Rewiring of cortical glucose metabolism fuels human brain cancer growth

Abstract

The brain avidly consumes glucose to fuel neurophysiology¹. Cancers of the brain, such as glioblastoma, relinquish physiological integrity and gain the ability to proliferate and invade healthy tissue². How brain cancers rewire glucose use to drive aggressive growth remains unclear. Here we infused ¹³C-labelled glucose into patients and mice with brain cancer, coupled with quantitative metabolic flux analysis, to map the fates of glucose-derived carbon in tumour versus cortex. Through direct and comprehensive measurements of carbon and nitrogen labelling in both cortex and glioma tissues, we identify profound metabolic transformations. In the human cortex, glucose carbons fuel essential physiological processes, including tricarboxylic acid cycle oxidation and neurotransmitter synthesis. Conversely, gliomas downregulate these processes and scavenge alternative carbon sources such as amino acids from the environment, repurposing glucose-derived carbons to generate molecules needed for proliferation and invasion. Targeting this metabolic rewiring in mice through dietary amino acid modulation selectively alters glioblastoma metabolism, slows tumour growth and augments the efficacy of standard-of-care treatments. These findings illuminate how aggressive brain tumours exploit glucose to suppress normal physiological activity in favour of malignant expansion and offer potential therapeutic strategies to enhance treatment outcomes.

Fig. 1. Glucose labelling of brain tumours and cortex.

a, Schematic of [U¹³C]glucose infusions and analyses of patients with glioma and patient-derived intracranial GBM mouse models. The diagram was created using BioRender. b, Example of magnetic-resonance-imaging-defined tissue acquisition. c, Arterial m+6 glucose enrichment in patients undergoing [U¹³C]glucose infusions at the indicated times after the start of infusion. Plasma from patient 8 could not be analysed at time 0. The triangles indicate the tissue resection and flash-freezing timepoint. d, Arterial m+6 glucose enrichment in mice undergoing [U¹³C]glucose infusions. Data are mean ± s.d. at each timepoint. e,f, Normalized (to labelled plasma glucose on a per-patient (e) or per-mouse (f) basis) enrichment of glycolytic intermediates in intracranial tissues from eight patients with glioma (e) or mouse models (f) infused with [U¹³C]glucose. Data are mean ± s.d. Comparisons between groups were performed using linear mixed-effects models with a random intercept for individual. For each metabolite, multiple pairwise comparisons across tissue types were adjusted using Holm’s method. Statistical tests were two-sided. g, H&E staining of GBM38 patient-derived xenograft (PDX) grown orthotopically (left) and MALDI image showing ¹³C enrichment of lactate (right) with the tissue maximum set at 100%. Imaging with a separate, independent instrument was performed once and produced similar results. Scale bar, 3 mm. FBP, fructose 1,6-bisphosphate; GAP, glyceraldehyde 3-phosphate; DHAP, dihydroxyacetone phosphate; 3PG/2PG, 3-phosphoglycerate/2-phosphoglycerate; PEP, phosphoenolpyruvate. Source data

Fig. 2. GBM modulates TCA cycle and neurotransmitters to channel glucose into nucleotides and NAD.

a, Intracranial enrichment of TCA cycle intermediates in patients with glioma infused with [U¹³C]glucose. b, Intracranial enrichment of TCA cycle intermediates in GBM-bearing mice infused with [U¹³C]glucose. c, Intracranial enrichment of amino acids derived from the TCA cycle in patients with glioma infused with [U¹³C]glucose. d, Intracranial enrichment of amino acids derived from the TCA cycle in GBM-bearing mice infused with [U¹³C]glucose. e, Malate enrichment in representative human cortex, non-enhancing tumour tissue and enhancing tumour tissue. The colour bar maximum is set at true enrichment. f, MALDI image showing normalized mean ¹³C enrichment of malate in brain from a GBM38 tumour-bearing mouse. GBM was defined by an overlay with H&E staining shown in Fig. 1g. The colour bar tissue maximum is normalized to 100%. Scale bar, 3 mm. g, Intracranial purine enrichment in patients with glioma infused with [U¹³C]glucose. h, Intracranial purine enrichment in GBM-bearing mice infused with [U¹³C]glucose. i, MALDI image of AMP m+5 signal intensity in brain from a GBM38 tumour-bearing mouse with tissue maximum set at 100%. GBM was defined by an overlay with H&E staining shown in Fig. 1g. Scale bar, 3 mm. j, Intracranial pyrimidine enrichment in patients with glioma infused with [U¹³C]glucose. k, Intracranial pyrimidine enrichment in GBM-bearing mice infused with [U¹³C]glucose. l, Intracranial enrichment of NAD and NADH in patients with glioma infused with [U¹³C]glucose. m, Intracranial enrichment of NAD and NADH in GBM-bearing mice infused with [U¹³C]glucose. For a–d, g, h and j–m, data are mean ± s.d. of metabolite enrichments normalized to labelled plasma glucose on a per-patient or per-mouse basis. Comparisons between groups were performed using linear mixed-effects models with a random intercept for individual, and multiple pairwise comparisons across tissue types were adjusted using Holm’s method. All statistical tests were two-sided. Source data

