Microcephaly and medulloblastoma may both result from mutations that compromise genomic stability. We report that ATR, which is mutated in the microcephalic disorder Seckel syndrome, sustains cerebellar growth by maintaining chromosomal integrity during postnatal neurogenesis. Atr deletion in cerebellar granule neuron progenitors (CGNPs) induced proliferation-associated DNA damage, p53 activation, apoptosis and cerebellar hypoplasia in mice. Co-deletions of either p53 or Bax and Bak prevented apoptosis in Atr-deleted CGNPs, but failed to fully rescue cerebellar growth. ATR-deficient CGNPs had impaired cell cycle checkpoint function and continued to proliferate, accumulating chromosomal abnormalities. RNA-Seq demonstrated that the transcriptional response to ATR-deficient proliferation was highly p53 dependent and markedly attenuated by p53 co-deletion. Acute ATR inhibition in vivo by nanoparticle-formulated VE-822 reproduced the developmental disruptions seen with Atr deletion. Genetic deletion of Atr blocked tumorigenesis in medulloblastoma-prone SmoM2 mice. Our data show that p53-driven apoptosis and cell cycle arrest – and, in the absence of p53, non-apoptotic cell death – redundantly limit growth in ATR-deficient progenitors. These mechanisms may be exploited for treatment of CGNP-derived medulloblastoma using ATR inhibition.
Appropriate brain development requires genomic fidelity. Diverse mutations that increase DNA damage can restrict brain growth, causing microcephaly (Lee and McKinnon, 2007; McMahon et al., 2014; Orii et al., 2006; Rosin et al., 2015; Williams et al., 2015). Combined with p53 deletion, mutations in DNA repair genes may produce unrestricted growth in medulloblastoma (Frappart et al., 2009; Lee and McKinnon, 2002; Tong et al., 2003), a malignant tumor of neural progenitors. The serine/threonine kinase ATR (ataxia telangiectasia and Rad3-related protein) mitigates proliferation-associated DNA damage (Brown and Baltimore, 2000) and has been implicated in microcephaly in Seckel syndrome (O'Driscoll et al., 2003). In mice, conditional deletion of Atr throughout the CNS during embryonic development similarly induces microcephaly, but particularly affects the cerebellum and ganglionic eminence (Lee et al., 2012). The precise mechanisms that cause DNA damage to impair growth specifically in the brain are unknown. We investigated the processes that restrict growth when Atr is deleted in the cerebellum in order to shed light on the pathogenesis of microcephaly associated with DNA damage and to determine whether the reliance of neural progenitors on ATR persists in progenitor-derived medulloblastoma.
Cerebellar growth depends on the postnatal proliferation of cerebellar granule neuron progenitors (CGNPs) in the external granular layer (EGL) along the outside of the cerebellum, which peaks between postnatal day (P) 5 and 7 (Hatten and Heintz, 1995). Although Atr deletion in nestin (NES)+ progenitors blocks cerebellar development prior to postnatal neurogenesis (Lee et al., 2012), the specific cellular processes restricting the growth of ATR-deficient progenitors have not been discerned. In vitro, ATR reduces DNA damage during proliferation by stabilizing stalled replication forks and by arresting the cell cycle to allow for DNA repair (Cimprich and Cortez, 2008). Replication fork collapse following ATR disruption produces DNA double-strand breaks (DSBs), which are considered the most toxic form of DNA damage (Couch et al., 2013). ATR also maintains S- and G2-phase checkpoints by phosphorylating CHK1 (CHEK1) in response to DNA damage (Smith et al., 2010). Determining how these functions combine in vivo to sustain neural progenitors is essential to understanding both the pathogenesis of Seckel syndrome and the requirement for ATR in brain development.
We found that ATR-deficient CGNPs continued to proliferate despite accumulating DNA damage, which induced population-wide, p53-dependent apoptosis. Blocking apoptosis in ATR-deficient CGNPs through co-deletion of the key apoptotic mediators Bax and Bak (Bak1) or co-deletion of p53 (Trp53) did not fully rescue the resulting cerebellar hypoplasia. Rather, premature cell cycle exit in Atr/Bax/Bak mutants or non-apoptotic cell death in Atr/p53 mutants redundantly limited CGNP population growth. Atr-deleted CGNPs also demonstrated diverse chromosomal abnormalities that were intensified by restriction of apoptosis. CGNPs are the cells of origin for sonic hedgehog (SHH) subgroup medulloblastomas; we found that Atr deletion blocked the tumorigenic effect of constitutive SHH activation in transgenic SmoM2 mice. These investigations define a crucial role for ATR in maintaining genomic integrity during brain development and suggest that ATR dependence might be exploited for medulloblastoma therapy.
Atr deletion induces CGNP apoptosis and cerebellar hypoplasia
To analyze ATR function in CGNPs, we generated mice with conditional Atr deletion (Brown and Baltimore, 2003) in the MATH1 lineage. Math1 (Atoh1) is expressed by CGNPs in a rostrocaudal progression beginning at the anterior margins of the cerebellar cortex at embryonic day (E) 12.5 (Hatten and Roussel, 2011). We crossed the Math1-Cre and AtrloxP/loxP mouse lines to generate Math1-Cre;AtrloxP/loxP (AtrM-cre) mice.
AtrM-cre mice were viable and fertile, but by P12 displayed tremor and ataxia, suggesting impaired postnatal neurogenesis. These animals were born with hindbrains of normal appearance, with the EGL spread over the primordial cerebellum. By P3, however, the EGL in AtrM-cre mice was noticeably thinner and less foliated than in Math1-Cre littermate controls with intact Atr (Math1-Cre;Atr+/+ or Math1-Cre;AtrloxP/+). Cerebellar hypoplasia in AtrM-cre mice became progressively more apparent with age (Fig. 1A).
