Optical Analysis of Glioma: Fourier-Transform Infrared Spectroscopy Reveals the IDH1 Mutation Status

This article is featured in Highlights of This Issue, p. 2465

Somatic mutations of the isocitrate dehydrogenase 1 and 2 (IDH1 and IDH2) genes are frequent and early causative events in the pathogenesis of a subset of gliomas, mainly low-grade gliomas and secondary glioblastomas (1–3). Patients with IDH-mutated tumors show improved survival compared with patients who have IDH wild-type gliomas of the same histologic grade (4). Hence, it became increasingly apparent that IDH mutation status accounts for much of the prognostic information previously rendered by histologic grading (5). The latter fact underscores the prognostic significance and the clinical impact of IDH mutations in patients with glioma.

IDH mutations inactivate the standard enzymatic activity and confer novel activity on IDH1 for conversion of α-ketoglutarate to 2-hydroxyglutarate (2HG). This gain of function causes an accumulation in glioma cells of 2HG (6), an oncometabolite that drives the oncogenic activity of IDH mutations contributing to malignant transformation and glioma genesis through several pathways (7). IDH mutations and/or 2HG accumulation represent ideal surrogate markers for both diagnosis and for monitoring the treatment response in affected tumors.

As testing for IDH1/2 mutations is now considered standard by international guidelines for glioma management, reliable and effective methods of detection are of paramount importance (8). IDH mutation detection methods include IHC, DNA sequencing, and PCR analysis, all of which rely on invasive tissue or cerebrospinal fluid sampling for mutation detection (9). One important limitation is that all these methods involve long analytic times, which take several days and expensive equipment, making a rapid diagnosis unlikely. In addition, several methods have been reported to test for the presence of 2HG in tumor samples (6) as well as noninvasively using magnetic resonance spectroscopic technology (10, 11). Quantitatively measuring 2HG in tumor cells over time may assist in monitoring for tumor recurrence, differentiate treatment effect from pseudoprogression, and assess tumor response to IDH-specific pharmacotherapies (12). However, no intraoperative 2HG detection method with a high accuracy has been described to date. Along this line of thinking, conclusive intraoperative confirmation of IDH mutations might guide the decision to pursue definitive neurosurgical resection in diffuse infiltrative glioma (13) or pave the way to IDH1-dependent in situ therapeutics and therapy.

“Vibrational spectroscopy” is the collective term used to describe two analytic techniques: infrared and Raman spectroscopy. They probe vibrational energy levels that are associated with the chemical bonds in the sample without the need for labeling. Both are nondestructive, noninvasive tools that deliver comprehensive information about the sample's composition in the form of a spectrum. Several reports highlighted the potential of vibrational spectroscopy for clinical applications (14–16). Moreover, intraoperative detection (17) and diagnosis (18) of glioma were recently reported. By using these techniques, it is possible to distinguish brain tissue from necrosis and tumor (19), to provide tumor type and grade (20), to identify the origin of brain metastases (21), and to discriminate tumors that differ in a single molecular alteration, such as hormone-producing and hormone-inactive pituitary adenoma (22).

The development of a simple, rapid, specific, and sensitive diagnostic technique for IDH mutations based on vibrational spectroscopy might provide valuable intraoperative diagnosis in affected patients. Therefore, we investigated the ability of Fourier-transform infrared (FT-IR) spectroscopy to analyze the IDH1 mutation status in glioma samples.

The near tetraploid primary glioblastoma cell line HT7606 has been described recently (23). The immortalized astrocytoma cell lines SVGp12 and U87-MG were authenticated using SNP profiling (Multiplexion GmbH) and cultured in Basal Minimal Eagle medium (catalog no. 41010-026) supplemented with 10% v/v heat-inactivated FBS (catalog no. A15-101), 2 mmol/L l-glutamine (catalog no. 25030-024), and 1% nonessential amino acids (catalog no. M11-003; all from Gibco). All cell lines were tested as free of Mycoplasma contamination using MycoAlert kit (Lonza Cologne GmbH).

