RNAseq Analysis of FABP4 Knockout Mouse Hippocampal Transcriptome Suggests a Role for WNT/β-Catenin in Preventing Obesity-Induced Cognitive Impairment

Microglial fatty-acid binding protein 4 (FABP4) is a regulator of neuroinflammation. We hypothesized that the link between lipid metabolism and inflammation indicates a role for FABP4 in regulating high fat diet (HFD)-induced cognitive decline. We have previously shown that obese FABP4 knockout mice exhibit decreased neuroinflammation and cognitive decline. FABP4 knockout and wild type mice were fed 60% HFD for 12 weeks starting at 15 weeks old. Hippocampal tissue was dissected and RNA-seq was performed to measure differentially expressed transcripts. Reactome molecular pathway analysis was utilized to examine differentially expressed pathways. Results showed that HFD-fed FABP4 knockout mice have a hippocampal transcriptome consistent with neuroprotection, including associations with decreased proinflammatory signaling, ER stress, apoptosis, and cognitive decline. This is accompanied by an increase in transcripts upregulating neurogenesis, synaptic plasticity, long-term potentiation, and spatial working memory. Pathway analysis revealed that mice lacking FABP4 had changes in metabolic function that support reduction in oxidative stress and inflammation, and improved energy homeostasis and cognitive function. Analysis suggested a role for WNT/β-Catenin signaling in the protection against insulin resistance, alleviating neuroinflammation and cognitive decline. Collectively, our work shows that FABP4 represents a potential target in alleviating HFD-induced neuroinflammation and cognitive decline and suggests a role for WNT/β-Catenin in this protection.

The FABP4 knockout mouse model (AKO) has allowed an increasingly comprehensive understanding of lipid metabolism, diabetes, and peripheral immune response [16][17][18][19][20]. Unlike obese wild type controls, AKO mice present with decreased TNF-α (a biomarker of obesity-related insulin resistance) in adipose tissue [5]. Fatty acid metabolism plays a central role in immunometabolic diseases, including diet-induced obesity and metabolic syndrome [16,21,22]. We hypothesized that the link between lipid metabolism and inflammation indicates a role for FABP4 in regulating HFD-induced cognitive decline. We have previously shown that AKO mice are protected against HFD-induced hippocampal proinflammatory cytokine expression and memory deficits [23,24]. These mice were fed HFD for 12 weeks starting at 15 weeks of age. Herein, we investigated the transcriptomics of the AKO model fed HFD compared to wild type (WT) mice fed the same diet. As our previous research showed no differences in hippocampal inflammation and cognitive decline between groups fed normal chow diet (NC), nor between HFD-fed and NC-fed AKO groups [23], only HFD hippocampal transcriptomes were analyzed. Our present results show that AKO mice present with a neuroprotective hippocampal transcriptome that correlates with a decrease in proinflammatory signaling, ER stress, apoptosis, and cognitive decline, and an increase in neurogenesis, synaptic plasticity, long-term potentiation, and spatial working memory. This is accompanied by an increase in WNT/β-Catenin signaling, suggesting a role for the pathway in preventing HFD-induced neuroinflammation and cognitive decline.

Transcript Analysis
We determined the transcripts that were differentially expressed in HFD-fed AKO mice compared to HFD-fed WT mice. AKO and WT mice were fed 60% HFD for 12 weeks starting at 15 weeks of age. Hippocampal tissue was dissected and RNAseq was performed to measure differentially expressed transcripts. Transcript analysis showed 1793 differentially expressed transcripts at log10 adjusted p-value (padj) ≤ 0.05 ( Figure 1).

β-Catenin Western Blot
We performed Western blotting for nuclear β-Catenin in HFD-fed AKO and WT mouse hippocampal tissue. Results showed that HFD-fed AKO mice had increased expression of nuclear β-Catenin compared to WT (Figure 2; p = 0.03).

