Research ArticleCancer

ATRX loss promotes tumor growth and impairs nonhomologous end joining DNA repair in glioma

See allHide authors and affiliations

Science Translational Medicine  02 Mar 2016:
Vol. 8, Issue 328, pp. 328ra28
DOI: 10.1126/scitranslmed.aac8228

Aggressive gliomas’ Achilles’ heel

ATRX is a protein that is often mutated in glioma, a lethal and relatively common brain tumor. Koschmann et al. developed a mouse model of ATRX-deficient glioma and discovered that these tumors grow more aggressively than their counterparts with wild-type ATRX. The reason this happens is that the loss of ATRX impairs DNA repair, resulting in genetically unstable tumors that can accumulate oncogenic mutations more quickly. However, because of their DNA repair defect, these tumors also proved to be more sensitive to treatments that damage the DNA, such as radiation and some types of chemotherapy. Consistent with these findings, the presence of ATRX mutation correlated with better outcomes in patients, because these tumors were more susceptible to treatment.


Recent work in human glioblastoma (GBM) has documented recurrent mutations in the histone chaperone protein ATRX. We developed an animal model of ATRX-deficient GBM and showed that loss of ATRX reduces median survival and increases genetic instability. Further, analysis of genome-wide data for human gliomas showed that ATRX mutation is associated with increased mutation rate at the single-nucleotide variant (SNV) level. In mouse tumors, ATRX deficiency impairs nonhomologous end joining and increases sensitivity to DNA-damaging agents that induce double-stranded DNA breaks. We propose that ATRX loss results in a genetically unstable tumor, which is more aggressive when left untreated but is more responsive to double-stranded DNA-damaging agents, resulting in improved overall survival.


Glioblastoma (GBM) is a lethal primary brain tumor with a median survival of less than 2 years. Recent work in human gliomas has documented recurrent mutations in the histone chaperone protein ATRX. ATRX mutation in glioma is primarily seen in adolescents and young adults (ages 10 to 30 years) (1). In pediatric patients, ATRX was reported to be mutated in 31% of patients with primary GBM [World Health Organization (WHO) grade IV glioma], often with concurrent mutation of TP53 and point mutation of the gene encoding the histone H3.3 variant H3F3A (1, 2). In adults (age >30 years), ATRX is mutated less frequently in primary GBM but is frequently found in lower-grade (WHO grade II/III) and secondary GBM (24). Recent profiling of adult grade II and III gliomas revealed that a majority (~75%) of the subtype of low-grade gliomas that carry TP53 and IDH1 mutations also harbor ATRX mutations, thus underscoring their fundamental role in gliomagenesis (4).

ATRX mutation is seen in at least 15 types of human cancers, including neuroblastoma, osteosarcoma, and pancreatic neuroendocrine tumors (PanNETs) (5, 6). However, the role of ATRX in tumorigenesis remains largely unknown. The ATRX protein likely plays an important epigenetic role, depositing histones at heterochromatin and telomeric DNA (7, 8). Previous characterization of human GBM and PanNETs has shown an association between ATRX loss and the maintenance of telomere length by alternative lengthening of telomere (ALT), or nontelomerase-based, pathways (1, 8). Transgenic loss of ATRX in a mouse model is embryonic-lethal, and postnatal conditional loss of ATRX alone impairs cortex development without causing tumor formation (7). Mutations in ATRX result in loss of ATRX protein by immunostaining and are thought to mediate loss of function (1).

The development of genetically engineered mouse models allows for the systematic evaluation of the contribution of specific genetic lesions to glial tumor development. The Sleeping Beauty (SB) transposase system is particularly well-suited to providing a platform for the rapid validation of proposed driver mutations (9). The SB system uses a synthetic plasmid DNA encoding a transposase gene that inserts a desired transposon DNA element stably into genomic DNA. With various combinations of human oncogenes and inhibitors of tumor suppressor genes, tumors resembling human GBM can be reliably generated (10, 11).

We used the SB transposase system to create an animal model of ATRX-deficient GBM. Loss of ATRX accelerated tumor growth and reduced median survival, uncovering the impact of ATRX loss on glioma tumor proliferation. We show that ATRX loss causes genetic instability in mouse GBM, including both microsatellite instability (MSI) and impaired telomere maintenance. In accordance with this, analysis of publicly available human glioma genome-wide data integrated from multiple sequencing platforms showed that ATRX mutations are associated with increased mutation rate at the single-nucleotide variant (SNV) level, but not at the chromosomal/copy number level. We also show that loss of ATRX results in impairment of nonhomologous end joining (NHEJ) activity and is strongly correlated with loss of activated (phospho-) DNA-dependent protein kinase catalytic subunit (pDNA-PKcs) staining. By uncovering the connection of ATRX mutation and impaired NHEJ, we provide a mechanism for genetic instability and an actionable therapeutic target for ATRX-deficient GBM. Together, these results provide insights into the role of ATRX mutations in human glioma.


