Metabolite ratiomics: Rapid diagnosis and classification of gliomas for intraoperative application during brain surgery

Significance

Gliomas are highly infiltrative brain tumors whose molecular heterogeneity complicates diagnosis and treatment. Maximal safe surgical resection offers the best chance for durable disease control, yet tumor cells often extend beyond MRI-defined margins, making intraoperative identification of residual disease challenging. Rapid and sensitive analysis is needed to measure glioma biomarkers during surgery to guide surgical decision-making about the extent of resection and local treatment. Here, we show that ratios of endogenous metabolites in brain tissue can distinguish glioma from nontumor tissue and classify tumor subtypes based on molecular features using a simple ambient ionization mass spectrometry approach, enabling fast, clinically compatible analysis with potential to improve margin assessment and enhance patient outcomes.

Abstract

Gliomas are diffusely infiltrative, requiring accurate and sensitive diagnostic tools during surgical resection. Mutations in isocitrate dehydrogenase enzyme (IDH) alter its normal activity and result in accumulation of a unique oncometabolite, 2-hydroxyglutarate (2HG), in tumor cells. However, IDH-wild-type (IDH-wt) gliomas are more common and typically more aggressive. Here, 86 human brain samples were analyzed by syringe touch spray mass spectrometry to evaluate molecular alterations in gliomas. Tandem mass spectrometry (MS/MS) experiments were performed to minimize isobaric and isomeric interferences, and a ratiometric approach (using only signals for endogenous compounds) was employed to improve reliability of the measurements. Multiple ratios for pairs of metabolite signal intensities were found to differentiate tumor samples from nontumor tissue and they stratified glioma subtypes by IDH mutation status and by tumor grade, with several ratios showing 100% agreement with independent clinical assignments (P < 0.0001, Mann–Whitney U test). Increased carnitine relative abundances were found across all glioma subtypes. Importantly, α-aminoadipate was significantly elevated in grade 4 glioma, which may correlate with increased tumor malignancy. Grade 3 gliomas showed markedly higher 2HG abundance than grade 4 samples. In addition, a selection of ratios showed a linear trend with varying tumor cell percentage. Overall, the multiple metabolic ratio alterations indicate a high degree of heterogeneity among glioma subtypes. Furthermore, there are metabolites that can differentiate glioma from nontumor tissue and this measurement can be made by a simple ratiometric method which has potential to be used intraoperatively for margin assessment without diverging from standard of care.

Gliomas are malignant brain tumors with extremely poor 5-y survival rate (1, 2). Rapid and accurate intraoperative diagnosis is important for treatment planning, surgical guidance, and patient management as intraoperative local therapies like gliadel or CART are being utilized at the time of surgery (3–6). Patients suspected of having glioma through preoperative MRI typically undergo craniotomy, followed a few weeks later by chemo-radiation which remains the standard-of-care for brain cancer. Complete tumor resection offers the initial best chance for better long-term control; however, because gliomas are diffusely infiltrative, surgeons must balance preservation of healthy tissue with removal of malignant regions. This challenge directly affects the extent of resection (EOR) and the volume of residual tumor. Progression-free survival (PFS) and overall survival (OS) are significantly affected by residual tumor volume (7–9). Recent studies show that patients who undergo supratotal resection (SpTR), compared with those who undergo gross total resection (GTR) experience prolonged PFS and OS, and there is an increasing interest in using SpTR to reduce infiltrated residual tumor beyond MRI-defined borders (8, 10–14). Current guidelines for accurate glioma diagnosis include multiple criteria, many of which are primarily determined by genetic testing or immunohistochemical staining (IHC) which are typically available only days after the surgery (15, 16). The mutation status of isocitrate dehydrogenase (IDH) is a key diagnostic factor, as patients with IDH-mutant (IDH-mut) gliomas generally have higher OS and elevated levels of 2-hydroxyglutarate (2HG) phenotype (17, 18). In its wild-type form, IDH catalyzes the conversion of isocitrate (iCit) to α-ketoglutarate (αKG), whereas mutant IDH catalyzes the reduction of αKG to 2HG, which is normally present only at trace concentrations (19–21). Glioblastoma (GBM), the most common and aggressive subtype of glioma, falls within the IDH-wt classification (22).

