This article was originally published online as an accepted preprint. The “Published Online” date corresponds to the preprint version. You can request a copy of the preprint by emailing the Biopolymers editorial office at firstname.lastname@example.org
Cancer is the one of the leading causes of the death. Surgery, radiation, and chemotherapy, alone or in combination, are the known modalities of treatment. Multidrug resistance (MDR), resistance to multiple and structurally uncorrelated drugs, is a serious hurdle in chemotherapy.1 Several transport proteins, termed as ABC (ATP Binding Cassette) transporters, are known to be involved in this phenomenon. A number of these ABC transporters are identified and the mechanism of action of some of these proteins is understood. A major physiological function of these proteins is to transport lipids, bile salts, toxic compounds, and peptides from inside to outside of the cell. P-glycoproteins (P-gp), multidrug-resistance-associated proteins (MRPs) are two such known ABC transporters.2–4 P-gp, a membrane-embedded glycoprotein, is coded by the MDR1 gene. Under resistance conditions, overexpression of this protein takes place and this is responsible for the active efflux of the anticancer drugs. MRPs are also membrane-embedded glycoproteins with additional transmembrane domains at the N-terminal end and they transport drugs in a glutathione conjugated form.2–4 These conditions eventually cause resistant cells to escape from cytotoxic effect of drugs and in turn the treatment. Identification of a resistant/sensitive phenotype in cancer cells from a patient can enhance the efficiency of the therapy and can thus lead to treatment plans tailored to the individual patient. Therefore, development of simpler, cost-effective, and less time-consuming methods for identification of a resistant/sensitive phenotype in cancer cell lines is of significant interest in cancer therapy.
Optical spectroscopy techniques such as fluorescence, Raman, and infrared, which are sensitive to biochemical composition of samples, have shown to discriminate normal and malignant tissues in oral, cervical, breast, and many other cancers.5–10 Raman and infrared vibrational spectroscopic techniques have also been exploited for investigating eukaryotic and prokaryotic cells.11–15 As can be seen from the literature, among these two methods, infrared, probably due to its higher sensitivity, has been most widely used. Several successful Fourier transform infrared (FTIR) studies for discrimination of Pap smear and other exfoliated cells have been reported.16–20 Studies on apoptosis and identification of MDR and sensitive phenotypes in a cell line have also been described,21–24 showing the potentials and applications of this approach.
Although FTIR is a widely used and sensitive technique, it suffers from lack of spatial resolution, which is on the order of few tens of microns, thus suitable for population measurement. Nevertheless, through the use of synchrotron sources or more recently of new IR imaging systems, the problem of spatial resolution could be partly circumvented and cells have been analyzed with a resolution of about 10 μ or less as claimed by certain reports. However, samples need still to be dried and FTIR spectroscopy is also prone to other sampling errors like uneven thickness.15 Raman microspectroscopy, on the other hand, is less sensitive but has confocal properties, is not perturbed by water, and offers higher spatial resolution, down to about the micron level, making it an adapted tool for single-cell analysis. In the present investigation, both FTIR and micro-Raman approaches have been compared for characterizing cell types and phenotypes. These techniques provide complementary information and, moreover, few studies on micro-Raman spectroscopy for cell typing have been reported despite the fact that it has more practical applications for in vivo and real-life conditions. Cell typing and MDR phenotype discrimination of sensitive and resistant HL60 promyelocytic leukemia and MCF7 breast cancer human cell lines were undertaken. The aim was to evaluate and compare both approaches for their feasible applications in cell type differentiation and characterization of resistance phenotype such as MDR. Principal components analysis (PCA) was used to analyze spectral data24; in the case of micro-Raman data, due to the higher spatial resolution, both point and mean spectra, representative of different cell populations, were considered.
MATERIALS AND METHODS
HL60 is a human promyelocytic leukemia sensitive cell line. Two multidrug-resistant derivative cell lines named HL60/DOX and HL60/DNR were obtained by exposure to 40 nM of doxorubicin (DOX) and 1 μM of daunorubicin (DNR), respectively.