We noted sparing of the posterior regions of the cerebellum in AtrM-cre mice (Fig. 1A, middle row). To determine if this sparing was the result of incomplete recombination in these regions or was due to reduced dependence on ATR, we used hGFAP-Cre to delete Atr throughout the CGNP population. hGFAP-Cre mice express Cre recombinase in neural stem cells, targeting all cells of the cerebellum except the Purkinje cells (Andrae et al., 2001). In the resulting hGFAP-Cre;AtrloxP/loxP (AtrG-cre) mice, we found hypoplasia across the entire EGL. The EGL was thinned by P0 and completely lost by P7. These mice showed more severe tremor and died by P18 (Fig. 1A, bottom row; Fig. S1A,B).
To understand the cause of cerebellar hypoplasia in Atr mutants, we analyzed patterns of proliferation, DNA damage and apoptosis in AtrM-cre and AtrG-cre mice by immunohistochemistry (IHC). Although prior studies showed that deletion of Atr in Nes-expressing progenitors blocks the prenatal proliferation of CGNP precursors (Lee et al., 2012), we observed no reduction in the postnatal proliferation of CGNPs in P3 AtrM-cre or P0 AtrG-cre cerebella, as demonstrated by the fraction of EGL cells expressing the mitosis marker phosphorylated histone H3 (pH3) (Fig. 1B,C, Fig. S1C) or the cycling cell marker proliferating cell nuclear antigen (PCNA) (Fig. S3D). However, we found significantly increased DNA damage as demonstrated by phosphorylated histone H2A.X (γH2A.X), increased activation of the cellular DNA damage response as shown by phosphorylated ATM (p-ATM) and p53 (p-p53), and increased apoptosis as shown by the levels of cleaved caspase 3 (cC3) (Fig. 1B,C, Fig. S1C,D). Atr-deleted CGNPs therefore proliferate robustly, but accrue extensive DNA damage and undergo apoptosis.
Deletion of Bax and Bak prevents cell death in Atr mutant CGNPs
Extensive apoptosis complicated further analysis of ATR-deficient CGNPs. Prior studies showed that deletion of the apoptosis gatekeeper Bax is sufficient to stabilize CGNPs with radiation-induced or proliferation-associated DNA damage (Chong et al., 2000; Garcia et al., 2013; Williams et al., 2015). We found, however, that apoptosis of Atr-deleted CGNPs and cerebellar hypoplasia were not prevented by Bax co-deletion in Math1-Cre;AtrloxP/loxP;BaxloxP/loxP (Atr;BaxM-cre) animals (Fig. 2A, Fig. S2A,B). Co-deletion of Atr and the Bax homolog Bak also failed to stabilize the CGNP population in Math1-Cre;AtrloxP/loxP;Bak−/− (Atr;BakM-cre) mice (Fig. S2A,B). However, combining genetic deletion of both Bax and Bak in ATR-deficient mice (Math1-Cre;AtrloxP/loxP;BaxloxP/loxP;Bak−/−, or Atr;Bax;BakM-cre) effectively prevented CGNP apoptosis, partially rescuing cerebellar growth (Fig. 2A,B, Fig. S2A,C). TUNEL staining did not detect non-apoptotic cell death in Atr;Bax;BakM-cre cerebella that might have been missed by cC3 studies (Fig. 2B,D). These data show that Atr deletion in CGNPs, in contrast to other pro-apoptotic stimuli such as radiation, activated both BAX- and BAK-driven apoptosis.
We compared markers of proliferation, DNA damage and the DNA damage response in CGNPs of P3 Atr;BaxM-cre, Atr;Bax;BakM-cre and Atr-intact Math1-Cre;BaxloxP/loxP;Bak−/− (Bax;BakM-cre) controls by IHC. The fractions of pH3+ and PCNA+ CGNPs were similar in Atr mutant and Atr-intact genotypes (Fig. 2D, Fig. S2C, Fig. S3C,D). CGNPs in Atr;BaxM-cre and Atr;Bax;BakM-cre mice showed similar levels of γH2A.X. However, p-ATM, p-p53, and the cell cycle inhibitor p21 (CDKN1A) were all markedly increased in Atr;Bax;BakM-cre CGNPs (Fig. 2B,D, Fig. S2B,C). Apoptosis-disabled, ATR-deficient CGNPs were thus able to activate p53 without undergoing cell death.
Atr and p53 double-mutant CGNPs undergo caspase-independent cell death
To determine the functional significance of p53 activation in CGNPs of AtrM-cre mice, we examined the phenotype of Math1-Cre;AtrloxP/loxP;p53loxP/loxP (Atr;p53M-cre) mice. p53 deletion prevents CGNP apoptosis after radiation-induced DNA damage (Herzog et al., 1998). p53 deletion did not completely rescue cerebellar hypoplasia in Atr mutants (Fig. 2A, Fig. S3A), despite effective suppression of the p53 axis, as demonstrated by the absence of p21 and p53 activation (Fig. 2C,D). The fraction of pH3+ and PCNA+ CGNPs in Atr;p53M-cre mice did not differ from Atr-intact controls at P3 (Fig. 2D, Fig. S3B,D). Atr;p53M-cre CGNPs accumulated significantly more DNA damage than AtrM-cre or Atr;Bax;BakM-cre CGNPs (Fig. 2D). This increased level of DNA damage did not trigger activation of caspase 3 (Fig. 2D). However, TUNEL staining demonstrated that cell death had occurred without caspase activation (Fig. 2C,D).
