BBAMCR-17255; No. of pages: 11; 4C: 2, 5, 6 Biochimica et Biophysica Acta xxx (2014) xxx–xxx

Contents lists available at ScienceDirect

Biochimica et Biophysica Acta journal homepage: www.elsevier.com/locate/bbamcr

3Q1

M. Marro a,1, C. Nieva a,b,1, R. Sanz-Pamplona c, A. Sierra b,⁎ a

7

a r t i c l e

8 9 10 11 12

Article history: Received 25 November 2013 Received in revised form 7 April 2014 Accepted 10 April 2014 Available online xxxx

13 14 Q2 15 16 17 18 19

Keywords: Biomarker Breast cancer Epithelial-to-mesenchymal transition (EMT) Lipid Metabolism Raman spectroscopy

i n f o

R O

ICFO-Institut de Ciències Fotòniques, Parc Mediterrani de la Tecnologia, Av Carl Friedrich Gauss, 3, 08860 Castelldefels, Barcelona, Spain IDIBELL-Institut d'Investigació Biomèdica de Bellvitge, Av. Castelldefels, Km 2.7, 08907 L'Hospitalet de Llobregat, Barcelona, Spain Unit of Biomarkers and Susceptibility, Catalan Institute of Oncology, IDIBELL, and CIBERESP, Barcelona, Spain

a b s t r a c t

E

D

P

In breast cancer the presence of cells undergoing the epithelial-to-mesenchymal transition is indicative of metastasis progression. Since metabolic features of breast tumour cells are critical in cancer progression and drug resistance, we hypothesized that the lipid content of malignant cells might be a useful indirect measure of cancer progression. In this study Multivariate Curve Resolution was applied to cellular Raman spectra to assess the metabolic composition of breast cancer cells undergoing the epithelial to mesenchymal transition. Multivariate Curve Resolution analysis led to the conclusion that this transition affects the lipid profile of cells, increasing tryptophan but maintaining a low fatty acid content in comparison with highly metastatic cells. Supporting those results, a Partial Least Square-Discriminant analysis was performed to test the ability of Raman spectroscopy to discriminate the initial steps of epithelial to mesenchymal transition in breast cancer cells. We achieved a high level of sensitivity and specificity, 94% and 100%, respectively. In conclusion, Raman microspectroscopy coupled with Multivariate Curve Resolution enables deconvolution and tracking of the molecular content of cancer cells during a biochemical process, being a powerful, rapid, reagent-free and non-invasive tool for identifying metabolic features of breast cancer cell aggressiveness at first stages of malignancy. © 2014 Published by Elsevier B.V.

C

c

T

b

E

37 35 34 36

1. Introduction

39 40

Genetically altered neoplastic cells have special metabolic requirements [1]. Recent molecular studies of cancer revealed that oncogenes directly affect cellular energy metabolism [2]. An early and universal feature of tumours is the activation of lipid metabolism secondary to the activation of the lipogenic enzymes in the malignant process [3]. Increased lipogenesis is a hallmark of many cancers including breast carcinomas [4]. Lipid metabolism related genes in breast cancer cells have been studied describing a lipogenic phenotype [3]. Gene silencing experiments with seven genes (ACACA, ELOVL1, FASN, INSIG1, SCAP, SCAP, SCD, THRSP) indicated that reduction of the lipidomic profiles disrupted the viability of breast cancer cells [5]. In addition, the constitutive activation of signalling cascades that stimulate cell growth has a profound impact on anabolic metabolism [6]. One of the principal mechanisms of aerobic glycolysis resides in

47 48 49 50 51 52 53

R

N C O

45 46

U

43 44

R

38

41 42

O

4 5 6

F

2

Molecular monitoring of epithelial-to-mesenchymal transition in breast cancer cells by means of Raman spectroscopy

1

⁎ Corresponding author at: Institut d'Investigacions Biomèdiques August Pi i SunyerIDIBAPS, Laboratory of Molecular and Translational Oncology, Centre de Recerca Biomèdica CELLEX-CRBC, Planta 1A, Casanova, 143, 08036 Barcelona, Spain. Tel.: +34 93 227 54 00x4801, +34 649790094 (Mobile). URLs: http://www.hospitalclinic.org/, http://www.idibaps.org/ (A. Sierra). 1 These authors contributed equally to this work.

the activation of hypoxia-inducible factor (HIF), a transcription factor activated by hypoxic, oncogenic, metabolic and oxidative stress, and also involved in epithelial-to-mesenchymal transition (EMT), the initial signal of the lethal metastatic phenotype of breast cancer cells [7]. Indeed, the preferential expression of EMT-related genes has been found in basal-like breast tumours, the most invasive breast carcinomas [8] with the worst prognosis and greatest resistance to chemotherapy [9]. Basal-like tumour cells express markers characteristic of the normal mammary gland myoepithelium, such as epidermal growth factor receptor (EGFR), p63 and basal cytokeratins CK14, CK5/6 and CK17 [10]. Cellular remodelling occurring as a consequence of EMT, whereby cells have altered responses to agents in the circulatory system or secondary tumour site, could be advantageous for the process of metastasis [11,12]. EMT confers mesenchymal properties on epithelial cells and has been closely associated with the acquisition of aggressive traits by carcinoma cells [13]. Moreover, the dynamic interactions among epithelial, self-renewal and mesenchymal gene programmes determine the plasticity of epithelial tumour-initiating cells [14]. A close relationship between changes in fatty acid metabolism is associated to cell motility and the proclivity to breast cancer cells undergoing EMT [15]. We have previously characterized the metabolic phenotype of breast cancer cells by using Raman microspectroscopy (RS) combined with Principal Component Analysis (PCA), based on a panel of small molecules derived from the global or targeted analysis

http://dx.doi.org/10.1016/j.bbamcr.2014.04.012 0167-4889/© 2014 Published by Elsevier B.V.

Please cite this article as: M. Marro, et al., Molecular monitoring of epithelial-to-mesenchymal transition in breast cancer cells by means of Raman spectroscopy, Biochim. Biophys. Acta (2014), http://dx.doi.org/10.1016/j.bbamcr.2014.04.012

20 21 22 23 24 25 26 27 28 29 30 31 32 33

54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 Q3 75 76 77

93 94 95 96 97 98 99 100 Q4 101 102 103 104 105 106 107 108 109 110 111 112

2. Material and methods

114

2.1. Cell cultures

115

124

MCF10A cells were obtained from the American Type Culture Collection and were grown in DMEM/F12 medium supplemented with 5% horse serum, 1 mM pyruvate, 2 mM L-glutamine, 0.01 mg/ml bovine insulin, 20 ng/ml EGF, 1 mg/ml hydrocortisone and 100 ng/ml tetanus toxin in 5% CO2–95% air at 37 °C in a humidified incubator. MDA-MB-435 cells maintained under standard conditions in a 1:1 (v/v) mixture of DMEM and Ham F12 medium (DMEM/F12) supplemented with 10% foetal bovine serum (FBS), 1 mM pyruvate and 2 mM L-glutamine, in the same incubator conditions as described above were used in some experiments [32].

125

2.2. Sample preparation

126

Each measurement required 3 × 105 cells, or 3 × 104 cells for MCF10A cells in confluent or sparse conditions. Cells were seeded in six-well plates (Becton Dickinson, NJ) over a quartz crystal (ESCO products, Oak Ridge, NJ), which was used to reduce the background signal. After 24 h, the cells were fixed with 4% cold paraformaldehyde (PFA) in PBS 1× for 15 min, washed 3 times for 10 min with PBS 1 × (until no PFA residues were present) and maintained in the same PBS 1× solution at 4 °C until measurements. The complete washing of the cells also ensures that no residues of different culture media or other chemicals are present and therefore the cells are measured under the same conditions.

