Gallic Acid is a D ual α/β -Secretase Modulator that Reverses Cognitive Impairment and Remediates Pathology in Alzheimer Mice

Several


Abstract
Several plant-derived compounds have demonstrated efficacy in pre-clinical Alzheimer's disease (AD) rodent models. Each of these compounds share a gallic acid (GA) moiety, and initial assays on this isolated molecule indicated that it might be responsible for the therapeutic benefits observed. To test this hypothesis in a more physiologically relevant setting, we investigated the effect of GA in the mutant human amyloid β-protein precursor/presenilin 1 (APP/PS1) transgenic AD mouse model. Beginning at 12 months, we orally administered GA (20 mg/kg) or vehicle once daily for 6 months to APP/PS1 mice that have accelerated Alzheimer-like pathology. At 18 months of age, GA therapy reversed impaired learning and memory as compared with vehicle and did not alter behavior in wild-type littermates. GA-treated APP/PS1 mice had mitigated cerebral amyloidosis, including brain parenchymal and cerebral vascular β-amyloid deposits, and decreased cerebral amyloid β-proteins. Beneficial effects co-occurred with reduced amyloidogenic and elevated nonamyloidogenic APP processing. Further, brain inflammation, gliosis, and oxidative stress were alleviated. We show that GA simultaneously elevates α-and reduces β-secretase activity, inhibits neuroinflammation and stabilizes brain oxidative stress in a pre-clinical mouse model of AD. We further demonstrate that GA increases abundance of the ADAM10 proprotein convertase furin and activates ADAM10, directly inhibits BACE1 activity but does not alter Adam10 or Bace1 transcription. Thus, our data reveal novel posttranslational mechanisms for GA. We suggest further examination of GA supplementation in humans will shed light on the exciting therapeutic potential of this molecule.
Alzheimer's disease (AD) 3 irreversibly affects memory and cognition in the elderly, beginning with brain changes decades before dementia.
While AD etiology is largely unknown, it has been suggested that a combination of genetic factors, accumulation of amyloid β-protein (Aβ) and neurofibrillary tangles, neuroinflammation, oxidative stress, mitochondrial dysfunction, and synaptic/neuronal loss conspires in the onset and progression of the disease (1).
Designer drugs currently available for AD have only mild symptomatic benefits (i.e., acetylcholinesterase inhibitors or N-methyl-Daspartate receptor antagonists), and do not modify disease. Further, such drugs typically incur unwanted side-effects, and chronic usage can bring about drug resistance. In this regard, "nutraceuticals" (i.e., compounds that are naturally present in the diet) typically have few if any side effects, are well-tolerated, and do not develop resistance.
These intriguing observations raised the question of whether GA might be the common nutraceutical moiety that mediates beneficial effects in the AD context.
GA is an aromatic carboxylic acid that belongs to the hydrolysable tannin family (10). GA has a molecular weight of 170.12 g/mol and a rational formula, C6H2(OH)3COOH.
The compound is an endogenous product of plants, and is distributed across different families of the vegetable kingdom (i.e., Anacardiaceae, Fabaceae, and Myrtaceae) as well as fungi of the genus Termitomyces (11). GA has garnered attention for having anti-oxidant (12)(13), antiinflammatory (9), anti-cancer (14), anti-viral (15), anti-melanogenic (12), and anti-ulcerogenic (16) properties. Further, it has been reported that GA has anti-anxiety (17), anti-depressant (18), anti-epileptic (19), and anti-dementia (20) effects in animal models. Due to potent anti-oxidant properties, GA and its derivatives are widely used as natural preservatives in food and beverages (11). In this report, we tested whether 6 months of oral GA therapy impacts AD-relevant phenotypes in the mutant human amyloid βprotein precursor/presenilin 1 (APP/PS1) transgenic AD mouse model. Mori et al. 3 52%) in the training phase of the test (Fig. 1A). In the retention phase, one-way analysis of variance (ANOVA) followed by post hoc comparisons disclosed statistically significant differences on recognition index between APP/PS1-V mice and the other three mouse groups ( Fig. 1B; **, p < 0.01 for APP/PS1-V versus all other mouse groups). GA-treated APP/PS1 mice had significantly increased novel object exploration frequency by 61% versus APP/PS1-V mice (50%). Notably, GA therapy completely reversed episodic memory impairment, which did not significantly differ from any of the WT mouse groups ( Fig. 1B; p > 0.05).
Next, we tested exploratory activity and spatial working memory in the alternation Ymaze task. One-way ANOVA followed by post hoc testing revealed significant differences for total arm entries between APP/PS1-V mice and the other three mouse groups ( Fig. 1C; ***, p < 0.001).
Mice spontaneously alternate arms in the Ymaze, such that they enter the three arms in sequence more often than by chance (50%, see dotted line in Fig. 1D); this behavioral phenotype is often interpreted as a spatial working memory index (21). As predicted, APP/PS1-V mice showed less tendency to alternate compared with nontransgenic littermates. One-way ANOVA followed by post hoc testing disclosed significant differences on Y-maze spontaneous alternation between APP/PS1-V mice and the other three mouse groups ( Fig. 1D; **, p < 0.01). Again, APP/PS1-GA mice did not significantly differ from any of the WT mouse groups. Therefore, 6 months of oral GA therapy completely restored defective spatial working memory in the alternation Y-maze.
Lastly, we tested hippocampus-dependent spatial reference learning and memory in the RAWM.
On day 1, overall ANOVA demonstrated main effects of block and genotype (p ≤ 0.001 for errors and escape latency). APP/PS1-V mice tended toward increased numbers of errors and longer escape time to reach the visible or hidden platform locations compared with the other three mouse groups. On day 2, overall ANOVA revealed main effects of block, genotype, and treatment (p < 0.001 for both errors and escape latency). Repeatedmeasures ANOVA followed by post hoc testing showed statistically significant differences between APP/PS1-V mice and the other three mouse groups ( Fig. 1E; **, p < 0.01 for errors and *, p < 0.05 for escape latency). APP/PS1-V mice had more errors and extended escape latencies to touch the visible or hidden platforms compared with the other tree mouse groups; yet, APP/PS1-GA mice finished the task with significantly less errors and shorter escape times, and their behavior did not significantly differ from wither WT mouse group (p > 0.05). Remarkably, there were no significant betweengroups differences (p > 0.05) on swim speed, nor did we observe thigmotaxis (characteristic of extreme anxiety-like behavior) in any of the mice tested. We can therefore reasonably conclude negligible impact of motivation, locomotor impairment, or anxiety to RAWM testing.
Importantly, we addressed sex as a biological variable by including it as a categorical covariate in multiple ANOVA models, which did not show significant main effects or interactive terms with sex (p > 0.05).
Further, we stratified all analyses by sex and did not find any significant differences (p > 0.05).
Collectively, these cognitive/behavioral tests show that 6 months of oral GA therapy completely remediates APP/PS1 transgene-related spatial reference learning and 4 memory impairment.
To verify whether reduced cerebral -amyloid burden was specific to a particular plaque type or was more general, we performed morphometric analysis by blindly assigning -amyloid plaques to one of three mutually exclusive categories based on maximum diameter: small (< 25 m); medium (between 25 and 50 m); or large (> 50 m). GA therapy consistently and significantly reduced all three plaque sizes: small (20-44%), medium (32-36%), and large (43-63%) (Fig. 2, C-E; *, p < 0.05; ***, p < 0.001). This effect was also independent of sex (data not shown).
Adding to brain parenchymal deposition of βamyloid as senile plaques, 83% of AD patients manifest cerebral vascular β-amyloid deposits known as cerebral amyloid angiopathy (CAA) (29). In this regard, we blindly counted Aβ antibody (4G8)-stained cerebral vascular βamyloid deposits in walls of penetrating arteries at the pial surface in RSC and EC regions, and in small arteries at the hippocampal fissure and brachium of the superior colliculus in the hippocampal area. APP/PS1 mice that received GA had significant reductions in mean numbers of CAA deposits in all three brain regions examined (20-29%; Fig. 3, A and B; *, p < 0.05; **, p < 0.01; ***, p < 0.001).