The coding regions of human IDH1 and IDHR^132H fused to a C-terminal His6- and myc-tags and flanked by AgeI and NotI restriction were synthesized (Eurofins MWG Biotech) and ligated into AgeI/NotI-digested lentiviral vector pHATtrick-puro (24). All vectors were confirmed by DNA sequencing. Lentiviral particle production and transduction of cells were performed as described previously (24).

Human brain tumor tissue was obtained during routine tumor surgery. The patients gave written consent; the study was approved by the ethics committee at the University Hospital Carl Gustav Carus, Technische Universität Dresden (Dresden, Germany; EK 323122008); and the study was conducted in accordance with the Declaration of Helsinki. A total of 98 tumor samples from 90 patients with glioma were analyzed (Supplementary Tables S1 and S2). The samples were either directly subjected to attenuated total reflection (ATR) FT-IR without any processing or snap frozen at −80 °C and embedded in tissue freezing medium (catalog no. 14020108926; Leica Microsystems) to prepare cryosections of 16-μm thickness on glass or CaF₂ slides for FT-IR spectroscopy.

After ATR FT-IR, the samples were formalin fixed and subjected to diagnostic histopathologic workup. IDH1 mutations were assessed using direct DNA sequencing, as reported elsewhere (2) or by standard IDH1 IHC.

IR spectra were acquired in transmission from dry cell pellet preparations on CaF₂ slides (n = 3 replicates for each group) or unstained cryosections of human brain tumors (n = 34). The area of measurement in the human brain tumor samples was chosen according to the evaluation of stained consecutive sections by a pathologist.

The system used was described elsewhere (25). Briefly, IR spectra were acquired in transmission mode with an FT-IR spectrometer TENSOR 27 equipped with an IR microscope HYPERION 3000 (both from Bruker Optics GmbH) and a 15× Cassegrain objective (0.4 NA). An area of 175 μm × 175 μm was imaged by a 64 × 64 mercury cadmium telluride focal-plane array detector using a spectral resolution of 6 cm⁻¹. A background spectrum was recorded from pure CaF₂ slide. A total of 200 interferograms were coadded and Fourier transformed by applying Blackman–Harris apodization without filling. Each transmission spectrum was ratioed to the background spectrum. The transmission spectra were converted to absorbance values. For each sample (both cells and tumor cryosections), a single area was imaged, and thus, 4096 spectra were acquired in approximately 1 minute.

An atmospheric compensation was performed to subtract contributions of residual water vapor bands from the spectra and 4 × 4 binning was applied in OPUS 7.2 (Bruker Optics GmbH).

ATR IR spectra were acquired from 64 fresh, unprocessed tumor biopsies. The system used was described elsewhere (26). Briefly, measurements were performed on a portable ALPHA-P ATR spectrometer (Bruker Optics GmbH) equipped with a diamond crystal. The instrument's anvil was used to force the contact between crystal and sample. Measurements were performed using a spectral resolution of 4 cm⁻¹ and coadding of 128 interferograms and Fourier transformed by applying Blackman–Harris apodization without filling. Spectra were ratioed against the background spectrum that was recorded without sample. One spectrum was obtained for each sample, and the measurement time was approximately 1 minute.

Further analysis was performed using the MATLAB package (MathWorks Inc.). Data were reduced to the spectral region of 950 to 1800 cm⁻¹. A linear baseline was applied, followed by normalization.

The differentiation of cell lines expressing wild-type IDH1 (IDH1-wt) or mutated IDH1 (IDH1-mut) was obtained using a classification by quadratic discriminant analysis (function classify in MATLAB) combined with manual feature selection based on the comparison with the 2HG reference spectrum and the difference spectrum of IDH1-mut versus IDH1-wt cell lines. The training set consisted of 25% of the dataset (1728 spectra).