Metabolic alteration is a hallmark of cancer (23, 24), and several studies have reported metabolic aberrations in brain tumors, including levels of choline (Cho), N-acetylaspartic acid (NAA), creatine (Cre), γ-amino butyric acid (GABA) (25–27). Although some studies report decreased Cre concentrations in cancer cells and suggest a tumor-suppressive role, other reports indicate that creatine uptake can support tumor growth and progression (26–30). N-guanidinoacetate (GAA) is an intermediate in creatine catabolism from breakdown of arginine (Arg) and has been reported to be increased in gliomas in several studies (30–32). Cho contributes to cell membrane structure and is involved in neurotransmitter synthesis and signaling, which is frequently elevated in glioma (26, 29, 33, 34). GABA participates in the regulation of cell proliferation, development, differentiation, and signaling (35). Studies have shown that higher-grade gliomas (HGG) are associated with lower GABA concentrations and reduced expression of GABA receptors (35–38). Carnitine (Car) is essential for transport of fatty acids into mitochondria for fatty acid oxidation (FAO). Reports have shown increased concentrations of carnitine in gliomas, consistent with its role in supporting energy metabolism in rapidly proliferating cancer cells. Moreover, inhibition of carnitine transport has been associated with apoptosis in glioma cells (25, 39, 40). α-Aminoadipate (AAA), an intermediate in lysine degradation, has been detected at high concentrations in glioma and has been correlated with tumor grade, with higher levels reported in GBMs (31, 41–43). Cystathionine (Cth), an intermediate in the transsulfuration pathway, has also been reported at increased concentrations in glioma (36, 37). Given the heterogeneity of glioma, investigation of metabolic aberrations can provide valuable insights for discriminating between cancerous and noncancerous tissue and improving diagnostic accuracy. Such metabolic discriminators could support intraoperative decision-making, helping to maximize tumor resection while minimizing neurological deficits. As discoveries and implementation of precision medicine for local therapies at the time of surgery become more readily available there can be significant impact of taking advantage of the “window of opportunity” and avoiding delays of even a few weeks in treatment.

IHC and genetic tests, currently used to diagnose glioma at the molecular level, are not compatible with intraoperative decision-making because results are typically not available during surgery. Microscopic evaluation of hematoxylin and eosin (H&E) stained tissue provides primarily morphological rather than molecular information. The unmet need for rapid molecular assessment has driven the development of techniques capable of delivering molecular information at the time of surgery. Magnetic resonance spectroscopy (MRS), fluorescence, Raman spectroscopy, and infrared spectroscopy have all been explored for intraoperative tissue analysis (44–47). In addition, imaging modalities have been developed for both preoperative and intraoperative applications (48, 49). Due to its high sensitivity and versatility, mass spectrometry (MS) has been used in clinical chemistry, and more recently, with the advent of ambient ionization mass spectrometry (AIMS) (50), it has attracted substantial attention in clinical oncology and intraoperative cancer diagnosis (51–55). Gas chromatography (GC) and liquid chromatography (LC) coupled to MS are routinely used for a variety of applications in clinical chemistry, including therapeutic drug monitoring and biomarker analysis (53). However, GC- and LC-MS require multiple sample preparation steps and the time needed for chromatographic separation generally precludes their use in intraoperative settings. The development of AIMS techniques (50, 56–58) has accelerated the analysis of biospecimens and opened the door to introducing MS into the operating room (OR). AIMS enables qualitative and quantitative analysis of untreated or minimally processed samples in complex matrices, including biological tissues and fluids (59, 60). To minimize isobaric, and in some cases isomeric, interferences from the matrix, tandem mass spectrometry (MS/MS) can be employed in conjunction with AIMS. Several studies have used AIMS methods, such as desorption electrospray ionization (DESI), MasSpec Pen, rapid evaporative ionization mass spectrometry (REIMS), and picosecond infrared laser mass spectrometry (PIRL-MS) to analyze tissue specimens (61–65). Touch spray (TS) is an ambient ionization method in which an absorbent substrate is used to collect a sample, followed by solvent addition and application of high voltage (HV) to generate ions from sprayed droplets. Due to its simplicity and ease of use, multiple TS-MS and related substrate-spray approaches have been developed, including paper spray (PS), swab TS, and probe electrospray ionization (PESI) (66–74). In PS and swab TS, the substrate is required to have a macroscopically sharp tip to facilitate formation of a Taylor cone and stable droplet ejection (69, 70). Although many of these techniques have been applied to biological samples, only a few, including swab TS and needle-biopsy spray, have been used for brain assessments (66, 68). Extraction nano electrospray ionization has also been explored for intraoperative glioma study and can offer higher extraction and ionization efficiencies (75). In this approach, a thin paper strip is used to collect the sample and it is then transferred into a pulled glass capillary prefilled with solvent; application of a high electric field ionizes the extracted analytes. Although extraction nano electrospray ionization is simple and efficient, tip clogging is frequent and the method often requires trained personnel, making it less ideal for routine clinical and intraoperative use. Beyond high analytical sensitivity and speed, intraoperative AIMS methods must meet additional practical requirements, including ease of use by non–mass spectrometry experts, sterilizability for in vivo sampling, minimal invasiveness, and disposability. These considerations should guide the development of AIMS methods and the refinement of existing techniques.

