Heterochromatic genome instability and neurodegeneration sharing similarities with Alzheimer’s disease in old Bmi1+/− mice

Sporadic Alzheimer’s disease (AD) is the most common cause of dementia. However, representative experimental models of AD have remained difficult to produce because of the disease’s uncertain origin. The Polycomb group protein BMI1 regulates chromatin compaction and gene silencing. BMI1 expression is abundant in adult brain neurons but down-regulated in AD brains. We show here that mice lacking one allele of Bmi1 (Bmi1+/−) develop normally but present with age cognitive deficits and neurodegeneration sharing similarities with AD. Bmi1+/− mice also transgenic for the amyloid beta precursor protein died prematurely and present aggravated disease. Loss of heterochromatin and DNA damage response (DDR) at repetitive DNA sequences were predominant in Bmi1+/− mouse neurons and inhibition of the DDR mitigated the amyloid and Tau phenotype. Heterochromatin anomalies and DDR at repetitive DNA sequences were also found in AD brains. Aging Bmi1+/− mice may thus represent an interesting model to identify and study novel pathogenic mechanisms related to AD.


Murine neuronal cultures.
After dissociation of E18.5 cortices, cells were plated at 1.5 × 10 5 cells/well on poly-L-lysine-coated 8-well cultures slides (BD Biosciences). Cells were maintained in normal medium composed of Neurobasal-A Medium (Invitrogen), Glutamax-I (Gibco), gentamycin (50 µg/ml; Gibco), B27 supplement (Gibco), nerve growth factor (50 ng/ml; Invitrogen) and BDNF (0,5 ng/ml; Invitrogen). For the assays using inhibitors, we added 2 μM of DMSO, 2 μM of ATM/ATR inhibitor (CGK733; Millipore) or 10 μM of ATM inhibitor (KU55933; KuDOS Pharmaceuticals) to the cell culture medium for 16 hours. NAC was added daily at 5 mM into the culture medium for 7 days 18 . Amyloid fractions and Immunoprecipitation. Cortices were minced and homogenized in a 10 ml TBS 1 × (50 mM Tris-Cl, pH 7.5. 150 mM NaCl)/protease inhibitor cocktail solution (Roche Diagnostics), using a tissue grinder. The homogenates were sonicated and ultra-centrifuged at 175,000 g for 1 h at 4 °C. The supernatants were collected as the soluble cellular fraction 24 . The dense pellet was re-homogenized in a 10 ml TBS 1×/1% Triton X-100/protease inhibitor cocktail solution, followed by sonication and centrifugation at 100,000 g for 1 h at 4 °C. The supernatants were recuperated as the insoluble cellular fraction 24 . For immunoprecipitation, protein samples and cellular fractions were incubated overnight with the primary antibody followed by incubation with the protein A/G beads (Millipore) for 2 hours the second day. Beads with bound immunocomplexes were washed with RIPA buffer (50 mM Tris-HCl, pH 7.5; 150 mM NaCl; 0.1%Tween 20; protease inhibitors Complete) and bound proteins were subsequently heat eluted with 1x Laemmli buffer.
Western blot. Cortices were minced and homogenized in 5 ml of complete mini protease inhibitor cocktail solution (100 mM Tris-HCl, pH7.5; 150 mM NaCl; 0.1% Tween 20; 2% SDS; protease inhibitors Complete; Roche Diagnostics) using a tissue grinder. The homogenates were sonicated and quantified using the Bradford reagent. For Western Blot, proteins were resolved by SDS-PAGE electrophoresis and transferred to a nitrocellulose blotting membrane (Bio-Rad). Membranes were incubated with the primary antibodies, treated with the corresponding horseradish peroxidase-conjugated secondary antibodies (Sigma), and developed using the Immobilon Western (Millipore).
Immunohistochemistry. Tissues were immersed in Formalin or 4% paraformaldehyde/3% sucrose in 0.1 M phosphate buffer, pH 7.4 for 16 h at room temperature on a rocking platform. Samples were washed three times in PBS and embedded in paraffin according to standard protocols. 8 µm thick sections were mounted on Super-Frost glass slides (Fisher Scientific) and processed for immunohistochemistry staining. Paraffin-embedded slices were analyzed by using the Vectastain ® ABC kit (Vector) according to the manufacturer instructions. Peroxidase substrates used are the Vector ® VIP (Violet) (Vector), and DAB (brown) (Sigma) 18 . Observations were made using the Zeiss imager Z2 microscope and images were captured with a digital camera. To obtain comparable results, we selected pyramidal neurons from the frontal cortex to quantify the number of chromocenter/neuron. Pyramidal Scientific REPORTS | (2019) 9:594 | DOI: 10.1038/s41598-018-37444-3 neuron identity was determined by the expression of NeuN, the cell morphology (using phase contrast imaging), the large nuclear diameter (when compared to astrocytes), and the presence of chromocenters and of a large nucleolus.