127 128 129 130 131 132 133 134 135 136

R

O

C

122 123

142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 Q5 175 176 177 178 179 180 181 182

N

120 121

140 141

U

118 119

R

113

116 117

F

91 92

O

89 90

138 139

R O

87 88

The Renishaw Raman system (Apply Innovation, Gloucestershire, UK) comprises a 514 nm laser that supplies an excitation beam of about 5 mW power, which is focused onto the sample via a microscope using an air 60× objective (Leica, 0.75 NA). The same objective collects the scattered light from the sample and directs it to the spectrometer. The spectrometer processes this scattered light, by rejecting the unwanted portion and separating the remainder into its constituent wavelengths. The Raman spectrum is recorded on a deep depletion charge-coupled device (CCD) detector (Renishaw RenCam). The recorded Raman spectrum is digitalized and displayed on a personal computer using Renishaw WiRE software, which allows the experimental parameters to be set. For experiments, one spectrum was collected per cell in the cytoplasm, near the nucleus but outside the endothelial reticulum [15], focusing light at 3 μ inside the quartz window. We were interested in obtaining the most number of cells measured rather than study the intracellular domain. That is why we acquired one Raman spectra per cell, obtaining a total of 20 spectra per cell line studied: MCF10 confluent, MCF10 sparse and MDA-MB-435. To ensure that changes in the spectra among different cell lines were due to its intrinsic biochemical content and not due to changes in other factors that could be out of control (setup alignment, background signals, temperature, conditions in sample preparation …), we measure all cell lines in the same day under exactly the same conditions. Replications of the experiment were performed in other different days and were comparable, providing similar results. The quartz in which the cells were grown and fixed was translated to a magnetic fluid chamber (Live Cell Instruments), filled with PBS and closed with other quartz coverslip. For Raman measurements the quartz containing the cells was placed on the top (with cells inside the fluid chamber) to be measured with the upright microscope. Before including Raman spectra in the multivariate statistical techniques, correct pre-processing must be performed. Spectra shown in Fig. 1 have been pre-processed using the following methods: subtracting fluorescence background [15], comic spike removal, smoothing (5 points averaging) and multiple scattering correction (MSC); MSC algorithm was used to correct for differences in global intensity among the spectra. We used a custom-made Labview and Matlab code to perform all these preprocessing methods. However because there is still the presence of background signals from quartz or PBS, with the use of MCR-ALS algorithm we subtracted those signals at the same time that we looked for molecular components (explained in the next section on Statistical analysis). For Raman spectral analysis, the region between 1015 and 1110 cm−1 was removed because it contained a background-related signal that reduced the quality and interpretability of our statistical models.

P

85 86

137

D

84

2.3. Raman spectroscopy

T

82 83

C

80 81

of metabolic profiles of cells. RS combined with PCA has emerged as a useful technique for stratifying breast cell malignancy, being well correlated with the EMT phenotype and the expression of SREBP-1c and ABCA1 genes [16]. Since, metabolic features of breast tumour cells are critical in cancer progression and drug resistance [9,17], there exists a need to develop alternative/adjunct, user-freed, cost effective, rapid, objective and unambiguous methods for the early detection and diagnosis of metastatic breast cancer cells. Recently, some unmixing algorithms have been used to decompose the Raman spectra into different molecular components in areas such as Raman imaging: Vertex Component Analysis (VCA) [18–20] and Multivariate Curve Resolution (MCR) [21]. In the last decade, MCR algorithm has been applied in a broad range of chemical analysis [22–24] and recently to Raman images [25,26]. Multivariate Curve Resolution-Alternating Least Squares (MCR-ALS) stands out because of its flexibility to include initial estimates and constrains in the iterative resolution procedure [21]. The output of the MCR-ALS analysis gives more chemically and physically understandable results than classical techniques such as PCA, where only a mathematical exploration of the data can be performed [27–30]. Thus, being a powerful candidate tool to deconvolve key molecular signatures from the Raman spectra and monitor the biochemical composition of the sample over the process studied [31]. In the present research we used MCR-ALS algorithm to study cellular RS, which allowed the deconvolution of meaningful molecular RS of biomolecules that are related to metabolites that change concentration in the cells under study. Furthermore, and supporting MCR results, a Partial Least Square-Discriminant analysis (PLS-DA) was performed to test the ability of discrimination with RS of the EMT from initiating breast cells, achieving a high level of sensitivity and specificity. Overall the results presented in this paper report the first application of MCR to RS to deconvolve and track the molecular content of initiating breast cancer cells during the EMT process, suggesting that RS combined with MCR and PLS-DA is a powerful diagnostic tool for identifying metabolic features of the first stages of malignancy and breast cancer cell aggressiveness.

E

78 79

M. Marro et al. / Biochimica et Biophysica Acta xxx (2014) xxx–xxx

E

2

Fig. 1. Raw Raman spectra from MCF10 confluent and MCF10 sparse cells and MDA-MB- Q15 435 metastatic cells. The spectra shown have been the background subtracted similarly as in [15] and a Multiple Scatter Correction was performed.

Please cite this article as: M. Marro, et al., Molecular monitoring of epithelial-to-mesenchymal transition in breast cancer cells by means of Raman spectroscopy, Biochim. Biophys. Acta (2014), http://dx.doi.org/10.1016/j.bbamcr.2014.04.012

M. Marro et al. / Biochimica et Biophysica Acta xxx (2014) xxx–xxx

Transcriptomic data from MCF10A cells cultured in sparse or confluent conditions with accession number GSE8430 [12] were downloaded from the Gene Expression Omnibus database [33]. In addition, a series of matrix including normalized data from GSE18070 that contains different strands of k-ras transfected MCF10A cells was downloaded from the GEO repository [33]. To assess gene expression differences, a t-test between all probe sets from three hybridized MIII samples (cells with ability to metastasize in vivo) and three MII samples (cells without metastasis ability) was conducted. Then the 33,853 genes included in the array (Affymetrix U133 Plus 2.0) were ranked based on their statistic expression level.

203 204 205 206 207 208 209 210 211 212 213 214 215 216 Q6 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231

235 236 237 238 239 240 241 242 243 244

C

201 202

E

199 200

R

197 198

R

195 196

N C O

193 194

U

191 192

F

234

189 190

O

2.5. Gene expression analysis

188

3. Results

R O

233

186 187

245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262

263

3.1. Multivariate Curve Resolution (MCR) algorithm provides an accurate 264 evaluation of Raman spectroscopy (RS) 265

P

232

Two multivariate techniques were performed on cellular Raman spectra, in order to evaluate the spectral differences among the cancerous cell lines studied and to develop a model allowing their discrimination and classification: MCR and PLS-DA. In Multivariate Curve Resolution (MCR) [21], a bilinear relation of the spectrum is assumed from the matrix, taking into account that the total Raman spectra contained in X matrix can be modelled as the contribution of pure molecular spectra (S) multiplied by the concentration of each pure molecular component (C), X = CST + E, where X is the Raman spectra acquired matrix, C is the concentration matrix transposed and E is the matrix of spectral errors. The algorithm used was Multivariate Curve Resolution-Alternating Least Squares (MCR-ALS). Using a first PCA on the dataset of spectra, the number of components necessary to explain the variance in the data was selected, accounting for 95% of the total variance. For ALS optimization, constraints of non-negativity in the spectra and concentrations were applied. A limit of 300 iterations was applied to reach the 0.95 confidence limits. MCR-ALS allowed visualization of the bands (biomolecular vibrations) responsible for the changes in metabolite concentrations in the different cell lines. Finally, five molecular components were deconvolved from the spectra. In order to deconvolve from the spectra the possible contributions of the background signals we include in the Raman spectra matrix X, 5 spectra obtained from the PBS solution at the same focal distance where the cells were measured (3 μm). In this way, the algorithm extracts one component that contains the spectral profiles of background signals from the quartz window and PBS solution. These five spectra are represented by stars in the plots and they always contain very low levels of other cellular components as expected. A remaining contribution from quartz signal was arising at 1100 cm− 1 in other components, and for this reason we exclude from the analysis this region to ensure that not the interference of the components deconvolved was mixed with background signals. PLS-DA [15] is a supervised classification method in which knowledge of the sample is included (in our case metastatic, malignant and benign MCF10A phenotypes). PLS-DA employs the fundamental principle of PCA but further rotates the component (latent variables, LVs) by maximizing the covariance between the spectral variation and group affinity so that the LVs explain the diagnostically relevant variations rather than the most prominent variations in the spectral dataset. The performance of the PLS-DA diagnostic algorithm was validated using leave-one-out cross validation methodology. The number of retained LVs was determined based on the minimal root mean square error of cross validation (RMSECV) curves. Multivariate statistical analysis was performed using the PLS toolbox (Eigenvector Research, Wenatchee, WA) in the Matlab (Mathworks Inc., Natick, MA) programming environment. Before including Raman spectra in the multivariate statistical techniques, correct preprocessing must be performed (see Raman spectroscopy section).

