Adaptive rewiring of purine metabolism promotes treatment resistance in H3K27M-mutant diffuse midline glioma

Abstract Background: Diffuse midline gliomas (DMG), including diffuse intrinsic pontine gliomas (DIPGs), are a fatal form of brain cancer. These tumors often carry a driver mutation on histone H3 converting lysine 27 to methionine (H3K27M). DMG-H3K27M are characterized by altered metabolism and resistance to standard of care radiation (RT), but how the H3K27M mediates the metabolic response to radiation and consequent treatment resistance is uncertain. Methods: We performed metabolomics on irradiated and untreated H3K27M isogenic DMG cell lines and observed an H3K27M-specific enrichment for purine synthesis pathways. We profiled the expression of purine synthesis enzymes in publicly available patient data and in our models, quantified purine synthetic flux using stable isotope tracing, and characterized the in vitro and in vivo response to de novo and salvage purine synthesis inhibition in combination with RT. Results: DMG-H3K27M cells activate purine metabolism in an H3K27M-specific fashion. In the absence of genotoxic treatment, H3K27M-expressing cells have higher relative activity of de novo synthesis and lower activity of purine salvage due to decreased expression of the purine salvage enzymes. Inhibition of de novo synthesis radiosensitized DMG-H3K27M cells in vitro and in vivo. Irradiated H3K27M cells adaptively upregulate purine salvage enzyme expression and pathway activity. Silencing the rate limiting enzyme in purine salvage, hypoxanthine guanine phosphoribosyl transferase (HGPRT) when combined with radiation markedly suppressed DMG-H3K27M tumor growth in vivo. Conclusions: H3K27M expressing cells rely on de novo purine synthesis but adaptively upregulate purine salvage in response to RT. Inhibiting purine salvage may help overcome treatment resistance in DMG-H3K27M tumors.


Background
Diffuse midline gliomas (DMG) are pediatric high-grade gliomas that arise in midline structures of the brain including the thalamus, cerebellum, and pons (1)(2)(3).In 2016, the World Health Organization described a subtype of DMG that carries a driver missense mutation in the tail domain of Histone H3 that converts the 27th residue from lysine (K) to methionine (M), termed H3K27M (1,4).Patients with DMG-H3K27M carry a dire prognosis with over 90% dying within 2 years of diagnosis (2).Treatment options are limited for patients with DMG-H3K27M tumors.Surgical resection is often unfeasible due to the eloquent function of the midline tissues in which they arise (2,3,5).DMG-H3K27M tumors also derive minimal bene t from systemic therapies (3,(5)(6)(7).Radiation therapy (RT) is currently the only treatment modality that provides meaningful bene t to patients and is the current standard of care(8-12).However, RT typically only extends patient survival by months, and tumors regrow within the high-dose radiation eld(8, 12,13).This suggests that DMG-H3K27M tumor cells can effectively adapt and resist RT, leading to regrowth.Simply increasing the RT dosage given is not feasible owing to limiting toxicity in normal tissue.Therefore, there is a critical need to selectively increase the sensitivity of DMG-H3K27Ms to RT.
The importance of the H3K27M mutation and its profound effect on metabolism suggest that this mutation may speci cally confer RT resistance in DMG-H3K27M tumors.Here, we identify purine metabolism as an H3K27M-speci c metabolic vulnerability that governs the RT response.Cells can synthesize the two main classes of purines (adenylates and guanylates) de novo from metabolic building blocks or salvage them from pre-formed nucleobases.We nd that H3K27M-expressing cells preferentially utilize de novo guanylate synthesis during unperturbed growth, likely due to increased expression of the enzymes utilized in de novo synthesis and decreased expression of those used in salvage.Inhibitors of de novo guanylate synthesis sensitize DMG-H3K27M models to RT in vitro and in vivo but do not cure tumors.Unexpectedly, we nd that H3K27M cells upregulate guanylate salvage in response to RT, which likely mediates resistance to inhibition of de novo purine synthesis.Knockdown of the rate limiting enzyme in guanylate salvage greatly increases RT e cacy in orthotopic H3K27M xenografts.These ndings indicate that guanylate salvage inhibition may be a promising strategy to overcome RT resistance in DMG-H3K27M.
Successful transduction was con rmed by GFP expression or puromycin selection and loss of HGPRT protein.

