GSK3‐ARC/Arg3.1 and GSK3‐Wnt signaling axes trigger amyloid‐β accumulation and neuroinflammation in middle‐aged Shugoshin 1 mice

Abstract The cerebral amyloid‐β accumulation that begins in middle age is considered the critical triggering event in the pathogenesis of late‐onset Alzheimer's disease (LOAD). However, the molecular mechanism remains elusive. The Shugoshin 1 (Sgo1−/+) mouse model, a model for mitotic cohesinopathy‐genomic instability that is observed in human AD at a higher rate, showed spontaneous accumulation of amyloid‐β in the brain at old age. With the model, novel insights into the molecular mechanism of LOAD development are anticipated. In this study, the initial appearance of cerebral amyloid‐β accumulation was determined as 15‐18 months of age (late middle age) in the Sgo1−/+ model. The amyloid‐β accumulation was associated with unexpected GSK3α/β inactivation, Wnt signaling activation, and ARC/Arg3.1 accumulation, suggesting involvement of both the GSK3‐Arc/Arg3.1 axis and the GSK3‐Wnt axis. As observed in human AD brains, neuroinflammation with IFN‐γ expression occurred with amyloid‐β accumulation and was pronounced in the aged (24‐month‐old) Sgo1−/+ model mice. AD‐relevant protein panels (oxidative stress defense, mitochondrial energy metabolism, and β‐oxidation and peroxisome) analysis indicated (a) early increases in Pdk1 and Phb in middle‐aged Sgo1−/+ brains, and (b) misregulations in 32 proteins among 130 proteins tested in old age. Thus, initial amyloid‐β accumulation in the Sgo1−/+ model is suggested to be triggered by GSK3 inactivation and the resulting Wnt activation and ARC/Arg3.1 accumulation. The model displayed characteristics and affected pathways similar to those of human LOAD including neuroinflammation, demonstrating its potential as a study tool for the LOAD development mechanism and for preclinical AD drug research and development.


| Unmet need for AD drug development
With an increasing number of patients and no effective treatments, AD drug represents a major unmet need in medicine. As of early 2020, more than 200 clinical trials for AD have failed, including 9 phase 3 trials since 2016 (Cummings et al., 2016;Cummings, Morstorf, & Zhong, 2014;Huang, Chao, & Hu, 2020). The high failure rate may be attributed to (a) incomplete understanding of human AD pathology development, and (b) insufficient representation of lateonset Alzheimer's disease (LOAD) models in the AD drug research and development (R&D) process.

| Hypotheses for load pathology development
Drug development is driven by hypotheses. Strong hypotheses that are well-supported by data are needed for clinical AD drug development. A number of hypotheses have been proposed for human AD development and its cause, including the "neuroinflammation hypothesis," the "cholinergic hypothesis," the "Tau hypothesis," the "infection hypothesis," and the currently prevailing "amyloid-β oligomer hypothesis" (Fan et al., 2020). However, which hypothesis will lead to effective clinical therapeutics remains to be seen. Some hypotheses lost support or have received less enthusiasm, due to limited efficacy with or side effects of target-specific drugs (e.g., acetylcholinesterase inhibitors and NSAIDs).

| Limited modeling for load
LOAD accounts for over 95% of all human AD. The remaining 2%-5% of cases are early-onset AD (EOAD) that are linked to genetic mutation(s) in genes involved in amyloid metabolism, such as Amyloid Precursor Protein (APP), Presenilin 1 (PSEN1), and Presenilin 2 (PSEN2). With insufficient knowledge in modeling LOAD, most preclinical AD drug R&D has employed rodent EOAD models. However, most LOAD patients do not carry mutation(s) in EOAD genes (Carmona, Hardy, & Guerreiro, 2018;Giri, Zhang, & Lü, 2016), suggesting other, and likely multiple, causes influencing LOAD pathology, such as APOE-associated transports. Many AD drug candidates that showed efficacy in EOAD mice have failed in clinical trials. These findings led to uncertainty concerning EOAD models for proper target representation in human AD drug R&D. For example, although many of the existing EOAD mice models display signs of microgliosis and astrogliosis with development of amyloid-β deposits, these mice generally do not show robust neuroinflammation comparable to that in human AD patients (Drummond & Wisniewski, 2017;Jankowsky & Zheng, 2017;Saito & Saido, 2018;Sasaguri et al., 2017). This uncertainty calls for a research model that does not rely on EOAD mutations, while representing some aspects of LOAD (i.e., LOAD model).
demonstrating its potential as a study tool for the LOAD development mechanism and for preclinical AD drug research and development.

