Aberrant ATM signaling and homology-directed DNA repair as a vulnerability of p53-mutant GBM to AZD1390-mediated radiosensitization

Abstract

ATM is a key mediator of radiation response, and pharmacological inhibition of ATM is a rational strategy to radiosensitize tumors. AZD1390 is a brain-penetrant ATM inhibitor and a potent radiosensitizer. This study evaluated the spectrum of radiosensitizing effects and the impact of TP53 mutation status in a panel of IDH1 wild-type (WT) glioblastoma (GBM) patient-derived xenografts (PDXs). AZD1390 suppressed radiation-induced ATM signaling, abrogated G₀-G₁ arrest, and promoted a proapoptotic response specifically in p53-mutant GBM in vitro. In a preclinical trial using 10 orthotopic GBM models, AZD1390/RT afforded benefit in a cohort of TP53-mutant tumors but not in TP53-WT PDXs. In mechanistic studies, increased endogenous DNA damage and constitutive ATM signaling were observed in TP53-mutant, but not in TP53-WT, PDXs. In plasmid-based reporter assays, GBM43 (TP53-mutant) showed elevated DNA repair capacity compared with that in GBM14 (p53-WT), whereas treatment with AZD1390 specifically suppressed homologous recombination (HR) efficiency, in part, by stalling RAD51 unloading. Furthermore, overexpression of a dominant-negative TP53 (p53DD) construct resulted in enhanced basal ATM signaling, HR activity, and AZD1390-mediated radiosensitization in GBM14. Analyzing RNA-seq data from TCGA showed up-regulation of HR pathway genes in TP53-mutant human GBM. Together, our results imply that increased basal ATM signaling and enhanced dependence on HR represent a unique susceptibility of TP53-mutant cells to ATM inhibitor–mediated radiosensitization.

INTRODUCTION

Radiation therapy (RT) is the most effective nonsurgical treatment for glioblastoma (GBM). Despite aggressive treatment with RT and temozolomide, long-term survival beyond 5 years is uncommon (1). Thus, strategies that enhance the efficacy of existing therapies are urgently needed. Both alkylating chemotherapy and RT generate DNA damage that is repaired through DNA damage response (DDR) pathways (2). Dysregulation of these pathways during tumor-igenesis enables nascent cancer cells to accumulate genomic alterations. The resulting defects in DDR may also reduce the capacity to repair DNA damage, and the differential capacity of normal versus neoplastic cells to respond to genotoxic insults provides the biological basis for the therapeutic window. Moreover, inefficient DNA damage repair may render neoplastic cells susceptible to inhibitors of DDR pathways, and this possibility has spurred the development of small-molecule DDR inhibitors as sensitizing agents (3). Ataxia telangiectasia mutated (ATM), a serine/threonine protein kinase, is a central mediator of DDR and a key therapeutic target (2, 4-6). ATM is recruited to DNA damage sites and subsequently phosphorylates hundreds of targets to orchestrate various aspects of DDR, including cell cycle checkpoint activation, chromatin relaxation at sites of damage, DNA repair, and apoptosis (7, 8). Several inhibitors of ATM enhance efficacy of genotoxic therapies in preclinical cancer models (9-11), and there is great interest in clinical development of ATM inhibitors as sensitizing strategies in a variety of malignancies.

Radiation-induced DNA double-strand breaks (DSBs) are repaired by nonhomologous end-joining (NHEJ) and homologous recombination (HR), and failure of these pathways to repair DSBs leads to growth arrest and cell death. HR requires a sister chromatid as a template and is therefore restricted to S and G₂ phases of the cell cycle; on the contrary, NHEJ is active throughout the cell cycle (12). With an important role for p53 in enforcing a G₀-G₁ arrest after radiation exposure (13, 14), HR may play a less prominent role in repair of DSBs in p53 wild-type (p53-WT) tumors. Moreover, p53 and ATM have multiple intersections within DNA repair pathways that may contribute to less potent sensitizing effects of ATM inhibitors in p53-WT tumors (15, 16), but the mechanistic underpinning of this effect is unclear. In this study, we evaluated the efficacy of the brain-penetrant ATM inhibitor AZD1390 in a panel of isocitrate dehydrogenase-1 wild type (IDH1-WT) GBM patient-derived xenografts (PDXs) with varying TP53 status. Here, we demonstrate that p53 dysfunction results in aberrant ATM signaling and elevated HR that render cells vulnerable to AZD1390-mediated radiosensitization.

RESULTS

Effects of AZD1390 on RT-induced DDR

The time course and dose response for AZD1390 were assessed in the TP53-mutant U251, a human glioma cell line. Treatment with AZD1390 resulted in dose-dependent inhibition of RT-induced ATM autophosphorylation at serine-1981 (pATM-Ser¹⁹⁸¹) and ATM-dependent phosphorylation of checkpoint kinase 2 (Chk2) at threonine-68 (pChk2-Thr⁶⁸) (Fig. 1A) with a significant suppression (P ≤ 0.05) observed at 30 nM or higher concentrations (fig. S1A). Phosphorylation of another direct target, serine-824 of KRAB-associated protein 1, KAP1 (pKAP1-Ser⁸²⁴), was significantly suppressed even at 3 nM AZD1390 (P = 0.03; Fig. 1A and fig. S1A). In a time-course analysis, 30 nM AZD1390 suppressed RT-induced pATM-Ser¹⁹⁸¹ and downstream signaling (Fig. 1B and fig. S1B). Despite redundancies in upstream kinases involved, RT-induced phosphorylation of serine-139 on H2A.X variant histone, H2AX (pH2AX-Ser¹³⁹ also known as γH2AX) was significantly inhibited by AZD1390 at the 2-hour time point, when signal for pH2AX-Ser¹³⁹ was at the peak (P = 0.004; fig. S1B). AZD1390 was similarly effective at suppressing RT-induced ATM signaling in GBM12 PDX-derived cell cultures, exhibiting even greater potency (Fig. 1, C and D, and fig. S2, A and B).

Fig. 1. AZD1390 inhibits ATM-dependent DDR and radiosensitizes GBM cells.

(A) Immunoblots showing dose-dependent inhibition of DDR signaling in U251 cells preincubated with graded AZD1390 ± 5-Gy RT; lysed 6 hours later; and analyzed for phosphorylated and total ATM, KAP1, and Chk2, with vinculin used as loading control. (B) Immunoblots for ATM signaling in U251 cells treated with 0 or 30 nM AZD1390 ± 5-Gy RT and analyzed 2, 6, or 24 hours after RT. (C) Immunoblots showing AZD1390 dose response in GBM12 as described for (A). D) Immunoblots showing time course analysis of ATM signaling in GBM12 as described for (B). (E) Immunofluorescent images showing 53BP1 (red) or γH2AX (green) foci in the nuclei (blue) of U251 cells treated as indicated and processed 1 hour after irradiation; scale bars, 10 μm. Bar graphs (bottom) show mean positivity ± SEM, N = 3, and comparison by two-sample t test, n = 3. (F) Clonogenic survival for U251 cells treated with 0 or 30 nM AZD1390 ± increasing doses of RT, normalized survival (means ± SEM, N = 4) fitted in LQ model, two-sample t test, n = 4. (G) Drug-exposure time-dependent increase in survival (means ± SEM, N = 3) in U251 cells treated with (or without) 30 nM AZD1390 ± 2.5-Gy RT, replacing drug-free medium at indicated time after RT; the effect of prolonged drug exposure was compared with that of 4-hour exposure, two-sample t test, n = 3. (H) Three-dimensional plots showing mapped surface bliss using normalized NS counts (means ± SEM, N = 4) for GBM12 cells treated with graded AZD1390 ± RT. (I) Radiosensitizing effect of 0 versus 30 nM AZD1390 in GBM12; the relative mean NS count (means ± SEM, N = 4) was fitted in LQ model and analyzed by two-sample t test, n = 4.

