A high‐fat diet exacerbates the Alzheimer's disease pathology in the hippocampus of the App NL−F/NL−F knock‐in mouse model

Abstract Insulin resistance and diabetes mellitus are major risk factors for Alzheimer's disease (AD), and studies with transgenic mouse models of AD have provided supportive evidence with some controversies. To overcome potential artifacts derived from transgenes, we used a knock‐in mouse model, AppNL−F/NL−F , which accumulates Aβ plaques from 6 months of age and shows mild cognitive impairment at 18 months of age, without the overproduction of APP. In the present study, 6‐month‐old male AppNL−F/NL−F and wild‐type mice were fed a regular or high‐fat diet (HFD) for 12 months. HFD treatment caused obesity and impaired glucose tolerance (i.e., T2DM conditions) in both wild‐type and AppNL−F/NL−F mice, but only the latter animals exhibited an impaired cognitive function accompanied by marked increases in both Aβ deposition and microgliosis as well as insulin resistance in the hippocampus. Furthermore, HFD‐fed AppNL−F/NL−F mice exhibited a significant decrease in volume of the granule cell layer in the dentate gyrus and an increased accumulation of 8‐oxoguanine, an oxidized guanine base, in the nuclei of granule cells. Gene expression profiling by microarrays revealed that the populations of the cell types in hippocampus were not significantly different between the two mouse lines, regardless of the diet. In addition, HFD treatment decreased the expression of the Aβ binding protein transthyretin (TTR) in AppNL−F/NL−F mice, suggesting that the depletion of TTR underlies the increased Aβ deposition in the hippocampus of HFD‐fed AppNL−F/NL−F mice.