RT and real-time PCR. All primers were designed to flank individual exons and tested by PCR in RT+ and RT− control extracts. Total RNA was isolated using TRIzol reagent (Invitrogen). The RNA was treated with DNaseI for 20 min prior to reverse transcription (RT) and PCR for satellite repeats, which are intron-less. Reverse transcription was performed using 1 µg of total RNA and the MML-V reverse transcriptase (Invitrogen). Real-time PCR was performed using the Platinum SYBRGreen SuperMix (Invitrogen) and a Real-Time PCR apparatus (BioRad). Primers sets used were as follow: p16Ink4a (F) 5_-CAACGCCCCGAACTCTTTC-3_; (R) 5_-GCAGAAGAGCTGCTACGTGAAC-3_, p19Arf (F) 5_-GGCTAGAGAGGATCTTGAGAAGAGG-3_; (R) 5_-GCCCATCATCATCACCTGGTCCAGG-3_. Primers sets used for Bmi1, IAP, Maj Sat, Min Sat, Line, Sine and IAP were as described 19 .
Chromatin Immunoprecipitaion (ChIP) assay. ChIP was performed using the ChIP Assay kit (Upstate).
Briefly, 50 mg of cortical tissue was frozen for 1 hour and then homogenized at RT according to the manufacturer's protocol. The tissue was sonicated on ice for 10 sec at 30% amplitude to shear the chromatin (Branson Digital Sonifier 450, Crystal Electronics, On. Canada). Sonicated materials were immunoprecipitated using designated antibodies.  5′ -T TGGAAAC GGGAT T TGTAGA-3′ , 5′ -C GGT T TC CAACATATGTGT T T T-3′ ;  LINE 5′ -TGGCT TGTGCTGTAAGATCG-3′ , 5′ -TCTGT TGGTGGTCT T T T TGTC-3′ ; SINE1 5 ′ -G AG C AC AC C C AT G C AC ATAC -3 ′ , 5 ′ -A A AG G C AT G C AC C T C TAC C AC C -3 ′ ; Ac t i n 5 ′ -T C G ATAT C C AC G T G AC AT C C A-3 ′ ; 5 ′ -G C AG C AT T T T T T TAC C C C C T C -3 ′ ; Hox A 7 5 ′ -G T G G G C A A AG AG T G G AT T T C -3 ′ ; 5 ′ -C C C C G AC A AC C T C ATAC C TA-3 ′ ; β -m a j o r 5′-CAGTGAGTGGCACAGCATCC-3′; 5′-CAGTCAGGTGCACCATGATGT-3′. ChIP-qPCR data was analyzed according to the Percent Input method 19 . First, the raw Ct of the diluted 1% input fraction is adjusted by subtracting 6.64 cycles (i.e. log2 of the dilution factor 100). Subsequently, the percent input of each IP fraction is calculated according to this equation: 100*2 (Adjusted Input Ct-Ct(IP) .
Behavioral tests. Behavioral  Slices were recovered at 32 °C for one hour and subsequently maintained at room temperature. Carbogenated ACSF containing bicuculline methbromide (5 µM) to block GABA A receptor-mediated inhibitory synaptic current was used to perfuse slices in all recordings. Postsynaptic responses, which were evoked by stimulating the Schaffer collateral-commissural pathway via constant current pulses (0.08 ms) delivered through a tungsten bipolar electrode (FHC), were recorded from the hippocampal CA1 region, amplified by a Multiclamp 700B (Axon) and stored in a PC for offline analysis using Clampfit (Axon). All recordings were performed at room temperature. Field excitatory postsynaptic potential (fEPSP) was evoked at 0.05 Hz and detected by an ACSF-filled glass electrode that was placed in the stratum radiatum of the hippocampal CA1 region. Long-term potentiation (LTP) of fEPSP was induced by a high-frequency tetanus (100 Hz, 100 pulses). Student's t-test was used to compare changes in fEPSP slope at 60 minutes after tetanus between WT and Bmi1 +/− groups. In order to determine if fEPSPs are potentiated by tetanus within each mouse group, paired Student's t-test was used to compare fEPSP slope recorded before (baseline) and at 60 minutes after tetanus.
