Gliocidin is a nicotinamide-mimetic prodrug that targets glioblastoma

Abstract

Glioblastoma is incurable and in urgent need of improved therapeutics¹. Here we identify a small compound, gliocidin, that kills glioblastoma cells while sparing non-tumour replicative cells. Gliocidin activity targets a de novo purine synthesis vulnerability in glioblastoma through indirect inhibition of inosine monophosphate dehydrogenase 2 (IMPDH2). IMPDH2 blockade reduces intracellular guanine nucleotide levels, causing nucleotide imbalance, replication stress and tumour cell death². Gliocidin is a prodrug that is anabolized into its tumoricidal metabolite, gliocidin-adenine dinucleotide (GAD), by the enzyme nicotinamide nucleotide adenylyltransferase 1 (NMNAT1) of the NAD⁺ salvage pathway. The cryo-electron microscopy structure of GAD together with IMPDH2 demonstrates its entry, deformation and blockade of the NAD⁺ pocket³. In vivo, gliocidin penetrates the blood-brain barrier and extends the survival of mice with orthotopic glioblastoma. The DNA alkylating agent temozolomide induces Nmnat1 expression, causing synergistic tumour cell killing and additional survival benefit in orthotopic patient-derived xenograft models. This study brings gliocidin to light as a prodrug with the potential to improve the survival of patients with glioblastoma.

Extended Data Fig. 1 |. Gliocidin dynamics and mTORC1 relationship.

a and b, Volcano plots depicting transcriptomics of HTS cells (a) and MEFs (b) treated with 1 μM Gliocidin for 12 h. c, GO Biological Process enrichment analysis of HTS cells treated with 1 μM Gliocidin versus DMSO. d and e, GO Biological Function enrichment analysis of enriched (d) and depleted hits (e) Gliocidin CRISPR-Cas9 screen. f, GO Cellular Components enrichment analysis of Gliocidin CRISPR-Cas9 screen depleted hits. g, Rank order plot highlighting enrichment of sgRNAs for mTOR regulators in the IC₅₀ Gliocidin versus DMSO CRISPR-Cas9 screen. h, Schematic of mTOR regulators color coded by positive (red) or negative (blue) β-scores. Dashed line represents unresolved mechanism. i and j, sgRNA validation of selected top enriched (i) and depleted (j) mTOR regulators from the CRISPR-Cas9 KO screen. NG2-3112 cells were treated with 0.125 μM Gliocidin for six days prior to cell counting. Data indicate mean ± SD, n = 4 and n = 3 biological replicates for (i) and (j), respectively. k, Cell count of NG2-3112 cells with non-targeting (NT) control, Pfas, Ppat and Gart single KO or co-KO with Hprt over a six-day proliferation assay. Data indicate mean ± SD, n = 3 biological replicates. For panel i-k, two-tailed P-values were determined by paired t-test. ns: not significant.

Extended Data Fig. 2 |. Guanosine supplementation rescue of Gliocidin.