In this study, we extend earlier work on brain cancer analysis based on the oncometabolite 2HG in two important ways: First, we measure a substantial number of metabolites (using ratios of ion intensities to cancel out biological and analytical variables other than intrinsic metabolite production) and second, we show that these measurements should be applicable to determining tumor margins during brain cancer surgery. A total of 86 human brain specimens, comprising 78 gliomas and 8 controls, was analyzed by syringe touch spray (STS) using tandem MS/MS measurements to assess molecular aberrations associated with diseased tissue (Fig. 1). Once the method was optimized, brain specimens were analyzed in both positive and negative ion modes to investigate molecular changes in glioma relative to control tissue. Several small molecules were targeted, and MS/MS was performed to record product ion spectra of each metabolite. Following STS mass spectrometry (STS-MS) analysis, data from the positive and negative modes were integrated, and ion ratios were calculated for pairs of metabolites. Statistically significant features were then selected and evaluated for their ability to discriminate glioma from nondiseased tissue. Data were further stratified by IDH mutation status and tumor grade. A subset of IDH-mut samples was used to evaluate metabolic ratios across varying tumor cell percentages (TCP) to assess the applicability of the STS-MS/MS method for margin assessment within the tumor cavity. Use of a ratiometric approach makes the measurements robust against signal fluctuations without the need for an internal standard, making it particularly well suited for intraoperative applications where accuracy and speed are most important.

Fig. 1.

The STS-MS workflow. (A) Schematic diagram of STS-MS sampling and analysis. Sampling is performed by briefly touching the tissue to the substrate for 2 to 3 s. The substrate-plunger is then reinserted into the syringe barrel, solvent is added, and application of a voltage (HV) initiates ionization for MS analysis. (B) Images of the substrate before and after contacting the sample, showing absorption of tissue residue. (C) Syringe with fixed needle is filled with solvent following sampling. (D) Upon application of HV to the needle, a Taylor cone is formed and ions are generated for MS measurement.

Results

In this study, 86 human brain tissue samples were analyzed by STS-MS/MS. In positive ion mode, ions at m/z 104 (choline and protonated GABA), m/z 118 (protonated valine and GAA), m/z 132 (protonated Cre and Leu), m/z 146 (protonated spermidine), m/z 156 (protonated His), m/z 162 (protonated Car and AAA), m/z 170 (potassium adduct of Cre), m/z 175 (protonated Arg), and m/z 223 (protonated Cth) were subjected to CID and the corresponding product-ion MS/MS spectra were acquired. In negative ion mode, MS/MS measurements included ions at m/z 146 (deprotonated Glu), m/z 147 (deprotonated 2HG), m/z 173 (deprotonated Arg), m/z 174 (deprotonated NAA and citrulline), m/z 191 (deprotonated Cit and iCit), and m/z 221 (deprotonated Cth). Product-ion spectra for authentic standards, along with representative spectra from brain samples for the ions listed above, are provided in SI Appendix, Figs. S1–S4. Metabolic ratios were then evaluated and compared across glioma and control samples, IDH-mut and IDH-wt tumors, and tumor grades. Binary comparisons were performed between glioma and control, IDH-wt and control, IDH-mut and control, IDH-wt and IDH-mut, grade 2 and grade 4, and grade 3 and grade 4. The most significant classifiers are listed in SI Appendix, Table S4, along with their statistical significance and clinical performance. To assess cutoff robustness for the selected ratios, patient-level bootstrap optimization correction was performed; the ratios retained high corrected area under the curves (AUCs) and balanced accuracies across the evaluated comparisons, supporting robust internal performance within the cohort (SI Appendix, Table S5).

Comparing Glioma Samples with Control Samples Using STS-MS.