Statistical analysis. The Student's t-test for unpaired samples was used to measure statistical differences. An analysis of variance (ANOVA) followed by the Dunnett test was used for multiple comparisons with one control group. In each figures, the criterion for significance (P value) was mentioned.

Results
Bmi1 +/− mice develop accelerated aging features and neurological symptoms. We previously reported that Bmi1 +/− mice have reduced median and maximal lifespan along with apparition of lens cataracts at higher frequencies than WT littermates, suggesting that Bmi1 hemi-deficiency may accelerate the aging process 18 . Alopecia and weight loss are common characteristics of premature aging in mice [26][27][28][29][30] . We observed that between the ages of 15-24 months (defined here as old mice), Bmi1 +/− mice exhibited alopecia and reduced body size (~25%) when compared with age-match WT littermates ( Fig. 1a,b). These pathological features were not present in 3-month and 1-year old Bmi1 +/− mice ( Fig. 1a,b). We performed the paw-clasping test to verify if old Bmi1 +/− mice presented pathological reflexes. Adult WT mice picked by the tail and slowly inclined towards a horizontal surface spread all four limbs in anticipation of contact. However, instead of limb extension, old Bmi1 +/− mice displayed an irregular clasping behavior characterized by the dragging of the fore paws towards the trunk and asymmetry in limb position (Fig. 1c). This phenotype is generally associated with a neurological syndrome 31 . The Bmi1 gene operates as a central regulator of apoptosis and senescence through repression of the p19 Arf / p53 and p16 Ink4a /pRb tumor suppressor axes. Senescence is usually excluded in post-mitotic cells such as neurons, yet increased SA β-galactosidase activity and expression of the senescence-associated genes p19 Arf and p16 Ink4 have been noted in cortical neurons of aged p73 +/− mice and of Bmi1 −/− pups 18,25,32 . Furthermore, expression of p16 Ink4 is considered as a biomarker of aging 33 . We first confirmed neuronal expression of Bmi1 in the cortex of old WT mice and its severe reduction in old Bmi1 +/− mice (Fig. 1d). Next, we compared these old WT and Bmi1 +/− mice for SA β-galactosidase activity and p16 Ink4 /p19 Arf expression. We observed robust SA β-galactosidase staining in pyramidal neurons and a 2-fold increase in the total number of SA β-galactosidase-positive neurons/ section in layers 2 and 3 of the frontal cortex of Bmi1 +/− mice (Fig. 1e,f). This was accompanied with significant increased in p16 Ink4 and p19 Arf expression (Fig. 1g). These findings suggested that in addition to affect lifespan, Bmi1 hemi-deficiency resulted in the apparition of an age-dependent neurological syndrome associated with neuronal senescence. Bmi1 +/− mice show altered spatial memory and reduced long-term potentiation. To test whether the neurological symptoms were associated with perturbed cognitive functions, we examined apparently healthy 15 month-old WT and Bmi1 +/− mice for spatial learning and memory aptitude using the Morris water and Barnes maze tests. For the Morris water navigation task, mice were trained for 4 days to locate the platform (training) and then evaluated on their capacity to remember the platform location (probe test). Bmi1 +/− mice performed poorly in this assay during both the training period and probe test, as demonstrated by the increased time (latency) required to reach the platform (Fig. 2a,b). Bmi1 +/− mice also exhibited an anxious phenotype as demonstrated by an increased thigmotactic behavior while performing the training (Fig. 2d) 34 . The swimming speed of aged Bmi1 +/− mice was also reduced (0.25 m/s for WT and 0.20 m/s for Bmi1 +/− ) (Fig. 2c). In contrast, the swimming speed of 1 year-old WT and Bmi1 +/− mice was nearly identical (not shown), thus consistent with the apparition of a progressive, age-related reduction in locomotors strength. The Barnes maze test is based on mice aversion to open enlightened spaces, which ultimately prompts them to find secure shelter. Bmi1 +/− mice were more prone to visiting non-target holes while searching for the hole with the escape box and the latency of escape was also increased (Fig. 2e). Although these results suggested impaired spatial memory formation, the reduction in swimming speed and the possibility of visual dysfunction 23 led us to perform electrophysiological analyses to further test this possibility.