Discussion
Obesity is known to be associated with peripheral and central chronic low-grade inflammation [99]. Peripherally, HFD is linked to low-grade systemic inflammation associated with the development of metabolic disorders such as type 2 diabetes mellitus [100]. In the brain, diets high in SFAs are known to be associated with neuroinflammation, cognitive impairment, and the development of neurodegenerative diseases [101]. An increase in circulating free fatty acids (FFAs) disrupts metabolic homeostasis [102]. This increase in FFAs can also activate Toll-like receptor 4 (TLR4) in microglia, leading to ER stress and increased inflammatory cytokine expression [103]. In turn, increased ER stress and neuroinflammation can lead to cognitive deficits and neurodegenerative diseases such as AD [102]. The HFD-fed FABP4 knockout model has revealed an altered hippocampal transcriptome that is shown to have changes in metabolic pathways and be neuroprotective against HFD-induced neuroinflammation and cognitive decline.
In the present study, the AKO model has shown an altered hippocampal transcriptome with changes in metabolic pathways. AKOs had a decrease in pathways relating to the electron transport chain (ETC). These pathways were "formation of ATP by chemiosmotic coupling", "cristae formation", "complex I biogenesis", "respiratory electron transport, ATP synthesis by chemiosmotic coupling, and heat production by uncoupling proteins", and "respiratory electron transport" (Supplemental Table S1). Importantly, the major downregulated transcripts in these pathways encode for subunits of NADH:ubiquinone oxidoreductase (Complex I) and ATP synthase (Complex V) of the ETC. Partial inhibition of Complex I has been shown to improve energy homeostasis, synaptic activity, long-term potentiation, and cognitive function while reducing oxidative stress and inflammation in the brain [68]. Targeting Complex V has been shown to decrease expression of an aged and dementia-associated transcriptome [86]. AKOs had an increase in the "citric acid (TCA) cycle and respiratory electron transport" pathway (Table 3). It is important to note that transcripts relating to the tricarboxylic acid cycle (TCA) were increased. These transcripts include Aco2, Cs, Ogdh, and Sdha. Downregulation of these transcripts have been observed in neurodegenerative transcriptomes and some have been associated with oxidative stress and insulin resistance [74,[76][77][78]. Our previous work has shown that FABP4 knockout confers protection against ROS production and ER stress through an

Discussion
Obesity is known to be associated with peripheral and central chronic low-grade inflammation [99]. Peripherally, HFD is linked to low-grade systemic inflammation associated with the development of metabolic disorders such as type 2 diabetes mellitus [100]. In the brain, diets high in SFAs are known to be associated with neuroinflammation, cognitive impairment, and the development of neurodegenerative diseases [101]. An increase in circulating free fatty acids (FFAs) disrupts metabolic homeostasis [102]. This increase in FFAs can also activate Toll-like receptor 4 (TLR4) in microglia, leading to ER stress and increased inflammatory cytokine expression [103]. In turn, increased ER stress and neuroinflammation can lead to cognitive deficits and neurodegenerative diseases such as AD [102]. The HFD-fed FABP4 knockout model has revealed an altered hippocampal transcriptome that is shown to have changes in metabolic pathways and be neuroprotective against HFD-induced neuroinflammation and cognitive decline.