Fig. 3. GBM elevates nucleotide biosynthesis fluxes.

a, Flux estimation scheme: to measure metabolite enrichment and estimate metabolic fluxes, analysis was performed using GBM and cortex tissues from intracranial GBM38-bearing mice infused with [U¹³C]glucose that were serially euthanized at varying timepoints (0, 30, 120 and 240 min). b, Purine synthesis fluxes in mouse cortex and orthotopic GBM38 tissues calculated from [U¹³C]glucose MFA. Statistical significance was determined by non-overlapping 95% confidence intervals; *P < 0.05. c, Pyrimidine synthesis fluxes in mouse cortex and orthotopic GBM38 tissues calculated from [U¹³C]glucose MFA. Statistical significance was determined by non-overlapping 95% confidence intervals. d, Validation of [U¹³C]glucose MFA through stable isotope labelling experiments comparing labelling of tumour and cortex tissues from intracranial GBM38-bearing mice infused with either ¹⁵N-amide-glutamine or ¹⁵N₄-inosine. e, Normalized (to labelled plasma glutamine on a per-mouse basis) enrichment of nucleotides in cortex and orthotopic tumour tissue from intracranial GBM38-bearing mice infused with ¹⁵N-amide-glutamine. Data are mean ± s.d. of cortex and tumour samples isolated from three mice containing one to three intracranial tumour foci each. Comparisons between groups were performed using linear mixed-effects models with a random intercept for mouse. Tests were two-sided. f, Normalized (to labelled plasma inosine on a per-mouse basis) enrichment of purines in cortex and orthotopic tumour tissue from intracranial GBM38-bearing mice infused with ¹⁵N₄-inosine. Data are mean ± s.d. of cortex and tumour samples from four mice containing one to two tumour foci each. Comparisons between groups were performed using linear mixed-effects models with a random intercept for mouse. Tests were two-sided. Hpx, hypoxanthine; MID, mass isotopologue distribution. The diagrams in a and d were created using BioRender. Source data

Fig. 4. Rewired serine metabolism enables GBM nucleotides and therapeutic resistance that can be overcome by dietary serine/glycine restriction.

a,b, The ratios of m+3 serine to m+1 serine in cortex and tumour tissues from patients with glioma (a) or intracranial GBM-bearing mice (b) infused with [U¹³C]glucose. Data are mean ± s.d. Comparisons between groups were performed using linear mixed-effects models with a random intercept for individual, and multiple pairwise comparisons across tissue types were adjusted using Holm’s method. All statistical tests were two-sided. c, ¹³C₃-serine (m+3) was infused into orthotopic GBM-bearing mice to compare serine uptake between cortex and intracranial GBM. d–f, The relative accumulation of infused m+3 serine into mouse cortex and intracranial HF2303 (d), GBM38 (e) or GBM12 (f) tumours. Data are mean ± s.d. Groups were compared using two-sided t-tests. g, Principal component analysis of metabolite levels from cortex and intracranial GBM38 tissues from mice on control diets or −SG diets. h, The relative metabolite levels (top PLS-DA) in orthotopic GBM38 tumours from mice fed control or −SG diets; each row represents a separate tumour. i–k, The survival of intracranial HF2303 (i), GBM38 (j) and GBM12 (k) tumour-bearing mice under altered dietary serine/glycine conditions alone or in combination with chemoradiation treatment. Two-sided log-rank tests were performed to compare survival curves between groups. l, Intracranial tumour-bearing mice on control diets or −SG diets were infused with [U¹³C]glucose and assessed for metabolite labelling. m, Isotopologue distributions of labelled serine in intracranial HF2303 tumour-bearing mice infused with [U¹³C]glucose. Data are mean ± s.d. n, Normalized (to labelled plasma glucose on a per-mouse basis) enrichment of nucleotides and NAD species in HF2303 tumours from intracranial GBM-bearing mice infused with [U¹³C]glucose. The diagrams in c and l were created using BioRender. Source data