RT-induced DNA breaks are recognized by signaling proteins, including γH2AX and p53 binding protein 1 (53BP1), which recruit repair proteins to damage sites. This damage-inducible protein localization can be visualized as foci (17). Consistent with the role of ATM as an upstream regulator of γH2AX (18), AZD1390 suppressed RT-induced γH2AX foci. However, 53BP1 is an ATM-independent sensor for DNA damage (19), and AZD1390 did not have a significant impact on RT-induced 53BP1 foci (Fig. 1E). As expected, AZD1390 did not significantly affect the extent of DSBs when measured at 2 hours after RT by a neutral comet assay (P = 0.63; fig. S3A). Furthermore, consistent with a role of ATM in cell cycle checkpoint control, cotreatment with 30 nM AZD1390 and 5-gray (Gy) RT significantly reduced G₀-G₁ phase in U251 cells compared with RT alone (8.5 ± 1.1 versus 24.6 ± 1.3%; P < 0.0001; fig. S3B). These data confirm inhibition of ATM signaling by AZD1390 in the low-nanomolar range.

Radiosensitization of GBM cells by AZD1390-mediated ATM inhibition

The impact of AZD1390 on radiation sensitivity was assessed in clonogenic and neurosphere (NS) survival assays using U251 and GBM12 cells, respectively. Consistent with inherent sensitivity to ionizing radiation in U251, 5-Gy RT alone decreased clonogenic survival compared with control [plating efficiency (PE): 0.007 ± 0.002 versus 0.27 ± 0.06, respectively; P = 0.002; fig. S3C]. Addition of AZD1390 enhanced RT-induced cytotoxicity with ~2-log difference in survival at concentrations of 10 nM or higher (P < 0.0001; fig. S3C). Across multiple radiation doses, cotreatment with 30 nM AZD1390 enhanced radiation sensitivity, resulting in a sensitizer enhancement ratio at 10% survival (SER₁₀) of 1.9 ± 0.1 as determined by a linear-quadratic (LQ) model (Fig. 1F). To identify optimal duration of ATM inhibition for maximum sensitization, U251 cells treated with 30 nM AZD1390 were irradiated, and then medium was replaced with drug-free medium at 4, 8, 12, 16, and 24 hours after irradiation. In comparison to 2.5-Gy RT alone, significant radiosensitization was observed with 4 hours of drug exposure (PE: 0.02 ± 0.001 versus 0.11 ± 0.004; P < 0.0001), whereas radiosensitization further increased with extended drug exposure of 12 hours or longer (Fig. 1G). In an NS assay, continuous exposure of GBM12 cells to 30 nM or higher AZD1390 alone decreased NS formation, RT had relatively more potent single-agent activity (NS count: 73.1 ± 0.3%, 34.8 ± 7.7%, and 28.0 ± 14.8% after 1-, 2.5-, and 5-Gy RT, respectively). Cotreatment with AZD1390 (10 nM or higher) and RT further reduced NS formation compared with either treatment alone (P < 0.05 for all RT doses tested; fig. S3D). Cotreatment also showed robust synergy when analyzed by Bliss independence model (Fig. 1H) and a potent radiosensitizing effect by LQ model (SER₁₀: 2.4 ± 0.3; Fig. 1I). These results indicate that AZD1390 can augment efficacy of RT in GBM.

The impact of ATM knockdown (ATMKD) in U251 cells was evaluated using CRISPR-Cas9 technology. ATMKD attenuated phosphorylation of KAP1-Ser⁸²⁴. However, Chk2-Thr⁶⁸ phosphorylation was not significantly affected by AZD1390 in ATMKD clones (Fig. 2A and fig. S4). As expected, AZD1390 had no single-agent activity in U251-ATMKD cells, and AZD1390-mediated radiosensitization was also significantly reduced (SER₁₀: 1.3 ± 0.3) in ATMKD clones compared with that in U251-EV cells (SER₁₀: 2.3 ± 0.4; P = 0.025; Fig. 2, B and C). With some residual ATM and Chk2-Thr⁶⁸ phosphorylation apparent by Western blotting, these results are consistent with radiosensitizing effects by ATM inhibition rather than off-target effects. In ATM-deficient cells, radiation response results in accumulation of unresolved RAD51 recombinase foci (20). Similarly, in ATMKD#22 cells, RT increased RAD51 foci (69.9 ± 3.6% and 62.8 ± 11.3% cells with ≥10 residual foci at 24 and 48 hours after RT, respectively), and cotreatment with AZD1390/RT was no different from RT alone (Fig. 2D). In contrast, RT did not significantly increase RAD51 foci in U251-EV cells, whereas AZD1390/RT cotreatment resulted in a two-to threefold increase in cells with residual RAD51 foci (Fig. 2D). Collectively, these data indicate that ATM deletion or inhibition renders GBM cells hypersensitive to radiation by preventing RAD51 nucleofilament resolution.

Fig. 2. Suppression of ATM potentiates efficacy of RT in mouse xenografts.

(A) Effect of AZD1390 on DDR signaling in U251 cells modified by CRISPR-Cas9 empty vectors (EVs) or ATM guide RNAs (clones #4 and #22) pretreated with 0 or 30 nM AZD1390 for 1 hour and subsequently with 0 or 5-Gy RT were lysed 2 hours later. (B) Clonogenic survival (means ± SEM, N = 3) for U251-EV and ATMKD#22 cells treated with 0 or 30 nM AZD1390 ± RT as indicated, and data fitted in LQ model, two sample t test, n = 3. (C) Floating bar plots (line at mean) for SER_10, two-sample t test, n = 3. (D) Immunofluorescent images for RAD 51 foci (red) and nuclei (blue) of U251-EV or ATMKD#22 cells treated as indicated and analyzed 24 or 48 hours later; floating bars (line at mean) for % positive cells (±SEM; N = 3) compared by two-sample t test, n = 3. Scale bars, 10 μm. DAPI, 4′,6-diamidino-2-phenylindole. (E) Animal setup for x-ray irradiation and dosimetry map (blue line) in sagittal (right) and coronal (bottom) planes for the prescribed dose of 2 Gy; isocenter shown by intersection of red and green lines. (F) Kaplan-Meier graphs for survival, for mice with orthotopic U251-EV xenografts, after treatment, starting on day 5 after inoculation, with placebo/sham, 2-Gy RT per day for 5 consecutive days, or AZD1390 (20 mg/kg) and RT, analyzed by log-rank test, n = 10 per treatment group; two animals randomized to AZD1390/RT arm suffered procedural complications were excluded. Red font in parentheses and arrows pointed at x axis indicate treatment days. Line graphs for body weight changes over time (bottom), two-sample t test. (G) Survival over time, for mice with orthotopic U251-ATMKD#22 xenografts, after treatment, starting on day 6 after inoculation, with sham RT or 2-Gy RT per day for 5 consecutive days, and analyzed by log-rank test, n = 10 per treatment group; two animals in control arm survived until the end, had no sign of tumor engraftment by postmortem H&E analysis, and were excluded from analysis. Red fonts in parentheses and arrows pointed at x axis indicate treatment days.

