In summary, we have identified profound growth-promoting rewiring of glucose metabolism in aggressive human brain cancers. These cancer-specific metabolic alterations are compelling targets with favourable therapeutic indices. Our research team and others are currently inhibiting these metabolic pathways in patients with brain cancer (NCT04477200, NCT05236036). We are hopeful that broader use of isotope tracing in clinical studies will help us to make additional fundamental observations about cancer metabolism and teach us which patients might benefit most from metabolically targeted therapies.

The human isotope tracing study was approved by the Institutional Review Board of the University of Michigan. After obtaining informed consent, eight patients with suspected GBM were enrolled onto our clinical study, which was performed perioperatively with standard-of-care craniotomies for tumour resection. Near the start of each procedure (approximately 2-4 h before tissue resection), patients received a bolus intravenous dose of [UC]glucose (8 g) followed by a continuous intravenous infusion of [UC]glucose at a rate of 4 g h. Arterial blood was collected into EDTA-coated vials every 30-60 min for plasma preparation and analysis until intracranial tissues of interest were isolated from each patient. After initial craniotomy and tumour exposure, stereotactic image guidance was used to identify radiographically defined enhancing tumour, non-enhancing FLAIR T2 hyperintense tumour and adjacent cortical tissues. These tissues were then resected and separated by a board-certified neurosurgeon (W.N.A.-H.), rinsed in cold PBS and immediately (typically less than 3 min after resection) flash-frozen in liquid nitrogen by the research team for further analysis. Cortex and non-enhancing tumour were obtained from all eight patients. Enhancing tumour tissue was obtained from only seven patients, because one tumour comprised entirely non-enhancing disease. Tumour content quantification in human samples (cortex, enhancing tumour, non-enhancing tumour) stained with H&E was performed by a board-certified neuropathologist (S.V.). Clinical information including Ki-67 index and IDH mutation status were abstracted from the medical record and clinical pathology report. Sample size was based on maximal enrollment rather than predetermination with statistical methods, and patients were not allocated into groups. Blinding was not performed, as the paired design comparing distinctly identifiable tumour and cortex from each participant made masking impractical during sample collection and analysis; however, the quantitative nature of the data obtained by mass spectrometry mitigated potential subjective bias in interpretation.

All animal studies were approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Michigan. Mice were housed in specific-pathogen-free conditions at a temperature of 74 °F and relative humidity between 30 and 70% on a light-dark cycle of 12 h on-12 h off with unfettered access to food and water. In experiments assessing baseline C labelling from [UC]glucose under control or serine/glycine-restricted conditions, all mice were placed on either control diets (TestDiet Baker Amino Acid Diet 5CC7) or serine/glycine-restricted diets (TestDiet Modified Baker Amino Acid Diet 5BJX) 3 days before orthotopic tumour implantation (described below) and remained on respective diets indefinitely. Before intracranial implantation and for all other experiments, mice were fed standard chow (PicoLab Laboratory Rodent Diet 5L0D).

Studies of intracranial tumour-bearing mice used three different patient-derived models that genomically and histologically reflect typical GBMs and capture heterogeneity characteristic of the disease. These models included the patient-derived HF2303 model, as well as the GBM12 and GBM38 PDXs from the PDX National Resource at Mayo Clinic. Tumour tissue was propagated subcutaneously in the flanks or brains of immunodeficient mice (B6.129S7-Rag1/J, obtained from Jackson Laboratory or bred in-house). To introduce GFP and firefly luciferase into GBM tissue and enable their fluorescence and luminescence detection, short-term explant cultures were generated from tumours and transduced by lentiviral infection (lenti-LEGO-Ig2-fluc-IRES-GFP-VSVG or lenti-CAG-dscGFP-T2ALuc-Efs-Puro-VSVG, obtained from the University of Michigan Vector Core). After infection, cells were enriched for GFP-positive populations by fluorescence-activated cell sorting or puromycin selection and then reintroduced to mice either as subcutaneous flank tumours or intracranial tumours. To generate orthotopic GBM brain tumours in B6.129S7-Rag1/J mice, 3 × 10 to 1 × 10 cells were implanted intracranially using a stereotactic rig or guide screws to the region of the brain calculated to be the striatum in anaesthetized mice. Tumour development was then confirmed by bioluminescence imaging as described in Supplementary Methods.