For the most reliable assessment, discovery of the most significant ratios was limited to tumor-core tissue samples. After identifying the most discriminative ratios (Wilcoxon rank-sum test), data from glioma samples were also evaluated. The ratios that best discriminate between 37 glioma samples (only core, all subtypes combined) and the eight control samples are shown in SI Appendix, Fig. S5 as box plots with median lines and whiskers extending to ±1.5 times the interquartile range (IQR). SI Appendix, Fig. S5A shows a decrease in NAA/Car ratio in glioma compared with controls, with clear separation between groups (P < 0.0001, Mann–Whitney U test). ROC analysis yielded an optimal cutoff for log(NAA/Car) of −0.222 (Youden index) and an AUC of 0.997 (SI Appendix, Table S4). A confusion matrix constructed using this threshold resulted in 97% sensitivity, 100% specificity, and 98% accuracy when STS-MS predictions were compared with clinical diagnoses (SI Appendix, Table S4). The log(Car/Glu) ratio was elevated in glioma samples compared to control samples (SI Appendix, Fig. S5B). While Cho/GABA ratios were higher in glioma, GABA/Car was lower in these samples (SI Appendix, Fig. S5 C and D). Discrimination became slightly weaker when core and margin samples were combined (SI Appendix, Fig. S6 A–D).

Discrimination of IDH-wt Samples from Control Samples Using STS-MS.

A total of 24 IDH-wt samples (tumor core) and eight nontumor samples (controls) were analyzed by STS-MS. The NAA/Car ratio was lower in IDH-wt samples and clearly distinguished IDH-wt from controls (P < 0.0001), with an AUC of 1.00 and 100% classification accuracy (Fig. 2A and SI Appendix, Table S4). Similarly, Glu/Car, NAA/AAA, and GABA/Car ratios were lower in IDH-wt samples and showed perfect separation between the two groups (P < 0.0001, Fig. 2 B–D) with 100% correct classification (see SI Appendix, Table S4 for more information). IDH-wt samples also showed reduced NAA/Cho values compared with control samples, which is shown in Fig. 2E (P < 0.0001). AAA/Cth was elevated in IDH-wt glioma compared with control samples (P < 0.0001; Fig. 2F), and at the optimal cutoff, 100% accuracy was achieved. SI Appendix, Fig. S7 shows box plots of the same ratios comparing combined core and margin glioma samples with controls. As expected, greater heterogeneity at the margins and variable tumor infiltration resulted in slightly poorer separation. Overall, several metabolic ratios demonstrated strong discrimination between IDH-wt glioma and noncancerous tissue, supporting the potential use of STS-MS/MS to guide surgical resection.

Fig. 2.

Box plots showing metabolite ratio differences between IDH-wt glioma (core) and control (nontumor) tissue. The fragment ions used for discrimination are represented according to acquisition polarity. Note that some of the ratios use data on ions of opposite polarities. (A) NAA/Car, (B) Glu/Car, (C) NAA/AAA, (D) GABA/Car, (E) NAA/Cho, (F) AAA/Cth.

Discrimination of IDH-mut Samples from Control Samples.

IDH-mut gliomas accumulate 2HG, making it a robust surrogate marker of cells harboring IDH mutations. Our previous studies demonstrated the use of the 2HG/Glu ratio to predict IDH mutation status (76). However, given the heterogeneity of gliomas, it is important to investigate additional metabolic ratios alongside the conventional 2HG/Glu approach. STS-MS data included 13 IDH-mut glioma samples (core tissue) and eight control tissues. As shown in SI Appendix, Fig. S8, nine classifiers perfectly differentiated IDH-mut glioma from controls with 100% agreement with clinical diagnoses (SI Appendix, Table S4); six of these ratios included 2HG in the metabolite pair (SI Appendix, Fig. S8 A–F) but Cho/GABA, Cho/Glu, and Car/Glu were also higher in IDH-mut glioma than in control samples (SI Appendix, Fig. S8 G–I). Including margin samples in the ratiomics analysis reduced discrimination (SI Appendix, Fig. S9 A–I), consistent with greater heterogeneity at the margins.

IDH-mut vs. IDH-wt Glioma.

IDH-mut glioma tissue samples were readily differentiated from IDH-wt samples when 2HG was included in the metabolite pair. The ratios 2HG/Glu, 2HG/Cho, Car/2HG, and AAA/2HG achieved 100% correct classification (SI Appendix, Fig. S10 A–D and Table S4). When core and margin data were combined, slight overlap was observed in SI Appendix, Fig. S11 box plots.

