A pilot study on metabolomic characterization of human glioblastomas and patient plasma

Purpose To determine whether recurrent GBMs are metabolically distinct from primary GBM, and whether patient plasma can be used as a liquid biopsy to reflect this difference. Methods In a single center cohort study, tissue and blood samples from 15 patients with glioblastoma (9 glioblastoma tissues at diagnosis, 3 pairs of tissue, and 6 pairs of plasma specimens at diagnosis and at recurrence) were analyzed. Results Several metabolites had significant alternations in both tumor and plasma specimens. In the tissue, the following representative metabolites had a significant increase in peak intensity at recurrence compared to diagnosis: N-alpha-methylhistamine (p = 0.037), glycerol-3-phosphate (p = 0.029), phosphocholine (p = 0.045), and succinic acid (p = 0.025). In patient plasma, metabolites that significantly increased at recurrence included: 2,4-difluorotoluene (p = 0.031), diatrizoic acid (p = 0.032), indole-3-acetate with (p = 0.029), urea (P = 0.025), pseudouridine (p = 0.042), and maltose (p = 0.035). Metabolites that significantly decreased in plasma at recurrence were: eicosenoic acid (p = 0.017), glucose-1-phosphate (p = 0.017), FA 18:2 (linoleic acid) (p = 0.017), arginine (p = 0.036), fatty acids 20:3 (homo-gamma-linolenic acid (p = 0.036), galactosamine (p = 0.007), and FA 18:3 (linolenic acid) (P = 0.012). Principal component analysis showed that the metabolomic profiles differ between tumor tissue and patient plasma. Conclusions Our data suggest that metabolomic profiles of human GBM tissue and patient plasma differ at diagnosis and at recurrence. Many metabolites involved in tumorigenesis and metabolomic flexibility were identified. A larger study using targeted metabolomic assay is warranted to measure the levels of these metabolites, which will help identify the metabolomic signatures in both GBM tissue and patient plasma for risk stratification, clinical outcome prediction, and development of new adjuvant metabolomic-targeting therapy.


Introduction
Glioblastoma (GBM) is a fatal tumor with a median survival of less than 2 years. [1] Most GBMs respond to initial therapeutic interventions of surgical resection and chemoradiation [2,3] but eventually, all patients will relapse. The range of progression free survival is wide, from a few months to more than 2 years. Understanding the differences in molecular features of GBM and its microenvironment at diagnosis and at recurrence can help identify biomarkers to monitor for treatment response, understand pathogenesis of treatment resistance, and predict outcome. It may also help identifying new therapeutic targets.
Surgical resection and standard chemoradiation improve survival but this initial effect is limited by the development of resistance. [4] While therapy resistance and the aggressiveness of GBM have been investigated at the genomic and transcriptomic levels, less is known about the metabolic phenotypes.
Altered metabolism is a hallmark of cancer. [5] Metabolomic changes are critical for tumor cells to undergo conversion to aggressive and treatment-resistant phenotypes.
[6] Recurrent, therapy resistant tumors develop within the high dose radiation eld, and the ability of recurrent tumors to resist therapy is, in part, due to metabolomic alternation within the tumor [7]. Tumor metabolism is in uenced by both cancer cell-intrinsic information (genome, epigenome, proteome, post-translational modi cations) and cell-extrinsic cues from the tumor microenvironment.
Targeting the metabolome has succeeded in a number of cancers including high-grade gliomas. [7] Glucose uptake can inform prognosis in a variety of cancers including glioma.
[8] Metabolomic pro ling in GBM tissue may provide important information on the differences in tumor responses to initial standard of care therapies and for understanding the differences between tumors at diagnosis and at recurrence. This may help predict tumor aggressiveness and patient prognosis. Early data from several metabolomic studies in small cohorts of GBM patients have yielded promising results. [9,10] GBM is a heterotrophic tumor but is also known to have highly heterogeneous lipid metabolism [11] and favors heterotrophy [12,13]. However, the comprehensive metabolomic pro le including metabolites, biogenic amines, and lipidomes have not been studied. The comparison between GBM tissue and patient blood specimens has not been performed. Therefore, it is unclear if patient blood can be used as a surrogate to predict tumor status in the brain.
We performed a study to test the hypothesis that recurrent GBMs are metabolically distinct from GBM at initial diagnosis, and patient plasma can be used as a liquid biopsy to re ect this difference. Using untargeted mass spectrometry, we pro led the metabolomes, lipidomes, and biogenic amines of human glioblastoma tissue and patient plasma both at diagnosis and at recurrence, and correlated metabolomic information with clinical data. We identi ed patterns of metabolomic remodeling in tumor tissue and patient plasma. These changes can pave the way for metabolomic signature identi cation for treatment response monitoring, risk strati cation, and outcome prediction.