The radiosensitizing effects of genetic depletion or pharmacologic inhibition of ATM were further evaluated in orthotopic U251 tumors established in athymic nude mice. The efficacy of RT (2 Gy × 5 delivered through opposed lateral beams with a 15-mm collimator; Fig. 2E), alone or combined with AZD1390 (20 mg/kg before RT), 5 to 9 days after inoculation, was evaluated with time to reach a moribund state as primary end point. The combination (AZD1390/RT) resulted in significant survival prolongation (median survival of 80 days versus 58 days with RT alone; P = 0.02), and both regimens extended survival compared with placebo/sham (median survival of 41 days; P < 0.001; Fig. 2F). Significant weight loss was observed 1 week after AZD1390/RT therapy (−14.5 ± 3.5%; P < 0.0001) but not after RT alone (−1.0 ± 2.2%; P = 0.8; Fig. 2F). With the radiation field encompassing the roof of the oral cavity, the delayed weight loss observed could be explained by enhanced mucosal toxicity.

Radiation-induced oral mucositis is a common toxicity of therapeutic radiation (21). To assess whether cotreatment with AZD1390 sensitizes oral mucosa to low-dose (fractionated) RT, nontumor-bearing C57BL6 mice were treated with same RT (2 Gy × 5) alone or AZD1390/RT combination regimens used in the brain irradiation study, except that the x-ray beam was directed to the oral cavity (fig. S5A). Although irradiation of the entire oral cavity with 2 Gy × 5 was well tolerated, addition of AZD1390 resulted in histologic evidence of mucositis and significant weight loss observed 1 week after completion of AZD1390/RT therapy (−13.7 ± 1.9% versus +3.5 ± 1.1%, with the RT alone; P = 0.0002; fig. S5, B and C). On the basis of this observation, subsequent irradiation studies were performed with modified RT targeting.

The efficacy of RT was evaluated in orthotopic tumors established with U251-ATMKD#22. Although sham treatment of these tumors had a relatively short median survival (25 days), RT (2 Gy × 5) on days 6 to 10 after inoculation extended median survival to 149 days (P < 0.0001; Fig. 2G). This study confirms the potent radiosensitizing effects of ATM depletion, but the cause of greater radiosensitization by ATMKD versus AZD1390 cotreatment is unclear. ATM functions within a complex of proteins involved in DNA repair, and depletion of protein potentially could result in more profound effects than kinase inhibition. Alternatively, these results may reflect suboptimal targeting by drug treatment as compared with knockdown.

Tissue binding of AZD1390 favors free drug partitioning into GBM

The achieved unbound (free) drug concentration in the brain and brain tumor can influence the efficacy of glioma therapeutics. To enable an extrapolation of our in vitro radiosensitizing effects, the unbound fraction of AZD1390 was measured in culture medium, murine plasma, murine brain, heterotopic GBM12, and human GBM tissues using rapid equilibrium dialysis. Samples were spiked with 5 μM AZD1390, and, 4 hours later, the fraction of drug that diffused into the dialysate was measured. This demonstrated an unbound fraction of 0.50 ± 0.05 in medium, 0.08 ± 0.002 in murine brain, 0.34 ± 0.04 in heterotopic GBM12, and 0.39 ± 0.03 in human GBM tissue (Fig. 3A). These data were then applied to our previous pharmacokinetic analysis of total AZD1390 measured in the tumor core and rim of orthotopic GBM12 xenografts and adjacent normal brain at 4 and 12 hours after a single oral dose (20 mg/kg) (22). Multiplying these published data by the unbound fraction provided an estimation of the free drug concentration of 265 ± 115 and 25 ± 12 nM in the tumor core and 56 ± 5 and 6.9 ± 4.0 nM in the tumor rim at 4 and 12 hours, respectively (Fig. 3B). These concentrations were then used to calculate free drug partitioning between tissue and plasma (Kpuu). At both the time points, the average Kpuu in the tumor core (1.34 ± 0.14 and 0.79 ± 0.44) was significantly higher than in normal brain (P < 0.0001 and P = 0.003, respectively; Fig. 3C).

Fig. 3. Free drug partitioning and pharmacodynamic effects of AZD1390.

(A) Bar graphs showing unbound fraction of AZD1390 in indicated mouse tissue types, resected patient GBM (GBM pt.), or culture medium. (B) Floating bar plots (line at mean) for estimated unbound AZD1390 in normal brain, tumor core, or rim of orthotopic GBM12 xenografts analyzed 4 or 12 hours after a single dose of AZD1390 (20 mg/kg), two-sample t test, n = 4 or 5 per group. (C) Floating bar plots (line at mean) for unbound partition coefficient (Kpuu) calculated for each tissue type for individual mouse, log-transformed data analyzed by two-sample t test, n = 4 or 5 per group. (D) Immunofluorescent images showing γH2AX foci (green) and nuclei (blue) in orthotopic GBM12 xenografts excised from athymic nude mice treated (10 days after inoculation) with (or without) single dose of AZD1390 (20 mg/kg) and/or 3 or 11 hours later with 5-Gy RT (n = 5 per group) and euthanized 1 hour after RT. Scale bars, 50 μm. (E to G) Long-term effects of RT (6 Gy × 5 fractions) ± AZD1390 on brain morphology and molecular integrity; images from H&E-stained coronal section of the brains (E), representative spectroscopic images constructed from the integrated areas of the spectra between 1050 and 1100 cm⁻¹ for nucleic acids (F), and FTIR spectra (G) obtained from the irradiated region (outlined in red), for individual mouse in each group; each spectrum is an average of measurements from four or five animals per treatment group.

The pharmacodynamic effects in GBM12 orthotopic xenografts in athymic nude mice were analyzed 4 or 12 hours after dosing with AZD1390 (20 mg/kg) combined with (or without) 5-Gy RT. Single-fraction 5-Gy RT, on day 10 after inoculation, resulted in increased γH2AX immunofluorescence 1 hour after RT, and this was suppressed in animals pretreated with AZD1390 for 3 or 11 hours earlier (Fig. 3D). With an in vitro 12-hour exposure to 30 nM AZD1390 (15 nM free drug) providing robust radiosensitizing effect, these data suggest that cells within the tumor core and rim would have adequate drug exposure, but isolated, invading tumor cells in the adjacent normal brain could have subtherapeutic exposures, especially at later time points after irradiation.

Impact of AZD1390 on late radiation effects in normal brain

Radiation-induced brain injury can critically affect tolerability of a radiosensitizing agent. To assess this risk, nontumor-bearing C57BL6 mice were treated (on day 0) with (or without) AZD1390, and radiation (6 Gy × 5 fractions) was delivered using a 2.5-mm collimator and partial arcs directed to the pons (fig. S6A). This regimen approximates the potential for late radiation effects of a clinical 6-week radiation schedule used in patients. Mice were observed weekly for 15 months. During this period, there was no significant difference in body weight, survival, or gross neurological deficits, including signs of paresis assessed just before euthanasia (fig. S6, B and C). After euthanasia, brains were collected, and histology was evaluated by a neuropathologist blinded to the treatment received; there was no evidence for radiation necrosis or other histopathologic abnormalities in either treatment group (Fig. 3E).