Fig. 5. Reprogramming of cortical glucose metabolism fuels GBM growth and therapy resistance.

Models of cortical metabolic rewiring in brain cancer: cortex (left) robustly takes up glucose, which it uses to fuel the TCA cycle and synthesize neuroregulatory metabolites including serine, glutamate, glutamine, GABA and aspartate. Brain tumours (middle) upregulate the uptake of environmental serine and reduce the fraction of glucose incorporated into the TCA cycle and glucose-derived neurotransmitter synthesis. Tumours also reroute glucose-derived carbons to synthesize nucleotides and NAD/NADH, used to drive tumour growth and resistance to chemoradiation. Restriction of dietary serine (right) forces multiple gliomas to reroute glucose carbon towards serine synthesis, which decreases nucleotide and NAD/NADH levels, slows tumour growth and sensitizes tumours to chemoradiation. The diagram was created using BioRender.

Extended Data Fig. 1. Transcriptional profiling of human brain tumours.

a, Heatmap of single sample gene set enrichment analysis (ssGSEA) scores using 150 signature genes of proneural, classical, and mesenchymal subtypes for 170 GBM samples of TCGA and three patients from this study based on the availability of bulk RNA-seq data. Samples were grouped by hierarchical clustering. Four methods were used to identify molecular subtypes: hierarchical clustering, support vector machine (SVM), k-nearest neighbour (KNN), ssGSEA. The latter three methods were used on GlioVis portal. Subtypes of TCGA samples have been predicted in Wang et al. and were used to assess GlioVis methods. b, Venn diagrams of percentages of cells assigned to each molecular subtype based on AUCell scores using scRNA-seq data. n represents the number of cells passed quality control for each patient tissue. c, Heatmap of AUCell score of 150 signature genes for molecular subtypes using scRNA-seq data. Abbreviations: ssGSEA (single sample gene set enrichment analysis), SVM (support vector machine), KNN (k-nearest neighbour).

Extended Data Fig. 2. Histological and metabolic validation of tumour and cortex sample separation.

a, Representative H&E stains of tissues resected from our cohort of 8 glioma patients. b, Percent tumour content in tissues from each patient was defined by a clinical neuropathologist (SV). Data are mean ± s.d. c, Levels of NAA in cortical tissue and tumour tissue (enhancing and non-enhancing) from human glioma patients undergoing surgical resection. d, Levels of 2-hydroxyglutarate in cortical tissue and tumour tissue (enhancing and non-enhancing) from human glioma patients undergoing surgical resection. e-i, Volcano plots of metabolite abundance determined by LC-MS were used to assess fold change in tumour metabolite levels compared to cortical metabolite levels as follows: e, GBM compared to cortex in orthotopic GBM38-bearing mice, f, GBM compared to cortex in orthotopic GBM12-bearing mice, and g, GBM compared to cortex in orthotopic HF2303-bearing mice, h, enhancing tumour compared to cortex in patients, and i, non-enhancing tumour compared to cortex in patients. j, Levels of GABA in intracranial tissues from orthotopic GBM-bearing mice. Data are mean ± s.d. k, Levels of GABA in cortical tissue and tumour tissue (enhancing and non-enhancing) from human glioma patients undergoing surgical resection. Abbreviations: NAA (N-acetylaspartate), NAAG (N-acetylaspartylglutamate), GABA (gamma-aminobutyric acid). Source data

Extended Data Fig. 3. Glucose-derived labelling of tissue and plasma metabolites in patients and mice with brain tumours.

