Feasibility study on the application of Raman spectroscopy in the diagnosis of glioma

Introduction

Glioma is the most common primary tumor of the nervous system, and malignant gliomas seriously threaten human health. Although the survival of patients has improved with the advancement of surgery, radiotherapy, and chemotherapy techniques, the treatment effect of malignant gliomas is still unsatisfactory. If the tumor boundary can be accurately and quickly determined during surgery, it will undoubtedly improve the surgical effect. Intraoperative neuronavigation technology can help distinguish potential boundaries of tumors, but significant errors may occur due to tissue displacement during surgery. The application of intraoperative magnetic resonance imaging can improve the total resection rate of gliomas, but it is expensive and cumbersome to operate, making it difficult to popularize in the current situation. This poses new requirements for techniques that can achieve rapid diagnosis and determine the boundaries of gliomas during surgery.

Raman spectroscopy (RS) is a phenomenon of light scattering, in which the incident laser interacts with molecular chemical bond vibrations, resulting in inelastic collisions and energy loss of the laser. This energy difference corresponds to the frequency of chemical bond vibrations, which is the fingerprint information of molecular vibrations. RS can reflect the chemical composition of the sample and analyze the composition and concentration of biological macromolecules such as proteins, nucleic acids, esters, sugars, etc. (1,2). It reflects the type and physiological and biochemical status of cells and can be used for cell identification. Therefore, RS analysis of biological samples has the characteristics of high identification accuracy and high accuracy. However, there is still limited research on the use of RS for glioma analysis. This study first used RS to compare the differences between multiple glioblastoma cell lines and astrocyte cell lines. Then, based on the research results of cell lines, the feasibility of RS in rapid diagnosis of glioblastoma was verified between clinical glioblastoma specimens and normal brain tissue. We present this article in accordance with the MDAR reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0705/rc).

Methods

Sample preparations

Cell lines

Three glioma cell lines, U87 (Warner Bio, Hefei, China, WN-10206), LN229 (Warner Bio, LN229), T98g (Warner Bio, WN-10180), and one normal human astrocyte cell line HA1800 (Warner Bio, WN-10737), were routinely cultured in culturing solution in 5% CO₂ at 37 ℃ for 48 h. After culturing in cell culture flasks, cells were removed from the flask surface using 0.25% trypsin-EDTA (Gibco Company, New York, USA) and washed three times in phosphate-buffered saline (PBS; Gibco Company, USA). Subsequently, the resuspended cells were collected by centrifugation and fixed with 4% paraformaldehyde in sterile deionized water. For RS measurements, the cells were mounted on coverslips and air-dried.

Tissue specimen

The glioma tumor tissues were harvested from three patients diagnosed as glioblastoma. The human brain tissue was from a patient with severe traumatic brain injury, whose brain tissue suffered severe contusion and laceration, accompanied by brain swelling, and we had to remove some of the damaged brain to save the patient. Fresh clinical specimens are stored in formalin and sent to the laboratory. A portion is frozen and sliced, and the slices are stored in a −20 ℃ freezer. Before starting RS data acquisition, take out the sample slices from the refrigerator and allow them to reach room temperature for approximately 5 minutes to evaporate the water vapor on the slices. Then, the sample-slice was detected under a Raman microscope for data collection. Before Raman signal acquisition, calibrate the instrument using a silicon wafer. The other part of the tissue is embedded in paraffin for hematoxylin and eosin (H&E) staining and observed under a regular optical microscope. Informed consent was obtained prior to obtaining clinical tissues. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of Shanghai General Hospital, Shanghai Jiaotong University (No. HEQ [2025]024).

RS collection and data pre-processing

The Raman characteristics of the sample were measured on a WITec alpha300 Raman microscope (manufactured by WITec GmbH, UIm, Germany) 532 nm laser, grating: 1,200 g/mm, spectral range 279–2,187 cm⁻¹, spectrometer center: 1,300 (rel. 1/cm), laser power: 15 mw, integration time: 8 s, 100× objective lens, spot size 360 nm. Before collection, calibrate the instrument using a silicon wafer and adjust the silicon wafer signal to 520.7 cm⁻¹. At least 100 valid single spectra were collected in the designated area for each tissue, and 15 cells were recorded for each cell line. RS analysis was conducted using Labspec 6 (Horiba, Japan), followed by routine data analysis, including cosmic ray removal, polynomial curve fitting for background subtraction, and normalization (area). The normalization of each spectrum was performed by referencing the integrated area under the curve, ranging from 300 to 1,800 cm⁻¹. Subsequently, machine learning training analysis was conducted using the preprocessed data.