We use MCR to efficiently disentangle pure molecular information encoded in complex spectrum such as Raman. To distinguish the metabolic features of cells with the EMT phenotype [37,38], we used MCF10A cells, benign breast tumour cells, unable to spread outside the basal membrane, despite their basal-like phenotype. It has been described that MCF10A sparse cultures presented many features of mesenchymal cancer when compared with MCF10A confluent [12]. The more aggressive MDA-MB-435 cell line showed an even more noticeable mesenchymal phenotype characterized by a high level of expression of vimentin and N-cadherin [39,40] and a down-regulation of E-cadherin [40]. Therefore we used MCF10A confluent and sparse culture conditions as a signal of the initiation of metastasis progression, to characterize the spectroscopic parameters of a prototype of metastasis-initiating cells in comparison with the highly metastatic MDA-MB-435 breast cancer cells, a widely recognized metastatic cells in which the degree of metastaticity has been related to the polyunsaturated fatty acid contents by Raman analysis [41]. Since we have previously distinguished confluent and sparse MCF10A, according to the total unsaturated fatty acids (TUFA) band intensity (in the C\H stretching region of the spectra (2800–3100 cm−1), [15], RS of both phenotypes (confluent and sparse MCF10A) was acquired in the fingerprint region (600–1800 cm−1). Fig. 1 shows the row spectrum of the three different phenotypic cells analysed (MCF10 sparse, confluent and MDA-MB-435) after background subtraction and MSC correction. Band characteristic of the Raman spectra of molecules is presented in Table 1. The MCR-ALS algorithm was then used to deconvolve from these raw spectra the metabolites that differed among the cell groups. One problem that could arise in resolution analysis methods is that the components deconvolved from the experimental Raman spectra are mixed with background signals and are therefore difficult to interpret. This problem was solved by adding to the data set a small number of spectra obtained from the PBS (buffer, not cells) at the same focal distance at which the cellular spectra were acquired. Including this extra information in the MCR-ALS algorithm allows the deconvolution of a component that contains only background spectra (quartz and water). Therefore, the remaining components only account for the differences in the metabolite content in the cells. Four pure spectra were deconvolved from inherent cellular components and are shown in Fig. 2. Score plots relating to different components are presented in Fig. 3. The control spectra from PBS were plotted on the score plot as blue crosses and were always around the lowest levels of

D

184 185

A pre-ranked GSEA (Gene Set Enrichment Analysis) using both lists of ranked genes was run. Only 81 gene sets were related to “fatty acid”, “lipid”, “glucose”, “cholesterol”, “EMT” and “tryptophan”; and those gene sets were described by Charafe et al. [34] including differentially expressed genes between different kinds of breast cancer that were included in the analysis. The statistical significance of the enrichment score was calculated by permuting the genes 1000 times as implemented in the GSEA software. Functional terms were considered to be significant at FDR q-value of 25%. The conventional statistical methods for microarray gene expression analysis choose a list of differentially expressed genes based on a p-value corrected by multi-testing. Using this astringent criterion, a large number of genes that do in fact contribute to the studied phenotype could be missed. The GSEA (gene set enrichment analysis) approach was designed to address the limitation of single gene analysis since it uses all transcriptomic information included in the array [35]. All computations were performed using R statistical software and Bioconductor [36].

E

2.4. Statistical analysis

T

183

3

Please cite this article as: M. Marro, et al., Molecular monitoring of epithelial-to-mesenchymal transition in breast cancer cells by means of Raman spectroscopy, Biochim. Biophys. Acta (2014), http://dx.doi.org/10.1016/j.bbamcr.2014.04.012

266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 Q7 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306

4 t1:1 t1:2 t1:3

M. Marro et al. / Biochimica et Biophysica Acta xxx (2014) xxx–xxx

Table 1 Band assignments for the Raman spectra corresponding to components 1, 2, 4 and 5 obtained by MCR analysis of sparse and confluent MCF10A cells and MDA-MB-435 cells. Assignments were based on Refs. [34–37]. PA: phenylalanine; FA: fatty acids; T: tryptophan; K: keratin; A: amide; PS: polysaccharides.

t1:4

Component 1

t1:5

Band (cm−1)

Assignment

Band (cm−1)

Assignment

Band (cm−1)

Assignment

Band (cm−1)

Assignment

t1:6 t1:7 t1:8 t1:9 t1:10 t1:11 t1:12 t1:13 t1:14 t1:15 t1:16

1001 1170 1204 1237 1300 1315 1336 1359 1448 1479 1587

PA PA T AIII T T T T T, CH2 deformation T, AII T

1123 1260 1300 1438

FA FA FA FA

940 986 1125 1187 1207 1337 1365 1452

PS PS PS PS PS PS PS PS

936 1001 1235 1240 1296 1335 1448 1650

K PA AIII, K AIII, K K K K K

F O

R O

P

D

E T C E R

317 318

R

315 316

Component 2 included bands assigned to fatty acids: 1123, 1260, 1300 and 1438 cm− 1. All the bands contained in component 5 were identified as keratin (936, 1296, 1335, 1448 and 1650 cm−1) and amide III (1235 and 1240 cm−1). Therefore, MCR resolved meaningful components in the spectra with respect to classical techniques such as Principal Component Analysis (PCA), demonstrating a high capacity to deconvolve the pure cell component spectra of molecules from a set of Raman spectra. Moreover, the outputs of the statistical analysis provided more chemically and physically understandable results than classical techniques such as PCA, via which only a mathematical exploration of the data can be performed and components have mixed molecular contributions and negative loadings.

O

313 314

the concentration of the different cellular metabolites, indicating the absence of those molecules from the spectra. With the help of previously published Raman databases [42–45] and measuring the Raman spectra of pure molecules, we identified component 1 as tryptophan, component 2 as fatty acids and component 5 as cytokeratin. Component 4 did not have a priori clear assignment, whereas component 3 accounted only for background contribution (Fig. 2). Table 1 lists the assigned metabolites in each component [43,44]. Component 1 included mainly phenylalanine and tryptophan. Two bands, at 1001 cm − 1 and 1170 cm − 1 , correspond to phenylalanine, and six to tryptophan: 1204, 1300, 1315, 1336, 1359 and 1587 cm− 1.

C

311 312

Component 5

N

309 310

Component 4

U

307 308

Component 2

Fig. 2. Component spectra obtained by Raman microspectroscopy deconvolved from MCR analysis using sparse and confluent MCF10A cells and MDA-MB-435 cells. The main characteristic bands of components 1, 2, 4 and 5 are indicated in each spectrum. Assignments were based on Refs. [34–36].

Please cite this article as: M. Marro, et al., Molecular monitoring of epithelial-to-mesenchymal transition in breast cancer cells by means of Raman spectroscopy, Biochim. Biophys. Acta (2014), http://dx.doi.org/10.1016/j.bbamcr.2014.04.012

319 320 321 322 323 324 325 326 327 328 329 330

5

E

C

T

E

D

P

R O

O

F

M. Marro et al. / Biochimica et Biophysica Acta xxx (2014) xxx–xxx

N C O

R

R

Fig. 3. MCR score plot for components represented in Fig. 2. Confluent MCF10A cells (green asterisks), sparse MCF10A cells (red triangles) and MDA-MB-435 cells (dark blue squares). Light blue crosses are control spectra obtained from the PBS solution at the same focal depth. The arrows indicate an increase in the epithelial-to-mesenchymal transition (EMT), the lipid phenotype (LP), the tryptophan (T) content or the polysaccharide (PS) content. a) Component 5 (keratin), a marker of EMT, is plotted with component 2, which was assigned to fatty acids, an indicator of the LP. b) Component 1, assigned to tryptophan, is plotted with component 2 (fatty acids). c) Keratin is plotted with component 1, assigned to tryptophan. MCF10A cells in both culture conditions are completely separated by differences in the levels of both components. d) Component 4, assigned hypothetically to polysaccharides, is plotted with component 2 (fatty acids). MDA-MB-435 cells are clearly split into two subpopulations: SP435A (with higher lipid content) and SP435B (with lower lipid content).

3.2. RS dissects the metabolic phenotype of EMT in breast cancer cells

332

We conducted an in-depth analysis of the most significant differences between confluent and sparse MCF10A cells to assess the different profiles of the spectral composition and to build a classification algorithm for the discrimination of EMT cells with the metastaticinitiating phenotype from the highly metastatic MDA-MB-435 cells. We plotted the contributions of component 2 (fatty acids) with respect to component 5 (keratin) (Fig. 3a). EMT appeared inversely correlated with keratin, with MDA-MB-435 cells having low keratin level whereas confluent MCF10A cells had the highest (Fig. 3a and c). Component 2 (fatty acid) separated two MDA-MB-435 cell populations: SP435A with the highest fatty acid content, and SP435B with a moderate fatty acid increase. In contrast, sparse MCF10A cells had a low keratin and increased fatty acid content compared with confluent MCF10A cells, more similar to SP435A. Indeed, the most lipid accumulating cells were sparse MCF10A and SP435A. Thus, RS accurately classified EMT based on a low keratin and a high lipid content, so-called lipid phenotype (LP). In addition, component 1 identified as tryptophan (T) clearly segregated sparse MCF10A from the rest of the cells, suggesting that high levels of this metabolite were an idiosyncratic element of the EMT phenotype (Fig. 3b and c). Taking into account scores on

