Effect of high fat diet on phenotype, brain transcriptome and lipidome in Alzheimer’s model mice

We examined the effect of chronic high fat diet (HFD) on amyloid deposition and cognition of 12-months old APP23 mice, and correlated the phenotype to brain transcriptome and lipidome. HFD significantly increased amyloid plaques and worsened cognitive performance compared to mice on normal diet (ND). RNA-seq results revealed that in HFD mice there was an increased expression of genes related to immune response, such as Trem2 and Tyrobp. We found a significant increase of TREM2 immunoreactivity in the cortex in response to HFD, most pronounced in female mice that correlated to the amyloid pathology. Down-regulated by HFD were genes related to neuron projections and synaptic transmission in agreement to the significantly deteriorated neurite morphology and cognition in these mice. To examine the effect of the diet on the brain lipidome, we performed Shotgun Lipidomics. While there was no difference in the total amounts of phospholipids of each class, we revealed that the levels of 24 lipid sub-species in the brain were significantly modulated by HFD. Network visualization of correlated lipids demonstrated overall imbalance with most prominent effect on cardiolipin molecular sub-species. This integrative approach demonstrates that HFD elicits a complex response at molecular, cellular and system levels in the CNS.

effects 8 . Here, using the same AD mouse model we aimed to integrate the effects of HFD on the AD-like phenotype, brain transcriptome and lipidome. We report that HFD significantly worsens cognitive performance in correlation with a decreased expression of genes related to neuronal projections and synaptic transmission. Conversely, HFD increased immune transcriptome in brain and TREM2 immunoreactivity. We identified specific changes in brain lipidome associated with HFD in APP expressing mice by lipidomics assays, suggesting a role for lipid sub-species in AD-like phenotype in mice.

Results
High fat diet aggravates AD-like phenotype of middle-aged APP23 mice. Effect on memory. The goal was to test the effect of chronic feeding (HFD) on the phenotype of middle-aged AD model mice. We used one-year old APP23 mice fed HFD for three months and compared their learning and memory performance to those of mice fed normal diet (ND). We used HFD that supplies 40% of the calories from fat (milk fat and corn oil) compared to 16% fat calories in the ND. We specifically chose to start diet treatment in middle age because in one-year old APP23 mice the amyloid phenotype is present but not excessive, and thus this model is more relevant to human AD pathology at relatively early stages of the disease. As shown on Fig. 1A, weight gain resulting . Statistic by two-way ANOVA. No interaction between trial day and diet; significant main effect of diet (F (1,110) = 11.24, p = 0.0011) and trial day (F (4,110) = 12.38, p < 0.0001). (C) Probe trial of MWM (latency to the target). Student's t-test. (B,C) N = 5-7 mice gender/group. (D-G) Amyloid plaques in cortex and hippocampus of APP23 mice on ND and HFD were analyzed by two-way ANOVA followed by Sidak's multiple comparison test. Representative images (D) and quantification (E) of Aβ deposits (6E10 staining). No interaction between diet and gender and significant effect of diet (F (1,12) = 29.92, p = 0.0001) and gender (F (1,12) = 40.58, p = 0.0001). On the graph is shown the result of Sidak's post-hoc test. Representative images (F) and quantitation (G) of X-34 fibrillary amyloid plaques. No interaction and significant main effect of diet (F (1,12) = 5.32, p = 0.004) and gender (F (1,12) = 12.11, p = 0.003). On the graph is shown the result of Sidak's post-hoc test. For D-G N = 4-5 male, 5 female mice per group. *p < 0.05 and **p < 0.01. from HFD was similar in male and female mice. To examine cognitive performance, we used Morris Water Maze (MWM). There was no significant difference in cognitive performance between male and female mice on either diet ( Supplementary Fig. S1) therefore the results are combined for both genders. As seen from Fig. 1B, HFD significantly affected the acquisition of spatial memory exemplified by the increased escape latency. Memory retention was also significantly reduced as indicated by the latency to enter the target quadrant in the probe trial of MWM (Fig. 1C).
