Spontaneous development of Alzheimer's disease‐associated brain pathology in a Shugoshin‐1 mouse cohesinopathy model

Summary Spontaneous late‐onset Alzheimer's disease (LOAD) accounts for more than 95% of all human AD. As mice do not normally develop AD and as understanding on molecular processes leading to spontaneous LOAD has been insufficient to successfully model LOAD in mouse, no mouse model for LOAD has been available. Existing mouse AD models are all early‐onset AD (EOAD) models that rely on forcible expression of AD‐associated protein(s), which may not recapitulate prerequisites for spontaneous LOAD. This limitation in AD modeling may contribute to the high failure rate of AD drugs in clinical trials. In this study, we hypothesized that genomic instability facilitates development of LOAD and tested two genomic instability mice models in the brain pathology at the old age. Shugoshin‐1 (Sgo1) haploinsufficient (∓) mice, a model of chromosome instability (CIN) with chromosomal and centrosomal cohesinopathy, spontaneously exhibited a major feature of AD pathology; amyloid beta accumulation that colocalized with phosphorylated Tau, beta‐secretase 1 (BACE), and mitotic marker phospho‐Histone H3 (p‐H3) in the brain. Another CIN model, spindle checkpoint‐defective BubR1−/+ haploinsufficient mice, did not exhibit the pathology at the same age, suggesting the prolonged mitosis‐origin of the AD pathology. RNA‐seq identified ten differentially expressed genes, among which seven genes have indicated association with AD pathology or neuronal functions (e.g., ARC, EBF3). Thus, the model represents a novel model that recapitulates spontaneous LOAD pathology in mouse. The Sgo1−/+ mouse may serve as a novel tool for investigating mechanisms of spontaneous progression of LOAD pathology, for early diagnosis markers, and for drug development.

oxidative stress, cholesterol metabolism, glycation, and other environmental and lifestyle factors are recognized as aggravating factors (Scheltens et al., 2016).
Major pathological features of the human AD brain include plaques of amyloid-b made of cleaved APP, tangles of Tau proteins, and congophilic cerebral amyloid angiopathy (Kitazawa, Medeiros, & Laferl, 2012;Onos, Sukoff Rizzo, Howell, & Sasner, 2016;Sasaguri et al., 2016;Scheltens et al., 2016). With insufficient knowledge on the cause, modeling LOAD in rodents has been an issue. Normally, mice do not develop AD, which has been interpreted as due to their shorter lifespan and sequence differences in APP and Tau. Various transgenic mouse models with modified APP, Tau, and others have been developed for EOAD (Kitazawa et al., 2012;Onos et al., 2016;Sasaguri et al., 2016). However, LOAD models are limited to apes and are practically nonexistent in rodents. AD drugs developed with EOAD models were explored for use as human LOAD therapy, under an assumption that drugs effective on EOAD would also be effective in treating LOAD. More than 98% of drugs tested in EOAD rodent models were ineffective in human LOAD patients in clinical trials (Cummings, Morstorf, & Zhong, 2014), raising concerns on current drug targets and on the validity of the EOAD models for human LOAD.
Genomic instability and aneuploidy have been suspected to cause or aggravate AD, as high rates of both are present in the human AD brain (Bajic, Spremo-Potparevic, Zivkovic, Isenovic, & Arendt, 2015). Further, genomic instability biomarkers have been associated with mild cognitive impairment and AD (Lee, Thomas, & Fenech, 2015). Aneuploidy may facilitate development of AD-like dementia, as 15% of patients with Down syndrome with chromosome 21 trisomy develop AD-like cognitive dysfunction in their 40s, and the rate increases to 50%-70% by age 60 (Bajic et al., 2015;Potter, 1991). However, a causal link between CIN, aneuploidy, and AD has not been established.
BubR1 is a mitotic spindle checkpoint component. BubR1 À/+ mice showed mitotic slippage in the cells and were colon cancerprone (Dai et al., 2004;Rao et al., 2005), and BubR1 H/H hypomorphic mice were identified as a model for premature aging (Baker et al., 2004). Neuronal cell division and axon growth were inhibited by siRNA-mediated BubR1 knockdown in the mouse brain (Yang et al., 2017). The above findings in BubR1 transgenic models led to a hypothesis that mitotic errors and CIN facilitate AD-like neurodegeneration.
With the current gap in LOAD study models, we hypothesized that LOAD development is facilitated by genomic instability with CIN. We tested whether the Sgo1 À/+ or BubR1 À/+ haploinsufficient mouse could serve as a model for spontaneous LOAD progression.