The classification function for human brain tumors was developed and tested using spectra of 34 cryosections. The algorithm was implemented in MATLAB as described elsewhere (27). The spectra of four IDH1-wt samples and four IDH1-mut samples were selected by the algorithm and used as training set. An optimal selection routine was applied on the training set to retrieve the spectral regions that contained the information discriminating IDH1-wt and IDH1-mut. The performance of the classification was assessed with the leave-one-out cross-validation method. After optimization of the classification rate on the training data, the remaining spectra served as an independent test set and were analyzed using those selected spectral regions and quadratic discriminant analysis.

Spectra of fresh brain tumor biopsies (n = 64) were classified using quadratic discriminant analysis based on spectral regions selected by the algorithm described above in combination with manual feature selection using the regions identified on glioma cryosections. Ten spectra of IDH1-wt samples and 10 spectra of IDH1-mut samples were selected as training set.

The probability of class membership was transformed into a color code and plotted for each sample.

Statistical analyses were performed with Prism 6.0 (Graph Pad Software).

SvGp12, U87-MG, and the primary glioblastoma cell line HT7606 were transduced with either IDH1-wt or IDH1-mut. The expression of IDH1-wt or IDH1^R132H protein was confirmed using Western blot analysis (Supplementary Fig. S1). IR spectra were obtained from three independent preparations for each cell line.

Figure 1 shows the mean spectra of the three cell lines expressing either IDH1-wt or IDH1-mut. They display the same overall spectral pattern of bands that is a general feature of cells and tissue (dotted lines, band assignment; see Table 1). The spectra are dominated by the bands of amide II and amide I at 1550 and 1650 cm⁻¹, respectively. The spectral region of 950 to 1200 cm⁻¹ results from the complex overlap of multiple molecular vibration bands of carbohydrates and nucleic acids. In this region, a cell line–specific spectral shape was observed (arrows in Fig. 1). For instance, U87-MG exhibits a small band at 1153 cm⁻¹, whereas HT7607 and SvGp12 show bands at 1173 cm⁻¹. Those characteristic spectral features of the cell lines were likewise found in IDH1-wt or IDH-mut–expressing cells.

To identify a potential spectral marker for IDH1 mutation, the IR spectra of the different cell lines were merged and grouped according to the IDH1 mutation status. Figure 2A shows the mean IR spectra with SD for IDH1-wt cell lines and IDH1-mut cell lines.

At first glance, the IR spectra look related. Spectral differences become recognizable and are clearly separable in the difference spectrum (Fig. 2B, black spectrum). The spectral regions at 1000 to 1100, 1300 to 1430, and 1560 to 1610 cm⁻¹ are more intense in IDH1-mut cell lines (i.e., regions where the difference spectrum is positive), whereas the band intensities in the regions at 1500 to 1545, 1625 to 1650, and 1690 to 1730 cm⁻¹ are lower compared with the spectra of IDH1-wt cells. The comparison with the IR spectra of 2HG (Fig. 2B, gray spectrum), which is elevated in the case of IDH1-mut in human tumors and engineered cell lines (6, 28), indicated higher band intensities in the spectra of IDH1-mut cells at similar band positions (dashed lines). This might suggest that those bands reflect the increase of 2HG in the IDH1-mut cell lines. In addition, further spectral differences were observed (Fig. 2B, black arrows). The positive band at 1050 cm⁻¹ is assigned to C-O stretching coupled with C-O bending of C-OH groups and might indicate elevated carbohydrates, whereas the negative bands in the difference spectrum in the region around 1740 cm⁻¹ are assigned to C=O stretching and might point to decreased lipid content in IDH1-mut cell lines. Moreover, bands assigned to vibrations of functional groups in proteins at 1522 and 1641 cm⁻¹ suggest an altered protein profile (29).