Effect on amyloid deposition. Our previous data demonstrated that HFD increases amyloid deposition in APP23 mice 8 . To confirm this, total Aβ plaques (including diffuse and fibrillar) were visualized using 6E10 anti-Aβ antibody (Fig. 1D) and compact amyloid plaques assessed by X-34 staining (Fig. 1F). As visible on Fig. 1E and G, there was a significant diet and gender effect on 6E10-positive Aβ plaques as well as on X-34 labeling compact amyloid plaques, respectively. While HFD affected males and females similarly the females had significantly more amyloid in each diet group. The effect of diet on plaques was not a result of an increase in full length Aβ precursor protein (APP), as it was unaffected by HFD ( Supplementary Fig. S4A and B). We also observed a significant decrease in ABCA1 protein level following HFD whereas APOE protein level was unchanged ( Supplementary  Fig. S4A and B). Collectively these results confirm our previous study 8 and we conclude that in this model HFD aggravates AD-like phenotype.
HFD affects the expression of genes involved in immune response, and those related to neuron differentiation and transcription. To examine if HFD affects transcriptome, we performed RNA-seq using RNA extracted from cortices of the same APP23 mice. As seen on Fig. 2A, we found similar numbers of differentially expressed genes: in females -597 up-(red) and 632 down-(blue) regulated genes were affected by HFD, and in males -631 up-and 654 down-regulated genes using a p-value of p < 0.05 as a cut off. Importantly, HFD induced similar changes in Gene Ontology (GO) categories in both genders (Supplementary Fig. S2A and B). GO analysis revealed that Biological Process categories most significantly up-regulated by the diet in both genders are "Immune system process" and "Innate immune response". Most significant down-regulated category in both genders was "Transcription" and "Neuron projections" in males, and "Nervous system development" and "Neuron projection development" in females (Supplementary Fig. S2A and B).
Next, to identify commonly affected genes in males and females, we compared the lists of differentially affected transcripts in both genders. We found 301 common genes which were significantly up-regulated and 165 that were significantly down-regulated by HFD in males and females. To examine the predicted activation state of diseases and function categories, we used Ingenuity Pathway Analysis (IPA) software and calculated "z-score". The results shown on Fig. 2B demonstrate that functions related to immune response, phagocytosis, migration of macrophages, apoptosis and inflammation had an increased activation state in the HFD group. In contrast, memory, transcription, cell cycle progression and quantity of neurons were among the most significant down-regulated functions in the HFD group.
Validation of RNA-seq results. For further validation, we chose genes from categories commonly affected in both genders namely "Immune response", "Neuron differentiation" and "Regulation of transcription". The genes were assessed by qPCR, Western Blot (WB), and immunohistochemistry (IHC). Because the fold change in males and females was similar, we combined the results for both genders. On Fig. 2C and D are shown RNA-seq and qPCR results, respectively, for genes related to immune response, including Trem2, Treml2, Tyrobp, Cx3cr1, Ccl3, Clec7a and complement components. As visible from Fig. 2B, another up-regulated category in mice on HFD was apoptosis that is interconnected with immune response 21,22 . On Supplementary Fig. S3 are shown RNA-seq results demonstrating statistically significantly increased genes by HFD, related to apoptosis -such as Aim2, Sp110 and Stat1.
The effect of HFD on mRNA expression of genes related to neuronal differentiation and regulation of transcription is shown on Fig. 2E (RNA-seq) and F (qPCR): included are genes related to neurogenesis (Gnrh1 and Dlk1), transcription factors (Klf4 and Nr4a3/NOR1), early growth response genes (Egr2 and Fosl2), and neurotransmitter release (Slc32a1/VGat). Dlk1 (delta-like homologue 1) is an atypical NOTCH ligand that is important for adult neurogenesis 23 . Klf4 (Krüppel-like factor-4) is a transcriptional factor expressed in neural stem cells 24 and has a role in neuronal differentiation 25 . To validate RNA-seq results we analyzed the level of proteins, coded by these genes, using WB. As seen on Supplementary Fig. S4C-F, DLK1 and KLF4 protein level was significantly decreased in mice on HFD.
We examined also protein level of vesicular GABA transporter (Slc32a1/VGat), which has a role in neurotransmitter release. Slc32a1/VGat is expressed in inhibitory GABAergic neurons and affects cognitive performance 26 . IHC results demonstrated a significant decrease of VGAT staining ( Supplementary Fig. S5A-D). In contrast, HFD had no effect on mRNA and protein level of vesicular glutamate transporter Slc17a7/VGut, also involved in neurotransmitter release 27 (Supplementary Fig. S5E-G). Altogether these results suggest selective HFD effect on genes regulating inhibitory and excitatory neurotransmitter release.