Atr;p53M-cre CGNPs showed abnormal morphologies that were different to the pyknotic nuclei found in the AtrM-cre EGL. These features included giant multinucleated cells and cells with micronuclei (Fig. S3E), consistent with published descriptions of p53-deficient cancer cells undergoing necrosis after DNA damage (Vakifahmetoglu et al., 2008). To examine whether Atr;p53M-cre CGNPs were dying through regulated necrosis, we compared the phosphorylation of MLKL, an essential step in the necroptosis pathway (Cai et al., 2014). We found scattered CGNPs showing phosphorylated MLKL (p-MLKL) in Math1-Cre, p53M-cre and Atr;p53M-cre mice, but no increase in the Atr;p53M-cre genotype (Fig. S3F). Thus, ATR/p53-deficient CGNPs undergo cell death without evidence of activating the apoptotic or necroptotic pathways.
Accelerated cell cycle exit in Atr;Bax;BakM-cre CGNPs
To account for the incomplete rescue of cerebellar growth in Atr;Bax;BakM-cre and Atr;p53M-cre mice despite the prevention of CGNP apoptosis, we investigated CGNP proliferation over time. We compared the mitotic rate of AtrM-cre, Atr;Bax;BakM-cre and Atr;p53M-cre CGNPs in the EGL with Atr-intact controls at P3, P5 and P7. We first determined that EGL mitotic rates did not vary significantly between the Atr-intact Math1-Cre, Bax;BakM-cre and p53M-cre control genotypes, allowing us to pool controls (Fig. S3G). The mitotic rates of AtrM-cre CGNPs did not differ from controls with any statistical significance at P3, P5 or P7: both groups showed a small decrease in proliferation by P7 (Fig. 2E). By contrast, the proliferation rate of Atr;Bax;BakM-cre CGNPs decreased markedly from P3 to P5 and from P5 to P7, and was significantly lower than in controls at both P5 and P7. These data show that non-proliferating cells were increased in the P5 and P7 EGL of Atr;Bax;BakM-cre mice but not in AtrM-cre mice with intact apoptosis. This increased cell cycle exit at P5 and P7 was p53 dependent, as the mitotic rate of apoptosis-disabled Atr;p53M-cre CGNPs was significantly higher than the Atr;Bax;BakM-cre CGNP mitotic rate at both P5 and P7.
We noted clusters of cells in P20 Atr;Bax;BakM-cre cerebella in the typically depopulated region where the EGL had been. These cells expressed the neuronal marker NEUN (RBFOX3) (Fig. 2F) and were PCNA− (data not shown), identifying them as differentiated neurons. The location of these neurons suggests that they derive from CGNPs, and similar ectopic neurons were not seen in AtrM-cre, Atr;p53M-cre or Atr-intact control cerebella. These observations provide evidence that premature, p53/p21-driven CGNP cell cycle exit after P3 limits the rescue effect of disabled apoptosis in Atr;Bax;BakM-cre cerebella.
Cell cycle checkpoint failure in ATR-deficient CGNPs with DNA damage
ATR coordinates the cellular response to replication stress by activating cell cycle checkpoints (Abraham, 2001). Our studies of cerebellar sections indicated that Atr-deleted CGNPs continue to proliferate after developing detectable DNA damage. To assess checkpoint function in Atr mutants, we used fluorescence-activated cell sorting (FACS) to analyze freshly isolated CGNPs from the cerebella of P3 mice with and without Atr deletion, including AtrM-cre, Atr;Bax;BakM-cre and Atr;p53M-cre knockout genotypes as well as Math1-Cre, Bax;BakM-cre and p53M-cre controls. All cells were stained for DNA content (FxCycle Violet), γH2A.X and pH3.
Atr-deleted genotypes showed a strongly γH2A.X+ subpopulation that was not seen in controls (Fig. S4A). To generate Atr-intact CGNPs with comparable DNA damage, we subjected P3 wild-type (WT) mice to X-ray radiation (2 Gy) and then isolated and analyzed CGNPs 2 h later. These WT irradiated CGNPs showed γH2A.X staining that was comparable in intensity to the γH2A.X+ CGNPs in Atr-deleted mice (Fig. S4A). DNA content staining effectively identified cells at G1, S and G2/M phases (Fig. S4A), and pH3 staining distinguished G2 from M phase (Fig. S4B). We found no statistically significant differences in cell cycle distribution between non-irradiated control genotypes (Fig. S4C), and these controls were pooled for further analysis.
Without γH2A.X selection, comparison of P3 CGNP cell cycle distribution between controls and Atr-deleted genotypes showed a significant difference only in the Atr;Bax;BakM-cre CGNPs. These cells showed enrichment of G2 at the expense of G1 phase (Fig. 3A), consistent with either delayed progression or G2 arrest. However, comparing the γH2A.X+ subsets from Atr-deleted genotypes and irradiated WT mice, we noted marked differences in M phase. All three Atr-deleted genotypes demonstrated CGNPs that were both γH2A.X+ and pH3+ (Fig. 3B). By contrast, pH3+ CGNPs were almost undetectable in the irradiated WT mice (Fig. 3B).
FACS results were confirmed by dual immunofluorescence (IF) for pH3 and γH2A.X or phospho-53BP1 (p-53BP1; 53BP1 is also known as TRP53BP1), another marker of DNA damage (Fig. 3C). First, in WT non-irradiated controls, pH3+ cells were always negative for γH2A.X and p-53BP1. Second, in irradiated WT mice, CGNPs that were γH2A.X+ or p-53BP1+ were never pH3+. Finally, in all Atr-deleted genotypes, γH2A.X+/pH3+ and p-53BP1+/pH3+ cells were common (Fig. 3C). The γH2A.X+/pH3+ CGNPs in Atr-deleted genotypes either entered M phase with DNA damage or persisted in M phase after acquiring DNA damage. These data do not conclusively show whether γH2A.X+, Atr-deleted CGNPs are able to complete mitosis. However, Atr deletion clearly permitted cells with DNA damage to be mitotic – a condition that is not seen in Atr-intact CGNPs.