a, Time course of m + 3 lactate in plasma from 8 patients infused with [U¹³C]glucose. Plasma from patient 8 could not be analysed at time 0. b, Time course of m + 3 lactate in plasma from orthotopic GBM38 bearing mice (3–10 per timepoint) infused with [U¹³C]glucose. Data are mean ± s.d. at each timepoint. c, Schema of glucose carbon (red circles) redistribution into glycolytic intermediates. Scrambling can occur via recombination with unlabelled intermediates in the pentose phosphate cycle (E4P, S7P, R5P, GAP). d, Normalized enrichment of m + 6 UDP-glucose in intracranial tissues from 8 glioma patients infused with [U¹³C]glucose. e, Normalized enrichment of m + 6 UDP-glucose enrichment in cortex and orthotopic GBM tissue isolated from intracranial tumour-bearing mice infused with [U¹³C]glucose. f, Schema of ¹³C-labelling in TCA cycle intermediates and neurotransmitters arising from m + 3 pyruvate. Red circles indicate entry through pyruvate dehydrogenase, and gold circles indicate entry through pyruvate carboxylase. Blue circles indicate labelling patterns possible on second TCA cycle turn. g-i, Isotopologue abundance levels of citrate (g), succinate (h), and malate (i) in cortex and GBM tissues at indicated time points from orthotopic GBM38-bearing mice infused with [U¹³C]glucose. j-l, Percent change in citrate (j), succinate (k), and malate (l) isotopologues from 120 to 240 min in mice infused with [U¹³C]-glucose. m, Schema of purine synthetic pathways. Green and yellow circles indicate glycine- and folate-derived carbons, respectively. Blue circles indicate R5P-derived carbons. Partial shading of these circles indicates that a variety of labelling patterns are possible. n, Normalized enrichment of R5P in intracranial tissues from 8 glioma patients infused with [U¹³C]glucose. o, Normalized enrichment of R5P in cortex and orthotopic GBM tissue isolated from intracranial tumour-bearing mice infused with [U¹³C]glucose. p-v, Normalized enrichment of IMP (p), inosine (q), AMP (r), ADP (s), GMP (t), GDP (u), and guanosine (v) in intracranial tissues from glioma patients infused with [U¹³C]glucose. w, Schema of pyrimidine synthetic pathways. Magenta indicates aspartate-derived carbons, blue indicates R5P-derived carbons, and yellow indicates folate-derived carbons. Partial shading is used to indicate that a variety of labelling patterns are possible. x,y, Normalized enrichment of UMP (x) and dTDP (y) in intracranial tissues from glioma patients infused with [U¹³C]glucose. z, Schematic of carbon incorporation into NAD and NADH. In d, e, n, and o, data are mean ± s.d. of metabolite enrichments normalized to labelled plasma glucose on a per-patient or per-mouse basis. Comparisons between groups were performed using linear mixed-effects models with a random intercept for individual, and multiple pairwise comparisons across tissue types were adjusted using Holm’s method. All statistical tests were two-sided. For mice, n = 4–16 samples per group were from 16 mice (4–7 mice for each orthotopic model with cortex samples from all mice pooled) were analysed. For humans, n = 7–8 samples per group were analysed. Some metabolites were not reliably detected in every tissue sample and were therefore not shown. In g, h, and i, data are mean ± s.d. of n = 3–9 samples from 1–3 mice per timepoint. In j, k, and l, data are mean ± s.d. Error bars are propagated from uncertainty in t = 120 and t = 240 min datapoints. n = 9 samples from 3 mice per time point. In c, f, m, w, and z, some intermediates are omitted from pathway diagrams for conciseness. Abbreviations: G6P (glucose 6-phosphate), F6P (fructose 6-phosphate), DHAP (dihydroxyacetone phosphate), GAP (glyceraldehyde 3-phosphate), 3PG (3-phosphoglycerate), R5P (ribose 5-phosphate), E4P (erythrose 4-phosphate), S7P (sedoheptulose 7-phosphate), PRPP (phosphoribosyl pyrophosphate), Me-THF (N10-formyltetrahydrofolate), 5,10-MeTHF (5,10-methylenetetrahydrofolate), NAM (nicotinamide). Source data

Extended Data Fig. 4. Spatially defined isotope labelling in cortex and GBM.