discovery of AD; however, its etiology is still not fully understood.
Genetic factors such as mutations in the APP, PSEN1, and PSEN2 genes only account for approximately 1% of AD cases, known as early-onset or familial AD (FAD). In the majority of AD cases in which the onset is late or sporadic, aging is the main risk factor, and mounting epidemiological evidence has suggested that people with type-2 diabetes mellitus (T2DM) or impaired glucose tolerance are at an increased risk for the development of AD (Arvanitakis et al., 2004;de la Monte, 2014;Ninomiya, 2019). Pathological studies have also shown a significant association between T2DM-related factors and the neuropathology of AD, such as Aβ plaque formation (Matsuzaki et al., 2010;Peila et al., 2002). The etiology of cognitive dysfunction in AD patients with T2DM is probably multifactorial, but the mechanisms underpinning this association are not yet fully understood.
High-fat diet (HFD) treatment is an established method to induce T2DM in experimental animals and has been applied to many AD mouse models. HFD treatment causes obesity and induces insulin resistance and impaired glucose metabolism in both the periphery and brain (Heydemann, 2016). Studies with AD mouse models have generally been performed using transgenic mice to induce Aβ accumulation and deposition by overexpressing a mutant amyloid precursor protein (APP) transgene, or with a combination of mutant APP and PSEN1 or MAPT transgenes. Some studies with 3xTg-AD mice carrying mutant APP and MAPT transgenes together with a Psen1 knock-in mutation have shown that HFD treatment increases brain Aβ levels with an exacerbated cognitive decline (Barron et al., 2013;Vandal et al., 2014). Other studies with 3xTg-AD mice found an exacerbated cognitive decline without alteration of the Aβ levels (Knight et al., 2014;Sah et al., 2017). Some studies with APP Swe / PS1 deltaE9 mice have also reported that HFD treatment increases the brain levels of Aβ with an exacerbated cognitive decline (Ettcheto et al., 2016;Walker et al., 2017), while others with APP Swe /PS1 deltaE9 mice reported that HFD treatment increases the Aβ load but does not enhance the cognitive decline (Bracko et al., 2020;Yeh et al., 2015). Moreover, it has been reported that an HFD protects the blood-brain barrier in Tg2576 mice carrying the APP Swe transgene (Elhaik Goldman et al., 2018). Thus, there have been some controversies in the results obtained using transgenic AD mouse models.
These contrasting results may be attributed to artificial effects of the transgenes introduced, such as their overexpression and ectopic expression.
A recent study using a novel knock-in mouse model of AD, App NL/NL carrying the Swedish "NL" mutation, showed that chronic HFD treatment does not trigger AD-associated pathological alterations, such as hippocampal amyloidosis, Tau phosphorylation, and cognitive impairment, although mild impairments in both hippocampal long-term potentiation and social memory were observed (Salas et al., 2018). Despite carrying a FAD mutation, this mouse line does not spontaneously develop Aβ deposition. Given the findings from transgenic mouse models of AD mice with HFD treatment, we hypothesized that prior Aβ deposition may be a prerequisite for HFD treatment to exacerbate the development or progression of AD. It is reasonable to assume that T2DM has an impact on amyloidosis since Aβ deposition starts at 10-20 years before the clinical onset of AD (Bateman et al., 2012), and T2DM increases the risk of AD development during this early stage without clinical symptoms of AD (Ninomiya, 2014).
To test this hypothesis, we applied HFD treatment to an App NL−F/NL−F knock-in mouse model of AD, which carries a humanized β-amyloid (Aβ) sequence with two pathogenic mutations: Swedish "NL" and Iberian "F" at the authentic mouse App locus, thereby increasing the production of the pathogenic Aβ without the overproduction of APP. App NL−F/NL−F mice accumulate Aβ plaques from 6 months of age and show very mild cognitive impairment at 18 months of age. The present study provides the first experimental evidence demonstrating that T2DM exacerbates pre-existing AD pathology, and that AD pathology does not induce T2DM in regular diet (RD)-fed App NL−F/NL−F mice. F I G U R E 1 A high-fat diet led to increased body weight, impaired glucose metabolism, and an impaired insulin signaling pathway in both wild-type and App NL−F/NL−F mice. (a) Six-month-old wild-type and App NL−F/NL−F mice were fed either a regular (RD) or high-fat diet (HFD), and their weights were plotted every 4 weeks for a period of 24 weeks (solid line) and then at 58 and 78 weeks of age (dotted line). (b) Intraperitoneal glucose tolerance test at 18 months of age. After 6 h of fasting, mice were injected with glucose (2 g/kg of body weight). Blood glucose levels were then monitored over time. (c) Representative Western blots showing the hippocampal levels of insulin signaling proteins (pIRS1 Y608 , pIRS1 S632/635 , IRS1, INSRβ, pAKT S473 , and AKT) in App NL−F/NL−F and wild-type mice fed an RD or HFD. (d) Relative ratios of pIRS1 Y608 /IRS1, pIRS1 S632/635 /IRS1, and pIRS1 S632/635 /pIRS1 Y608 in the blots using β-actin as a loading control. (e) Quantification of protein levels of INSRβ in the blot using β-actin as a loading control. (f) Quantification of the protein levels of pAKT S473 and AKT and the pAKT S473 / AKT ratio in the blots using β-actin as a loading control. The bar graph shows the protein/β-actin ratio relative to RD-fed wild-type mice. The data were expressed as the mean ± SEM, n = 17 (solid line) or 10 (dotted line) for RD-fed wild-type mice (Wild-type•RD), n = 12 (solid line) or 5 (dotted line) for HFD-fed wild-type mice (Wild-type•HFD), n = 29 (solid line) or 13 (dotted line) for RD-fed App NL−F/NL−F mice (App NL−F/NL−F •RD), and n = 22 (solid line) or 12 (dotted line) for HFD-fed App NL−F/NL−F mice (App NL−F/NL−F •HFD) in (a); n = 4-5 for all groups in (b). n = 3 for all groups in (d-f). The results were statistically analyzed by a MANOVA (p values for effects are shown) with data from 25 to 49 weeks of age in (a) or 0 to 120 min in (b), and a two-way ANOVA followed by post hoc Tukey's Honest Significant Difference (HSD) test was applied to the data for each week, ** p < 0.01 and *** p < 0.001 for App NL−F/NL−F •HFD vs. App NL−F/NL−F •RD; $$ p < 0.01 and $$$ p < 0.001 for App NL−F/NL−F •HFD vs. Wild-type•RD; and # p < 0.05, ## p < 0.01, ### p < 0.001 for Wild-type•HFD vs. Wild-type•RD. NS, not significant. Weeks 58 and 78 in (a) were excluded from the two-way repeated measures ANOVA and instead subjected to a one-way ANOVA with post hoc Tukey's HSD test. The results in (d-f) were statistically analyzed by a two-way ANOVA (p values for each analysis shown) followed by post hoc Tukey's HSD test, * p < 0.05, ** p < 0.01 and *** p < 0.001 2 | RE SULTS