Sample sizes for animal experiments were determined by preliminary studies and the level of observed effect. Stratified randomization was used to achieve similar starting median tumour luminescence across groups. Blinding was not possible due to the complexity of multistep procedures: group allocation, animal treatments, animal monitoring, data collection and analyses were performed independently by separate investigators, including technicians unaware of hypotheses and without stake in outcome, minimizing bias. For all experiments, B6.129S7-Rag1tm1Mom/J mice (Jackson 002216 or bred in-house) were used, except in the long-term serine/glycine restriction experiment described in Supplementary Methods, which used C57BL/6J mice (Jackson 000664) without tumours. Mice were aged 4-12 weeks at the start of experiments. All experiments used a mix of male and female mice, except for the time-course glucose tracing experiment, which comprised only female mice and was validated using non-glucose tracers in both male and female mice. The maximum allowed tumour sizes and end points were predetermined in agreement with IACUC. For subcutaneous flank tumours, these were as follows: tumour diameter exceeds 2 cm in any single dimension; tumour ulcerates greater than half its surface area; ulceration has effusion, appears infected or has haemorrhage; tumour develops in an area that impairs normal movement or physiological behaviour. For intracranial GBM-bearing mice, humane end points were determined based on clinical observations in collaboration with veterinarians with University of Michigan's Unit for Laboratory Animal Medicine. In all experiments, we followed their numerical scoring system, assigning points for appearance, physical parameters, behaviours and body condition. When predetermined scores were reached, mice meeting these end-stage criteria were humanely euthanized. No experiments exceeded tumour sizes or end points predetermined with IACUC.

At approximately 3-6 weeks after orthotopic tumour implantation, intracranial GBM-bearing mice underwent dual catheterizations, with one catheter surgically placed into the jugular vein (for tracer administration) and a second catheter placed into the carotid artery (for plasma collection during infusion). Mice were then allowed to recover from surgery for 4-5 days. Conscious, free-moving, undisturbed mice were then delivered stable isotope tracers through the intravenous line as follows: [UC]glucose administered as a bolus dose (0.4 mg per g) followed by a continuous infusion (12 ng per g per min) for a total of 4 h; N-amide-glutamine administered as a bolus dose (0.28 mg per g) followed by a continuous infusion (5 µg per g per min) for 6 h; N-inosine administered as a bolus dose (0.8 mg) followed by a continuous infusion (0.65 µg per g per min) for 6 h; or C-serine delivered continuously (6.3 µg per g per min) for 6 h.

During infusions, blood was collected periodically through the arterial line into EDTA-coated vials and used to prepare plasma. At the end of infusions, ketamine (50 mg per kg) was administered into the intravenous line to rapidly induce anaesthesia. Mice were then decapitated, and tissues were extracted on dry ice. To separate orthotopic GBM from mouse cortex, we performed microdissection aided by fluorescent light that enabled us to distinguish GFP-positive tumour from GFP-negative cortex. All tissues were then immediately (less than 3 min after anaesthesia) flash-frozen in liquid nitrogen for further analysis. In experiments measuring the spatial distribution of labelled intracranial metabolites, GBM-bearing mice were administered two intraperitoneal injections of either saline or [UC]glucose at dosages of 2 g per kg 30 min apart. At 30 min after the second injection, mice were administered a lethal dose of isoflurane (inhalation) and then decapitated. Whole brains were rapidly collected and flash-frozen in liquid nitrogen and then cryosectioned for further analysis.

To estimate nucleotide synthesis fluxes from isotopologue time-course data, in vivo MFA methods were developed. These are described in detail in Supplementary Methods and corresponding Supplementary Tables 3-13. To estimate serine contributions in brain tumours using single-timepoint data, a steady-state method was used as described in Supplementary Methods and corresponding Supplementary Table 14.