Oral GA therapy has dual effects on α-and βsecretases
Mitigation of cerebral amyloid pathology by long-term GA oral treatment could be owed to 1) shifting equilibrium toward brain-to-blood Aβ efflux (30), 2) decreasing expression of APP or PS1 transgenes, or 3) modulating APP cleavage. To begin addressing these possibilities, we assayed plasma A1-40 and A1-42 species in peripheral blood samples from each APP/PS1 mouse group, but did not detect between-groups differences (data not shown).
We then compared transgene-derived APP or PS1 proteins in brain homogenates from GA-and vehicletreated APP/PS1 mice by Western blotting, and found similar band densities between-groups, showing that long-term GA oral therapy does not alter transgene-derived APP (Fig. 4A) or PS1 (data not shown) proteins.
Collectively, our biochemical results suggest that GA therapy exerts dual activities on elevating nonamyloidogenic and reducing amyloidogenic APP processing. We moved on to examine a disintegrin and metalloproteinase domain-containing protein 10 (ADAM10, an αsecretase candidate) and β-site APP cleaving enzyme 1 (BACE1, β-secretase) proteins in brain homogenates from each group of APP/PS1 mice by Western blotting. Both precursor ADAM10 (pADAM10, 90-kDa band) and mature (active) ADAM10 (mADAM10, 68-kDa band) protein expressions were significantly increased in brain homogenates from GA-treated mice (Fig. 5, A-C; ***, p < 0.001 for pADAM10 and mADAM10). ADAM10 is regulated by the proprotein convertase, furin, which activates the enzyme by cleaving the zymogen into the mature form (31). Therefore, we investigated furin expression in brain homogenates from each group of APP/PS1 mice by Western blotting. Results showed that furin protein abundance was significantly increased in brains from GA-compared with vehicle-treated mice (Fig. 5, A and D; ***, p < 0.001). By contrast, BACE1 protein expression was reduced in brain homogenates from GAversus vehicle-treated APP/PS1 mice (Fig. 5, A and E; **, p < 0.01, APP/PS1-GA compared with APP/PS1-V mice). To assess if GA altered transcription of Adam10 or Bace1 in the brain, we assayed both mRNA species by quantitative real-time PCR (Q-PCR). We did not observe any between-groups differences (Fig. 6, A and B), suggesting that GA functions at the posttranscriptional level.
To test whether GA therapy impacted secretase activity, we assayed α-and β-secretase enzymatic kinetics in brain homogenates from APP/PS1-GA versus APP/PS1-V mice. Repeated-measures ANOVA showed that αsecretase activity kinetics were significantly accelerated while kinetics for β-secretase activity were reduced in brain homogenates from GAversus vehicle-treated APP/PS1 mice (Fig. 6, C and D; ***, p < 0.001). Therefore, GA therapy exerts dual effects on α-and β-secretases by modulating protein levels and enzymatic activity of both secretases.
To assess whether GA directly inhibited βsecretase, we assayed GA dose-response in a cellfree β-secretase activity inhibitor assay. The result was positive, as GA dose-dependently inhibited β-secretase enzymatic activity, with statistical significance at the 6.25 and 12.5 μM doses ( Fig. 6E; *, p < 0.05). These results support that GA therapy upregulates α-secretase and downregulates β-secretase activity, with a net positive shift toward nonamyloidogenic APP processing.