| An HFD altered the metabolic parameters and brain insulin signaling in both wild-type and App NL−F/NL−F mice
To examine whether or not App NL−F/NL−F mice respond to HFD treatment in a similar way to wild-type mice, wild-type and App NL−F/NL−F male mice were fed either an RD or an HFD for 12 months starting at 6 months (25 weeks) of age. As shown in Figure 1a, HFD treatment significantly increased the body weights in both genotypes compared to the RD-fed groups (Tukey's HSD test, p < 0.0001). There were no phenotypical changes between the HFD-fed wild-type and App NL−F/NL−F mice, and their increased body weights were maintained throughout the experiment. Both genotypes of mice fed an HFD exhibited a significantly lower intake of food and water in comparison with those fed an RD ( Figure S1a,b); however, the calorie intake calculated from the amounts of diet consumed was significantly increased in both genotypes of mice fed an HFD. To evaluate the effect of an HFD on their glucose metabolism, we measured the fasting blood glucose levels every other week up to 49 weeks of age. As expected, HFD treatment significantly increased fasting blood glucose levels from the second week of treatment in both genotypes of mice ( Figure   S1c). To examine whether HFD treatment induces a diabetic condition, mice were subjected to an intraperitoneal glucose tolerance test (IPGTT) at 18 months of age. As shown in Figure 1b, HFD treatment caused significantly impaired glucose tolerance, regardless of the genotype, indicating that an HFD increases body weight and disrupts glucose metabolism in both wild-type and App NL−F/NL−F mice.

Subsequently, we assessed the hippocampal insulin signaling by
Western blotting (Figure 1c). Consistent with the peripheral glucose metabolism impairment, we found an increase in basal levels of IRS1 phosphorylation at Ser 632/635 , which negatively regulates the IRS1 function by reducing its association with PI3-kinase, in both wild-type and App NL−F/NL−F mice fed an HFD, while the basal level of IRS1 phosphorylation at Tyr 608 , which generates a docking site for the PI3-kinase, was slightly decreased only in the HFD-fed App NL−F/NL−F mice, thereby significantly increasing the pIRS1 S632/635 /pIRS1 Y608 ratio ( Figure 1d).
The total protein levels of insulin receptor β (INSRβ) and AKT were not significantly altered by the diet in either genotype, but the basal level of AKT phosphorylation at Ser 473 was significantly increased only in the HFD-fed App NL−F/NL−F mice (Figure 1e,f). The hyperphosphorylation of AKT is known to increase phosphorylation of serine residues of IRS1 through activation of mTOR pathway (Copps & White, 2012).
These results suggest that HFD induces a more intense hippocampal insulin resistance in App NL−F/NL−F mice than in wild-type mice.