333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351

U

331

tryptophan and fatty acids (Fig. 3b), the aggressive metastatic phenotype of MDA-MB-435 cells was strongly associated with an increase in phospholipids with low tryptophan content (Fig. 3b). On the other hand, MDA-MB-435 cells were split into two clear groups, one with low lipid content and the other with the highest lipid content, according to its heterogeneity [46]. Moreover, in Fig. 3d, it was possible to distinguish the highest fatty acid SP435A cells (with a lipid phenotype) from SP435B cells (with a low lipid concentration and increased levels of component 4), suggesting that this component may act as an alternative fuel source for MDA-MB-435 cells with a lower lipid content [46]. In addition, some bands in component 4 (see Table 1) can be assigned to polysaccharides (PS). Therefore, component 4 could be related to glucose metabolism. Furthermore, the cloud of sparse MCF10A cells was displaced with respect to confluent MCF10A cells in Fig. 2d. Sparse MCF10A showed a high LP and low PS content whereas confluent MCF10A have higher PS and lower LP which supports PS as an alternative fuel source for low lipid accumulating cells. To support MCR results and investigate the biomarker usefulness of RS for diagnostic applications, we performed a PLS-DA model in order to explore the ability of RS to differentiate between sparse and confluent MCF10A and MDA-MB-435 cells. The model (Fig. 4) gave 100% sensitivity and 100% specificity for discrimination of sparse MCF10A, 94%

Please cite this article as: M. Marro, et al., Molecular monitoring of epithelial-to-mesenchymal transition in breast cancer cells by means of Raman spectroscopy, Biochim. Biophys. Acta (2014), http://dx.doi.org/10.1016/j.bbamcr.2014.04.012

352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 Q8 369 370 371 372 373

M. Marro et al. / Biochimica et Biophysica Acta xxx (2014) xxx–xxx

P

R O

O

F

6

380

3.3. Differential expression of EMT genes supports RS metabolic phenotype

381

We performed functional enrichment analyses to support the spectral differences between confluent and sparse MCF10A cells according to transcriptomic analysis. We used public transcriptomic data downloaded from Gene Expression Omnibus database GSE8430 [12] to search for functional differences between sparse and confluent cells. This dataset contains a normalized log ratio (z-score) for the differences in expression of 11,500 probe sets representing 9300 genes from the CNIO Homo sapiens Oncochip 2. Genes were ranked by their median z-score and gene set enrichment analysis (GSEA) algorithm [47] was applied to identify specific functions in the list of pre-ranked genes. A total of 71 gene sets including the keywords “fatty acid”, “lipid”, “glucose”, “cholesterol”, “EMT” and “tryptophan”, and those gene sets described by Charafe et al. [34], including genes that are differentially expressed in different kinds of breast cancer, were tested (Table 2). Genes that were downregulated in luminal-like breast cancer cell lines in comparison with mesenchymal-like ones [34] were expressed in sparse MCF10A (FDR b 0.000), whereas confluent MCF10A contained genes that were upregulated in luminal-like breast cancer cell lines in comparison with the mesenchymal-like ones (FDR b 0.000). As expected, sparse cells showed a significant enrichment in genes that were upregulated in EMT whereas confluent cells were enriched in EMT downregulated genes, pointing to a mesenchymal phenotype. Surprisingly, both up- and downregulated EMT genes described by Jechlinger et al. [48] were significant gene sets in confluent cells, but downregulated genes were much more statistically significant (FDR = 0.008). Fatty acid and glucose metabolic functions differentiated confluent and sparse MCF10A cells. The metabolic phenotype of sparse MCF10A

388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408

C

E

R

R

O

386 387

C

384 385

N

382 383

U

376 377

cells was characterized by genes involved in glucose deprivation (FDR = 0.0020) and glucose transport (FDR = 0.120), whereas in confluent MCF10A cells the profiling showed an enrichment in genes involved in glucose metabolism (FDR = 0.032). Moreover, differences included increased fatty acid metabolism (FDR = 0.240) and fatty acid oxidation (FDR = 0.150) in confluent MCF10A. Since fatty acid metabolism was increased in these cells their glucose dependence was minimized with regard to sparse cells, suggesting that confluent and sparse MCF10A cells have a different metabolic phenotype in both conditions. One of the first metabolic alterations observed in cancer cells was the higher dependence upon glucose for their growth, the so-called Warburg effect, indicating that the EMT process in sparse MCF10A cells altered their metabolism, giving them more oncogenic properties [49]. Moreover, tryptophan metabolism and tryptophan catabolism were statistically significant functions (FDR = 0.190 and 0.136 respectively) in confluent cells, which expressed lower levels of tryptophan than sparse cells. Due to the low number of metabolism-related genes represented in the GEO2603 chip that hybridized with sparse and confluent MCF10A cells (Supplementary Table 1), an alternative analysis using data from Papageorgis et al. [50] was conducted. To assess differences in gene expression, a t-test between all probe sets from three hybridized MII samples (k-ras transfected MCF10A cells without metastasis ability) and MIII samples (k-ras transfected MCF10A cells with ability to metastasize in vivo) was conducted at the transcriptomic level. The 33,853 genes included in the array (Affymetrix U133 Plus 2.0) were ranked by the t-statistic. In this way, positive values corresponded to a MIII phenotype whereas negative values were characteristic of MII. A GSEA using the same gene sets as those in the GSE8430 study was then run. A total of 33,853 genes were hopefully hybridized on the chip, so more metabolism-related genes were included in the functional enrichment analysis. Significantly enriched functions at FDR 25% were found such as regulation of insulin secretion by free fatty acids (FDR = 0.150) in MIII cells, whereas in MII cells (see Table 3) glucose transporters (FDR = 0.189), lipid rafts (FRD = 0.029), phospholipid binding (FDR = 0.124) and cholesterol biosynthesis (FDR = 0.002), among others, were enhanced.

T

378 379 Q9

sensitivity and 89% specificity for confluent MCF10A and 80% sensitivity and 100% specificity for MDA-MB-435 using the cross validation method venetian blinds with 7 splits. Overall there was a clear stratification method to differentiate the characteristic fatty acid accumulation, which enriched by the increased tryptophan concentration in sparse MCF10A cells performed an initial progression phenotype.

E

374 375

D

Fig. 4. Results of PLS-DA analysis to explore the ability of RS to differentiate between sparse and confluent MCF10A cells and MDA-MB-435 cells. The model gives 100% sensitivity and 100% specificity for sparse MCF10A discrimination (a), 94% sensitivity and 89% specificity for confluent MCF10A discrimination (b), and 80% sensitivity and 100% specificity for MDA-MB-435 identification (c) using the cross-validation method venetian blinds with 7 splits.

Please cite this article as: M. Marro, et al., Molecular monitoring of epithelial-to-mesenchymal transition in breast cancer cells by means of Raman spectroscopy, Biochim. Biophys. Acta (2014), http://dx.doi.org/10.1016/j.bbamcr.2014.04.012

409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445

N

Table 2 Functional enrichment in sparse and confluent MCF10 A cells.

t2:3

GSEA gene set standard name

t2:4 t2:5 t2:6 t2:7

Sparse Cui glucose deprivation Reactome glucose transport Sarrio epithelial mesenchymal transition up

t2:8

Gotzmann epithelial to mesenchymal transition up Charafe breast cancer luminal vs mesenchymal down Charafe breast cancer basal vs mesenchymal down

t2:9 t2:10 t2:11 t2:12 t2:13

t2:14 t2:15

Confluent Glucose metabolic process

t2:16 t2:17 t2:18

KEGG fatty acid metabolism Reactome activated AMPK stimulates fatty acid oxidation in muscle KEGG tryptophan metabolism Reactome tryptophan catabolism Jechlinger epithelial to mesenchymal transition down

t2:19

Gotzmann epithelial to mesenchymal transition down

t2:20

Sarrio epithelial mesenchymal transition down

t2:21

Jechlinger epithelial to mesenchymal transition up

t2:22 t2:23 t2:24

Charafe breast cancer basal vs mesenchymal up Charafe breast cancer luminal vs mesenchymal up Charafe breast cancer luminal vs basal down

C

O

Brief description

R

Representative genes up-regulated in MiaPaCa2 cells (pancreatic cancer) under glucose-deprived conditions. Genes involved in glucose transport Genes up-regulated in MCF10A cells (breast cancer) grown at low (mesenchymal phenotype) compared to those grown at high (epithelial, basal-like phenotype) confluence. Genes up-regulated in MMH-RT cells (hepatocytes displaying an invasive, metastatic phenotype) during epithelial to mesenchymal transition (EMT). Genes down-regulated in luminal-like breast cancer cell lines compared to the mesenchymal-like ones Genes down-regulated in basal-like breast cancer cell lines as compared to the mesenchymal-like ones.

R

E

C

T

Genes annotated by the GO term GO: 0006006. The chemical reactions and pathways involving glucose, the aldohexose gluco-hexose. D-glucose is dextrorotatory and is sometimes known as dextrose; it is an important source of energy for living organisms and is found free as well as combined in homo- and hetero-oligosaccharides and polysaccharides. Fatty acid metabolism Genes involved in activated AMPK stimulates fatty-acid oxidation in muscle

E

D

P

Tryptophan metabolism Genes involved in tryptophan catabolism Genes down-regulated during epithelial to mesenchymal transition (EMT) induced by TGFB1 in the EpH4 cells (mammary epithelium cell line transformed by HRAS. Genes down-regulated in MMH-RT cells (hepatocytes displaying an invasive, metastatic phenotype) during epithelial to mesenchymal transition (EMT). Genes down-regulated in MCF10A cells (breast cancer) grown at low (mesenchymal phenotype) compared to those grown at high (epithelial, basal-like phenotype) confluence. Genes up-regulated during epithelial to mesenchymal transition (EMT) induced by TGFB1 in the EpH4 cells (mammary epithelium cell line transformed by HRAS. Genes up-regulated in basal-like breast cancer cell lines as compared to the mesenchymal-like ones. Genes up-regulated in luminal-like breast cancer cell lines compared to the mesenchymal-like ones. Genes down-regulated in luminal-like breast cancer cell lines compared to the basal-like ones.