GA attenuates neuroinflammation and oxidative stress in APP/PS1 brains
GA reportedly has anti-inflammatory and anti-oxidant properties (9,(12)(13). Therefore, we examined putative effects of GA on neuroinflammatory processes and oxidative stress in APP/PS1 brains. We began this line of investigation by elucidating cerebral β-amyloid deposit-associated gliosis (i.e., astrocytosis and microgliosis) by quantitative immunohistochemical analyses.
We observed positive immunostaining for the astrocytic activation marker, glial fibrillary acidic protein (GFAP), and the structural marker of activated microglia, ionized calcium-binding adapter molecule 1 (Iba1), in glial somata and processes. Numerous minute GFAP-positive astrocytic processes were distributed between neurons, and GFAP expression was co-localized with dystrophic neurites near β-amyloid deposits (Fig.  7A).
We also investigated protein expression of two key brain oxidative stress markers: superoxide dismutase 1 (SOD1) and glutathione peroxidase 1 (GPx1). Western blots of brain homogenates from GA-treated APP/PS1 mice showed that both SOD1 and GPx1 proteins were decreased compared with APP/PS1-V brains (Fig.  8A), and densitometry confirmed statistically significant reductions (Fig. 8, B and C; ***, p < 0.001). Amelioration of oxidative stress was complete, because abundance of either protein did not significantly differ from either WT mouse group (p > 0.05). These effects were sexindependent (data not shown). Collectively, these data suggest that 6-month oral GA therapy alleviates neuroinflammation and stabilizes oxidative stress in APP/PS1 brains.

