Modeling and Quantifying Biochemical Changes in ... - ACS Publications

Anal. Chem. 2008, 80, 8406–8415

Modeling and Quantifying Biochemical Changes in C6 Tumor Gliomas by Fourier Transform Infrared Imaging Abdelilah Beljebbar,*,† Nadia Amharref,† Antoine Le ´ ve`ques,† Sylvain Dukic,† Lydie Venteo,‡ ‡ ‡ Laurence Schneider, Michel Pluot, and Michel Manfait† Unite´ Me´DIAN, Universite´ de Reims Champagne-Ardenne, UMR CNRS 6237-MEDYC, IFR 53, UFR de Pharmacie, 51 Rue Cognacq-Jay, 51096 Reims Cedex, France, and Laboratoire Central d’Anatomie et de Cytologie Pathologiques, CHU Robert Debre´, Avenue du Ge´ne´ral Koenig, 51092 Reims Cedex, France The purpose of the study was to investigate molecular changes associated with glioma tissues using FT-IR microspectroscopic imaging (FT-IRM). A multivariate statistical analysis allowed one to successfully discriminate between normal, tumoral, peri-tumoral, and necrotic tissue structures. Structural changes were mainly related to qualitative and quantitative changes in lipid content, proteins, and nucleic acids that can be used as spectroscopic markers for this pathology. We have developed a spectroscopic model of glioma to quantify these chemical changes. The model constructed includes individual FTIR spectra of normal and glioma brain constituents such as lipids, DNA, and proteins (measured on delipidized tissue). Modeling of FT-IR spectra yielded fit coefficients reflecting the chemical changes associated with a tumor. Our results demonstrate the ability of FT-IRM to assess the importance and distribution of each individual constituent and its variation in normal brain structures as well as in the different pathological states of glioma. We demonstrated that (i) cholesterol and phosphatidylethanolamine contributions are highest in corpus callosum and anterior commissure but decrease gradually towards the cortex surface as well as in the tumor, (ii) phosphatidylcholine contribution is highest in the cortex and decreases in the tumor, (iii) galactocerebroside is localized only in white, but not in gray matter, and decreases in the vital tumor region while the necrosis area shows a higher concentration of this cerebroside, (iv) DNA and oleic acid increase in the tumor as compared to gray matter. This approach could, in the future, contribute to enhance diagnostic accuracy, improve the grading, prognosis, and play a vital role in therapeutic strategy and monitoring. Glioblastoma is the most frequent brain tumor and accounts for approximately 12% to 15% of all brain tumors.1 Glioblastomas are among the most aggressive malignant human neoplasms, with * To whom correspondence should be addressed. Dr. Abdelilah Beljebbar Unite´ Me´DIAN, CNRS UMR 6237 IFR 53 UFR de Pharmacie, Universite´ de Reims Champagne-Ardenne 51, rue Cognacq-Jay, 51096 Reims Cedex, France. Phone: (33) 3 2691-3554. Fax: (33) 3 2691-3550. E-mail: [email protected]. † UMR CNRS 6237-MEDYC. ‡ CHU Robert Debre´.

8406

Analytical Chemistry, Vol. 80, No. 22, November 15, 2008

a mean total length of disease in patients with primary glioblastoma of less than 15 months.2 This tumor is characterized by a diffuse local invasion of the normal parenchyma which renders complete surgical resection of cancerous tissue impossible.3 This invasion process implies modifications of the adjacent extracellular matrix to allow cell migration.4 A better understanding of the factors favoring glioblastoma tumor invasion are critically important for developing novel, and more effective strategies to treat this cancer. Fourier transform infrared microspectroscopy (FT-IR) spectroscopy is a vibrational spectroscopic technique, with a high molecular specificity, that can be applied in medical diagnostics.5-8 An infrared spectrum of a biological sample is composed of characteristic absorption bands originating from all infrared-active vibrational modes of biomolecules present in the tissue, such as proteins, lipids, and nucleic acids.9 Biological tissues are the complex mixture of a large number of these biomolecules. As a disease leads to changes in the intrinsic molecular composition of affected tissues, these changes should be reflected in the spectra. Almost all recent studies on brain glioma, using infrared microspectroscopy, reveal that the principal difference between tumor and healthy brain frozen tissue was associated to the changes in the lipid composition especially neutral lipids, phospholipids.10-12 Indeed, the accumulation of some lipids such (1) Hill, C. I.; Nixon, C. S.; Ruehmeier, J. L.; Wolf, L. M. Phys. Ther. 2002, 82, 496–502. (2) Stupp, R.; Mason, W. P.; van den Bent, M. J.; Weller, M.; Fisher, B.; Taphoorn, M. J.; Belanger, K.; Brandes, A. A.; Marosi, C.; Bogdahn, U.; Curschmann, J.; Janzer, R. C.; Ludwin, S. K.; Gorlia, T.; Allgeier, A.; Lacombe, D.; Cairncross, J. G.; Eisenhauer, E.; Mirimanoff, R. O. N. Engl. J. Med. 2005, 10, 987–996. (3) Claes, A.; Idema, A. J.; Wesseling, P. 2007, 114, 443–458. (4) Liotta, A.; Kohn, E. C. Nature 2001, 411, 375–379. (5) Hammody, Z.; Argov, S.; Sahu, R. K.; Cagnano, E.; Moreh, R.; Mordechai, S. Analyst 2008, 133, 372–378. (6) Martin, F. L.; Fullwood, N. J. Proc. Natl. Acad. Sci. U.S.A. 2007, 18, 104– 151. (7) Petibois, C.; Drogat, B.; Bikfalvi, A.; De´le´ris, G.; Moenner, M. FEBS Lett. 2007, 27, 5469–5474. (8) Bhargava, R. T. Anal. Bioanal. Chem. 2007, 389, 1155–1169. (9) Parker, F. S. Application of Infrared Spectroscopy in Biochemistry, Biology and Medicine; Plenum: New York, 1971. (10) Steiner, G.; Shaw, A.; Choo-Smith, L. P.; Abuid, M. H.; Schackert, G.; Sobottka, S.; Steller, W.; Salzer, R.; Mantsch, H. H. Biopolymers 2003, 72, 464–471. (11) Krafft, C.; Thummler, K.; Sobottka, S. B.; Schackert, G.; Salzer, R. Biopolymers 2006, 82, 301–305. 10.1021/ac800990y CCC: $40.75  2008 American Chemical Society Published on Web 10/21/2008