Additionally, in each of the Atr-deleted genotypes, we found that γH2A.X+ CGNPs were distributed across all phases of the cell cycle, with relative enrichment in S, G2 and M phases, compared with γH2A.X+ CGNPs from irradiated WT mice (Fig. 3D). Consistent with these data, IF studies showed colocalization of γH2A.X with markers of G1 (cyclin D1), S (BrdU) and G2 (cyclin B1) phases (Fig. S4D) in AtrM-cre cerebella. G1 depletion in Atr-deleted, γH2A.X+ CGNPs could indicate either loss of H2A.X phosphorylation in G1 cells through DNA repair, or slowed return of γH2A.X+ cells to G1. The abundance of γH2A.X+ cells in each phase of the cell cycle, however, suggests that CGNPs with Atr deletion and DNA damage can progress through the cell cycle and, in the case of Atr;Bax;BakM-cre, have not all permanently arrested by P3. Altogether, at this single time point, most ATR-deficient CGNPs did not show evidence of impaired cell cycle function. However, Atr deletion increased the fraction of CGNPs with DNA damage and altered the barriers that normally prevent them from undergoing mitosis.
FACS data confirmed the increased cell death seen in AtrM-cre and Atr;p53M-cre CGNPs, which both showed increased sub-G1 fractions. We did not detect an increase in sub-G1 γH2A.X+ CGNPs in Atr;Bax;BakM-cre cerebella (Fig. 3E). These increases in sub-G1 fractions are consistent with our observations of apoptotic cell death in the AtrM-cre genotype and non-apoptotic cell death in the Atr;p53M-cre genotype.
ATR protects CGNP chromosome integrity
Our finding of impaired checkpoint function suggested that genomic fidelity might be impaired in Atr-deleted CGNPs. In vitro studies have shown incomplete DNA replication and chromosomal fragmentation in Atr mutant cells (Brown and Baltimore, 2000). To determine if chromosomal damage accumulates in ATR-deficient cells in vivo and to quantify the damage, we karyotyped Atr-deleted CGNPs freshly harvested from P3 cerebella. We dissociated CGNPs from AtrM-cre, Atr;Bax;BakM-cre and Atr;p53M-cre mice and Math1-Cre, Bax;BakM-cre and p53M-cre controls. Dissociated cells were treated ex vivo with Colcemid, and metaphase spreads were analyzed by conventional karyotyping and spectral karyotyping (SKY). We noted frequent chromosome breaks in all Atr mutant genotypes and a spectrum of abnormalities, including dicentric chromosomes, end-to-end fusions, deletions, translocations, and whole chromosome gains or losses (Fig. 4A). In addition, Atr;p53M-cre CGNP chromosomes showed segregation defects (premature sister chromatid separation). These abnormalities were not detected in controls (AtrM-cre 35%, P<0.01; Atr;Bax;BakM-cre 48%, P<0.01; Atr;p53M-cre 49%, P<0.01; Student's t-test).
For statistical analysis, we divided metaphases with varying degrees of chromosomal damage into four bins: 1-10 chromosome breaks/cell; >10 chromosome breaks/cell and <50% chromosomes fragmented (mild); >10 chromosome breaks/cell, >50% chromosomes fragmented, and some preserved chromosome morphology (moderate); and >10 chromosome breaks/cell, >50% chromosomes fragmented, and complete loss of chromosome morphology (severe) (Fig. 4B). In all Atr-deleted genotypes, the frequencies of cells with all degrees of chromosomal damage were increased, but the increases in AtrM-cre cells with moderate and severe damage were not statistically significant. Moderate and severe damage, however, were particularly enriched in the apoptosis-incompetent Atr;Bax;BaxM-cre and Atr;p53M-cre genotypes (Fig. 4C). These data show that CGNPs require ATR to maintain chromosomal integrity during proliferation, and that apoptosis limits the accumulation of genetic abnormalities by removing damaged cells from the population. Karyotype analyses also reaffirm that CGNPs with damaged DNA progress to M phase in Atr mutants.
Transcriptomic adaptations to ATR deficiency are predominantly p53 driven
We used RNA-Seq to determine the transcriptomic alterations induced by proliferation-associated DNA damage in the absence of Atr. Despite a wide body of literature on the cellular consequences of ATR loss, the impact on the transcriptome had not previously been reported. The prolonged survival of Atr-deleted CGNPs in Atr;Bax;BaxM-cre and Atr;p53M-cre mice allowed us to analyze mRNA abundance without losing cells with DNA damage from the population through cell death. We isolated and purified total RNA for RNA-Seq analysis from the cerebella of apoptosis-incompetent, ATR-deficient P3 mice (Atr;Bax;BakM-cre and Atr;p53M-cre) and corresponding age-matched controls (Bax;BakM-cre and p53M-cre). The abundance of mRNAs may be altered by changes in either production or degradation. ATR has been implicated in RNA processing (Chandris et al., 2010), and Atr deletion may alter both gene transcription and mRNA stability. Our data thus identify transcriptomic, rather than transcriptional changes.