a, Time course study of murine GBM cells treated with 1 μM Gliocidin assayed for replication stress and mTORC1 activity. b, Gliocidin dose-response curves of non-targeting control (NT) or Hprt knockout NG2-3112 cells with or without 20 μM guanosine. P-values determined by comparison between guanosine treated to non-treated control. c, Gliocidin dose-response curves of PDX GBM cells with ribonucleosides. Statistical analyses determined by comparison with Gliocidin only treated samples. d, Western blot of Gliocidin-treated NG2-775 cells with guanosine supplementation. e, Cell counts of NG2-775 cells treated with 2 μM Gliocidin for 96 h. f, Fold change of steady state quantification of purine metabolites in Gliocidin-treated NG2-775 cells. Data indicate mean ± SD, n = 3 technical replicates. g, CRISPR-Cas9 screen rank order plot highlighting essentiality score (β-score) of de novo guanine nucleotide synthesis genes. h, Cell counts of control and Impdh1 knockout NG2-3112 cells over six days. i, Western blot validation of Gmps and Impdh2 overexpressing cells. j and k, IMPDH2 activity assay of Gliocidin and MPA plotted against activity (j) and absorbance at 340 nm (k). l, Gliocidin dose-response curve of NMNAT1 KD PDX-160329-1 cells. m, Western blot validation of NMNAT1 KD cells. n, LC-MS/MS heatmap of purine metabolite levels (fold changes) for NG2-3112 NT, Nrk1 and Nmnat1 KO cells treated with 6 h Gliocidin, n = 3 technical replicates (R1-3). o, Dose-response curves of NG2-3112 treated with MPA and Mizoribine. P-values determined by comparison to sgNT samples. For panels b, c, e, h, l and o, data indicate mean ± SD, n = 3 biological replicates; two-tailed P-values were determined by unpaired (f) and paired (e, h) t-test. For panels b, c, l and o, P-values were determined by two-way ANOVA. ****P < 1.0e-15, ns: not significant. All western blots were repeated three times independently.

Extended Data Fig. 3 |. A fraction of Gliocidin is converted into GAD.

a, c and e, Chemical structures of Gliocidin-R (a), Gliocidin-MN (c) and GAD (e) synthetic standards used for LC-MS. Observed mass is m/z found in NG2-3112 cells treated with Gliocidin. b, d and f, Extracted ion chromatograms of Gliocidin-R (m/z = 337.0853) (b), Gliocidin-MN (m/z = 417.0516) (d) and GAD (m/z = 746.1041) (f) in Gliocidin-treated NG2-3112 cells in comparison with synthetic standard. g, Comparison of MS/MS fragmentation spectra of Gliocidin-R, -MN and GAD between Gliocidin-treated cells to synthetic standard. h, in GAD spectrum, numbers highlighted in red are characteristic ions, assigned as 1: adenine, 2: Gliocidin, 3: AMP, 4: ADP, 5: Gliocidin-diphosphate ions. i and j, LC-MS/MS quantification of Gliocidin-R (i) and Gliocidin-MN (j) levels in NG2-3112 control and KO cells treated with vehicle or Gliocidin. k and l, LC-MS/MS quantification of Gliocidin-R (k) and GAD (l) levels in NG2-3112 cells treated with vehicle, Gliocidin, FK866 or combination of both drugs for 6 h. m-o, LC-MS/MS quantification of GAD (m), NR (n), NMN (o) levels in NG2-3112 NT and Nt5c3b KO cells treated with vehicle or Gliocidin. p-s, LC-MS/MS quantification of GAD (p), NAD⁺ (q), Gliocidin-MN (r) and Gliocidin-R (s) levels in NG2-3112 cells treated with Gliocidin over time. t and u, Dose-response curves of Gliocidin (t) and FK866-treated (u) NG2-3112 cells co-supplemented with guanosine or NMN. v and w, Concentration ratio of GAD to Gliocidin-MN (v) and NAD⁺ to GAD (w) over time in Gliocidin-treated NG2-3112 cells. Data indicate mean ± SD, n = 3 technical replicates. ****P < 1.0e-15, ns: not significant. Two-tailed P-values were determined by unpaired t-test. For t and u, n = 3 biological replicates, P-values were determined by two-way ANOVA. Statistical significance is determined by comparing to Gliocidin (t) and FK866-treated only (u) samples in corresponding panels.