LTP is a cellular surrogate for spatial memory formation 35 . Using the same animals as in the behavioral tests, LTP of the Schaffer collateral pathway was induced in the CA1 region of the hippocampus. We found that a commonly used protocol for inducing LTP (100 Hz, 100 pulses) enhanced fEPSP slope in WT mice at 60 minutes after tetanus (Percent potentiation: 48.9 ± 5.2%, data from 6 slices and 3 mice). Compared with fEPSP recorded before tetanus (baseline), WT slices displayed significant potentiation (P = 0.036; paired Student's t-test) (Fig. 2f-h). Slices from Bmi1 +/− mice exhibited weaker potentiation than WT (11.7 ± 11.2%, data from 6 slices, 4 mice; transgenic vs. WT: P = 0.045, Student's t-test) and failed to show significant potentiation when compared to baseline at 60 minutes after tetanus (P = 0.357; paired Student's t-test) (Fig. 2f-h). Our findings demonstrated that LTP formation in Bmi1 hemi-deficient mice was significantly impaired, thus strengthening the hypothesis of impaired spatial memory formation. Old Bmi1 +/− mice present a pathology showing some features of AD. We previously showed that most neuronal anomalies in Bmi1 −/− mice were mediated by p53 activation 18 . Using immuno-blot, we observed increased p53 levels in the cortex of 15-20 month-old Bmi1 +/− mice when compared to WT littermates (Fig. 3a).
Hallmarks of AD and related dementias are accumulation of p-Tau, extra-cellular amyloid plaques, reduced synaptophysin immunoreactivity and neuronal loss. By immunoblot using antibodies that recognize p-Tau at Ser202 (AT-8) or Ser396 and Ser404 (PHF1), we observed the presence of p-Tau accumulation in Bmi1 +/− cortices (Fig. 3a). In cultured human neurons, BMI1 knockdown results in an increase in total Tau levels 20 . Similarly, we observed that total Tau levels were increased in the cortex of 22-24 month-old Bmi1 +/− mice (Fig. S1a,b).
Remarkably, these anomalies correlated with reduced synaptophysin levels (indicative of synaptic atrophy) and increased p-JNK immunoreactivity (Fig. 3a) 36,37 . By immuno-histochemistry (IHC), we observed p-Tau immunoreactivity in the neuron's perikarya of 20 month-old WT and Bmi1 +/− mice. However, large p-Tau deposits in the neuronal cell body and strong immunoreactivity in track fibers of the cortical white matter were only observed in Bmi1 +/− mice (Fig. 3b). The presence of p-Tau tangles was not observed. We next performed immunoblot on mouse cortices using an antibody (FCA3542) that can recognize the mouse and human amyloid peptide. Notably, we could detect the presence of a ~12-14 kDa band only in the Bmi1 +/− brain samples (Fig. 3c-Input 20%). This band was also revealed using FCA3542 after immunoprecipitation (IP) with another anti-amyloid antibody (DE2B4), but not after IP with IgG ( Fig. 3c-IP). Antibody specificity was also confirmed using cortical extracts from human APP Swe/Dutch/Iowa (APP) transgenic mice ( Fig. 3c-IP). On Bmi1 +/− brain sections at 20 months of age, we observed that amyloid was preferentially found in the neuronal soma (Fig. 3d). Amyloid plaques were not observed. Cleavage of APP by the β-secretase Bace1 generates the ~12-14 kDa C99 fragment (C99). Notably, we observed increased Bace1 levels in cortical neurons of Bmi1 +/− mice (Fig. 3a,e). To distinguish between accelerated aging and neurodegeneration, we compared the brains of 24 month-old WT mice with that of 15 month-old Bmi1 +/− mice for the expression of the ~12-14 kDa fragment. Notably, the amyloid-related band was not detectable in the cortex of old WT mice (Fig. 3f). Likewise, p-JNK and p-Tau (AT-8) were not observed in the 24 month-old WT mice (Fig. 3g).