as cholesterol ester is a marker for necrosis which often occurs in brain metastasis.13 The FT-IR technique probes the total constituents that are present in the tissue. However, it is difficult to separate between the spectral contribution of each individual tissue components (lipids, protein, DNA, and/or RNA) even with multivariate statistical analysis such as principal component analysis.14 In order to provide information about the chemical basis for diagnosis and to understand the relationship between a tissue IR spectrum and its disease state, previous FT-IR work has developed models to identify subtle structural changes in DNA at various stages of tumor development.15 This quantification is based on the assumptions that the IR spectrum measured on tissue is a linear combination of the spectra of its individual components. In the present work, a FT-IR statistical model was developed to extract relative absorbance percentage fractions of normal and glioma brain constituents (proteins, lipids, nucleic acid) from IR spectra to provide information about the chemical basis for diagnosis and treatment of the brain tumor. This model, consisting of spectra acquired from individual normal and glioma brain constituent such as lipids (extracted from white matter and gray matter separately), DNA, and protein (measured on delipidized tissue) was used as a linear combination to fit measured spectra on brain tissues. MATERIALS AND METHODS Sample Preparation. All animal procedures adhered to the “Principles of laboratory animal care” (NIH publication no. 85-23, revised 1985). Male Wistar rats weighing 273 ± 28 g (mean ± SD) were purchased from Harland (Paris, France). The glioma tumor were obtained by a C6 glioma cell suspension injection in brain parenchyma as described elsewhere.12 Rats were anesthetized with isoflurane, placed into a stereotactic head holder (Phymep, France), a small burr hole was drilled into the right side of the skull (anterior 1 mm; lateral 3 mm; lateral depth 4 mm, according to the bregma) and then, a tumor cell suspension (5 × 106 cells in 10 µL) was injected with a syringe over 2 min. Control and tumor bearing rats were sacrificed 20 days after tumor cell injection. After brain excision, tissue samples were snap-frozen by immersion in methyl-butane cooled down in liquid nitrogen and stored at -80 °C. Two adjacent sections were cut from each sample using a cryomicrotome. One section, 10 µm thick, was placed onto infrared transparent calcium fluoride (CaF2) slides for infrared imaging. The second section, 10 µm thick, was placed on a microscope glass slide and stained with hematoxylin and eosin (H&E) for histopathological examination. Infrared Reference Spectra. Reference spectra were obtained on the following specific pure chemical compounds purchased from Sigma-Aldrich (Saint Quentin Fallavier, France): cholesterol (99%), cholesteryl oleate (3β-hydroxy-5-cholestene 3-oleate, > 98%), (12) Amharref, N.; Beljebbar, A.; Dukic, S.; Venteo, L.; Schneider, L.; Pluot, M.; Vistelle, R.; Manfait, M. Biochim. Biophys. Acta 2006, 1758, 892–899. (13) Sijens, P. E.; Levendag, P. C.; Vecht, C. J.; van Dijk, P.; Oudkerk, M. NMR Biomed. 1996, 9, 65–71. (14) Manoharan, R.; Shafer, K.; Perelman, L.; Wu, J.; Chen, K.; Deinum, G.; Fitzmaurice, M.; Myles, J.; Crowe, J.; Dasari, R.; Feld, M. Photochem. Photobiol. 1996, 7, 15–22. (15) Malins, D. C.; Gilman, N. K.; Green, V. M.; Wheeler, T. M.; Barker, E. A.; Vinson, M. A.; Sayeeduddin, M.; Hellstro ¨m, K. E.; Anderson, K. M. Proc. Natl. Acad. Sci. U.S.A. 2004, 101, 11428–11431.

phosphatidylcholine (1,2-diacyl-sn-glycero-3-phosphocholine, 99%, from egg yolk), phosphatidylethanolamine (1,2-dihexadecanoylsn-glycero-3phosphoethanolamine, 99%), phosphatidylserin (1,2diacyl-sn-glycero-3-phospho-L-serine, 99%, from bovine brain), sphingomyelin (N-acyl-D-sphingosine-1-phosphocholine, 99%, from bovine brain), galactocerebroside (ceramide β-D-galactoside, 99%, from bovine brain), linoleic acid (cis-9,cis-12-octadecadienoic acid, > 99%), oleic acid (cis-9-octadecenoic acid, 99%), and DNA from calf thymus. Each lipid was dissolved in a mix of methanol (HPLC grade, CARLOERBA-SDS, Peypin, France) and chloroform (HPLC grade, CARLOERBA-SDS, Peypin, France) (v/v). DNA was dissolved in phosphate buffered saline at pH 7.4. These references components were placed onto a CaF2 slide and dried in air before infrared measurement. Extraction of Lipids from Brain Tissue. Tissue samples were obtained from healthy rat brain. The white matter (corpus callosum) and the gray matter (cortex) were separated and homogenized with 2 mL of a chloroform/methanol mixture (2:1, v/v) by sonication.16 After centrifugation at 2500g for 10 min (BR 3.11, Jouan, France), the chloroform phase was withdrawn. Then, 2 mL of a chloroform/methanol mix (2:1) were added to the supernatant and homogenized by sonication and centrifugation at 2500g for 10 min (BR 3.11, Jouan, France). The lower phase was then collected and added to those obtained after the first centrifugation. Chloroform (1 mL) and distilled water (1 mL) was added. The mixture was shaken to form an emulsion and briefly centrifuged at 2000g for 2 min (BR 3.11, Jouan, France). The lower phase was withdrawn, and the chloroform was evaporated to dryness at ambient temperature in a vacuum evaporator (RC 10.10, Jouan, France). Delipidation of Brain Tissues. After IR measurements on healthy and glioma tissues, the latter were passed in a chloroform/ methanol mix (1/2; v/v) to remove lipids and rinsed several times with distilled water.17 IR maps were recorded on the same areas as those obtained before delipidization with the same experimental conditions. These protein models obtained from white matter, gray matter, and tumor were included in the model set. To prevent protein structural alterations in tissue biochemistry after solvent treatment, we have compared protein models obtained before and after delipidation in terms of secondary structures of tissue proteins. FT-IRM Microspectrometer. Spectra were collected using an FT-IR imaging system (SPOTLIGHT, Perkin-Elmer, France) coupled to a FT-IR spectrometer (Spectrum 300, Perkin-Elmer, France). This system is equipped with a liquid N2 cooled mercurycadmium-telluride MCT line detector comprised of 16 pixel elements. The microscope was equipped with a movable, software controlled, x, y stage. In this study, FT-IR images were collected from selected sites with a spatial resolution of 25 µm/pixel, in transmission mode in the 4000-720 cm-1 ranges, with a final spectral resolution of 4 cm-1. Data acquisition was carried out by means of the Spotlight software package supplied by Perkin-Elmer. Data Analysis. To allow meaningful comparisons, all FT-IRM data were uniformly preprocessed. After atmospheric correction, data were cut to the fingerprint region (950-1800) and converted (16) Dasgupta, S.; Glushka, J.; van Halbeek, H.; Hogan, E. L. Arch. Biochem. Biophys. 1994, 310, 373–384. (17) Folch, J.; Ascoli, I.; Lees, M.; Meath, Ja.; Lebaron, N. J. Biol. Chem. 1951, 191, 833–841.


