Cyclin A2 promotes DNA repair in the brain during both development and aging

Cyclin A2 loss delays embryonic forebrain development

In order to quantitatively describe the neuropathology of Cyclin A2 loss in the VZ/SVZ, we performed high-resolution analyses of the CCNA2-/- brains using unbiased stereological methodologies. We generated CCNA2-null brains by intercrossing CCNA2fl/fl mice with Nestin-cre mice. Ablation of CCNA2 was confirmed by immunohistochemical staining (Supplemental Fig. S1). We focused our analyses on the VZ/SVZ of E14.5 and E17.5 mice. At E14.5, most radial glia divide symmetrically to expand the progenitor pool [13], while at later ages radial glia divide asymmetrically to self-renew and generate new neurons [14]. VZ/SVZ and cortical plate (CP) volumes and total number of cleaved caspase-3 positive cells in the entire VZ/SVZ and CP were determined in CCNA2fl/fl, Nestin-cre brains and compared to controls using unbiased stereology (Fig. 1A-C, Supplemental Table S1). E14.5 CCNA2fl/fl, Nestin-cre mice showed greater than 4-fold reduction in VZ/SVZ volume and greater than 2-fold reduction in CP volume (Fig. 1A-B).

By E17.5, the CP and VZ/SVZ volumes were not significantly different between groups (p = 0.068 and p = 0.5, respectively). We conclude that during the E14.5 ->E17.5 period, the amount of growth of the CCNA2^fl/fl, Nestin-cre VZ/SVZ was greater than that of the control VZ/SVZ.

To investigate the underlying cause of the early size reduction, we examined apoptosis in the VZ/SVZ and CP of these embryos by measuring the total number of cleaved caspase 3-positive cells in the VZ/SVZ and CP. We found greater than 5-fold increase in apoptosis in the VZ/SVZ and CP of CCNA2^fl/fl, Nestin-cre embryos compared to controls at E14.5. E17.5 mice displayed approximately 2.5-fold increase in apoptosis in both structures (Fig. 1C). Notably, the volume occupied by apoptotic cells in the VZ/SVZ was vastly increased by CCNA2 ablation compared to the increase observed in the cerebellar granule neuron progenitor cells of the external granule layer (Supplemental Table S1). These data indicate that forebrain neural progenitors are actually more sensitive to CCNA2 ablation than cerebellar neuron progenitors, yet cerebellar morphogenesis is more adversely affected post-CCNA2 ablation. To test if the observed growth delay was a property of neural progenitors in general, or was restricted to the embryonic period, we utilized an in vitro approach. We generated CCNA2-null neural progenitors dissected from the CCNA2^fl/fl forebrain ganglionic eminence and expanded in vitro. CCNA2 ablation was accomplished by infection with adenovirus encoding cre-recombinase. CCNA2-null neural progenitors showed increased BrdU incorporation, as well as delayed growth (Fig. 1F-I). Comparing the overall apoptosis rates to the overall proliferation rates, we concluded that the increase in apoptosis in the VZ/SVZ had minimal impact on the overall growth of the embryonic forebrain whereas apoptosis has more detrimental effects on cerebellar morphogenesis. Furthermore, we were perplexed by our finding of compensatory growth in CCNA2-null forebrains despite loss of a crucial cell cycle gene.

Theoretical models of compensatory forebrain growth

We sought to understand the paradoxical effects of CCNA2 ablation on embryonic neurogenesis. In particular, we interpreted the results from Fig. 1 in terms of the interplay between proliferation, self-renewal, apoptosis, and differentiation. The volume trajectories for both lines suggest autoregulatory saturation of the neuronal population by day E15. This plateau is similar to that shown in previous observations in the literature [8,15]. The simplest model of such population dynamics is logistic growth*¹, where proliferation rate decreases as a population approaches a limiting capacity. Logistic growth has been shown to well-explain the growth of many tissues including in the brain [16]. Since the main feature of logistic growth is a decrease in proliferation rate as a population approaches a certain threshold or carrying capacity*, two populations that initially grow at the same rate may converge to the same steady size over a sufficient amount of time. For this reason, it provides the most simplistic mechanism for the slow-down in the normal growth rate that would allow the CCNA2-null embryos to catch up to their CCNA2-intact counterparts.

In order to better understand the growth of CCNA2^fl/fl, Nestin-cre brains, we considered several hypotheses to explain the data. These hypotheses are presented as theoretical growth curves in Fig. 2A. The blue curve represents logistic growth of CCNA2-intact brains. Growth mediated by more rapid cell cycle (black curve) cannot explain the growth of CCNA2^fl/fl, Nestin-cre brains as it would require the CCNA2^fl/fl, Nestin-cre VZ/SVZ volume to be increased at E14.5, in addition to showing more rapid cell cycle transit in cells lacking a critical S-phase cyclin. Elevated apoptosis would result in a prolonged lag phase (green curve), yet our measurements of apoptosis do not show a sufficiently elevated programmed cell death rate to cause such a lag. Thus, an increased lag phase model cannot explain the growth of CCNA2^fl/fl, Nestin-cre brains because it would require the apoptosis rates to be much higher than what we observed in order to delay the transition from lag to log phase. The linear growth model (yellow curve), corresponding to a fixed-size proliferative pool, cannot explain the growth of CCNA2^fl/fl, Nestin-cre brains because it would either require apoptosis rates to be high relative to proliferation rates, which is not the case, or it would require that the VZ/SVZ stem cell niche be populated only by a single cell type [1].

Our considerations of these alternative models left us with only our delayed development model (red curve) to explain the data. To test this hypothesis, we generated a mathematical model of mouse forebrain development using previously reported data on cell cycle timings available in the scientific literature, as well as data from our own unbiased estimates of VZ/SVZ size during embryonic development (Fig. 1 and 2, and Supplemental Tables S2-S4). The variables that we track in our model are the total volumes of cells in each of the G1/S/G2/M phases, existing within each of the VZ/SVZ niches, which have a given mitotic age. The parameters in this model are the rates of transitions between each stage of the cell cycle. We incorporated logistic growth mechanism in determining cell fate, with the rate of Type I pro-liferative divisions (Fig. 2B) decreasing linearly as a function of total volume.