| Middle-aged Sgo1 À/+ expressed Amyloid Beta
Precursor Protein-Binding Family B Member 1 (APBB1) at a higher amount in the whole blood To identify biomarkers for CIN and cohesinopathy in whole blood RNA, we performed comparative whole blood RNA-seq analysis on 12-month-old Sgo1 À/+ and wild-type mice. Among differentially expressed genes (p < 0.05), Amyloid Beta Precursor Protein-Binding Family B Member 1 (APBB1) was notable, with a 3.63-fold increase compared with wild-type control (Supporting Information Figure S1). APBB1 encodes a protein involved in DNA damage repair, interacts with APP, and is thought to promote AD. The pilot result at a younger age led us to suspect that the brains of Sgo1 À/+ mice would show signs of neurodegenerative disease similar to AD.

| Sgo1 À/+ brains accumulated amyloid-b by 24 months of age
An aging-and-carcinogenesis study cohort provided Sgo1 À/+ brains at older ages (24-25 months) corresponding to human old age over 65. We also collected brains from BubR1 À/+ haploinsufficient mice to determine whether they show brain aging and AD pathology, as the initial hypothesis was focusing on CIN and AD. Increased accumulations of amyloid-b (i.e., increase in amyloid-b/APP ratio) were observed in brain extracts from Sgo1 À/+ mice, but not from control littermate wild-type or BubR1 À/+ mice (Figure 1a,b). There was no significant difference in the total amount of phosphorylated TAU ( Figure 1a).
With the assumption that CIN would affect AD pathology, we were puzzled by the result that our haploinsufficient BubR1 À/+ model did not show more amyloid-b in the brain than did wild-type mice. A major difference between the Sgo1 À/+ cohesinopathy model and the BubR1 À/+ mitotic checkpoint-defective model is the mitotic checkpoint function and existence (or absence) of prolonged mitosis. Sgo1 À/+ brains showed higher expression of mitotic marker phospho-Histone H3 (p-H3), consistent with prolonged mitosis, while BubR1 À/+ brains did not (Figure 1a,b). and (b) cytoplasmic staining enriched in living cells. In addition to T A B L E 1 Differentially expressed genes in 24-m-old Sgo1 À/+ brain compared with age-matched wild type (FDR-p < 0.05, twofold cutoff)

Max group mean
The immunofluorescence results support the hypothesis that the source of accumulation of amyloid-b is p-H3-positive cells that also coexpress BACE and p-TAU. Overall, the results suggested that F I G U R E 2 Sgo1 À/+ brains displayed AD-associated pathology that may originate from mitotic cells (a) Sgo1 À/+ brain shows cells with APP/ amyloid-b and phosphorylated TAU. Next, we tested Congo red staining for amyloidosis, which did not provide clear staining in the Sgo1 À/+ (not shown). Lack of Congo red staining suggested that degree of amyloid accumulation is not as high in this Sgo1 À/+ model as existing EOAD mouse models that typically express a few-to-several-fold amount of total amyloids compared with controls and show Congo red staining (Kitazawa et al., 2012;Onos et al., 2016;Sasaguri et al., 2016). The result was in agreement with immunoblots in Figure 1 indicating only mild increase in total amyloids (amyloid-b and APP combined) in Sgo1 À/+ compared with agematched wild type and BubR1 À/+ . The modest increase in total amyloids suggests that the Sgo1 À/+ model may recapitulate relatively early phase of spontaneous LOAD development.