Flash-frozen tissue samples were homogenized in cold (-80 °C) 80% methanol. For plasma analysis, 100% methanol at -80 °C was added to plasma samples to yield a final methanol concentration of 80%. Insoluble material was then precipitated from all samples by centrifugation at 4 °C, and supernatants containing soluble metabolites were dried by nitrogen purging or vacuum centrifugation. Dried metabolites were reconstituted in 50% methanol for LC-MS analysis. Isotope labelling was determined using an Agilent system consisting of an Infinity Lab II UPLC coupled with either a 6545 or 6230 QTOF mass spectrometer (Agilent Technologies) operated by the University of Michigan Metabolomics Core, and data were analysed with values corrected for natural isotope abundance using MassHunter Profinder 10.0. We used control samples without C or N labelling to ensure that labelled isotopologues from mice and patients were not from contaminating species, although it is possible that isobars present only in traced samples could still contaminate peaks. For targeted analyses of metabolite abundances, the samples were prepared as described above and then analysed using the Agilent 1290 Infinity II LC-6470 Triple Quadrupole tandem mass spectrometry system (Agilent Technologies). For compound optimization, calibration and data acquisition, Agilent MassHunter Quantitative Analysis Software v.B.08.02 was used. Retention times, transitions, buffers and gradient compositions are described in Supplementary Methods and Supplementary Table 15. For in vitro patient-derived gliomaspheres, MS and analyses of serine, glycine and phosphoglycerate are described in Supplementary Methods.

Standard microscopy slides with mounted tissue were vacuum desiccated for 20 min before matrix coating. After desiccation, the slides were sprayed with NEDC matrix (10 mg ml, 1:1 ACN:HO) using an M3+ sprayer (HTX Technologies; flow rate, 75 µl min; temperature, 70 °C; velocity, 1,000 mm min; track spacing, 1 mm; pattern, crisscross; drying time, 0 s). The slides were mounted into a MTP Slide Adapter II (Bruker Daltonics) before analysis.

MALDI imaging data were acquired using the timsTOF fleX MALDI-2 mass spectrometer (Bruker Daltonics) operating in transmission mode with a 20 µm raster size, acquiring m/z 100-600. The laser (Bruker Daltonics; SmartBeam 3D, 355 nm, 5,000 Hz repetition rate) used a 16 µm beam scan, resulting in a 20 µm × 20 µm ablation area. Taurine was used as a lock mass ([M-H]1-, m/z 124.0074).

MS imaging data were visualized using SCiLS Lab 2023b, with single fractional enrichment, normalized mean enrichment and fractionalized enrichment images generated in SCiLS Lab using an in-house script using the SCiLS REST API (Bruker Daltonics; v.6.2.114), written in R (v.4.2.2), using RStudio (2022.12.0 build 353). A segmentation algorithm built into SCiLS Lab was used to create four data-driven regions corresponding to the healthy and GBM tissue in the C dosed and control tissues (normalization, total ion count; denoising, weak; method, bisecting k-means; metric, Manhattan). The relative isotopologue intensity of these regions was also calculated with another in-house script implemented through the SCiLS REST API.

Tentative annotations were performed using MetaboScape 2023 (Bruker Daltonics) using target lists of known biological molecules generated with the TASQ 2022 software (Bruker Daltonics; amino acids, glycolysis, citrate cycle). The molecular formula of target molecules was used to calculate an accurate mass for each target. Annotations required a mass error of less than 3.0 ppm. In total, 72 features were annotated using this limited list, with annotated peaks having a mass accuracy of 1.1 ppm.

Spatial metabolite imaging and analysis of human clinical specimens are detailed in Supplementary Methods.

In experiments evaluating the influence of RT on metabolism, cannulated mice were anaesthetized by 2% isoflurane inhalation and then treated with cranially directed RT at a dose of 8 Gy or sham RT with a lead shield keeping the cranium exposed. Immediately after RT (less than 5 min), the mice were removed from anaesthesia and administered tracer infusions as described above.

In studies assessing animal survival, mice were placed onto either a control diet (TestDiet Baker Amino Acid Diet, 5CC7) or a modified diet containing 0% serine and 0% glycine with all other amino acids adjusted to account for serine and glycine reduction (TestDiet Modified Baker Amino Acid Diet, 5BJX) 3 days before orthotopic tumour implantation. The mice were then implanted with luciferase-positive intracranial tumours and maintained on respective diets for the remainder of experimentation. Once bioluminescence imaging (described in the Supplementary Methods) detected intracranial tumours with flux values of 10-10 p s (9 to 50 days after implantation), the mice on each diet were randomized into control or chemoradiation-treated groups. On each treatment day, mice received both TMZ (oral administration) and cranial RT (under isoflurane anaesthesia with a lead shield keeping cranium exposed), with RT administered 1 h after TMZ treatment. GBM12-bearing mice were administered chemoradiation as three consecutive days of TMZ (25 mg per kg) and cranial RT (2 Gy). GBM38-bearing mice were administered chemoradiation as 3 consecutive days of TMZ (50 mg per kg) and cranial RT (2 Gy). HF2303-bearing mice were administered chemoradiation in two cycles, with 13 days between cycles. The first cycle consisted of 4 consecutive daily doses of TMZ (25 mg per kg) and RT (2 Gy). The second cycle consisted of 2 consecutive treatment days, a 2 day gap, and then 1 treatment day. A schematic of the treatment schemes is provided in Extended Data Fig. 11j-l. Overall survival was determined using the Kaplan-Meier method, with log-rank tests used to compare survival curves.