a, H&E staining of brains from orthotopic GBM bearing mice intraperitoneally injected with either vehicle (left) as a negative control or [U¹³C]glucose (right). b-h, MALDI-MS was used to determine ¹³C enrichment of lactate (b), malate (c), aspartate (d), GABA (e), glutamate (f), glutamine (g), and AMP m + 5 (h). Tissue maximum is set to 100%. i-k, Tissues resected from brain cancer patients receiving either [U¹³C]glucose infusion (patients 1–8) or no infusion (UL 1–2) were assessed by MALDI-MS for ¹³C isotope labelling of malate (i), glutamate (j), and glutamine (k). Colour bar maximum set at true enrichment. l-n, Quantification of ¹³C-labelling of malate (l), glutamate (m), and glutamine (n) for all data points (pixel values) from spatial MALDI-MS scans shown in panels i-k and Fig. 2e. In box plots, centre line represents median, box limits represent upper and lower quartiles, and whiskers show 1.5x interquartile range. Outlier points are hidden. Violin plots are trimmed to the range of the data. All violins have the same maximum width. For a and b, the images of ¹³C-labelled tissues are also shown in Fig. 1g. For c, the image of ¹³C-labelled tissue is also shown in Fig. 2f. For h, the image of ¹³C-labelled tissue is also shown in Fig. 2i. For i, the images corresponding to patient 7 are also shown in Fig. 2e. Abbreviations: UL (unlabelled). Source data

Extended Data Fig. 5. Transcriptional analysis of neurotransmitter signalling and serine metabolism genes in GBM and cortex.

a,b, UMAP visualization based on expression of all genes in the GSE165595 RNA-seq dataset (a) for cortex and GBM tissues and GSE59612 RNA-seq dataset (b) for cortex, enhancing, and non-enhancing tissues. Number of nearest neighbours in UMAP is 13 for (a) and 15 for (b). n represents number of samples for each group. c, Volcano plot of differentially expressed neurotransmitter genes in tumour compared to the cortex in both datasets. d,e, Transcriptional expression of neurotransmitter-related genes in GSE165595 (d) and GSE59612 (e). f, Volcano plot of differentially expressed serine synthesis and serine transporter genes in tumour compared to the cortex in both datasets. g,h, Transcriptional expression of serine-related genes in GSE165595 (g) and GSE59612 (h). In (c) and (f), Log2 fold change represents the ratio of mean expression of a gene in tumour over cortex. Positive Log2 fold change indicates higher expression in tumour. P-values were adjusted using Benjamini-Hochberg test with the false discovery rate <0.05 and log2 fold change threshold was set to 1. In d-e and g-h, expression values are shown in log2(1+transcript per million) unit. Abbreviations: UMAP (Uniform Manifold Approximation and Projection).

Extended Data Fig. 6. Dynamic incorporation of glucose carbons into nucleotide metabolites and in vivo modelling frameworks.

a, Time-dependent enrichment profiles of purine metabolites IMP, GDP, guanosine, AMP, and inosine in orthotopic GBM38-bearing mice infused with [U¹³C]glucose. Data are mean ± s.d. of n = 3–9 samples from 1–3 mice per timepoint. b, Time-dependent enrichment of pyrimidine metabolites UMP and uridine in orthotopic GBM38-bearing mice infused with [U¹³C]glucose. Data are mean ± s.d. of n = 3–9 samples from 1–3 mice per timepoint. c-d, Biochemical interconversions and model boundaries used for metabolic modelling of purine synthesis (c) and pyrimidine synthesis (d). e, Relative abundance of nucleotide species in cortical and GBM tissues from orthotopic tumour bearing mice. f, Schema of dynamic metabolic flux analysis pipeline. The diagram was created using BioRender. g-h, Isotope enrichment (g) and relative abundance (h) of purine metabolites IMP, GDP, guanosine, AMP, and inosine in tissues from orthotopic GBM38-bearing mice treated with a single dose of cranial RT (8 Gy) and then immediately infused with [U¹³C]glucose. Data are mean ± s.d. of n = 3–9 samples from 1–3 mice per timepoint. For a, b, and g, time-course enrichment data normalization to plasma glucose is detailed at https://github.com/baharm1/iMFA/. Abbreviations: PRPP (phosphoribosyl pyrophosphate), R1P (ribose 1-phosphate), Me-THF (10-formyltetrahydrofolate). Source data

Extended Data Fig. 7. Dynamic metabolic response to radiation therapy in GBM and cortex.