Generation of ATRX-deficient mouse GBM using SB model

To assess the impact of ATRX loss on GBM, we developed an endogenous mouse model using the SB transposase system (10). We cloned an ATRX knockdown sequence (shATRX)] into a plasmid with flanking sequences recognized by the SB transposase for insertion into the host genomic DNA (Fig. 1A and fig. S1). We induced GBM in mice by injecting plasmids encoding (i) SB transposase/firefly luciferase, (ii) shp53, and (iii) NRAS, with or without (iv) shATRX, into the lateral ventricle of neonatal mice. Transfection efficiency and tumor growth were monitored by in vivo imaging of luminescence (Fig. 1B). ATRX loss was tested in the context of overexpression of the oncogene NRAS and a short hairpin against p53, because ATRX mutation by itself is not associated with cancer development in humans (12), and glial tumors with ATRX mutations almost always include TP53 mutations (1). The receptor tyrosine kinase (RTK)–RAS–phosphatidylinositol 3-kinase (PI3K) pathway is mutated in a large percentage of adult and pediatric high-grade gliomas (1, 13, 14). Thus, many genetically engineered animal models of GBM have taken advantage of the glioma-promoting effects of NRAS, either by activating mutations or through loss of NF1 expression, which results in NRAS up-regulation (table S1).

Fig. 1. SB mouse model represents ATRX-deficient GBM.

(A) Constructs of SB plasmids, including the plasmid with short hairpin against ATRX. Purple brackets represent IR/DR (inverted repeat/direct repeat) sequences; the area between these IR/DR sequences is recognized by the SB transposase for insertion into the host genomic DNA. miR-30 sequences represent the 5′ and 3′ flanking sequences of the 300-nucleotide primary microRNA molecule. PGK, phosphoglycerate kinase; CMV, cytomegalovirus. (B) Plasmid insertion and tumor growth are monitored by in vivo luminescence. ROI, region of interest. (C and D) Hematoxylin and eosin staining of an 8-day-old mouse brain (7 days after injection) (C), including inset of subventricular zone lining the lateral ventricle (D). (E) Immunofluorescence of the region depicted in (D), with transfected cells (GFP-positive, to the right of the dotted line) showing ATRX reduction at 7 days after injection. Asterisks indicate a tumor cell that is positive for expression of GFP and negative for ATRX. DAPI, 4′,6-diamidino-2-phenylindole. (F) IHC staining of transfected cells/tumors for markers associated with neural stem cells, including GFAP, OLIG2, and Nestin in mice injected with shp53/NRAS/shATRX.

After injection of plasmids (shp53, NRAS, and shATRX), the mice were euthanized at early time points to characterize the model. At 7 days after injection, transfected cells [green fluorescent protein (GFP)–positive] were found within 50 to 100 μm of the lateral ventricles and showed early loss of ATRX expression by immunohistochemistry (IHC) (Fig. 1, C to E). To determine whether all three plasmids (shp53, NRAS, and shATRX) were being expressed in the same cells, we injected mice with plasmids expressing single fluorescent markers: (i) shp53-GFP (green), (ii) NRAS-Katushka (red), and (iii) shATRX-noGFP [detected by IHC with immunofluorescent blue secondary antibody (405 nm)]. At 15 days after injection, cells showed evidence of cotransfection with all plasmids (fig. S2).

Under both experimental conditions (shp53/NRAS with or without shATRX), GFP-positive transfected cells coexpressed glial fibrillary acidic protein (GFAP) and Nestin, markers of neural stem cells (Fig. 1F) (15), and not myosin VIIa, a marker of ependymal cells (fig. S3) (16). By day 15, we found multiple clusters of proliferative cells, which were GFAP-negative by this time point but remained OLIG2- and Nestin-positive. This expression pattern persisted through late-stage tumors from moribund animals (Fig. 1F and figs. S4 to S7). Tumors in moribund animals showed strong phospho–extracellular signal–regulated kinase (pERK) staining (fig. S6), which is downstream of RAS and expressed in human GBM (17).