Both the wild-type MCF7 (MCF7/WT) human breast cancer cell line and the multidrug-resistant subclone resistant to verapamil (MCF7/VP) were supplied by Dr. J. Robert (University of Bordeaux, France). MCF7/VP cells were isolated by stepwise selection in increasing concentrations of VP16.25, 26 Resistance was maintained by exposure to 1 μM of VP16 for 2–3 days every 3 weeks.
All cell lines were grown in RPMI-1640 medium (Life Technologies, Cergy Pontoise, France) and supplemented with 10% heated-inactivated fetal calf serum (Life Technologies) and 2 mML-glutamine (Sigma, St. Quentin Fallavier, France). Antibiotics, fungizone 3 mg/L (Bristol-Meyers Squibb, Paris, France), and gentamycin 0.1 mg/mL (Sigma, France) were added. Cells were maintained at 37°C in a humidified atmosphere containing 5% CO2.
Cell Preparation for Spectroscopy
Cells were seeded in 25-cm2 flasks with medium. After 48 h in exponential growth conditions, cells were washed twice with 0.9% NaCl, scraped, and centrifuged at 1000 rpm for 5 min. Cell pellets were resuspended in 3 mL 0.9% NaCl. The concentration of cells in the suspension was determined. Cells were then diluted in 0.9% NaCl to obtain a concentration of 1.106 cells/mL. For micro-Raman measurements, cell suspensions were centrifuged at 1000 rpm for 5 min to obtain pellets. For FTIR studies, cell suspensions were dried under mild vacuum on a zinc selenide sample wheel.
Micro-Raman spectra were recorded on a commercial Raman microspectrometer (LabRam, Horiba Jobin Yvon, France). The instrumentation employed in the present investigation is described elsewhere.27 The 632.8-nm irradiation of a He–Ne laser was used for excitation. The experimental conditions were as follows: laser power of 3 mW at a sample with a 100 × objective; grating 1000 g/mm, and 3 acquisitions of 90 s each. These parameters were kept constant for all measurements. Spectra were collected at several points of the cell pellet. Because the Raman microspectrometer collects data from a volume of approximately a few micron cube of the sample, spectra depend very much on the site on which the laser has been focused.28 In other words, spectra of even adjacent areas can be quite different. In order to get a good representative spectrum, 25 or more spectra from adjacent positions in a 50 × 50 μm2 selected site, were recorded at 10-μm intervals and a mean spectrum was taken as representative of the site. This was repeated in triplicate for each sample. Spectral acquisition and all preliminary spectral corrections such as baseline subtraction, normalization, and spectral analysis were carried out using the LabSpec 4.04 software (Jobin Yvon Horiba) Baseline of Raman spectra was corrected by fitting and subtracting a third-order polynomial function. Baseline-corrected Raman spectra were then vector normalized using entire spectral range of 600–1800 cm−1.
FTIR spectra of dried cell suspensions were recorded using an Equinox 55 spectrometer (Bruker, France), equipped with a KBr beam splitter and a DTGS detector. The system was used in association with a sample changer module (MICOR-ID). This module can host a zinc selenide sample carrier wheel capable of accommodating 15 samples at a time. The whole system was continuously purged with dry air (FTIR purge gas generator model 75-62, Whatman, France). The zero position of the wheel left blank is used to collect the reference spectrum before each measurement. Thin IR transparent films of the cells were prepared at every position of the sample carrier wheel after dehydration of 35 μL of cell solution for 45 min under moderate vacuum conditions.11 Transmission spectra were acquired in the 3800–800-cm−1 range using 64 scans at 4-cm−1 resolution. These experimental conditions were kept constant for all measurements. Under these conditions, spectral acquisition of all 15 spectra could be completed in less than 30 min. Spectra were corrected for baseline using a scattering correction procedure, and vector normalized using the OPUS NT software (Bruker, France). They were then imported into the LabSpec 2.0 software for PCA analysis. Mean spectra were calculated from triplicates of independent measurements.