Glioblastoma tissue specimens
The UC Davis Pathology Biorepository at the Comprehensive Cancer Center provides high quality and well-characterized human brain tumor tissue specimens. In this centralized biorepository, all samples were collected after patients' informed consents and underwent quality control by a clinical neuropathologist. From January 2010 to July 2022, a total of 12 fresh frozen GBM specimens were identi ed and obtained from the biorepository for metabolomic analysis. Limited deidenti ed clinical information was abstracted from the medical records within the scope of the approved IRB protocol of the biorepository.

Plasma collection
The UC Davis Department of Neurosurgery (Dr. Orin Bloch Laboratory) has an IRB approved protocol to collect blood samples from GBM patients at diagnosis and throughout treatment, along with access to clinical information of these patients. All procedures performed in this study were in accordance with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. From January 2018 to July 2022, a total of 12 plasma specimens were selected for metabolomic analysis.
Untargeted metabolomic, biogenic amine, and lipidomic analyses for GBM tissue and patient plasma

Sample preparation and extraction
Metabolites and biogenic amines were extracted as previously described. [14] Blood plasma or serum was extracted following the protocols rst published by V. Matyash et al. [15] Using this protocol, lipid extracts in methyl tert-butyl ether phase (MTBA) were separated from proteins and polar hydrophilic small molecules (in the methanol/water phase) in a way that the lipids were found in the top layer of liquidliquid separations, rather than in the bottom layer. Decanting the top layer therefore ensured that the extracts were not contaminated by proteins or polar compounds. The top layer was used for lipidomics while the bottom layer (methanol/water phase) was very suitable for the hydrophilic interaction liquid chromatography-mass spectrometry (HILIC-MS) investigations.
Primary metabolism data were acquired by gas chromatography (GC)-MS using an Agilent 7890A GC coupled to a Leco Pegasus HT TOF mass spectrometer as previously described. [14] Brie y, extracts were dried down, derivatized by methoxyamination and trimethylsilylation, and injected in splitless mode with a temperature gradient from 50-330 o C. Mass spectra were acquired at 17 Hz from 85-500 Da. [14] Raw data processing and metabolite annotation Acquired raw LC-MS and LC-MS//MS data were processed as previously described. [14] Raw CSH-C18-TOF (for lipidomics) and HILIC-TTOF (for polar metabolites pro ling) MS data were processed using MS-Dial 4.9, data-independent MS/MS deconvolution for comprehensive metabolome analysis, for untargeted peak-picking, peak alignment and annotation of related peaks. [14] Raw GC-TOF MS data les were processed using ChromaTOF and metabolomics BinBase database. [14] Statistical analysis Descriptive statistics were used to characterize baseline patient and treatment characteristics. Individual metabolite abundance comparisons at diagnosis and at relapse were performed using GraphPad Prism 9 (version 9.5, San Diego, CA). MetaboAnalyst 5.0 MetaboAnalyst 5.0 (McGill University, Montreal, QC, Canada) (http://www.metaboanalyst.ca) [16] was used to generate principal component analysis (PCA), score plots, heat maps and volcano plots. The processed peak heights with their annotation were imported to MetaboAnalyst, normalized to the total sample median and auto scaled. Paired or unpaired Student's t-Test was used to identify signi cantly altered metabolites between the compared groups (pvalue of less than 0.05 was considered signi cant). ChemRICH (Chemrich. ehnlab.ucdavis.edu), a statistical enrichment approach based on chemical similarity rather than sparse biochemical knowledge annotations was used to group the metabolites. ChemRICH sets have a self-contained size where pvalues do not rely on the size of a background database.