Discussion
As natural dietary compounds present throughout evolution, nutraceuticals are well-tolerated, either alone or in combination with other therapies. GA is a plant-derived nutraceutical that has anti-inflammatory and anti-oxidant properties (11), and we were specifically interested in GA because this moiety is present in other nutraceuticals that we have shown to be therapeutic in preclinical AD model mice (2)(3)(4)(5). We therefore tested if administering oral GA therapy for 6 months to 12-month-old APP/PS1 mice modified cognitive function and AD-like pathology.
Results showed complete mitigation of impaired learning and memory, reduced cerebral amyloidosis, and attenuated neuroinflammation and oxidative stress. These beneficial effects occurred with increased ADAM10 and reduced BACE1 protein abundance and enzymatic activity, beneficially shifting the balance toward nonamyloidogenic APP metabolism. When taken together, these results provide preclinical evidence supporting oral GA therapy as an AD prophylaxis.
According to the Registry of Toxic Effects of Chemical Substances (RTECS), the GA acute intraperitoneal lethal dose for 50% survival (LD50) is 4,300 mg/kg in the mouse. The noobserved-adverse-effect level (NOAEL) for mouse oral GA is 5,000 mg/kg (32). Tolerable daily intake (TDI) for humans can be extrapolated from rodent NOAEL threshold data by dividing the NOAEL by the uncertainty factor (i.e., interspecies and individual variation) (33)(34). Assuming a default uncertainty factor of 100 (34), the 5,000 mg/kg NOAEL yields human TDI for GA of 50 mg/kg, or 3,000 mg for a 60 kg person.
In the present study, we orally administered 20 mg/kg/day to mice.
To extrapolate this mouse dose to humans, we multiply 20 mg/kg by the mouse factor of 3, and then divide by the human factor of 37 (35). This calculation results in a GA human equivalent dose of 1.622 mg/kg, which equates to 97.32 mg of GA for a 60 kg person. The dose we gave to mice is orders of magnitude below tolerable levels, and there was no morbidity or mortality observed in this study. Furthermore, we did not observe pathological findings in any major organs at autopsy, including the skin, hair, brain, eye, ear, lung, heart, liver, digestive tract, or kidney. Even so, it remains possible that long-term upregulation of ADAM10 and/or 7 downregulation of BACE1 might lead to unwanted side effects. Therefore, investigating the toxicity profile of long-term GA treatment in humans is warranted.
In terms of bioavailability, GA oral administration produces maximum absorption at 60 min for rodents (36) and 87 min for humans (37). A dual mechanism has been suggested for gastric absorption of GA: rapid for intact GA and slow for conjugated derivatives (i.e., 4-Omethylgallic acid) (38). GA has low molecular weight and have been shown to cross the bloodbrain-barrier in rodents (39).
Interestingly, newly generated Aβ species enter into systemic equilibrium between soluble and deposited forms (40).
Aβ lowering strategies, including secretase targeting, have become essential therapeutic approaches in AD. In this report, we show that GA therapy provides beneficial effects on ameliorating cerebral amyloid pathology including parenchymal and cerebral vascular βamyloid deposits, concomitant with decreasing Aβ species. We show that nonamyloidogenic sAPP-α protein expression is significantly increased in brain homogenates from GA-treated APP/PS1 mice; whereas amyloidogenic sAPP-β and β-CTFs (i.e., C99 and P-C99) are significantly decreased. These data suggest that GA dually impacts both amyloidogenic and nonamyloidogenic pathways, with a net positive effect of shifting the balance toward reducing Aβ production.
In this report, we show that GA is a dual modulator of α/β-secretase activity. Specifically, our data support that GA enhances ADAM10 and inhibits BACE1 activity, with a net positive effect to promote nonamyloidogenic APP processing. Importantly, GA did not alter brain transcription of Adam10 and Bace1, suggesting a post-transcriptional mode of action. Since GA therapy significantly elevated mature (active) ADAM10 in APP/PS1 brains, we hypothesized a post-translational mechanism. ADAM10 is produced in an inactive form, and the proprotein convertase, furin, cleaves the ADAM10 zymogen to activate it (31). In this regard, we show increased protein expression of furin in brain homogenates from GA-treated mice.
Increased brain furin abundance is associated with elevated mADAM10. Unexpectedly though, pADAM10 is also increased in the brain following GA therapy. We previously observed increases in both forms of ADAM10 after EGCG treatment (3), and a similar positive feedback loop may exist with GA, whereby ADAM10 cleavage promotes pADAM10 protein expression. Further, we conducted experiments in a cell-free β-secretase activity assay and found that GA directly inhibits BACE1 enzymatic activity post-translationally.
We further show that elevation of active ADAM10 correlates with increased α-secretase APP cleavage and sAPP-α protein abundance in APP/PS1 brains. The sAPP-α generated by αsecretase APP cleavage has both neurotrophic and neuroprotective properties (45)(46) and has been shown to enhance long-term potentiation (47). Interestingly, our previous study showed that sAPP-α is capable of inhibiting BACE1 though physical interactions (48). In support, others have shown that sAPP-α can autoregulate APP metabolism (49-51). Therefore, it remains possible that GA promotion of α-secretase could indirectly inhibit β-secretase activity.