PCA for Spectral Data Treatment
PCA is generally described as an ordination technique for describing the variation in a multivariate data set. It is used as an effective approach to visualize and to mine information from data tables with a large number of variables such as those contained in FTIR and Raman data. It is the workhorse in chemometrics and its objective is to provide the most compact representation of all the variation in a data table. The iterative NIPALS algorithm, implemented in the LabSpec 2.0 software, is one of the many methods that exist for finding the eigenvectors. In our analysis, six principal components were used to explain almost 98% of total variance in the spectra. These principal components were further analyzed for identification of discriminating principal components using discriminant analysis algorithms of STATISTICA (Statsoft, Tulsa, OK, USA).
Discriminating analysis is a method used to reveal those variables that make it possible to differentiate between two or several groups. This approach is based on multivariate variance analyses between two groups but the program builds a discriminating function, which is a linear combination of the multivariate tests for each pair of groups. Among the various modes of calculation suggested by the STASTICA software, the ascending incremental analysis was used. This algorithm builds a model while testing with variables and by including with each cycle that which discriminates best between the groups. In the present analysis, scores of all factors obtained by PCA were fed into the program to determine the most discriminant principal components useful to discriminate the different cell types/phenotypes.
RESULTS AND DISCUSSION
In the present study, the HL60 cell line used is from the circulatory system and MCF7 cell line is of tissue origin. Therefore, it is expected that spectral profile of HL60 will differ from that of MCF7. Figure 1A shows the mean normalized micro-Raman spectra of HL6O and MCF7 cell lines. Differences in the spectral profiles, which include intensity changes of the amide III peak, several bands in the 1000–1200-cm−1 region, and the band at 936 cm−1, can be observed. In addition to these differences in intensities, the presence of a new peak at 1554 cm−1 is also seen in the spectrum of HL60 cells. For the FTIR spectra (Figure 1B), the differences are weaker and are seen mainly in the 900–1800-cm−1 absorbing region including fatty acids, proteins, nucleic acids, and polysaccharides. These weak differences are obviously difficult to assess by visible inspection only.
Highly objective evaluation is one of the major advantages of spectroscopic methods over conventional diagnostic methods like histopathology or cytology. This can be achieved by processing spectral data with multivariate statistical methods that can highlight subtle differences. In the present study, PCA was employed for this purpose. Contribution of factors [principal components (PCs)], also referred to as scores, has been used for discrimination. A discriminant analysis procedure has been applied in order to choose which scores plots reflect the best discrimination. PCA of mean micro-Raman point spectra and mean FTIR spectra of HL60 and MCF sensitive cell lines are shown in Figure 2A and B, respectively. As the results show, a good differentiation of the two cell types can be achieved with both techniques, indicating the feasibility of cell typing by both vibrational spectroscopy methods with particular emphasis on the Raman results. As it has been mentioned earlier, several FTIR studies16–23 have been reported in cell type discrimination but there have been only few studies on the Raman spectroscopic approach.24 When the loadings plots were inspected (data not shown here), the first component could be assigned to a mean spectrum. For the loadings of higher factors, we found several regions—namely in the 1000–1200-cm−1 region, amide I region, δCH2 region, and the 700–800-cm−1 region—that would be responsible for the clear differentiation. For example, in the 1400–1700-cm−1 region, the loadings were in opposite directions for mean Raman spectra of HL60- and MCF7-sensitive cells. Thus, different classes of molecules are involved, including proteins and lipids. PCA is a mathematical device (eigenvalue–eigenvector problem) by which the features in the whole data set of thousands of points are resolved into a few significant eigenvectors, which can express the entire data set with their scores for each spectrum. The eigenvectors may have spectral significance only when the data set consists of a single class of molecules—say, for example, a set of spectra of various lipids. When the data set consists of several classes of molecules and different types of mixtures, the eigenvectors may not represent any particular spectral/class characteristics but only factors essential to regenerate the data. This is clear from the PCA process. In PCA, the mean of all spectra is calculated and each spectrum is subtracted from the mean, leaving the variations from the mean. These variations are subjected to eigenvalue–eigenvector solution, giving the significant eigenvectors that are required to express all the variations to the desired degree of fit. Therefore, it is noteworthy to point out that relating PCA separation to spectral features that vary between sample types is more obvious when dealing with single-class molecules. In our case, because cells are composed of different classes of macromolecules with varying compositions, such type of assignment is less obvious. Furthermore, in the case of MDR resistance, although it is known that a class of ABC transport proteins are involved in this phenomenon, multivariate statistical analysis will not be able to depict specifically these proteins among so many present in the cell. This is because only very subtle differences exist and therefore the resulting analysis reflects an overall information of the cell status.