K E Y W O R D S
amyloid-β, cohesinopathy, genomic instability, late-onset Alzheimer's disease, mitosis, mouse model, neuroinflammation, Shugoshin1 and other molecular analysis. The International Mouse Phenotyping Consortium (IMPC) database reports an abnormal behavior phenotype in Sgo1 tm1a(EUCOMM)Wtsi allele mice, suggesting the likelihood of AD-like cognitive function/behavioral issues with Sgo1 defects (The IMPC database website: http://www.mouse pheno type.org/data/ genes/ MGI:19196 65#secti on-assoc iations). Previously, spontaneous accumulation of cerebral amyloid-β was not thought to occur in mice, and it was attributed to the protein sequence difference and shorter lifespan (Drummond & Wisniewski, 2017;Sasaguri et al., 2017). The Sgo1 −/+ mouse provided a novel observation counter to the previously prevailing notion. There is the potential for the Sgo1 −/+ mitotic cohesion defect-genomic instability model to serve as a LOAD model portraying aspects of the disease that were not represented in previous models.

| A novel integrated hypothesis for amyloid-β accumulation; the "amyloid-β accumulation cycle"
Amyloid-β accumulation in the Sgo1 −/+ model was originally hypothesized to be linked to three events: (a) aging, (b) mitotic re-entry, and (c) prolonged mitosis [the "Three-hit" hypothesis] . Recently, incorporating evidence suggestive of mitotic dysregulations as a common underlying, if not causal, event for both early-onset and late-onset human AD, an integrated hypothesis for amyloid-β accumulation, the "amyloid-β accumulation cycle," was proposed (Rao, Asch, Carr, & Yamada, 2020). The "amyloid-β accumulation cycle" hypothesis purports the occurrence of vicious cycles of events leading to amyloid-β accumulation, as follows: initial increase in amyloid-β, growth signaling activation and inflammation triggered by the amyloid-β exposure, mitotic re-entry, accumulation of amyloid-β during (quasi-)mitotic state, mitotic catastrophe, and release of more amyloid-β from dead cells into microenvironment, leading to another cycle. The cycle is enabled by unique characteristics of amyloid-β. Amyloid-β is neurotoxic, can activate growth signaling, is pro-inflammatory, can interfere with mitosis (aneuploidogenic), and can be generated during mitosis or in quasimitosis conditions. The cycle may occur at any time during LOAD development, in early and in late stages. Yet, amyloid-β catabolism, which can antagonize the cycle and may decline over age, may be a factor influencing the late-onset accumulation of amyloid-β (Rao et al., 2020).

| Purpose of this study
The observation of amyloid-β accumulation in the aged Sgo1 −/+ model mice raised two major questions; (a) whether the mouse model reflects or recapitulates some other aspects of human LOAD pathology and its development process, such as neuroinflammation and oxidative stress, and (b) in light of the "amyloid-β accumulation cycle" hypothesis, which growth signaling is/are responsible for the amyloid-β accumulation in the model. In the present study, we (a) determined the timing of amyloid-β appearance over age, (b) tested a hypothesis that human AD-relevant age-associated events (i.e., neuroinflammation, misregulations in proteins involved in oxidative stress, mitochondrial energy metabolism, or β-oxidation/peroxisome) are associated with amyloid-β accumulation in the Sgo1 −/+ model, and (c) identified the GSK3-ARC/Arg3.1 and GSK3-Wnt signaling axes as candidates for triggering amyloid-β accumulation in middle age.