8
While there is evidence linking cognitive decline with cerebral amyloid pathology in AD patients (52-53), a subset of elderly without cognitive impairment have striking AD pathology including 'senile' plaques and/or neurofibrillary tangles (54-55).
When considering preclinical mouse models of AD, accumulating evidence has shown that improved cognitive function occurs in parallel with mitigation of cerebral amyloid pathology in a subset of studies (4, 23-27, 56-59), whereas others have pointed out de-coupling of cognitive function from AD-like pathology (60-61). This discrepancy has led to an alternate hypothesis that soluble, oligomeric forms of Aβ might impair cognition due to synaptotoxicity and neurotoxicity (62)(63)(64)(65).
Supporting this, a positive link has been reported between increased Aβ oligomer levels and behavioral impairment in rodent models (57, 61). Accordingly, we show that Aβ oligomers are significantly decreased after oral GA therapy.
Shifting from protective to destructive functions of glial cells may contribute to neurotoxicity and synaptotoxicity in neurodegenerative diseases. In the AD brain, reactive gliosis co-exists in temporal and spatial proximity to β-amyloid plaques, and diffuse βamyloid deposits attract and activate cytokinesecreting microglia, which can in turn activate astrocytes that cooperatively produce cytokines. This self-propagating "cytokine cycle" promotes further release of acute-phase reactants, proinflammatory cytokines, and immunostimulatory molecules, leading to generation of reactive oxygen species that are toxic at supraphysiologic levels.
Our understanding of role that neuroinflammatory processes play in AD progression is becoming clearer. While there is consensus that glial cells ultimately fail to clear β-amyloid deposits (23,(66)(67)(68)(69)(70)(71)(72), it is now appreciated that not all forms of gliosis are deleterious, and this is underscored by welldocumented dichotomous roles of microglia and astrocytes in exacerbating or mitigating AD pathology depending on context (67)(68)(69).
Here, we show that oral GA therapy dampens β-amyloid plaque-associated microgliosis and astrocytosis. One possibility is that GA exerts anti-inflammatory effects that are independent of Aβ lowering.
However, neuroinflammation and cerebral amyloidosis generally correlate in human AD and in mouse models, making it complex if not impossible to tease them apart in vivo.
Antioxidants are essential biomolecules that protect the body against the hazardous effects of oxidative stress and imbalanced redox homeostasis (73). Most oxidative stress occurs in the form of intracellular superoxide anions, singlet oxygen, hydrogen peroxide, and other reactive oxygen species that are generated in mitochondria, the cellular organelle responsible for a diverse array of functions such as metabolism, ATP synthesis, Ca 2+ signaling, cell survival, and growth (74). Oxidative stress has been implicated in AD etiology (75)(76), and GA is a potent anti-oxidant owing to its free radical scavenging properties. For example, GA has electron-donating sites on the benzene ring (i.e., 3-, 4-, and 5-hydroxyl) yielding phenoxyl radical resonance structures conferring stability to this intermediate or even terminating free radical chain reactions altogether. Additionally, GA has a carboxylic acid group that can act as a lipid anchor to protect against lipid peroxidation. Indeed, we show that GA therapy completely stabilizes brain expression of SOD1 and GPx1 to baseline levels in APP/PS1 mice, which did not differ from nontransgenic littermates.
In conclusion, we report that administering 6month oral GA therapy to the APP/PS1 mouse model of cerebral amyloidosis completely remediates behavioral deficits, ameliorates cerebral amyloidosis, and reduces Aβ abundance. GA treatment further alleviates 9 neuroinflammation and stabilizes oxidative stress. If the APP/PS1 mouse model reflects the clinical syndrome, then dietary supplementation with GA represents a viable therapeutic modality for AD.