The difference spectrum indicated that FT-IR spectroscopy is sensitive to metabolic changes induced by the IDH1 mutation. Therefore, we tested whether IDH1-mut and IDH1-wt cell spectra can be mathematically discriminated. Sixteen positions of the spectra were selected, either corresponding to bands of the 2HG spectrum or to features in the difference spectrum. The following positions were chosen: (i) the band positions that were consistent with bands of 2HG (1072, 1174, 1203, 1311, 1344, 1416, 1450, 1568, and 1589 cm⁻¹; dashed lines in Fig. 2B); (ii) other bands of 2HG that were not identified in the difference spectrum (999, 1236, and 1267 cm⁻¹; gray arrows in Fig. 2B); and (iii) other maxima and minima of the difference spectrum (1043, 1522, 1641, and 1740 cm⁻¹; black arrows in Fig. 2B). On the basis of the absorbance at those spectral positions, the spectra were grouped in IDH1-wt and IDH1-mut with a correct rate of 92.2% by quadratic discriminant analysis (correct rate test set: 92.1%; reclassification rate training set: 92.6%). Figure 2C visualizes the result of the classification. Each lane represents one sample. The probability of class membership was calculated for all 256 spectra of one sample and is shown using color coding (see Supplementary Fig. S2). In all samples, the majority of the spectra were assigned to the correct class; a reliable discrimination was obtained for all cell lines. This shows that FT-IR spectroscopic data comprise the information about the IDH1 mutation status, and that this information can be extracted in homogenous cell culture systems.

In the next step, we employed the technique to analyze cryosections of human brain tumor samples (n = 34; see Supplementary Table S1). This allowed the selection of the measurement position on solid tumor according to reference pathology on consecutive sections. Many spectra were obtained for each sample to generate large datasets suitable for mathematical analysis. The mean spectra of all tumor samples are shown in Fig. 3A and display a high variability. This reflects the changes in tissue composition that vary among patients and among tumor types. As an example, two tumors exhibited an altered spectral shape at 1113 cm⁻¹ that is assigned to ν(PO₄⁻)_sym vibrations characteristic for acid phosphate of hydroxyapatite (Fig. 3A, arrow) and might indicate the presence of calcifications that are abundant in glioma (30). The corresponding difference spectrum (Fig. 3B) shows the strongest spectral features at 1113 cm⁻¹, representing the calcifications that were only found in IDH1-mut tumors. Furthermore, it indicated alterations in bands characteristic of nervous tissue (compare Table 1): Lower intensities around 1230 cm⁻¹ indicate less phosphate groups and might be interpreted either as diminished DNA content or lower content of phospholipids in IDH1-mut tumor tissue. Similar cell numbers were found in IDH1-wt and IDH1-mut samples, suggesting similar DNA content (Supplementary Fig. S3). Therefore, alterations in the region around 1230 cm⁻¹ rather represent decreased or altered phospholipids in IDH1-mut tumors. This is consistent with the lower intensity of the spectral band at 1740 cm⁻¹ that is assigned to ν(C=O) and can be used for the assessment of overall lipid content in nervous tissue (Table 1).

The changes in the bands around 1550 and 1650 cm⁻¹ might reflect variations in the protein profile. In addition, the more intense bands in the region 1005 to 1030 cm⁻¹ suggest variations in carbohydrates. The interpretation of the band at 1352 cm⁻¹, which is increased in IDH1-mut tumor samples, is difficult. This band might reflect elevated levels of 2HG (band position at 1344 cm⁻¹ in the reference spectrum of 2HG and at 1348 cm⁻¹ in the difference spectrum of cell lines), but no other bands characteristic for 2HG were identified (compare Fig. 2B). Therefore, it might rather indicate an increase in proteins, especially collagen of the extracellular matrix (band of CH₂ wagging at 1340 cm⁻¹).

We used quadratic discriminant analysis to exploit the spectral information for the prediction of IDH1 mutation status. Because of the high variability of human tissue samples, we employed an automatic selection routine to identify spectral regions suitable for classification. Five spectral regions were selected and are indicated in Fig. 3A and B as dashed gray lines (1103, 1362, 1441, 1485, and 1553 cm⁻¹). After using four samples of each group as training set, the algorithm provided a probability for the assignment to IDH1-wt or IDH1-mut class for each spectrum. This probability of class membership is shown in Fig. 4 for all samples. In 10 cases, all 256 spectra of the sample were assigned to the correct class. For the other cases, the sample's spectra were assigned to both groups. However, the majority (>70%) of spectra were assigned correctly to either IDH1-wt or IDH1-mut in 30 samples, retrieving the correct information about IDH1 mutation status from the spectroscopic dataset (see Supplementary Table S1 for the result of the classification broken down by samples).