In the process of optimizing and characterizing this model, we established a large cohort of mice injected with shp53/NRAS plasmids (n = 47). We then compared their survival with our experimental group of mice injected with shp53/NRAS/shATRX plasmids (n = 19). The median survival of mice injected with shp53/NRAS/shATRX was significantly decreased (69 days) compared to that of mice injected with shp53/NRAS (84 days, P = 0.0032) (Fig. 2A). All tumors (with or without shATRX) showed histological hallmarks of human GBM, including pseudopalisading necrosis (Fig. 2B, black arrows). Loss of ATRX was localized only within tumors generated by injection of the shATRX-expressing plasmid and not in the adjacent normal cortex or choroid plexus (Fig. 2C). ATRX loss was evident throughout the entire tumor, and mice injected with shp53/NRAS/shATRX had larger tumors at earlier time points (Fig. 2C and table S2). These data show that ATRX loss accelerates GBM tumor growth and reduces survival in mice bearing GBM.

Fig. 2. ATRX loss decreases median survival in mice bearing SB-generated GBMs.

(A) Kaplan-Meier survival curves of C57BL/6 mice bearing SB-induced tumors (P = 0.0032, Mantel log-rank test). (B) Representative brain tumors with histologic hallmarks of GBM: pseudopalisading necrosis and nuclear atypia (black arrows and blue arrow, respectively). (C) Tumors with addition of shATRX plasmid are larger at earlier time points than tumors with shp53/NRAS alone and show ATRX loss throughout the tumor. Asterisk in the left panel indicates a region in a p53/NRAS tumor with positive expression of ATRX. Asterisk in the right panel shows a region in a p53/NRAS/shATRX tumor which is negative for ATRX.

MSI in ATRX-deficient GBM

ATRX encodes a protein with a DNA binding finger and a SWI2/SNF2-like adenosine triphosphatase (ATPase) motif, making it a member of a family of adenosine triphosphate (ATP)–dependent chromatin-associated proteins (18). Other members of this class participate in DNA damage repair (DDR) pathways, in particular nucleotide excision repair and double-stranded break repair (19). Previous work has shown that ATRX is recruited to sites of DNA damage (20). Thus, we hypothesized that ATRX may play a role in maintaining genetic stability in GBM.

One type of genetic instability that is seen in human GBM is MSI. Microsatellites are repeated mononucleotide and dinucleotide sequences that are prone to error during DNA replication. Loss of DNA mismatch repair results in variation of the length of microsatellite sequences in some human tumors, including GBM (21, 22). Both ATRX mutation (1) and MSI (21) are seen more frequently in younger patients with GBM. We used established primer sets for mouse microsatellite sequences (23, 24) to assess our mouse GBMs and determine whether ATRX loss resulted in the presence of MSI.

With established MSI software parameters, tumors were scored for differences in predominant microsatellite [polymerase chain reaction (PCR) fragment] length between tumor and control tail DNA. Tumors with ATRX loss showed greater rates of MSI using four distinct microsatellite markers (Fig. 3, A and B, and table S3). Overall, ATRX loss increased the rate of instability fivefold (P = 0.014; n = 48 total tumor versus control DNA comparisons). The rate of MSI in ATRX-deficient mouse tumors (21%) was comparable to the rate of MSI in human pediatric GBM (19%) (21).

Fig. 3. ATRX loss increases MSI in mouse GBM and SNV frequency in human glioma.

(A) Representative microsatellite lengths from a tumor with MSI-high positivity (log2 ratio of >4 for predominant tumor sample allele versus control mouse tail DNA) using marker D1Mit62. Plot shows a shift in populations of allele lengths in p53/NRAS/shATRX tumor (orange) compared to control mouse tail DNA (gray). Predominant allele size in this example is two nucleotides longer (orange arrow) than control mouse tail DNA allele (gray arrow). (B) Comparison of the percentage of tumors with MSI positivity using four independent MSI primer sets (mBat26, D7mit91, D1mit62, and D6mit59) (n = 48 tumor versus control DNA comparisons). Data sre means ± SEM; *P < 0.05 using two-sided χ2 analysis. (C) Analysis of matched human tumor/germline integrated sequencing data sets showing SNV frequency in tumors by ATRX mutational status (GBM, WHO grade IV; pediatric GBM excludes diffuse intrinsic pontine glioma; HGG, high-grade glioma, WHO grades III and IV). **P < 0.005 using unpaired Mann-Whitney test. (D) Analysis of significance of contribution to SNV rate by ATRX and TP53 mutational status, using a two-way analysis of variance (ANOVA) model in adult GBM (n = 290) and pediatric GBM (n = 128). Each data point represents an individual human tumor; line represents mean ± SEM.