Cohort description
Patient demographics are described in Table 1. In the tissue cohort, a total of 12 specimens from 9 patients were analyzed (9 specimens at diagnosis, 3 of which had paired specimens at recurrence). The mean age at diagnosis was 49 years. There was a male predominance of 67%. Most of this cohort was non-Hispanic White in race/ethnicity. In the plasma cohort, a total of 12 paired specimens (diagnosis and at recurrence) from 6 GBM patients were analyzed. The mean age was 54 years. There was a male predominance of 67%. The original GBM tissue pathology all showed IDH wild type. Additional pathology features including EGFR, MGMT, and ATRX status are shown in Table 1. The mean progression free survival of this cohort was 14 months. The mean overall survival was 17 months.
Unsupervised exploratory analysis on GBM tissue metabolomic, biogenic amine, and lipidomic pro ling The GBM tissue cohort included 9 tumor tissue specimens at diagnosis, 3 of which had paired tissue specimens at recurrence. A complete list of signi cantly altered metabolites is available in Supplemental Table 1. A summary of these changes is shown in Fig. 1. Principal component analysis (PCA) showed two grouped clustering trends with signi cant overlap (Fig. 1, A). A heat map of the top 50 signi cantly altered metabolites, lipids, and biogenic amines, showed differences in cluster trends between tissue samples at diagnosis and at recurrence (Fig. 1, B). There were several signi cantly upregulated compounds (Fig. 1, C) in the lipidomic and biogenic amine analysis. Also included in Fig. 1 are scattered column plots for compounds with signi cant change in abundance at diagnosis and at recurrence (Fig. 1 When analyzing the three paired tissue specimens at diagnosis and at recurrence, again the PCA plot demonstrated two overlapping clusters (Fig. 2, A). The heat map showed more visible separation between tissue samples at diagnosis and at relapse (Fig. 2, B). We found 19 compounds that were signi cantly altered. Volcano plot (using p-value < 0.05 and Fold Change cutoff 1.5) revealed that 3 metabolites were upregulated, and 6 metabolites were down regulated (Fig. 2, C). A complete list of signi cantly altered metabolites is available in Supplemental Table 2. Among these metabolites, we found that the level the 2methylbutyryl-L-carnitine (P = 0.02) and an unknown compound with eluting at 1.84 minutes with accurate mass 186.1075 Da (P = 0.04) were signi cantly higher in primary tumors than recurrent GBM (Fig. 2, D).
Unsupervised exploratory analysis on plasma metabolomic, biogenic amine, and lipidomic characteristics All 6 patients enrolled in this cohort had paired plasma specimens at diagnosis and at recurrence.
However, these were not the same patients whose GBM tissue were studied in the cohort above. PCA analysis showed evident separation between the metabolomic pro les at diagnosis and at recurrence, except for patient 1, whose metabolomic pro les were similar at diagnosis and at recurrence (Fig. 3, A). The heatmap of the top 50 altered metabolites showed visible differences between plasma at diagnosis and at recurrence (Fig. 3, B). Again, the metabolite pro le of patient #1 at recurrence was similar to its status at diagnosis, while the other patients' pro les demonstrated differences.
Combined analyses on glioblastoma tissue and patient plasma at diagnosis and at recurrence Both the PCA and heatmap analyses showed separate metabolic pro les from tissues at diagnosis and plasma at diagnosis (Fig. 4, A and B). Similarly, the metabolomic pro le from tissue at recurrence separated from plasma at recurrence (Fig. 4, C and D). When placing all four groups of data in one plot (Fig. 4, C and D), we again saw inter-specimen differences between the metabolomic pro les in tissue and plasma, but there were no intra-specimen differences at diagnosis and at recurrence.