To evaluate the possibility of neuronal loss, we calculated the total number of cortical neurons per section in the frontal cortex using a pan-neuronal marker (NeuN). When compared to WT littermates, 15-20 month-old Bmi1 +/− mice displayed a ~20% reduction in NeuN-positive neurons (Fig. 3h). To examine if the lower neuronal density observed in Bmi1 +/− mice correlated with increased apoptosis, we quantified the number of neurons positive for activated caspase-3 in the frontal cortex. This revealed a ~2 fold increase in the frequency of apoptotic neurons in old Bmi1 +/− mice (Fig. 3i). Reduced neuronal density and increased apoptosis were also observed in the hippocampus of 15-17 month-old Bmi1 +/− mice (Fig. S2a,b), which was consistent with the behavioral and LTP results. We next analyzed the brains from rare surviving 22-25 month-old Bmi1 +/− mice using antibodies against amyloid (n = 7). A single 24 month-old Bmi1 +/− animal presented amyloid deposits that were present in all cortical areas, with other brain regions being generally less affected (Fig. 3j). Interestingly, the  pattern of amyloid immuno-reactivity was slightly distinct when using another amyloid antibody (Fig. S3a,b). Intra-neuronal accumulation of amyloid and reduced synaptophysin immuno-reactivity was also observed in the hippocampus of another Bmi1 +/− mouse (Fig. S3c,d). This revealed that old Bmi1 +/− mice could spontaneously develop a disorder sharing some similarities with AD.

Bmi1 hemi-deficiency cooperates with APP-mediated neurodegeneration. Genetic cooperation
between two mutations can be indicative of protein-protein interactions and/or convergent molecular pathways.
To test if Bmi1 hemi-deficiency could cooperate or not with FAD-associated mutations, we crossed Bmi1 +/− mice with human APP transgenic mice. From this, we obtained 4 genotypes in F1: WT, Bmi1 +/− , APP and APP/ Bmi1 +/− mice. Kaplan Mayer analysis revealed that although both Bmi1 +/− and APP mice had reduced maximal and median lifespan compared to WT littermates, the median lifespan of the APP/Bmi1 +/− animals was further reduced by ~45% when compared to APP mice (Fig. 4a). The difference was significant despite the limitations of using a small data set. When analyzed by IHC at 6 months of age for p-Tau, we observed almost no signal in WT and Bmi1 +/− mice, and weak immunoreactivity in APP mice. Notably, while neuronal accumulation of p-Tau was rare in APP mice and absent in Bmi1 +/− mice, it was relatively robust and abundant in APP/Bmi1 +/− mice (Fig. 4b,c). Similarly, although intra-neuronal amyloid immunoreactivity was weak to absent in Bmi1 +/− mice at this young age, it was easily detectable in APP mice (Fig. 4d). Notably, the number of immunoreactive neurons and the signal strength for amyloid were greatly enhanced in APP/Bmi1 +/− mice (Fig. 4d,e). Other parameters of disease severity were also enhanced in APP/Bmi1 +/− mice, including reactive gliosis, microglia activation, synaptic atrophy and apoptosis (Fig. S4). Taken together, these results suggested genetic cooperation between Bmi1 hemi-deficiency and mutant APP in neurodegeneration.

Loss of heterochromatin and DDR in cortical neurons of Bmi1 +/− mice. BMI1 is required for hetero-
chromatin maintenance in somatic cells 19 . We thus investigated if the AD-like phenotype of Bmi1 +/− mice was also associated with heterochromatin alterations. By Western blot analysis on whole cortices, we observed reduced levels for histones H2A ub and H3K9 me3 in 15 month-old Bmi1 +/− mice (Fig. 5a). To test if this correlated with an abnormal chromatin structure, we performed IHC. We observed that in contrast to WT mice, where 2-3 large chromocenters/ neuron were present, chromocenters of Bmi1 +/− neurons were smaller and more numerous, suggesting heterochromatin de-nucleation (Fig. 5b). Consistently, this was associated with transcriptional de-repression of satellite repeats, which are normally silenced by heterochromatin (Fig. 5c) [38][39][40] . One prediction of heterochromatin de-nucleation is activation of a DDR. The Ataxia Telengectasia Mutated (ATM) and Ataxia Telengectasia and RAD-3 Related (ATR) kinases work at the apex of the DDR [41][42][43] . Using immunoblot, we observed accumulation of phospho-ATM at Serine 1981 (p-ATM) and phospho-ATR at Serine 428 (p-ATR) in the cortices of Bmi1 +/− mice (Fig. 5d). The histone mark γH2AX is present at sites of DNA damage 44 . We observed nuclear accumulation of γH2AX in cortical neurons only in Bmi1 +/− mice, suggesting that the DDR occurs primarily in neurons (Fig. 5e).