Radiation exposure can induce changes in the macromolecules of biological tissues. Fourier transform infrared (FTIR) spectroscopic imaging, which enables label-free detection of biomolecules (23), was used to assess long-term effects of 30-Gy RT in the brain region irradiated 15 months prior. FTIR images based on the integrated area of the spectral regions representing phosphates of DNA, β sheet protein conformations, and lipids revealed no significant differences in their distribution pattern among treatment groups (Fig. 3F). Averaged spectra from the region of interest, representing the area targeted with radiation beam, on brain slices showed increases, which were not significant, in absorbance for asymmetric phosphodiester group representative of nucleic acid (~1235 cm⁻¹) and methylene group (−CH₂−) representative of lipid concentration (~2925 cm⁻¹) for animals treated either with RT or with AZD1390/RT (Fig. 3G) that could reflect elevated oxidative stress and increased products of lipid peroxidation (24, 25). However, barring small differences in the second derivatives (fig. S6D) for the nucleic acids and carbohydrate (~1050 to 1450 cm⁻¹), there was no major shift in peak positions, indicating preserved molecular integrity (26). Collectively, these results suggest that AZD1390 combined with RT does not pose an obvious risk of late radiation injury as compared to RT alone.

Varied radiosensitizing effects of AZD1390 in GBM PDX lines

The effects of AZD1390 on response to radiation were evaluated across a panel of GBM PDXs with distinct molecular characteristics (table S1). Consistent with ATM-dependent p53 activation (27), AZD1390 significantly suppressed RT-induced expression of p21, transcriptionally regulated by p53, in p53-WT PDXs (P = 0.02; Fig. 4, A and B). Furthermore, cotreatment with AZD1390 attenuated RT-induced pATM-Ser¹⁹⁸¹, pKAP1-Ser⁸²⁴, and pChk2-T⁶⁸ in vitro in all PDX lines except pChk2-T⁶⁸ in GBM108 (Fig. 4, A and C). In a subset of four PDXs evaluated in extreme limiting dilution clonogenic assays (ELDAs), AZD1390 decreased clonogenic survival in TP53-mutant GBM22 and GBM43 but had no effect in TP53-WT GBM14 and GBM39 (Fig. 4, D and E). RT had varied effects across four GBM models, whereas AZD1390 augmented RT-induced cytotoxicity in TP53-mutant GBM22 and GBM43 PDXs (survival: 0.4 and 0.5% versus 5.3 and 1.8% with RT alone, respectively; P < 0.0001; Fig. 4E). Likewise, among TP53-WT PDXs, AZD1390/RT decreased clonogenic potential in GBM39 but not in GBM14 (Fig. 4E). AZD1390/RT increased apoptosis in GBM43 (61.7 ± 2.4% versus 19.6 ± 0.6% with RT alone; P < 0.001) but not in GBM14 (fig. S7). Thus, despite robust ATM inhibition across all PDXs, only a subset of PDXs were sensitized to RT-induced cytotoxicity by AZD1390.

Fig. 4. Genotype-dependent differences in DNA damage signaling and radiosensitizing effects of AZD1390.

(A) Representative immunoblots showing DNA damage signaling in specified PDX lines in vitro. Primary cells for each GBM PDX line were treated with 0 or 30 nM AZD1390 ± 5-Gy RT, lysed 6 hours after RT, and analyzed for the indicated proteins; β-actin was used as a loading control, and p21 was used as an indicator of p53 transcriptional activity. (B and C) Floating bar plots (line at mean) showing signal intensity for p21 normalized to β-actin (B) and the signal intensities for signaling proteins normalized to appropriate loading controls (C); data points represent average signal intensities, for the protein analyzed, from four PDX models analyzed per group analyzed in two independent experiments (N = 2), and two-sample t test was applied to log-transformed data, n = 4. (D) Graphs showing limiting dilution clonogenic survival after treatment with (or without) 10 nM AZD1390 ± 2.5-Gy RT for indicated PDX lines; clonogenic growth scored 15 days after irradiation was plotted and analyzed using ELDA webtool: http://bioinf.wehi.edu.au/software/limdil. (E) Bar graphs showing frequency of clonogenic cells as estimated by ELDA algorithm; data represents means ± SEM from three independent assays for each PDX line, two-sample t test, n = 3.

ATM signaling can control the efficiency of HR, and persistent RAD51 foci in irradiated cells can reflect a failure to resolve HR intermediates. Therefore, the proportion of cells with residual RAD51 foci (≥10 foci per cell) was evaluated. In GBM14, the addition of AZD1390 had an increase in RAD51 foci only at the 48-hour time point (P = 0.05; fig. S8). In comparison, persistent RAD51 foci were significantly greater at both 24- and 48-hour time points with AZD1390/RT cotreatment in GBM43 (P = 0.02 and P = 0.0004, respectively; fig. S8). RAD51 foci were mostly localized to nuclei expressing geminin, which accumulates in S, G₂, and M phases of the cell cycle (fig. S8). Furthermore, AZD1390/RT cotreatment decreased G₀-G₁ with concomitant increase in cells with 4N or higher DNA content in both GBM14 and GBM43 PDX lines. The fraction of GBM14 cells in G₀-G₁ (46.2 ± 0.7%) at 48 hours after treatment was significantly higher than in similarly treated GBM43 cells (P < 0.0001; fig. S9). The more pronounced G₁ arrest in GBM14 cells may be an important factor limiting the overall importance of HR-mediated repair in this PDX.

Genotype-dependent differences in therapeutic response to AZD1390/RT

The in vivo radiosensitizing effects of AZD1390 were further studied in the same PDXs, as well as a ninth PDX (GBM26) that grows poorly in vitro, and in U251. For these studies, orthotopic tumors were treated (on days 5 to 9, GBM43 and U251; days 7 to 11, GBM12; days 8 to 12, GBM10; days 13 to 17, GBM26, GBM39, and GBM108; days 11 to 15, GBM06 and GBM22; or days 17 to 21, GBM14) after inoculation with AZD1390 and/or modified RT targeting (Fig. 5A). The smaller collimator size (10 mm) and superior shift of the inferior border improved the tolerability of the combination. Although AZD1390 monotherapy was mostly ineffective compared to placebo except in three PDXs: GBM10 (median survival: 34 days versus 28 days; P = 0.01), GBM26 (88 days versus 58 days; P = 0.006), and GBM22 (40 days versus 25 days; P = 0.002). RT afforded significant but variable survival benefit for most lines (P < 0.05 compared with control group; Fig. 5, B and C). Cotreatment with AZD1390/RT significantly extended survival benefit (P < 0.05 compared with RT alone) in five of nine PDX lines (including all four TP53-mutant PDXs and GBM39, which is TP53-WT, MDM4-amplified); however, no survival benefit was observed in GBM10, GBM14, GBM26, or GBM108 (all TP53-WT) models (Fig. 5, B and C). Comparing mean survival ratios (fold change in median survival with treatment versus control) across the entire panel showed no significant difference between AZD1390/RT versus RT (2.2 versus 1.7; P = 0.06). However, stratifying GBM models based on TP53 status demonstrated enhanced response to AZD1390/RT in TP53-mutant models (survival ratio: 2.2 versus 1.2 with RT alone; P < 0.01), but not in TP53-WT models (Fig. 5D). Collectively, AZD1390 is an effective radiosensitizer for a subset of GBM, especially those harboring TP53 mutations.