a, Fluxes through purine synthesis after RT estimated by analysing orthotopic tumour and cortex tissues from GBM38-bearing mice that were treated with whole cranial RT (8 Gy), immediately (<5 min post-RT) infused with [U¹³C]glucose and serially euthanized at varying timepoints following treatment. b-r, Flux values for the following reactions were approximated using time-course enrichment as described in Supplementary Methods: de novo IMP synthesis (b), conversion of IMP to GMP (c), conversion of IMP to AMP (d), conversion of PRPP + hypoxanthine to IMP (e), conversion of PRPP + guanine to GMP (f), conversion of R1P + guanine to guanosine (g), conversion of adenosine to AMP (h), conversion of GMP to GDP (i), GDP exit from model boundary (j), conversion of GMP to guanosine (k), conversion of guanosine + R1P to guanine (l), AMP exit from model boundary (m), conversion of R1P + hypoxanthine to inosine (n), conversion of IMP to inosine (o), conversion of inosine to R1P + hypoxanthine (p), conversion of adenosine to inosine (q), and conversion of AMP to IMP (r). Solid lines indicate estimated fluxes, and shaded regions indicate 95% confidence intervals. Fluxes were approximated using 3–9 samples per timepoint with samples originating from 1–3 mice per group and 3 samples per mouse. s, Levels of m + 4 IMP in orthotopic GBM38 tumours harvested from control or cranial RT (8 Gy)-treated mice infused with ¹⁵N₄-inosine. Data are mean ± s.d. of tumour samples from 4–5 mice per group containing 1–2 tumour foci each. t-v, Relative levels of de novo m + 3 GMP (t), de novo m + 2 IMP or (u) de novo m + 2 AMP (v) in orthotopic GBM38 tumours harvested from control or cranial RT (8 Gy)-treated mice infused with ¹⁵N-glutamine. Data are mean ± s.d. of tumour samples isolated from 3 mice per group containing 1–3 intracranial tumour foci each. In s-v, groups were compared by t-test. Abbreviations: PRPP (phosphoribosyl pyrophosphate), R1P (ribose 1-phosphate). Source data

Extended Data Fig. 8. Rewiring of glucose-derived serine synthesis and its environmental uptake in tumours.

a, Normalized (to labelled plasma glucose on a per-mouse basis) enrichment of serine in cortex and orthotopic GBM tissue isolated from intracranial tumour-bearing mice infused with [U¹³C]glucose. b, Normalized (to labelled plasma glucose on a per-patient basis) enrichment of serine in intracranial tissues from 8 glioma patients infused with [U¹³C]glucose. c, Serine isotopologue distributions (normalized to total serine labelling) in intracranial tissues from 8 human brain cancer patients infused with [U¹³C]glucose. d, Percent of serine that contains one (m + 1) or three (m + 3) tracer-derived ¹³C atoms in intracranial tissues from 8 glioma patients infused with [U¹³C]glucose. e, Ratios of m + 3 serine to m + 1 serine in intracranial tissues from 8 human brain cancer patients infused with [U¹³C]glucose. f, Percent of serine that contains one (m + 1) or three (m + 3) tracer-derived ¹³C atoms in cortex and orthotopic GBM tissue isolated from intracranial tumour-bearing mice infused with [U¹³C]glucose. g, Isotopic enrichment of serine in brain tumours from patients infused with [U¹³C]glucose were plotted from supplementary data reported by Courtney et al. h-j, Isotopic enrichment of serine in different paediatric tumour types including paediatric neuroblastoma (h), paediatric sarcoma (i), and a variety of other miscellaneous paediatric cancers (j) was previously reported as supplementary information by Johnston et al. and re-plotted here. k, The work of Courtney et al. includes supplementary information reporting enrichment of serine in clear cell renal cell carcinoma (CCRC) and adjacent kidney tissue. l, The work of Hui et al. includes ¹³C serine labelling information in multiple tissue types from mice infused with [U¹³C]glucose. m, Isotopologue distributions of plasma serine and tissue phosphoglycerate in cortex and orthotopic GBM tissue isolated from intracranial tumour-bearing mice infused with [U¹³C]glucose. n, Isotopologue distributions of plasma serine and tissue phosphoglycerate in intracranial tissues from 8 glioma patients infused with [U¹³C]glucose. o, Metabolic flux analysis model used to estimate scores of de novo serine synthesis to serine uptake ratios in brain cancer patients and orthotopic GBM-bearing mice infused with [U¹³C]glucose. The model was used to quantify serine acquisition routes in cortex and gliomas. p, Relative (to cortex) reliance on serine uptake compared to glucose-driven de novo serine synthesis in mice and humans infused with [U¹³C]glucose. Significance levels of tumour vs. adjacent cortex were tested by comparing 95% confidence intervals with p < 0.05 (indicated by asterisk) defined by non-overlapping groups. q, Ratio of contribution of de novo serine synthesis to serine uptake in human cortex and brain cancers. Error bars represent 95% confidence intervals. In a, b, d, f, m, and n, data are mean ± s.d. In a, b, d, and f, comparisons between groups were performed using linear mixed effects models with Holm’s correction. In a and f, n = 4–16 samples per group from 16 mice (4–7 mice for each orthotopic model with cortex samples from all mice pooled). In b and d, n = 7–8 samples per group. In g-l, data are mean ± s.d. Comparisons between groups were performed using linear mixed effects models. Source data