Increased single variant rate in human glial tumors with ATRX mutations

The finding of MSI in ATRX-deficient mouse GBM pointed to ATRX playing a role in maintaining genetic stability at the sequence level. To validate this finding in human glioma, we evaluated whether ATRX mutation was associated with increased SNV rate in genome-wide data from multiple data sets. We retrieved multiple publicly available genome-wide data sets available in the European Genome Archive (EGA). We then integrated these multiple sequencing platforms, along with additional pediatric high-grade glioma samples (deposited at EGA accession no. EGAS00001001436), to produce full somatic sequence and copy number information on 293 pediatric high-grade glioma samples (up to age 30 years), of which 38 of 293 (13%) samples contained ATRX mutations. We also retrieved and analyzed 290 GBM samples (age >30 years) available from The Cancer Genome Atlas (TCGA) (10, 25). After analyzing our integrated pediatric and adult data sets, we did not observe any clustering of mutations in the SNF2-DNA binding region of the gene (fig. S8), as has been described by others (1). A higher average SNV rate was seen in ATRX-mutated pediatric high-grade glioma in all anatomical locations (P = 0.0038) and particularly in pediatric GBM (P = 0.0005), but not in adult GBM (P = 0.71) (Fig. 3C).

ATRX mutation is frequently found with concurrent TP53 mutation in human glioma. Because loss of p53 function may be permissive of genetic instability in tumor cells (26), we evaluated whether our finding of increased somatic variant rate in ATRX-mutated tumors was confounded by TP53 mutation in human GBM. TP53 mutation was not a predictor of variant rate in either data set, whereas ATRX was a predictor in pediatric nonbrainstem GBM (P = 0.027 ANOVA), but not in adult GBM (Fig. 3D and table S4). IDH1 mutation was also not a predictor of variant rate in pediatric or adult data sets, with ANOVA P values of 0.14 (adult GBM) and 0.19 (pediatric GBM) (table S4).

Association between ATRX loss and copy number alterations

Previous analysis of pediatric GBM has shown an association between tumors with concurrent mutations in H3F3A, TP53, and ATRX and copy number alterations (1). ATRX mutations were not associated with total copy number alterations, nor gains or losses considered separately, in the pediatric data set, or with percentage genome fraction altered in the adult glioma data set (Fig. 4A).

Fig. 4. ATRX loss does not cause structural/chromosomal alterations or change chromosome count.

(A) Analysis of chromosomal alterations, including copy number alterations, in the integrated pediatric GBM data sets, and percentage of genomic alterations in the adult TCGA data set show no difference by ATRX mutational status. Line represents mean ± SEM; P ≥ 0.05 using unpaired Mann-Whitney test; NS, not significant. (B) Chromosome count by metaphase preparation of independent GBM neurosphere cultures. Line represents mean ± SEM; P ≥ 0.05 using unpaired Mann-Whitney test. Both groups had similar coefficients of variation [31.4% in p53/NRAS tumor cells (n = 15) and 30.7% in p53/NRAS/shATRX tumor cells (n = 20)].

To further examine the impact of ATRX on stability at the chromosomal level, we investigated whether ATRX loss was associated with karyotypic changes in our mouse GBMs. We generated primary cell cultures from SB-generated mouse GBMs and demonstrated that tumor neurosphere cultures generated with p53/NRAS/shATRX tumors showed stable reduction of ATRX expression compared to p53/NRAS cells (fig. S9). With a previously established method (27), karyotypes of multiple independent GBM cell cultures showed similar mean chromosomal counts and coefficients of variation between neurospheres with and without shATRX (Fig. 4B).

Assessment for ALT in mouse GBM

Previous characterization of human GBM has shown an association between mutations of ATRX and the maintenance of telomere length by ALT, or nontelomerase-based, pathways (1, 8). To establish a causal link, we assessed our mouse GBMs for evidence of impact on telomere lengthening. We isolated DNA from GFP-positive mouse GBM tumor tissue (with or without shATRX) and noted no difference in telomere lengths in tumors with ATRX loss by established telomere quantitative PCR (qPCR) primers (fig. S10 and table S5) (28). We then surveyed DNA extracted from tumors and neurospheres for c-circle amplification, which has been shown to be a specific assay for ALT (29). C-circles were found in a subset of tumor and neurosphere samples with ATRX loss, but not in DNA extracted from normal mouse brains (fig. S11).

Recently, ultrabright spots seen on human tumor tissue by telomeric fluorescence in situ hybridization (FISH) have been demonstrated as a sensitive indicator for the presence of ALT (8). Most human PanNETs have been found to harbor both ATRX mutations and ALT assessed by telomeric FISH (8). We therefore hybridized FISH probes against human PanNETs to be used as positive controls and noted distinct bright spots (Fig. 5A, white arrows) in a subset of cells in PanNETs, which are consistent with tumor ALT positivity.

Fig. 5. ATRX-deficient mouse GBMs display ALT.