Extended Data Fig. 4 |. NMNAT and NRK isozymes activate Gliocidin.

a. Dose-response curve for NG2-3112 cells treated with MPA and Mizoribine. b and c, Dose-response curves for PDX-160329-1 (b) and PDX-180911-1 (c) vector control (Ctrl) and human Nmnat1 (hNmnat1) overexpressed (OE) cells treated with Gliocidin, MPA or Mizoribine. d, Western blot validating high human NMNAT1 expression in GBM PDX cells. e and f, Quantification of Gliocidin-R and -MN levels in NG2-3112 NMNAT1 OE cells by LC-MS/MS. Cells were treated with vehicle or 2.5 μM Gliocidin for 10 h. g, Dose-response curves of vector control, human NMNAT1, NMNAT2 and NMNAT3 overexpressed NG2-3112 cells treated with Gliocidin. h, Western blot validating human NMNAT1, NMNAT2 and NMNAT3 expression in NG2-3112 cells. i-k, Quantification of GAD, -MN and -R levels in NG2-3112 cells overexpressing NMNAT1, NMNAT2 and NMNAT3 by LC-MS/MS. Cells were treated with vehicle or 2.5 μM Gliocidin for 10 h. l, Western blot validating mouse NRK1 and NRK2 overexpression in NG2-3112 cells. m-o, LC-MS/MS quantification of GAD, -MN and -R levels in NG2-3112 cells overexpressing Nrk1 and Nrk2. All data indicate mean ± SD. ****P < 1.0e-15, ns: not significant. For panel a-c and g, n = 3 biological replicates and P-values were determined by two-way ANOVA. Statistical significance is determined by comparing to corresponding control samples. For panel e-f, i-k, and m-o, n = 3 technical replicates and two-tailed P-values were determined by unpaired t-test. All western blots were repeated in three times independently.

Extended Data Fig. 5 |. Gliocidin-R conserves Gliocidin MoA and requires nucleoside transporter for cell entry.

a, Dose-response curves for NG2-3112 cells treated with Gliocidin and Gliocidin-R. b, Dose-response curves for NG2-3112 cells treated with Gliocidin-R co-supplemented with ribonucleosides. c, Dose-response curves for NG2-3112 non-targeting (NT) control and Nmnat1 KO cells treated with Gliocidin-R. d and e, Dose-response curves for PDX-180911-1 (d) and PDX-160329-1 (e) control or human Nmnat1 cDNA overexpressed cells (hNmnat1 OE) treated with Gliocidin-R. f and g, Dose-response curves for NG2-3112 NT, Slc28a1, Slc28a2, Slc28a3, Slc29a1, Slc29a2 and Slc29a3 KO cells treated with Gliocidin-R (f) and Gliocidin (g). h-m, LC-MS/MS quantification of Gliocidin-R levels in NG2-3112 NT and nucleoside transporter gene KO cells treated with 0.3125 μM Gliocidin-R for three hours. (n) Quantification of Gliocidin levels in NG2-3112 NT and Slc29a1 KO cells treated with 0.3125 μM Gliocidin for three hours. All data indicate mean ± SD, n = 3, biological (a-g) and technical replicates (h-n). ****P < 1.0e-15, ns: not significant. For panel a-g, P-values were determined by two-way ANOVA. Statistical significance was determined by comparing to corresponding control samples. For panel h-n, two-tailed P-values were determined by unpaired t-test.

Extended Data Fig. 6 |. GAD is not a general NAD⁺ inhibitor.

a, Human IMPDH2 activity assay of different control and drug treatment conditions plotted against absorbance at 340 nm and time (minutes). b, Human NAD⁺ dependent G6PD activity assay of G6PDi-1 (positive control), Gliocidin, -R, -MN and GAD. c, Human G6PD activity assay of different control and drug treatment conditions plotted against absorbance at 340 nm and time (minutes). d, Human NADPH dependent DHFR activity assay of MTX (positive control), Gliocidin, -R, -MN and GAD. Activity calculated by rate of decrease in 340 nm absorbance due to the consumption of NADP⁺ to produced NADPH. e, Human DHFR activity assay of different control and drug treatment conditions plotted against absorbance at 340 nm and time (minutes). f, Representative cryo-EM micrograph of IMPDH2 mixed with IMP and GAD (scale bar, 50 nm). g, Fourier shell correlation (FSC) of the two unfiltered half-maps from non-uniform refinement is colored black and of the two half-maps following density modification in blue. The map versus model FSC with the density modified map is colored red. The FSC 0.143 and 0.5 cutoffs are indicated by dotted lines. h, Angular distribution plot of the IMPDH2 reconstruction. i, Cryo-EM map of IMPDH2 colored by local resolution. j, Electron density (contoured to 6σ) corresponding to IMP and C331, indicating the absence of a covalent bond between the C331 side chain and the C2 position of IMP. k, Close-up view of the IMP and GAD binding pocket of IMPDH2. l, Electron density (contoured to 6σ) of IMP, GAD and of amino acid residues and water molecules that coordinate GAD within the IMPDH2 active site.