| Samples
Twenty-four-month-old animals and brain tissue samples were obtained as described in Rao, Farooqui, Zhang, et al. (2018). Twelvemonth-old animals and brain tissue samples were obtained from a previous study described in Yamada et al., (2015). Fifteen-and eighteen-month-old Sgo1 −/+ brain tissue samples were obtained in the present study (N = 4-5) from both genders (2-5 each). We have not observed significant differences in male/female birth ratio, cancer or other disease development, and Aβ accumulation, between genders in Sgo1 −/+ mice. At sample collection, whole brains were cut in half at the midline. Left hemispheres were fixed in 10% buffered formalin, then were embedded in paraffin and processed onto slides in the CCPDD histopathology core for immunohistochemistry or immunofluorescence. Right hemispheres were flash-frozen in liquid nitrogen and stored at −80°C. For immunoblots and select pathway protein panel analyses, the rear parts of forebrains, including cortex and hippocampus, were excised and used.

| Pathway protein panel analysis
Frozen mouse brains (mouse cerebrum, including cortex and hippocampus, and excluding olfactory bulb, cerebellum, and medulla) were extracted in RIPA buffer with 250 mM NaCl, with added protease inhibitor cocktail (Sigma-Aldrich) and proteasome inhibitor MG132 10 µM (Sigma-Aldrich). Extracts were cleared with 5000 rpm for 5 min. Protein concentrations of the supernatants were estimated with a Nanodrop spectrophotometer (Thermo Fisher Scientific).

| Quantitative mass spectrometry
The samples were mixed with 100 µl 1% SDS, 20 µl of our Bovine Serum Albumin (BSA) internal standard, mixed, and heated for 15 min. The proteins precipitated with 1 ml acetone. The dried protein pellet was reconstituted in 60 µl Laemmli sample buffer and 20 µl (20 µg) was used to run a short (1.5-cm) SDS-PAGE gel. The gels were fixed and stained. Each sample was cut from the gel as the entire lane and divided into smaller pieces. The gel pieces were washed to remove the Coomassie blue, then reduced, alkylated, and digested overnight with trypsin. The mixture of peptides was extracted from the gel, evaporated to dryness in a SpeedVac, and reconstituted in 200 µl 1% acetic acid for analysis.
The analyses were carried out on a TSQ Quantiva triple quadrupole mass spectrometry system. The HPLC was conducted on an Ultimate 3000 nanoflow system with a 10 cm × 75 µm i.d. C18 reversed-phase capillary column. 5-µl aliquots were injected and the peptide was eluted with a 60-min gradient of acetonitrile in 0.1% formic acid.
The mass spectrometer was operated in the selected reaction monitoring mode. For each protein, the method was developed to measure two ideal peptides. Assays for multiple proteins were bundled together in larger panels. Data were analyzed using the program Skyline to determine the integrated peak area of the appropriate chromatographic peaks. The response for each protein was calculated as the geometric mean of the peptide areas. These values were normalized to the response for the BSA standard and to the total ion current. The samples were also analyzed on our Thermo QEx system in the LC-full scan MS mode. The total ion current in those analyses is an indication of the amount of material present in the sample and may be useful for normalization. For final analysis, we used total ion current for normalization, as we found some variations among BSA signals. Additional "universal detection" runs, high-resolution accurate mass (HRAM), were also done using our orbitrap system (Thermo Scientific QEx plus), as an additional type of data that can be re-interrogated as needed.
Overall, as above, four groups (12-month-old wild-type, 12-month-old Sgo1 −/+ , 24-month-old wild-type, and 24-month-old Sgo1 −/+ ; N = 5 each) with a total of 20 samples were simultaneously processed for quantification. The amounts of proteins (i.e., representative peptides) in the panels were quantified. Total ion current was used for normalization. An antioxidant protein panel (49 proteins), a mitochondria and energy metabolism panel (47 proteins), and a β-oxidation and peroxisome panel (37 proteins) were analyzed. The panels were 2018 July/August version. Subtracting overlapping proteins in the panels, a total of 130 proteins were analyzed (see Figure S1).
Control wild-type mice did not show signs of amyloid-β, even at 24 months of age. Expression of another major AD marker, p-TAU, was weak if present, when tested with immunoblots with pTAU S404 and S262 antibodies. However, localized accumulation of pTAU S262 in cell bodies was observed with IHC ( Figure 2b). Expression of aging biomarker p21 WAF1/CIP1 , the variants of which are also a risk factor for human AD (Yates et al., 2015), prematurely increased in 15-to-18month-old Sgo1 −/+ mice. The p21 increase was concurrent with Aβ increase. Expression of p21 WAF1/CIP1 remained low in wild-type mice even at 24 months of age. Thus, cerebral amyloid-β accumulation in Sgo1 −/+ mice is late-onset occurring past middle age, progressive with age, and concurrent with p21 WAF1/CIP1 increase.