Four of the 34 tumor samples were not assigned to the correct class by the algorithm (indicated by asterisk in Fig. 4). However, no obvious spectral differences or indications for contamination were observed by looking at the misclassified spectra. It is important to state that spectra with distinct features such as calcifications (band at 1113 cm⁻¹; compare Fig. 3A arrow) were correctly classified (labelled with “C” in Fig. 4).

We further tested our approach on fresh, unfixed bulk samples of human glioma (n = 64; Supplementary Table S2) to approximate a potential clinical application for intraoperative tissue analysis. Biopsies were directly obtained from the neurosurgeon and subjected to ATR FT-IR spectroscopy, a variant of infrared spectroscopy that allows the analysis of bulk samples. However, it has the limitation that the spectral signal of tissue is overlaid in some regions to the spectral signal originating from water (see Supplementary Fig. S4). Measurements were performed within minutes after removal (median, 6.5 minutes; range, 1–20 minutes) according to established protocols (26). The tissue spectra were obtained in approximately 1 minute and integrate the information from the region of sample that is in contact with the measurement crystal (area of ∼0.5 × 0.5 mm). The spectra of all samples are shown in Fig. 5A and confirm the high variability of human glioma similar to the spectra of tissue sections (compare Fig. 3A). Interestingly, the spectra with high band intensities in the region of around 1080 cm⁻¹ fell into the group of IDH1-mut tumors. Quadratic discriminant analysis was performed on the basis of the spectral regions 1103, 1362, 1406, 1446, 1457, and 1558 cm⁻¹ (dashed lines in Fig. 5A) and assigned 55 of the 64 spectra correctly to the IDH1-mut or IDH1-wt group (Fig. 5B; correct rate: 86%, correct rate test set: 82%, reclassification rate training set: 95%). Four of the nine misclassified samples were identified as infiltrating tumor, and three were obtained from recurrent glioma, whereas two misclassified samples did not have any specific histologic or clinical features.

Here, we show for the first time that FT-IR spectroscopy of tissue samples provides information about a specific genetic alteration. We focused on IDH1 mutation in brain tumors, as it is known to alter cell metabolism and therefore, constitutes an ideal target for vibrational spectroscopic techniques. FT-IR spectroscopy does not directly detect the missense mutation at the level of the DNA or the replacement of arginine at 132 of the IDH1 protein but rather addresses changes in the biochemistry of the cell that are the consequence of the IDH1 mutation. The spectral changes might reflect both metabolic and epigenetics alterations (methylation of histones), which can be addressed by vibrational spectroscopy (31, 32).

The most prominent metabolic change in IDH1-mut tumor cells is the increased level of 2HG (6). Spectral features that might be consistent with 2HG increase were detected in the difference spectra of the IDH1-mut versus IDH1-wt cell lines. Glioma cell lines that were genetically modified to express IDH1-mut were shown to express high levels of 2HG (28, 33, 34). However, the increase of 2HG was not obvious in the difference spectrum of human glioma. It was either masked by overlapping contributions of other tissue constituents, or the difference in 2HG may be not evident when merging patients' datasets. The level of 2HG is highly variable among patients [4.50–31.56 μmol/g (6), up to 2,000 nmol/mg in secondary glioblastoma (10), 2.53 ± 0.75 μmol/g in glioma (35), ∼10–50 μg/mg protein (36)]. Furthermore, an overlap in 2HG levels in IDH1-wt and IDH1-mut glioma was shown (37), and some IDH1-mut tumor tissue might not even show detectable levels of 2HG (38). Therefore, metabolic changes other than production of 2HG account for the spectral differences observed in our study.