Extended Data Fig. 7 |. Blood-brain-barrier penetrance of Gliocidin.

a, Pharmacokinetic study of mice injected with a single dose of Gliocidin (50 mg/kg I.P.) and quantified for Gliocidin in plasma and brain over time. b, Pharmacokinetic parameters of Gliocidin in plasma and brain. c, Pharmacokinetic study as in panel a and quantified for Gliocidin-Riboside in plasma and brain over time. d-f, Biodistribution study of mice treated with a single dose of Gliocidin (50 mg/kg I.P.). Organs and tissues were harvested at 6, 24 and 48 h after drug administration. Gliocidin (d) and Gliocidin-R (e) were measured by LC-MS. Gliocidin-MN (f) was only detected in the small intestine for up to 24 h. All data indicate mean ± SD. For each time point in d-f, tissues of n = 3 independent mice were harvested for analysis from a single experiment. In cases where fewer than three individual values are shown per tissue per time point, Gliocidin metabolites were not detectable. g, Body weight measurement of wildtype C57BL/6 mice administered with vehicle or Gliocidin over a 28-day period (50 mg/kg, I.P., two doses per day; n = 5 per treatment group). P-values were determined by two-way ANOVA; ns: not significant.

Extended Data Fig. 8 |. No GAD impairment of peripheral immune and hematopoietic stem cell compartments.

Ten wildtype C57BL/6 mice n = 5 mice per group) dosed with vehicle or Gliocidin (50 mg/kg, I.P., two doses per day) for seven days were harvested for immune and hematopoietic subpopulation analysis. a, Splenic immune population analyzed for total cellularity, b, neutrophils and monocytes, c, B-cells, d, CD4 and e, CD8 T-cell subsets. f, Bone marrow analyzed for total cellularity; g, enumeration of LSK (HSC-enriched) cell subsets (LT HSC: Lineage⁻Sca-1⁺cKit⁺ CD34⁻CD48⁻Flt3⁻CD150⁺; ST HSC: Lineage⁻Sca-1⁺cKit⁺ CD48⁻Flt3⁻CD150⁻; MPP2: Lineage⁻Sca-1⁺cKit⁺ CD48⁺Flt3⁻CD150⁻; MPP3: Lineage⁻Sca-1⁺cKit⁺ CD48⁺Flt3⁻CD150⁺; MPP4/LMPP: Lineage⁻Sca-1⁺cKit⁺ Flt3⁺CD150⁻); h, myeloid progenitor (MP) cell subsets (CMP: Lineage⁻IL7Ra⁻Sca-1⁻Kit⁺CD34⁺Fcgr^lo; GMP: Lineage⁻IL7Ra⁻Sca-1⁻Kit⁺CD34⁺Fcgr^hi; MEP: Lineage⁻IL7Ra⁻Sca-1⁻Kit⁺CD34⁻Fcgr^lo); and i, common lymphoid progenitor cells (CLP: Lineage⁻IL7Ra⁺Flt3⁺Sca^mid/lo Kit^lo) All data indicate mean ± SD. ns: not significant. Two-tailed P-values were determined by unpaired t-test. LSK: Lin-Sca1⁺c-Kit⁺, CMP: common myeloid progenitors, GMP: Granulocyte-monocyte progenitors, MEP: megakaryocyte/erythrocyte progenitors, MPP: multipotent progenitors, ST-HSC: short-term HSC, LMPP: lymphoid-primed multipotent progenitors.