Ethics statement
All experiments were performed in accordance with the guidelines of the National Institutes of Health, and all animal studies were approved by the Saitama Medical University Institutional Animal Care and Use Committee. Mice were humanely cared for during all experiments, and all efforts were made to minimize suffering.
GA was obtained from Sigma-Aldrich (St. Louis, MO, USA). Twenty mg of GA was resuspended in 25 μl of 100% DMSO and then diluted with distilled water to a final concentration of 0.2% DMSO. The regent was freshly prepared daily prior to each treatment. We randomly assigned APP/PS1 mice to two treatment groups (n = 12 per condition; six each sex): GA (APP/PS1-GA) or vehicle (distilled water containing DMSO at a final concentration of 0.2%; APP/PS1-V).
Additionally, WT littermates received the same two treatments (n = 12 per group; six of each sex) as follows: WT-GA or vehicle (WT-V). To determine if treatment prevented versus reversed kinetics of cerebral amyloid accumulation, we included twenty untreated APP/PS1 mice at 12 months of age (APP/PS1-12M mice, six of each sex) for analysis of β-amyloid pathology. After baseline behavioral assessment just prior to dosing (at 12 months of age), animals were orally treated (via gavage using a 6202 gastric tube with a rounded silicone-coated tip, Fuchigami Kikai, Kyoto, Japan) with GA (20 mg/kg) or vehicle once daily for 6 months. Mice were housed in a specific pathogen-free barrier facility with a 12/12-h light/dark cycle and ad libitum access to food and water.

Behavioral analyses
To assess episodic memory, mice were habituated in a cage for 4 h, and two objects of different shapes were concurrently provided for 10 min. The number of times that the animal explored the familiar object (defined as number of instances where an animal directed its nose 2 cm or less from the object) were counted for the initial 5 min of exposure (training phase). To test memory retention on the following day, one of the familiar objects was replaced with a different shaped novel object and explorations were recorded for 5 min (retention test). The recognition index, taken as a measurement of episodic memory, is reported as frequency (%) of explorations of the novel versus familiar object (78).
To measure exploratory activity and spatial working memory, mice were individually placed in one arm of a radially symmetric Y-maze made of opaque gray acrylic (arms, 40 cm long and 4 cm wide; walls, 30 cm tall), and the sequence of arm entries and total number of entries were counted over a period of 8 min, beginning when the animal first entered the central area. Percent alternation was defined as entries into sequentially different arms on consecutive occasions using the following formula: % alternation = number of alternations / (number of total arm entries -2) × 100% (21).
To assess spatial reference learning and memory, the RAWM test was done over 2 days and consisted of triangular wedges in a circular pool (80 cm diameter) configured to form swim lanes that enclosed a central open space. Mice naïve to the task were placed in the pool and allowed to search for the platform for 60 s. Animals were dropped into a random start arm and allowed to swim until they located and climbed onto the platform (goal) over a period of 60 s. Latency to locate the platform and errors were recorded. Each mouse was given a total of 15 trials. On day 1, the goal alternated between visible and hidden, although on day 2 the goal was always hidden. All data were organized as individual blocks of three trials each. The goal arms remained in the same location for both days, whereas the start arm was randomly altered. All trials were done at the same time of day (± 1 h), during the animals' light phase. To avoid interference with behavioral testing, treatment was carried out 1 h after conclusion of behavioral testing (79).

Brain tissue preparation
At 18 months of age, anesthesia was induced and maintained with isoflurane (1.5% to 2% and then 0.5%).
Mice were euthanized by transcardial perfusion with ice-cold physiological saline containing heparin (10 units/ml). Brains were isolated and quartered (sagittally at the level of the longitudinal fissure of the cerebrum, and then coronally at the level of the anterior commissure).
Left anterior hemispheres were weighed and snap-frozen at -80°C for Western blotting.
Left posterior hemispheres were immersed in 4% paraformaldehyde at 4°C overnight and routinely processed in paraffin.
Right anterior hemispheres were sliced into two pieces, weighed, and snap-frozen at -80°C for α-and βsecretase activity assays.
Right posterior hemispheres were sliced into two pieces. One piece including hippocampus region was weighed and snap-frozen at -80°C for sandwich ELISA. The other piece was weighted and immersed in RNA stabilization solution (RNAlater ® , Qiagen, Valencia, CA, USA) and then snap-frozen at -80°C for Q-PCR analyses.

Immunohistochemistry
Five coronal paraffin sections (per set) were cut with a 100 μm interval and 5 μm thickness spanning bregma -2.92 to -3.64 mm (80). Three sets of five sections were prepared for analyses of β-amyloid plaques, astrocytosis, and microgliosis.
Primary antibodies were as follows: biotinylated anti-Aβ17-24 monoclonal Osaka, Japan). Immunohistochemistry was performed using a Vectastain ABC Elite kit (Vector Laboratories, Burlingame, CA, USA) coupled with the diaminobenzidine reaction, except that the biotinylated secondary antibody step was omitted for the biotinylated Aβ17-24 monoclonal antibody.