| The middle-aged Aβ accumulation is concurrent with IFN-γ-mediated neuroinflammation
Neuroinflammation is proposed to be a major disease modifier for AD. Yet, its causal role in AD development remains controversial. We questioned whether IFN-γ expression-mediated neuroinflammation

F I G U R E 3 (Continued)
is a triggering event for Aβ accumulation, and investigated the timing when IFN-γ expression became prominent. The amount of IFN-γ in Sgo1 −/+ brain was low at 12 months of age and increased over time (Figure 4a). We next investigated whether the increase of IFN-γ at 18 months of age is accompanied by activation of microglia.
However, clear signs of microglia congressing around Aβ-positive cells were not obtained (not shown). Iba1, a microglia marker, did not indicate a significant increase (Figure 4b). In 18-month samples, only diffused IFN-γ was seen, and distinct localization was observed only in aged 24-month samples (Figure 4c). Wild-type controls did F I G U R E 4 IFN-γ -mediated neuroinflammation was concurrent with, but did not precede, Aβ accumulation. (a) IFN-γ increased by age 18 months in Sgo1 −/+ . The amount of IFN-γ in Sgo1 −/+ brain was low at 12 months of age, but increased by 18 months of age. Protein amounts, measured by immunoblot and normalized with actin amount, are plotted and compared. (b) Microglia infiltration may not accompany IFN-γ increase at the 12-18 month transition. Microglia marker Iba1 (measured as in (a)  The "amyloid-β accumulation cycle" hypothesis (Rao et al., 2020). The "amyloid-β accumulation cycle" hypothesis purports the occurrence of vicious cycles of events leading to amyloid-β accumulation (see Introduction). Among a few mysteries in the hypothesis is the growth signaling driving the amyloid-β accumulation cycle. (b) Key growth signaling pathways that are misregulated in human AD patients. AKT, AMPK, MAPK, and GSK3 are among the growth signaling misregulated in human AD and proposed to be involved in the disease process. GSK3 targets ARC/Arg3.1 and β-catenin with ubiquitylation-mediated proteolysis. (c) Phosphorylated GSK3 α and β (inactive forms) increased in Aβ-accumulating Sgo1 −/+ . We tested components of the growth signaling in (b). Amounts of pGSK3α (S21) and pGSK3β (S9), inactive forms of GSK3, increased significantly in Aβ-accumulating Sgo1 −/+ , while the total amount of GSK3 showed only a minor change. Consistently, ARC/Arg3.1 amount also significantly increased. pMAPK 42/44 , pAMPK, PCNA, and pTBK1 did not show significant change.
(d) Nuclear accumulation of pGSK3α (S21) in Sgo1 −/+ . Consistent with immunoblots in (c) Aβ-accumulating Sgo1 −/+ showed accumulation of pGSK3α in the nucleus, both in the hippocampus and in the cortex. Age-matched wild-type showed no such pGSK3α accumulation. Enlarged panels for localization details. (e) Cytoplasmic accumulation of pGSK3β (S9) in Sgo1 −/+ . Aβ -accumulating Sgo1 −/+ showed accumulation of pGSK3β in the cytoplasm, both in the hippocampus and in the cortex. Enlarged panels for localization details. (f) ARC/Arg3.1 was accumulated in the nucleo-cytoplasm. ARC/Arg3.1 was accumulated in the nucleo-cytoplasm, in