Extended Data Fig. 9 |. GAD activity in immunocompromised setting.

a, T2-weighted brain MRIs were performed on vehicle and Gliocidin-treated athymic nude mice at day 0 and day 15 of treatment. Tumor regions are highlighted in red dashed lines. b, Kaplan-Meier curves of NST-1329 orthotopic GBM mice (athymic nude) administered with vehicle or Gliocidin (50 mg/kg, I.P., two doses per day). P-value was determined by Log-rank test. Animal numbers and median survival are labeled in parentheses.

Extended Data Fig. 10 |. Gliocidin/Temozolomide combination attenuates GBM transcriptional signatures.

a, Experiment schematic for single cell RNA seq in Fig. 5d. Gliocidin and TMZ are administered accordingly at indicated time points for corresponding treatment groups. b, UMAP projections of integrated dataset from pooled samples (n = 3 in each group, total of 12 samples) of Vehicle (Veh), TMZ, Gliocidin and Combo treatment groups. Total of 84,647 cells (Veh: 24,637, TMZ: 21,413, Gliocidin: 21,075 and Combo: 17,522 cells). Cell clusters are defined by GS (green): Glioma-stem cell, Mix (gray): undefined mixed population, PS (red): proliferation S-phase, PM (orange): proliferation M-phase, OL (blue): oligodendrocytic lineage. c, Visualization of integrated cells of interest in each of its corresponding clusters shown as UMAP projection. Enriched cells were identified with gene signatures of interest and AUCell automatic thresholds. d, Summary of enriched cells of interest shown in panel c expressed as percentage of its corresponding treatment group. e, Frequency histogram depicting distribution of cells for log(NMNAT1 expression score+1). Cells with score >0.025 are defined as NMNAT1^high (colored in red on UMAP depiction) and <0.025 as NMNAT1^low (colored in yellow on UMAP depiction). f, Western blot of PDX-161010-1 cells treated with Temozolomide (TMZ) and blotted for NMNAT1 expression. g, Cell count assay of PDX-161010-1 cells treated with DMSO, 200 μM TMZ, 5 μM Gliocidin and combination of both drugs for 96 h. Data indicate mean ± SD, n = 3 biological replicates. Two-tailed P-values were determined by paired t-test. h, Gliocidin dose-response curve of vector-control and NMNAT1 OE PDX-160403-1-R. Data indicate mean ± SD. ****P < 1.0e-15. Statistical analyses were performed using two-way ANOVA. i, Western blot validating human NMNAT1 overexpression in GBM PDX-160403-1-R cells. All western blots were repeated three times independently.

Fig. 1 |. Anti-GBM activity of gliocidin and genome-wide CRISPR–Cas9 knockout screen.

a, Chemical structure of gliocidin (N-(pyridin-3-yl) thiophene-2-carboxamide). b, Gliocidin dose–response curves of early-passage primary HTS (GBM) cells and primary MEFs. Data are shown as the mean ± s.d., n = 3 biological replicates. c, Western blot analysis of mouse HTS (GBM) cells and MEFs treated with 1 μM gliocidin for 24 and 48 h assayed for apoptosis activity. c, cleaved protein. d, Schematic of the experimental workflow for the genome-wide CRISPR–Cas9 screen conducted in NG2-cre-derived 3112 mouse GBM cells. e, Merged plot of positively (red) and negatively (blue) enriched sgRNAs, with enrichment shown as the β score for IC₅₀ (x axis) and IC80 (y axis). Gliocidin-treated cells were compared with vehicle (DMSO)-treated cells. Relevant enriched (Nmnat1, Nrk1, Pfas, Ppat, Gart, Mios and Lamtor2) and depleted (Nt5c3b, Tsc2, Nprl2 and Nprl3) sgRNAs are highlighted. f, Western blot analysis of downstream mTOR effectors and apoptosis markers in NG2-3112 cells treated with 2 μM gliocidin for 48 h with or without 20 nM rapamycin. p, phosphorylated. g, Live cell ratio of gliocidin- versus DMSO-treated cells with knockout of individual DNPS genes alone or in combination with Hprt over a 6-day course. NT, non-targeting. Data are shown as the mean ± s.d., n = 3 biological replicates. Two-tailed P values were determined by paired t test; NS, not significant. All western blots were repeated three times independently.