Extended Data Fig. 9. Differential metabolic effects of exogenous serine restriction on patient-derived GBM models and mouse cortex.

a, Abundance of serine, glycine, and phosphoglycerate in 374gs gliomaspheres labelled with [U¹³C]glucose in control media or media without serine and glycine. b, Mass isotopologue distributions of serine, glycine, and phosphoglycerate in 374gs gliomaspheres labelled with [U¹³C]glucose in control media or media without serine and glycine. In a and b, data are mean ± s.e.m. n = 9 samples per group from 3 independent experiments each with 3 samples per condition. c, Heatmap of relative nucleotide abundance in 374gs gliomaspheres under control or environmental serine/glycine-depleted conditions. Each column represents one of 9 biological replicates from 3 independent experiments each with 3 samples per condition. A white box containing an X indicates not detected in this replicate. d, Metabolite levels in cortical tissue from intracranial GBM38-bearing mice on a control or serine/glycine-restricted diet. e, Relative serine and phosphoserine levels in GBM38 tumours and cortex from mice fed control or Ser/Gly-restricted diets. Data are mean ± s.d. Comparisons between groups were performed by t-test. n = 4–5 mice per group. f, UMP:Ser label ratios in HF2303 tumours from mice on control or -SG diets. g, IMP:Ser label ratios in HF2303 tumours from mice on control or -SG diets. h, AMP:Ser label ratios in HF2303 tumours from mice on control or -SG diets. i, GMP:Ser ratios in HF2303 tumours from mice on control or -SG diets. f-i, Label ratios are defined as ratio (relative to control) of ¹³C-glucose-derived nucleotide enrichments (% labelled carbons) to glycolysis-derived m + 3 serine (% of total serine) in HF2303 tumours from mice infused with [U¹³C]glucose. Data are mean ± s.d. Data were generated from 3–4 mice per group with 1–2 tumour foci per mouse. Comparisons between groups were performed by two-sided t-test. j, Isotopologue distributions of labelled serine in intracranial GBM12 tumour-bearing mice infused with [U¹³C]glucose. k, Heatmap showing normalized (to labelled plasma glucose on a per-mouse basis) enrichment of biosynthesis intermediates in GBM12 tumours from intracranial GBM-bearing mice infused with [U¹³C]glucose. l, UMP:Ser ratios in GBM12 tumours from mice on control or -SG diets. m, IMP:Ser ratios in GBM12 tumours from mice on control or -SG diets. n, AMP:Ser ratios in GBM12 tumours from mice on control or -SG diets. o, GMP:Ser ratios in GBM12 tumours from mice on control or -SG diets. l-o, Label ratios are defined as ratio (relative to control) of ¹³C-glucose-derived nucleotide enrichments (% labelled carbons) to glycolysis-derived m + 3 serine (% of total serine) in GBM12 tumours from mice infused with [U¹³C]glucose. Data are mean ± s.d. Data were generated from 5 mice per group. Comparisons between groups were performed by two-sided t-test. Source data

Extended Data Fig. 10. Biochemical and physiological effects of dietary serine/glycine restriction in mice.