Comparing Grade 2 vs. Grade 4 and Grade 3 vs. Grade 4 Samples.

Knowledge of tumor grade is valuable because it can influence treatment planning and the extent of resection. Tumor grading is commonly determined by microscopic evaluation of stained tissue. Here, molecular features obtained by STS-MS were evaluated for their association with tumor grade to support additional glioma subtyping. No strongly discriminative ratios were observed between grade 2 and grade 3 gliomas; therefore, this comparison is not discussed further.

A total of 20 grade 4 and 8 grade 2 glioma tissue samples were analyzed by STS-MS to identify molecular features that differentiate these grades. Several ratios showed strong discriminatory power. In particular, AAA/Cre, Arg/Car, and His/AAA provided perfect separation between grade 2 and grade 4 samples (P < 0.0001), with AUCs of 1.00 and 100% agreement when compared with clinical diagnoses (Fig. 3 A–C and SI Appendix, Table S4).

Fig. 3.

Box plots showing separation of grade 4 from grade 2 (A–C) and grade 3 (D–F) glioma tumors based on STS-MS results: (A) AAA/Cre, (B) Arg/Car, (C) His/AAA, (D) 2HG/Glu (E) AAA/2HG, and (F) Car/Cho.

Seven grade 3 and 20 grade 4 glioma tissue samples were classified using STS-MS. Six ratios provided complete separation between grade 3 and 4 (Fig. 3 D–F and SI Appendix, Fig. S12 A–C), achieving 100% classification accuracy. These ratios were 2HG/Glu, AAA/2HG, Car/Cho, 2HG/iCit, AAA/Cre, and 2HG/GAA (SI Appendix, Table S4). Grade 3 gliomas showed higher 2HG, whereas grade 4 samples showed higher AAA. Several ratios included 2HG in the metabolite pair and they might therefore be influenced by IDH mutation status; however, ratios without 2HG (Car/Cho and AAA/Cre) also showed excellent discrimination between grade 3 and grade 4 glioma samples. Collectively, these results demonstrate the potential of STS-MS to predict tumor grade based on molecular features, and simultaneously distinguish tumor from control.

Correlation of TCP to Metabolic Ratios: Potential Application for Margin Assessment.

So far, this study has focused on identifying discriminative features between glioma subtypes and between glioma and noncancerous tissue using tumor core specimens. However, STS-MS is intended as a clinically compatible intraoperative analytical tool, particularly for margin assessment (Fig. 4A). Therefore, 2HG, the best-studied biomarker, was selected to evaluate ratiometric changes as TCP varied. Seven IDH-mut samples were mixed with noncancerous control tissue to generate mixtures with estimated TCP values ranging from 0 to 100%, based on the mixing ratios. Thus, the TCP values used in this analysis were not pathologist assigned values. The corresponding 2HG/Glu were recorded (SI Appendix, Fig. S13 A–D). A linear increase in log(2HG/Glu) was observed with increasing TCP, with higher 2HG/Glu values corresponding to higher TCP. In all curves, dashed lines indicate the optimal cutoff value that identifies IDH mutation derived from tumor-core analysis. Another ratio, Glu/Car, which differentiated all glioma subtypes, was investigated across TCP (Fig. 4 B and C) and also showed a linear relationship. Because the 2HG/Glu and Glu/Car ratios at 0% TCP were very low and high, respectively, and fell outside the linear range, we focused on the linear portion of the data. In two IDH-wt cases, Glu/Car ratios in margin samples tended to be closer to the cutoff than those in core samples (Fig. 4 D and E). This suggests that cutoff values established using tumor-core tissue may not directly translate to margin assessment, where tumor heterogeneity is typically greater, consistent with our previous study (61). Accordingly, margin evaluation requires careful consideration to ensure accurate interpretation. Collectively, these results highlight the potential of STS-MS for intraoperative margin assessment.

Fig. 4.

(A) Application of TST-MS for intraoperative margin assessment. (B and C) Glu/Car vs. TCP in glioma samples show an approximately linear relationship. (D and E) Glu/Car ratios in margin tissues tend to be less consistent and closer to the cutoff value.

