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 quantified how metabolite levels differed between normal brain and DMG-H3K27M xenograft tumor tissue. Using fluorescence-guided microdissection, we separated GFP and luciferase-expressing DIPGXIII (DIPGXIII-GFP/LUC) orthotopic tumors from surrounding normal brain and quantified their metabolites. The brain-specific metabolite N-acetylaspartate was higher in cortex compared to DMG tumors, indicating that our fluorescent-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 specifically by the H3K27M mutation(24) while others might be related to alternative mutations or oncogenic transformation in general. To understand how the H3K27M mutation specifically 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 confirmed the presence of the mutation in parental cells and the corresponding lack of H3K27me3 (Fig S1B). We quantified 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 findings are consistent with prior studies and confirm 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 influences the metabolic response to RT, thereby conferring RT resistance(33–35). 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 identified numerous metabolites that changed following RT in either the H3K27M or H3K27M-KO cells (Fig. 1D-E and Supplemental 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 Table 2).
To better understand which metabolic changes were H3K27M-specific, 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-specific fashion.
Purine metabolic flux and enzyme expression are influenced 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 efficacy 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 flux. 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 fluxes in response to RT. We tracked de novo synthesis (DNS) using 15N-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 significantly higher ratio of 15N-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 15N-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-specific preference for DNS-derived AMP (Fig. 2C, S4D and F).
Given the H3K27M-specific 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 efficacy 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 efficacy 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 efficacy 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 efficacy 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 significant manner. Combination MMF + RT significantly extended survival over vehicle controls (38d vs 26d, p = 0.006), but did not cure tumors (Fig. 4D). These findings 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 efficacy 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 15N into GMP but not AMP, than did H3K27M-KO cells (Fig. 2B, S4E-F). RT increased 15N incorporation into GMP in H3K27M-KO cells, but H3K27M cells showed no change (Fig. 5A). We also observe decreased 15N-labeled AMP in both H3K27M and H3K27M-KO cells, though this effect is greater in H3K27M-KO cells (Fig S8A). These findings 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 efficacy 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 first experiment (33d vs 26d), suggesting that purine salvage is dispensable for initial tumor growth (Fig. 4D and 6E). However, irradiation of HGPRT-deficient tumors significantly 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 findings indicate that purine salvage through HGPRT may mediate RT resistance in DMG-H3K27M.