a, Abundance of serine in plasma from intracranial GBM-bearing mice fed either control diets or serine/glycine-restricted diets for 4 weeks. Data are mean ± s.e.m., and groups were compared by t-test. n = 21–25 mice bearing intracranial HF2303, GBM38 or GBM12 tumours per group from 5 independent experiments. b, Weights of mice fed control diets or serine/glycine-restricted diets. Data are mean ± s.e.m., and differences between control diet groups and serine/glycine-restricted diet groups for each sex were compared by t-test with Holm-Šídák correction. n = 5 mice per group. *p ≤ 0.05; **p ≤ 0.01, ***p ≤ 0.001. c-x, Complete blood counts (c-r) and serum chemistries (s-x) were analysed for C57BL/6 J mice fed either a control diet or serine/glycine-restricted diet for 9 months. c, Lymphocyte percentage of white blood cells and lymphocyte counts. d, Monocyte percentage of white blood cells and monocyte counts. e, Neutrophil percentage of white blood cells and neutrophil counts. f, Eosinophil percentage of white blood cells and eosinophil counts. g, Basophil percentage of white blood cells and basophil counts. h, White blood cell counts. i, White blood cell distributions. Data are mean ± s.d. j, Red blood cell counts. k, Platelet counts. l, Haemoglobin concentrations. m, Haematocrit. n, Mean corpuscular volume. o, Mean corpuscular haemoglobin. p, Mean corpuscular haemoglobin concentration. q, Mean platelet volume. r, Red blood cell distribution width. s, Creatinine. t, Blood urea nitrogen. u, Albumin. v, Sodium. w, Potassium. x, Chloride. c-x, Lines represent means. Groups were compared by t-test with n = 6 mice per group except for albumin measurements in panel u (n = 2–3 mice per group), in which experimentation was limited by sample volume and no statistical comparison could be made. Abbreviations: LYM (lymphocyte), MONO (monocyte), NEU (neutrophil), EOS (eosinophil), BAS (basophil), WBC (white blood cell), RBC (red blood cell), PLT (platelet), HGB (haemoglobin), HCT (haematocrit), MCV (mean corpuscular volume), MCH (mean corpuscular haemoglobin), MCHC (mean corpuscular haemoglobin concentration), MPV (mean platelet volume), RDW (RBC distribution width), CREA (creatinine), BUN (blood urea nitrogen), ALB (albumin), Na (sodium), K (potassium), Cl (chloride). Source data

Extended Data Fig. 11. Effects of dietary serine/glycine restriction on intracranial tumour burden in GBM-bearing mice.

a, Representative bioluminescence imaging of luciferase-positive HF2303 tumours grown orthotopically in mice on control or -SG diets. b, Representative H&E staining of HF2303 tumour-containing brain tissue from mice on control or -SG diets. One control image shows only tumour tissue (top left, uniform dark purple staining) due to tumour size exceeding field. c, Representative Ki-67 immunohistochemistry of intracranial HF2303 tumour tissue from mice on control or -SG diets and quantification of Ki-67 positivity. Quantitative data are mean ± s.d. with n = 8 mice per group. Comparisons between groups were performed by t-test. d, Representative bioluminescence imaging of luciferase-positive GBM38 tumours grown orthotopically in mice on control or serine/glycine-restricted diets. e, Representative H&E staining of GBM38 tumour-containing brain tissue from mice on control or -SG diets. f, Representative Ki-67 immunohistochemistry of intracranial GBM38 tumour tissue from mice on control or -SG diets and quantification of Ki-67 positivity. Quantitative data are mean ± s.d. with n = 4–5 mice per group. Comparisons between groups were performed by t-test. g, Representative bioluminescence imaging of luciferase-positive GBM12 tumours grown orthotopically in mice on control or serine/glycine-restricted diets. h, Representative H&E staining of GBM12 tumour-containing brain tissue from mice on control or -SG diets. i, Representative Ki-67 immunohistochemistry of intracranial GBM12 tumour tissue from mice on control or -SG diets and quantification of Ki-67 positivity. Quantitative data are mean ± s.d. with n = 4–6 mice per group. Comparisons between groups were performed by t-test. j-l, Chemoradiation treatment schemes for intracranial GBM-bearing mice on control or -SG diets used in therapeutic efficacy studies of Fig. 4i–k. Tumour models used were HF2303 (j), GBM38 (k), and GBM12 (l). Source data