As mentioned earlier, the Raman microprobe used in the present study collects data from a microvolume of several μm3. Due to this collection mode, spectra measured even at two adjacent points can be expected to exhibit differences. As a result, spectra collected from different points on the cell pellet can be very different and may be a drawback for cell type discrimination as the strong variability can influence the analysis. However, this does not appear to be the case under our experimental conditions because PCA of Raman point spectra, as shown in Figure 2C, provided a very good discrimination based on scores of PCs 1 and 2. Again, when loadings of these factors corresponding to Raman point spectra of HL60- and MCF7-sensitive cell lines were inspected, they were found to consist of several differences in the whole spectrum range, implicating different macromolecules like proteins and lipids. In fact, in our present study on cell pellets, a confocal hole of 800 μm was used in order to increase the Raman sensitivity. Because of this and the diffusing nature of cell pellets, the probing volume could be more than specified (1 × 1 × 2 μm3 for a confocal hole of 140 μm, using a silicium sample) by the instrument manufacturer. Based on the specifications provided by the manufacturer, the laser spot size with the 100 × objective was about 4–5 μm; the depth of penetration of the laser at the 800-μm confocal hole, for a nondiffusible sample, was about 6–10 μm. The size of a typical cell used here is in the order of 15 μm. As a result, in a cell pellet, the probing beam encounters a stack of cells covering the whole and partial volume of any cell. Therefore, scattering can be expected from different organelles like the nucleus, mitochondria, and other subcellular compartments of different cells. This condition results in a spectrum containing an average information that is representative of the given cell line and that can still be used for discrimination.
Vibrational spectroscopic methods are sensitive to biochemical composition of cells and can be expected to discriminate the different resistant phenotypes in a cell line. Keeping this in view, Raman and FTIR spectral profiles of the sensitive phenotype of the promyelocytic leukemia HL60S cell line and its multidrug-resistant phenotypes (HL60/DOX: resistant to doxorubicin; HL60/DNR: resistant to daunorubicin) were recorded. The anticancer drugs doxorubicin and daunorubicin have been shown to be substrates of both classes of transporters P-gp and MRP.3 In the DOX-resistant phenotype (HL60/DOX), MRP1 was shown to be the transporter while P-gp was the predominant transporter in the DNR-resistant phenotype (HL60/DNR).29, 30 Mean Raman spectra of HL60S (dark curve), HL60/DOX (gray curve), and HL60/DNR (darker gray curve) are displayed in Figure 3A. As can be seen from the plots, differences in spectral profiles are very subtle for Raman data. Typical FTIR spectra of these cell lines are also shown in Figure 3B and the main spectral variations seem to take place in the 900–1700-cm−1 region. Differences concern variable protein (amide I and II, 1500–1700 cm−1) and lipid contents, a change in relative intensity of the two bands in the 1400–1500 cm−1 region, and a change in spectral profile below 1400 cm−1. PCA was again used to analyze the cell phenotypes and the results are shown in Figure 4. Mean Raman and FTIR spectra produced a good discrimination of all three phenotypes using PCs 3 and 5 (Figure 4A) and PCs 2 and 3 (Figure 4B), respectively, with, however, a better clear-cut in the FTIR classification. For the Raman data, sensitive and resistant phenotypes of the HL60 cell line also formed three clusters, but with a small overlap between clusters of the two resistant phenotypes. Three clusters, corresponding to each phenotype, could be achieved with FTIR spectra. Also, within each cluster the distribution is more homogeneous, indicating highly reproducible FTIR data. From the analysis of the FTIR spectra, it can be seen that component 3 contributes highly to the differentiation of the sensitive and resistant cell lines. Also, in the cluster of resistant cells, two subgroups corresponding respectively to HL60DOX and HL60DNR can be identified.