both the hippocampus and the cortex, in Sgo1 −/+ . Enlarged panels for localization details. In wild-type, IHC signals for ARC/Arg3.1 were much weaker, if any. (g) Another GSK3 target β-catenin was enriched in the nuclei of Sgo1 −/+ , indicating Wnt signaling activation. Consistent with GSK3 inactivation, nuclear translocation of β-catenin, a sign of canonical Wnt signaling activation and cell fate toward cell cycle and mitosis, was observed in Sgo1 −/+ as distinct nucleo-cytoplasmic signals in both the hippocampus and the cortex. β-catenin IHC signals in wild-type were weak, if any (enlarged panels) 3.5 | WNT and GSK3α/β signaling were misregulated at the middle age/late middle age transition in Sgo1 −/+ mice Based on the "amyloid-β accumulation cycle" hypothesis ( Figure 5a), we predicted that one or more growth signaling pathways may be activated in amyloid-β-accumulating middle-aged Sgo1 −/+ brains.
Several key growth signaling pathways that are misregulated in human AD patients (Figure 5b) were selected and investigated to determine whether they indicate aberrant activation at the middle age (12 months) to late middle age (18 months) transition in the Sgo1 −/+ mice, at which time Aβ accumulation occurred.
Activated GSK3 leads proteasomal degradation of β-catenin (canonical Wnt signaling) and degradation of ARC/Arg3.1 (Gozdz et al., 2017). ARC/Arg3.1 was increased in AD patients and was suggested to be a causal protein for AD development (Wu et al., 2011). Thus, we tested the GSK3-ARC/Arg3.1 axis. We detected significant in- We also tested activation of the cGAS-STING pathway, which is involved in detection of cytosolic DNA and innate immunity against virus. The cGAS-STING pathway can be activated with genomic instability and generation of fragmented DNA in cytoplasm (Bakhoum et al., 2018), and thus was suspected to be activated in the model.
However, signs of significant activation of the cGAS-STING pathway were not observed with the markers tested (i.e., pTBK1, cGAS, pST-ING, and pIRF3) in Sgo1 −/+ brains (not shown).
Our results suggest that accumulation of Aβ in middle-/late middle-aged Sgo1 −/+ mice is driven by, at least in part, misregulation of the GSK3-ARC/Arg3.1 axis and activation of Wnt signaling. This GSK3 inactivation is quite different from EOAD mouse models with hyper-activated GSK3 (see Section 4).

| Protein panel analysis
Next, we explored other aging-associated factors as contributors to late-onset amyloid-β accumulation in Sgo1 −/+ brains. The analyses included quantitative mass spectrometry-based protein expression panels for 49 antioxidant proteins, 47 mitochondrial energy metabolism proteins, and 37 β-oxidation and peroxisome proteins. Not counting overlapping proteins, the analysis quantified 130 proteins in four groups of mice (12-month-old wild-type and Sgo1 −/+ mice, 24-month-old wild-type and Sgo1 −/+ mice). Figure 6a shows examples of proteins indicating an Sgo1 −/+ -specific increase in 24-month-old brains. Figure 6b shows examples of proteins indicating both an Sgo1 −/+ -specific increase at 24 months and an agedependent increase in Sgo1 −/+ brains (12 months vs. 24 months).