Histological analyses were performed as previously described by H&E staining and Ki-67 staining. The primary antibody for Ki-67 (BD Biosciences, 550609) was used at a 1:1,000 dilution. Immunohistochemical staining was performed using the Vectastain Elite ABC Kit (PK-6102), which included the secondary antibody (1:200). Further details and quantification are described in Supplementary Methods.

To quantify the incorporation of stable isotope tracers into metabolites, we calculated mean enrichment, defined as the average percentage of labelled atoms (C or N) within a metabolite pool, corrected for natural abundance:

where n is the number of C or N atoms in the metabolite, i represents isotopologues and m represents isotopologue abundance. This metric captures the total fractional contribution of tracers to metabolites of interest, averaging labelling across all isotopologues. Alternatively, specific isotopologue fractions or relative abundance were used to track specific metabolic pathways. When appropriate and where indicated, tissue enrichment values were normalized to plasma tracer enrichment. Detailed analysis methods are described in the figure legends. Isotopologues and mean enrichment percentages were determined using MassHunter Profinder v.10.0 (Agilent Technologies).

Descriptive statistics (mean, s.d.) for metabolite enrichments are graphically presented. Comparisons between groups were performed using linear models for unpaired data and linear mixed-effects models for paired (for example, cortex and tumour from the same individual) or repeated-measures data (for example, in cases of multiple tumour foci extracted from a single mouse). Holm's method was used to adjust for multiple comparisons performed for each metabolite. All tests were two-sided, and statistical significance was assessed at the 0.05 significance level. Statistical comparisons and parameters are further described in the figure legends. Analyses were performed using R v.4.4.2 or GraphPad Prism 10.

No commonly misidentified cell lines were used in this study. The GBM374gs gliomasphere model was authenticated regularly via short tandem repeat fingerprinting using the GenePrint 10 system (Laragen). All experiments involving GBM12, GBM38 and HF2303 patient-derived models were performed in vivo, with the exception of in vitro culturing for lentiviral GFP and luciferase transduction, and were not authenticated beyond primary isolation; in vivo passage and marker retention ensured identity of these models. All models cultured in vitro were tested regularly for mycoplasma using the MycoAlert Mycoplasma Detection Kit (Lonza) and confirmed negative. Comprehensive experimental and analytical details including metabolite imaging of clinical samples, LC-MS, subtype and transcriptional analyses of our data and others, development of MFA methods, in vitro gliomasphere experiments, additional in vivo methods, and histological staining and quantification protocols are found in Supplementary Methods, containing refs. .

Human data were obtained from the same cohort of patients throughout the study. Labelled human samples were analysed from eight patients, comprising n = 8 cortex samples, n = 7 enhancing tumour samples, and n = 8 non-enhancing tumour samples. One patient lacked enhancing tumour tissue. Mouse LC-MS data corresponding to [UC]glucose labelling and metabolite abundance in HF2303 and GBM12 models were from the same cohorts throughout the study. Mouse LC-MS data corresponding to single-timepoint [UC]glucose labelling in GBM38-bearing mice were from the same cohort throughout the study. For all mouse LC-MS data showing single-timepoint labelling from [UC]glucose, samples were analysed from sixteen total mice with n = 16 cortex samples and comprised: seven GBM38-bearing mice with n = 7 tumour foci and n = 7 cortex samples; five GBM12-bearing mice with n = 5 tumour foci and n = 5 cortex samples; and four HF2303-bearing mice with five tumour foci, one of which was diffuse and cortically contaminated and was therefore excluded from analysis, yielding n = 4 tumour foci and n = 4 cortex samples. One of these HF2303-bearing mice exhibited suboptimal data quality for plasma glucose measurements used to normalize tissue labelling. Therefore, plasma glucose enrichment for this mouse was estimated using the mean value derived from remaining mice in the group. Mouse data corresponding to single-timepoint metabolite abundance in GBM38-bearing mice were from the same cohort throughout the study. All time course [UC]glucose labelling data were from the same cohort of mice. All data showing in vivo isotope labelling from N-glutamine were from the same cohort of mice. All data showing in vivo isotope labelling from N-inosine were from the same cohort of mice. Mouse tissue spatial imaging data and matched H&E staining shown throughout this study are from the same C-labelled and negative control unlabelled mice. In developing our hypothesis-generating metabolic flux models of nucleotide synthesis, the need for multiple timepoints constrained sample numbers at each timepoint, and exclusions were necessary to minimize variance within groups and ensure data robustness. Specific exclusion criteria are described at https://github.com/baharm1/iMFA/, and model-based predictions were validated by multiple orthogonal tracers. Mice failing to form intracranial tumours post-orthotopic implantation (<5%) or dying 0-5 days after implantation (<5%) or jugular/carotid catheter placement (<10%), likely due to surgical complications, were excluded. In LC-MS experiments, metabolites below detection thresholds were excluded.