The software R 3.6.3 performs principal component analysis (PCA) and linear discriminant analysis (LDA) analysis on two sets of samples. (I) PCA analysis chart: Using R 3.6.3, call two packages, Facto Mine R and factoextra, use PCA function to perform PCA analysis on the intercepted graph, and use fviz_pca_ind function to draw the PCA analysis chart, with Dim1 as the horizontal axis and Dim2 as the vertical axis. (II) LDA analysis graph: Using R 3.6.3, call the MASS and care packages, use the lda function to perform LDA analysis on the truncated graph, and use the gglot function to plot LD1 and LD2. (III) Peak intensity plot: Use R 3.6.3 to obtain the mean and standard deviation of each group of samples, call ggplot2 and RColorBrewer packages, and plot using the gglot function. The mean of the graph is represented by solid lines, and the standard deviation is represented by shadows. Use machine learning support vector machine (SVM) algorithm for data analysis. Take 80% of the data from each group as the training set, and the remaining 20% as the testing set. Calculate the sensitivity, specificity, and accuracy of two sets of data.

Statistical analysis

Data are presented as mean ± standard error of the mean (SEM), with each value derived from at least three biological replicate experiments. Student’s t-test was used to analyze the differences between two groups. P<0.05 was considered statistically significant (*P<0.05, **P<0.01, ***P<0.001, ****P<0.0001).

Results

RS profiling of cell lines

About 4,500 RS from 600 cells were recorded, with each cell line containing spectra from 150 cells. The average RS for each cell line is shown in Figure 1A. The RS from bottom to top represent HA1800, Ln229, T98g, and U87 cells, respectively. The solid line represents the average RS for each cell line, and the shaded area indicates ±1.96 standard deviations (approximately 95% confidence interval of the spectrum).

Figure 1 Basic characteristics of RS in glioma cell lines and normal astrocyte cell lines. (A) Average RS of four cell line samples: U87, T98g, Ln229, and Ha 1800. (B) LDA visualization shows distinct clusters of normal astrocytes and glioma cells and the four cell lines can be distinguished by this method. (C) The visualization results indicate that in PCA analysis, the four cell lines cannot be distinguished by this method. a.u., arbitrary units; LDA, linear discriminant analysis; PCA, principal component analysis; RS, Raman spectroscopy.

RS possesses an inherent fingerprint region (300 to 1,800 cm⁻¹) that represents the unique Raman phenotypic information of cells or tissues. We employed LDA, a multivariate dimensionality reduction technique, to distinguish inherent differences and visualize the data in a lower dimension. Using LDA, we obtained cluster visualization results of Raman spectra in glioma cell lines and normal human astrocyte cell lines (Figure 1B). Through the LDA cluster map, we can distinguish glioma cell lines from normal astrocytes. However, the visualization results indicate that this method cannot distinguish the four cell lines in PCA (Figure 1C).

Analysis of machine learning SVM model results

SVM is a common function package based on the R language. When conducting analysis, it is generally divided into training and testing sets. Two sets of data with a total of 600 graphs are randomly grouped, with 80% of them being the training set and the remaining 20% being the testing set. The SVM model is first constructed based on 80% of the data, and then the remaining 20% is tested to obtain the final model prediction matching rate, which is the accuracy of using the model for identification. The results showed that all four cell lines could be well judged based on RS. Therefore, SVM is reliable as the basic model for later validation (Table 1).

Distinguishing normal astrocyte cells from glioma cells

The attribution of peaks refers to the labeling of statistically significant peaks in each group of samples based on their averages, and then comparing them with existing databases or publicly published literature to show if there are biological macromolecules that match the peak position. However, since the biological macromolecule library does not contain information about all molecules, the position of the peak may not necessarily be attributed. The peaks that appear this time are relatively regular, showing a certain trend of change and statistical significance. Between glioma cells and normal astrocytes, characteristic bands can be observed at various positions.