We noted that Atr deletion evoked a richer, more complex pattern of transcriptomic changes in the Bax/Bak co-deleted background compared with the p53 co-deleted background. Principal component analysis (PCA) demonstrated that the second component of variation explained 16% of variance in global gene expression, and separated Atr;Bax;BakM-cre from Bax;BakM-cre (P=4.41×10−7, Hotelling's t-square test) but not Atr;p53M-cre from p53M-cre (P=0.20, Hotelling's t-square test) (Fig. 5A). This difference in separation on PCA indicates that p53-deleted CGNPs are more restricted in their transcriptomic response to ATR deficiency. Consistent with reduced transcriptional responsiveness with Atr/p53 co-deletion, the number of genes differentially expressed in Atr;Bax;BakM-cre cerebella compared with controls was significantly greater than in Atr;p53M-cre cerebella compared with controls (P=2.8×10−52, Fisher's exact test). Using a false discovery rate (FDR) of <0.01 and considering only genes meeting a fold-change (FC) criterion of |log2(knockout/control)|>1.5, we identified 339 genes differentially expressed by ATR-deficient cells in the Bax/Bak-deleted background, as compared with 56 genes differentially expressed by ATR-deficient cells in the p53-deleted background, with nine genes represented in both sets (Fig. 5B, Tables S1 and S2). This overlap demonstrates a high degree of correspondence between the gene sets (P=2.0×10−8, hypergeometric test). However, P-values for differentially expressed genes were consistently smaller in the p53-deleted comparison, as visualized in plots of FC versus significance (Fig. 5C), indicating an overall smaller transcriptomic effect of Atr deletion in the absence of p53. Expression microarray analysis comparing mRNA from Atr;Bax;BakM-cre versus Bax;BakM-cre cerebella defined a gene set that corresponded well with the differential gene set defined by RNA-Seq (Fig. S5A). By contrast, microarray analysis was unable to identify any differentially expressed genes in the Atr;p53M-cre versus p53M-cre comparison, consistent with an overall low signal-to-noise ratio in the absence of p53.
Pathway analysis of the 339 genes differentially expressed in Atr;Bax;BaxM-cre CGNPs demonstrated a strong correspondence between these genes and the molecular signature of p53-dependent transcription. Activation of p53 can induce both apoptosis and cell cycle arrest (Zuckerman et al., 2009), and we noted that regulators of both processes were enriched in the Atr;Bax;BaxM-cre gene set (Table 1). By contrast, we did not identify any pattern in Atr;p53M-cre CGNPs that could be attributed to a specific transcriptional regulator (Table S3) and, consistent with the lack of increased MLKL activation, we found no elevation of necroptosis-associated genes such as RIP kinases.
We used IF to determine the protein expression patterns of trophinin (TRO), which was among the nine genes identified as differentially expressed in both Atr-deleted genotypes, and for which we were able to obtain effective antibodies. Examining Bax;BakM-cre and p53M-cre control brains, we found that TRO was expressed in all differentiated neurons of the brain, but absent in progenitor regions (Fig. 5D). In Atr-deleted Atr;Bax;BaxM-cre and Atr;p53M-cre cerebella, however, TRO was detected within the CGNP layer (EGL), indicating upregulation of the protein as well as the RNA.
From the larger set of 124 genes that were upregulated in the Atr mutant cerebella of both p53-deleted and Bax/Bak-deleted backgrounds with FDR<0.01, irrespective of FC, we selected Eif4ebp1 for further analysis. EIF4EBP1 is known to regulate protein translation downstream of mTOR (Fingar et al., 2002), a PIKK kinase homologous to ATR (Lovejoy and Cortez, 2009). Atr-deleted CGNPs demonstrated marked upregulation of EIF4EBP1 protein. In control cerebella, EIF4EBP1 was limited to a subset of CGNPs at the outer margin of the EGL, whereas in Atr-deleted cerebella EIF4EBP1 was detected throughout the EGL (Fig. 5D). Our in situ IF data, along with our microarray results, thereby validate our RNA-Seq findings.
We further examined our RNA-Seq data to determine whether Atr deletion produced recurrent point mutations or other changes in RNA sequence. We did not detect a significant change in mismatch rate associated with Atr deletion (Fig. S5B). Fusion transcript detection software suggested a small number of fusion products, only one of which, with unknown biological function, emerged with statistical significance (Fig. S5C). Together with our karyotype analysis, these data suggest that, although Atr deletion significantly altered chromosome structure, it did not substantially increase the rate of point mutations or recurrent fusion transcripts.
ATR is required for medulloblastoma tumorigenesis
Prior studies have shown that deletion of diverse DNA repair pathway genes, combined with p53 deletion, induces medulloblastoma in mice (Frappart et al., 2007, 2009; Holcomb et al., 2006; Lee and McKinnon, 2002). To determine if Atr mutation predisposes mice to medulloblastoma, we followed the viability and neurologic function of 12 Atr;p53M-cre and 12 Atr;Bax;BakM-cre mice for more than 300 days (supplementary Materials and Methods). These mice showed no progressive neurological changes or other evidence of tumorigenesis, and their cerebella remained hypoplastic (data not shown). The absence of tumors despite impaired genomic stability and p53 deletion suggests that medulloblastoma tumorigenesis, like postnatal neurogenesis, might require ATR function.
We directly tested this suggested requirement for ATR in medulloblastoma tumorigenesis by deleting Atr in tumor-prone hGFAP-Cre;SmoM2 mice (Schuller et al., 2008). CGNPs in SmoM2G-cre mice with intact Atr gave rise to medulloblastoma with 100% frequency by P7, causing the mice to die from tumor progression by P20 (Fig. 6A). By contrast, Atr deletion in SmoM2;AtrG-cre animals completely blocked tumor formation, as seen and quantified by Hematoxylin and Eosin (H&E) staining and by IF for PCNA, which marks proliferating tumor cells (Fig. 6A). The inability of Atr-deleted CGNPs in SmoM2;AtrG-cre mice to give rise to tumors suggests that ATR inhibition might effectively restrict medulloblastoma growth.
ATR inhibitor administered in vivo induces DNA damage specifically in CGNPs
To determine whether acute disruption of ATR function during postnatal neurogenesis in WT mice would recapitulate conditional Atr deletion, we developed a novel formulation of the small molecule ATR inhibitor VE-822 (pVE-822) (Charrier et al., 2011; Fokas et al., 2012). We first demonstrated that VE-822 induces DNA damage and apoptosis in isolated CGNPs in vitro in a time- and dose-dependent manner (Fig. S6A). To enhance delivery of VE-822 across the blood-brain barrier, we encapsulated VE-822 in poly(2-oxazoline) micelles, generating pVE-822 (Fig. S6B,C). We found that pVE-822, administered by intraperitoneal (IP) injection, induced γH2A.X and cC3 in the EGL, without affecting the surrounding differentiated neurons in the cerebellum or cortical neurons (Fig. 6B,C). As with Atr deletion, acute ATR inhibition by pVE-822 promoted upregulation of EIF4EBP1 in CGNPs, demonstrating that EIF4EBP1 is an effective biomarker of ATR disruption (Fig. 6D). These results show that pVE-822 effectively crosses the blood-brain barrier in bioactive concentrations after IP administration, and that acute, global loss of ATR function specifically disrupts CGNPs. The absence of neuronal toxicity associated with pVE-822 administration is consistent with recent literature demonstrating that ATR inhibition does not damage non-proliferating, differentiated neurons (Kemp and Sancar, 2016).