In the present study, the AKO model has shown an altered hippocampal transcriptome with changes in metabolic pathways. AKOs had a decrease in pathways relating to the electron transport chain (ETC). These pathways were "formation of ATP by chemiosmotic coupling", "cristae formation", "complex I biogenesis", "respiratory electron transport, ATP synthesis by chemiosmotic coupling, and heat production by uncoupling proteins", and "respiratory electron transport" (Supplemental Table S1). Importantly, the major downregulated transcripts in these pathways encode for subunits of NADH:ubiquinone oxidoreductase (Complex I) and ATP synthase (Complex V) of the ETC. Partial inhibition of Complex I has been shown to improve energy homeostasis, synaptic activity, long-term potentiation, and cognitive function while reducing oxidative stress and inflammation in the brain [68]. Targeting Complex V has been shown to decrease expression of an aged and dementia-associated transcriptome [86]. AKOs had an increase in the "citric acid (TCA) cycle and respiratory electron transport" pathway (Table 3). It is important to note that transcripts relating to the tricarboxylic acid cycle (TCA) were increased. These transcripts include Aco2, Cs, Ogdh, and Sdha. Downregulation of these transcripts have been observed in neurodegenerative transcriptomes and some have been associated with oxidative stress and insulin resistance [74,[76][77][78]. Our previous work has shown that FABP4 knockout confers protection against ROS production and ER stress through an increase in UCP2 [20]. As such, our present results provide further evidence that FABP4 knockout leads to protective metabolic changes in HFD-fed mice. The AKO model has also shown an increase in a pathway relating to Golgi transport. This pathway is named "transport to the Golgi and subsequent modification" (Supplemental Table S1). The upregulated genes in this pathway include regulators of ER to Golgi transport [104,105] and a decrease in some of these transcripts have been associated with ER stress [106] and neurodegenerative diseases [107,108]. As Golgi function and transport is important for lipid metabolism [109], changes in transcripts relating to these pathways provide further insight on the neuroprotection conferred via FABP4 knockout. Differentially expressed genes relating to lipid metabolism in AKOs were Apoa2 (Table 1) and Sort1 (Table 2). Apoa2 was downregulated in AKOs and encodes for a high-density lipoprotein. Certain polymorphisms of Apoa2 are known to be associated with obesity [110]. Sort1 was upregulated in AKOs and has been shown to modulate low-density lipoprotein uptake in macrophages [111]. Together, these changes in metabolic and Golgi transport pathways, along with genes associated with changes in lipid metabolism, show that HFD-fed AKO mice have a hippocampal transcriptome consistent with neuroprotection.
Previously, we have shown that HFD-fed AKO mice exhibit an alleviation of hippocampal inflammatory cytokine signaling compared to HFD-fed WT mice [23]. In the present study, HFD-fed AKO mice had transcriptome changes consistent with a decrease in neuroinflammation. AKOs had decreased expression of Tpx2 and Dmd (Table 1), along with increased expression of S1pr3 ( Table 2). Silencing of Tpx2 has been shown to decrease inflammation as marked by decreases in TNF-α, IL-6, and IL-8 protein levels [28]. Mice deficient of dystrophin, the protein encoded by Dmd, exhibit decreased infiltration of CD3+ T cells [29]. Knockdown of S1pr3 has been shown to exacerbate neuroinflammation as measured by IBA1 and TNF-α [45]. HFD-fed AKO mice also had transcriptome changes consistent with a decrease in ER stress. AKOs had decreased expression of Lonp2 (Table 1), and increased expression of Hspa5 and Smpd4 (Table 2). ER stress is known to cause a decrease in LONP2 expression [31]. HSPA5 is known to be a regulator of ER stress in the brain [42] and loss of SMPD4 has been shown to induce ER stress in the brain [43]. HFD is known to cause ER stress in the hippocampus [112] and ER stress can trigger inflammatory responses in the brain, leading to neurodegeneration [113]. Therefore, the transcriptomic changes relating to decreases in ER stress provide a potential mechanism for the decrease in neuroinflammation.
We have also previously shown that HFD-fed AKO mice exhibit an alleviation of cognitive decline and memory deficits compared to HFD-fed WT mice [23]. In the present study, HFD-fed AKO mice had transcriptome changes consistent with a decrease in cognitive decline. AKOs had decreased expression of Apoa2 (Table 1). Increased APOA2 expression is associated with cognitive impairment and late-life dementia [27]. AKOs also had transcriptome changes consistent with a decrease in apoptosis. Expression of Hspa5 and Smpd4 were increased ( Table 2). HSPA5 is known to be an antiapoptotic factor for cells undergoing ER stress [42]. Loss of SMPD4 has been shown to lead to apoptosis under stress conditions [43]. As apoptosis leading to neuronal cell death has been associated with AD pathogenesis [114], an increase in antiapoptotic factors could contribute to a neuroprotective profile against cognitive decline. AKOs also had a transcriptome negatively associated with AD. Expression of Eif4e2 and Wtap were decreased (Table 1), and expression of Atp2a2 was increased ( Table 2). The phosphorylated state of EIF4E is associated with hyperphosphorylated tau and AD [36]. WTAP is a component of the complex responsible for m6A methylation [40], which is increased in AD mice [41]. Expression of ATP2A2 has been shown to be decreased in the AD brain [58]. As cognitive decline is a hallmark of AD, a transcriptome negatively associated with AD further reveals the neuroprotective profile of the AKO hippocampus.