Notably, increases in Phb and
NADH:Ubiquinone Oxidoreductase Core Subunit V1 (NDUFV1) was among the critical hippocampal genes and pathways that might be involved in the pathogenesis of human AD, identified via bioinformatics . Fatty acid-binding protein 3 (Fabp3) is a human AD biomarker in cerebrospinal fluid and in sera (Chiasserini et al., 2017;Höglund et al., 2017). A Phb2 decrease in Sgo1 −/+ mice may additionally contribute to cohesinopathy, as depletion of Phb2 by RNA interference caused premature sister-chromatid separation and mitotic arrest by spindle-checkpoint activation, a near-identical phenotype to that of the Sgo1 defect, indicating a functional similarity of Phb2 with Sgo1 during mitosis (Takata et al., 2007).

| D ISCUSS I ON
The present study identified 15-18 months of age (late middle age) as the age when spontaneous cerebral amyloid-β accumulation became evident in the Sgo1 −/+ model mice. In human LOAD, cerebral amyloid-β also begins to accumulate in middle to late middle age, 15-20 years before the cognitive symptoms of AD manifest. Thus, the Sgo1 −/+ model, carrying no mutations in APP or PSEN1, recapitulates the late-onset and sporadic aspect of amyloid-β accumulation. This identification of amyloid-β accumulation timing helps to establish experimental conditions for using the model for testing an AD drug candidate, especially for disease intervention initiated in middle age. To which degree the internal pathology of Aβ accumulation affects the animals' cognition, behavior, and memory as external LOAD symptoms is a major research interest, which will be addressed in the future studies.
While we investigated potential causes for the late-onset Aβ accumulation, the Sgo1 −/+ model revealed two surprises that are different from conventional EOAD mouse models; (1) GSK3 activation status, and (2) β-catenin and Wnt signaling behavior. GSK3 is thought to be involved in Aβ generation (Phiel, Wilson, Lee, & Klein, 2003). In the 5xFAD EOAD mouse model, both the GSK3α and β isoforms are hyperactive, and GSK3 inhibitor or GSK3α shRNA ameliorated senile plaque formation (Avrahami et al., 2013). In the PS1 APP EOAD mouse model, β-catenin and p-β-catenin amounts are comparable to those in wild-type controls at 6 months of age, yet at 18 months of age, PS1 APP mice displayed a significant decrease in β-catenin.
GSK-3β inhibitory phosphorylation (S-9) showed a marked decrease by 18 months in PS1 APP mice (Jimenez et al., 2011). These results in EOAD mice are consistent with an interpretation that GSK3 activation is a contributor of AD development, and GSK3 inhibition would be beneficial to manage AD.
However, the present study indicated age-associated GSK3α/β inactivation that is concurrent with Wnt activation, ARC/Arg3.1 ac- Among various neuroinflammatory proteins proposed to be involved in human AD (see Figure 3a), INF-λ and TNF-α were co-expressed in brains of Sgo1 −/+ mice in the present study. Several reports suggest the involvement of INF-λ and TNF-α in human AD pathology development. INF-λ and TNF-α levels were higher, as was nitric oxide production, in AD patients in mild and severe stages compared with patients in earlier phases (moderate stage and mild cognitive impairment), indicating progressive increases in INF-λ and TNF-α in human AD patients (Belkhelfa et al., 2014). Bhaskar et al. (2014) showed that activated TNF-α and the c-Jun Kinase (JNK) signaling pathway led neuronal cells to cell cycle progression toward the mitotic cycle, which was followed by neuronal cell death. This sequence of events is consistent with the aforementioned "amyloid-β accumulation cycle" hypothesis (Rao et al., 2020). Mouse primary astrocytes treated with both INF-λ and TNF-α displayed significantly increased levels of astrocytic APP, BACE1 (an APP-Aβ conversion enzyme), and secreted Aβ40, suggesting a role of INF-λ and TNF-α as priming factors for astrocytes to produce amyloid-β (Zhao, O'Connor, & Vassar, 2011). From the present study, targeting INF-λ and/or TNF-α, and assessing the effects on amyloid-β accumulation, has emerged as a new approach of interest.