Histone Puri cation
Histone puri cation was performed as described previously (24).H3K27M isogenic cells (> 1.0x10^6) were suspended in 1mL hypotonic lysis buffer and lysed by mechanical shearing on a rotator at 4C. Nuclei were isolated and suspended in 0.4N H 2 SO 4 overnight to precipitate histones.Samples were clari ed and trichloroacetic acid was added dropwise to the histone-containing supernatant, which were mixed and incubated on ice.Histones were pelleted, washed with ice-cold acetone, and air-dried.Histones were resuspended in ddH2O.Protein content was quanti ed using a BCA assay.

Steady-State Metabolomics
H3K27M-isogenic cells (2-3x10^6 cells/biological replicate) were plated in 10cm dishes in TSM media and were grown to be small neurospheres (approximately 2 days in culture).Media was changed 2hrs before treatment.Cells were irradiated with 4Gy RT using a Philips RT250 (Kimtron Medical) at a dose rate of approximately 2 Gy/minute at the University of Michigan Rogel Comprehensive Cancer Center Experimental Irradiation Shared Resource (Ann Arbor, MI).Cells were incubated for 2hrs at 37°C.One plate from each group was analyzed by BCA assay quantify protein for standardization.After incubation, all cells were collected in individual 15mL conical tubes and pelleted.Supernatants were removed and metabolites were extracted with ice-cold 80% methanol on dry ice.Pellets were resuspended and incubated for on dry ice.Samples were centrifuged to clarify the metabolite extract and supernatants were transferred to new 15mL tubes.This step was repeated twice to ensure clarity.Metabolite extracts equating to roughly 1000µg of protein were transferred to 1.5mL microcentrifuge tubes before drying in a speed vacuum centrifuge or under a nitrogen blower.Liquid chromatography/mass spectrometery (LC/MS) analyses were performed as previously described(28).Agilent MassHunter Quantitative Analysis B.09.00 QqQ software was used to integrate and quantitate areas (Agilent Technologies).

Steady-State Metabolomics Data Analysis
Metabolite lists were trimmed to exclude metabolites that fell below the noise threshold.Remaining metabolite abundances were median centered.H3K27M isogenic cells were analyzed in pairs.All metabolites whose abundance changed by a Log2 FC of |0.15| following RT in either H3K27M or H3K27M-KO cells were selected and combined into a single list.These lists were ordered by the absolute value of difference in post-RT Log2FC between K27M and KO cells.The top 25 metabolites were selected for pathway enrichment analysis using MetaboAnalyst 5.0 where the top 10 most signi cantly represented pathways were selected (29).Statistical analyses and heatmap construction were performed using GraphPad Prism 9.0.