Assuming a conservatively rapid clearance rate for apoptotic cells, we found that apoptosis did not significantly affect the population dynamics* in the E14.5 to E17.5 period. Although the apoptosis rate increases many-fold, the apoptosis rate remains mathematically insignificant relative to the growth rate (Supplemental Table S1). Since we have determined that apoptosis is insignificant, a decrease in growth rate is likely the determinant of the observed developmental delay. For this reason, we expected that CCNA2 loss in a neural stem cell lengthens the cell cycle, a finding that has previously been identified in CCNA2-null mouse embryonic fibroblasts [17]. We assume that such cell cycle lengthening occurs during S and G2 phases as these are the phases during which Cyclin A2 protein is present.

Simulations of the VZ/SVZ volume are given in Fig. 2D, where we assumed that the SVZ has no volume at E13.5, and that all radial glia present at E13.5 have the capability to divide (on average) approximately five more times before terminally differentiating. We arrived at the number five in order to match the characteristics of the observations in our unbiased measurements for the CCNA2-intact controls. We then tested simulations as to how a delayed forebrain stem cell niche would behave.

At E14.5, the measured control VZ/SVZ volume is approximately 4 times that of CCNA2^fl/fl, Nestin-cre. Supposing that this cre expresses at E10.5 [18], and that the divisions in this period are primarily Type I symmetric divisions (Fig. 2B), which have a mean transit time of 9.2 hrs, then proliferative CCNA2^fl/fl, Nestin-cre radial glia have a mean transit time of 10.6 hrs. Isolating the excess time of 1.4 hrs to the S and G2 phases suggests that they are elongated by 25%. This also suggests that at E13.5, CCNA2^fl/fl Nestin-cre radial glia are capable on average of approximately 1 more division than in control animals. For this reason, we assumed that the radial glia are developmentally delayed, giving them the capability to divide six times before terminal differentiation during the E14.5-E17.5 time period. These simulations thus shed light as to why increased forebrain apoptosis does not affect overall growth rate in the CCNA2^fl/fl, Nestin-cre animals.

In summary, the notion that developmental delay results in compensatory growth in CCNA2^fl/fl Nestin-cre is best explained by a predominance of symmetric divisions (Type 1 in Fig. 2B) that expand the VZ/SVZ between E14.5 and E17.5, while neural progenitors in the control brains undergo more asymmetric and terminal divisions, slowing the growth of the control VZ/SVZ. Stated otherwise, the CCNA2^fl/fl Nestin-cre brains behave similarly to control brains, although delayed by approximately one day.

Our developmental delay model’s predictions include (1) that the CCNA2-null neural progenitor cells have a prolonged cell cycle time, (2) the total output of intermediate progenitors would be decreased early in development, (3) more neurons would be produced at later time points, and (4) at later time points the mutants would have a higher overall level of proliferation. In order to test the predictions of our model, we examined VZ/SVZ cytoarchitecture and proliferation in E14.5 embryos and P0 pups. We injected an E14.5 timed-pregnant mouse with CldU, euthanized the pregnant dam 2.5 hours later, and harvested embryos. We first examined VZ/SVZ cytoarchitecture in these embryos. Neural progenitor cells are characterized by expression of Pax6, localized adjacent to the ventricle. As they divide and mature into intermediate progenitors, they express Tbr2 and are located deeper within the VZ/SVZ [19]. We observed appropriately-localized Pax6 and Tbr2, indicating that loss of CCNA2 does not perturb VZ/SVZ cytoarchitecture (Fig. 3A-C). We quantified Tbr2 and found that there was no change in the number of Tbr2-positive cells in individual sections (Fig. 3D). However, in concordance with our model, there was a reduction in the total number of Tbr2 cells when we normalized the volumes for the smaller size of the CCNA2^fl/fl, Nestin-cre VZ/SVZ (Fig. 3D).

To confirm that the cell cycle was lengthened in CCNA2^fl/fl, Nestin-cre neural progenitors, we examined the localization of CldU-positive nuclei within the VZ/SVZ. Neural progenitors undergo interkinetic nuclear migration (INM), a process in which the localization of the nucleus within the cell depends upon cell cycle phase. During S-phase, the nucleus is located away from the ventricle, deep within the VZ/SVZ (Fig. 3A, E). Upon completion of DNA synthesis, the nucleus migrates back to the ventricular surface, where the cell divides. Cells located outside the prominent band of CldU-positive nuclei deep in the SVZ would therefore have completed S-phase and begun INM toward the ventricle. Increasing the duration of S phase would result in fewer CldU-positive nuclei that have completed DNA synthesis and begun their migration back to the ventricular surface. With this in mind, we measured the number of CldU-positive nuclei adjacent to the ventricle. Very few of these cells are noted in CCNA2^fl/fl, Nestin-cre embryos compared to controls (Fig. 3E-F), indicative of a lengthened S-phase. In summary, CCNA2 loss results in increased brain growth during E14.5->E17.5, despite showing prolonged cell cycle time and increased apoptosis.

Between E14.5 and E17.5, wild-type VZ is known to plateau and decay due to differentiation [8,15]. Due to this normal growth plateau, the CCNA2^fl/fl, Nestin-cre VZ recovers in volume relative to the control. This recovery, however, is misleading, as the cumulative neuronal output is predicted to be significantly depressed (Fig. 2E) in the mutant animals. Indeed, our model suggests that the CNS stem cell niche would exhibit behaviors at P0 characteristic of earlier embryonic stages. Stated otherwise, proliferation in CCNA2^fl/fl, Nestin-cre animals would remain elevated and a significant number of neurons would be generated later in development from the forebrain stem cell niche.