HFD increases TREM2 protein level in correlation with Aβ deposition. The importance of TREM2
as a risk factor for AD prompted further evaluation of its protein level by IHC. As shown on Fig. 3A and B, we observed a significant increase in TREM2 staining in cortex of females fed HFD compared to those on ND, and a trend toward increase in males but the difference was not significant. We also found that Trem2 mRNA (Fig. 3C) as well as TREM2 staining (Fig. 3D) were highly correlated to the levels of 6E10-positive Aβ plaques. This data suggests that amyloid load affects Trem2 mRNA expression and the corresponding TREM2 protein, and potentially accounts for the higher level of TREM2 in females that are more afflicted by the amyloid pathology. . Positive z-score (red) indicates a predicted upregulated function and negative z-score (blue) is a down-regulated function. (C and D) Significantly changed genes associated with "Immune response" and "Inflammation" in RNA-seq datasets (C) and qPCR validation (D). (E and F) Genes related to "Neuron Differentiation" and "Regulation of transcription" in RNA-seq (E) and qPCR validation assays (F). Fold change (FC) of ND was calculated from the number of reads per million (RPM) for each gene and each sample. For C and E, statistics is by edgeR. For D and F, statistics is by t-test, *p < 0.05; **p < 0.01. N = 9-10.
Scientific RepoRts | 7: 4307 | DOI:10.1038/s41598-017-04412-2 HFD affects neurite morphology. RNA-seq results revealed that HFD decreased the expression of genes related to neuron projections, axon guidance and synaptic transmission. To test if the changes in transcriptome reflect on the neuronal phenotype, we compared the dendrite architecture in ND and HFD groups using staining for neurite marker microtubule-associated protein 2 (MAP2). As shown on Fig. 4, we found that in mice fed HFD there was a significant decrease in the dendritic length ( Fig. 4B) and number of segments and branches ( Fig. 4C and D), suggesting that HFD detrimentally affects neurite morphology.
HFD significantly impacts brain lipidome in AD model mice. To get further insight into the effects of HFD on the phenotype, we examined brain lipidome of mice on HFD and ND using Multidimensional Mass Spectrometry based Shot Gun Lipidomics (MDMS-SL) 28 . We did not find a significant gender effect on brain lipidome and for further analysis combined the data for both genders. We identified 17 major classes of lipids (for abbreviation see Fig. 5A) comprised of 228 molecular species within those classes. We then applied Principal Component Analysis (PCA) to further process the abundance matrix of observed variables (lipid molecular species) and to calculate Principal Components that account for the highest possible variance in the datasets. We noticed that HFD samples clustered closely and were separated from ND replicates (Fig. 5B). While there was no difference in the total amounts of lipids of each class, we identified that the levels of 24 molecular species in the brain of APP23 mice were significantly modulated by HFD ( Fig. 5C and Supplementary Table S1). To elucidate additional changes in brain lipidomes, canonical correlation analysis was performed 28 and the correlation matrices were further processed to visualize the correlations (Fig. 6). The noticeable difference in the upper right corner of the correlation matrices (panel A-ND, panel B-HFD) indicates a change in the relationships of the related lipids (CL, ST, PIP, and PIP2). In HFD there is a stronger negative correlation of CL to ST, PIP and PIP2 than in the ND. This is accompanied by the higher positive correlation between ST, PIP, PIP2 in HFD. Network visualization of correlated lipid pairs in ND and HFD lipidomes confirmed these observations. On Fig. 6C-F the most significant correlations for both diets in the matrices (p < 0.005, r > 0.9) are presented by chord diagrams. The comparison of the negative correlations in both lipidomes clearly demonstrate that CL in HFD mice strongly and negatively correlates to PIP and PIP2 (panel F). The overall imbalance seen in HFD lipidome, when correlating the lipids in both groups is indicative of alterations by diet and is most prominent in CL, ST, PIP and PIP2 phospholipids.