Stable Isotope Tracing
H3K27M isogenic cells (> 4.0x10^6 cells/replicate) were plated in 10cm dishes in TSM and grown into small neurospheres.Approximately 30min before RT, cells were pelleted and washed and replated with TSM media with stable isotope tracer molecules replacing the naturally occurring metabolites (2mM 2,8deuterium hypoxanthine (2D-Hpx) or 4mM 15 N-glutamine [Cambridge Laboratories Inc, Cat#: DLM-8658 and NLM-557, respectively]).One replicate plate from each group was set aside and given unlabeled TSM media to act as unlabeled control samples.Untreated and irradiated cells (4Gy) were incubated at 37°C for 3hrs after RT at which point metabolites were extracted.LC/MS Analysis was performed on an Agilent system consisting of an In nity Lab II UPLC with a 10-port valve coupled with a 6545 QToF mass spectrometer (Agilent Technologies) using a JetStream ESI source in negative mode.Source parameters: Gas Temp: 250 °C, Gas Flow: 13 L/min, Nebulizer: 35 psi, Sheath Gas Temp: 325 °C, Sheath Gas Flow: 12 L/min, Capillary: 3500 V, Nozzle Voltage: 1500 V.
Chromatographic separation was performed on an Agilent ZORBAX RRHD Extend 80Å C18, 2.1 × 150 mm, 1.8 µm column with an Agilent ZORBAX SB-C8, 2.1 mm × 30 mm, 3.5 µm guard column.The column temperature was 35 °C.Mobile phase A consisted of 97:3 water/ methanol and mobile phase B was 100% methanol; both A and B contained tributylamine and glacial acetic acid at concentrations of 10mM and 15mM, respectively.The column was back ushed with mobile phase C (100% acetonitrile, no additives) between injections for column cleaning.
The LC gradient was as follows: 0-2.5 min, 0% B; 2.5-7.5 min, linear ramp to 20% B min, 7.5-13 min, linear ramp to 45% B; 13-21 min linear ramp to 99% B and held at 99% B until 25 min.At 25 minutes, the 10-port valve was switched to reverse ow (back-ush) through the column, and the solvent composition changed to 95% C and held there for 3 min.From 28 to 28.5 min, the ow rate was ramped to 0.8 mL/min, held until 32.5 min, then reduced to 0.6mL/min.From 32.5 to 33.25 the solvent was ramped from 99-0% C while ow was simultaneously ramped down from 0.6-0.4mL/minand held until 39.9 min., at which point ow rate was returned to starting conditions at 0.25mL/min.The 10-port valve was returned to restore forward ow through the column at 40 min.An isocratic pump was used to introduce reference mass solution through the reference nebulizer for dynamic mass correction.Total run time was 30 min.The injection volume was 5µL.Data was analyzed using MassHunter ProFinder 8.0 software (Agilent Technologies).
Tumor expression data mRNA-seq Z-score expression and patient survival data from Mackay et al. (2017) was obtained using PedCBioPortal (Children's Hospital of Philadelphia Research Initiative) (2,30,31).Data was ltered for patients who had con rmed WT Histone H3 or H3K27M mutant (H3F3A or HIST1H3B) and the Z-score for each gene of interest was extracted and sorted based on mutational status.Statistical analyses were performed using a two-tailed t-test in GraphPad Prism 9.0.
Long-term neurosphere assay and live cell imaging H3K27M cells (3x10^5-4x10^5) were cultured in 6-well dishes for 48hrs until they formed small spheres which were then treated with mycophenolic acid (MPA) for 6hrs prior to irradiation in the previously described apparatus.Approximately 18hrs later, spheres were dissociated and replated at low density (500 cells/well) in 96-well plates.Plates were incubated for 5-10 days in a BioSpa incubator (Agilent Technologies) and imaged every 24hrs in a Cytation 5 plate reader with the 4x objective.Gen5 software (Agilent Technologies) was used to analyze the data.A 2D area threshold of 2600um^2 was used to pick spheres, approximating the 50-colony threshold used in standard clonogenic survival assays.End-point sphere numbers were used to calculate the surviving fraction of cells.The Dmid of the 0µM MPA control was divided by the Dmid of the MPA-treated groups to calculate the Enhancement Ratio (ER) to display changes in RT e cacy.

Stereotactic orthotopic implantation
Rag1-KO C57BL/6 mice were anesthetized and provided carprofen analgesic before removing the scalp fur and sterilizing the incision site.An incision was made along the midline of the scalp and a small hole in the skull was made using an electric burr hole drill or handheld drill bit at coordinates 2-3mm lateral, 0.5-1mm rostral or caudal (depending on the injection rig used) from the bregma.Approximately 1.5x10^5-2.0x10^5cells (~ 5.0x10^4 cells/µL) in a volume of 3µL were implanted in the cortex at depth of approximately 2.5mm using 10µL Hamilton syringes.After injections, wounds were sutured and were given triple antibiotic ointment.Atipamezole was given reverse the anesthetic in heated cages.Mice were given diet gel supplement and monitored for 10 days, including a second dose of carprofen the day following the surgery.For some experiments, a PhD Ultra multi-syringe pump (Harvard Apparatus) was utilized to facilitate simultaneous implantation of up to 8 mice at once.A minimum of 6 mice were used per treatment group in subsequent in vivo experiments.