Fig. 2 |. Gliocidin targets the guanine nucleotide dependence of GBM cells.

a, Schematic of the de novo (pink) and salvage (blue) purine synthesis pathways. PRPP, phosphoribosyl pyrophosphate. b,c, Dose–response curves of gliocidin-treated NG2-3112 (b) and NST-1329 (c) GBM cells supplemented with ribonucleosides. Data are shown as the mean ± s.d., n = 3 biological replicates. d,e, Western blot analysis of gliocidin-treated NG2-3112 (d) and NST-1329 (e) cells with or without guanosine supplementation assayed for replication stress activity. Experiments were repeated three times independently. f,g, Relative cell counts of NG2-3112 (f) and NST-1329 (g) cells after 96 h of gliocidin treatment, with or without guanosine supplementation. Data are shown as the mean ± s.d., n = 3 biological replicates. h,i, Steady-state quantification of the indicated metabolites in NG2-3112 (h) and NST-1329 (i) cells treated with 2 μM gliocidin or DMSO for 6 h, measured by LC–MS/MS. Data correspond to the fold-change differences in the respective metabolite between gliocidin-treated and DMSO-treated cells and are shown as the mean ± s.d., n = 3 technical replicates. j, Schematic summary of the purine metabolite changes following treatment with gliocidin. k, Relative cell counts for NG2-3112 cells with non-targeting (NT) control sgRNA or sgRNAs against Impdh2 and Gmps over a 6-day proliferation assay. Data are shown as the mean ± s.d., n = 3 biological replicates. l, Dose–response curves of NG2-3112 cells with vector control (Ctrl) or Gmps or Impdh2 overexpression (OE) treated with gliocidin. Data are shown as the mean ± s.d., n = 3 biological replicates. ****P < 1.0 × 10⁻¹⁵. For b, c and l, statistical analyses were performed using two-way ANOVA; for f, g and k, two-tailed P values were determined by paired t test; and for h and i, two-tailed P values were determined by unpaired t test.

Fig. 3 |. The gliocidin prodrug is activated by NAD⁺ salvage enzymes.

a, Rank-order plot of gliocidin gene dependence showing enriched and depleted sgRNAs following treatment with IC₈₀ gliocidin versus DMSO in the CRISPR–Cas9 screen. b, Schematic of the NAD⁺ salvage pathway. c, Dose–response curves for NG2-3112 cells with non-targeting control sgRNA or sgRNAs against Nrk1, Nmnat1 and Nmnat2 treated with gliocidin. Data are shown as the mean ± s.d., n = 3 biological replicates. d, Western blot analysis of NG2-3112 cells with non-targeting sgRNA or sgRNAs against Nrk1 and Nmnat1 treated with 1 μM gliocidin for 6 h. Cell lysates were blotted for p-CHK1, CHK1, NMNAT1 and NRK1. Experiments were repeated three times independently. e, Diagram of the proposed gliocidin drug metabolism pathway. f, LC–MS/MS quantification of GAD levels in NG2-3112 cells with non-targeting sgRNA or sgRNAs against Nrk1 and Nmnat1 treated with vehicle or gliocidin. a.u., arbitrary units. g, LC–MS/MS quantification of gliocidin-MN levels in NG2-3112 cells treated with vehicle, 2.5 μM gliocidin, FK866 or a combination of both drugs for 6 h. h,i, LC–MS/MS quantification of gliocidin-R (h) and gliocidin-MN (i) levels in NG2-3112 cells with non-targeting sgRNA or sgRNA against Nt5c3b treated with vehicle or gliocidin. j, LC–MS/MS quantification of GAD levels in NG2-3112 cells with vector control or overexpression of human NMNAT1 that were treated with vehicle or gliocidin. k, Dose–response curves of NG2-3112 cells with vector control or overexpression of human NMNAT1 that were treated with gliocidin. Data are shown as the mean ± s.d., n = 3 technical (f–j) and n = 4 biological (k) replicates. ****P < 1.0 × 10⁻¹⁵. For c and k, P values were determined by two-way ANOVA; for f–j, two-tailed P values were determined by unpaired t test.