HFD significantly modulates genes associated with mitochondrial components and apoptosis. The phospholipid composition of cellular membranes is important for cell signaling including apoptosis and phagocytosis, and phospholipid imbalance is closely linked to mitochondrial dysfunction 29 . To have a better understanding of the interaction between transcriptome and lipidome (Figs 5 and 6), we searched our RNA-seq results for genes that code for components of mitochondrial membranes or are associated with mitochondrial biogenesis. We compared our RNA-seq data set with mitochondria cellular component genes from MSigDB (Molecular Signatures Database v5.2, http://software.broadinstitute.org/). Figure 7A shows the protein interaction network of significantly affected genes coding for proteins associated with components of mitochondrial membranes (Cox6a2 30 , Shmt2 31 , Tomm20 and Timm10 32, 33 ) and matrix (Abat 34 and ATP synthases) as well as mitochondrial transcription factors (Tfb2m) 35 by HFD. As shown the most connected protein was SHMT2 associated with mitochondrial biogenesis and immune response 31 . Among the genes related to mitochondrial component and function, most differentially affected genes by HFD were represented in Fig. 7B. Interestingly, we identified Nr4a subfamily (Nr4a1/Nur77 and Nr4a3/Nor1 36,37 ) which regulates apoptosis and inflammation was significantly decreased in HFD (Fig. 7B).

Discussion
Our study demonstrates that HFD affects cognitive performance and amyloid pathology in APP23 transgenic mice. The diet also significantly affected the transcriptome and increased the expression level of genes related to immune response and inflammation, such as Trem2, Tyrobp, P2ry12, Dock2, a number of cytokines, chemokines, and TLRs. Conversely, the expression of the genes associated with neuronal projection development and synaptic transmission were decreased in APP23 mice fed HFD. Interestingly, in those mice there was a significant decrease of vesicular GABA transporter (Slc32a1/VGat), but not of Slc17a7/VGlut. Collectively, gene ontology categories that were significantly decreased by HFD corresponded to a phenotype presented by significantly deteriorated neurite morphology and spatial memory test. The integration of transcriptomics and lipidomics data demonstrated that the effect of HFD on behavior and AD-like phenotype in an animal model of AD is extremely multidimensional and is a result of a complex response at molecular, cellular and system levels in the CNS.
The increased mRNA level of Trem2, was accompanied by an increased protein level determined by IHC. Interestingly, TREM2 immunoreactivity was predominantly increased in female APP23 mice. Overall, when all mice were analyzed, there was a very high correlation between TREM2 immunostaining and Aβ plaques. The association of TREM2 with amyloid deposition is somewhat controversial and the reports, inconsistent: some groups have reported an increase and other a decrease in amyloid plaque load in a variety of animal models including Trem2 deficient and mice expressing different TREM2 variants 38,39 . It has been also reported that Trem2 deficiency exacerbates amyloid pathology late in disease progression 40 , suggesting that amyloid pathology is linked to TREM2 level. A recent study demonstrated that a type of western diet enriched on carbohydrates, and different from the diet we have used, increased TREM2 immunoreactivity in association with micro-and astrogliosis 18 . Thus, an overall inflammatory response elicited by the increased amyloid deposition and diet could contribute to the worsened cognitive performance.
The main question in this study was if HFD affected gene expression/brain transcriptome and if changes in transcription profiles correlated to changes in brain lipidome? Lipids play critical roles in cellular signaling, they are major factors in membrane protein assembly, and the abnormal lipid metabolism is strongly associated with AD progression. Here, in addition to mRNA-seq we applied MDMS-SL to quantify global phospholipid changes induced by diet. The results of our lipidomics analysis demonstrate that amounts of individual molecular species of anionic phospholipid classes PS, PG and PIP3 were increased in HFD fed mice. While changes in total mass of the majority of lipid classes are important in different aspects of brain metabolism, changes in molecular speciation of PIP3 and PS are particularly important for activation of immune phagocytic receptors, phagocytosis and apoptosis [41][42][43] . It has been recently shown, for example, that TREM2 senses anionic lipids 38 and TREM2 variants identified by GWAS as increasing the risk of AD, affect its binding 44 . Previous studies demonstrated that these anionic phospholipids are exposed on the outer leaflet of cell membranes and become accessible in conditions following neuronal damage inflicted by brain trauma 45 or neurotoxic effects of Aβ 46 . In the resting cell, PS is enriched on the cytoplasmic leaflet of the plasma membrane by flippases, including P4-ATPases, and ABC transporters. However, the cells undergoing apoptosis expose PS on the cell surface and then it acts as an "eat me" signal 47,48 . In this study we found changes in expression of several genes related to PS-mediated apoptosis-C1q (C1qa, C1qb and C1qc) (Fig. 2C and D), Tim-3/Havcr2, Scarf1 and Atp8b2 (Supplementary Fig. S3). Multiple reports by other groups suggest that C1q and TIM-3/HAVCR2 directly bind to PS on the cell surface, and C1q bound to PS provides a docking point for the scavenger receptor SCARF1. These interactions lead to increased apoptosis and initiate phagocytosis signaling pathways 49,50 .