8407

to their first derivative and smoothed using a seven point Savitzky-Golay algorithm. The resulting spectra were then normalized using a standard normal variate (SNV) procedure.18 All data measured on several healthy and glioblastoma tissues were pooled in one data set, processed at the same time to extract all features describing both normal and tumor tissues, and displayed as pseudocolor maps with the same color scale. In this way, we can easily determine all their common and discriminating features by comparing their infrared maps. A multivariate statistical analysis (principal component analysis (PCA) and K-Means (KM)) was performed on this data set. PCA was performed on the data set to remove redundancies at different locations in the spectra by finding the independent sources of variation in all spectra and to reduce the number of variables describing the data set. K-means clustering was performed on these principal component scores. Pseudocolor maps based on cluster analysis were then constructed by assigning a color to each spectral cluster. The cluster models were calculated by averaging the clusters and used for the interpretation of the chemical or biochemical differences between clusters. Cluster averaged spectra were obtained by the mean of the absorbance spectra associated with each group. Investigation of Linear Independence of the Model Set. Before applying multiple least square (MLS), two tests were performed to verify if the model set (based on reference spectra) is an independent set of spectra. First, each individual spectrum of the model set was fitted with all other spectra of the model set. For spectra that formed an independent set, a very poor fit was obtained, judged by the large infrared features in the fit residuals (obtained by subtracting the fit result from the measured spectrum). On the other hand, spectra that give a very good fit (form a dependent set) cannot be accounted for in the databases. In order to minimize error due to nonorthogonality of the reference spectra, spectra that could be fitted for more than 95% by the other reference spectra were excluded from the reference. The relative uncertainty of the fit itself was expressed as the part of the spectrum that could not be fitted by the reference spectra, i.e., as the quotient of the absolute spectral content of the residual and the spectral content of the spectrum that is fitted (which is 1 due to normalization). Only components with a high orthogonal relationship were selected for fitting. One supplementary test was added for each of the components in the model set and was fitted with the complete model set. It was verified that only the spectrum of that component, as included in the model set, contributed to the fit around 100%. The results of these tests show that the model set provides indeed a sufficiently independent set of spectra for the purpose of this study. The result of this test will help to choose an independent set which is a prerequisite before applying the multiple least-squares fitting procedure. To validate the model set, a test was then performed on the database in order to evaluate the efficiency of the fit procedure to accurately determine the relative amount of the individual constituent. Two lipid mixtures were then realized, blindly, by using different reference lipids. The measured spectra were then fitted with all reference spectra to identify and quantify each individual compound present in the mixtures. (18) Barnes, R. J.; Dhanoa, M. S.; Lister, S. J. Appl. Spectrosc. 1989, 43, 772– 777.

8408


Multiple Least-Squares Fitting Procedure. An infrared spectrum measured on tissue is a linear superposition of spectra of individual chemical compounds (like proteins, lipids, and nucleic acids). A multiple least square (MLS) is applied to fit the spectrum measured on the tissue sample under investigation with a number of reference spectra (pure lipids, proteins, and nucleic acids). In our study, before application of the MLS method, the spectra measured on tissues were converted to their first derivatives. Each spectrum was then fitted with a set of reference spectra, offset (second order polynomial) and slope to minimize the Mie scattering effect. These fit coefficients were normalized to one, and maps representing the molecular distribution and/or contribution of each individual brain constituent from normal and tumor tissues were built. The colorbar represents the molecular absorbance percentage of each constituent. This method will allow a better understanding of the changes between normal and glioma tissues. The accuracy of the model was established by judging the residuals obtained by subtracting the measured spectra from the theoretical one. Estimation of Errors in the Fit Results. An independent set of fit spectra will yield unique fit results. The objective of any biochemical fitting is to minimize the sum of residuals. If the above sum of residuals was to be minimized, the residual plot would tend to flatten toward the midline. To determine the error in our fit coefficient, we used chi-squared analysis. Chi-squared analysis is a well-known method for estimating the goodness of a fit as well as the error associated with model fitting. No clear remaining spectral features were visible in the residuals, thus indicating that virtually all of the changes in molecular composition that were reflected in the FT-IR spectra have been accounted for. Indeed, the successful application of the MLS technique requires the careful elimination of water vapor and random noise from the spectra as these may affect the fit results. An estimate was made of the influence of noise on the quantitative fit results. For this purpose, noise was added to the FT-IR spectra to artificially decrease the signal-to-noise ratio by a factor of 3 and a new set of fit coefficients was calculated. For each of the spectrum, this procedure was repeated 100 times. The standard deviation thus obtained for each of the fit coefficients was used as an estimate for the error in the fit coefficients. The processing analysis on FT-IRM data was performed with Matlab (version 7.02, MathWorks, Inc., Matick, MA). RESULTS AND DISCUSSION The objective of this study was to determine tissue modifications between tumor and healthy brain tissues and to quantify biochemical changes in relation with lipid, proteins, and nucleic acids between these tissues. These changes are then used as spectroscopic markers associated with the tumor. Indeed, FT-IRM, with high spatially resolved morphological and biochemical information could be used as a diagnostic tool, complementary to histopathology in order to understand the molecular changes associated with tissue alteration. As the FT-IR image contrast is based on the vibrational signature of the tissue intrinsic components, FT-IRM imaging does not require the use of dyes, tags, or stains. On the other hand, histopathological diagnosis depends on visualization of the sample morphology by staining techniques. In spectral diagnosis, objective intrinsic structural and morpho-