(A) Detection of ALT using telomeric FISH assay showing characteristic ultrabright spots in human PanNETs (white arrows, positive control). A distinct population of cells with increased telomere signal is seen in ATRX-deficient tumors (white dotted circle). (B) CTCF in arbitrary units for telomeric FISH signal (data points represent individual cells from three tumors under each condition); black dotted circles denote cells qualitatively showing ultrabright spots, consistent with ALT. Line represents mean ± SD; *P < 0.05 using unpaired Mann-Whitney test.

In our mouse tumors, we noted a population of tumor cells with higher fluorescence signal in ATRX-deficient mouse GBM (Fig. 5A, white dotted circle) that was not seen in control tumors or striatal cells. To quantify the difference, we calculated corrected total cell fluorescence (CTCF) in tumor cells randomly chosen from multiple tumors under each experimental condition. The mean CTCF from p53/NRAS tumors was similar to that of mouse striatal cells but was increased in p53/NRAS/shATRX tumor cells (Fig. 5B). Again, we saw a distinct set of cells with higher CTCF in both p53/NRAS/shATRX tumors and human PanNETs (Fig. 5B, black dotted circles), consistent with ALT physiology.

ATRX loss and impaired NHEJ

To determine the mechanism by which ATRX loss affects genomic and telomeric stability, we assessed its impact on DDR pathways. We used reporter plasmids previously designed to quantify homologous recombination (HR) and NHEJ efficacy (30). Hepa 1-6 cells were transfected with linearized reporters harboring a DNA-damaged GFP that can be restored by NHEJ or HR, depending on the plasmid system used. Cotransfection with shATRX reduced NHEJ function by 50% (P = 0.0024), but HR activity remained unaffected (Fig. 6A) when quantified by flow cytometry (percentage of GFP-positive cells after transfection with shATRX or shSCRAMBLE normalized to mean control; Fig. 6B and fig. S12).

Fig. 6. ATRX loss reduces NHEJ repair.

(A) Reporter assay with GFP expression that is restored by NHEJ or HR in the appropriate plasmids. Addition of shATRX impairs NHEJ (assay performed in triplicate). (B) Flow cytometric quantification of HR and NHEJ activity as assessed by the percentage of GFP-positive cells normalized to control (differences in HR activity are nonsignificant). Line represents mean ± SD; **P < 0.005 using unpaired t test. (C) Loss of ATRX in mouse GBM decreases pDNA-PKcs immunostaining (showing representative results of three mouse tumors per condition). Dotted line represents distinction between tumor and nontumor brain.

To confirm that the NHEJ pathway was impaired in vivo in mouse GBM with ATRX loss, we investigated whether key NHEJ pathway proteins were affected. Previous studies have shown that the proteins, namely, Ku70 and Ku80, first bind to double-stranded DNA breaks and then recruit the DNA-PKcs. DNA-PKcs can then recruit and phosphorylate other NHEJ pathway proteins, as well as phosphorylate itself to promote its activity (pDNA-PKcs). These activated pathway proteins then bridge broken ends to facilitate rejoining (30).

We stained mouse GBMs (with and without shATRX) and found that ATRX loss was strongly correlated with loss of pDNA-PKcs by immunofluorescence staining (Fig. 6C and table S6). Tumors with p53/NRAS (n = 3) alone showed robust pDNA-PKcs staining throughout the tumor, but it was almost absent in all tumors with p53/NRAS/shATRX (n = 3). We observed staining for pDNA-PKcs in the nontumor striatum of all animals in both groups (Fig. 6C). IHC for the NHEJ DNA repair enzymes Ku70 and XRCC4 revealed no overt changes in expression (fig. S13). Similar results were obtained for HR repair pathway enzymes RAD51 and BRCA1 (fig. S14) and base excision repair protein PARP1 [poly(adenosine diphosphate–ribose) polymerase 1] (table S6). On the basis of our finding of increased MSI in ATRX-deficient mouse GBM, we also surveyed for mismatch repair proteins known to affect MSI (MLH1, MSH6, and PMS2) and found similar expression patterns in tumors with and without ATRX loss (fig. S15).