Fig. 5. Genotype-dependent differences in in vivo radiosensitizing effect of AZD1390.

(A) Animal setup for opposed lateral beam irradiation and dosimetry map for brain irradiation using athymic nude mice 7 days after inoculation of GBM12 PDX cells. Posterior-anterior radiograph showing isocenter targeting (left) and radiation dosimetry in coronal (middle) and sagittal (right) planes through head for prescribed dose of 2 Gy. (B) Kaplan-Meier plots showing survival over time for each cohort of mice with intracranial xenografts of TP53-WT (left column) or TP53-mutant (right column) PDXs (n = 5 per group, except for GBM10, where results from two independent studies were combined). Red fonts in parentheses and arrows pointed at x axis indicate treatment days. P, placebo; A, AZD1390 (20 mg/kg per day) alone; RT, 2 Gy per day for five consecutive days; A/RT, AZD1390 + RT. (C) Table summarizing same results with TP53 status, median survival time (days), and survival ratios for each PDX model. *P < 0.05, log-rank test for P versus A or RT versus A + RT; survival ratio represents fold change in median survival with treatment over placebo. (D) Box and whisker plots (with individual data points representing median survival ratios for each PDX line) show the comparison of efficacy between RT versus A + RT for all models (left) or select TP53-WT (middle) or TP53-mutant (right) models analyzed by Wilcoxon rank sum test.

Aberrant ATM signaling and elevated HR in TP53-mutant GBM

Consistent with a prominent role for ATM in HR, AZD1390 reduced HR efficiency (but not NHEJ) in a plasmid reconstitution assay in TP53-mutant GBM43 but had no effect on HR in TP53-WT GBM14 cells in vitro (Fig. 6A). Likewise, ATM inhibition or ATMKD decreased HR efficiency and promoted sensitivity to veliparib (fig. S10, A to D) in U251. In clonogenics, HR disruption by RAD51 knockdown rendered U251 cells hypersensitive to RT, with concomitant decrease in AZD1390-mediated radiosensitization (SER₁₀: 1.1 ± 0.1 versus 2.0 ± 0.2 for U251-siControl; P = 0.002; fig. S10, E and F). These results are consistent with AZD1390-mediated suppression of HR in GBM PDX cells.

Fig. 6. Aberrant ATM activation associated with elevated DNA damage and HR activity in TP53-mutant GBMs.

(A) Floating bar plots (line at the mean) show relative NHEJ (left) and HR (right) efficiency in GBM43 (TP53-mutant) and GBM14 (TP53-WT) cells treated as indicated, two-sample t test, n = 3 and 5 for NHEJ and HR, respectively. (B) Top-ranked pathways associated with differentially expressed genes between TP53-mutant versus TP53-WT in the TCGA Firehose Legacy GBM dataset. (C) Pie charts for NHEJ (left) or HR (right) genes; highlighted in pink are those significantly up-regulated in TP53-mutant GBM. (D) Kaplan-Meier plots showing progression-free survival among patients with TP53-mutant (n = 40) or TP53-WT (n = 56) GBM, who received RT and expressed low versus high BRCA1 mRNA compared by log-rank test. (E) Immunoblots (left) showing total and phosphorylated ATM, KAP1, and Chk2 proteins in untreated TP53-mutant and TP53-WT GBM lines; β-actin was loading control; floating bar plots (right), with line at mean, are for signal intensities normalized to β-actin; P values calculated by two-sample t test applied to log-transformed data, n = 5. (F) Representative images (scale bars, 10 μm) show basal γH2AX (red) and DAPI (blue) in specified PDX lines in vitro; graphs (right) showing mean (±SD) for γH2AX positivity (≥10 foci per nucleus) in PDX lines stratified into TP53-mutant and TP53-WT groups; P values by two-sample t test, n = 3.

The impact of p53 on DNA repair pathways was further explored using gene expression data from the GBM the Cancer Genome Atlas program (TCGA) Firehose Legacy dataset. When comparing TP53-mutant and -WT tumors, we identified differentially expressed genes (DEGs; 1900 overexpressed and 1016 underexpressed; false discovery rate < 0.05). Because transcriptomic changes influence multiple biologic processes, functional annotation clustering and pathway analysis for DEGs were performed using DAVID Bioinformatics Resources. These analyses revealed DNA replication, cell cycle regulation, Fanconi anemia, and HR among pathways enriched in TP53-mutant group (Fig. 6B and table S2). In a direct evaluation, nearly half of HR genes (29 of 63) were up-regulated in TP53-mutant GBM, whereas only 2 of 13 NHEJ genes were up-regulated (Fig. 6C). Furthermore, most HR genes up-regulated in the patient dataset also were up-regulated in the collection of TP53-mutant Mayo GBM PDXs (fig. S11A). BRCA1 mRNA expression showed an inverse correlation with progression-free survival (P = 0.002) and overall survival (P = 0.05) in TP53-mutant GBM (Fig. 6D and fig. S11B). These results suggest that elevated HR in TP53-mutant GBM is associated with transcriptional changes potentially induced by genomic stress. Consistent with this hypothesis, we observed significantly elevated endogenous pATM-Ser¹⁹⁸¹ (P = 0.006) and pChk2-The⁶⁸ (P = 0.02) and a similar trend for pKAP1-Ser⁸²⁴ (P = 0.06) in TP53-mutant lines compared with TP53-WT lines (Fig. 6E). Moreover, the relative fraction of nuclei with endogenous γH2AX (≥10 foci per nucleus) trended higher but was not statistically significant in TP53-mutant PDXs (GBM6, GBM22, and GBM43) compared to that in TP53-WT PDXs (GBM14, GBM39, and GBM108; P = 0.06; Fig. 6F). These results imply that heightened endogenous DNA damage and transcriptomic alterations are likely associated with elevated HR capacity in TP53-mutant GBM.

Effects of p53 transactivation on DDR and AZD1390-mediated radiosensitization

Mutant p53 may exert dominant-negative effects on transcriptional activity of coexpressed WT p53 (p53-WT) protein. To examine effects of p53 transactivation, we stably transduced GBM14 (TP53-WT) cells with lentiviral constructs encoding green fluorescent protein (GFP) or a p53 C-terminal dominant-negative fragment (amino acids 302 to 390; p53DD) (28). As expected, p53DD overexpression blocked p21 induction after radiation (Fig. 7A). A significant increase in basal pKAP1-Ser⁸²⁴ was observed in GBM14-p53DD compared with GBM14-GFP cells (P = 0.02), whereas RT-induced pKAP1-Ser⁸²⁴ was unaffected (Fig. 7A). Similarly, the nuclei with ≥10 γH2AX foci were ~3-fold higher in GBM14-p53DD than in GBM14-GFP cells (P < 0.0001; Fig. 7B). These results imply that blocking p53 transactivation activity promotes endogenous DNA damage and aberrant ATM signaling. In addition, pKAP1-Ser⁸²⁴, pChk2-Ther⁶⁸, and pChk1-Ser³⁴⁵ were also more pronounced at 48 hours after RT in GBM14-p53DD cells than in GBM14-GFP counterparts. At this time point, cotreatment of either subline with AZD1390/RT suppressed pATM-Ser¹⁹⁸¹, pKAP1-Ser⁸²⁴, and pChk2-Thr⁶⁸ compared with RT alone but induced pChk1-Ser³⁴⁵, with a significant increase seen in GBM14-GFP cells (P = 0.04; Fig. 7C). Consistent with a role of p53 transactivation in modulating DNA repair, GBM14-p53DD cells had nearly 2.5-fold higher HR efficiency than GBM14-GFP cells (P = 0.0004; Fig. 7D). These results indicate that loss of p53 transactivation enhances ATM signaling and HR activity in GBM cells.