To test these predictions, we examined proliferation in the VZ/SVZ of P0 mice. We injected P0 mice with BrdU and euthanized them 30 minutes later. We then stained for BrdU (Fig. 4A), pH3, and Ki67. Consistent with our model, we observed an increase in all three of these proliferation markers in the VZ/SVZ of CCNA2^fl/fl, Nestin-cre mice (Fig. 4B). Furthermore, we measured neuronal output in newborn mice by determining the total number of NeuN-positive cells in the combined VZ/SVZ and intermediate zone (IZ) with unbiased stereological quantification of NeuN immunohistochemistry. We found that the total number of neurons in the combined VZ/SVZ and intermediate zone (IZ) was significantly increased, as predicted by our model (Fig. 4C-F). Thus, we conclude that CCNA2 loss does not affect stem cell self-renewal or migration to the CP and the forebrain develops appropriately, yet remains developmentally immature relative to a control brain. Furthermore, our experimental data and our mathematical model predict a maximal limit to the size of the forebrain stem cell niche, which we term the carrying capacity (V_max). V_max is constant between the control and CCNA2^fl/fl, Nestin-cre brains. Since at E14.5 the CCNA2-null neural progenitor cells are at a different point in the growth curve toward V_max, our model predicts that levels of proliferative cells would be elevated relative to controls as the CCNA2-null cells would not experience the autoinhibitory mechanisms acting on the control stem cell niche during equivalent epochs.

Mechanisms of increased cell cycle time in CCNA2 deficient cells

The lengthening of the cell cycle in CCNA2^fl/fl, Nestin-cre mice is perhaps not surprising, given the loss of an important cell cycle regulator. However, Cyclin E has been shown previously to compensate for loss of Cyclin A2 [17], leading us to ask if there were other underlying reasons for the lengthened cell cycle. One of the major causes of cell cycle arrest leading to a lengthened cell cycle is DNA damage. The function of Cyclin A2 with its effector kinase CDK2 is blocked during the DNA damage checkpoint, which stops progression through the cell cycle [20]. However, our results suggest that Cyclin A2 may have a more active role in the DNA damage response. Therefore, we asked if Cyclin A2 was involved in promoting DNA repair in addition to its well-characterized cell cycle roles. To answer these biochemical questions, we utilized an in vitro cell culture model. First, we tested the extent to which Cyclin A2 localized to sites of DNA damage after ionizing radiation (IR), which induces DNA double-strand breaks (DSBs). Cells were exposed to 2 Gy of X-ray radiation, and the localization of Cyclin A2 with the DNA double-strand break marker γH2AX was tested by proximity ligation assay (PLA). Epitopes located within 30 nm of each other elicit fluorescent punctae [21], which we quantified by automated counting in ImageJ. We observed an approximately three-fold increase in PLA signals in the nuclei during recovery from X-irradiation (Fig. 5A-C), which is indicative of localization of Cyclin A2 at double-strand breaks. We conclude that Cyclin A2 is found at DSB sites after ionizing radiation.

We next tested the hypothesis that CCNA2 played a role in the signal transduction cascade of the DNA damage response. One of the main hallmarks of the DNA signal transduction cascade is phosphorylation of H2AX [22]. In normal cells, IR treatment results in acutely high levels of phosphorylated H2AX (γH2AX); the levels of this phosphorylated histone returns to baseline following DNA repair. We therefore asked if the phosphorylation of H2AX was affected by loss of Cyclin A2. We used a previously published and validated lentiviral approach to silence CCNA2 expression (Fig. 5D) [23]. Comet assay found no change in DNA damage between CCNA2-silenced cells and control cells (Fig. 5E-F). We then analyzed levels of γH2AX in CCNA2-silenced cells. Quantification of γH2AX normally involves counting of nuclear foci. In our case, we observed such a large γH2AX response to irradiation that we were unable to count distinct foci; we thus analyzed the percent of the nucleus that was positive for γH2AX signal. CCNA2-silenced cells phosphorylated H2AX post-IR at lower levels than did control cells. Furthermore, CCNA2-silenced cells showed delayed DNA repair kinetics 1 and 4 hours post-IR, indicating that the damaged DNA was repaired at a reduced rate (Fig. 5G-I). Thus, reduced Cyclin A2 protein levels results in decreased H2AX phosphorylation with a functional, albeit slowed, DNA damage response.

With this in mind, we tested the hypothesis that CCNA2 was a mediator of DSB repair by performing DNA repair assays in CCNA2-silenced cells. Specifically, we utilized the DRGFP homologous recombination (HR) and EJ5GFP non-homologous end joining (NHEJ) assays [24,25], which required us to modify our CCNA2-silencing methodology (Fig. 6A-B). In these assays, DSB resolution by the pathway in question is determined by manifestation of GFP fluorescence detected by flow cytometry. CCNA2-silenced cells showed a mean reduction of 40% in HR and a mean reduction of 67% in NHEJ (Fig. 6C-D and Supplemental Fig. S2). These data are in line with reductions in GFP fluorescence observed by silencing other DNA repair genes such as RNF8 [26]. In toto, our finding showing blunted γH2AX formation and deficient HR and NHEJ DSB repair in CCNA2-silenced cells support the notion that CCNA2 functions early in the DSB signal transduction cascade. Such a finding would result in radiation sensitivity in CCNA2-silenced cells. Lentiviral CCNA2-silenced cells were thus subjected to the clonogenic assay [27] after exposure to 0, 3, 6, or 9 Gy of X-rays. Cells deficient in CCNA2 were significantly more sensitive than control cells (Fig. 6E). We conclude that DSB repair is deficient in CCNA2 silenced cells.

CCNA2-null neural progenitor cells of the VZ/SVZ show evidence of intrinsic DNA repair defects

We next asked if loss of Cyclin A2 had a similar effect in neural progenitors found in the VZ/SVZ forebrain stem cell niche. To achieve this, we evaluated E14.5 embryos and P0 pups in which CCNA2 was ablated in neural progenitors by intercrossing CCNA2^fl/fl mice with Nestin-cre mice. We examined the VZ/SVZ of these animals for hallmarks of DNA damage experienced during normal development. Such DNA damage causes cell cycle arrest, resulting in a slowed cell cycle. A common physiological cause of cell cycle arrest includes DSB formation caused by replication fork collapse or generation of reactive oxygen species (reviewed in [28,29]). With this in mind, we examined VZ/SVZ neural progenitor cells in vivo for increased Rad51, a homologous recombination protein elevated in cells undergoing HR [30]. We noted an increased proportion of VZ/SVZ neural progenitor cells expressed RAD51 in CCNA2^fl/fl, Nestin-cre animals at both E14.5 and P0, supporting the notion that these cells were undergoing DNA repair (Fig. 7A-B, E-F). This also supports the notion that loss of Cyclin A2 does not fully block DNA repair, but decreases the rate of repair as evidenced by the persistence of Rad51 expression in cells outside the VZ/SVZ. We then investigated if CCNA2^fl/fl, Nestin-cre neural progenitor cells showed altered DNA damage (Fig. 7C-D). We quantified γH2AX in the dorsal VZ/SVZ and observed a reduction in the mean number ofγH2AX foci, in concordance with our cell culture data (Fig. 5J-K). This is in contrast to brains deficient in NBS and ATM, two high-level regulators of the DNA damage response that showed an increase in γH2AX in the VZ/SVZ during embryonic development [31]. This further supports the importance of H2AX phosphory-lation on Cyclin A2 function. We conclude that CCNA2-null neural progenitor cells show evidence of active DSB repair in vivo.