Image analysis
Images were acquired and quantified using SimplePCI software (Hamamatsu Photonics, Shizuoka, Japan). Images of five 5-μm sections through each anatomic ROI (i.e., RSC, EC, and H) were captured based on anatomical criteria (80), and we set a threshold optical density that discriminated staining from background. Selection bias was controlled for by analyzing each ROI in its entirety. For Aβ, GFAP, and Iba1 burden analyses, data are reported as the percentage of positive pixels captured divided by the full area captured.
For β-amyloid plaque morphometric analysis, diameters (maximum lengths) of β-amyloid plaques were blindly measured and assigned to one of three mutually exclusive plaque size categories (< 25, 25-50, or > 50 μm). For quantitative analysis of CAA, numbers of Aβ antibody-positive cerebral vessels were blindly counted in each ROI.
Remaining pellets were treated with 5 M guanidine HCl and dissolved by occasional mixing on ice for 30 min and centrifuged at 18,800 × g for 30 min at 4°C. Supernatants were then collected; this is taken as the guanidine HCl-soluble fraction. Aβ1-40 and Aβ1-42 species were separately quantified in each sample in duplicate by sandwich ELISA (IBL, Gunma, Japan) (82).
Aβ oligomers were quantified in the TNE-soluble fraction in duplicate by human amyloid β oligomers (82E1specific) assay (IBL) (83). All samples fell within the linear range of the standard curve.

Western blotting
Brain homogenates were lysed using TissueLyser LT in TBS containing protease inhibitor mixture (Sigma-Aldrich) followed by TNE buffer.
Homogenates were then centrifuged at 18,800 × g for 30 min at 4°C, and supernatants were collected, and aliquots corresponding to 10 μg of total protein were electrophoretically separated using Tris glycine gels. Electrophoresed proteins were transferred to polyvinylidene difluoride membranes (Bio-Rad, Richmond, CA, USA) that were blocked for 1 h at ambient temperature. Membranes were then hybridized for 1 h at ambient temperature with primary antibodies as follows: Membranes were then rinsed and incubated for 1 h at ambient temperature with appropriate horseradish peroxidase-conjugated secondary antibodies.
After additional rinsing, membranes were incubated for 5 min at ambient temperature with enhanced chemiluminescence substrate (SuperSignal West Dura Extended Duration Substrate, Thermo Fisher Scientific), exposed to film, and developed. Western blots were done for each brain (n = 6, three mice each sex per group), and quantitative data were averaged.

The α-and β-secretase activity assays
To determine α-and β-secretase activity in brain homogenates, we used available assay kits according to our published methods (2,4). Briefly, for α-secretase activity assays (SensoLyte® 520 TACE (α-Secretase) Activity Assay Kit, Anaspec, Inc., Fremont, CA), brains were lysed using TissueLyser LT in ice-cold 3x 12 volume of extraction buffer containing 1% NP-40 and centrifuged at 2,000 × g for 15 min. For β-secretase activity assays (β-Secretase Activity Fluorometric Assay Kit, BioVision Inc., Milpitas, CA), brains were lysed using TissueLyser LT in ice-cold 3x volume of extraction buffer and centrifuged at 10,000 × g for 5 min. Subsequently, both supernatants were collected and kept on ice. Appropriate protein amounts of brain homogenate, assay buffer, and substrate were added in duplicate to a 96-well plate and incubated in the dark at 25 o C for α-secretase activity assays and 37 o C for β-secretase activity assays for kinetics analyses.
Following incubation, fluorescence was monitored (490 nm excitation and 520 nm emission for α-secretase activity; 345 nm excitation and 450 nm emission for β-secretase activity) at 25 o C at 30, 45, and 60 min using a SH-9000 microplate fluorimeter with SF6 software (CORONA ELECTRIC, Hitachinaka, Ibaraki, Japan). Background was determined from negative controls (omission of brain homogenate or fluorogenic substrate).

Cell-free β-secretase inhibitor assay
To directly test the effect of GA on βsecretase activity, we used an available assay based on secretase-specific peptides conjugated to DABCYL/EDANS fluorogenic reporter molecules (Cayman Chemical, Ann Arbor, MI) in accordance with the manufacturer's instructions.