At the characteristic bands of 519, 719, 759, and 957 cm⁻¹, the RS intensities of glioma cells increases; however, they decreased at the bands of 1,266 and 1,655 cm⁻¹ comparing to normal astrocytes (Figure 2).

Figure 2 Significant peak differences in RS between human glioma cells and normal astrocyts. (A) The averaged RS intensity and RS peak positions of the glioma cells and normal astrocytes. (B) Statistical comparison of difference-peak intensities of using single-cell quantification. ****, P<0.0001. RS, Raman spectroscopy.

Localization of Raman microscope detection position on tissue slices

Slice the tissue embedded in paraffin and perform H&E staining. Then, locate the location of the tumor tissue or cells under a regular optical microscope, and find the same location under a Raman microscope to determine that the area to be observed, calculated, and analyzed under a Raman microscope is tumor tissue or cells. The comparison and positioning of the two kinds of microscopes are convenient and feasible, as shown in Figure 3.

Figure 3 The location of clinic tissue specimen for Roman microscope detection position. (A-E) The process for human brain tissues from patients with cerebral hernia caused by contusion and laceration: (A) contused brain tissue under Raman microscope (bar: 2 mm, at low magnification); (B) H&E staining on the same section as (A) under optical microscope (bar: 2 mm, at low magnification); (C) magnifying the square portion in (A) under Raman microscope (bar: 0.4 mm, at medium magnification); (D) the magnification of the square in (B), showing nerve tissue in the visible field (bar 0.1 mm at medium magnification); (E) combined with H&E staining in (D), Raman spectroscopy sampling and data collection were performed on the determined nerve tissue site. (F) Glioma tissue under Raman microscope (bar: 1 mm, low magnification). (G) H&E staining on the same section as (F) under optical microscope (bar: 1 mm, at low magnification). (H) Magnifying the square portion in (F) under Raman microscope (bar: 0.2 mm, at medium magnification). (I) The magnification of the square in (G), showing glioma tissue in the visible field (bar: 0.1 mm, at medium magnification). (J) Combined with H&E staining in (I), Raman spectroscopy sampling and collection were performed on the identified glioma enriched area. H&E, hematoxylin and eosin.

Quality control of clinic tissue specimen

After determining the scanning sites of clinical tissue specimen, each site was scanned with 100 valid single spectrums, and then computer modeling and analysis calculations were performed. The averaged RS for each site is shown in Figure 4. The RS from bottom to top represent clinical glioma and decayed human brain tissue, respectively. The solid line represents the average Raman spectra of each cell line, and the shaded area indicates ±1.96 standard deviations (approximately 95% confidence interval of the spectra).

Figure 4 The average RS for rotten human brain tissue and clinic glioma tissue. a.u., arbitrary units; RS, Raman spectroscopy.

Peak assignment and statistical analysis between clinic tissue specimen

Based on the LDA algorithm confirmed by intercellular comparison analysis, an SVM model was constructed, and then RS between clinic glioma tissues and rotten human brain tissues were compared and analyzed. The spectra peak trend between both kinds of clinic tissues also follows a certain pattern, which is consistent with the trend between cells, and has statistical significance. At the characteristic bands of 519, 719, 759, and 957 cm⁻¹, the Raman intensity of glioma tissue has increased; however, their Raman intensities decreased at the bands of 1,266 and 1,655 cm⁻¹ comparing to rotten human brain tissues (Figure 5).

Figure 5 Significant peak differences in RS between human glioma tissues and rotten human brain tissues. (A) The averaged RS intensity and RS peak positions of the glioma tissues and rotten human brain tissues. (B) Statistical comparison of difference-peak intensities basing glioma tissues and human brain tissues. **, P<0.01; ***, P<0.001; ****, P<0.0001. RS, Raman spectroscopy.

Discussion

The application of RS in medicine is becoming increasingly widespread, and there has been extensive research in the detection of microorganisms (3). The diagnosis and classification of tumors by RS have gradually become a hot topic. Zhang et al. used RS technology to classify breast cancer (4). RS also was used to distinguish nasopharyngeal cancer from normal tissue (5), and classified esophageal cancer (6). Even, RS microscopy technology can be used to distinguish and identify different stages of stem cell development (7). We have conducted research on glioma with RS technology, and have achieved consistent results in glioma cell lines and clinical specimens, indicating that RS has the potential to distinguish glioma cells from normal tissue cells. As this research continues to deepen, it may enable pathologists to break away from the so-called visual confirmation and quickly complete diagnosis through spectral information analysis to achieve Raman imaging. This can be used to establish images based on the concentration and category of biomolecules detected by RS, achieving real-time imaging during surgery. This is also an important goal of RS in neurosurgery.