| An HFD impaired the cognitive function only in App NL−F/NL−F mice
We next examined whether HFD treatment impairs the cognitive function by performing a Morris water maze test at 18 months of age. Mice were trained to find a hidden platform underwater for 11 consecutive days and then subjected to a probe test without a platform ( Figure 2a). During training, App NL−F/NL−F mice fed an HFD showed a significantly increased (p < 0.0001) escape latency to the platform in comparison with all other groups, with significant effects of genotype, diet, and interaction between genotype and diet ( Figure 2b). During the probe test at 24 h after the last training, App NL−F/NL−F •HFD mice, but not the other three groups of mice, exhibited no preference for the target quadrant among all quadrants ( Figure 2c), indicating that HFD administration significantly impairs memory retrieval only in App NL−F/NL−F mice. Furthermore, the frequency of virtual platform crossing was decreased, and the time to the target quadrant was increased in App NL−F/NL−F mice fed an HFD compared to all other groups; however, the differences did not reach statistical significance ( Figure S2a,b). The swimming speed did not differ markedly among the groups ( Figure S2c). Next, we examined the levels of pre-and post-synaptic proteins in the hippocampus by Western blotting (Figure 2d). As shown in Figure  These results indicate that an HFD disturbs synaptic integrity only (e-f) Quantification of protein levels in blots using β-actin as a loading control. The bar graph shows the protein/β-actin ratio relative to RD-fed wild-type mice. Data are expressed as the mean ± SEM, n = 11-14 for all groups for (b-c) and n = 3 for (e-f). Statistical analyses for (b) were performed by a two-way repeated measures ANOVA (p values for effects are shown) where *** p < 0.001 for App NL−F/NL−F •HFD vs. all three other groups, followed by post hoc Tukey's HSD test, where $ p < 0.05 for App NL−F/NL−F •HFD vs. Wild-type•RD. A nonparametric comparison with the % time spent in the target quadrant performed using the Steel method (c), where *** p < 0.001. For (e, f), a two-way ANOVA (p values for each analysis shown) was performed followed by post hoc Tukey's HSD test, * p < 0.05, ** p < 0.01 than App NL−F/NL−F mice fed an RD, while the levels of SDS-soluble Aβ were not affected by diet. We also examined the phosphorylated Tau accumulation by immunohistochemistry using an anti-Tau (AT8) antibody. No positive signal was detected in App NL−F/NL−F mice fed an RD or those fed an HFD ( Figure S3b).

| An HFD increased Aβ deposition in App NL−F/NL−F mice
To clarify whether or not HFD treatment affects either APP processing or Aβ formation, we examined the levels of APP and APPderived fragments by Western blotting ( Figure S4a). App NL−F/NL−F mice fed an HFD showed a significant decrease in full-length APP (FL-APP) detected by the anti-APP A4 (22C11) that recognizes Nterminal residues (a.a. 66-81) of APP. Two other antibodies (6E10, APP-CT) also showed that an HFD caused a similar but not significant decrease in FL-APP in App NL−F/NL−F mice. In contrast, an HFD slightly increased FL-APP in wild-type mice ( Figure S4b-d). Humanized Aβ detected by 6E10 was only found in App NL−F/NL−F mice with no diet effect ( Figure S4e). Accordingly, significantly higher levels of β secretase-cleaved CTFβ fragment were detected in App NL−F/NL−F mice than in wild-type mice, and they were slightly increased by an HFD ( Figure S4f). In contrast, slightly lower levels of α secretase-cleaved CTFα were detected in App NL−F/NL−F mice than in wild-type mice, with a slight decrease due to an HFD ( Figure S4g). These results suggest that APP processing by β secretase to generate Aβ may be slightly but not significantly enhanced by an HFD in App NL−F/NL−F mice.
Next, we examined the hippocampal gene expression profiles of all four groups of mice using a microarray analysis of hippocampal RNA and confirmed no significant difference in the expression of genes involved in APP processing and Aβ clearance (Table S1). Taken together, these results showed that an HFD did not significantly alter the APP processing or Aβ clearance in App NL−F/NL−F mice.

| An HFD increased the hippocampal expression of genes involved in glial activation in App NL−F/NL−F mice
Next, we compared the expression of specific marker genes for five major types of brain cells: astrocytes, oligodendrocytes, microglia, neural stem cells/progenitor cells (NSCs/NPCs), and neurons to assess the cell population changes caused by diet and genotype (Table 1). We observed an increased expression of markers related to microglial (C1qa, C1qb, and Cd68) and astrocytic (Gfap) activation in App NL−F/NL−F mice fed an RD or HFD and also in HFD-fed wild-type mice in comparison with RD-fed wild-type mice. App NL−F/NL−F mice fed an HFD exhibited the highest expression of these genes, suggesting that an HFD exacerbates glial activation in the App NL−F/NL−F hippocampus, but has a limited effect in the wild-type hippocampus.
No differences were observed among the 4 groups in the expression of markers related to the three other cell types (neuron, NSC/NPC, and oligodendrocyte) ( Table 1).
To confirm the activated state of glial cells in hippocampus, we performed immunofluorescence microscopy of coronal brain sections using anti-CD68 antibody (Figure 4a