Perhaps the most significant changes in response to HFD, revealed in this study, were associated with CL -a major phospholipid of the inner mitochondrial membrane. Recently, it has been demonstrated that alterations of CL profile in brain of AD animal model are associated with synaptic mitochondrial dysfunction and oxidative stress 51 . In brain of HFD fed mice we identified 5 elevated CL molecular sub-species (68:4; 70;4; 72:4; 74:7; 84:20) and 2 (72:6; 76:11) with decreased amounts (Supplementary Table 1). Since the amount of CL in the outer mitochondrial membrane and intra-mitochondrial CL content are tightly regulated, significant changes in either direction from the threshold result in substantial, even detrimental consequences 52 . If these changes occur simultaneously with changes in brain transcriptome,-expression level of genes coding for proteins in the mitochondrial membranes in particular, responsible for cholesterol and protein transport 32, 33 , if there is already an undergoing neurodegenerative process, the final outcome of an external stimulus like HFD could be serious and long lasting.
Collectively, the results of our study demonstrate that HFD increases mRNA expression and protein level of genes related to immune/inflammatory response and decreases the level of genes related to neuronal differentiation and synaptic transmission. Interconnected alteration of membrane phospholipid composition and likely mitochondrial dysfunction emphasize the complexity of the neurotoxic effects of HFD, reflected in aggravated AD-like phenotype and behavioral changes that we begin to understand.

Animals and Diets. All animal experiments were approved by the University of Pittsburgh Institutional
Animal Care and Use Committee. All procedures were carried out in accordance with the approved guidelines. APP23 transgenic mice (C57Bl6 background) expressing human APP751 familial Swedish AD mutation (APPK670N, M671L) were crossed to wild-type C57BL/6 mice to generate APP23 hemizygous mice 53 . Male and female APP23 mice (mean age 11.7 months) were randomly assigned to ND (Prolab Isopro RMH 3000, Lab Diet) or HFD (D12079B RD Western Diet, Research Diets). HFD supplemented 40% calories from fat; 16.8% from protein and 42.6% from carbohydrates. Fat was provided as anhydrous milk fat and corn oil. The mice had free access to water and food. After 3 months of feeding, behavior was assessed at mean age of 14.7 months.
Morris Water Maze. Morris Water Maze (MWM) task was performed as previously described 8 . Briefly, the water maze pool was filled with water (20-22 °C) made opaque by non-toxic paint and visual cues were placed on the walls surrounding the pool. The hidden platform (10 cm diameter, submerged 1 cm below water level) remained in the target quadrant for the duration of testing. Performance was recorded with video tracking software (AnyMaze; Stoelting Co.). Spatial Acquisition: All mice performed four 60-second trials with different pool entry points and 5-minute trial intervals for five consecutive training days. Probe Trial: A single 60-sec probe trial was performed 24 hours after the last spatial acquisition trial where the hidden platform was removed.
Processing of mouse brain tissue. Mice were anesthetized with Avertin (250 mg/kg of body weight, i.p.) and perfused transcardially with cold 0.1 M PBS, pH 7.4: left hemisphere fixed in 4% paraformaldehyde for histology. The cortex was dissected from the other and snap frozen on dry ice for RNA and protein isolation 54 . All procedures were as reported previously 53 . Briefly, OCT-embedded hemibrains were cut in the coronal plane at 30 μm sections and stored in a glycol-based cryoprotectant at −20 °C. Serial sections were selected 700 μm apart, starting from approximately 150 μm caudal to the first dentate gyrus appearance.