But there are still many challenges to achieving this goal. One is the analysis of massive and interrelated RS data. Only biomarkers with sufficient specificity and biological significance can become the gold standard for RS diagnosis of glioma, which is the fundamental prerequisite for the application of RS on glioma diagnosis and grading. Secondly, massive data analysis requires complex and clever intelligent algorithms. Currently, PCA, LDA, SVM, and Tree boosted are widely used, but deeper machine learning is needed to reflect the interrelationships between biomolecules. It is precisely because of this that we first conducted comparative studies using multiple cell lines in vitro. Through coarse and fine extraction, layer by layer screening, we selected 6 peaks from a wide range of 300 to 1,800 cm⁻¹, and then validated them in clinical specimens of gliomas. The differential results obtained are consistent.

But what are the chemical bonds or components represented by these RS band positions? Do they conform to the biological characteristics of tumor cells? Among these 6 RS band positions (519, 719, 759 and 957 cm⁻¹, respectively), the Raman intensities in four of them increased at these characteristic bands in both glioma cells and glioma tissues than them in control groups (normal human astrocyte cell line HA1800 and human brain tissue). There are also two RS bands at 1,266 and 1,655 cm⁻¹, where the Raman intensities decreased in both glioma cells and tissues than them in control groups.

The RS bands at 519, 759, 719 and 957 cm⁻¹ mainly reflect the level of phospholipids in tissue cells, such as phosphatidylinositol (PI) and phospholipids; phosphatidylcholine (PC), sphingomyelin, phosphatidylethanolamine (PE), and cholesterol. It also reflects the abundance of tryptophan in tissue cells. PI is an important substrate of PI3K, widely involved in processes such as cell proliferation and apoptosis. In gliomas, PI3K increases in most cases and requires more substrates to play its role in promoting tumor cell proliferation (8). This can explain why the abundance of PI is higher in gliomas than in normal brain tissues or normal astrocyte cell line. During the metabolic process of glycerophospholipids, intermediate or final products, PE, PC, phosphatidylglycerol (PG) and PI were highly intensified in human tissue and cell line model of grade III and GBM (9). Similarly, the concentration of PC and PE were also elevated with increasing breast cancer grade, indicating that the glycerophospholipid synthesis rate increases with oncogenesis and tumor progression (10). PC and PE comprised the most phospholipid content in human cell membranes (11,12). When magnetic resonance imaging (MRI) scan, it can be found that the choline peak is often higher than the peritumoral tissue or normal brain tissue (13), which is consistent with our research results.

Cancers are prone to be in the status of immunosuppression, and it was found that inhibiting the metabolism of tryptophan can improve the effectiveness of tumor immunotherapy. The increased expression of indoleamine 2,3-dioxygenase [an enzyme that converts tryptophan (Trp) into kynurenine (Kyn)] leads to elevated levels of Kyn in various cancers, accompanied by reduced levels of Trp in cancers (14). Moreover, the metabolism and transformation of tryptophan have been accelerated in various cancers, as its metabolites have a pro cancer effect. These researches indicated that the content of tryptophan was lower in malignant tumors comparing with the corresponding of normal tissues, which is also the result of our research here. Additionally, the other two RS band positions at 1,266 and 1,655 cm⁻¹ predominantly confirm the existence of α-helix conformation of protein molecules. In fact, the proportion of α-helix and random coil in proteins decreases with tumor malignancy (15). Our results are consistent with those of previous studies.

So, we believe that RS has the potential to improve treatment outcomes by helping to detect and identify healthy and cancerous brain tissue, blood vessels, and even to confirm specific biomarkers. Although this is very promising, designing a Raman system for neurosurgical applications requires important technical support in both hardware implementation and data science methods, and continuous research in this area is still needed to achieve it.