| p-H3 expression in Sgo1 was observed both in the cortex and in the hippocampus, while p-H3 expression is limited to the hippocampus in wild type
Dentate Gyrus (DG) and subgranular zone in hippocampus are known to be sites for adult neurogenesis (Bordiuk, Smith, Morin, & Sem€ enov, 2014). Hippocampus is also known to be the site functionally affected by LOAD, leading to the primary LOAD symptom of memory defect.
We tested whether amyloid-b-and-p-H3-positive cells in Sgo1 À/+ localize in a particular area (e.g., hippocampus) in the brain. Amyloid-band-p-H3-positive cells in Sgo1 À/+ appeared both in the cortex and in the hippocampus (Figure 3a,b). The p-H3-positive cell percentages in Sgo1 À/+ were estimated as 23.96 ∓ 15.32% in the cortex and 16.64 ∓ 6.48% in the hippocampus (Figure 3c). In control wild type (photograph in Supporting Information Figure S3), amyloid-b-positive cells were hardly present, and p-H3-positive cells were localizing in the hippocampus (10.31 ∓ 6.31%), but not in the cortex (2.33 ∓ 3.82%) (Figure 3c). The data demonstrate that amyloid-b-and-p-H3-positive cells characteristically (p < 0.05) appear in the cortex of Sgo1 À/+ , although there is a modest (nonsignificant) increase of p-H3 in the hippocampus of Sgo1 À/+ compared with wild type as well.
2.5 | Differentially expressed genes in Sgo1 À/+ brain To elucidate the molecular basis for the AD-associated brain pathology in Sgo1 À/+ , we used RNA-seq to compare mRNA expression profiles in 24-month-old brains. With p < 0.05 and twofold cutoff, ten genes were identified (Table 1). ARC, PMCH, Gm20388, and F I G U R E 2 (Continued).
RAO ET AL.
| 5 of 11 AA465934 were overexpressed, while Shisa8, Ebf3, DAO, Slc6a5, PPP1r17, and PCP2 were underexpressed (Figure 4). Among the ten genes, seven had known connections to AD and/or neuronal function. In wild-type brains (picture in Supporting Information Figure S3), p-H3 signals located exclusively in hippocampus (p = 0.005). In Sgo1 À/+ brains, high expression of p-H3 in the cortex was observed, hence p-H3 expression was not limited to hippocampus in Sgo1 À/+ . Percentages of p-H3-positive cells in Sgo1 À/+ cortex were significantly higher compared with those in cortex of age-matched wild-type control (p = 0.0003). Although p-H3 expression in hippocampus was modestly higher in Sgo1 À/+ compared with wild type, the difference was not significant (p = 0.1153 * * * * * * * F I G U R E 4 Differential expressions of AD and/or neuronal function-related genes in the Sgo1 À/+ brain. Heat map. Blue color indicates lower expression, and red indicates higher expression. Gm20388, AA465934, PMCH, and ARC were more highly expressed in Sgo1 À/+ mice, while Shisa8, Ebf3, DAO, Slc6a5, PPP1r17, and PCP2 were expressed at lower levels. Genes marked with asterisk had known connections to AD and/or neuronal function (see text)