Sensitivity of ATRX-deficient GBM tumor cells to double-stranded DNA-damaging treatment

Inhibition of NHEJ in cancer cells induces an accumulation of double-stranded DNA breaks, which increases susceptibility to radiation (31). Because our data show that ATRX loss impairs NHEJ, we hypothesized that ATRX-deficient GBM cells would have increased sensitivity to DNA-damaging agents that primarily induce double-stranded breaks. Indeed, ATRX-deficient GBM cells were more sensitive in vitro to treatment with radiation and previously published doses of doxorubicin, irinotecan (SN-38), and topotecan (Fig. 7A) (3235). In contrast, treatment of GBM cells with agents that primarily induce single-stranded defects [CCNU (1-(2-chloroethyl)-3-cyclohexyl-1-nitrosourea) and temozolomide (32)] was not affected by ATRX status. To confirm our findings in vivo, we assessed bioluminescence in mice with GBMs treated with whole-brain radiation. ATRX-deficient mice showed reduced growth at days 4 and 10 after radiation compared to controls (mice with shp53/NRAS) [Fig. 7B (treated) and fig. S16 (untreated)].

Fig. 7. ATRX-deficient GBM cells are sensitive to double-stranded DNA-damaging treatments.

(A) In vitro data showing proliferation of mouse GBM cell cultures (with or without shATRX) after exposure to escalating doses of cytotoxic agents. ATRX-deficient tumor cells have reduced proliferation only after exposure to agents that induce double-stranded breaks. Line represents mean ± SEM; *P < 0.05 , **P < 0.01, ***P < 0.001 using F test of logIC50 (median inhibitory concentration). (B) Schematic of whole-brain radiation for mice with GBM (with or without shATRX). Tumor growth assessed by in vivo luminescence is reduced in ATRX-deficient tumors (representative mice and their luminescence values are shown). Plot on right shows average tumor luminescence for all mice at early time points after radiation (n = 6 mice in each group). Line represents mean ± SEM; *P < 0.05 using unpaired t test.

We found that ATRX-deficient tumors showed reduction in pDNA-PKcs at multiple time points after radiation (fig. S17), whereas control tumors and nontumor brain showed both pDNA-PKcs and ATRX expression. In vitro, we found that ATRX-deficient tumor cells showed an increase in γH2A.X expression, a sensitive marker for double-stranded DNA breaks, 24 hours after treatment with doxorubicin (35), compared to control tumor cells (fig. S18A). In vivo, we saw increased γH2A.X expression in ATRX-deficient tumors 24 hours after radiation (single dose of 6 Gy) (fig. S18B).

ATRX mutation and improved survival in treated human GBM

Recent data have shown that adults with treated ATRX-mutated GBM have a survival advantage (3, 4). We used our integrated human glioma genome-wide data set to confirm that ATRX mutation provides a survival advantage in pediatric high-grade glioma patients as well (P = 0.0035, Fig. 8A).

Fig. 8. Pediatric patients with high-grade glioma and ATRX mutation survive longer.

(A) Kaplan-Meier curve based on genome-wide data from multiple pediatric data sets (n = 293) showing the survival benefit of ATRX mutation in treated patients. (B) Schematic of impact of ATRX loss on GBM.

In contrast, our data indicate that ATRX loss in untreated mice confers a survival disadvantage (Fig. 2). Increased sensitivity of ATRX-deficient GBM to radiation and/or chemotherapy that induces double-stranded breaks could explain this apparent discrepancy. Thus, we propose that ATRX loss causes impaired NHEJ and genetic instability in glioma, which is more aggressive when untreated but is more responsive to double-stranded DNA-damaging therapy, ultimately resulting in improved overall survival (Fig. 8B).


Recurrent mutations in ATRX point to its critical role in tumor progression in multiple human cancers (5, 6). Despite this, little is known about the molecular mechanism by which it affects tumor development and growth. We used an animal model of ATRX-deficient GBM to uncover the impact of ATRX loss on glioma tumor proliferation and loss of genetic stability. We show that ATRX reduction impairs NHEJ and pDNA-PKcs recruitment, thus providing a mechanism for genetic instability and a molecular target. These results provide insights into the impact of ATRX mutations in human glioma, and possibly other human tumors.

Our data link ATRX to regulation of NHEJ. Other chromatin remodelers are known to play important roles in providing proteins access to damaged DNA and telomeres (36). We hypothesize that reduced ATRX causes conformational changes in heterochromatin, restricting the access of NHEJ proteins such as DNA-PKcs to damaged DNA. Our finding of impaired NHEJ is consistent with previous research, which has demonstrated that ALT is dependent on HR of sister telomere chromatids (37). Our results suggest that a relative increase in HR in relation to NHEJ in ATRX-deficient glioma may be supportive of ALT (proposed schematic, fig. S19).

Our experimental data reproduce ATRX loss resulting in ALT in an animal model. Human (non-GBM) immortalized cell lines with ALT have been associated with genetic instability and altered DNA damage response (38). Our data substantiate the connection between ATRX, ALT, and the DNA damage response. Our animal model showed ALT in ATRX-deficient tumors by telomere FISH and in a subset of our ATRX-deficient tumors and tumor neurospheres by c-circle assay but did not display elongated telomeres by telomere qPCR. Overall, our results suggest that ATRX loss in GBM is capable of producing ALT but that additional factors and/or species differences may also play a role. Mouse telomeres are much longer than human telomeres (28); thus, we hypothesize that this species difference could influence ALT assays and ALT physiology.