Conclusions

Data collected by Raman microscopy scanning and modeling using SVM algorithm can accurately identify glioma cell lines and ordinary glial cells; and significant differences in RS band intensity between benign and malignant cells can be determined. These differences also exist in clinical specimens of glioblastoma and brain tissue specimens.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the MDAR reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0705/rc

Data Sharing Statement: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0705/dss

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0705/prf

Funding: This work was supported by Songjiang (Shanghai) District Science and Technology Key Project (No. 21SJKJGG103).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0705/coif). All authors report that this work was supported by Songjiang (Shanghai) District Science and Technology Key Project (No. 21SJKJGG103). The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of Shanghai General Hospital, Shanghai Jiaotong University (No. HEQ [2025]024). Informed consent was obtained from all participants prior to obtaining clinical tissues.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Mamede AP, Santos IP, Batista de Carvalho ALM, et al. A New Look into Cancer-A Review on the Contribution of Vibrational Spectroscopy on Early Diagnosis and Surgery Guidance. Cancers (Basel) 2021;13:5336. [Crossref] [PubMed]
2. Li X, Chen H, Zhang S, et al. Blood plasma resonance Raman spectroscopy combined with multivariate analysis for esophageal cancer detection. J Biophotonics 2021;14:e202100010. [Crossref] [PubMed]
3. Walter A, Reinicke M, Bocklitz T, et al. Raman spectroscopic detection of physiology changes in plasmid-bearing Escherichia coli with and without antibiotic treatment. Anal Bioanal Chem 2011;400:2763-73. [Crossref] [PubMed]
4. Zhang L, Li C, Peng D, et al. Raman spectroscopy and machine learning for the classification of breast cancers. Spectrochim Acta A Mol Biomol Spectrosc 2022;264:120300. [Crossref] [PubMed]
5. Li Y, Su S, Zhang Y, et al. Accuracy of Raman spectroscopy in discrimination of nasopharyngeal carcinoma from normal samples: a systematic review and meta-analysis. J Cancer Res Clin Oncol 2019;145:1811-21. [Crossref] [PubMed]
6. Huang W, Shang Q, Xiao X, et al. Raman spectroscopy and machine learning for the classification of esophageal squamous carcinoma. Spectrochim Acta A Mol Biomol Spectrosc 2022;281:121654. [Crossref] [PubMed]
7. Hsu CC, Xu J, Brinkhof B, et al. A single-cell Raman-based platform to identify developmental stages of human pluripotent stem cell-derived neurons. Proc Natl Acad Sci U S A 2020;117:18412-23. [Crossref] [PubMed]
8. Goncalves MD, Hopkins BD, Cantley LC. Phosphatidylinositol 3-Kinase, Growth Disorders, and Cancer. N Engl J Med 2018;379:2052-62. [Crossref] [PubMed]
9. Abdul Rashid K, Ibrahim K, Wong JHD, et al. Lipid Alterations in Glioma: A Systematic Review. Metabolites 2022;12:1280. [Crossref] [PubMed]
10. Beckonert O, Monnerjahn J, Bonk U, et al. Visualizing metabolic changes in breast-cancer tissue using 1H-NMR spectroscopy and self-organizing maps. NMR Biomed 2003;16:1-11. [Crossref] [PubMed]
11. Patel D, Witt SN. Ethanolamine and Phosphatidylethanolamine: Partners in Health and Disease. Oxid Med Cell Longev 2017;2017:4829180. [Crossref] [PubMed]
12. Sonkar K, Ayyappan V, Tressler CM, et al. Focus on the glycerophosphocholine pathway in choline phospholipid metabolism of cancer. NMR Biomed 2019;32:e4112. [Crossref] [PubMed]
13. McCarthy L, Verma G, Hangel G, et al. Application of 7T MRS to High-Grade Gliomas. AJNR Am J Neuroradiol 2022;43:1378-95. [Crossref] [PubMed]
14. Perez-Castro L, Garcia R, Venkateswaran N, et al. Tryptophan and its metabolites in normal physiology and cancer etiology. FEBS J 2023;290:7-27. [Crossref] [PubMed]
15. Xue M, Ye S, Ma X, et al. Single-Vesicle Infrared Nanoscopy for Noninvasive Tumor Malignancy Diagnosis. J Am Chem Soc 2022;144:20278-87. [Crossref] [PubMed]