Fig. 4 |. GAD occupies the IMPDH2 NAD⁺-binding pocket.

a, Assay measuring the inhibitory activity of MPA (positive control), gliocidin, gliocidin-R, gliocidin-MN and GAD against human IMPDH2. b, Electron density corresponding to IMP and GAD contoured to 6σ. c, Interactions of GAD (magenta) with IMP and IMPDH2 residues (yellow); A46 and Q469 belonging to a neighbouring protomer are marked with an asterisk. Hydrogen-bond interactions with active site residues are represented by thick dashed lines; interactions shared with NAD⁺ are shown in black, while the hydrogen bond with D274 unique to GAD is shown in green. Hydrogen-bond interactions coordinated by water molecules (red spheres) are indicated by narrow black dashed lines. d, Conformational differences between NAD⁺ and GAD. Monomers of the NAD⁺-bound (grey) and GAD-bound (yellow, with GAD in magenta) structures are superimposed on their main chain atoms. The arrow indicates displacement of the pyridine ring of GAD relative to the corresponding ring in NAD⁺. Residues 324–343 and 437–444 have been omitted for clarity. e, Close-up view of the π–π stacking interaction between IMP and the pyridine ring of NAD⁺ that facilitates hydride transfer. f, Close-up view of IMP and GAD in the same region as in e.

Fig. 5 |. Gliocidin prolongs survival of mice with orthotopic GBM derived from PDX cells.

a, T2-weighted brain MRI of mice performed at day 0 (before treatment) and day 21 (21 days after treatment). Tumour regions are highlighted in red dashed lines. b, Kaplan–Meier curves of mice with orthotopic NST-1329 GBM (C57BL/6) administered vehicle, TMZ, gliocidin or a combination of gliocidin and TMZ. P values were determined by log-rank test. c, Kaplan–Meier curves of mice with orthotopic PDX-161010-1 GBM treated with vehicle, gliocidin, TMZ or a combination of TMZ and gliocidin. P values were determined by log-rank test. d, Uniform manifold approximation and projection (UMAP) plots of NMNAT1^high (red) and NMNAT1^low (yellow) cells for each treatment group. The number (red) at the top right of each UMAP plot represents the abundance (percentage) of NMNAT1^high cells in each treatment group. e, Ex vivo GFP fluorescence imaging of dissected brain from mice with PDX-160403-1-R tumours overexpressing NMNAT1 treated with vehicle (n = 4) or gliocidin (n = 4; 50 mg kg⁻¹, i.p., two doses per day) for 7 days. f, Quantification of the signal intensity in e. Bar plot data are shown as the mean ± s.d. Total flux was quantified. The two-tailed P value was determined by Mann–Whitney test. g, Kaplan–Meier curves of mice with orthotopic PDX-160403-1-R GBM with vector control or NMNAT1 overexpression treated with vehicle or gliocidin (50 mg kg⁻¹, i.p., two doses per day). P values were determined by log-rank test. For b, c and g, animal numbers and median survival are included in brackets.