Our data show that ATR is required to mitigate endogenous DNA damage during postnatal CGNP proliferation in order to maintain genomic stability. A prior study that deleted Atr using Nes-Cre, which is expressed by E9.5 with an effect by E10.5, found that embryonic CGNP precursors exit the cell cycle prematurely, blocking cerebellar development before postnatal CGNP proliferation (Lee et al., 2012). By contrast, in AtrG-cre and AtrM-cre mice, where Cre expression does not occur until E11.5 and E12.5 (Andrae et al., 2001; Machold and Fishell, 2005), respectively, CGNPs proliferated normally in the postnatal period and the effect of Atr deletion was not observed until P0 and P3, respectively. Postnatal proliferation in Atr-deleted CGNPs produced widespread DNA damage that activated ATM and p53 and induced apoptosis, resulting in cerebellar hypoplasia.
Blocking apoptosis by co-deletion of Bax and Bak or p53 did not fully rescue cerebellar hypoplasia. Cerebellar growth failure persisted despite attenuation of apoptosis due to inappropriate cell cycle exit and differentiation in Atr;Bax;BakM-cre CGNPs or caspase-independent cell death in Atr;p53M-cre CGNPs. Karyotype analysis showed that Atr-deleted CGNPs developed extensive chromosomal abnormalities, and RNA-Seq analysis revealed the p53 pathway to be the predominant driver of the transcriptomic response to ATR-deficient proliferation.
Atr deletion in medulloblastoma-prone mice (SmoM2;AtrG-cre) completely abrogated tumor formation, suggesting a therapeutic potential for ATR inhibition as a novel treatment for medulloblastoma. The anti-tumor effect of targeting ATR might be magnified in our model, as early deletion of Atr by hGFAP-Cre depletes the pool of CGNPs from which tumors may originate. However, CGNPs were present in P0 AtrG-cre mice (Fig. 1A), and SmoM2 expression consistently failed to induce tumor growth from these cells in SmoM2;AtrG-cre animals. Future work will assess the therapeutic potential of the ATR inhibitor pVE-822 in mice with established medulloblastoma. Indeed, we have already shown here that acute in vivo ATR inhibition by pVE-822 in WT mice reproduces the effects of conditional Atr deletion. Taken together, these results define a crucial role for ATR in maintaining genomic integrity in rapidly proliferating neural progenitors and in medulloblastoma cells.
The finding that 5-9% of Atr-deleted CGNPs were cC3+ demonstrates a strong induction of apoptosis. Unlike CGNP death after radiation, the death of these cells is not expected to be a synchronous process. Moreover, dying cells are cC3+ for only a limited time because they are rapidly cleared by phagocytosis and cC3 is known to have a short half-life (Elliott and Ravichandran, 2010; Walsh et al., 2011). Detecting 5-9% of cells expressing cC3 thus indicates a high rate of cell death.
Apoptosis in Atr-deleted CGNPs, in contrast to irradiated Atr-intact CGNPs (Chong et al., 2000; Williams et al., 2015), was not prevented by co-deletion of Bax, but was prevented by co-deletion of both Bax and Bak. In other cell types, BAK-driven apoptosis can be activated by genotoxic stress during mitosis (Chu et al., 2012; Flores et al., 2012). Atr-deleted CGNPs with DNA damage were markedly more prone to be mitotic than irradiated CGNPs in Atr-intact mice. The mitosis of Atr-deleted CGNPs with DNA damage might activate BAK-driven apoptosis that is not activated by radiation in WT mice.
The dependence of CGNPs on ATR to prevent DNA damage suggests that these cells are particularly prone to endogenous replication stress. ATR is activated in response to stalled DNA replication forks (Nam and Cortez, 2011), which can be caused by exogenous (Harley et al., 2016; Harper et al., 2010) or endogenous (Willis et al., 2013) factors. We have previously shown that normal CGNP proliferation produces endogenous DNA damage, detectable as small γH2A.X+ foci (Williams et al., 2015). We now show that ATR mitigates endogenous, proliferation-associated DNA damage in CGNPs in order to maintain replication fidelity.
To account for the phenotype of Atr deletion we propose a model in which: (1) rapid proliferation promotes replication fork stalling from endogenous sources (Mirkin and Mirkin, 2007); (2) absence of ATR prolongs and collapses forks with consequent formation of DSBs (Couch et al., 2013); (3) ATM is activated and recruited to sites of DSBs (Shiloh, 2003); and (4) p-ATM activates p53, which ultimately induces apoptosis (Shiloh and Ziv, 2013) through BAX and BAK, and cell cycle arrest through p21 (Fig. 7). Because DNA damage and apoptosis occur throughout the entire CGNP population, cerebellar growth is effectively halted.