CCNA2-null cells of the hippocampus show evidence of intrinsic DNA repair defects

The above data indicates that Cyclin A2 is important in the development of the VZ/SVZ that gives rise to the cerebral cortex. We next asked if Cyclin A2 was necessary for the development of other stem cell niches in the brain. We previously demonstrated that Cyclin A2 is critically important for the development of the cerebellum [3]. Similar to the above results in the VZ/SVZ, growth of the cerebellar EGL stem cell niche was delayed with an increase in apoptosis. However, in contrast to the VZ/SVZ, the cerebellum fails to form appropriately. These contrasting results led us to hypothesize that Cyclin A2 was necessary for the development and function during adulthood of another stem cell niche in the CNS, the dentate gyrus of the hippocampus. To test this hypothesis, we used different cre drivers to ablate Cyclin A2 expression in the brain at different times during development. We used Nestin-cre for embryonic ablation, and CamkIIα-cre for ablation in the adult after the brain was fully developed. The CamkIIα-cre mouse was first developed to drive cre expression in the CA1 layer of the hippocampus, although it displays widespread expression in the hippocampus and cerebral cortex [32]. We performed unbiased stereological volume measurements of the dentate gyrus as we performed in the VZ/SVZ. We found that embryonic ablation of Cyclin A2 led to a drastic reduction in size of the dentate gyrus immediately after birth (Fig. 8A). Unlike the VZ/SVZ, the size of the dentate gyrus did not recover. However, loss of Cyclin A2 after the brain was fully formed by interbreeding with CamkIIα-cre mice had no effect on the size of the dentate gyrus at 4 or 8 months after birth (Fig. 8A).

We next asked if, similar to the VZ/SVZ, loss of Cyclin A2 led to decreased γH2AX in the hippocampus. We examined both the dentate gyrus and the CA1 layer of the hippocampus, the original target for cre expression in the CamkIIα-cre mouse. We found no significant change in γH2AX in the dentate gyrus in 4 month old animals (Fig. 8B-C). However, we did observe a decrease in γH2AX foci in the CA1 layer in 4 month old animals, and observed a time-dependent decrease in γH2AX in controls in both hippocampal areas in the controls. This time-dependent decrease in γH2AX was much less pronounced in experimental animals (Fig. 8D-E).

Next, we asked if loss of Cyclin A2 in the developed brain led to appreciable behavioral changes. Based on our γH2AX results in the hippocampus, we hypothesized that these mice would have defects in learning and memory. We therefore subjected 8 month old animals to a battery of behavioral tests. Mice without Cyclin A2 performed significantly worse in the Barnes Maze [33], which tests spatial learning and memory, and fear conditioning [34], which tests contextual memory (Fig. 9A-D). We additionally observed a trend toward decreased performance in the passive avoidance test, although it did not meet our threshold for statistical significance. We did not observe differences in other behavioral tests including anxiety-like responses (open field and elevated plus maze tests), depressive-like responses (forced swim and tail suspension tests), sociability (social preference test), or in marble burying behavior, although there was a small decrease in pain sensitivity and a small increase in Rotarod performance (Supplemental Fig. S3). These results underscore the differential requirement of Cyclin A2 in the development of different structures of the brain. Furthermore, they support a novel role for Cyclin A2 in DNA repair in post-mitotic neurons, which is unexpected given its major role as an S-phase regulator in cycling cells.

Non-canonical functions of Cyclin A2

DSBs comprise lethal forms of DNA damage, with even a single DSB being capable of killing a cell or disturbing its genomic integrity [35]. The DNA damage response has been well-characterized [36], and the importance of several of these genes in mouse neural development has been described [31,37–40]. The commonality of these mouse models lies in their hypomorphic cerebellum, which lacks its characteristic lamination and foliation. We previously reported the effect of Cyclin A2 deletion on cerebellar development [3]. Mice without Cyclin A2 during embryonic neural development displays striking similarities to the aforementioned DNA repair mouse mutants, with the same defects in cerebellar morphology. However, we observe a decrease in H2AX phosphorylation, while mutants of many of the other DNA repair genes accumulate γH2AX foci, lending further credence to the hypothesis that Cyclin A2 is a high-level regulator of the DNA damage response.

The most commonly employed DNA repair mechanisms for DSB include homologous re-combination (HR) and non-homologous end joining (NHEJ), and these mechanisms show predilection for distinct cell cycle phases [41]. Specifically, according to this notion NHEJ would be most robust for repair of DNA damage incurred during G1, while HR would be more common for repair during S and G2 [42]. Cyclins and CDKs have been implicated in DSB repair, including Cyclin D1, Cyclin A1, CDK1, and CDK2 [43–45]. Cyclin A2 has previously been shown to interact with both BRCA1 [46] and BRCA2 [47], although to our knowledge we are the first to report that Cyclin A2 is found at the site of DNA damage. It is unexpected however that although it is an S-phase cyclin, Cyclin A2 regulates both HR and NHEJ. This supports our notion that Cyclin A2 is a regulator of DNA damage in post-mitotic neurons, and provides an explanation for our previous finding that Cyclin A2 is expressed in the adult brain [3].