Fig. 7. Loss of p53 transactivation function promotes endogenous ATM and HR activity.

(A) Immunoblots (left) showing effect of constitutively overexpressed dominant-negative p53 (p53DD) on endogenous and RT-induced KAP1 phosphorylation and p21 expression in GBM14 cells in vitro; KAP1 and β-actin served as loading controls; floating bar plots (right), with line at mean, are for signal intensities normalized to β-actin analyzed by two-sample t test applied to log-transformed data, n = 3. (B) Representative images of endogenous γH2AX (red) in the nuclei (blue) of GBM14-GFP and GBM14-p53DD cells in vitro (scale bars,10 μm); bar graphs (right) showing means ± SEM for γH2AX (>10 foci per nucleus) positivity from three independent experiments; P values by two-sample t test, n = 8 or 9 view fields. (C) Immunoblots (left) showing effect of AZD1390 on RT-induced DNA damage signaling in GBM14-GFP and GBM14-p53DD cells in vitro, phosphorylated and total ATM, KAP1, and Chk2 and Chk1 proteins 48 hours after specified treatments; β-actin was loading control; floating bar plots (right), with line at mean, are for signal intensities normalized to β-actin; two-sample t test applied to log-transformed data, n = 3. (D) Floating bar plots (with line at mean) for relative NHEJ and HR efficiency in GBM14-GFP and GBM14-p53DD cells; each data point represents average of three independent measurements, P values by two-sample t test, n = 3. (E) Line graphs of relative confluence and YoYo-3 positivity over time in GBM14-GFP and GBM14-p53DD cells treated with (or without) 10 nM AZD1390 ± 2.5-Gy RT in vitro; arrow indicates data points analyzed by two-sample t test, n = 3.

The impact of p53 transactivation on AZD1390-mediated radiosensitization was assessed using live-cell monitoring. For GBM14-GFP cells, the cell confluence progressed to nearly 100% during 120 hours of incubation. Although RT significantly blocked the progression to confluence (P = 0.0001), in comparison, AZD1390 had only a subtle (but significant, P = 0.03) effect on cell confluence (Fig. 7E). Consistent with our annexin V apoptosis data from GBM14, there was no significant increase in YoYo-3 uptake, a fluorescent marker of cell death (29), with RT (±AZD1390) in GBM14-GFP cells (Fig. 7E). In comparison, GBM14-p53DD cells grew faster, RT alone decreased cell confluence to some extent, and AZD1390 had no significant effect. In contrast, AZD1390/RT significantly reduced the progression to confluence (P = 0.0002 versus RT; Fig. 7E). Furthermore, unlike GBM14-GFP cells, treatment of GBM14-p53DD cells with AZD1390 or RT alone induced a modest but significant increase in YoYo-3 uptake [P < 0.0001; compared with dimethyl sulfoxide (DMSO) control] that further increased to ~40% with AZD1390/RT (P < 0.0001 as compared with RT alone; Fig. 7E). Together, these results imply that elevated ATM signaling and HR activity associated with p53 dysfunction represent a key vulnerability underlying AZD1390-mediated radiosensitization in TP53-mutant GBM.

DISCUSSION

ATM orchestrates the cellular response to DSBs, and multiple pharmacological inhibitors of ATM kinase have demonstrated radiosensitizing effects in various preclinical models, including GBM (9, 15, 30). Distribution of small-molecule inhibitors into the central nervous system (CNS) is a key consideration for GBM because all GBM have regions with a relatively intact blood-brain barrier (31). In this context, AstraZeneca optimized the ATM inhibitor AZD1390 for enhanced brain distribution and previously demonstrated robust radiosensitizing effects in orthotopic xenografts of a lung cancer brain metastasis and a syngeneic GBM cell line (30). In the current study, efficacy of AZD1390 combined with radiation was evaluated in orthotopic xenografts of U251 and nine GBM PDXs using clinically relevant radiation fractionation schedules. These experiments demonstrate radiosensitization in all TP53-mutant models and one of the five TP53-WT PDXs (GBM39, MDM4-amplified). Consistent with elevated genomic stress, TP53-mutant PDXs had perturbed DDR signaling. These data, along with evaluation of impact of AZD1390 on late radiation effects, provide insights to support development of brain-penetrant ATM inhibitors as radiosensitizers for GBM.

The repair of DSBs is primarily performed by NHEJ and HR. NHEJ is active throughout the cell cycle, but the HR is restricted to S-G₂ phases when a sister chromatid is available as a repair template (12). DNA-dependent protein kinase is a key driver of NHEJ, whereas ATM predominantly regulates HR. In response to DNA damage, both kinases and related ATM and Rad3-related (ATR) kinase signal through partially overlapping pathways to regulate cell cycle checkpoints, DNA repair, replication, and cell fate determination (32, 33). This radiation-induced DDR signaling leads to stabilization and activation of WT p53, which enforces G₁ checkpoint through transcriptional up-regulation of p21 (32). In cells with intact p53 function, G₁ arrest prevents cells from entering S phase and thereby precludes DNA repair by HR. Although ATM inhibition can disrupt radiation-inducible G₁ checkpoint, in tumors with ATM-independent checkpoint activity, G₁ arrest may preclude radiosensitizing effects of AZD1390-mediated suppression of HR. Retinoblastoma (RB) protein has pleiotropic roles in cell cycle control and DSB repair (34). Although ATM-dependent phosphorylation of serine-29 on E2F transcription factor 1 (pE2F1-Ser²⁹) can promote RB-mediated DSB end resection (35), RB is infrequently mutated in GBM, and the impact of RB pathway on AZD1390-mediated radiosensitization remains unclear.

p53 has a major impact on DNA repair pathway choice (36-38). Along with p53, 53BP1 accumulation prevents DSB end resection and thereby promotes repair by NHEJ (37, 39, 40). Although 53BP1 accumulates in DSB-induced foci through all cell cycle phases, 53BP1 is evicted from the core of S phase foci in a breast cancer type 1 susceptibility protein (BRCA1)–dependent manner (39). Whether p53 influences 53BP1 displacement from these foci remains unclear. However, given the role of p53 in facilitating NHEJ (37), p53 dysfunction likely shifts the balance of DNA repair toward HR. Given this elevated reliance on HR, AZD1390-mediated suppression of HR likely hampers DSB repair during S-G₂ phases and could account for enhanced radiosensitization in TP53-mutant tumors. Accordingly, HR disruption by RAD51 knockdown rendered U251 cells hypersensitive to RT, which is congruent with reported importance of HR for survival in irradiated p53-deficient mouse embryonic stem cells (41). These studies support the concept that p53 dysfunction promotes reliance on HR for DSB repair.