| DISCUSSION
Our results suggest that accumulated amyloid-b originated from p-H3-positive prolonged mitotic cells, which later die and leave extracellular deposits including amyloid-b and p-TAU that may become seeds for "plaques and tangles." Cells with accumulated amyloid-b were specifically observed in Sgo1 À/+ model mice with intact spindle checkpoint and not in mitotic checkpoint-defective BubR1 À/+ model mice. There is supporting evidence suggesting that mitotic cells are involved in amyloid-b accumulation in human LOAD: (a) Human neurofibrillary tangles colocalized with MPM2 antigens, another mitotic marker (Kondratick & Vandr e, 1996); (b) Abnormal Tau phosphorylation of the Alzheimer-type also occurred during mitosis in human neuroblastoma SY5Y cells overexpressing Tau (Delobel et al., 2002); (c) APP Thr668 phosphorylation in mitosis correlated with increased processing of APP to generate Ab and the C-terminal fragment of APP (Judge, Hornbeck, Potter, & Padmanabhan, 2011); (d) Although p-H3 localization is usually limited in chromatin in many other organs, human AD brain showed a cytoplasmic, diffused pattern of p-H3 (Ogawa et al., 2003), which was recapitulated in the Sgo1 À/+ mouse brain (Figures 2d and 3a,b). These reports strongly suggest that human LOAD development can be aided by prolonged mitosis, which the Sgo1 À/+ model recapitulates. Indeed, in human LOAD, models incorporating the critical role of mitotic cells have been proposed, such as a "simple linear model" that states that human AD pathology develops from mitotic cycle-reentering neurons that later die (Herrup, 2010), and the "two-hit model" of human LOAD (Zhu, Lee, Perry, & Smith, 2007;Zhu, Raina, Perry, & Smith, 2004) that purports that LOAD development occurs with (a) oxidative stress and (b) mitotic reentry.
Although the direct trigger for mitotic cycle reentry in Sgo1 À/+ model mice remains unclear, studies on roles of cell cycle regulators, such as Cdk5 (Zhang et al., 2008), and on effects of genes identified through RNA-seq in this study on the cell cycle, are warranted.
The Sgo1 À/+ haploinsufficient mouse is a model that displays two direct effects of a reduction in Sgo1, both of which lead to prolonged mitosis via the spindle checkpoint: (a) cohesinopathy in mitotic chromosome; and (b) defects in centrosome integrity (Yamada et al., 2012). Whether the AD pathology is caused by cohesinopathy, centrosome defect, or the common consequence that is prolonged mitosis must be distinguished. Cohesinopathy in humans leads to diseases with cancer proneness, developmental malformation, and/or intellectual disability and behavioral issues, such as Cornelia de Lange syndrome or mutations in STAG1 or STAG2 (Kline et al., 2017;Kumar et al., 2015). The symptoms suggest that maintenance of chromosome cohesion may play a role more critical than previously anticipated in brain functions.
Mutations in centrosomal genes are often connected to developmental malformations of the brain, such as autosomal recessive primary microcephaly, microcephalic osteodysplastic primordial dwarfism type II, and Seckel syndrome, in humans (Nigg, Caj anek, & Arquint, 2014). These rare human diseases have not been studied in the context of AD, in part because they are rare yet patients do not survive long, and because mental retardation symptom is difficult to distinguish from cognitive dysfunction of AD. However, the use of corresponding mouse models should provide guidance. At this moment, we do not think Sgo1 is the only target gene to create spontaneous LOAD mouse model. We speculate that functional equivalent of Sgo1 mutation can also create LOAD model. Indeed, functional equivalent of Sgo1 mutation, such as accumulation of aneuploid cells (Bajic et al., 2015;Potter, 1991) or increase in cells reentering mitotic cycle, does occur in human LOAD (Herrup, 2010;Zhu et al., 2004Zhu et al., , 2007. RAO ET AL.

| 7 of 11
Our results also suggest that prolonged mitosis and/or mitotic spindle checkpoint may have potential as a therapeutic target for AD. Agreeing with this prediction, Flavopiridol, a CDK inhibitor, reversed memory impairment in amyloid-b-injected AD mouse model, suggesting the detrimental role of prolonged mitosis in AD (Leggio et al., 2016).
Here, we present evidence that the Sgo1 À/+ haploinsufficient mouse model displays AD-like brain pathology at an age equivalent to human old age. The model will allow us to test genetic interactions between known AD-associated genes (e.g., APOE, ARC) through simple breeding, as well as the influence from environmen-

| Immunoblots
Frozen brain samples (mouse cerebrum including cortex and hippocampus, excluding olfactory bulb, cerebellum, medulla) were extracted in extraction buffer and subjected to immunoblots following our standard protocol (Rao et al., 2016;Yamada et al., 2012).
Blots were quantified using IMAGEJ 1.43 software (NIH). b-actin blots were used for loading control and normalization.

| Immunofluorescence
Formalin-fixed brain hemispheres were embedded in paraffin and sectioned onto slides. After deparaffinization, antigen retrieval, sodium borohydride treatment, CuSO 4 treatment, and blocking, the slides were treated with primary antibodies for 16 hr, then with secondary fluorescent antibodies for 1 hr, followed by brief DAPI staining and sealing with antifade. Sodium borohydride and CuSO 4 were used to minimize autofluorescence by Shiff-base and by Lipofuscin, respectively (Schnell, Staines, & Wessendorf, 1999;Spitzer, Sammons, & Price, 2011).

| Quantification of Immunofluorescence
In the experiment in Figure 3 and Supporting Information Figure

| RNA-seq
Comparative RNA sequencing was performed as in (Rao et al., 2016;Yamada et al., 2016). The total RNA was extracted from frozen brain

| Statistical analysis (RNA-seq)
We used Student's t test to analyze the data. Statistical significance was evaluated by algorithms integral to the aforementioned software. FDR-adjusted p values of <0.05 were considered significant.

| Data and materials availability
The RNA-seq dataset was deposited to the NIH-GEO database (accession number GSE115185) and will be available there on June which provided RNA-seq-bioinformatics service.

CONF LICT OF I NTEREST
The authors declare no conflict of interests.