We additionally found that H2AX phosphorylation is inhibited by CCNA2 loss. These data would indicate that Cyclin A2 lies very early in the DNA repair signal transduction cascade. Further supporting this conclusion are findings that show minimal reduction in HR efficiency by upstream effectors and near complete blockage of HR by downstream effectors of HR. For example, silencing of ATM, one of the first genes involved in the DNA damage response results in no significant difference in HR efficiency [48]. In contrast, silencing Rad51 expression results in near complete blockage of HR [43]. These results are more suggestive of a role for Cyclin A2 in the regulation of the DNA damage response rather than in the actual resolution of the DNA double-strand break. Such a notion is in line with known Cyclin A2 functions as a DNA replication fork initiator during S-phase. In S-phase, the chromatin landscape undergoes dramatic alterations as the entire genome is duplicated. Elegantly reviewed by Alabert and Groth, accurate cell cycle progression requires replication initiation at 30,000 to 50,000 origins; no clear DNA consensus sequence for replication origin exists, and it is thought that chromatin structure is a major recognizing factor for replication fork initiation [49]. CCNA2’s role as an initiator of replication origin firing [50] thus insinuates a potential role as a chromatin sensor. In summary, our data allows us to conclude that Cyclin A2’s well established role as a chromatin sensor may include an ability for this protein to sense DNA double strand breaks.

Developmental and aging roles for Cyclin A2

Using CCNA2 ablation as a model, we were able to show cell cycle defects lead to a shift in the quantities of cells produced by the neural stem cell niche. Stated otherwise, the forebrain stem cell niche adapts to cell cycle dysfunction by inducing a developmental delay,which allows compensatory accelerated growth mediated by progenitor cells sensing their population size relative to V_max. This delay is well explained by a standard logistic growth mechanism, whereby the rate of population growth depends on the size of the population. This compensation is done at the expense of emerging from the embryonic environment with a decreased computational capacity since the number of neurons generated is lower. The insights obtained from this study carry significant implications for pediatric neurology. Two of the most common environmental fetal genome stressors include alcohol exposure and gestational diabetes. Fetal alcohol exposure induces DNA damage due to aldehyde-mediated DNA cross-linking [51]. Diabetes increases DNA damage, presumably through elevation in reactive oxygen species, sensitizes DNA to damage, and delays DNA repair kinetics [52]. Although the specific chemical changes to DNA mediated by aldehydes and hyperglycemia differ, many types of DNA damaging agents ultimately converge, directly or indirectly, to DNA double-strand break repair [53]. Of particular relevance to our study is the known association of children of gestational diabetic mothers to manifest with subtle developmental delays [54,55]. The implications of our work provide an underlying mechanism as to how fetal stressors, even if transient, can result in long-term developmental delays that manifest months after exposure.

In addition to the role for Cyclin A2 in the repair of DNA damage encountered during neurogenesis, we have further uncovered a role for Cyclin A2 in DNA repair of adult hippocampal neurons. This finding underscores the critical importance of DNA repair in all aspects of life. Indeed, DNA repair malfunction is linked with diseases of the nervous system that are primarily seen early in life such as ataxia-telangiectasia which is caused by mutations in the DNA damage response kinase ATM [56]. At the other end of life, many neurodegenerative diseases have been linked with DNA damage. For example, familial amyotrophic lateral sclerosis and frontal temporal lobar degeneration have been linked to mutations in the FUS gene which lead to increased DNA damage in neurons [57,58]. Furthermore, recent evidence has suggested that accumulation of Aβ plaques reduces levels of the well-known DNA repair protein BRCA1 in neurons, leading to the accumulation of DNA damage [59]. Our work demonstrates the differential outcomes resulting from loss of DNA repair function during both development and aging. Supporting this notion are studies on the tumor suppressor BRCA1 that demonstrated that ablation of BRCA1 by Nestin-cre lead to severe brain growth defects [60]. Additionally, silencing of BRCA1 in the adult dentate gyrus leads to a similar behavioral phenotype as we observed in our adult-ablated CCNA2 mice [59]. These studies underscore the importance of the DNA damage response in embryonic neural progenitors as well as in adult neurons.

That Cyclin A2 possesses conserved roles in development and adulthood is at first counter-intuitive. Although the adult brain shows less proliferation than the embryonic brain, the adult brain requires continued cellular turnover in the subgranular zone of hippocampus and the subventricular zone. Furthermore, endogenous DNA damage is an expected finding in both cell types. For example, highly proliferative neural precursor cells in the VZ/SVZ would undergo DNA replication stress, whereas reactive oxygen species-mediated DNA damage represents a well-characterized finding in adult neurons. Nevertheless, we were surprised to find similar phenotypes in DNA damage detection (i.e., a diminished level of endogenous γH2AX in proliferative neural precursor cells and in post-mitotic neurons. We propose that Cyclin A2’s function in both homologous recombination and non-homologous end joining-mediated repair permits its role in these two time points. Indeed, the notion of proteins playing conserved roles in both embryonic development and aging has been well-established, and was recently elegantly reviewed by Kovacs, et al. [61]. We submit that through Cyclin A2’s non-canonical roles in DNA repair detection, that it has joined the expanding list of genes showing crucial roles in both embryonic development and in the aging brain.

Animal husbandry and tissue processing

All work involving animals was performed under the auspices of a protocol approved by the Ohio State University Institutional Animal Care and Use Committee. CCNA2^fl/fl and Nestin-cre transgenic mice were bred as described previously for embryonic ablation of CCNA2 expression [3]. Briefly, timed-pregnant females were euthanized at embryonic days 14.5 or 17.5, and embryos were drop-fixed in 4% PFA overnight, cryoprotected in 30% sucrose, and snap frozen in OCT. For cell cycle length analyses, an E14.5 pregnant dam was injected with CldU at 50 mg/kg. Embryos were harvested 2.5 hours after CldU administration and drop fixed in 4% PFA. Post-natal day 0 pups were anesthetized with ketamine (30 mg/mL) and xylazine (2 mg/mL) and perfused transcardially with PBS and 4% PFA. P0 pups were pulsed with BrdU for 30 minutes before perfusion as necessary for each experiment. Brains were post-fixed overnight in 4% PFA, cryoprotected in 30% sucrose, and snap frozen in OCT. Cryosections of OCT-embedded brains were taken at 5, 12, or 50 μm and stained with pertinent antibodies. Control mice are littermates of CCNA2^fl/fl, Nestin-cre mice with preserved CCNA2.