WT p53 also has pleiotropic transcriptional effects on DNA repair (36-38, 42). Several p53 target genes induced by radiation encode for proteins important for DNA repair (43-45), cell cycle arrest, and replication-associated DNA repair (46, 47). Pertinent to HR, RAD51 and RECQL4 are both direct p53 target genes (48, 49), whereas several other HR genes can be suppressed by p53-mediated activation of the dimerization partner, RB-like, E2F, and multivulval class B (DREAM) complex (50-52). In this context, multiple HR pathway genes were elevated in TP53-mutant GBM in human (TCGA) and xenograft (Mayo GBM National Resource) RNA sequencing datasets. Those genes include MRE11, RBBP8, and DNA2, involved in DSB end resection; PALB2 and BRCA1, involved in nucleofilament formation; and WRN helicase, essential for dissociation of holiday junctions that arise during HR (53). Although we did not analyze transcriptional regulation of HR genes, these data provide another possible mechanism for increased reliance on HR for DSB repair in TP53-mutant tumors.

ATM modulates multiple aspects of HR, including DSB end resection and resolution of complex HR intermediates (54, 55). Accordingly, in a plasmid-based reporter assay, ATM inhibition suppressed HR efficiency. Likewise, consistent with synthetic lethality between poly(adenosine 5′-diphosphate–ribose) polymerase and HR deficiency, pretreatment with AZD1390 rendered U251 cells sensitive to veliparib, and a similar but less marked effect was seen with ATMKD. Furthermore, ATM inhibition or ATMKD led to persistent RT-induced RAD51 foci, suggestive of stalled HR activity. This accords with prior experience with irradiated ataxia telangiectasia cells and potentially highlights compensatory roles of other related kinases (56, 57). Disrupted RAD51-nucleofilament resolution, regulated by ATM (50, 53), may contribute toward persistent RAD51 foci observed in cells deprived of ATM function. Whether this phenomenon associated with AZD1390 treatment reflects excessive single-stranded DNA exposed by unrestricted origin firing despite DNA damage or stalled repair remains to be evaluated. However, persistence of RAD51 foci implies that suppression of HR is a key mechanism underpinning radiosensitizing effects of ATM inhibition.

Minimizing the risk of elevated radiation–induced normal tissue injury is a key factor in the clinical development of radiosensitizers (58). Although ATM deficiency in humans and animals results in hypersensitivity to clinical radiotherapy, there are marked tissue–specific differences in radiation-enhancing effects (59). As demonstrated in our initial studies (Fig. 2E) and subsequent analyses of oral cavity irradiation with or without concurrent AZD1390, the oral mucosa is profoundly sensitized by ATM inhibition. These data are consistent with enhanced radiosensitivity of the gut and bone marrow in ATM-deficient mice (59, 60). Asymmetric division of stem cells after radiation is responsible for restoring tissue homeostasis, and the profound sensitization of oral mucosa by AZD1390 likely reflects sensitization of the stem cell compartment (61). These results contrast with protection against radiation-induced neuronal apoptosis in juvenile ATM-null mice (62). In addition, coadministration of AZD1390 with high-dose RT (6 Gy × 5 fractions, 30 Gy in total) delivered to the pons did not result in overt neurological defects in mice. Although AZD1390-mediated radiosensitization needs to be carefully evaluated across all CNS compartments, data generated so far suggest that the addition of ATM inhibitors to brain irradiation will likely provide a tolerable CNS toxicity profile.

Because HR repair occurs during S-G₂ phase of cell cycle, AZD1390-mediated radiosensitization may be muted in nondividing terminally differentiated cells in the brain. However, the radiosensitizing effects on actively dividing stem and progenitor cells in hippocampi warrant careful risk assessment. AZD1390 is a brain-penetrant ATM inhibitor, and concentrations in normal brain can reach the threshold required to sensitize actively dividing GBM cells (22). Especially with even higher concentrations of AZD1390 achievable in the human brain, evaluating the impact of AZD1390 combined with brain irradiation on cognitive ability will be relevant. Besides brain tissues, radiosensitizers can increase radiation toxicity in tissues adjacent to the brain, as illustrated by the weight loss and mucositis in our studies. These data highlight the importance of limiting the radiation doses to acutely responding tissues (oral and nasal mucosa, lacrimal glands, cornea, and inner and middle ear), when combined with ATM inhibitors.

AZD1390 is a promising radiosensitizer for the treatment of brain tumors, and the safety and tolerability of AZD1390 combined with brain irradiation are being tested in a phase 1 trial (NCT03423628). Assuming that this combination moves forward into phase 2 or 3 clinical trials, identification of efficacy biomarkers will be crucial. Across 10 GBM models, AZD1390/RT combination extended survival in four TP53-mutant PDX models. However, the magnitude of this benefit was relatively modest for GBM6 and GBM12. Conversely, although AZD1390 did not radiosensitize the majority of TP53-WT tumors tested, the survival extension for GBM39 was similar to the best sensitized TP53-mutant PDX. In the original report of AZD1390, differential radiosensitizing effects based on TP53 status were apparent in vitro, but in vivo evaluation in heterotopic GBM PDXs showed enhanced tumor regression in three of four TP53-WT models (30). Although not tested, IDH1-mutant tumors are commonly TP53 mutant (63), and understanding whether these tumors would benefit from ATM inhibitor combined with radiation will be important. Using a different ATM inhibitor, overexpression of a dominant-negative p53-281G in U87 was required for effective radiosensitization of orthotopic tumors by KU60019 (15). Acknowledging that p53 functions within multiple pathways, these data collectively suggest that mutant TP53 is an important but not exclusive biomarker for efficacious AZD1390-mediated radiosensitization.

There are several limitations to our study. The analysis of only 10 glioma models limits our ability to comprehensively evaluate how the spectrum of genetic alterations influence treatment outcome. For example, PTEN deletion is common in GBM and was reported to limit the radiosensitizing effects of ATM deletion in a model of brainstem glioma (64). Likewise, the interplay between inherent radiation sensitivity and radiosensitizing effects of ATM inhibition needs to be explored. In a second model of brainstem glioma, ATM deficiency rendered TP53- and CDKN2A-codeleted tumors sensitive to radiation (65). This is highly relevant for IDH1-mutant tumors with cooccurrence of TP53 mutation and CDKN2A/B homozygous deletion, an established biomarker of poor prognosis in otherwise low-grade gliomas (63, 66). Beyond affecting DDR, p53 and ATM are critical mediators of apoptosis and senescence in response to unrepaired DNA damage (67). Senescent tumor cells, within the appropriate microenvironmental milieu, can return to the cell cycle and regrow as a more aggressive tumor (68). Thus, understanding the influence of ATM inhibitors on cell fate after irradiation is another compelling area of study.

In summary, this study demonstrates radiosensitizing effects of AZD1390 in a subset of orthotopic GBM models. Loss of p53 transactivation results in elevated endogenous DNA damage and aberrant ATM signaling that renders GBM cells vulnerable to AZD1390-mediated radiosensitization. Together, this study provides a strong rationale for functional assessment of p53 and ATM as predictor of ATM inhibitor–induced radiosensitization.