Figure 1. Photomicrographs (H&E staining) of healthy (A) and glioma (D) brain tissue sections. Pseudocolor FT-IR maps B and E, based on 14-means cluster analysis, were measured on healthy (map A) and tumor (map D), respectively. After IR analysis, the same tissues were passed in a chloroform/methanol mix (1/2; v/v) to remove lipids and rinsed several times with distilled water. New IR maps were then recorded on the same areas as those mapped before delipidization using the same experimental conditions and are represented in parts C and F.

logical information, on measured area, were used to better understand the tissue transformation. Discrimination between Healthy and Tumor Tissues by Clustering Analysis. Multivariate statistical analysis was performed on a data set containing all FT-IRM measurements obtained from healthy and tumor brain tissues. Pseudocolor maps (Figure 1B,E) obtained were compared to histopathology examination (Figure 1A,D) in order to correlate each pseudocolor feature with its anatomical counterpart. Twelve clusters describing both healthy and cancer features were extracted and pseudo FTIR maps were constructed with the same color scale. White color represents the area where no tissue was present. In the pseudocolor map obtained from healthy brain tissue (Figure 1B), six clusters were sufficient to describe all normal brain features. FTIRM images allow clear identification of the anatomical structures of the rat brain (Figure 1A,B). This pseudocolor map displayed some brain structures such as white matter (corpus callosum (CC) and commissura anterior (CA)) associated with cluster 5, four layers associated with gray matter cortex (clusters 1, 2, 3, and 6), and then the caudate putamen (clusters 3, 4, and 6). Comparison between parts A and B of Figure 1 shows that the FT-IRM image provides more information on the cortex than standard histopathology. Four layers were identified from the cortex, whereas, H&E staining did not allow one to discriminate between the layers in the cortex. In fact, the cerebral cortex is partitioned into six

layers numbered I-VI.19 The neurons of these layers may be categorized into two major cell types: pyramidal cells and stellate cells. This result was confirmed by our previous study where Luxol Fast Blue staining was combined with cresyl violet coloration to visualize all cortex layers.12 This study showed that FTIRM pseudocolor maps were clearly similar to LFB-CV staining for visualizing myelin distribution in healthy brain tissues. Figure 1E displays an infrared map obtained from tumor brain tissue. This pseudocolor map shows some common structures with normal tissue such as CC (model 5). Other structures were restricted to tumor tissue (models 7, 9, 10, 11, 12, and 14). Model 9 was located in the tumor and seems to constitute the viable tumor tissue. Two clusters were associated with the necrotic part of the tumor (clusters 7 and 12). Cluster 7 observed in the border of the necrotic zone seems to correlate with the pseudopalisading formation observed in the perinecrotic area. Cluster 12 correlates with the center of the necrosis (full necrosis) of the tumor. Glioblastoma is the most malignant brain tumor and aggressively proliferates and invades surrounding normal brain tissues. The tumor peripheric zone is described by two clusters (clusters 14 and 11), and more precisely, cluster 11 appears as the tumor invasive structure. In fact, brain parenchyma in direct vicinity of tumor tissue appeared modified, and these modifications could (19) Martin, J. H. Anatomical substrates for somatic sensation. Principle of Neural Science, 2nd ed.; Kandel, E. R., Schwartz, J. H., Eds.; Elsevier: New York, 1985; Vol. 30, pp 1-315.


8409

Figure 2. Representative cluster mean FT-IR spectra extracted from pseudocolor maps: (A) 14 models describing healthy and glioma brain tissue sections before delipidation and (B) 8 models associated to all tissues features after delipidation of these tissues.

be associated with the brain edema. Indeed, brain tumors contain tumor vessels that may have different structural properties favoring the formation of edema within and around the tumor. Ide et al. found a correlation between peritumoral brain edema and cortical destruction by the tumor.20 This structure is not visible with histopathological H&E staining. In our previous study, immunohistochemical Ki-67 and MT1-MMP staining were used to visualize the proliferative and invasive activities of glioma, respectively,21 and were clearly correlated with clusters that encoded the surrounding tumor area. This correlation between the immunohistochemistry staining and spectroscopic data showed the potential of vibrational spectroscopy to provide spectroscopic markers associated with the proliferative and invasive properties of glioblastoma. Cluster mean infrared spectra are shown in Figure 2. The attribution of all infrared bands has been reported elsewhere.12 The comparison between these spectra shows that in the malignant brain tumor, (i) the 1466/1396 cm-1 ratio (amino acid side chain from peptides and proteins at 1466 and 1396 cm-1, associated with the CH2 scissoring and C-H bending vibrations) decreased, (ii) the band at 1740 cm-1 (stretching mode of CdO groups of lipids) decreased and even disappeared when compared to the corresponding bands in healthy tissues. This result confirms (20) Ide, M.; Jimbo, M.; Kubo, O.; Yamamoto, M.; Takeyama, E.; Imanaga, H. Acta Neurochir. Suppl. 1994, 60, 369–372. (21) Amharref, N.; Beljebbar, A.; Dukic, S.; Venteo, L.; Schneider, L.; Pluot, M.; Manfait, M. Biochim. Biophys. Acta 2007, 1768, 2605–2615.