CCNA2^fl/fl and CamkIIα-cre transgenic mice were intercrossed for adult ablation of CCNA2 in the hippocampus and cerebral cortex. 4 month old mice were perfused as above. Brains were bisected sagittally and processed for unbiased stereology cresyl violet staining or OCT embedding. 8 month old mice were subjected to a battery of behavioral tests, then sacrificed and brains processed. Control mice were littermates of CCNA2^fl/fl, CamkIIα-cre mice with preserved CCNA2.

Unbiased stereology

Derivation of mathematical model

Unbiased stereological* values were used to guide the development of a mathematical model of forebrain development [63]. While a natural description of proliferative populations would be through continuous time stochastic processes such as branching processes or birth-death processes [64–67], our main interest is in the situation where population size is already exponentially high, making any stochastic effects mathematically insignificant. For this reason, we formulated our model through deterministic ordinary differential equations*. Note however that these equations can be interpreted as a mean-field approximation of an associated stochastic branching process.

Radial glia cells in embryonic mice are known to undergo three types of division [1,68]. The first is proliferation, a type of symmetric self-renewal, where two radial glia daughter cells are produced. The second is an asymmetric self-renewal where a migratory neuron is produced and a radial glial cell is renewed. The third is a different type of asymmetric division where a neuron is produced along with an intermediate progenitor cell in the SVZ, and the radial glia does not undergo self-renewal. The intermediate progenitor undergoes another division to yield two migratory neurons. We incorporated all of these types of division into a mathematical model. The number of cells present in each of the stages is given by a system of ordinary differential equations presented in Supplemental Tables S2 and S3.

The terms denote rates of exit from each of the states and are denoted graphically in Fig. 2C and in Supplemental Table S4. These flows can be understood in terms of cycle transition times on the cell-by-cell basis. They are parameterized* by transition rates, which are related to the mean times taken for cells to progress between the different stages in cell cycle. In turn, the timing is determined by the detailed biochemistry of the cells under influence from both intracellular and extracellular environments. While the precise biochemistry with respect to changes in Cyclin A2 expression is not known with certainty, we were still able to investigate the effects of Cyclin A2 on development by examining how the growth of the VZ/SVZ is perturbed. We performed this task by first parameterizing our model using experimental observations found in the literature. Prior researchers have reported the timing of the various cell-cycle transitions in wild-type mice in the embryonic ages. We used this timing information to deduce equivalent rates for the transitions (shown in Supplemental Tables S4 and S5).

The transitions in this model generally follow first-order kinetics. The exceptions are the transition rates out of G1 for the radial glia cells. G1 phase is known to lengthen in the course of the development of the VZ/SVZ, concomitant with a shift towards the production of differentiated cells and away from proliferative precursors [15]. We take this fact to imply that the effective average G1 cycle length can be given as




where is the G1 cycle length for asymmetric division into differentiated cells, is the G1 cycle length for proliferation of the neural stem cells, and is the G1 cycle length for asymmetric divisions of the radial glia that are associated with self-renewal. The terms describe the relative proportion of each of the different types of mitosis. As developmental age increases, and the number or density of cells increases, there is a transition from proliferative-type divisions to renewal to differentiation (neurogenesis). The biological basis for these transitions from proliferation to self-renewal and neurogenesis is complicated, involving both intrinsic and extrinsic signaling factors, which change temporally. These factors include the developmental age of the cells, presence and activation of cell-surface signaling proteins such as αE-catenin [69], Notch [70], and Jagged [71], as well as diffusing growth factors like HH amongst others.

In order to determine when differentiation occurs, we utilized the data of Takahashi et al. [15], where the authors reported the fraction of cells exiting cell cycle into neurogenesis as a function of developmental age (e.g., a cell of age one divides into two cells of age two). We interpreted these data as a survival probability for the mitotic capability of the progenitors, where we assume that progenitors exit cell cycle by asymmetrically differentiating into an intermediate progenitor and a neuron. In our model we assume that radial glia differentiate when they reach a preset number of divisions . In our model, wildtype radial glia have K=5 available divisions at E13.5, which corresponds to approximately 2 divisions at E14.5.

We assume that a separate biochemical switch that depends on interactions with other radial glia determines the proportion of proliferative cells versus self-renewing cells. We model this switch by assuming that can be described as a function of the population size. We suspect on the basis of our data that this is so, as growth is slower in the wildtype VZ/SVZ in the days between E14.5 and E17.5 compared to in the mutant, which trails the wildtype in VZ/SVZ volume as well as CP volume. We postulate that this difference in growth rates can be due to the decreased inhibition of mutant growth, which can be explained through a logistic growth mechanism. The logistic growth model for population dynamics assumes that the overall growth rate is proportional to the size of the population and the amount of available resources (in our case, space, as well as growth factors) [72]. It has also previously been noted that the brain appears to develop according to a logistic growth curve [73]. We model logistic growth by setting p using the expression

where the quantity is the effective carrying capacity of the VZ. The model is assumed to start at day E13.5 with initial conditions given to yield the total cumulative volume observed in the data at E14.5. We assumed that initially all cells in the simulation are of the same mitotic age, and that the proportion of cells in each of the phases of cell cycle is weighted by the time spent in each phase.

The growth of the cortical plate depends heavily on the migration parameter contained in the migration flow . The latency involved in migration of the differentiated neuron to the cortical plate depends on the changing dimensions of the developing cortex. As the distance between the SVZ and the CP increases, it is expected that neurons take longer to reach the CP. Understanding the kinetics of this process, while interesting in itself, is outside of the scope of the questions that we hope to answer with this model. For this reason, we report the cumulative output of differentiated neurons from the VZ/SVZ rather than the volume of the cortical plate. Yet, we retain this migration in the model to emphasize that the dynamics of the cortical plate volume can be computed provided one has a description of the migration rate.

Cell culture and viral infections

All cell culture was performed in a 37°C humidified incubator set at 5% CO₂ in room air. U251 glioblastoma cells were cultured in DMEM/F12 supplemented with non-essential amino acids and glutamine with 10% fetal bovine serum. Previously published and experimentally-validated CCNA2-targeting or scrambled non-targeting shRNA lentiviral constructs were a kind gift of Dr. Benedicte Lemmers and Dr. Jean-Marie Blanchard and were produced as described [23]. U251 cells were transduced with lentivirus encoding either CCNA2-targeting or scrambled shRNA. Transduced cells were selected by culturing in medium containing 1.0 μg/mL puromycin, and maintained in puromycin medium for all experiments. Cells were irradiated with 160 kV X-rays using a RadSource RS2000 Biological Irradiator at the appropriate doses for each experiment. Cells were fixed with 4% PFA before immunocytochemical processing.