To localize DNA damage on the genome, we performed chromatin immuno-precipitation (ChIP) experiments on whole cortices. As expected, Bmi1, H3K9 me3 and H2A ub were enriched at genomic repeats (Minor and Major satellite repeats and LINE elements) and at the Hoxa7 locus in WT mice and were depleted in Bmi1 +/− mice (Fig. 5f) 19 . In contrast, while p-ATR and γH2AX were nearly absent in WT mice, they were preferentially and highly enriched at genomic repeats in Bmi1 +/− mice (Fig. 5f).
To test if heterochromatin anomalies were present before disease onset, we performed immuno-blot, IHC and confocal microscopy on cortices from 2-3 month-old WT and Bmi1 +/− mice. This revealed reduced H3K9 me3 levels and fragmented H3K9 me3 immunostaining in Bmi1 +/− pyramidal neurons when compared to WT (Figs 5g  and S5a,b). Reduced H3K9 me3 levels were also present in cultured embryonic day 18.5 neurons, as measured using ChIP analyses (Fig. S5c). Despite the presence of chromatin anomalies at the earliest step of neurogenesis, we failed to detect a DDR or the expression of AD-related markers in 3 month-old Bmi1 +/− mice using immuno-blot (Fig. S5d). Mild accumulation of p53 was however noticed in Bmi1 +/− mice (Fig. S5d). This suggested that in Bmi1 +/− mice, loss of heterochromatin is a very early event in the disease process.
To understand the basis of p53 activation, we cultured WT, Bmi1 −/− and Bmi1 +/− embryonic day 18.5 cortical neurons to test if free radical-mediated DNA damage was responsible for p53 accumulation 18,47,48 . Using N-Acetyl Cysteine (NAC), a free radical scavenger, we found that reducing free radicals could slightly improve p53 levels in Bmi1 −/− and Bmi1 +/− neurons (Fig. 6c). The p19 Arf protein can stabilize p53, and p19 Arf is up regulated in Bmi1 −/− neurons 18,49 . However, co-deletion of Bmi1 and Ink4a (encoding for p16 Ink4a and p19 Arf ) could not prevent p53 accumulation (Fig. 6d). In response to DNA damage, the ATM and ATR kinases can target p53 for phosphorylation 45,50,51 . We found that inhibition of ATM alone could not prevent p53 accumulation, but that inhibition of ATM and ATR largely mitigated p53 accumulation in Bmi1 −/− neurons (Fig. 6e). Residual p53 levels were however observed. We also observed the presence of p-ATM and phospho-Chk1 (a downstream target of ATR) in Bmi1 −/− neurons and found that ATM/ATRi was sufficient to prevent their accumulation (Fig. 6f). Remarkably, p-Tau accumulation in Bmi1 −/− neurons could be also prevented after exposure to ATM/ATRi (Fig. 6f). These results suggested a theoretical model where AD-related neuronal pathologies observed in Bmi1-deficient mice are triggered in part by a ATM/ATR-driven DDR working upstream of p53, but where Bmi1/Ring1 ubiquitin ligase activity and Bmi1-mediated ROS repression are also likely required to prevent p53 accumulation (Fig. 6g) 52 .

Loss of heterochromatin and DDR in AD brains.
We next investigated the possible relevance of our findings in the context of AD. We first confirmed the robust expression of BMI1 in cortical neurons from aged controls and BMI1 depletion in neurons from AD patients (Fig. 7a). To test for possible heterochromatin Quantitative PCR was performed in triplicate for each DNA sequence. All data are presented as fold of input. Bmi1 and H2A ub reduction (p < 0.05), and p-ATR and γH2AX accumulation (p < 0.01) in Bmi1 +/− mice are significant (Two-way ANOVA analysis). (bottom panel) Fold differences for γH2AX and p-ATR accumulation in Bmi1 +/− mice were measured relative to WT levels for each DNA sequences. (g) Cortical slices from 45-day old mice. Areas showed are located in layers 2-3 of the frontal cortex. In WT samples, large pyramidal neurons were strongly labeled by the H3K9 me3 antibody and H3K9 me3 co-localized with DAPI. In Bmi1 +/− samples, large pyramidal neurons were weakly labeled by the H3K9 me3 antibody and the H3K9me3 signal was fragmented. Bottom panels: Quantification of the images showed in (f).