Bioluminescent imaging (BLI) and mouse treatment
Mice were administered sterile-ltered 30mg/mL luciferin solution (Syd Laboratories) via IP injection.Ten minutes post-injection, BLI signal was measured using an IVIS Spectrum imaging station and Luminescent Flux Values were obtained for each tumor.BLI ux values were used to randomize mice into four treatment groups (10 mice/group) based on BLI ux average signal.BLI was repeated 1-2x weekly during e cacy studies.Treatment groups included: Vehicle Control (0.5% w/v methylcellulose, 0.1% Tween-80 in ddH2O, mycophenolate mofetil (MMF) alone, RT alone, and MMF + RT.MMF (150mg/kg) was administered by oral gavage for 11 consecutive days starting between 13-18 days postimplantation.Mice were sedated using 2.5% iso urane and RT was administered in 2Gy fractions on days 2, 4-5, 8-9, and 11 of MMF treatment using an Orthovoltage irradiator.Mice were euthanized upon development of neurological symptoms and survival data was analyzed using the Kaplan-Meier method using GraphPad Prism 9.0.

Results
The H3K27M mutation confers global changes to cellular metabolism.
Altered cellular metabolism is a hallmark of cancer (32).To begin to understand how the H3K27M mutation alters metabolism and facilitates treatment resistance, we quanti ed how metabolite levels differed between normal brain and DMG-H3K27M xenograft tumor tissue.Using uorescence-guided microdissection, we separated GFP and luciferase-expressing DIPGXIII (DIPGXIII-GFP/LUC) orthotopic tumors from surrounding normal brain and quanti ed their metabolites.The brain-speci c metabolite Nacetylaspartate was higher in cortex compared to DMG tumors, indicating that our uorescent-based separation was successful (Fig S1A).Numerous other metabolites including dGTP/ATP, citrate/isocitrate, and UDPglcNAC were elevated in DMG while adenine, ureidosuccinate, and aspartate were lower indicating that DMG tumor tissue possesses a distinct metabolome compared to normal brain (Fig. 1A).To understand the biology of these dysregulated metabolites, we performed metabolic pathway enrichment analysis and found enrichment of pyrimidine and methionine metabolism in DMG-H3K27M tumors (blue bars), consistent with recent reports of the importance of these pathways in H3K27M tumors, as well as purine metabolism (Fig. 1B) (25,26,29).Some of these metabolic alterations might be caused speci cally by the H3K27M mutation(24) while others might be related to alternative mutations or oncogenic transformation in general.To understand how the H3K27M mutation speci cally affected metabolism, we used isogenic patient-derived DMG-H3K27M cell line pairs (DIPGXIII and BT245) in which the H3K27M mutation has been removed from the parental cell line using CRISPR/Cas9 (DIPGXIII-KO and BT245-KO) (20).We con rmed the presence of the mutation in parental cells and the corresponding lack of H3K27me3 (Fig S1B).We quanti ed metabolites in these isogenic pairs using LC/MS and found that H3K27M-expressing cells have an altered metabolome compared to H3K27M-KO counterparts (Fig. 1B, S2, and Supplemental Table 1).These ndings are consistent with prior studies and con rm that the oncohistone plays an active role in altering DMG metabolism (24).
RT induces changes in purine metabolism in H3K27M cells.
We hypothesized that the H3K27M mutation in uences the metabolic response to RT, thereby conferring RT resistance (33)(34)(35).Using the H3K27M-isogenic models, we performed steady-state metabolomics on untreated and RT-treated cells two hours after RT to capture early shifts in cellular metabolism that might mediate RT resistance (Fig. 1D-E).We identi ed numerous metabolites that changed following RT in either the H3K27M or H3K27M-KO cells (Fig. 1D-E and Supplemental To better understand which metabolic changes were H3K27M-speci c, we calculated the differences in fold-change (FC) following RT between H3K27M and H3K27M-KO