Discussion
In this study, we investigated the comprehensive untargeted metabolomic, lipidomic, and biogenic amine pro les of GBM tissue and patient plasma specimens at diagnosis and at recurrence. Despite a small overall cohort size, our result showed that many metabolites were altered in GBM tissue and patient plasma at recurrence when compared to diagnosis. Our study demonstrated the feasibility of studying GBM tissue and patient blood specimen longitudinally using metabolomic methodology.
GBM display marked metabolic heterogeneity in their microenvironments. [17] Both glucose and lipid metabolisms are abnormally regulated in GBM tissues. [18,19] In our study, we observed several metabolites that had changed in abundance at recurrence when compared to diagnosis. Many of these metabolites were also identi ed in a recently published study on the metabolic hallmarks of gliomas. [20] Speci cally, we identi ed 2-methylbutyryl-L-carnitine and ecgonine that were known to re ect tumor metabolic exibility in brain tumor tissues at diagnosis and at recurrence. Carnitine serves as a "shuttlemolecule" that allows fatty acid acyl moieties to enter the mitochondrial matrix for oxidization via the beta-oxidation pathway [21]. We found that the 2-methylbutyryl-L-carnitine level was signi cantly reduced in recurrent tumors compared to initial GBM tissue. Carnitine transporter modulation has been thought to be a potential target for cancer treatment. [21] In addition, we also found many altered levels of lipids in GBM tissue at recurrence when compared to initial diagnosis. Our ndings are in line with previously published data suggesting lipid metabolic alterations in GBM. [22] In addition, mannitol was upregulated in recurrent tissue compared to the original tumor, suggesting BBB permeability changes after surgical resection and chemoradiation. This may suggest mannitol as a vehicle to guide targeted treatment.
The list of metabolites with signi cantly altered abundance and fold changes differ when comparing the unpaired samples (9 tissue samples at diagnosis vs 3 tissue samples at recurrence) and the paired samples (3 tissue samples from the same patients at diagnosis and at recurrence). Also, despite a small sample size, the paired tissues samples at diagnosis and at recurrence demonstrate a clearer trend in the differences in compound abundance. Therefore, paired samples are recommended in future studies given their better capacity as an internal control.
In both the PCA plot from patient plasma and the heatmap generated from the top 50 altered metabolites, we found that the metabolomic pro les differed between specimens at diagnosis and at recurrence, except for one patient with early recurrence. The metabolomic pro le of this patient's plasma at diagnosis was different from the rest of the plasma specimens at diagnosis and was similar to the group pattern at recurrence. This interesting nding needs to be validated in a larger cohort of patients with treatment refractory tumors and early recurrence. This metabolomic signature may indicate a high risk and poor prognosis.
In the plasma cohort, we found several signi cantly altered metabolites with large fold changes. For example, 2,4-di uorotoluene increased in patient plasma at recurrence; this metabolite is incorporated into DNA and undergoes replication by DNA polymerase enzymes. [23] The change suggests rapid growth of recurrent tumors. Diatrizoic acid also increased at recurrence. However, this is a contrast agent used during imaging. Indole-3-acetate is an indol-3-yl carboxylic acid anion and has a role as a human metabolite. We again found several other compounds involved in glucose and lipid metabolism. The overall pattern changes of these compounds need to be validated in targeted metabolomics for their potential candidacy as biomarkers for treatment response and tumor recurrence. The enrichment of these metabolites in recurrent GBM tissue may suggest that targeting metabolic activity can be a potential adjuvant targeted treatment for GBM patients.
Interestingly, when plotting all four groups in the same PCA plot, the brain tissue and plasma specimens separated distinctly from each other regardless of disease status (Fig. 4). There was no overlap between specimen types and, the intra-specimen comparison at diagnosis and at recurrence became smaller. This suggested that plasma metabolite patterns are not re ective of brain tumor tissue, and therefore, two sets of biomarker panels are necessary for tissue and plasma. The pattern of tumor microenvironment can be further altered through end organ metabolism in the plasma. Also, plasma is affected by systemically administered medication for peri-operative care.
Our study has limitations. First, we must acknowledge that the sample size in this pilot study is small. Second, although we have paired data within the brain tissue and plasma cohorts at diagnosis and at recurrence, we were unable to identify if the tissue and plasma specimens were from the same patients due to the restrictions of the UC Davis biorepository consents for GBM tissue. In addition, we did not have normal brain tissue or plasma for comparison. These limitations prevented us from identifying small changes in metabolite abundance. Therefore, only signi cant, and large fold changes were analyzed. A bigger cohort with paired specimens will be emphasized in future studies. The current study paved the way for the next targeted metabolomics, lipidomic, and biogenic amine studies validating and further investigating these pro les.

Conclusions
Our data suggest that metabolomic pro les of human GBM tissue and patient plasma differ at diagnosis and at recurrence. Many metabolites involved in tumorigenesis and metabolomic exibility were identi ed. A larger study using target metabolomic assay is warranted to measure the levels of these metabolites, which will help identify the metabolomic signatures in both GBM tissue and patient plasma for risk strati cation, clinical outcome prediction, and development of new adjuvant metabolomictargeting therapy.

Declarations
Author Contributions: All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by Yin Allison Liu, Orwa Aboud, Lina Dahabiyeh, Orin Bloch, and Oliver Fiehn. The rst draft of the manuscript was written by Yin Allison Liu, and all authors commented on previous versions of the manuscript. All authors read and approved the nal manuscript.
Data Availability: The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.  (E) Signi cantly altered metabolite clusters by ChemRich, a statistical enrichment approach based on chemical similarity. The size of the dots is in proportion with the level of alteration in each cluster of metabolites. Red indicates increased peak value and blue indicates decreased peak value. (F) Box plots demonstrating metabolite peak value changes at diagnosis and at recurrence in brain tumor tissues. Values were determined as peak height from LC/MC analysis. Single asterisk indicates a p value of <0.05.
Double asterisks indicate a p value of <0.01.