Discussion

Both the extent of brain cancer resection and accurate and fast intraoperative molecular diagnosis have important clinical relevance for patient outcomes, therapies, and ultimately survival and quality of life. In this study, STS-MS/MS was used to investigate discriminative metabolic ratios in glioma subtypes vs. nontumor tissues. Performing MS/MS reduces matrix-related interferences and improves measurement reliability. While several isomeric and isobaric species could not be distinguished using a unit-resolution mass scan, they exhibited distinct fragmentation patterns in product ion MS/MS scans. These included GABA and Cho at m/z 104, GAA and valine at m/z 118, Cre and Leu at m/z 132, and Car and AAA at m/z 162 in positive ion mode, as well as NAA and citrulline at m/z 174, and iCit /citrate at m/z 191 in the negative ion mode. A high-resolution mass spectrometer operated in conventional single stage mode is not expected to improve on the unit resolution MS/MS performance achieved in this study because isomeric ions will not be discriminated.

Using a ratiometric approach, normalizing signals to endogenous compounds, minimizes signal fluctuations and nonbiological sample variability. The ratiometric approach has been used in MRS for brain cancer diagnosis, and parallel advantages should accrue to mass spectrometry which offers higher analytical sensitivity and molecular specificity, particularly useful for margin assessments. In the STS-MS ratiometric results, glioma samples showed higher Cho/GABA ratios, which may reflect downregulation of GABA receptors or increased expression of succinic semialdehyde dehydrogenase (SSADH), a key enzyme in GABA oxidation, in highly proliferative glioma cells (77, 78). In addition, elevated Cho has been reported in glioma tumor tissue relative to normal brain parenchyma, and Cho levels may predict malignancy grade and survival (26). Higher Car/Glu ratios observed in glioma samples (including both IDH-wt and IDH-mut) may reflect reduced glutamate abundance, particularly in IDH-mut tumors, due to enhanced glutaminolysis (glutamine → Glu → αKG → 2HG) (77, 79). Reduced intracellular Glu may also result from downregulation of excitatory amino acid transporters (EAATs) and upregulation of the cystine/glutamate exchanger (SXC) in glioma cells (80). Increased Car levels may be associated with overexpression of carnitine palmitoyltransferase 1 (CPT1) and the organic cation/carnitine transporter (OCTN2), supporting fatty acid oxidation and energy metabolism in glioma (39, 40, 81). Moreover, several studies report significantly reduced NAA concentrations in glioma. Although the underlying biology remains unclear, this reduction may reflect decreased neuronal density. (25, 82–84). Consistent with our previous reports, 2HG/Glu provided clear separation between IDH-mut and control samples and between IDH-mut and IDH-wt tumors, reflecting the neomorphic gain of function of mutant IDH that drives 2HG accumulation (60, 61). Importantly, several less-studied metabolites demonstrated discriminatory ratios. Cth, an intermediate in the transsulfuration pathway, has been reported to be upregulated in glioma, particularly in IDH-mut tumors with 1p/19q codeletions (36, 37); however, Cth has not been investigated in IDH-wt glioma previously. When comparing grade 2 and grade 4 gliomas, the finding indicates that AAA is a highly informative marker of HGG, correlating with tumor aggressiveness, a result that is consistent with prior reports (31, 41–43). In addition, comparison of grade 3 and grade 4 glioma samples identified 2HG-containing ratios as the strongest discriminators, yielding perfect separation between the two groups. Reduced 2HG/GAA ratios were also observed in grade 4 relative to grade 3, consistent with reports of unexpectedly high GAA expression in glioma (30–32). Previous studies have reported reduced citrulline abundance in glioma. Notably, both GAA and citrulline are linked to arginine (Arg) metabolism (30–32). Despite the complexity of the metabolome, STS-MS demonstrated strong potential for discriminating glioma subtypes from control tissue and for distinguishing among glioma subtypes.

The STS-MS technique is highly promising for intraoperative delineation of tumor boundaries and for tumor subtyping based on IDH mutation status and grade. STS-MS integrates sampling, extraction, and ionization into a simple workflow, while producing MS data of comparable quality to that obtained using standard nano-electrospray ionization (nESI)-MS (SI Appendix, Fig. S14). With the use of surgically approved materials, the feasibility of in vivo sampling could be explored, potentially further enhancing the value of STS-MS assessments during brain surgery. The estimated turnaround time for STS-MS analysis is approximately 3 min. Although margin assessment would require additional time, the substantial clinical value of this information may justify the added time.