R O

Source

Nominal p-value

FDR q-value

Pubmed: 17409444 REACTOME database Pubmed: 18281472

0.002 0.046 b0.000

0.002 0.120 b0.000

Pubmed: 16607286

0.009

0.031

Pubmed: 16288205 Pubmed: 16288205

b0.000 0.238

b0.000 0.144

GO database

0.012

0.032

KEGG database REACTOME database

0.032 0.043

0.24 0.15

KEGG database REACTOME database Pubmed: 14562044

0.100 0.146 0.002

0.190 0.136 0.008

Pubmed: 16607286

b0.000

0.016

Pubmed: 18281472

0.002

0.015

0.030

0.067

b0.000 b0.000 b0.000

b0.000 b0.000 b0.000

O

Pubmed: 14562044

F

Pubmed: 16288205 Pubmed: 16288205 Pubmed: 16288205

M. Marro et al. / Biochimica et Biophysica Acta xxx (2014) xxx–xxx

Please cite this article as: M. Marro, et al., Molecular monitoring of epithelial-to-mesenchymal transition in breast cancer cells by means of Raman spectroscopy, Biochim. Biophys. Acta (2014), http://dx.doi.org/10.1016/j.bbamcr.2014.04.012

t2:1 t2:2

U

7

8

Table 3 Functional enrichment in MIII and MII cells.

U

t3:3

GSEA gene set standard name

t3:4 t3:5 t3:6

MIII Reactome regulation of insulin secretion by free fatty acids Jechlinger epithelial to mesenchymal transition up

t3:7 t3:8

Anastassiou cancer mesenchymal transition signature Gotzmann epithelial to mesenchymal transition up

t3:9

Sarrio epithelial mesenchymal transition up

t3:10 t3:11 t3:12 t3:13 t3:14 t3:15 t3:16

Charafe breast cancer luminal vs mesenchymal down Charafe breast cancer basal vs mesenchymal down Charafe breast cancer luminal vs basal down

N

MII Reactome glucose and other sugar SLC transporters Lipid raft

Lipid homeostasis

t3:18

Lipid transporter activity

t3:19

Phospholipid binding

t3:20 t3:21 t3:22

Reactome cholesterol biosynthesis Reactome transformation of lanosterol to cholesterol Reactome synthesis of bile acids and bile salts via 24-hydroxycholesterol Goering blood HDL cholesterol QTL trans

t3:24

Source

Nominal p-value

FDR q-value

Genes involved in regulation of insulin secretion by free fatty acids Genes up-regulated during epithelial to mesenchymal transition (EMT) induced by TGFB1 in the EpH4 cells (mammary epithelium cell line transformed by HRAS). Genes in the ‘mesenchymal transition signature’ common to all invasive cancer types. Genes up-regulated in MMH-RT cells (hepatocytes displaying an invasive, metastatic phenotype) during epithelial to mesenchymal transition (EMT). Genes up-regulated in MCF10A cells (breast cancer) grown at low (mesenchymal phenotype) compared to those grown at high (epithelial, basal-like phenotype) confluency. Genes down-regulated in luminal-like breast cancer cell lines compared to the mesenchymal-like ones Genes down-regulated in basal-like breast cancer cell lines as compared to the mesenchymal-like ones. Genes down-regulated in luminal-like breast cancer cell lines as compared to the basal-like ones.

REACTOME database Pubmed: 14562044

0.050 b0.000

0.150 b0.000

Pubmed: 22208948 Pubmed: 16607286

b0.000 b0.000

b0.000 b0.000

Pubmed: 18281472

0.004

0.013

Pubmed: 16288205 Pubmed: 16288205 Pubmed: 16288205

b0.000 b0.000 b0.000

b0.000 b0.000 0.005

Genes involved in glucose and other sugar SLC transporters Genes annotated by the GO term GO: 0045121. Specialized membrane domains composed mainly of cholesterol and sphingolipids, and relatively poor in polyunsaturated lipids such as glycerophospholipids. The formation of these membrane domains is promoted by the presence of cholesterol in the lipid bilayer: the rigid hexagonal rings of cholesterol can pack tightly against the saturated hydrocarbon chains of some membrane lipids, allowing these lipids to assemble into cohesive units floating in the mass of loosely packed polyunsaturated plasma membrane components. Genes annotated by the GO term GO: 0055088. Any of the processes involved in the maintenance of an internal equilibrium of lipid within an organism or cell. Genes annotated by the GO term GO: 0005319. Enables the directed movement of lipids into, out of, within or between cells. Genes annotated by the GO term GO: 0005543. Interacting selectively with phospholipids, a class of lipids containing phosphoric acid as a mono- or diester. Genes involved in cholesterol biosynthesis Genes involved in transformation of lanosterol to cholesterol Genes involved in synthesis of bile acids and bile salts via 24-hydroxycholesterol

Database: REACTOME Database: GO

0.027 b0.000

0.189 0.029

Database: GO

0.010

0.066

Database: GO

0.012

0.115

Database: GO

0.010

0.124

b0.000 0.008 0.033

0.002 0.018 0.050

0.075

0.074

Database: REACTOME

0.080

0.086

Pubmed: 16607286

0.017

0.173

Pubmed: 14562044

0.058

0.112

Pubmed: 16288205 Pubmed: 16288205 Pubmed: 16288205

b0.000 b0.000 0.072

b0.000 b0.000 0.198

C

O

R

R

E

C

T

t3:17

t3:23

Brief description

t3:25

Reactome synthesis of bile acids and bile salts via 7-alphahydroxycholesterol Gotzmann epithelial to mesenchymal transition down

t3:26

Jechlinger epithelial to mesenchymal transition down

t3:27 t3:28 t3:29

Charafe breast cancer luminal vs mesenchymal up Charafe breast cancer basal vs mesenchymal up Charafe breast cancer luminal vs basal up

E

D

P

Database: REACTOME Database: REACTOME Database: REACTOME

R O

Top scoring trans-regulated expression quantitative trait loci (eQTL) influencing blood levels of high-density lipoprotein (HDL) cholesterol. Genes involved in synthesis of bile acids and bile salts via 7-alpha-hydroxycholesterol

Pubmed: 17873875

O

Genes down-regulated in MMH-RT cells (hepatocytes displaying an invasive, metastatic phenotype) during epithelial to mesenchymal transition (EMT). Genes down-regulated during epithelial to mesenchymal transition (EMT) induced by TGFB1 in the EpH4 cells (mammary epithelium cell line transformed by HRAS). Genes up-regulated in luminal-like breast cancer cell lines as compared to the mesenchymal-like ones. Genes up-regulated in basal-like breast cancer cell lines compared to the mesenchymal-like ones. Genes up-regulated in luminal-like breast cancer cell lines compared to the basal-like ones.

F

M. Marro et al. / Biochimica et Biophysica Acta xxx (2014) xxx–xxx

Please cite this article as: M. Marro, et al., Molecular monitoring of epithelial-to-mesenchymal transition in breast cancer cells by means of Raman spectroscopy, Biochim. Biophys. Acta (2014), http://dx.doi.org/10.1016/j.bbamcr.2014.04.012