According to the mutator hypothesis of oncogenesis, early mutations in “caretaker genes” can drive further tumor development (39). Our data show that ATRX plays this role in glioma because it is required for NHEJ DNA repair. It is possible that the genetic instability in ATRX-deficient GBM drives proliferation by affecting cell cycle control or differentiation, as has been shown in other genetically unstable tumor models (40, 41). Additionally, impaired apoptotic signaling through defective DNA-PKcs phosphorylation (42) and/or concurrent TP53 mutations could provide an additional proliferative advantage to ATRX-mutated tumors.

We show that ATRX is implicated in maintaining stability at the sequence level in our animal model and in our pediatric human data sets. Notably, in our adult human data sets, we did not see a difference in SNV rate between ATRX-mutated and nonmutated tumors, possibly making this finding most applicable to younger patients. The choice of DDR pathway influences the quality of the repair and the introduction of new somatic mutations and chromosomal rearrangements (43). Recent work has highlighted that NHEJ is actually a combination of two related but distinct processes: (i) canonical NHEJ (C-NHEJ), which is the traditional pathway involving repair of double-stranded breaks using DNA-PKcs, and (ii) alternative NHEJ (A-NHEJ), which is driven by PARP1, uses microhomology-mediated end joining, and is associated with deletions at repair junctions (44). Because our tumors (with or without ATRX loss) showed robust PARP1 staining (table S6), it is possible that the reduction in C-NHEJ in ATRX-deficient glioma results in mutagenesis from use of the more error-prone A-NHEJ pathway. Increase in A-NHEJ in ATRX-deficient glioma could potentially explain the increase in point mutations that result in both increased somatic mutation rate and MSI (which is also the accumulation of single-stranded mutations), thus explaining the phenotypic features seen in our animal and human data.

An important feature of endogenous animal models of glioma is the choice of genetic drivers. One possible limitation of our model is the use of NRAS mutation, which is only rarely mutated in human GBM. Nevertheless, RAS provides a useful driver of GBM formation, which retains relevant histologic features of the human disease while activating the RTK-RAS-PI3K pathway, for which most GBMs contain signal alterations (13, 14). Although ATRX mutation co-occurs with IDH-R132H mutation in adult glioma, it almost never does in pediatric GBM and instead frequently co-occurs with histone H3.3 mutations. In both cases, it almost always co-occurs with TP53 mutation (14). This overlap highlights the fact that ATRX mutation can affect tumor development with multiple other glioma drivers, as long as TP53 signaling is impaired, a feature our model also encapsulates.

Our results have translational relevance in that we modeled ATRX loss in mice to identify its role in furthering GBM progression and therapy response. We believe that this seemingly paradoxical role will assist in the design of future therapy for ATRX-mutated glioma. For example, detection of ATRX mutation and/or loss of ATRX by immunostaining may be evidence of a treatment-responsive subtype of glioma that would encourage the use of radiation and/or chemotherapy agents that induce double-stranded breaks. Our data showing both a reduction in NHEJ activity and an increase in response to double-stranded DNA-damaging agents in tumor cells with ATRX loss are consistent with previous research showing that NHEJ inhibition reduces double-stranded break repair and increases sensitivity to radiation (31). Further preclinical studies could highlight regimens that more selectively target the defects in NHEJ and double-stranded break repair in ATRX-deficient glioma. Our data raise the possibility that topoisomerase inhibitors (topotecan or irinotecan) might be clinically useful to exploit this defect.

These results lay a foundation for uncovering the molecular impact of ATRX mutations in the pathogenesis and response to therapeutics in human GBM. Additionally, this mouse model provides a platform for the development of targeted therapy for GBM patients harboring ATRX mutations.


Materials and Methods

Fig. S1. SB-responsive shATRX plasmids are cloned to explore the impact of ATRX on GBM development.

Fig. S2. Cells cotransfected with shp53, shATRX, and NRAS plasmids are characterized at 15 days after injection.

Fig. S3. SB-mediated transfected cells are distinct from ependymal cells in mice.

Fig. S4. Characterization of p53/NRAS/shATRX mice 7 days after injection shows tumor cells expressing GFAP and Nestin.

Fig. S5. p53/NRAS/shATRX mice at 21 days after injection show tumor cells expressing OLIG2 and Nestin.