8410


that the development of tumor was characterized by a reduction in total lipid content. This reduction was also observed in the invasive area (model 11), which is composed of healthy and tumoral cells. This result is in agreement with those obtained in brain diseases.11,22 This lipid reduction in malignant tissues could be related to the fast growth of tumor cells which need more energy.23 Indeed, it is known that, in developing brain tumors, structural and functional cell changes take place in which lipids play a crucial role. Yet, qualitative and quantitative aspects of lipid changes in brain tumors of different degrees of malignancy are still the subject of numerous studies.10,11,24 Discrimination between Delipidized Healthy and Tumor Tissues by Clustering Analysis. Quantitative and qualitative use of lipid as a spectroscopic marker to discriminate between normal and tumor tissue as well as between low and high grade, and hence prognosis, were very essential for an early and accurate diagnostic of tumor and development of novel and more efficient therapies. To evaluate the spectroscopic changes between normal and tumoral brain tissues in term of proteins and nucleic acid, the same tissues measured above were delipidized and the same regions were mapped again by FT-IRM. The same statistical analysis was performed on these data, and pseudocolor maps on (22) Kneipp, J.; Lasch, P.; Baldauf, E.; Beekes, M.; Naumann, D. Biochim. Biophys. Acta 2000, 1501, 189–199. (23) Wang, J. S.; Shi, J. S.; Xu, Y. Z.; Duan, X. Y.; Zhang, L.; Weng, S. F.; Wu, J. G. World J. Gastroenterol. 2003, 9, 1897–1899. (24) Campanella, R. J. Neurosurg. Sci. 1992, 36, 11–25.

delipidized normal and tumor tissues were constructed (parts C and F of Figure 1). On normal tissues, only two clusters were sufficient to describe all normal features. The cluster 1 encoded the white matter (CC and CA), and cluster 2 described the gray matter (cortex). The comparison between the normal tissue before and after delipidation (parts B and C of Figure 1) confirmed that the four cortex layers identified in Figure 1B were associated with the changes in the lipid content because these layers disappeared after delipidation. On the other hand, by removing lipids we were still able to discriminate between tumor (cluster 6), the peritumoral region (cluster 3), and necrosis (cluster 5). This result demonstrated that proteins and nucleic acid can be used as spectroscopic biomarkers to differentiate between normal brain structures, glioma, and necrosis Cluster mean infrared spectra are compared in Figure 2B. The comparison between cluster mean spectra of normal tissue before and after delipidation shows the disappearance of the band associated with lipids such as the band at (i) 1740 cm-1 (arises from the stretching mode of CdO groups of lipids), (ii) 1466 and 1382 cm-1 are due to CH2 and CH3 bending of cholesterol and phospholipids, (iii) a disappearance of the band at 1068 cm-1 due to C-C stretching, and (iv) a decrease of the lipid contribution in the bands at 1082 and 1236 cm-1, these two remaining bands were associated with symmetric and asymmetric stretching vibration of the PO2- band. Almost all previous work reported on frozen tissues demonstrated that lipid content in tissues can be used as spectroscopic markers to discriminate between healthy, intermediate stages, and tumor tissues. To perform retrospective studies requires the use of paraffin-embedded samples available in tumor tissue banks. These tissues were delipidated with chemical solvent such as xylene and alcohol. In this case we need to find other spectroscopic markers than lipid to discriminate between normal and pathologic tissues. Previous work on chemical deparaffinized brain tissues has demonstrated that glioma tissues have a characteristic chemical signature, different from normal tissues based on protein and nucleic acids.25 Other work has used the delipidized tissues to find protein models that were then included in the reference data set with different lipids and nucleic acids present in the tissues to quantify biochemical changes associated with the development of pathologies (breast cancer and atherosclerosis).26,27 Characterization of Molecular Distribution of Brain Constituents (Lipids, Proteins, And Nucleic Acids) between Normal and Tumor Tissues. Test of Independency of the Reference Database. FT-IR is an analytical chemical technique used to study molecular structure; however, when it is applied to tissues, the resulting spectra reflect the total biochemical composition of these tissues. Almost all molecules absorb infrared light, and each type of molecule only absorbs infrared light at certain frequencies. This property provides a unique absorption spectral pattern or fingerprint through the entire infrared light spectrum. It provides a way to identify the molecule type (qualitative analysis) and the amount (25) Bambery, K. R.; Schu ¨ ltke, E.; Wood, B. R.; Rigley MacDonald, S. T.; Ataelmannan, K.; Griebel, R. W.; Juurlink, B. H.; McNaughton, D. A. Biochim. Biophys. Acta 2006, 1758, 900–907. (26) Haka, A. S.; Shafer-Peltier, K. E.; Fitzmaurice, M.; Crowe, J.; Dasari, R. R.; Feld, M. S. Proc. Natl. Acad. Sci. U.S.A. 2005, 102, 12371–12376. (27) Ro ¨mer, T. J.; Brennan, J. F.; Fitzmaurice, M.; Feldstein, M. L.; Deinum, G.; Myles, J. L.; Kramer, J. R.; Lees, R. S.; Feld, M. S. Circulation 1998, 10, 878–885.