To generate CCNA2^fl/fl neurospheres, CCNA2^fl/fl males were intercrossed with CCNA2^fl/fl females to generate homozygous pups. Postnatal neural progenitor cultures were generated by dissecting the VZ/SVZ from brain, and cultured in suspension with DMEM/F12 media supplemented by bFGF/EGF, N2, B27, glutamax, penicillin/streptomycin, and heparin. For the proliferation assays, neurospheres were dissociated with trypsin to single cells and plated onto PDL/laminin coated chamber slides followed by infection with 1 μl adenovirus (adenovirus-GFP for control, adenovirus-cre-GFP as the experimental group, Vector Biolabs). Cre-mediated excision was confirmed by amplification of genomic DNA with the following primers: LoxP1 #1a: 5’-CGCAGCAGAAGCTCAAGACTCGAC-3’; LoxP1 #2: 5’-TCTACATCCTAATGCAATGCCTGG-3’; LoxP2 4a: 5’-TGTACAGCATGGACTCCGAGCGAC-3’. Cells were pulsed with BrdU for 30 minutes, fixed, and immunostained for BrdU and pH3.

Proximity ligation assay

Fixed cells were per-meabilized for 5 minutes with PBS containing 0.1% Triton X-100, then blocked with blocking solution (PBS with 0.1% Triton X-100 and 5% normal goat serum) for 30 minutes. Cells were incubated overnight at 4°C with primary antibodies against Cyclin A2 and γH2AX diluted in blocking solution. Control cells were single-labeled with either Cyclin A2 or γH2AX antibody. Coverslips were then washed 3x5 minutes with 3 mL PBS. PLA was performed according to manufacturer’s instructions (O-link Biosciences). Maximum intensity projection images from .czi files were thresholded and number of foci in each nucleus was automatically counted using ImageJ.

Comet assay

Cells were irradiated at 2 Gy. Control cells were not irradiated. Cells were dissociated with accutase, quenched with cold medium, and harvested by centrifugation. Cell pellets were washed with ice-cold PBS, and prepared for comet assays. Cells were mixed with 1% low-melting point (LMP) agarose, and 100 μL of cells in agarose was pipetted onto an agarose-coated slide and covered with a coverglass. Slides were chilled on an ice-cold aluminum plate for 10 minutes and coverglass was removed. The cell-agarose mixture was covered with 1% LMP agarose and covered with a coverglass. Slides were chilled as before. Coverslips were removed, and slides were processed as described previously [74]. Slides were stained with ethidium bromide, visualized at 20x magnification, and scored using OpenComet software [75].

Homologous recombination and non-homologous end joining assays

Plasmids pDRGFP (Addgene #26475) and pCBASceI (Addgene #26477) were obtained from Maria Jasin via the Addgene repository [24]. PimEJ5GFP (Addgene #44026) was obtained from Jeremy Stark via the Addgene repository [25]. U251 cells were transfected with pDRGFP or pimEJ5GFP, and cells with the integrated plasmids were selected with medium containing puromycin at a concentration of 1.0 μg/mL and expanded for subsequent experiments. To measure levels of homologous recombination or non-homologous end joining, cells were plated in two 6-well plates at a concentration of 150,000 cells/well. Cells were transfected overnight with either CCNA2 siRNA (Invitrogen siRNA #4390824) or control siRNA #1 (Invitrogen catalog #4390843) at a concentration of 10 nM with JetPrime (Polyplus Transfection) following the manufacturer’s protocol. Following transfection, cells were re-plated at a density of 150,000 cells/well in 6-well plates and incubated overnight. Cells were then transfected overnight with CCNA2 or control siRNA together with 2 μg pCBASceI plasmid to induce a double-strand break at the DRGFP or EJ5GFP locus. Parallel controls were performed with pCAGGS for gating of cells for flow cytometry. DRGFP cells were re-fed with fresh puromycin medium every other day for 6 days before dissociation with accutase. EJ5GFP cells were refed with fresh medium without puromycin every other day for 5 days before dissociation with accutase. GFP+ cells were quantified using a Becton-Dickinson FACSCalibur flow cytometer. The threshold for GFP-negative cells was set using pCAGGS-transfected cells, and cells with fluorescence above this threshold were counted as GFP-positive. Gating strategy is shown in Supplemental Fig. S2. 20,000 cells were analyzed, and percentage of GFP-positive cells was determined relative to the total population. Relative levels of HR or NHEJ were determined by normalizing the percentage of GFP-positive cells in each condition to the percentage of GFP-positive cells in control conditions transfected with control siRNA and pCBASceI plasmid.

Clonogenic assay

U251 cells infected with lentivirus expressing shRNA targeting CCNA2 or non-targeting scrambled shRNA were seeded at clonal densities in 6 well plates and allowed to adhere for 1 hour at 37°C. Cells were then irradiated at 0, 3, 6, or 9 Gy at room temperature, and allowed to recover and grow for approximately 1 week in the cell culture incubator. Cells were re-fed with fresh medium every other day. After growth, cells were fixed for 15 minutes in 4% PFA, stained for 30 minutes with 0.05% crystal violet, washed with water, and dried. Colonies were manually counted on a Zeiss Discovery V.8 Stereomicroscope, and surviving fraction (SF) determined for each dose using the following formula: SF=#cells counted/(#cells plated * plating efficiency).

Western blotting

Cells were lysed in lysis buffer containing 50 mM Tris-HCl, pH 6.8, 1% SDS, 1 mM EDTA, 5% glycerol, and 1x Halt protease/phosphatase inhibitor (Pierce). Total cell lysates from U251 CCNA2-knockdown or scrambled cells, or U251 CCNA2 or control siRNA-transfected cells were resolved on an SDS-PAGE gel and transferred to a nitrocellulose membrane. For determination of CCNA2 silencing, membranes were probed with antibodies against Cyclin A2 and GAPDH as a loading control. Blots were visualized using Cy3 or Cy5 ECL-Plex secondary antibodies with a Typhoon scanner or with HRP-conjugated secondary antibodies with a Bio-Rad Chemidoc XRS. Quantification of Western blot images was performed by densitometric analysis using ImageJ.