| An HFD increased nuclear accumulation of 8-oxoguannine and altered gene expression in the hippocampus of App NL−F/NL−F mice
Amyloid deposition and glial activation are known to induce oxidative stress, which causes the oxidation of various molecules, including DNA, resulting in cellular dysfunction (Nakabeppu, 2019). This The mean intensities were normalized to Ponceau S staining (SDS sol) or total protein (FA ext). Data expressed as the mean ± SEM, n = 4 for all experiments. Four brain slices per mouse were examined for (b) and (d). The results were statistically analyzed by an unpaired t test, * p < 0.05, ** p < 0.01, and *** p < 0.001 raises a question as to whether an HFD increases oxidative stress and cellular damage in the App NL−F/NL−F hippocampus. To examine the extent of cellular damage in the hippocampus, we performed cresyl violet staining of brain sections from wild-type and App NL−F/NL−F mice fed an RD or HFD (Figure 5a). We found that App NL−F/NL−F mice tended to show a reduced GCL volume in comparison with wildtype mice (Student's t test, p = 0.0484) and that HFD treatment also decreased the volume in both genotypes of mice to some extent.
As a result, HFD-fed App NL−F/NL−F mice exhibited a significantly reduced GCL volume in comparison with RD-fed wild-type mice (2way ANOVA, Tukey's HSD post hoc test, p = 0.0151) (Figure 5b).  Table S3, HFD treatment significantly altered the expression of several genes (±1.5-fold change, minimum raw intensity >100) in the hippocampus. We found that HFD treatment of App NL−F/NL−F mice tended to decrease the expression of Ttr, a gene encoding the amyloid binding protein transthyretin (TTR), to 15% or 19% of the levels seen in wild-type or RD-fed App NL−F/NL−F mice, respectively. However, the change did not reach statistical significance. As a result, we examined the levels of TTR in hippocampus by immunofluorescence microscopy using an anti-TTR antibody. As shown in Figure 6a, HFDfed App NL−F/NL−F mice exhibited the lowest TTR IR in the hippocampus among all groups. Quantification of the TTR IR revealed that HFD treatment of App NL−F/NL−F mice significantly decreased the levels of TTR in both the GCL and the CA1sp zones but not in the MSS zone of the hippocampus (Figure 6b). Quantification of the TTR IR in the cortex also revealed that HFD treatment decreased the cortical levels of TTR in App NL−F/NL−F mice ( Figure S7a,b). In addition, we examined the total hippocampal levels of TTR by Western blotting (Figure 6c).
We confirmed that HFD-fed App NL−F/NL−F mice had the lowest level of hippocampal TTR among all groups, and that this level was significantly lower than in RD-fed or HFD-fed wild-type mice (Figure 6d).
Finally, multi-immunofluorescence microscopy for Aβ and TTR demonstrated the intracellular co-localization of Aβ and TTR in the GCL of RD-fed App NL−F/NL−F mice and its apparent reduction by HFD treatment (Figure 6e).

| DISCUSS ION
The major conclusion of the present study is that chronic HFD treatment caused obesity and impaired glucose tolerance (i.e., in- Salas et al. showed that chronic HFD treatment from 2 to 18 months of age does not trigger any AD pathology in male App NL/NL mice, except for a mild impairment in both hippocampal long-term potentiation and social memory. In contrast, we found that chronic HFD treatment from 6 to 18 months of age does aggravate hippocampal AD pathology in male App NL−F/NL−F mice, with cognitive impairment in an MWM test. In the aforementioned study, App NL/NL mice were fed an HFD containing 60% kcal from fat for 16 months. In contrast, in our study, App NL−F/NL−F mice were fed an HFD containing 40% kcal from fat for 12 months. In both cases, mice developed similar levels of metabolic syndrome or T2DM, namely increased body weight and fasting blood glucose levels, and impaired glucose tolerance. Because in comparison with App NL/NL mice, App NL−F/NL−F mice have an additional Iberian "F" mutation in the App gene, which is known to increase Aβ42 production by altering the γ secretase processing of the C-terminal end of Aβ (Saito et al., 2014), our results clearly indicate that the Iberian "F" mutation is essential to make App NL−F/NL−F mice susceptible to T2DM as a risk factor for the TA B L E 1 The altered expression of marker genes for various brain cell types in the hippocampi of wild-type and App NL−F/NL−F mice fed an RD or HFD