Similar spectral regions allowed the classification of IDH1-mut and IDH1-wt tumors in fresh samples or cryosections (compare dashed lines in Figs. 3A and 5A). Therefore, band assignment to functional groups might help to gain insight into the underlying biochemical differences. The band at 1103 cm⁻¹ is assigned to C-O stretching vibrations and might indicate changes of carbohydrates. The bands at 1362 and 1441 (sections), 1446 (fresh tissue), and 1457 cm⁻¹ (fresh tissue) can be related to CH₂ bending in fatty acids. The spectral position at 1406 (fresh tissue) and 1485 cm⁻¹ (sections) can be assigned to COO⁻ symmetric stretching vibrations and to C-H deformation, respectively. Those can be likewise attributed to fatty acids and proteins. However, the spectral positions at 1558 (sections) and 1553 cm⁻¹ (fresh tissue) are clearly related to amide II and, therefore, to proteins (29, 39). In combination with the analysis of the difference spectra, this underlines the importance of lipid- and protein-related spectral bands for recognition of IDH1 mutation status in human glioma.

It is well-known that IDH1 mutations cause global changes in cellular metabolism and signaling (40). The pathways that depend on normal IDH1 activity are impaired, levels of NAPDH are strongly reduced, and gene expression is altered by hypermethylation (3). In the current study, we assigned spectral changes to variations in (phospho-)lipid content and protein profile based on FT-IR spectroscopic datasets of both cell lines and human glioma. This is consistent with the changed phospholipid profile found in experimental and human glioma. In both models, IDH1-mut tumors displayed decreased levels of phosphoethanolamine and increased levels of glycerophosphocholine (41), and a significant decrease in phosphocholine was observed in IDH1-mut cell lines (28). Interestingly, increased band intensities at 1380, 1455, and 1465 cm⁻¹ were found in hypoxic glioma cells analyzed by IR spectroscopy and were interpreted as an altered lipid signal (39). IDH1-mut glioma are known to express higher levels of HIF-1α that is usually upregulated in hypoxic conditions (42).

IR spectroscopy does not allow the extraction of information on specific proteins if analyzing tissue samples. However, the data indicated a genotype-related altered protein profile. This is consistent with variations of amino acid levels in IDH1-mutated cell lines and tumor tissue that have been reported by others: IDH1-mutated cell lines showed elevated levels of glycine, serine, threonine, asparagine, phenylalanine, tyrosine, tryptophan, and methionine, whereas aspartate was decreased (40). Furthermore, n-acetylated amino acids were found to be reduced in IDH1-mut cells. Lower levels of amino acids (phenylalanine, glutamine, and glutamate) and amino acid derivate (n-acetyl-glutamate, n-acetyl-aspartate, and n-acetyl-histidine) were observed in human glioma (43). In contrast, IDH1 mutation does not alter the levels of glycolytic or pentose phosphate pathway intermediates (28, 40, 43).

Comprehensive interpretation and understanding of the entire molecular changes that are pictured in the IR spectra of IDH1-mut versus IDH1-wt cells and tumors were not possible in the frame of this study. This might be in part due to the fact that bands assigned to certain molecular vibrations are overlapping. Related to this intrinsic biochemical unspecificity, underlying causes of sample misclassification (four cryosections and nine fresh biopsies) remain unclear. Analysis of misclassified spectra did not provide conclusive results due to tumor heterogeneity that was likewise reflected in high interpatient variability of the IR spectra. However, the finding that infiltrative tumors were among the misclassified fresh biopsies might indicate a limitation of the technique and/or of the classification strategy. Further research is needed for evaluation of vibrational spectroscopy for detection of IDH1 mutation in the infiltration zone.

In the current study, spectroscopic data were shown to comprise the information about the IDH1 mutation status of human glioma that constitutes a prognostic factor. Furthermore, determination of the IDH1 mutation is essential for current glioma grading according to the World Health Organization (WHO; ref. 44) and might influence therapeutic decisions or frequency of screening after glioma resection.