cell lines for each metabolite (Fig. 1D-E and Supplemental Table 2).We noticed that numerous purine species including xanthine, hypoxanthine, guanine, AMP, and ADP/dGDP responded differently to RT in H3K27M mutant cells compared to KO controls (Fig. 1D-E, Supplemental Table 2).Metabolite set enrichment analysis of the top 25 metabolites highlighted purine metabolism in both H3K27M-isogenic cell line pairs (Fig. 1F-G, H).Other purine metabolism-related pathways found included the pentose phosphate pathway (DIPGXIII) (Fig. 1F) that creates the ribose-5-phosphate sugars needed for nucleotides, and glycine/serine/threonine metabolism (BT245) that generates metabolites needed for purine ring construction (Fig. 1G).We reanalyzed DIPGXIII-GFP/LUC tumor data to focus on purine species and found that tumor tissue has an inverse purine metabolic phenotype to that of the normal tissue (Fig. 1I).Interestingly, our group has shown that purine metabolism mediates treatment resistance in adult glioblastoma (GBM), another form of astrocytoma( 16).Together, these observations suggest that purine metabolism changes following RT in an H3K27M-speci c fashion.
Purine metabolic ux and enzyme expression are in uenced by the H3K27M mutation.
We and others have found that purine species, especially guanylates, contribute to treatment resistance in brain tumors(16, 36).We reasoned that targeting purine synthesis could increase RT e cacy in DMG-H3K27M.Purine nucleotides are produced through either the de novo or the salvage synthesis pathways (Fig. 2A).To determine which of these pathways DMG-H3K27M tumors may rely on, we performed stable isotope tracing to measure their baseline ux.This technique utilizes nutrients carrying one or more nonradioactive atoms that are heavier than the naturally occurring form, allowing for their detection by mass spectrometry.Heavy isotope labeling patterns in metabolites formed from the tracer molecule then allows us to track how cells change their metabolic uxes in response to RT.We tracked de novo synthesis (DNS) using 15 N-glutamine as it is critical for the formation of newly synthesized purine rings (Fig S4B).Purine salvage was measured using 2D-Hpx that is converted into the common purine precursor, IMP, which can then be converted into GMP or AMP (Fig S4A).Here, we observed that DIPGXIII H3K27M-expressing cells have a signi cantly higher ratio of 15 N-glutamine:2D-Hpx labeling of GMP than do H3K27M-KO cells, indicating that H3K27M cells prefer DNS to create GMP (Fig. 2B).This is further evidenced by higher 15  Given the H3K27M-speci c differences in GMP synthesis, we assessed the expression of the rate limiting enzymes in both de novo guanylate synthesis (IMPDH1 and IMPDH2) and guanylate salvage (HGPRT).
Using publicly available RNAseq data from patients bearing pediatric high-grade gliomas (pHGG)(2), we found that H3K27M pHGG tumors expressed less HPRT1, which encodes the rate-limiting guanylate salvage enzyme HGPRT (Fig. 1D).At the protein level, we observed increased total expression of IMPDH1 protein (Fig. 2E) and decreased total expression of HGPRT (Fig. 2F) in H3K27M cells compared to H3K27M-KO counterparts.We found no difference in the expression of IMPDH1 or IMPDH2 between H3WT or H3K27M pHGG tumors (Fig S5A -B).We also did not observe a difference in the expression of GMPS, the gene-encoding the enzyme downstream of IMPDH1/2 (Fig S5C).
Together, these results suggest that H3K27M cells inactivate guanylate salvage due to decreased expression of HGPRT and have increased reliance on de novo guanylate synthesis that is facilitated by increased IMPDH protein expression (Fig. 3A).Consistent with our tracing results, we did not observe similar patterns in the enzymes used to synthesize adenylates or in those used in the upstream steps of DNS (Fig S5D -F).
De novo guanylate synthesis inhibition increases RT e cacy in H3K27M models.