Cell type
Marker gene pathogenesis of AD. App NL/NL mice mostly produce humanized Aβ40, which is detected as both soluble and insoluble forms, due to the Swedish "NL" mutation being upstream of the β secretase processing site of Aβ. However, they do not develop Aβ plaques at all (Saito et al., 2014;Salas et al., 2018). On the other hand, App NL−F/NL−F mice produce similar levels of humanized Aβ but dominantly Aβ42 (Aβ42/ Aβ40 > 5), and thus develop Aβ plaques in both the cortex and hippocampus (Masuda et al., 2016;Saito et al., 2014). Taken  It has been well-documented that an HFD or T2DM interfere with hippocampal functioning (Cordner & Tamashiro, 2015;Kanoski & Davidson, 2011), and App NL/NL mice have also been shown to exhibit mild impairment in hippocampal long-term potentiation after chronic HFD treatment. We therefore examined HFD-induced alter- The affected zone contains niches for synapses from the perforant path and the Schaffer collateral pathway in the hippocampal network; thus, it is pivotal for learning and memory (Neves et al., 2008). Aβ (Kawarabayashi et al., 2001). Because there was no significant change in APP processing toward Aβ formation, HFD treatment likely promoted Aβ aggregation and deposition by the suppression of its clearance. This is consistent with a recent study reporting that HFD increased SDS-insoluble levels of Aβ in the cortex of a transgenic mouse line carrying the Swedish mutation, with a reduced Aβ clearance independent of insulin-degrading enzyme or neprilysin (Wakabayashi et al., 2019). Although soluble prefibrillar Aβ oligomers are classified as the most toxic species (Kayed & Lasagna-Reeves, 2013), plaques can release toxic Aβ intermediates, thus functioning as "reservoirs" of toxic oligomers (Haass & Selkoe, 2007;Thal et al., 2015). It is likely that chronic HFD treatment increases the levels of We found that HFD-fed App NL−F/NL−F mice exhibited a significantly reduced GCL volume with much higher levels of 8-oxoG accumulation in the nuclei of the granule cells in the GCL in comparison with RD-fed mice. The increased accumulation of 8-oxoG, a marker of oxidative stress, in the granule cells, and to a lesser extent in the CA1 pyramidal neurons, suggests that the production of reactive oxygen species in these neurons is enhanced by the increased toxic Aβ accumulation or by the consequence of microgliosis; thus, it may cause hippocampal atrophy without cell death by affecting dendrites stability and architecture as previously reported (Schoenfeld et al., 2017). The granule cells in the GCL or pyramidal cells in the CA1 themselves are likely to produce toxic Aβ because the MSS zone, which contains niches for synapses to their dendrites, accumulates the highest levels of Aβ. Moreover, we found that the expression of TTR in these neurons is significantly decreased in HFD-fed

