenetic alterations, plastic signal transduction, stemness (ability for self-renewal and differentiation) and/or variations within the tumour microenvironment (e.g., hypoxia, vascularisation and inter-cellular interactions). The phenotypic diversity of cancer cells is correlated to their migratory and metastatic nature, as well as to chemoresistance [52,58,59], and has severe implications on patients prognosis since metastasis is one of the main reasons of cancer-related mortality and drug resistance is a major limiting factor in therapy response and clinical outcome. Furthermore, the multiple morphological signatures of cancer cells, marked by distinct morphodynamics, have been found to be associated to their ability to invade (upon epithelial-to-mesenchymal transition (EMT)) and migrate to distant loci [52,58]. A molecular level approach on this intra-tumour heterogeneity, such as the one currently performed, will allow to determine the main chemical differences between distinct types (and sub-types) of cancer, and is expected to help understand tumour evolution and assist in diagnosis and treatment.
Breast cancer, in particular, is a highly heterogeneous disease triggered by a variety of genetic alterations in mammary epithelial cells. It is associated to a large cell plasticity and a remarkable ability to adapt to changing physiological conditions (through rapid genetic/epigenetic reprogramming and transcriptomic/proteomic changes) [8,60]. Among breast tumours, TNBC s are those with a largest heterogeneity and highest mortality rates. To this date, there are no reliable biomarkers to identify these types of tumours, nor FDAapproved targeted therapies available for TNBC patients. The differentiation presently obtained between TNBC and non-TNBC cell lines is therefore a relevant result with a potential high clinical impact. This distinction was mainly based on DNA and protein vibrational signatures, which were found to predominate in the former while the signals from lipids appeared to be prevalent in non-TNBC cells. This agrees with the enhanced cell cycle and cell division constituents and pathways typical of triple-negative cancer cells.
Discrimination within the TNBC MSL and BL1 cell lines, in turn, rested on major DNA and lipid contributions for the mesenchymal stem cell-like MDA-MB-231 cells, as compared to a higher influence of proteins and membrane phospholipids for the basal-like MDA-MB-468 and HCC-1143 cells. These findings are compatible with the increased expression of genes involved in cell proliferation and differentiation, cell motility and growth signalling pathways in TNBC MSL tumours. In fact, the MDA-MB-231 cells, displaying a stellate morphology, are the most aggressive TNBC cells based on a low expression of cell-cell adhesion genes (low claudin levels, a key protein component of cellular junctions) coupled to a very high expression of epithelial-to-mesenchymal transition genes [14,61,62].
Finally, separation between the TNBC basal-like low and high-metastatic potential cell lines was established upon the larger impact of the protein constituents in the metastatic MDA-MB-468 ductal carcinoma (displaying a grape-like morphology) relative to HCC-1143. Actually, matrix enzymes such as metalloproteinases (MMPs) and heparanase are responsible for degradation of the extracellular matrix, which prompts epithelial-to-mesenchymal transition thus facilitating cancer cell motility, invasion and metastasis characteristic of the MDA-MB-468 cells.

Conclusions
Single-point FTIR and Raman acquisition were performed on different human breast cancer cells, delivering specific vibrational spectroscopic signatures that allowed us to identify biochemical changes between cellular subtypes with high accuracy.
Upon a chemometric (PCA) analysis, several degrees of discrimination were achieved among the cell lines under study: TNBC from non-TNBC, TNBC MSL from TNBC BL1 and TNBC BL1 with either low-metastatic or high-metastatic potential. Actually, besides a reliable detection of triple-negative human breast cancer, a highly specific differentiation between distinct subtypes of TNBC was obtained, which is an innovative and highly promising result, particularly aiming at an application in the clinics. The results currently obtained give an important insight over the expected subtle spectral changes among TNBC cell lines, bringing complementary knowledge to the already documented differences between normal and malignant tissues.
This molecular approach to a highly heterogeneous tumour such as breast cancer may greatly contribute to gain an in-depth understanding of the molecular changes underlying cell plasticity during neoplastic progression, which will hopefully allow an early identification of low prognosis breast cancers and the development of novel therapeutic strategies leading to an improved clinical outcome.
These results may also pave the way for a bench-top to bedside translation of FTIR and Raman microspectroscopy techniques, in a near future (to assist histopathological analysis), within the current efforts to apply cutting-edge technologies in personalised health oncology.