t3:1 t3:2

M. Marro et al. / Biochimica et Biophysica Acta xxx (2014) xxx–xxx

4. Discussion

450 451 452 453 454 455 456 457 458

To our knowledge this study is the first time MCR has been used to 462 deconvolve and track the molecular content of cancer cells during the 463 EMT process, which involves biochemical changes linked with the 464 early metastatic phenotype of cancer cells. We have demonstrated 465 that the unsupervised computational MCR analysis resolves chemically 466 and physically meaningful component spectra from a set of mixed 467 Raman spectra, allowing the assignment of pure molecular spectra to 468 each component deconvolved and its contribution (concentration). 469 MCR is more powerful than the classical PCA technique, in which 470 the spectral signature in principal components can contain negative 471 regions, impeding the assignment of pure molecular spectra [30]. There472 fore, MCR is a good candidate for tracking the biochemical composition 473 of a biomedical sample during a transition involving molecular changes 474 such as in the early diagnosis of breast cancer. These results which are in 475 tune with findings of earlier studies carried out with transcriptomics 476 and molecular biological techniques, using normal and cancer cells, fur477 ther demonstrate the power of RS approaches in diagnostic applications 478 Q10 and in the knowledge of molecular processes of cancer cells. 479 During the last two decades MCR has been applied to a wide range of 480 chemical data [22–24] and has been adapted to Raman spectral data 481 from biological tissues for imaging their molecular content [25,38]. 482 High-contrast images of uterine tissue derived using Raman spectra 483 with an empty modelling approach of Multivariate Curve Resolution484 Alternating Least Squares [42] have been obtained. We have shown 485 the ability of MCR to track the biochemical composition of a biological 486 sample during a process involving molecular changes. These molecular 487 changes are sometimes masked by background signals in the Raman 488 spectra of cells. This is because cells need to be grown on a substrate 489 (normally glass) that contains Raman and fluorescence bands, and 490 when acquiring the cellular spectra, the focal point needs to be very 491 close to the window, which in some cases covers part of the confocal 492 volume. The ability of MCR to deconvolve pure component spectra 493 and the flexibility of MCR-ALS algorithm to include specific constrains 494 allow these undesired components to be removed and only the inherent 495 changes occurring in cellular RS to be analysed. Our results demonstrate 496 an improvement in the molecular characterization of cells when using 497 RS coupled with MCR-ALS. 498 MCR performed on the RS of confluent and sparse MCF10A cells 499 identified the metabolic profile of EMT, which is characterized by a de500 crease in the cytokeratin content and an increase in the fatty acid and 501 tryptophan content. These results are consistent with the gene profile 502 attributed to altered lipid metabolism functions like fatty acid metabo503 lism and fatty acid oxidation, which increase in malignant cells [51]. 504 Moreover, the PLS-DA analysis of the cellular Raman spectra classified 505 EMT cells in the fingerprint region with 100% sensitivity and 100% 506 specificity (see Fig. 4), indicating that Raman spectra provide powerful 507 information allowing stratification of cells by their metastasis initiation 508 process phenotype.

F

460

448 449

N C O

R

R

E

C

T

E

D

P

R O

461

U

MCF10A cells in confluent conditions had the highest content of cytokeratin according to the composition of their pre-EMT cytoskeleton. In contrast, MCF10A cells that undergo EMT and MDA-MB-435 cells contain abundant vimentin cytoskeleton [15]. Overall, there was a clear difference between confluent and sparse MCF10A cells, reflecting the increased EMT of the latter. These cells can thus be stratified using the lipid-accumulating phenotype profile, which has been found in the most aggressive cancer cells and is related to their increased migration in vitro [52]. Studies comparing the Raman spectrum of human breast cancerous and non-cancerous tissues also showed that fatty acid content and lipid profile were significantly different [53,54]. Moreover, the increased tryptophan concentration in sparse MCF10A compared to confluent and MDA-MB-435 cells might be secondary to the catabolic activation of L-tryptophan by indoleamine 2,3-dioxygenase (IDO1). The consumption of this amino acid by cancer cells has been related to immune system evasion [55]. Furthermore, the role of IDO1 in tumour cells has been associated with cell cycle regulatory genes, increased proliferation and inhibition of apoptosis [56]. IDO inhibition with siRNA leads to diminished breast cancer cell proliferation [57]. The activation of genes associated with the tryptophan metabolism has been reported in other types of carcinoma such as the cervical carcinoma, indicating a common metabolic change linked to malignancy [58,59]. Tryptophan metabolism could exert immunological and non-immunological effects, in turn improving cancer progression. Further studies in breast cancer cells are needed to understand the specific role of this metabolite in the EMT process. The EMT is one of the main mechanisms involved in breast cancer metastasis and most likely contributes to metastasis in all types of carcinoma [49,60,61]. Data obtained by the evaluation of RS using MCR were sensitive enough to detect differences in the breast cancer basal phenotype, since all analysed cells have a basal phenotype and share some similarities. The MDA-MB-435 cell population is basically fed by components 2 (fatty acid) and 4 (polysaccharides). Indeed, MDA-MB435 cells can be split into two clear groups, one with a low lipid content (fatty acid) and higher levels of polysaccharides and the other with a higher lipid content, indicating the important role of lipids as a fuel source for metastatic cells under nutritional deprivation, which in turn could influence metastasis organ-specificity [46,62]. Cancer is a very heterogeneous disease with a large number of genetic alterations. The molecular characterization of cancer cells provides interesting insights into breast cancer taxonomy but its implementation in clinical diagnosis is difficult because it is very expensive [63]. RS together with multivariate analysis has shown a high ability to differentiate cells with different degrees of malignancy according to their biochemical composition. This was validated by gene expression analyses, which confirmed that these differences in cell composition were closely associated with alterations in lipid metabolism that are associated with malignancy. Compared with other decomposition techniques used in RS, such as Principal Component Analysis, MCR analysis gives more chemically and physically understandable results. There is increasing evidence of the association between phospholipids and cancer. The analysis of MCF10A cells showed initial changes in the lipid content of cells associated with their EMT phenotype. In carcinoma cells EMT is related to higher aggressiveness and invasive and metastatic potential [64]. Activation of lipid metabolism is an early event in carcinogenesis and also a hallmark of many cancers [5]. In our attempt to study the metabolic changes involved in the EMT process we analysed sparse MCF10A cells, which show many features of post-EMT cells in in vitro cell culture, such as increased motility [12]. Our results clearly indicate that an increase in phospholipids is associated with the EMT process in breast cells. Characterization of this initial malignant process, which is closely related to metastasis progression, might be useful for diagnostic purposes. The essential hallmarks of cancer are intertwined with altered cancer cell-intrinsic metabolism, either as a consequence or as a cause [2]. New approaches based on a panel of small molecules derived from the analysis of

O

459

Interestingly, upregulated genes from the dataset “Sarrio epithelial mesenchymal transition up” were significantly associated with the MIII phenotype, suggesting a parallel between the MIII vs. MII comparison and the sparse MCF10A vs. confluent MCF10A comparison (FDR = 0.013). Along with this gene set, other sets including genes upregulated in EMT were found to be associated with MIII cells, whereas datasets including downregulated genes in EMT were associated with MII cells. As expected, MIII cells showed a more mesenchymal phenotype, with significant gene sets being those genes described by Charafe et al. [34] as downregulated in basal and luminal breast cancer sub-types. According to these transcriptomic public data, we inferred that fatty acid and glucose metabolic functions differentiate confluent and sparse MCF10A cells, then supporting the RS profile.

446 447

9

Please cite this article as: M. Marro, et al., Molecular monitoring of epithelial-to-mesenchymal transition in breast cancer cells by means of Raman spectroscopy, Biochim. Biophys. Acta (2014), http://dx.doi.org/10.1016/j.bbamcr.2014.04.012

509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574

581 582 583

[18]

[19]

[20]

[21]

586 587

We would like to thank Vanessa Hernández for her expert technical assistance. We are grateful to Dr. J. Ferrier for language advice. This 588 study was supported by a grant from the Spanish Ministry of Health 589 and Consumer Affairs FIS-PI10/00057 and by grants from MIIN 590 FIS2008-00114 and 2009-SGR-159 from the Generalitat de Catalunya, 591 Fundació Privada Cellex Barcelona and AECC (Spanish Association 592 Q11 Against Cancer) Scientific Foundation, respectively. 593