Histology and Immunohistochemistry. All procedures were as reported previously 53 . Briefly, brain sections mounted on slides were washed in PBS and stained with X-34 (1,4-bis (3-carboxy-4hydroxyphenylethenyl)-benzene, 100 µM, provided by W. Klunk, University of Pittsburgh). Following the staining, sections were washed in water, incubated in 0.2% NaOH in 80% unbuffered ethanol, washed again in water, and then soaked in PBS. For 6E10 staining, Adjacent sections were blocked for endogenous peroxidases, avidin-biotin quenched, antigen retrieval performed with 70% formic acid, and tissue blocked with 3% normal goat serum. Sections were then immunostained with 6E10 biotinylated antibody (1:1000, SIG-39340, Covance) at room temperature before being developed with Vectastain ABC Elite kit (Vector Laboratories, Burlingame, CA) and a DAB substrate (SK-4100, Vector Laboratories).
For neurite morphometry analysis, sectioning was performed as described above and sections sampled starting 450 μm from the first appearance of the dentate gyrus and every 360 μm (4 total sections). Sections were rinsed in PBS and permeabilized with 1% Triton X-100 in PBS for 5 hr at 4 °C. The sections were then washed in PBS and blocked with 10% goat serum at RT. The sections were incubated for 72 hr at 4 °C with MAP2 primary antibody (1:2000, MAB378, Millipore) and rinsed with PBS and then incubated with Cy3-conjugated anti-mouse secondary antibody.
TREM2 staining was performed as in refs 39 and 55. Briefly, sections were rinsed in PBS before antigen retrieval (RV1000, Biocare Medical, 1:10), endogenous peroxidases (0.3% H 2 O 2 ), avidin, and biotin (PK6100, ABC Elite Vector Laboratories), and hydrophobic interactions (2% donkey serum) were blocked. Sections then incubated in TREM2 primary antibody (AF179, R&D Systems) and developed with DAB substrate. To ensure the specificity of TREM2 staining we used the brain tissue from a Trem2 KO mice and a staining without a primary antibody as negative control.
Quantitative Analysis. Nikon Eclipse 90i microscope (10X magnification) and Nikon NIS Element (Nikon) software were used for evaluating the amyloid pathology in the cortex and hippocampus of fixed brain sections and the pathology is presented as a percentage of area covered by X-34 or 6E10 stain. For MAP2 imaging in CA1, 4 adjacent confocal images were captured using the granule cell layer as a landmark (Olympus Fluoview 1000, 60X magnification). The FilamentTracer (Imaris, version 7.1.1, Bitplane) was utilized to determine neuronal patterning in the hippocampal CA1 region. Unbiased examination and quantitation of neurite size and length required manual introduction. Neurite length was normalized to the number of H33342 stained nuclei.
RNA isolation, qPCR and sequencing. All procedures were performed as before 54,56 . RNA was isolated from cortex and purified using RNeasy mini kit (Qiagen) per the manufacturer's protocol. RNA quality was assessed using 2100 Bioanalyzer (Agilent Technologies) and samples with RIN > 8 were further used for sequencing. Sequencing libraries were generated using mRNA Library Prep Reagent Set (Illumina) according to the manufacturer's recommendations. The libraries were sequenced on Illumina HiSeq2000 at the Functional Genomics Core, UPenn, Philadelphia (http://fgc.genomics.upenn.edu/). All sequencing datasets, FASTQ files in this study have been processed using Subread-featureCounts-limma/voom pipeline 57, 58 and the differential expression analyzed by edgeR (v3.14.0) package http://bioconductor.org/packages/release/bioc/html/edgeR.html in R (v3.2.4).

Multi-dimensional Mass Spectrometry-based Shotgun Lipidomics (MDMS-SL). Quantitative
analysis was performed on a triple-quadrupole mass spectrometer (Thermo Fisher Scientific) equipped with an automated nanospray apparatus NanoMate and Xcalibur system [61][62][63] . Internal standards for quantification of individual molecular species of the major lipid classes are added to each brain tissue sample at the start of extraction procedure. Liquid Chromatography/Electro Spray Ionization Lipidomics-Mass Spectrometry (LC/