HFD-fed wild-type RD-fed App NL−F/NL−F HFD-fed App NL−F/NL−F
App NL−F/NL−F mice in comparison with RD-fed mice. TTR is a multifunctional protein that can bind Aβ peptide and suppress its aggregation, and which also promotes its clearance (Ghadami et al., 2020;Vieira & Saraiva, 2014). It has been shown that the hemizygous deletion of the Ttr gene in APP Swe /PS1 deltaE9 mice results in accelerated Aβ deposition (Choi et al., 2007), and that the overexpression of a human TTR transgene was ameliorative in the APP23 mice (Buxbaum et al., 2008). These results strongly suggest that the decreased expression of TTR in the hippocampal neurons of HFD-fed App NL−F/NL−F mice is one of the direct causes of the increased Aβ deposition. We F I G U R E 4 A high-fat diet promoted microglial activation in the hippocampus of App NL−F/NL−F mice. (a) Representative immunofluorescence images of CD68-positive microglial cells in the hippocampus of App NL−F/NL−F and wild-type mice fed an RD or HFD. Scale bar = 200 µm (for full image) and 100 µm (for magnified images), DG, dentate gyrus. (b) Quantification of CD68 intensity in three different zones of the hippocampus shows the increased expression of CD68 in App NL−F/NL−F mice fed an HFD. (c) Z-stack projections of multi-immunofluorescence images for Aβ and CD68 shows that CD68-positive microglia were clustered surrounding Aβ plaques (top). Magnified view of the dotted boxes for CD68 signal (bottom). Scale bar = 50 µm. (d) Western blots showing the hippocampal levels of IL-1β in App NL−F/NL−F and wild-type mice fed an RD or HFD. (e) Quantification of IL-1β levels in blots using β-actin as a loading control. The bar graph shows the IL-1β/β-actin ratio relative to RD-fed wild-type mice. Data are expressed as the mean ± SEM, n = 4 and four brain slices per mouse were examined for (b) and n = 3 for (e). The results were statistically analyzed by a two-way ANOVA (p values for each analysis shown) followed by post hoc Tukey's HSD test, where * p < 0.05 and ** p < 0.01 therefore speculate that decreased TTR levels may be responsible for the HFD-induced reduction in Aβ clearance previously reported by Wakabayashi et al., 2019.
TTR is mainly expressed in the liver and choroid plexus and secreted into the blood or cerebrospinal fluid; furthermore, its levels are known to be decreased in AD patients in comparison with agematched controls (Velayudhan et al., 2012). Recently, it has been shown that TTR is expressed in the neurons in the cortex or hippocampus in both humans and mice (Li & Buxbaum, 2011;Oka et al., 2016), as we observed in the present study. The expression of TTR was reportedly altered in the brain of various mouse models of AD and closely associated with the level of oxidative stress (Oka et al., 2016;Sharma et al., 2019;Stein & Johnson, 2002). 8-OxoG, or its repair reaction, is known to cause epigenetic alterations in the gene expression under oxidative conditions (Ba & Boldogh, 2018), suggesting a functional contribution of 8-oxoG to the altered expression of TTR in the hippocampal neurons.
We recently demonstrated that the increased accumulation of 8-oxoG in the granule cells in the GCL impairs hippocampal neurogenesis, thus inducing hippocampal atrophy and mild cognitive impairment in aged female mice (Haruyama et al., 2019). In the present study, we only examined male mice in order to avoid the strong effects of sex hormones or the estrus cycle; however, it is likely that female App NL−F/NL−F mice may exhibit a much higher susceptibility to HFD; thus, the AD pathology would be more strongly exacerbated because sex-dependent TTR modulation of brain Aβ levels or adult neurogenesis has been reported (Oliveira et al., 2011;Vancamp et al., 2019). Further studies are necessary to better understand the mechanism by which HFD treatment causes alterations in the gene expression, production, or clearance of the Aβ peptide, oxidative stress, and inflammatory responses. In particular, answering the question as to how T2DM affects the interaction of TTR and Aβ in the AD brain will help to establish new strategies for AD treatment.

| HFD treatment and metabolic assessment
To avoid strong effects of sex hormones or the estrus cycle in female mice, we used male App NL−F/NL−F and wild-type mice. At 6 months of age (25 weeks), mice were individually housed and fed ad libitum with either an RD (CA-1, 13.8% of total calories from fat, Clea Japan shows the TTR/β-actin ratio relative to RD-fed wild-type mice. (e) Orthogonal view of multi-immunofluorescence microscopy for Aβ and TTR shows the intracellular TTR signal co-localized with Aβ in App NL−F/NL−F mice fed an RD. In magnified bitmap images from the dotted boxes (bottom), the co-localized red and green pixels are shown in white. The percentage of the co-localized pixels among total TTR pixels is shown in brackets. Scale bar = 20 µm for full images and 10 µm for magnified images. Data are expressed as the mean ± SEM, n = 3 for all experiments, and 3 brain slices per mouse were examined for (b). The results were statistically analyzed by a two-way ANOVA (p values are shown for each analysis) followed by post hoc Tukey's HSD test, * p < 0.05, and ** p < 0.01 Inc., Tokyo, Japan) or HFD (custom diet, 40% of total calories from fat and 0.15% from cholesterol, Oriental Yeast Co., Tokyo, Japan) until the end of the experiment (18 months of age). Using an animal balance (DH-R610N, Shinko Denshi Co., Ltd.), the body weight of all mice was measured and recorded once a week from 25 to 50 weeks of age and then at 58 and 78 weeks of age. The weekly intakes of food and water were calculated by subtracting the remaining amounts of food and water at the end of the week from the amount fed to each mouse at the beginning of the week. Every other week, blood samples were collected from tail after 6 h fasting, and the fasting blood glucose level was measured using a Freestyle Flash glucometer (NIPRO Co., Ltd, Osaka, Japan). At 18 months of age, mice were subjected to an intraperitoneal glucose (2 g/kg) tolerance test (IPGTT). Briefly, mice were fasted for 6 h before the IPGTT and blood samples were collected from the tail at the following time points: before glucose injection (0 min), and 30, 60, 90, and 120 min after glucose injection.