Fig. S6. Moribund p53/NRAS/shATRX mice show tumor cells expressing OLIG2, Nestin, and pERK.

Fig. S7. p53/NRAS/shATRX mice show loss of GFAP expression by 15 days after injection.

Fig. S8. ATRX mutations do not cluster in SNF2/helicase domain in human glioma.

Fig. S9. Primary GBM cell cultures are generated from SB-induced mouse tumors.

Fig. S10. Telomere length measured by qPCR is not different in tumors with or without ATRX.

Fig. S11. ALT was assessed by c-circle assay with DNA from mouse tumors and neurospheres.

Fig. S12. Cells transfected with shATRX plasmid show reduction in NHEJ by flow cytometry.

Fig. S13. NHEJ pathway is characterized by IHC.

Fig. S14. HR pathway is characterized by IHC.

Fig. S15. Mismatch repair pathway is characterized by IHC.

Fig. S16. SB mouse tumor growth (without radiation treatment) is characterized by luminescence.

Fig. S17. Tumors with ATRX loss show reduction in pDNA-PKcs before and after radiation treatment.

Fig. S18. Mouse tumors with ATRX loss show increased expression of γH2A.X in vitro and in vivo.

Fig. S19. Detailed schematic shows proposed impact of ATRX loss on GBM progression.

Table S1. Animal models of glioma all have RTK-RAS-PI3K alterations.

Table S2. Mice injected with p53/NRAS/shATRX have faster-growing tumors.

Table S3. MSI rate is increased in p53/NRAS/shATRX tumors.

Table S4. IDH1 and TP53 mutations do not alter SNV rate calculated by two-way ANOVA model.

Table S5. Telomere qPCR average telomere length ratio worksheet and standard curve show similar results in all experimental groups.

Table S6. Immunofluorescence analysis shows reduced pDNA-PKcs expression in ATRX-deficient mouse GBM.

Table S7. Antibodies used for tissue analysis are summarized in table form.


  1. Acknowledgments: We thank J. Ohlfest (University of Minnesota, deceased) for his support of our implementation of the SB model. C.K. wishes to thank P. Robertson and H. Garton for their academic support. We gratefully acknowledge P. Jenkins and the Department of Neurosurgery at the University of Michigan Medical School for their support of our work. We are also grateful to K. Muraszko for her academic leadership and D. Tomford, S. Napolitan, M. Dahlgren, and C. Shaw for superb administrative support. This study makes use of data generated by the St. Jude Children’s Research Hospital–Washington University Pediatric Cancer Genome Project, the principal investigator C. Hawkins and the Hospital for Sick Children, the McGill University Health Centre and the DKFZ (Deutsches Krebsforschungszentrum)–University of Heidelberg Pediatric Brain Tumour Consortium, and the Institute of Cancer Research (ICR)–Institut Gustav Roussy–Hospital Sant Joan de Déu collaborative group. Funding: This work was supported by NIH/National Institute of Neurological Disorders and Stroke (NINDS) grants 1RO1-NS 054193, 1RO1-NS 061107, and 1RO1-NS082311 to P.R.L.; grants 1UO1-NS052465, 1RO1-NS 057711, and 1RO1-NS074387 to M.G.C., and grant NIH/National Cancer Institute R01CA172380 to A. Meeker. C.K. was supported by the St. Baldrick’s Foundation Fellowship and the Alex’s Lemonade Stand Foundation/Northwestern Mutual Young Investigator Grant. C.J., A. Mackay, and J.F.-S. acknowledge National Health Service funding to the National Institute for Health Research Biomedical Research Centre at The Royal Marsden and the ICR, and the INSTINCT network funded by The Brain Tumour Charity, Great Ormond Street Hospital Children’s Charity, and Children with Cancer UK. Author contributions: C.K. carried out the animal and in vitro studies and drafted the manuscript. A.-A.C., F.J.N., D.T., F.M., N.K., M.D., L.M., J.K., R.L., Y.L., and S.R. participated in the animal and in vitro studies. L.Z. performed statistical analysis. A. Meeker and J.A.B.-C. assisted with the design and interpretation of ALT studies (FISH and c-circle). H.A. provided human PanNET samples. D.F. assisted with the design and interpretation of metaphase preparation/chromosome counting. A. Mackay, J.F.-S., and C.J. performed analysis of human pediatric glioma data sets. V.G. assisted with the design and interpretation of DDR plasmid assays. M.G.C. and P.R.L. conceived and supervised the study, participated in its design and coordination, and helped to draft and edit the manuscript. All authors read and approved the final manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: We have deposited previously unpublished sequence data for additional pediatric high-grade glioma samples (C.J., EGAS00001001436).
View Abstract

Navigate This Article