Scientific REPORTS | (2019) 9:594 | DOI:10.1038/s41598-018-37444-3 anomalies, we performed IHC on frontal cortex sections 53 . In control samples, all neurons were labeled with H3K9 me3 and presented ~3 heterochromatic foci (or chromocenters)/neuron (Fig. 7b,c). In AD samples, ~38% of all pyramidal neurons showed highly reduced or absent H3K9 me3 nuclear staining, and the neuron's chromocenters were smaller and more numerous (Fig. 7b,c). To test the possibility of genomic instability, we analyzed the expression of DDR proteins. By Western blot analysis, we observed accumulation of p-ATM, p-ATR and p-CHK1 in AD brains, but not in age-match controls (Fig. 7d,d'). The presence of a DDR in AD brains also correlated with reduced H3K9 me3 levels (Fig. 7d). In contrast, accumulation of DDR proteins was not observed in FAD brains, with the exception of one sample presenting p-ATM accumulation (Fig. 7e). Likewise, H3K9 me3 levels were unaffected in FAD brains (Fig. 7e). Using IHC, we also confirmed that DDR in AD brains occurred primarily in neurons (Fig. 7f). To establish where the DDR was located on the genome, we performed ChIP experiments using frontal cortex samples. When compared to controls, we found significant enrichment for p-ATR and γH2AX at genomic repeats (SATIII, SATA and McBOX) in AD brains (Fig. 7g). In contrast, there was no enrichment at HOXC13, β-GLOBIN or ACTIN (Fig. 7g). Although there was a general trend for p-ATM enrichment at genomic repeats in AD samples, it was not significant. BMI1 and H3K9 me3 were also enriched at HOXC13 and at genomic repeats in control brains and were depleted in AD brains (Fig. 7g). Taken together, these findings revealed remarkable similarities between the neuronal genomic instability phenotype of Bmi1 +/− mice and AD patients.

Discussion
We have found that Bmi1 +/− mice develop with age a neurological and progeroid syndrome characterized by alopecia, weight loss, abnormal paw clasping reflex and neuronal senescence. Memory deficit as measured by their performance in the Morris water maze and Barnes test correlated with impairment in LTP formation as well as with synaptic atrophy and neuronal loss. Pathological brain anomalies such as p-Tau and amyloid accumulation were also found in old Bmi1 +/− mice. Remarkably, when compared to Bmi1 +/− or APP mice alone, Bmi1 +/− /APP mice showed significantly shorter lifespan and increased disease severity. Genetic and cell culture studies further revealed that inhibition of the DDR and/or p53 could mitigate the accumulation of p-Tau and amyloid in neurons from Bmi1-null mice, suggesting that the neuronal response to genomic instability plays a role in the disease process. At last, loss of heterochromatin in neurons and accumulation of DNA damage at genomic repeats were found in Bmi1 +/− mice and AD brains.
One possible interpretation of our data is that Bmi1 +/− mice age faster and thus present age-related neurological anomalies earlier than WT mice. This possibility is supported by our observation that some pathological marks, including p-Tau (using PHF1, which recognizes an epitope around Ser396 and Ser404) and SA β-galactosidase activity, although much less abundant, were also detected in the brains of old WT mice. However, other pathological marks, such as p-Tau (using AT-8, which recognizes Ser202), p-JNK and amyloid (C99) accumulation were not detected even in 24 month-old WT mice, suggesting that Bmi1 hemi-deficiency represents a pathological brain aging process that is distinct from end-stage physiological brain aging.
Using the Bmi1 +/− and Bmi1 −/− mouse models, we observed activation of DDR proteins and of p53 in neurons. This is consistent with previous work showing that Bmi1 is important to maintain genomic stability [15][16][17][18]54 . Interestingly, over-expression of Chk1 or Chk2 was shown to exacerbate Tau toxicity in a Drosophila model of neurodegeneration by specific phosphorylation of human Tau at Ser262 55 . Likewise, JNK activation may promote Tau phosphorylation 32 . Thus, several kinases potentially targeting Tau are activated in Bmi1 +/− mice. It however remains to be investigated by which mechanisms Bace1 level is increased and why the amyloid metabolism is affected in Bmi1 +/− mice. Our results showed that amyloid accumulation was mitigated after p53 deletion and we recently reported that p53 activation in human neurons might perturb the clearance of misfolded proteins by interfering with proteasome function 20 . Thus, a plausible mechanism to explain our results is that activated p53 also impairs proteasome function in mouse neurons. Further work is however needed to validate this.