Q-PCR
We quantified brain Adam10, Bace1, and βactin mRNA abundance by Q-PCR. Total RNA was extracted using the RNeasy Mini Kit (Qiagen), and first-strand cDNA synthesis was carried out using the QuantiTect Reverse Transcription Kit (Qiagen) in accordance with the manufacturer's instructions. We diluted cDNA 1:1 in H2O and performed Q-PCR analysis for all genes of interest using cDNA-specific TaqMan primer/probe sets (TaqMan Gene Expression Assays, Thermo Fisher Scientific) on an ABI 7500 Fast Real-time PCR instrument (Thermo Fisher Scientific).
Each 20-μl reaction mixture contained 2 μl of cDNA with 1 μl of TaqMan Gene Expression Assay reagent, 10 μl of TaqMan Fast Advanced Master Mix (Thermo Fisher Scientific), and 7 μl of H2O. Thermocycler conditions consisted of: 95°C for 15 s, followed by 40 cycles of 95°C for 1 s and 60°C for 20 s. TaqMan probe/primer sets were as follows: Bace1 (number Mm00478664_m1), Adam10 (number Mm00545742_m1) and βactin (number Mm00607939_s1; used as an internal reference control) (Thermo Fisher Scientific). Samples that were not subjected to reverse transcription were run in parallel as negative controls to rule out genomic DNA contamination, and a no template control was also included for each primer set. The cycle threshold number (CT) method (84) was used to determine relative amounts of initial target cDNA in each sample. Results are expressed relative to vehicle-treated WT mouse brains.

Statistical analysis
Data are presented as means with associated standard deviations. A hierarchical analysis strategy was used in which we first conducted an overall ANOVA (repeated measures was used for behavioral data) to assess significance of main effects and interactive terms. If the model was significant, post hoc testing was done with Tukey's HSD or Dunnett's T3 methods, where appropriateness was determined based on Levene's test for equality of the variance. In instances of multiple mean comparisons, oneway ANOVA was used, followed by post hoc comparison of the means using Bonferroni's or Dunnett's T3 methods (depending on Levene's test for equality of the variance). The α levels were set at 0.05 for all analyses. All analyses were performed using the Statistical Package for the Social Sciences, release 23.0 (IBM SPSS, Armonk, NY, USA).

DATA AVAILABILITY
All data described in this manuscript are contained within the manuscript.

CONFLICT OF INTEREST
The authors declare that they have no conflicts of interest with the contents of this article. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.    We also used N-terminal amyloid-β1-17 monoclonal antibody (mAb 82E1), which detects amyloidogenic APP cleavage fragments including Aβ monomers and oligomers as well as phospho-C99 (P-C99) and nonphospho-C99 (C99). Actin is included as an internal reference control. Densitometry is shown for ratios of sAPP-α or sAPP-β to APP (B-C) and for C-99 or P-C99 to actin (D). E, Abundance of (N) 82E1 Aβ oligomers in detergent-soluble brain homogenates (measured by sandwich ELISA) is shown. Data were obtained from APP/PS1 mice treated with vehicle (APP/PS1-V, n = 12) or GA (APP/PS1-GA, n = 12) for 6 months starting at 12 months of age. Data for B-E are presented as standard deviations of the means. Statistical comparisons for B-C and E are between-groups, and for D, betweengroups for each protein. **, p < 0.01; ***, p < 0.001 for APP/PS1-GA versus APP/PS1-V mice.   A, Representative images were captured from APP/PS1 mice treated with vehicle (APP/PS1-V) or GA (APP/PS1-GA) for 6 months starting at 12 months of age (mouse age at sacrifice = 18 months). Immunohistochemistry for glial fibrillary acidic protein (GFAP) and ionized calciumbinding adapter molecule 1 (Iba1) shows β-amyloid deposit-associated astrocytosis (left) and microgliosis (right) for each group. Brain regions shown include: retrosplenial cortex (RSC, top), hippocampus (H, middle), and entorhinal cortex (EC, bottom). Quantitative image analysis for astrocytosis (B) or microgliosis (C) burden is shown. Brain regions are shown on the x axis. Data for B-C are presented as standard deviations of the means. Statistical comparisons for B-C are within each brain region and between-groups. **, p < 0.01; ***, p < 0.001 for APP/PS1-GA versus APP/PS1-V mice. actin (C). Data for B-C were obtained from APP/PS1 mice treated with vehicle (APP/PS1-V, n = 12) or GA (APP/PS1-GA, n = 12), and also for wild-type littermates that received vehicle (WT-V, n = 12) or GA (WT-GA, n = 12) for 6 months commencing at 12 months of age (mouse age at sacrifice = 18 months). Data for B-C are presented as standard deviations of the means. Statistical comparisons for B-C are between-groups. ***, p < 0.001 for APP/PS1-V versus the other treated mice.