References

594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648

[1] V. Chi, G.L. Semenza, Oncogenic alterations of metabolism, TIBS 24 (1999) 68–72. [2] R.G. Jones, C.B. Thompson, Tumor suppressors and cell metabolism: a recipe for cancer growth, Genes Dev. 23 (2009) 537–548. [3] J.A. Menendez, R. Lupu, Fatty acid synthase and the lipogenic phenotype in cancer pathogenesis, Nat. Rev. Cancer 7 (2007) 763–777. [4] V. Chajes, M. Lanson, F.L. Fetissof, C. Huillery, P. Bougnoux, Membrane fatty acids of breast carcinoma: contribution of host fatty acids and tumor properties, Int. J. Cancer 63 (1995) 169–175. [5] M. Hilvo, C. Denkert, L. Lehtinen, B. Müller, S. Brockmöller, T. Seppänen-Laakso, J. Budczies, E. Bucher, L. Yetukuri, S. Castillo, E. Berg, H. Nygren, M. Sysi-Aho, J.L. Griffin, O. Fiehn, S. Loibl, C. Richter-Ehrenstein, C. Radke, T. Hyötyläinen, O. Kallioniemi, K. Iljin, M. Oresic, Novel therasnostic opportunities offered by characterization of altered membrane lipid metabolism in breast cancer progression, Cancer Res. 71 (2011) 3236–3245. [6] G. Kroemer, J. Pouyssegur, Tumor cell metabolism: cancer's Achilles' heel, Cancer Cell 13 (2008) 472–482. [7] W. Luo, R. Chang, J. Zhong, A. Pandey, G.L. Semenza, Histone demethylase JMJD2C is a coactivator for hypoxia-inducible factor 1 that is required for breast cancer progression, Proc. Natl. Acad. Sci. U. S. A. 109 (2012) E3367–E3376. [8] T. Sørlie, C.M. Perou, R. Tibshirani, T. Aas, S. Geisler, H. Johnsen, T. Hastie, M.B. Eisen, M. van de Rijn, S.S. Jeffrey, T. Thorsen, H. Quist, J.C. Matese, P.O. Brown, D. Botstein, P. E. Lønning, A.L. Børresen-Dale, Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications, Proc. Natl. Acad. Sci. U. S. A. 11 (2001) 10869–10874. [9] E. Rysman, K. Brusselmans, K. Scheys, L. Timmermans, R. Derua, S. Munck, P.P. Van Veldhoven, D. Waltregny, V.W. Daniëls, J. Machiels, F. Vanderhoydonc, K. Smans, E. Waelkens, G. Verhoeven, J.V. Swinnen, De novo lipogenesis protects cancer cells from free radicals and chemotherapeutics by promoting membrane lipid saturation, Cancer Res. 70 (2010) 8117–8126. [10] T. Sørlie, R. Tibshirani, J. Parker, T. Hastie, J.S. Marron, A. Nobel, S. Deng, H. Johnsen, R. Pesich, S. Geisler, J. Demeter, C.M. Perou, P.E. Lønning, P.O. Brown, A.L. BørresenDale, D. Botstein, Repeated observation of breast tumor subtypes in independent gene expression data sets, Proc. Natl. Acad. Sci. U. S. A. 100 (2003) 8418–8423. [11] F.M. Davis, P.A. Kenny, E.T. Soo, B.J. van Denderen, E.W. Thompson, P.J. Cabot, M.O. Parat, S.J. Roberts-Thomson, G.R. Monteith, Remodeling of purinergic receptormediated Ca2+ signaling as a consequence of EGF-induced epithelial–mesenchymal transition in breast cancer cells, PLoS One 6 (2011) e23464, http://dx.doi.org/10. 1371/journal.pone.0023464. [12] D. Sarrió, S.M. Rodriguez-Pinilla, D. Hardisson, A. Cano, G. Moreno-Bueno, J. Palacios, Epithelial–mesenchymal transition in breast cancer relates to the basal-like phenotype, Cancer Res. 68 (2008) 989–997. [13] C. Scheel, R.A. Weinberg, Cancer stem cells and epithelial–mesenchymal transition: concepts and molecular links, Semin. Cancer Biol. 22 (2012) 396–403. [14] T. Celià-Terrassa, O. Meca-Cortés, F. Mateo, A.M. de Paz, N. Rubio, A. Arnal-Estapé, B. J. Ell, R. Bermudo, A. Díaz, M. Guerra-Rebollo, J.J. Lozano, C. Estarás, C. Ulloa, D. Álvarez-Simón, J. Milà, R. Vilella, R. Paciucci, M. Martínez-Balbás, A.G. de Herreros, R. R. Gomis, Y. Kang, J. Blanco, P.L. Fernández, T.M. Thomson, Epithelial–mesenchymal transition can suppress major attributes of human epithelial tumor-initiating cells, J. Clin. Invest. 122 (2012) 1849–1868. [15] C. Nieva, M. Marro, N. Santana-Codina, S. Rao, D. Petrov, A. Sierra, The lipid phenotype of breast cancer cells characterized by Raman microspectroscopy: towards a stratification of malignancy, PLoS One 7 (2012) e46456, http://dx.doi.org/10.1371/ journal.pone.0046456. [16] C.J. Creighton, S. Massarweh, S. Huang, A. Tsimelzon, S.G. Hilsenbeck, C.K. Osborne, J. Shou, L. Malorni, R. Schiff, Development of resistance to targeted therapies

[22]

[23]

[24]

[25]

[26]

[27] [28]

E

[29]

[30]

T

C

E

R

R

O

C

N

U

F

Acknowledgements

579 580

[17]

O

585

577 578

transforms the clinically associated molecular profile subtype of breast tumor xenografts, Cancer Res. 68 (2008) 7493–7501. A. Vazquez-Martin, R. Colomer, J. Brunet, R. Lupu, J.A. Menendez, Overexpression of fatty acid synthase gene activates HER1/HER2 tyrosine kinase receptors in human breast epithelial cells, Cell Prolif. 41 (2008) 59–85. M. Hedegaard, C. Matthaus, S. Hassing, C. Krafft, M. Diem, J. Popp, Spectral unmixing and clustering algorithms for assessment of single cells by Raman microscopic imaging, Theor. Chem. Acc. 130 (2011) 1249–1260. C. Krafft, B. Belay, N. Bergner, B. Romeike, R. Reichart, R. Kalff, J. Popp, Advances in optical biopsy-correlation of malignancy and cell density of primary brain tumors using Raman microspectroscopic imaging, Analyst 137 (2012) 5533–5537. M. Miljkovic, T. Chernenko, M. Romeo, B. Bird, C. Matthaus, M. Diem, Label-free imaging of human cells: algorithms for image reconstruction of Raman hyperspectral datasets, Analyst 135 (2010) 2002–2013. J. Jaumot, R. Gargallo, A. de Juan, R. Tauler, A graphical user-friendly interface for MCR-ALS: a new tool for multivariate curve resolution in MATLAB, Chemom. Intell. Lab. Syst. 76 (2005) 101–110. T. Azzouz, R. Tauler, Application of multivariate curve resolution alternating least squares (MCR-ALS) to the quantitative analysis of pharmaceutical and agricultural samples, Talanta 74 (2008) 1201–2010. R. Tauler, S. Lacorte, M. Guillamon, S. Cespedes, P. Viana, D. Barcelo, Chemometric modeling of main contamination sources in surface waters of Portugal, Environ. Toxicol. Chem. 23 (2004) 565–575. M. Garrido, F. Rius, M. Larrechi, Multivariate curve resolution-alternating least squares (MCR-ALS) applied to spectroscopic data from monitoring chemical reactions processes, Anal. Bioanal. Chem. 390 (2008) 2059–2066. S. Piqueras, L. Duponchel, R. Tauler, A. de Juan, Resolution and segmentation of hyperspectral biomedical images by Multivariate Curve Resolution-Alternating Least Squares, Anal. Chim. Acta. 705 (2011) 182–192. C. Huang, M. Ando, H. Hamaguchi, S. Shigeto, Disentangling dynamic changes of multiple cellular components during the yeast cell cycle by in vivo multivariate Raman imaging, Anal. Chem. 84 (2012) 5661–5668. J. Chan, D. Lieu, Label-free biochemical characterization of stem cells using vibrational spectroscopy, J. Biophotonics 2 (2009) 656-628. U. Neugebauer, U. Schmid, K. Baumann, W. Ziebuhr, S. Kozitskaya, V. Deckert, M. Schmitt, J. Popp, Towards a detailed understanding of bacterial metabolism— spectroscopic characterization of Staphylococcus epidermidis, ChemPhysChem 8 (2007) 124–137. A.T. Harris, M. Garg, X.B. Yang, S.E. Fisher, J. Kirkham, D.A. Smith, D.P. Martin-Hirsch, A.S. High, Raman spectroscopy and advanced mathematical modelling in the discrimination of human thyroid cell lines, Head Neck Oncol. 1 (2009) 38–49. F. Bonnier, H. Byrne, Understanding the molecular information contained in principal component analysis of vibrational spectra of biological systems, Analyst 137 (2012) 322–332. M. Marro, A. Taubes, A. Abernathy, S. Balint, B. Moreno, B. Sanchez-Dalmau, E. Martinez-Lapiscina, I. Amat-Roldan, D. Petrov, P. Villoslada, Dynamic molecular monitoring of retina inflammation by in vivo Raman spectroscopy coupled with multivariate analysis, J. Biophotonics (2013), http://dx.doi.org/10.1002/jbio. 201300101. A.F. Chambers, MDA-MB-435 and M14 cell lines: identical but not M14 melanoma? Cancer Res. 69 (2009) 5292–5293. R. Edgar, M. Domrachev, A.E. Lash, Gene Expression Omnibus: NCBI gene expression and hybridization array data repository, Nucleic Acids Res. 1 (2002) 207–210. J.E. Charafe, C. Ginestier, F. Monville, P. Finetti, J. Adélaïde, N. Cervera, S. Fekairi, L. Xerri, J. Jacquemier, D. Birnbaum, F. Bertucci, Gene expression profiling of breast cell lines identifies potential new basal markers, Oncogene 25 (2006) 2273–2284. J. Shi, M.G. Walker, Gene Set Enrichment Analysis (GSEA) for interpreting gene expression profiles, Curr. Bioinforma. (2007) (ISBN 3-900051-07-0 URL). R: Development Core Team, R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing, Vienna, Austria, 2004. (http://www.r-project.org ). N.P. Gunasinghe, A. Wells, E.W. Thompson, H.J. Hugo, Mesenchymal–epithelial transition (MET) as a mechanism for metastatic colonisation in breast cancer, Cancer Metastasis Rev. 31 (2012) 469–478, http://dx.doi.org/10.1007/s10555-012-9377-5. E. Tomaskovic-Crook, E.W. Thompson, J.P. Thiery, Epithelial to mesenchymal transition and breast cancer, Breast Cancer Res. 11 (2009) 213, http://dx.doi.org/10.1186/ bcr2416. S.J. Lee, H.J. Seol, H.W. Lee, W.Y. Kang, B.G. Kang, J. Jin, M.Y. Jo, Y. Jin, J.I. Lee, K.M. Joo, D.H. Nam, Gene silencing of c-Met leads to brain metastasis inhibitory effects, Clin. Exp. Metastasis 30 (2013) 845–854. R.B. Hazan, G.R. Phillips, R.F. Qiao, L. Norton, S.A. Aaronson, Exogenous expression of N-cadherin in breast cancer cells induces cell migration, invasion, and metastasis, J. Cell Biol. 148 (2000) 779–790. L. Hartsuiker, N.J. Zeijen, L.W. Terstappen, C. Otto, Analyst 135 (2010) 3220–3226. Z. Movasaghi, S. Rehman, I. Rehman, Raman spectroscopy of biological tissues, Appl. Spectrosc. Rev. 42 (2007) 493–541. C. Krafft, L. Neudert, T. Simat, R. Salzer, Near infrared Raman spectra of human brain lipids, Spectrochim. Acta A 61 (2005) 1529–1535. S. Devpura, J. Thakur, S. Sethi, V. Naik, R. Naik, Diagnosis of head and neck squamous cell carcinoma using Raman spectroscopy: tongue tissues, J. Raman Spectrosc. 43 (2012) 490–496. J. Gelder, K. Gussem, P. Vandenabeele, L. Moens, Reference database of Raman spectra of biological molecules, J. Raman Spectrosc. 38 (2007) 1133–1147. N. Santana, C. Nieva, M. Marro, S. Rao, D. Petrov, A. Sierra, Raman microspectroscopy is a tool to identify the metastatic ability of breast tumors, International Workshop on Biophotonics, 2011, http://dx.doi.org/10.1109/IWBP.2011.5954866.