The heterochromatin island hypothesis of aging stipulates that epigenome instability at the constitutive heterochromatin is a driving force of cellular aging 56,57 . This hypothesis may be particularly relevant in post-mitotic neurons, where cell division associated with telomere attrition is irrelevant to the cellular aging process. Notably, inactivation of chromatin remodeling proteins, such as those of the NURD complex, results in DDR and in cellular senescence that are preceded by heterochromatin alterations 58 . Furthermore, modifications of constitutive heterochromatin in response to an acute stress can result in epigenome instability and perturbation of gene expression program, which are hallmarks of cellular aging 59 . In mice, deletion of the SUV39h1/h2 methyltransferases leads to reduced viability, loss of H3K9 me3 and genomic instability 60 . In Drosophila, perturbation of H3K9 me2 levels through inactivation of Su(var)3-9 results in genomic instability and constitutive DDR at the heterochromatin, in both somatic and germ-line cells 61 . We showed here that heterochromatin alterations in Bmi1 +/− mice are present at the time of neurogenesis and thus before activation of DDR proteins and neurodegeneration. Notably, this correlated with preferential accumulation of DNA damage (as marked by γH2Ax) at repetitive DNA sequences. Similarly, we found that DNA damage accumulation was predominant at genomic repeats in AD brains. Exactly why loss of heterochromatin compaction strongly correlates with DNA damage accumulation at genomic repeats is unknown. It is possible that repetitive DNA sequences are intrinsically unstable upon loss of heterochromatin compaction and thus prone to de novo DNA damage formation. It is also possible that the the frontal cortex of age-match control (n = 2) and AD (n = 2) patients. p-ATM is predominant in AD neurons and accumulates in both cytosolic (blue arrowhead) and nuclear (black arrowhead) compartments. (g) ChIP experiments were performed on frontal cortex extracts of age-match control (n = 6) and AD brains (n = 6). All data are represented as fold of input. (A) Note BMI1 and H3K9 me3 enrichment at McBOX, SATIII and SATA in control samples. BMI1 was also found at HOXC13. BMI1 and H3K9 me3 were depleted at all loci in AD samples (bottom). Fold differences between control and AD samples for p-ATM, p-ATR and γH2AX accumulation. Values are mean ± SEM. *P < 0.05; **P < 0.01; ***P < 0.001; Two way-ANOVA test was performed for multiple gene analysis and Student's t-test for single gene analysis.
Scientific REPORTS | (2019) 9:594 | DOI:10.1038/s41598-018-37444-3 efficiency of DNA repair is reduced upon BMI1 deficiency or that recognition of the H3K9 me3 mark is important to activate the process of DNA repair at these specific regions 15,16,62 . It remains however to be tested whether heterochromatin anomalies observed in Bmi1 +/− mice are sufficient to induce the DDR and neurodegeneration. Interstitial amyloid plaques and Tau tangles represent the classical pathological hallmarks of AD. While synaptic atrophy, neuronal loss and behavioral anomalies are present in 15-month old Bmi1 +/− mice, the animals do not develop Tau tangles or amyloid plaques (with the exception of one animal). Rather, old Bmi1 +/− mice present intra-neuronal accumulation of amyloid and show moderate levels of p-Tau deposits in the neuronal cell body and the axon. The absence of amyloid plaques formation in Bmi1 +/− mice is not surprising since the mouse amyloid peptide is known to be less prone to aggregation than the human one 63 . Also, transgenic mice over-expressing wild-type human APP rarely develop amyloid plaques, in contrast with those over-expressing APP with FAD-associated mutations 64 . Notably, massive neurodegeneration was reported in APP/PSEN1 transgenic mice (APP(SL)PS1KI mice), which only develop intra-neuronal accumulation of amyloid 65 . Thus, intra-neuronal accumulation of oligomeric amyloid species, including the C99 fragment, may be sufficient to induce neurodegeneration 66,67 . Nevertheless, the absence of amyloid plaques and Tau tangles as well as the late onset neuronal disease should be taken into account before using Bmi1 +/− mice as a model of AD.
In conclusion, we demonstrated that loss of one Bmi1 allele in mice results in age-related neurodegeneration sharing some similarities with AD. Bmi1 +/− mice may thus represent an interesting animal model to identify new pathogenic mechanisms related to AD.

Data Availability Statement
The authors declare that they will make available all data presented in this manuscript.