We used FT-IR spectroscopy on cryosections of frozen tissue samples to validate the method. Furthermore, ATR FT-IR spectroscopy of fresh, unfixed biopsies of human glioma confirmed our findings despite the disturbance due to the spectral contributions of water. As an alternative vibrational spectroscopic technique, which compared with IR spectroscopy is relatively insensitive to water content inside the tissue, Raman spectroscopy can be employed as well. Furthermore, fiber-based technical solutions might be applied to obtain spectroscopic information even in situ, before removing suspicious tissue, as recently shown using Raman spectroscopy for analysis of human glioma (17). In that study, diagnostic information was provided within 0.2 seconds using a hand-held Raman probe in patients with glioma.

Therefore, an intraoperative assessment of the IDH1 mutation status in glioma by vibrational spectroscopy is technically feasible and manageable in the clinical context. Perspectively, this opens the possibility to analyze brain tumors at multiple positions during regular resection and to obtain intraoperative diagnostic information about IDH1 mutations within seconds. From a clinical perspective, a rapid assessment of the IDH mutation status facilitates intraoperative neurosurgical decisions. It helps distinguish tumor from nonneoplastic lesions, especially in limited tissue, and therefore, decreases the risk from additional samplings in stereotactic biopsies (13). Furthermore, unlike primary glioblastoma, IDH1-mutant gliomas benefit from complete resection of enhancing and nonenhancing disease (45).

On the basis of the current findings, we expect that other clinically relevant mutations and (epi)genetic changes that influence metabolics or tissue composition are amenable to vibrational spectroscopy. Besides IDH1 mutation, other diagnostic, prognostic, and predictive molecular markers of brain tumors like α-thalassemia/mental retardation syndrome X-linked (ATRX) gene mutation, 1p/19q codeletion, and telomerase reverse transcriptase (TERT) promoter mutation or O6-methylguanine-DNA methyltransferase (MGMT) promoter methylation are of paramount clinical importance and guide treatment decisions (46).

In conclusion, optical spectroscopy constitutes a translational approach for brain tumor analysis, although with the limitation that the threshold of tumor infiltration required to produce a spectral difference remains unknown. In perspective, fast optical biopsies in situ might allow the adaptation of resection strategy and the immediate application of local therapies. Personalized treatment regimens could be started early in the therapeutic process based on the spectral fingerprint of the patient's tumor.

No potential conflicts of interest were disclosed.

Conception and design: O. Uckermann, T.A. Juratli, E. Koch, M. Kirsch

Development of methodology: O. Uckermann, R. Galli, M. Conde, K. Geiger, G. Steiner

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): O. Uckermann, T.A. Juratli, M. Conde, R. Wiedemuth, D. Krex, K. Geiger, M. Kirsch

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): O. Uckermann, T.A. Juratli, R. Galli, M. Conde, K. Geiger, G. Steiner, M. Kirsch

Writing, review, and/or revision of the manuscript: O. Uckermann, T.A. Juratli, R. Galli, D. Krex, A. Temme, G. Schackert, E. Koch, G. Steiner, M. Kirsch

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): O. Uckermann, A. Temme, G. Schackert, M. Kirsch

Study supervision: M. Kirsch

Other (neuropathologic evaluation): K. Geiger

Financial support was received from Bundesministerium für Bildung und Forschung (BMBF “EndoCARS” 13N13807; to M. Kirsch and E. Koch, and “MediCARS” 13N10777; to G. Schackert, M. Kirsch, and E. Koch), Medical Faculty of TU Dresden (Habilitationsförderung für Frauen; to O. Uckermann), and European Union and Free State of Saxony (EU-ESF GlioMath-DD; to A. Temme). The authors thank Elke Leipnitz for excellent technical assistance and Dr. Barbara Klink, Institut für Klinische Genetik, Medizinische Fakultät Carl Gustav Carus, TU Dresden, Dresden, Germany, for providing the original SvGp12 cell line.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.