or quantity of this molecule in the sample (quantitative analysis). Before applying the MLS procedure, a test of independency was performed on the model set of lipid reference spectra, which were chosen to represent the main kinds of lipids in the brain tissue.28 In fact, each spectrum of this database was fitted with all remaining reference spectra. As a result, a very poor fit was obtained for each spectrum, as judged by the large infrared features in the fit residuals (obtained by subtracting the fit result from the measured pure lipid spectrum). This result demonstrates that our reference spectra form an independent set and can be used to estimate the relative absorbance percentage fractions of each lipid in normal and glioma brain lipid extracts. Indeed, two lipid mixtures were then prepared, blindly, by using different reference lipids. The measured spectra were then fitted with all reference spectra (nine lipid spectra) to identify and quantify each individual compound present in the mixtures. We have successfully identified the lipids used in the mixtures (cholesterol, phosphatidylethanolamine phosphatidylcholine, galactocerebroside, and sphingomyelin) with an estimated error of about 5% (data not shown). This result shows that (i) even if these lipids share some common features, each of them has unique absorption spectral patterns (considered as a fingerprint) and (ii) the power of multivariate methods used the entire infrared light spectrum instead of the use of an isolated spectral feature. Several authors have used multivariate methods for the quantification of individual constituents in multicomponent mixtures. Antoon et al. have applied the least-squares method to the infrared spectra of a multicomponent mixture in order to quantify individual polymer components and mineral composition even in those cases where there is partial or complete overlap of the infrared spectral features.29 Haaland and Easterling have used MLS to estimate some organic solvent isomers and their mixtures.30 This multivariate method combined to FT-IR was used to estimate individual cell and tissue constituents that contribute to the early detection and rapid staging of many diseases. Wang et al. have investigated formalin-fixed Barrett’s esophagus for predicting the underlying histopathology, which will contribute to the early detection and rapid staging of many diseases.31 They reported that DNA, protein, glycogen, and glycoprotein comprise the principal sources of infrared absorption in the 950-1 800 cm-1 regime. The concentrations of these biomolecules can be quantified by using a partial least-squares fit and used to classify disease states with high sensitivity, specificity, and accuracy. Their model provides a very accurate fit despite the presence of a number of common spectral features or molecular interactions and variations that may introduce broadening of individual peaks. Mourant et al. have investigated the biochemical differences in mammalian cell cultures at different growth stages by the FT-IR technique.32 They used as the reference data set the major biochemical com(28) Olsson, N. U.; Harding, A. J.; Harper, C.; Salem, N. J. Chromatogr., B 1996, 681, 213–218. (29) Antoon, M. K.; Koenig, J. H.; Koenig, J. L. Appl. Spectrosc. 1977, 31, 518– 524. (30) Haaland, D. M.; Easterling, R. G.; Vopicka, D. A. Appl. Spectrosc. 1985, 39, 73–84. (31) Wang, T. D.; Triadafilopoulos, G.; Crawford, J. M.; Dixon, L. R.; Bhandari, T.; Sahbaie, P.; Friedland, S.; Soetikno, R.; Contag, C. H. 2007, 104, 1586415869. (32) Mourant, J. R.; Yamada, Y. R.; Carpenter, S.; Dominique, L. R.; Freyer, J. P. Biophys. J. 2003, 85, 1938–1947.


8411

Figure 3. FT-IR spectra (solid lines) of clusters averaging of CC (A), cortex (B), tumor (C), and necrosis (D) were fitted with a set of spectra from individual components (dotted lines) to quantify the chemical composition of the brain features. The curve under each spectrum shows the difference of the spectrum and the model fit, i.e., the residual spectral features that could not be fitted.

ponents present in the cell such as RNA, DNA, lipid, protein, and glycogen FT-IR. They identified the biochemical changes responsible for the spectral differences between cells in the exponential and plateau phases of growth. This MLS method has been frequently used in Raman spectroscopy. Romer et al. have quantified chemical composition of coronary atherosclerosis by free cholesterol, cholesterol ester, calcium salts, triglyceride and phospholipids, β-carotene, and delipidized arterial tissue.27 Haka et al. developed a chemical/ morphological model of breast tissue based on reference spectra: cell cytoplasm, cell nucleus, fat, β-carotene, collagen, calcium hydroxyapatite, calcium oxalate dihydrate, and cholesterol-like lipid deposits.26 Koljenvic et al. have quantified individual constituent of porcine brain using cholesterol, galactocerebroside, DL-Rphosphatidylcholine disteroyl and sphingomyelin, DNA, and BSA.33 Caspers et al. have measured the molecular concentration profiles in the skin (serine, glycine, pyrrolidone-5-carboxylic acid, arginine, ornithine, citrulline, alanine, histidine, urocanic acid) and for the sweat constituents (lactate and urea).34 Quantification of Biochemical Changes between Normal and Glioma Brain Tissues. To estimate the goodness of a fit, cluster averaged spectra of CC (cluster 5), cortex (cluster 1), tumor (cluster 9), and necrosis (cluster 12) were fitted with a set of spectra from individual components (dotted lines) to quantify the chemical composition of the brain features (Figure 3). The (33) Koljenovic, S.; Schut, T. B.; Vincent, A.; Kros, J. M.; Puppels, G. J. Anal. Chem. 2005, 77, 7958–7965. (34) Caspers, P. J.; Lucassen, G. W.; Carter, E. A.; Bruining, H. A.; Puppels, G. J. J. Invest. Dermatol. 2001, 116, 434–442.

8412


curve under each spectrum shows the difference of the spectrum and the model fit, i.e., the residual spectral features that could not be fitted. After fit analysis, no clear remaining spectral features are visible in the residual fit indicating that virtually all of the changes in molecular composition that are reflected in the FT-IR spectra have been accounted for. This work on detailed FT-IR maps may provide insight into different contributions of biochemical changes associated with the development of a glioma tumor and to understand the effects of these contributions in spectral discrimination between healthy and tumor tissues. In fact, the structure of the brain is not homogeneous. It is important to determine on each point of the tissue the contribution of the main constituents. Figure 4 shows the biomolecular distribution of each individual brain constituent (lipids, proteins, and nucleic acid) between normal, glioma, and necrosis features by a MLS procedure. Major lipids found in brain tissues can be classified into neutral lipid and phospholipids or gray matter and white matter lipids. The FT-IR maps from white and gray matter show large differences, representing the difference in molecular composition of these brain tissues. As result of MLS, lipids such as cholesterol was localized only in white matter (CC and CA) but not in gray matter (Figure 4A). In fact, cholesterol constitutes the highest amount of neutral lipids and count for approximately 10% of the dry weight of total lipid in adult brain, most of which are localized in myelin.35 The concentration of cholesteryl ester (cholesteryl (35) Davison, A. N. Adv. Lipid Res. 1965, 3, 171–196.

Figure 4. Biochemical distribution of each individual brain constituent (lipids, proteins, and nucleic acid) between normal, glioma, and necrosis features by a MLS procedure. To determine the molecular composition of tissues composition, the relative absorbance percentage of the various components were rescaled to add up to 100%. The spatial and biochemical information obtained from these maps can be used to identify which biochemical markers could be more potential indicators of such variations between normal and cancer tissues.