Increased reliance on de novo guanylate synthesis in H3K27M expressing cells suggests that inhibition of this pathway may have utility as monotherapy (37) or in combination with RT (Fig. 3A).Inhibition of de novo guanylate synthesis using the IMPDH inhibitor mycophenolic acid (MPA) radiosensitizes adult GBM brain tumors and is being evaluated in a Phase 0/1 clinical trial (NCT04477200) in human GBM patients(16, 38).To explore a similar combination strategy in H3K27M cells, we treated DIPGXIII and BT245 patient-derived H3K27M-expressing cells with 0-10uM MPA in combination with increasing doses of RT (0-6Gy) to determine its effect on radiosensitivity.By measuring neurosphere formation over time, we observed dose-dependent decreases in the surviving fraction of DIPGXIII and BT245 neurospheres following RT which was augmented upon addition of MPA at either concentration (1µM and 10µM), leading to increased RT enhancement ratios (DIPGXIII: 1.45 and 1.78, BT245: 1.33 and 1.33) (Fig. 3B-C).
This can be observed visually where we see that combined RT and MPA reduces neurosphere size and number in both of our H3K27M-expressing models, even at a low dosage of radiation (Fig. 3D, S6).Interestingly, we and others have observed single agent e cacy of MPA in DMG-H3K27M cells where we see reduced sphere size and number with MPA alone (Fig. 3D, S6).This suggests that DMG-H3K27M cells were more dependent on DNS at baseline, consistent with our isotope tracing studies and with previous reports (39).Taken together, these results show that IMPDH inhibition might help overcome RT resistance in H3K27M-expressing cells.
With these promising in vitro results, we next wanted to test the in vivo e cacy of RT in combination with IMPDH inhibition in H3K27M expressing tumors.Previous work from our group and others has shown that mycophenolate mofetil (MMF), the pro-drug of MPA, has e cacy in intracranial adult GBM models(16) and we sought to employ a similar strategy here.DIPGXIII-GFP/LUC cells were orthotopically implanted into the cortex and monitored by bioluminescent imaging (BLI).Once tumors were detectable, mice were randomized and treated with RT alone, MMF alone, combined RT and MMF, or vehicle control (Fig. 4A).Mouse weight was largely unaffected by the treatment course (Fig S7).Unlike our in vitro experiments, MMF alone had little effect on tumor size measured by BLI (Fig. 4B).Both RT and MMF + RT decreased tumor bioluminescence, but in both groups, tumors eventually regrew, and we did not observe marked difference between the two groups (Fig. 4B-C).MMF alone had no effect on median survival versus vehicle controls (27d vs 26d, respectively, p = 0.6).RT alone increased mouse survival over vehicle control (31.5d), but not in a statistically signi cant manner.Combination MMF + RT signi cantly extended survival over vehicle controls (38d vs 26d, p = 0.006), but did not cure tumors (Fig. 4D).These ndings suggest that while combination MMF + RT treatment extends survival, there may be resistance mechanisms employed by DMG-H3K27M tumors to evade MMF + RT treatment.
H3K27M cells upregulate purine salvage in response to RT.
We reasoned that DMG-H3K27M tumors might upregulate purine metabolism in response to RT and limit the e cacy of MMF in vivo.We employed stable isotope tracing to measure the activity of both de novo and salvage purine synthesis following RT (Fig. 5, S8).We showed previously that without RT, H3K27M cells converted more glutamine-derived 15 N into GMP but not AMP, than did H3K27M-KO cells (Fig. 2B,   S4E-F).RT increased 15 N incorporation into GMP in H3K27M-KO cells, but H3K27M cells showed no change (Fig. 5A).We also observe decreased 15 N-labeled AMP in both H3K27M and H3K27M-KO cells, though this effect is greater in H3K27M-KO cells (Fig S8A).These ndings suggest that H3K27M cells do not increase DNS following RT.