Apoptosis functions to cull CGNPs that acquire chromosomal abnormalities; thus, disabling apoptosis in ATR-deficient mice increases chromosomal fragmentation. p21 upregulation in Atr;Bax;BakM-cre CGNPs demonstrates that p53-induced cell cycle exit acts as an additional barrier to the propagation of genomic abnormalities in apoptosis-incompetent, Atr-deleted CGNPs. Genomic stability in CGNPs may be particularly important for preventing their rapid postnatal proliferation from degenerating into tumorigenesis. Medulloblastoma, the most common malignant pediatric brain tumor, arises from hindbrain neural progenitors including CGNPs (Yang et al., 2008). Chromosome damage has been observed in medulloblastoma, frequently in association with p53 mutation (Rausch et al., 2012). Importantly, despite severe chromosomal abnormalities in Atr;Bax;BakM-cre and Atr;p53M-cre mice, we never observed tumor formation. Indeed, we found that medulloblastoma tumorigenesis, like cerebellar development, requires ATR.
Since the physiological requirement for ATR in neural progenitors is recapitulated in the pathological growth of medulloblastoma, ATR may be a promising target for novel anti-tumor therapy. pVE-822 effectively crosses the blood-brain barrier in bioactive concentrations and induces DNA damage in proliferating cells. By damaging DNA in proliferating cells, pVE-822 might have a similar toxicity to conventional chemotherapeutic agents such as etoposide. An advantage of ATR inhibition over standard chemotherapy, however, is suggested by our data showing that Atr deletion kills cells even in the absence of p53. Whether the activity of pVE-822 against proliferating, non-tumor cells will limit its therapeutic potential remains to be tested.
MATERIALS AND METHODS
We generated AtrM-cre and AtrG-cre mice by crossing AtrloxP/loxP (Brown and Baltimore, 2003) mice with the Math1-Cre (Jackson Labs, stock #011104) (Matei et al., 2005) and hGFAP-Cre (Jackson Labs, stock #012886) (Zhuo et al., 2001) mouse lines, respectively. BaxloxP/loxP;Bak−/− mice (Takeuchi et al., 2005) were obtained from Jackson Labs (stock #006329). To generate mice with co-deletion of Atr/Bax/Bak, Atr/Bax, Atr/Bak and Bax/Bak, we crossed AtrM-cre and BaxloxP/loxP;Bak−/− animals, intercrossed the progeny, and selected mice of Atr;Bax;BakM-cre, Atr;BaxM-cre, Atr;BakM-cre and Bax;BakM-cre genotypes. We generated Atr;p53M-cre mice by crossing AtrM-cre with p53loxP/loxP mice (Jonkers et al., 2001) provided by the NCI (strain #01XC2). Medulloblastoma-prone animals with and without Atr deletion were born from the cross between hGFAP-Cre;AtrloxP/+ and SmoM2loxP/loxP;AtrloxP/loxP, in which tumorigenesis is induced by deletion of a loxP-flanked Stop cassette between the Smo promoter and coding region (Schuller et al., 2008). All mice were of species Mus musculus and crossed into the C57BL/6 background through at least five generations. We used equal numbers of male and female mice, as we did not observe any differences based on sex. Numbers (n) in figures indicate biological replicates, which were determined so as to measure a 25% difference in means with power=80% and α=0.05. Animal use was in keeping with the policies of the University of North Carolina at Chapel Hill Institutional Animal Use and Care Committee. P3 mice of the indicated genotypes received 2 Gy whole-body X-ray irradiation and were sacrificed 2 h later.
Immunostaining of cerebellar sections
Mouse brains were processed and immunostained as previously described (Gershon et al., 2013). Primary antibodies are listed in the supplementary Materials and Methods. Secondary antibodies of the indicated fluorophores were used at 1:2000 for IF. Cell death was detected by TUNEL assay [ThermoFisher Scientific (TFS) #C10617] (Galluzzi et al., 2009). DAPI and Hematoxylin were used as nuclear counterstains. BrdU analysis is described in the supplementary Materials and Methods.
Quantification of immunostaining
Stained slides were digitally imaged and positively stained cells were counted using Aperio Software (Aperio Technologies) for chromogen-stained slides or Tissue Studio (Definiens) for fluorescence, as previously described (Williams et al., 2015). The entire EGL region in each section was manually annotated and used for quantifications, which were normalized to the total number of nucleated cells in the designated region. The measurement of cerebellar cellularity is described in the supplementary Materials and Methods.
CGNPs from P3 Atr-deleted mice and controls were isolated as previously described (Lee et al., 2009). Briefly, we separated cerebella from the rest of the brain, removed the meninges, and dissociated the tissue in 20 units/ml papain (Worthington Biochemical Corporation #PDS) at 37°C for 15 min. CGNPs were then purified from the dissociated cerebellar tissue by successive rounds of centrifugation, discontinuous density gradient, and mesh filtering.
Flow cytometry on ATR-deficient CGNPs was performed by first fixing and permeabilizing (TFS #GAS-004) isolated CGNPs suspended in HBSS containing 33 mM glucose, as previously described (Stahl et al., 2008). CGNPs were then stained successively for DNA damage with e660-conjugated anti-γH2A.X Ser139 (eBioscience #50-9865) at 1:10, for M phase with 488-conjugated anti-pH3 (Cell Signaling Technology #9708) at 1:25, and for DNA content with FxCycle Violet (TFS #F-10347) at 1:50. Technical controls included no stain, single-stained and fluorescence-minus-one samples. FACS was performed on an LSR Fortessa (BD Biosciences). For all experiments, 10,000-50,000 cells were counted. Analysis of FACS data was performed using FlowJo v10.0.8 (FlowJo).