MATERIALS AND METHODS

Study design

The objective of this study was to investigate radiosensitizing effects of AZD1390 in a panel of genetically diverse GBM PDXs and to delineate underlying mechanism of radiosensitization. Immunoblotting and immunofluorescence were used to validate the target inhibition and flow cytometry to assess impact on cell cycle progression, apoptosis, and DNA repair capacity. To ensure rigor and reproducibility, all in vitro experiments were repeated at least three times. The exact number of biological replicates and repeats are described in each figure legend. The sample size for in vivo studies was estimated on the basis of prior experience with the experimental models and reported treatment conditions. Power analysis was not performed. Animals were randomized into treatment groups before initiate treatment and were evaluated by staff blinded to treatments by flipping the cage cards to conceal the information. The time to death or euthanasia at moribund state as per guidelines of Institutional Animal Care and Use Committee (IACUC) was primary end point for efficacy studies, whereas animals were euthanized at prespecified time after dosing for pharmacodynamic or toxicity studies. Accidental deaths from procedural complications or failed tumor engraftment as determined by post hoc hematoxylin and eosin (H&E) analysis were excluded from analysis. Standard computational approaches were used for data analyses and synergy prediction.

Cell culture, transfection, lentiviral transduction, immunofluorescence, Western blotting, flow cytometry, and DNA repair assays

U251 and PDX cells (table S1) were cultured as previously reported (69, 70). Detailed methodology for cell culture, genetic modifications by CRISPR-Cas9 or small interfering RNA, and cell-based in vitro assays, including clonogenic growth, cell cycle, apoptosis, and plasmid reconstitution–based DNA repair assays, is available in Supplementary Materials and Methods.

Animal studies

Studies were approved by the IACUC (approval #A00006074). Orthotopic xenografts were established in the standard female athymic nude mice for National Cancer Institute (NCI) studies (NCr-Foxn1^nu/nu mice, strain code: 553), aged 6 to 7 weeks (Charles River Laboratories, Wilmington, MA). Intracranial xenografts were established as described previously (70). Mice with established xenografts were randomized (n = 5 to 10 per group) and treated (days 5 to 9, GBM43 and U251; days 7 to 11, GBM12; days 8 to 12, GBM10; days 13 to 17, GBM26, GBM39, and GBM108; days 11 to 15, GBM06 and GBM22; or days 17 to 21, GBM14) after inoculation, with AZD1390, RT, or concomitant AZD1390 and RT. AZD1390 (20 mg/kg once per treatment day) was administered 1 hour before daily fraction of RT, and RT (2 Gy × 5) was delivered through opposed lateral beams (225 kVp) using X-Rad SmART on small animal stereotactic irradiator (225 kVp/20 mA) using a 2-mm Al filter for imaging and 0.3-mm Cu filter for therapy. Treatment planning and dosimetry performed using SmART-ATP (Scientific Solutions B.V.) software package. After treatment, mice were monitored daily by staff blinded to treatment and euthanized when moribund as determined by weight loss exceeding 20%, inability to reach food/water, immobility, hunched posture, lethargy, seizures, circling, and/or paralysis as per IACUC guidelines. For pharmacodynamic assessments, mice with orthotopic GBM12 were randomized and treated, on day 10 after inoculation, with single dose of sham RT, vehicle + RT, or concomitant AZD1390 and RT. AZD1390 (20 mg/kg) was dosed 3 or 11 hours before 5-Gy RT. Mice were euthanized 1 hour after irradiation, and brains were frozen in Tissue-Tek optimal cutting temperature (OCT) compound (Sakura Finetek USA Inc., Torrance, CA, USA, catalog #4583), sectioned, and analyzed by immunofluorescence imaging for γH2AX.

To assess AZD1390-mediated radiosensitization of oral mucosa, nontumor-bearing female C57BL/6 mice were randomized and treated (days 0 to 4) with RT or AZD1390/RT, AZD1390 (20 mg/kg) was administered 1 hour before daily fraction of RT, and RT was delivered at 2 Gy × 5 fractions using opposed lateral x-ray beams (225 kVp) through a 10-mm collimator and directed to the oral cavity. The body weights of mice were recorded daily. In a separate study, the animals were treated as above and 5 days later euthanized by CO₂ inhalation, and surgically removed tongues were formalin-fixed and paraffin-embedded. Trans-verse sections (5 μm in thickness) from each tissue were mounted on glass slides and stained with H&E for histologic assessment.

To assess neurotoxicity, 6-week-old non–tumor-bearing NCI C57BL6/6NCr, strain code: 556 mice (Charles River Laboratories) were randomized (n = 5 per group), and starting the same day (day 0) mice were treated with vehicle, RT, or AZD1390/RT. RT (6 Gy × 5 over 2 weeks) used a 2.5-mm collimator and arc delivery of x-rays (225 kVp) to the midbrain. The body weights of mice and survival were recorded. At day 457, mice were euthanized, and brains were fixed in 4% paraformaldehyde and frozen in OCT. Coronal 5-μm sections through the irradiated region of each brain were mounted on low-emissivity glass slides and washed to remove the residual OCT. Dried samples were then either H&E-stained for histologic examination or subjected to FTIR using PerkinElmer Spotlight 400 (PerkinElmer) equipped with a highly sensitive linear mercury-cadmium-telluride (MCT) detector array. Images were acquired in reflection mode in the mid-infrared range of 700 to 4000 cm⁻¹ at a spectral resolution of 8 cm⁻¹ and with 120 coadditions for 25-μm pixel size at line scan of 2.2 cm/s. All the sections were imaged separately and processed, and an average spectrum of the radiated region on sections from each group was included in the final analysis. To obtain the biomolecular distribution across the section, infrared images were first minimum noise fraction (MNF)–transformed to remove as much noise as possible. Thresholding approach was then used to clear out background and unwanted debris. The resulting images were baseline-corrected and normalized with respect to amide I peak (~1652 cm⁻¹) to minimize the light scattering and thickness-induced artifacts (23). To effectively compare differences between groups, raw data were MNF-transformed, and their respective vector-normalized second derivative plots were considered to monitor peak shifts and hidden shoulders within the spectra over four spectral regions: nucleic acids and carbohydrates (1030 to 1480 cm⁻¹), proteins and ester carbonyl (1480 to 1800 cm⁻¹), amide I (1600 to 1700 cm⁻¹), and lipids (2800 to 3020 cm⁻¹). Each region was smoothened with 11-point Savitzky-Golay algorithm of order 3. Vector normalization was used to not require any further normalization with respect to a specific peak.

TCGA data analysis

TCGA datasets were downloaded in August 2022 from the Broad Firehose and TCGA, respectively. RNA sequencing data for Firehose Legacy patient samples and available RNA-Seq data for PDXs were subdivided into TP53-WT and TP53-mutant groups, and relative gene expression was analyzed using cBioPortal. Patients who received RT were divided by BRCA1 expression of 0.5 z score above (high expressor) or below (low expressor) the median for correlative analysis.

Statistical analyses

Statistical comparisons across groups performed using the bidirectional two-sample t test or Wilcoxon rank-sum test, with log transformations as appropriate. The survival fractions in clonogenic (or NS) assays represent normalized PE or % NS count relative to controls and were fitted to an LQ model. SER₁₀ calculated as fold change in radiation dose (D) at survival fraction 0.1 by addition of AZD1390, where D for each condition is extrapolated by solving the equation {0.1 = e^{−(αD + βD2)}}. Median survival of orthotopic mice models estimated by the Kaplan-Meier method were compared by log-rank test. P values less than 0.05 were considered significant.