Previously we found that H3K27M cells converted less hypoxanthine into GMP than did H3K27M-KO cells at baseline, which suggested that the H3K27M mutation slows guanylate salvage (Figs.2B, 3A).Unexpectedly, both H3K27M cells and H3K27M-KO cells increased hypoxanthine salvage into GMP post-RT (Fig. 5B).This leads to a roughly sixty percent reduction in ratio of Gln:Hpx incorporation into GMP, highlighting increased reliance on salvage synthesis after RT in H3K27M cells (Fig. 5C).There was no change in hypoxanthine-derived AMP in either H3K27M or H3K27M-KO cells (Fig S8B).The ability for H3K27M cells to increase guanylate salvage following RT could account for their resistance to MMF + RT treatment.
To understand how H3K27M cells can upregulate purine salvage but not DNS following RT, we examined the expression of the rate limiting enzymes in these pathways.Consistent with an inability to upregulate DNS, H3K27M cells did not show increased IMPDH1 protein expression following RT (Fig. 5D).However, RT increased HGPRT expression in H3K27M-expressing DIPGXIII cells (Fig. 5E).Thus, while H3K27M mutant cells appear to rely on MPA-sensitive DNS in the unperturbed state, they upregulate MPA-resistant purine salvage synthesis following RT, which may mediate resistance to DNS inhibitors like MMF in vivo (Fig. 5F).
HPRT1 loss leads to extended survival in mice bearing DMG-H3K27M xenografts.
HGPRT can salvage both hypoxanthine and guanine (Fig. 3A).Salvaged hypoxanthine is converted to IMP whose conversion into GMP is blocked by MPA/MMF (Fig. 6A).Salvaged guanine, by contrast, directly forms GMP and bypasses IMPDH1 inhibition.Thus, upregulation of guanine salvage would cause MMF resistance while upregulation of hypoxanthine salvage would not, though both could be mediated by increased HGPRT expression.Our initial neurosphere assays (Fig. 3) were performed in TSM media where hypoxanthine is the dominant available purine base (40).However, the mouse tumor microenvironment is vastly different than the cell culture dish regarding metabolite availability.To determine which bases were available for guanylate salvage in DMG-H3K27M tumors, we analyzed both guanine and hypoxanthine levels in orthotopic DIPGXIII-GFP/LUC tumors and contralateral cortex.While DIPGXIII-GFP/LUC tumors and normal brain contained roughly the same abundance of hypoxanthine, tumor tissue possessed a 34-fold higher abundance of guanine (Fig. 6B).High intratumoral guanine coupled with increased salvage activity following RT (Figs. 5B) and RT-induced increases in HGPRT expression (Fig. 5D) suggest that HGPRT-mediated salvage of guanine can bypass IMPDH1 inhibition in H3K27M cells, leading to continued RT resistance (Fig. 6A) despite MMF treatment.
We then wanted to determine if HGPRT inhibition could increase RT e cacy in DMG-H3K27M xenograft tumors.No blood-brain barrier (BBB)-penetrant HGPRT inhibitors currently exist, so we utilized a pooled shRNA to knockdown HPRT1 expression in DIPGXIII-GFP/LUC cells (DIPGXIII-GFP/LUC-shHPRT1) (Fig. 6C).These cells were then implanted into the cortices of Rag1-KO mice.Tumor-bearing mice were RT-treated as previously described (Fig. 4A).Mouse weight was largely unaffected by the treatment regimen (Fig. S8).RT alone greatly reduced BLI signal in mice bearing DIPGXIII-GFP/LUC-shHPRT1 tumors (Fig. 6D).Median survival in DIPGXIII-GFP/LUC-shHPRT1 tumors was similar to vehicle control mice in our rst experiment (33d vs 26d), suggesting that purine salvage is dispensable for initial tumor growth (Fig. 4D and 6E).However, irradiation of HGPRT-de cient tumors signi cantly extended survival (> 75d) and led to multiple complete responses (Fig. 6E-F).Lastly, using publicly available patient tumor data, patients bearing H3K27M tumors had an inverse correlation between median survival and HPRT1 expression (Fig. 6G) (2).Together, these ndings indicate that purine salvage through HGPRT may mediate RT resistance in DMG-H3K27M.