Cytogenetic analysis was performed on Atr-deleted CGNP metaphase spreads. Freshly isolated CGNPs from P3 animals were treated for 30 min with 100 nM Colcemid (TFS #15210-040) in Neurobasal-A Medium (TFS #10888-022) supplemented with 1× GlutaMAX-I (TFS #35050), 1× penicillin-streptomycin and 25 mM KCl at 37°C to block cell cycle progression in M phase. CGNPs were then were resuspended in 75 mM KCl, incubated at 37°C for 5 min, and fixed in methanol:acetic acid (3:1). The fixed cell suspension was dropped onto slides, stained in 0.08 µg/ml DAPI in 2× SSC for 3 min, and mounted in antifade solution (Vector Laboratories #H-1200). The stained slides were scanned using a Zeiss Axioplan 2i epifluorescence microscope equipped with a megapixel CCD camera (CV-M4+CL, JAI) controlled by Isis 5.2 imaging software (Metasystems International). Chromatid breaks were counted as single-break events, tri-radials and quadri-radials as two-break events each, and other complex chromatid exchanges were converted into the minimum number of breaks required for their theoretical reconstruction. Chromosome aberrations (fragments, rings, dicentric/tricentrics and large marker chromosomes) were recorded and the breaks required for these rearrangements were not added to the frequency of chromatid breaks.
The frequency of mismatches or of transcript fusions associated with Atr deletion were determined as described in the supplementary Materials and Methods.
P3 WT CGNPs subject to VE-822 or vehicle were analyzed for γH2A.X and cC3 (with β-actin loading control) by western blot as described in the supplementary Materials and Methods.
Metaphase spreads from the fixed cell suspensions were hybridized with SKY painting probes according to the manufacturer's protocol (Applied Spectral Imaging). SKY images were acquired with an SD300 Spectracube (Applied Spectral Imaging) mounted on a Nikon Eclipse E800 microscope using a custom-designed optical filter (SKY-1) (Chroma Technology). For each sample, a minimum of 20, but usually 50, metaphases were captured and fully karyotyped. The breakpoints on the SKY-painted chromosomes were determined by comparison with the corresponding DAPI karyotype and chromosomal abnormalities were described according to ISCN (2013).
RNA-Seq and differential expression analysis
For RNA-Seq analysis, total RNA was purified using the RNeasy Mini Kit (Qiagen #74104) from freshly dissected whole P3 cerebella. RNA quality and quantity were assessed by spectrophotometry and capillary gel electrophoresis. We generated stranded mRNA libraries using an Illumina TruSeq Stranded mRNA Library Prep Kit (Illumina #RS-122-2101). 60 ng/µl poly(A)-selected RNA from each sample was run in two lanes of a HiSeq 2000 sequencing instrument (Illumina) for 100 cycles of multiplexed paired-end reads.
We performed read pseudo-alignment and quantification using Kallisto v0.42.4 [Bray et al. (2015) preprint] with 200 bootstraps/paired-end read, aligning against the mouse mm9 genomic assembly, GRCm38 transcriptome definition. Downstream analysis was performed using the edgeR v3.12.0 software package (Robinson et al., 2010) in R-Studio v0.99.491 with R v2.11.1 (R Core Team, 2014). To assess differential expression, we used a general linear model in edgeR comparing Atr-deleted with control, setting the minimum counts per million at 2 and requiring greater than half the samples to pass the count threshold for a given transcript. Microarray validation is detailed in the supplementary Materials and Methods.
Biological pathway enrichment was determined by comparing transcriptomic data with Kyoto Encyclopedia of Genes and Genomes v77.1 (Kanehisa and Goto, 2000), WikiPathways (Kutmon et al., 2016) and Gene Ontology v1.2 (Ashburner et al., 2000) databases using EnrichNet v1.1 (Glaab et al., 2012), Panther v10.0 (Mi et al., 2013), and WebGestalt (Wang et al., 2013). Only genes above the differential expression significance threshold of log2(FC)>1.5 and FDR<0.01 were used in pathway analysis.
pVE-822 in vivo administration
For in vivo studies, WT mice were subject to IP injection with 60 mg/kg pVE-822 (see supplementary Materials and Methods for formulation) or an equal volume of vehicle every 12 h for three total injections from P3-5. 12 h following the final injection, animals were sacrificed and brains were prepared for immunostaining.
We thank the UNC Center for Gastrointestinal Biology and Disease Histology Core for processing tissue sections and Hematoxylin and Eosin staining; the UNC Translational Pathology Laboratory for help in staining and digitizing cerebellar sections; the UNC Flow Cytometry Core Facility for FACS assistance; Jeremy Simon (UNC Neuroscience Center) for transcript fusion analysis on the RNA-Seq data; Matthew Soloway (UNC Lineberger Comprehensive Cancer Center) for uploading RNA-Seq and microarray data to GEO; Jing Gao (UNC Eshelman School of Pharmacy) for technical assistance with formulating pVE-822; Eric Brown (University of Pennsylvania, Pennsylvania, PA, USA) for AtrloxP/loxP mice; David Rowitch (UCSF, San Francisco, CA, USA) and Robert Wechsler-Reya (Sanford-Burnham Medical Research Institute, La Jolla, CA, USA) for Math1-Cre mice; and Eva Anton (UNC Neuroscience Center) for hGFAP-Cre mice.
P.Y.L. and T.R.G. conceived and designed the experiments and wrote the manuscript. G.J.N. and C.S. obtained and analyzed karyotype and SKY data. M.S.-P. and A.V.K. conceived the nanoparticle formulation of VE-822 in collaboration with P.Y.L. and T.R.G., and prepared and analyzed pVE-822 with the assistance of D.H. J.S.P. performed mutational analysis and consulted on transcriptomic studies. All other experiments were performed by P.Y.L. and analyzed by P.Y.L. and T.R.G.
This work was supported by the National Institute of Neurological Disorders and Stroke [R01NS088219 to T.R.G.]; the National Cancer Institute [F30CA192832 to P.Y.L., P30CA008748 to G.J.N. and C.S., P30CA016086 to J.S.P., and U01CA151806 to D.H., M.S.-P. and A.V.K.]; and Alex's Lemonade Stand Foundation for Childhood Cancer (M.S.-P. and T.R.G.). Deposited in PMC for release after 12 months.
RNA-Seq and microarray data are available at Gene Expression Omnibus under accession number GSE85394 (www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE85394).
The authors declare no competing or financial interests.