Stable isotope infusion studies in human patients inherently lack full repeatability because of substantial variability among individuals, clinical-protocol-related factors and restricted tissue quantities. Labelled glucose infusion studies of intracranial GBM38-bearing mice were performed three or more times. In vivo glucose labelling studies were further repeated by five or more serial labelled glucose injection experiments using intracranial HF2303 tumour-bearing mice, intracranial GBM38 tumour-bearing mice, and two additional intracranial models not presented. In vivo serine tracer uptake measurements were replicated via serial tracer injection into intracranial GBM38-bearing mice. Studies with amide-labelled glutamine infusions showed consistent GBM versus cortex label patterns in three or more independent experiments. In vitro gliomasphere experiments were performed three times. All attempts at replicating the experiments were successful.

n values for Fig. 3 are shown in the figure caption. For Fig. 4a, samples were analysed from eight patients, comprising n = 8 cortex samples, n = 7 enhancing tumour samples and n = 8 non-enhancing tumour samples. For Fig. 4b, samples were analysed from 16 total mice with n = 16 cortex samples and comprised 7 GBM38-bearing mice with n = 7 tumour foci and n = 7 cortex samples, 5 GBM12-bearing mice with n = 5 tumour foci and n = 5 cortex samples, and 4 HF2303-bearing mice with 5 tumour foci, one of which was diffuse and cortically contaminated and was therefore excluded from analysis, yielding n = 4 tumour foci and n = 4 cortex samples. For Fig. 4d, samples were analysed from 5 GBM-bearing mice, comprising n = 5 cortex samples and n = 5 tumour samples. For Fig. 4e, samples were analysed from 5 GBM-bearing mice, comprising n = 4 cortex samples and n = 4 tumour samples. For Fig. 4f, samples were analysed from five GBM-bearing mice, comprising n = 5 cortex samples and n = 5 tumour samples. For Fig. 4i, animal numbers are as follows: control diet, n = 7; -SG diet, n = 8; control diet with chemoradiation, n = 8; -SG diet with chemoradiation, n = 9. For Fig. 4j, animals were examined over three independent experiments with total mouse numbers as follows: control diet, n = 14; -SG diet, n = 14; control diet with chemoradiation, n = 11; -SG diet with chemoradiation, n = 9. In one of these experiments, three mice were excluded due to lack of detectable brain tumour and one mouse was excluded due to an unrelated condition requiring its humane euthanasia. For Fig. 4k, animals were examined over two independent experiments with total mouse numbers as follows: control diet, n = 18; -SG diet, n = 21; control diet with chemoradiation, n = 10; -SG diet with chemoradiation, n = 13. For Fig. 4m, data corresponding to control diets are from the same HF2303-bearing mice described in Fig. 4b. Moreover, data corresponding to -SG diets were obtained from four HF2303-bearing mice with four cortex samples and four tumour samples. Thus, sample numbers are as follows: for control diet, n = 4 cortex samples and n = 4 tumour samples; and, for the -SG diet, n = 4 cortex samples and n = 4 tumour samples. For Fig. 4n, data are from the same mice as described in Fig. 4b,m. These samples comprised n = 4 control tumour samples and n = 4 -SG tumour samples.

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.