Discussion
In this study, we have de ned new metabolic strategies to overcome treatment resistance in devastating DMG-H3K27M tumors.Using steady-state and stable isotope tracing metabolomics in patient-derived H3K27M-isogenic cell lines, we found that H3K27M cells preferentially rely on the DNS of guanylates.
This reliance is likely due to low HGPRT and high IMPDH1 expression in H3K27M cells.Inhibiting DNS of guanylates potentiated the effects of RT on H3K27M models in vitro and in vivo but did not cure tumors.Intriguingly, H3K27M cells upregulated guanylate salvage in response to RT, and DMG-H3K27M xenografts had abundant intratumoral guanine.This suggested that RT-induced guanylate salvage might be an adaptive mechanism of RT resistance.Consistent with this model, inhibition of guanylate salvage in DMG-H3K27M tumors overcame RT resistance and led to complete responses in several mice.Together, our ndings suggest that DMG-H3K27M tumors rely on both de novo and salvage guanylate synthesis, with salvage preferentially contributing to radiation resistance.
Our ndings add to the growing body of work showing that altered metabolism is a targetable vulnerability in DMG.These tumors cells preferentially rely on de novo pyrimidine synthesis and methionine consumption (25,26).Like others(39), we have found that DMG-H3K27M tumors also rely on de novo guanylate synthesis for survival.However, none of these seminal ndings have considered how H3K27M-mediated metabolic changes are related to RT resistance.Here, we unexpectedly discovered that activation of guanylate salvage may be an adaptive mechanism to resist RT.Thus, our work suggests that while strategies targeting methionine, pyrimidine, or de novo guanylate synthetic metabolism may be effective as monotherapies, targeting radiation-induced guanylate salvage in combination with RT may be an effective combination strategy to combat DMG-H3K27M tumors.
Our discoveries have important implications for the treatment of DMG-H3K27M patients.Nearly every patient with DMG-H3K27M receives RT, so a drug that selectively overcomes DMG-H3K27M RT resistance by inhibiting guanylate salvage could immediately be tested in combination with RT in patients.In this study, we utilized a genetic approach to inhibit guanylate salvage, which cannot readily be translated to patients.6-mercaptopurine (6-MP) is an HGPRT inhibitor that is effectively used in acute lymphoblastic leukemia (ALL) patients to maintain remission and has been shown to induce apoptosis in renal cancer cells (41,42).However, it was shown to have poor BBB penetrance in preclinical studies (41).Acyclic nucleoside phosphonates (ANPs) are being investigated as antimalarial compounds that target Plasmodium spp.HGPRT/HG(X)PRT (43).These ANPs were shown to complex with human HGPRT, however it is unknown if they can cross the blood-brain barrier.Preclinical studies to test the pharmacokinetics of molecules like ANPs and drug discovery directed at improving the BBB penetrance of existing inhibitors like 6-MP or development of new molecules could yield new strategies to target RTinduced guanylate salvage in DMG-H3K27M tumors.
These studies present lingering questions that still need to be answered.While we know that RT induces HGPRT expression, we do not know the molecular mechanism behind this upregulation.The rapid change in HGPRT expression (< 3hrs) suggests an epigenetic mechanism that could be related to the underlying H3K27M mutation.Additionally, we do not know if the ratio of DNS to salvage changes further as time progresses.We also do not know the source of the abundant guanine in our xenograft model, which could be due to high turn-over from malignant cells or secretion from non-cancerous cells such as atrocytes (39,44).Lastly, we do not know if the therapeutic window for inhibiting guanylate salvage will be large enough between normal brain and DMG-H3K27M tissue to improve treatment e cacy.The normal brain salvages purines and HGPRT loss underlies Lesch-Nyan disease, though it is possible that temporary inhibition of HGPRT in normal tissues may be tolerated (45).

Conclusions
Together, our studies suggest that multiple routes of purine synthesis are important for DMG growth.
Combining de novo purine synthesis inhibition with RT, an approach that we are utilizing clinically for adult patients with GBM(38), may therefore be less promising in DMG.Further preclinical study is needed to determine if combining small molecule inhibition of purine salvage with radiation can improve RT e cacy and be translated to clinical trials to improve DMG-H3K27M patient outcomes.
N-glutamine labeling (Fig S4C) and lower 2D-Hpx labeling (Fig S4E) and of GMP in H3K27M cells than in H3K27M-KO cells.There was no H3K27M-speci c preference for DNS-derived AMP (Fig. 2C, S4D and F).

Figure 2 Increased
Figure 2

Figure 3 De
Figure 3

Figure 4 Combination
Figure 4

Table 2 )
. Metabolites like glutamine in the DIPGXIII cells and aspartate in the BT245 model had similar responses to RT in both H3K27M and H3K27M-KO cells (Fig S3A-B).Other metabolites like xanthine in the DIPGXIII model and dGDP/ADP in the BT245 model had responses to RT that varied depending on the presence of the H3K27M mutation (Fig S3C-D, Supplemental Table2).
Supplementary FilesThis is a list of supplementary les associated with this preprint.Click to download.