The transcriptomic changes in HFD-fed AKO mice also reveals differentially expressed transcripts that relate to neurite outgrowth and neurogenesis. Expression of Tmod2 was decreased (Table 1), and expression of Gpr68 and Sufu were increased ( Table 2). Knockdown of Tmod2 leads to an increase in neurite outgrowth in vitro [26]. There is evidence to show that GPR68 has a positive role in neurogenesis [52]. Deletion of Sufu has been shown to decrease neurogenesis in the dentate gyrus [53]. As a decrease in neurogenesis is linked to cognitive deficits and neurodegenerative disease [115], transcriptomic changes that promote neurogenesis would further confer protection against cognitive decline. HFD-fed AKO mice also had transcriptomic changes relating to synaptic plasticity and long-term potentiation. Expression of Brd1, Adra2a, Mdga1, and Grik2 were increased ( Table 2). The BRD1 protein is needed for H3K14 acetylation [63], which is associated with synaptic plasticity [50]. The activation of α2-adrenergic receptors has been shown to promote long-term potentiation in the mouse brain [46]. Mice lacking MDGA1 and GRIK2 have exhibited compromised hippocampal long-term potentiation [47,48]. Long-term potentiation and synaptic plasticity are critical for forming memories and preventing cognitive impairment [116]. The AKO hippocampal transcriptome further reveals protection against cognitive decline by having upregulated transcripts that promote synaptic plasticity.
The WNT/β-Catenin signaling pathway has been shown to be neuroprotective against cognitive decline and neurodegenerative disease. Mechanisms that confer this protection include increasing neuronal survival, neurogenesis, synaptic plasticity, and blood-brain barrier integrity, as well as decreasing amyloid-β production and tau hyperphosphorylation [83,[117][118][119]. HFD-fed AKO mice exhibited transcript and pathway changes consistent with an increase in WNT/β-Catenin signaling. After Fabp4, the next most downregulated transcript was Sox6 (Table 1), which has been shown to inhibit WNT/β-Catenin signaling in adipocytes [25]. The third most upregulated transcript was Nox1 (Table 2), which has been shown to increase WNT/β-Catenin signaling [54]. Reactome analysis revealed that the most upregulated pathway was "disassembly of the destruction complex and recruitment of AXIN to the membrane" (Table 3). Within this pathway, transcripts that were increased included Ctnnb1, Dvl3, Fzd1, Ppp2cb, Ppp2r1a, Ppp2r5a, and Ppp2r5e (Table 3). These transcripts encode for proteins that are positive regulators of WNT/β-Catenin signaling, and have roles in neuronal survival, synaptic plasticity, neuronal differentiation, and improved spatial memory [80,[83][84][85]. Ccnd1, a target of WNT/β-Catenin signaling [120], was also found to be increased in AKOs (Log FC: 0.63, padj: 3.62 × 10 −2 ). As CCND1 has been shown to be a positive regulator of neurogenesis [121], the increase in this target gene further supports the neuroprotective profile of AKOs. To verify the increase in WNT/β-Catenin signaling, we performed Western blotting for nuclear β-Catenin in HFD-fed AKO and WT mouse hippocampal tissues. Results showed that AKO mice had an 80% increase in nuclear β-Catenin expression compared to WT (Figure 2). This is higher than the increase in Ctnnb1 transcript, which was 30% (Table 3). These results further show that loss of FABP4 confers neuroprotection against HFD-induced cognitive decline and suggests a role for WNT/β-Catenin in this protection. It is important to note that Csnk1a1, Csnk1g2, and Gsk3b were also upregulated in this pathway (Table 3). These transcripts encode for proteins that are canonically negative regulators of WNT/β-Catenin signaling. These proteins are known to have roles in other pathways in the brain. Casein kinase 1 is known to be a negative regulator of TGF-β [82], whose overproduction in the brain has been linked with glucose intolerance [122]. GSK3B also has many roles in the brain including regulating neurogenesis, axon growth, and synaptic plasticity [123]. These roles may explain why these transcripts were upregulated in AKOs.