oleate) was mainly detected in white matter (Figure 4C) and decreased from the inner layer to the outer layer of the cortex. The highest levels of cholesteryl ester in rat are exhibited at early fetal and postnatal ages.36 These levels decrease rapidly with age, and a small (but significant) amount of cholesteryl ester was found in adult rat. This finding is in accord with the previous works on the quantification of brain lipid.37,21 In fact, white matter consists of bundles of axons covered by a sheet of myelin. Myelin is a lipoprotein consisting of 70-85% lipid. Almost all brain cholesterol is a product of local synthesis without any significant exchange with circulation and with an efficient recycling in the brain.38 On the other hand, no cholesterol was found in the tumor even in the necrotic part of tissue (Figure 4B). We expected some cholesterol and/or cholesteryl ester in the partial necrotic part of the tissues (Figure 4D). Koljenvic et al. have discriminated vital from necrotic glioblastoma tissues by Raman microspectroscopy.33 They demonstrated that necrotic tissue contains higher levels of (36) Suzuki, K. Chemistry and Metabolism of Brain Lipids. In Basic Neurochemistry; Albers, R. W., Ed.; Little, Brown, and Company: Boston, MA, 1972; pp 207-227. (37) Bradford, H. F. Chemical Neurobiology; W.H. Freeman and Co.: New York, 1986. (38) Bjo ¨rkhem, I.; Meaney, S. Arterioscler. Thromb. Vasc. Biol. 2004, 24, 806– 815.

cholesterol than vital tumor tissue. Yamada et al. came to the same conclusions by comparing necrotic and vital carcinoma tissues.39 Phosphatidylethanolamine (PE) was highly distributed in white matter, and the concentration was decreased in the cortex layers (Figure 4K). Brain gray matter is characterized by a higher phosphatidylcholine (PC) content compared to brain white matter (Figure 4I). The inner layer of neuron membranes primarily contains phosphatidylethanolamine, whereas in the outer layer phosphatidylcholine predominates. This result can explain the differences between the cortex layers found in the clustering analysis (Figure 1B, clusters 6, 4, 2, and 1). Olsen et al. have quantified, by chromatography, the total lipid extract from white and gray matter.28 They demonstrated that PE was more concentrated in gray matter (30.7%) than white matter (19.6%). In our study, we have quantified point by point the distribution of PE in normal tissues while this previous work has quantified total lipid extracted from white matter and gray matter. Indeed, it is difficult to precisely quantify the total lipids from the cortex because the distribution of different kinds of brain lipids was different between cortex layers. NMR spectroscopy on human brain extract showed main constituents of the biological membrane, namely, PC, PE, (39) Yamada, T.; Miyoshi, N.; Ogawa, T.; Akao, K.; Fukuda, M.; Ogasawara, T.; Kitagawa, Y.; Sano, K. Clin. Cancer Res. 2002, 8, 2010–2014.


8413

PS (phosphatidylserine), and SMY (sphingomyelin).40 They demonstrated that PC was 43.6% in the gray matter and 28.1% in the white matter, and PE was 28.1% in the gray matter and 26.8% in the white matter Galactocerebroside is localized only in white matter but not in gray matter (Figure 4G). Galactocerebroside has long been shown to be localized in the oligodendrocytes constituting the myelin sheath.41,42 Galactocerebroside is the main glycolipid component and the cell membrane of the myelinating cell contains a high percentage of this compound.43 This suggests that galactocerebroside may play an important role in the successive layering of cell membranes, a process unique to myelination. O’Brien et al. reported that white matter exhibits higher galactocerebroside than gray matter.44 The galactocerebroside content decreases in the vital tumor region (Figure 4H) because the oligodendroglia (myelin-forming cells) die and myelination ceases very early. On the other hand, the necrosis area shows a higher concentration of this cerebroside. This brain constituent can be used as spectroscopic marker for early detection of necrosis. In fact, the grade of malignancy and prognosis, in particular glioblastoma multiforme, is based on the presence of necrosis.36 No SMY was found in either gray or white matter (Figure 4E). Olsen et al. have reported that the sphingomyelin content in white and gray matter were, respectively, 4.4 and 3.2 in percentage of total lipids.28 Indeed, Laule et al. have demonstrated that myelin of rat has less sphingomyelin than bovine or human myelin.45 Folch et al. showed that the sphingomyelin concentration increases in the brain during growth and development in rats.46 All these results could explain the absence of the sphingomyelin in rat brain because of the very low concentration of this lipid in the white matter of rats. On the other hand, this lipid content was slightly higher in the tumor and decreased in the necrotic tissue (Figure 4F). Ledwozyw et al. reoprted that glioma tumors contain a greater amount of SMY as compared to normal cortex tissue.43 PS was equally distributed in the gray matter layers but to a less extent in the white matter (Figure 4M) and disappeared in the tumor (Figure 4N). The tumor tissue was characterized by a decrease in the PC and PS content compared to white matter and gray matter (parts J and N of Figure 4, respectively). The decreased concentration of these phospholipids in tumors may be due to increased phospholipid degradation, which can result in the modification of composition, structure, and stability of the membranes and thus to membrane dysfunction. The decrease in phospholipid levels may also be due to the decreased concentration of free fatty acids in tumor tissues. Phospholipids are major substrates for lipid peroxidation in cells. These changes in phospholipids content may therefore be partly responsible for the (40) Paul, C. R.; Lenkinski, R. E.; Boyko, O. B.; Gur, R. E.; Gur, R. C.; Cecil, K. Proton Magnetic Resonance Spectroscopy of Patients with First Episode of Schizophrenia. In International Society for Magnetic Resonance in Medicine, New York, Book of Abstracts; 1996, p 1001. (41) Steck, A. J.; Perruisseau, G. J. Neurol. Sci. 1980, 47, 135–144. (42) Gjerset, R. A.; Fakhrai, H.; Shawler, D. L.; Turla, S.; Dorigo, O.; GroverBardwick, A.; Mercola, D.; Wen, S. F.; Collins, H.; Lin, H.; Garcia, M. V.; Kruse, C. A.; Royston, I.; Sobo, R. E. In Vitro Cell Dev. Biol.: Anim. 1995, 31, 207–214. (43) Ledwozyw, A.; Lutnicki, K. Acta Physiol. Hung. 1992, 79, 381–387. (44) O’Brien, J. S.; Sampson, E. L. J. Lipid Res. 1965, 6, 537–544. (45) Laule, I.; Vavasour, S.; Kolind, D.; Li, T.; Traboulsee, G.; Moore, A.; MacKay, A. L. Neurotherapeutics 2007, 4, 460–484. (46) Folch-Pi, J. In Biochemistry of the Developing Nervous System; Waelsch, H., Ed.; Academic Press: New York, 1955; Vol. 121.