Another set of results, indicating protein misregulations in Sgo1 −/+ and human AD in common pathways, also suggests the utility of Sgo1 −/+ mice as a study model for LOAD development. A critical point in interpreting the protein panel data is whether the misregulation is causal to AD, or is compensatory/antagonizing to AD development. As 24-month-old Sgo1 −/+ mice show amyloid-β accumulation colocalizing with p-TAU, but not extensive neurofibrillary tangles or neurodegeneration Rao, Farooqui, Zhang, et al., 2018), we suspect that the Sgo1 −/+ mouse model represents a relatively early phase in LOAD development, and speculate that many of the misregulations are compensatory.
Itemized tests and validation will be needed.
If the notion that the Sgo1 −/+ model represents an early phase of LOAD is correct, the model may be useful in two ways: (a) to test AD intervention drug candidates and assess them with amyloid-β reduction, and (b) to challenge the mouse with AD facilitator candidates and validate the "facilitator" with assessments of whether advanced AD pathology (i.e., Aβ plaques and neurofibrillary tangles [NFT]) would develop. Alternatively, if the mild plaque/NFT pathology in current Sgo1 −/+ model is due to the differences in human and mouse protein structures, developing a "humanized" Sgo1 −/+ model may yield a model that indeed recapitulates "plaques and tangles." Such a model would be a highly LOAD-relevant research tool.
Elucidating the mechanism by which amyloid-β starts accumulating in brains in non-symptomatic early phases would reveal effective interventions for LOAD. In the present study, we observed accumulation of amyloid-β and its co-localization with neuroinflammatory markers, specifically in the aged Sgo1 −/+ model mice and not in agematched control mice. The results suggest new possible scenarios occurring at an early stage of amyloid-β accumulation; (a) amyloid-β accumulation during prolonged mitosis (Rao et al., 2020;Rao, Farooqui, Zhang, et al., 2018) triggers inflammation markers, possibly as a part of mitotic catastrophe; and, alternatively, (b) cells with accumulated inflammation markers go through prolonged mitosis and accumulate amyloid-β (Rao et al., 2020;Rao, Farooqui, Zhang, et al., 2018).
Overall, our present findings (a) caution that the use of GSK3 inhibitors in middle age may be a potential facilitator of amyloid-β accumulation, (b) further add to the body of knowledge of the Sgo1 −/+ model's similarities to human AD, suggesting its potential as an animal model of spontaneous amyloid-β accumulation and LOAD, and (c) support the model to claim a unique niche among existing and prospective AD research models.

ACK N OWLED G EM ENTS
We thank Ms. Kathy Kyler for editorial aid, and Ms. Elizabeth Cambron and Ms. McCoy Taylor for administrative aid. We thank Dr. Michael Kinter in Oklahoma Medical Research Foundation (OMRF) multiplex protein quantification core facility for proteomic panel analysis.

CO N FLI C T S O F I NTE R E S T
The authors declare no conflicts of interest.

AUTH O R S ' CO NTR I B UTI O N S
H.Y. Yamada contributed to all aspects of the project. M. Farooqui and A. Madhavaram contributed key data generation. Y. Zhang contributed animal maintenance and sample collection. C.V. Rao and A.S. Asch provided material support and intellectual input.

E TH I C A L A PPROVA L
Not applicable for human subjects. Animal usage has been reviewed and approved by the OUHSC IACUC.

DATA AVA I L A B I L I T Y S TAT E M E N T
Raw data are available upon request from the corresponding author(s). Material transfer is subjected to negotiation with the PI and OUHSC. A certain use of Sgo1 −/+ mouse is protected by intellectual property rights of the University of Oklahoma.