| Morris water maze test
A Morris water maze test was performed at 18 months of age in all groups, as described previously (Haruyama et al., 2019) with some modifications. Detailed procedures can be found online in the Appendix S1.

| Brain sample preparation
After the final behavioral test, mice were sacrificed, and their brains were dissected as previously described (Oka et al., 2016

| Dot blotting of soluble and insoluble Aβ
Dot blotting was performed using a dot blotting assay system (Bio-Dot® Microfiltration Apparatus, Bio-Rad Laboratories, Hercules, CA, USA). The procedure was performed according to the manufacturer's guidelines. For each sample, 2.0 μg of SDS soluble or 2 µl of formic acid-soluble protein diluted in 100 μl of TBS was applied to a PVDF membrane in triplicate. After blotting, the membrane was detached from the apparatus and blocked with 5% skim milk in TBS +0.1% Tween 20 (TBST) for 1 h at room temperature, followed by incubation with mouse anti-human Aβ 82E1 antibody, which recognizes the N-terminal end of humanized Aβ, overnight at 4°C. The next day, the membrane was washed and incubated with appropriate HRP-conjugated secondary antibodies for 1 h at room temperature. The blot was TBST, incubated in luminol HRP substrate (EZWestLumi plus, ATTO, Tokyo, Japan), and imaged on an EZ capture MG (ATTO). The intensity of the dots was measured using ImageJ 1.52 (NIH, Bethesda, MD, USA).

| Western blotting
Western blotting was performed to measure the levels of different proteins in the hippocampus. Detailed procedures can be found online in the Appendix S1.

| Immunofluorescence microscopy and histochemical analyses
Immunofluorescence microscopy and histochemical analyses were performed as previously described (Castillo et al., 2017;Haruyama et al., 2019). Detailed procedures can be found online in the Appendix S1.

| RNA isolation and the microarray analyses
Preparation of RNA and the microarray analysis were performed as previously described (Castillo et al., 2017). Total RNA extracted from frozen hippocampi using Isogen (Nippon Gene, Tokyo, Japan) was subjected to microarray analysis with an Affymetrix Mouse Gene 2.0 ST Array. The generated CEL files were imported into the Transcriptome Analysis Console 4.0 software program (Affymetrix), and gene-level estimates were obtained for all transcript clusters.
All microarray data were deposited in the GEO database (accession number GSE152539).

| Statistical analyses
All statistical analyses were performed using JMP Pro 14.2.0 (SAS Institute Japan Ltd., Tokyo, Japan). All t tests were two-tailed. A multivariate analysis of variance (MANOVA) and two-way ANOVA were used to assess the interaction between factors. When significant interactions were detected, a post hoc Tukey HSD test was used to adjust for multiple comparisons. p values of <0.05 were considered to indicate statistical significance.

ACK N OWLED G M ENTS
This work was partly supported by grants from the Japan Society for the Promotion of Science (grant numbers 22221004, 17H01391 to Y.N.). We thank Dr. Brian Quinn for editing a draft of this manuscript, and Kaoru Nakabeppu and Tomoko Koizumi for their technical assistance. We also thank Daisuke Tsuchimoto for his helpful discussions.
Finally, we would like to pay our gratitude and respects to our late technical assistant, Setsuko Kitamura, for her invaluable contributions to our research.

CO N FLI C T O F I NTE R E S T
None declared.