8414


decrease in lipid peroxidation in tumor tissues. Martin et al. have compared the fatty acid composition of tumor tissue from glioma with that of normal brain tissue using gas-liquid chromatography.47 They reported that phosphatidylserine and phosphatidylethanolamine phospholipid classes were reduced in the glioma samples. Oleic acid was located in the white matter (Figure 4O) rather than gray matter. On the other hand, this fatty acid content increased in the tumor and particularly in the perinecrotic region (Figure 4P). Lombardi et al. have revealed an elevation in the monounsaturated fatty acid (oleic acid) level in the highly malignant glioblastoma.48 On the other hand, linoleic acid was present in gray matter rather than white matter (Figure 4Q). Naidu et al. have demonstrated that intratumoral administration of linoleic acid is a possible approach to the treatment of human glial tumors.49 Previous work has reported that a significant rise in oleic content in tumor tissue was observed especially around the necrosis.33 Our last study, using nile red staining revealed the presence of phospholipids and neutral lipids in the perinecrotic area and an absence of neutral lipids in the fully necrotic zone.21 No change in DNA content was observed between white matter and gray matter (Figure 4S). The overall nucleic acid content was higher in tumor tissues compared to normal tissues (Figure 4T). Among these components, the nucleic acids are vital as they are altered in cancerous conditions where cell division and growth are affected. The processes of transcription and DNA duplication during carcinogenesis would lead to increased signals from unwound DNA. The common underlying effects of malignancy are changes in the nuclear components which lead to an altered metabolism and biochemical composition in the malignant tissues. These changes would leave a signature in the FT-IR spectra that is characteristic of nucleic acids. This result suggests that spectral features of nucleic acids may be a sensitive marker for discrimination between normal and tumor. Figure 4U,V displays the distribution of proteins in the white and gray matter and in the tumor. These proteins were highly distributed in the tumor and decreased from the cortex to white matter. Previous work demonstrated that the biological characteristics of intracerebral glioma are defined by numerous proteins, which implies that the aggressiveness of gliomas can not be predicted by single or small numbers of protein markers but rather by combinations of many proteins.50 These proteins could be predictive markers for the aggressiveness of gliomas and could be direct and rational targets for antiglioma therapy. FT-IR spectroscopy can provide a total simultaneous chemical analysis (on lipid, proteins and nucleic acid) in a nondestructive and noninvasive manner. The methods adopted do not require any sample preparation. Indeed, FT-IRM imaging, with high spatially resolved morphological and biochemical information can be used as a diagnostic tool, complementary to histopathology in (47) Martin, D. D.; Robbins, M. E.; Spector, A. A.; Wen, B. C.; Hussey, D. H. Lipids 1996, 31, 1283–1288. (48) Lombardi, V.; Valko, L.; Valko, M.; Scozzafava, A.; Morris, H.; Melnik, M.; Svitel, J.; Budesˇinsky´, M.; Pelna´r, J.; Steno, J.; Liptaj, T.; Zalibera, L.; Budinska´, J.; Zlatosˇ, J.; Giuliani, A.; Mascolo, L.; Leibfritz, D. Cell. Mol. Neurobiol. 1997, 17, 521–535. (49) Naidu, M. R.; Das, U. N.; Kishan, A. Prostaglandins, Leukotrienes, Essent. Fatty Acids. 1992, 45, 181–184. (50) Yasuo, I.; Tsukasa, S.; Takaki, H.; Yuichiro, N.; Hiroshi, I.; Masaki, T.; Akira, Y. Cancer Res. 2004, 64, 2496–2501.

order to understand the molecular changes associated with tissue alteration. On the other hand, Thin Layer Chromatography (TLC) method is a very sensitive analytical technique that can be used for separation of individual constituents in a mixture. This technique, however, requires sample preparation (destructive method) and extraction. Indeed, all lipids used in this study were known to be present in the brain determined by the TLC technique.51 The aim of our study was to map individual lipids in different brain tissues (white matter, gray matter, tumor, necrosis, and peritumor zone). Chromatographic techniques provide information on the quantification of total lipid in normal brain structures and in the tumor, but the extraction of lipids from necrotic and perinecrotic areas of the tissue is difficult. In fact, the structure of the brain is not homogeneous. It is important to determine at each point on the tissue, the contribution of the main constituents. For example the cortex layer is composed of four different layers with different lipid composition. With TLC, it is only possible to quantify the total lipids in the whole cortex but not in each cortex layer. CONCLUSION This study demonstrates the potential of FT-IRM in successfully discriminating between normal, tumoral, peri-tumoral, and necrotic tissues in brain structures. Our results confirm that (i) white matter is characterized by a higher lipid content than gray matter, (ii) the development of a tumor is characterized by a reduction in total lipid content, (iii) proteins and nucleic acids can be used as spectroscopic biomarkers to differentiate between (51) Cherayil, G. D.; Scaria, K. S. J. Lipid Res. 1970, 11, 378–381.

normal, tumoral, and necrotic brain structures, and (iv) pseudocolor maps contain more information than simple histopathological studies. We have accurately quantified the contributions of biochemical changes associated with the development of a tumor. Detailed infrared spectroscopic mapping can be used as a tool for obtaining more information from tissue sections than present day histological techniques. We have shown heterogeneity in the spectra between normal and tumoral tissues. The use of fitting procedures on tissue spectra with pure compounds known to be found in normal or tumoral brain tissues revealed which tissue components contributed to the spectral discrimination. Our results demonstrate that white matter contains a higher concentration of PE, a slightly higher percentage of cholesterol and galactocerebroside, and less accumulation of PC than gray matter. As such, FT-IR spectra of gray matter were characterized by stronger signal contributions from proteins when compared to white matter. Glioma tissues contain a high amount of sphingomyelin, nucleic acids, and oleic acid. On the other hand, necrotic tissues were characterized by a higher content of galactocerebroside and PS as compared to the tumor. ACKNOWLEDGMENT This work was supported by La Ligue contre le Cancer de Marne, France. The authors would like to thank Dr. Ganesh Sockalingum for providing assistance in the preparation of the manuscript. Received for review May 14, 2008. Accepted August 27, 2008. AC800990Y


8415

Modeling and Quantifying Biochemical Changes in ... - ACS Publications

Recommend Documents