To demonstrate the feasibility of STS-MS for margin assessment, IDH-mut tumor tissue was mixed with normal human brain tissue at defined proportions to mimic specimens with varying TCPs. IDH-mut tumors were selected because they harbor a well-characterized mutation with a robust surrogate marker, 2HG. This approach could also be extended beyond gliomas to other tumors harboring IDH mutations. Our previous intraoperative study showed that biopsies resected from tumor margins can be classified as either positive or negative depending on the cutoff derived from tumor-core tissue (61). However, because margins are highly heterogeneous, distinguishing infiltrated tissue from normal tissue using a threshold established from core assessments is challenging. Therefore, understanding how TCP relates to metabolic ratios can provide valuable insight. Across varying TCPs, log(2HG/Glu) and log(Glu/Car) showed a linear correlation with TCP. When the core-derived cutoff was overlaid on these plots, several notable patterns emerged. IDH-mut samples with high maximum 2HG/Glu ratios (at 100% TCP) had fewer mixtures falling below the cutoff, whereas samples with lower maximum 2HG/Glu ratios at 100% TCP had a substantial fraction of 2HG/Glu values below the cutoff. Notably, even mixtures corresponding to ~80% TCP could fall below the threshold and be classified as negative, representing potential false-negative margin assessments. A similar trend was also observed in Glu/Car ratios. When comparing Glu/Car ratios, margin samples showed higher heterogeneity and values closer to cutoff. These observations suggest that metabolite ratios could support margin assessment even when values fall below the classification threshold and may also enable estimation of TCP.

Among the 24,492 computed ratios, many differed across tissue types; however, only the most discriminative ratios are reported here. Use of multiple metabolic ratios can improve measurement robustness, particularly given the heterogeneous nature of gliomas. Importantly, this strategy may provide added value for IDH-wild-type gliomas, for which 2HG detection is not informative and rapid intraoperative molecular assessment remains challenging. As larger cohorts become available, the metabolite ratios could be incorporated into multivariate statistical models for faster and more comprehensive intraoperative tumor assessment. It should be noted also that the cutoff values assigned in this study should not be considered definitive clinical decision limits. The reported performance parameters reflect agreement with the gold-standard diagnoses within this proof-of-concept cohort. Overall, these results demonstrate the applicability of STS-MS/MS for rapid and straightforward tissue analysis for intraoperative tumor diagnosis and subtyping. Multiple metabolite ion ratios differentiated tumor grades and glioma subtypes from control tissue. The information could support surgical decision-making by helping maximize tumor resection while minimizing neurological deficits and enabling targeted local therapies, ultimately improving patient outcomes.

Conclusions

Here, a total of 86 human brain samples were analyzed by STS-MS/MS. Multiple metabolite ratios enabled classification of glioma vs. control tissue, discrimination of IDH-mut from IDH-wt tumors, and stratification of gliomas by grade. MS/MS was used to reduce matrix interferences and, where possible, to distinguish isobaric and isomeric ions that cannot be resolved with unit-resolution mass spectrometers. Metabolic ratios were employed to minimize signal fluctuations and sample collection differences. Glioma samples showed distinct metabolite ratios when stratified by subtype and when compared with control brain samples. Variation in TCP relative to log(metabolite ratios) showed a linear correlation, supporting the potential of STS-MS for margin assessment even when margin samples showed high heterogeneity. Overall, STS-MS demonstrated the ability to subtype gliomas and to discriminate tumor from nontumor tissue using endogenous molecular features. Future studies using a larger, independent sample cohort, ideally in an intraoperative setting, are needed to support the clinical translation of STS-MS as an intraoperative diagnostic tool for providing complementary information during tumor resection, particularly at the margins. One could also note that a more sophisticated use of machine learning would likely improve this ratiomics approach. The resulting molecular information could help guide surgical resection, and inform treatment planning in the context of personalized medicine.

Materials and Methods

Tissue Specimens.

Tissue samples included 75 fresh-frozen intraoperative brain biopsies provided by Mayo Clinic to Purdue University (85). Sample collection and analysis were conducted under Mayo Clinic Institutional Review Board (IRB) protocol No. 19-010725 and Purdue University IRB No. 2020-516 (exempt) as part of the study protocol. The Purdue IRB specifically authorizes the authors’ analysis by mass spectrometry of these Mayo Clinic tissue samples. Analysis of the four additional brain tissue samples purchased from the Biorepository of the Methodist Research Institute is authorized under Purdue University IRB No. 1203011967 (exempt), while that of the seven IDH-mutant glioma samples obtained from the Cooperative Human Tissue Network is authorized under Purdue University IRB No. 2024-1417 (exempt). All samples were deidentified prior to shipment and stored at −80 °C until MS analysis.

Clinical Diagnosis.