Antibodies and histology

Antibodies used in this study for immunohistochemistry (IHC) and western blotting (WB) were as follows: BrdU, GeneTex #gtx26326, IHC 1:600 Cleaved caspase 3, Cell Signaling Technologies #9664S, IHC 1:100 dilution; CldU, Novus Biologicals # NB500-169 IHC 1:600 dilution; Cyclin A2, Santa Cruz Biotechnology #sc-596 IHC 1:100, WB 1:1000; GAPDH, Millipore #MAB374, WB 1:1000; γH2AX, Millipore #05-636, IHC 1:200; γH2AX, Cell Signaling Technologies #9718, IHC 1:400; NeuN, Millipore #MAB377, IHC 1:1000; Pax6, Santa Cruz Biotechnology #sc-7750 IHC 1:100 or DSHB #PAX6 IHC 1:5; pH3, Cell Signaling Technologies #2650, IHC 1:100; Rad51, Calbiochem #PC130, 1:200 for IHC on tissue, 1:300 for cell culture; Tbr2 was a kind gift from Dr. Robert Hevner and was used at 1:500 dilution for IHC. Alexa Fluor-conjugated secondary antibodies were from Invitrogen and were used at a 1:1000 dilution.

For unbiased stereology analyses of E14.5 and E17.5 embryos, 50 μm cryosections were labeled with antibodies raised against cleaved caspase 3. For unbiased stereology analyses of P0 brains, 50 μm cryosections were labeled with antibodies raised against NeuN. Stains were developed with the Dako EnVision HRP kit. Nuclei were counterstained with hematoxylin and tissue sections were dehydrated for 2 minutes each in 10%, 20%, 40%, 80%, 95%, and twice in 100% dehydrant. Sections were incubated for 1 minute each in sequential xylene baths, and coverslips were mounted with Permount.

For immunofluorescent analysis, tissue sections were subjected to antigen retrieval with 10 mM citrate buffer pH 6.0 in a pressure cooker for 4 minutes or Pelco BioWave at 95°C for 15 minutes, blocked with blocking solution, and labeled with the appropriate primary antibodies for each experiment. Sections were washed with PBS and incubated with the appropriate secondary antibodies and counterstained with DAPI. Sections were washed and coverslips mounted with Prolong Gold (Invitrogen). Cultured cells were processed in a similar fashion without antigen retrieval.

Microscopy and imaging

Fluorescent images were captured with a Carl Zeiss Axio Imager Z.1 with LSM700 confocal laser scanning microscope with 10x, 20x, or 63x objectives as necessary, or with a Nikon 80i epifluorescent microscope with Hamatasu Orca-R2 camera. For analysis of total Rad51-positive cells, single optical slice images or Z-stacks were captured at 200x magnification. For analysis of γH2AX in tissue sections, Z-stacks with step size 0.38 μm encompassing the entire thickness of the histological preparation were acquired at 630x magnification. For analysis of γH2AX or PLA foci in cultured cells, Z-stacks encompassing the entire thickness of the nucleus were acquired. All Z-stack analyses for cultured cells used a 63x objective with a pinhole of 1 Airy unit, scan speed of 6, averaging of 1, and resolution of 512x512. Maximal projections of Z-stacks were analyzed in ImageJ to count foci or proportion of the nucleus positive for each marker. For analysis of Cyclin A2, Pax6, Tbr2, and CldU staining, images of the dorsal cortex were captured at 100x or 200x magnification. 100x images were assembled as a montage and placed on a black background for visual clarity. 200x images were captured as Z-stacks in .czi files and converted to .tif images. Images were counted in a standard counting frame (100 μm medial-lateral, 180 μm dorsal-ventral starting at the boundary of the lateral ventricle) on the dorsal cortex and when necessary divided into 100 x 10 μm bins as described previously [76]. Counts for Tbr2 utilized total counts within the entire counting frame. To account for the decreased size of the brain, Tbr2 counts were multiplied by the size of the VZ/SVZ obtained by unbiased stereology measurements. Counts within the first bin along the ventricle wall were used for CldU quantification. To overcome the limitations of quantifications of Rad51 foci in tissue sections, we instead performed fluorescent intensity measurements in E14.5 animals or counted proportion of Rad51-positive cells in P0 animals. To achieve unbiased quantifications of γH2AX in neuroanatomical structures, we used a similar methodology to count foci in counting frames rather than in individual cells. Quantification of γH2AX in E14.5 embryos was performed using 630x magnification images with a standard counting frame of 50 μm medial-lateral x 90 μm dorsal-ventral to include the VZ/SVZ. Quantification of γH2AX in 4 month old and 8 month old brains was performed using 630x magnification images with a standard counting frame of 50 μm medial-lateral x 40 μm dorsal-ventral to include the dentate gyrus or CA1 layer. For γH2AX counting, nuclear signals smaller than 0.8 μm diameter were counted [77]. We have extensive experience quantifying γH2AX in the brain, and note that there is background signal that is generally extra-nuclear. To avoid parameterizing the data, we chose to count events within a nucleus in a defined area to maintain a normal distribution. All images were acquired as greyscale images and presented as pseudocolored images.

Behavioral testing

All behavioral testing was performed by the Ohio State University Rodent Behavioral Core Facility. CCNA2^fl/fl, CamkIIα-cre mice and control littermates with preserved CCNA2 were subjected to the following battery of behavioral tests. Twelve controls and seven CCNA2^fl/fl, CamkIIα-cre mice of mixed sex were used.

Statistical analysis

Descriptive and inferential statistics were performed in Microsoft Excel or R v.3.0.1. Student’s t-test for pairwise comparison or ANOVA with Tukey’s HSD correction for multiple comparisons were used as appropriate for each experiment as indicated in figure legends.

Ethics Approval

All work involving animals was performed under the auspices of a protocol approved by the Ohio State University Institutional Animal Care and Use Committee. Every effort was taken to minimize suffering of animals in this study.