Analysis of micro-Raman spectra was also carried out on randomly picked point spectra from several different experiments as described before (Figure 4C). In this case, contributions of PCs 4 and 5 produced two reasonably exclusive clusters corresponding to the sensitive and resistant phenotypes, but no clear discrimination of the two resistant phenotypes could be reached. In order to explore any possible discrimination of point spectra of the two resistant phenotypes, PCA of the spectra of the resistant phenotypes alone was carried out in a second step. In this analysis, discrimination of the two different phenotypes was achieved based on scores of PCs 2 and 4, as shown in Figure 4D. Therefore, for Raman data, a two-step classification procedure seems necessary—the first step to separate the sensitive from the resistant phenotype and the second step to delineate between the resistant ones. Thus, the results obtained indicate the feasibility of using both point and mean micro-Raman spectra for MDR phenotype discrimination in the HL60 cell line.
Breast cancer cell line MCF7 and its subclone resistant to verapamil (MCF7/VP) are the other cells studied in this investigation. P-gp is the mechanism implicated in MDR of resistant phenotype MCF7/VP.31 In this case, differences in Raman spectral profiles were found to be very minute, as shown in Figure 5A, where mean Raman spectra of the sensitive MCF7S cell line (dark curve) and of its resistant counterpart MCF7/VP (gray curve) are presented. Typical mean FTIR spectra of these cell lines are also shown in Figure 5B. The differences in FTIR spectral profiles noted previously are also observable in the case of the MCF7 cell line. Because mean spectra and point spectra gave reasonably good discrimination of cell type as well as phenotype, randomly picked Raman point spectra were used for comparing the MCF7 cells. PCA was carried out on the data in order to test if the resistance phenotype could be discriminated. Scores of PC 3, which were positive for the resistant cell type and negative for the sensitive type, were plotted against PC4 in order to test group separation (Figure 6A). The loadings of factor 3 exhibited several positive peaks such as in the amide I band, δCH2, 600–800-cm−1, and 1000–1200-cm−1 frequency ranges. Major negative peaks could be seen in the 800–1000-cm−1 region. In the case of factor 4 loadings, major positive bands could be seen at around 700 cm−1, in the 800–1200 cm−1 and δCH2 regions, and negative peaks in the amide I and 1200–1300-cm−1 regions. Many of these bands can be assigned to vibrational modes of biomolecules such as lipids, proteins, and amino acid side chains in proteins.32, 33 Again, it should be noted that PCA will reveal an overall change, including various classes of complex molecules. Although scores of PC3 seemed to produce a demarcation between the sensitive and resistant MCF7 cells, some overlapping of the Raman point spectra could be observed. This observation stems from the fact that the differences in the Raman spectra are very subtle and point spectra can introduce some irreproducibility. However, taken together, these results are quite promising and strengthen further, after improvement of the method, the feasibility of using micro-Raman spectra in MDR phenotype discrimination, specifically in a single-cell analysis point of view. In a similar manner as has been observed previously, PCA analysis of FTIR spectra again gave a much better discrimination. This is shown in Figure 6B, where the plot of scores of PCs 2 and 3 clearly reveals two distinct populations corresponding respectively to the sensitive and resistant phenotypes of MCF7 breast cancer cell lines. As could be seen from the FTIR data, the spectral differences were more apparent between the two cell lines, which could explain the better discrimination between them.
Our objective in this study was to explore the potentials of two complementary methods, micro-Raman and FTIR spectroscopy, for the discrimination of cell types and phenotypes. We demonstrate here that both Raman and FTIR methods are amenable to discriminate cell type and a resistance phenotype such as MDR. However, the FTIR approach seems to provide better discrimination than Raman due to a higher reproducibility in the data. Nevertheless, keeping in view the perspectives of the single-cell approach and measurements in a straightforward manner with minimum sample preparation, the discrimination achieved using the micro-Raman method appears quite promising with an improvement of both technique sensitivity and data analysis. Because both point spectra as well as mean spectra provided more or less the same and reproducible discrimination, the micro-Raman spectroscopy method for screening of cell smears or pellets for cell typing and resistance phenotyping (MDR or other resistance phenotypes) seems to be a feasible perspective.