IHC and/or PCR were performed to determine glioma subtype and IDH mutation status in accordance with the fifth edition of the World Health Organization (WHO) Classification of Tumors of the Central Nervous System. Clinical information was provided by Mayo Clinic, the Biorepository of the Methodist Research Institute, and the Cooperative Human Tissue Network. Details of the tissue samples are provided in SI Appendix, Tables S1 and S2.

STS and MS Analysis.

Syringe touch spray was performed using a sterile insulin syringe with fixed needle [Sol-Millenium, 31G (0.25 mm × 8 mm)]. Whatman 1 filter paper was laser-cut into a circular disk with a diameter of 2.8 mm. Paper disks were washed with water-MeOH (1:1) and then dried. The syringe plunger was removed and the filter paper was affixed to the plunger sealing. After brain samples had been thawed at room temperature, the substrate was brought into contact with the tissue for 3 s and then allowed to dry. Each brain biopsy was sampled using three individual syringes. Once the substrate had dried, the plunger was reinserted into the syringe barrel and loaded with 50 μL of a solvent mixture containing MeOH-water-formic acid (95:5:0.1). The syringe was mounted on a syringe pump (Harvard Apparatus, Pump 11 Elite) and operated at a flow rate of 2 μL min⁻¹, positioned 17 mm from mass spectrometer inlet. An HV was applied to the needle to generate droplets and form ions (4.0 kV in the positive and 3.9 kV in the negative modes). The reproducibility of STS-MS measurements is reported in SI Appendix, Tables S6 and S7.

To correlate TCP with changes in metabolic ratios, seven IDH-mut samples and one control sample (normal autopsied brain) were homogenized using ceramic bead tubes (2 mL kit, 1.4 mm ceramic beads; Fisherbrand), followed by addition of 80% MeOH. Tumor homogenates were mixed with the control homogenate to generate mixtures spanning 0 to 100% TCP. The mixtures were sampled using the syringe-based tool and analyzed by STS-MS.

An initial list of metabolites was prepared, and collision-induced dissociation (CID) experiments were performed for the metabolites of interest using optimized collision energies. Product-ion MS/MS spectra were acquired by averaging spectra for each precursor ion for 45 s using a linear ion trap (Finnigan LTQ, Thermo Fisher Scientific). The finalized metabolite list is provided in SI Appendix, Table S3. MS experiments were performed three times in both positive and negative ion modes, and for each sample using a different syringe each time. Metabolite identification and confirmation were performed using high-resolution mass spectrometry (Thermo Orbitrap) together with MS/MS spectra acquired from authentic standards. The time required for each experimental step, along with the estimated total turnaround time, is provided in SI Appendix, Table S8.

Data Processing.

For each sample, product-ion spectra were recorded for each precursor ion of interest in the form of a single .raw file. For data processing, the .raw files were converted to mzXML format using MSConvertGUI (ProteoWizard). Custom MATLAB scripts were used to extract each product-ion scan, filter scans using a blank-based intensity cutoff (mean + 3 SD), and average the remaining scans. Signals corresponding to known and unknown compounds were then selected from each spectrum, and data from the positive and negative ion modes were combined. Metabolic ratios were calculated iteratively by dividing each selected peak by every other selected peak, yielding 24,492 ratios. For binary comparisons, Mann–Whitney U tests (P < 0.05 considered significant; STATA SE 19) were performed on the averaged log-transformed ratios to identify the most discriminative features between groups. Because STS-MS is a semiquantitative technique, a ratiometric approach helps minimize signal fluctuations and sample variations. To assess IDH mutation status, the intensity of 2HG fragment ions was measured relative to glutamate (Glu), as previously reported (60).

Receiver operating characteristic (ROC) curves, areas under the curve (AUC), and optimal cutoff values (based on the Youden index) for IDH mutation status determination were obtained by analyzing the processed MS data in STATA SE 19. Classification performance was evaluated using sensitivity, specificity, and accuracy calculated from the confusion matrix comparing STS-MS predictions with clinical diagnoses.

Patient-level bootstrap optimization correction was performed using 1,000 bootstrap iterations. Samples were aggregated using the median ratio value. In each bootstrap iteration, the data were sampled with replacement, the cutoff was optimized using the Youden index, and optimism-corrected AUC, sensitivity, specificity, and balanced accuracy were calculated.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

The datasets analyzed during this study are available in the Purdue University Research Repository (PURR), https://purr.purdue.edu/publications/5097/1 (86).