R O

584

metabolic profiles of cells are being developed to characterize cancer cell-specific metabolism [65]. Many observations during the early period of cancer biology research identified metabolic alterations such as the Warburg effect [49]. In this scenario RS supported by MCR can provide a useful tool with which to assess the biomolecular composition of tumour cells and metastasis-initiating cells. The applications of spectroscopic methods in cancer detection open new possibilities in the early stage diagnosis of breast cancer. Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.bbamcr.2014.04.012.

P

575 576

M. Marro et al. / Biochimica et Biophysica Acta xxx (2014) xxx–xxx

D

10

[31]

[32] [33] [34]

[35] [36]

[37]

[38]

[39]

[40]

[41] [42] [43] [44]

[45] [46]

Please cite this article as: M. Marro, et al., Molecular monitoring of epithelial-to-mesenchymal transition in breast cancer cells by means of Raman spectroscopy, Biochim. Biophys. Acta (2014), http://dx.doi.org/10.1016/j.bbamcr.2014.04.012

649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 Q12 682 683 Q13 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 Q14 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734

M. Marro et al. / Biochimica et Biophysica Acta xxx (2014) xxx–xxx [47] A. Subramanian, P. Tamayo, V.K. Mootha, S. Mukherjee, B.L. Ebert, M.A. Gillette, A. Paulovich, S.L. Pomeroy, T.R. Golub, E.S. Lander, J.P. Mesirov, Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles, Proc. Natl. Acad. Sci. U. S. A. 102 (2005) 15545–15550. [48] M. Jechlinger, S. Grunert, I.H. Tamir, E. Janda, S. Lüdemann, T. Waerner, P. Seither, A. Weith, H. Beug, N. Kraut, Expression profiling of epithelial plasticity in tumor progression, Oncogene 22 (2003) 7155–7169. [49] W.H. Koppenol, P.L. Bounds, C.V. Dang, Otto Warburg's contributions to current concepts of cancer metabolism, Nat. Rev. Cancer 11 (2011) 325–337. [50] P. Papageorgis, A.W. Lambert, S. Ozturk, F. Gao, H. Pan, U. Manne, Y.O. Alekseyev, A. Thiagalingam, H.M. Abdolmaleky, M. Lenburg, S. Thiagalingam, Smad signaling is required to maintain epigenetic silencing during breast cancer progression, Cancer Res. 70 (2010) 968–978. [51] P. Chinnaiyan, E. Kensicki, G. Bloom, A. Prabhu, B. Sarcar, S. Kahali, S. Eschrich, X. Qu, P. Forsyth, R. Gillies, The metabolomic signature of malignant glioma reflects accelerated anabolic metabolism, Cancer Res. 72 (2012) 5878–5888. [52] C.J. Antalis, A. Uchida, K.K. Buhman, R.A. Siddiqui, Migration of MDA-MB-231 breast cancer cells depends on the availability of exogenous lipids and cholesterol esterification, Clin. Exp. Metastasis 28 (2011) 733–741. [53] H. Abramczyk, B. Brozek-Pluska, J. Surmacki, J. Jablonska-Gajewicz, R. Kordek, Raman optical biopsy of human breast cancer, Prog. Biophys. Mol. Biol. 108 (2012) 74–81. [54] H. Abramczyk, B. Brozek-Pluska, J. Surmacki, J. Jablonska, R. Kordek, Raman spectroscopy and imaging: applications in human breast cancer diagnosis, Analyst 137 (2012) 3773–3781. [55] M.T. Travers, I.F. Gow, M.C. Barber, J. Thomson, D.B. Shennan, Indoleamine 2,3dioxygenase activity and L-tryptophan transport in human breast cancer cells, Biochim. Biophys. Acta 1661 (2004) 106–112. [56] V. Levina, Y. Su, E. Gorelik, Immunological and nonimmunological effects of indoleamine 2,3-dioxygenase on breast tumor growth and spontaneous metastasis

[57]

[58]

[59] [60]

[61]

[64]

N C O

R

R

E

C

T

E

D

P

[65]

R O

[63]

O

F

[62]

formation, Clin. Dev. Immunol. 173029 (2012), http://dx.doi.org/10.1155/2012/ 173029. L.D. Kumar, A.R. Clarke, Gene manipulation through the use of small interfering RNA (siRNA): from in vitro to in vivo applications, Adv. Drug Deliv. Rev. 59 (2007) 87–100. L. Huang, M. Zheng, Q.M. Zhou, M.Y. Zhang, W.H. Jia, J.P. Yun, H.Y. Wang, Identification of a gene-expression signature for predicting lymph node metastasis in patients with early stage cervical carcinoma, Cancer 117 (2011) 3363–3373. M. Platten, W. Wick, B.J. Van den Eynde, Tryptophan catabolism in cancer: beyond IDO and tryptophan depletion, Cancer Res. 72 (2012) 5435–5440. C.J. Creighton, J.C. Chang, J.M. Rosen, Epithelial–mesenchymal transition (EMT) in tumor-initiating cells and its clinical implications in breast cancer, J. Mammary Gland Biol. Neoplasia 15 (2010) 253–260. J. Fuxe, T. Vincent, A. Garcia de Herreros, Transcriptional crosstalk between TGF-β and stem cell pathways in tumor cell invasion: role of EMT promoting Smad complexes, Cell Cycle 9 (2010) 2363–2674. V. Stresing, E. Baltziskueta, N. Rubio, J. Blanco, M.C. Arriba, J. Valls, M. Janier, P. Clézardin, R. Sanz-Pamplona, C. Nieva, M. Marro, D. Petrov, A. Sierra, Peroxiredoxin 2 specifically regulates the oxidative and metabolic-stress response of human breast cancer cells, Oncogene 32 (2012) 724–735. R. Simon, Challenges of microarray data and the evaluation of gene expression profile signatures, Cancer Invest. 26 (2008) 327–332. D.J. Drasin, T.P. Robin, H.L. Ford, Breast cancer epithelial-to-mesenchymal transition: examining the functional consequences of plasticity, Breast Cancer Res. 13 (2011) 226, http://dx.doi.org/10.1186/bcr3037. N.E. Simpson, V.P. Tryndyak, F.A. Beland, I.P. Pogribny, An in vitro investigation of metabolically sensitive biomarkers in breast cancer progression, Breast Cancer Res. Treat. 133 (2012) 959–968.

U

735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763

11

Please cite this article as: M. Marro, et al., Molecular monitoring of epithelial-to-mesenchymal transition in breast cancer cells by means of Raman spectroscopy, Biochim. Biophys. Acta (2014), http://dx.doi.org/10.1016/j.bbamcr.2014.04.012

764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792