FABP4 is expressed in peripheral macrophages and microglia. In the periphery, knockout of FABP4 in macrophages leads to a decrease in inflammatory cytokine secretion [124]. We have previously shown that FABP4 is expressed in microglia [24]. As microglia activation is an important aspect of HFD-induced neuroinflammation, we decided to examine the changes in microglial marker transcripts in AKOs. The transcripts Csf1r and Kcnk6 were both significantly upregulated (Table 4). Csf1r encodes for the receptor of colony stimulating factor 1, a positive regulator of the M2 microglial transcriptome [87]. KCNK6 is associated with a homeostatic microglial phenotype [98]. Other upregulated microglial transcripts include positive regulators of memory and long-term potentiation [92,93], and neuronal cell differentiation and survival [90,97]. Together HFD-fed AKO mice have changes in microglial transcripts consistent with homeostatic microglia and protection against cognitive decline. Future studies will include the immunohistochemical staining of these microglial markers to provide further insight on their regulation in the AKO hippocampus.
The HFD-fed FABP4 knockout model has revealed an altered hippocampal transcriptome that is shown to have changes in metabolic pathways and be neuroprotective against HFD-induced neuroinflammation and cognitive decline. We and others have shown that knockout of FABP4 leads to an alleviation of HFD-induced peripheral and central inflammation, insulin insensitivity, and cognitive decline [16,20,23]. In the present study, we have shown that this alleviation in the brain is associated with changes in metabolic pathways, Golgi transport, and WNT/β-Catenin signaling. As there is much supporting evidence that the WNT/β-Catenin pathway is neuroprotective against cognitive decline, our study suggests a role for WNT/β-Catenin in the FABP4 knockout-induced alleviation of cognitive decline. Other studies have shown that WNT/β-Catenin has a role in the regulation of microgliosis. A decrease in WNT/β-Catenin signaling has been shown to lead to proinflammatory microglial activation in the developing brain [125]. Wnt-3a, an activator of WNT/β-Catenin signaling, causes a decrease in the expression of inducible nitric oxide synthase (iNOS) and TNF-α and a decrease in microgliosis [126], while increasing insulin sensitivity [127]. Other activators of WNT/β-Catenin signaling have been shown to cause a switch in microglia from a proinflammatory phenotype to an anti-inflammatory one after ischemic stroke [128]. As many targets of WNT/β-Catenin were found to be unchanged in this study, further work is necessary to determine the mechanism and role of WNT/β-Catenin signaling in the alleviation of HFD-induced neuroinflammation and cognitive decline caused by FABP4 knockout. These studies will likely focus on isolated microglia from whole brain tissue. Single-cell RNA sequencing (scRNA-seq) has been used to define microglial immunoheterogeneity in models of AD [129,130]. Future studies using scRNA-seq to elucidate the microglial transcriptome changes that occur with FABP4 knockout in HFD-fed mice will provide further clarity on the mechanisms that lead to neuroprotection.

Mouse Model of Obesity and Immunometabolism
The FABP knockout mouse model (also known as aP2 −/− or AKO mouse) carries a null mutation in FABP4 bred to wild-type C57BL/6J mouse background [16]. We refer to this transgenic mouse line (and tissue derived from these mice) as AKO mice. Genotyping was confirmed using PCR as previously described [16,124]. Mice were maintained in a 12:12 h light/dark cycle in a temperature-controlled room (21-22 • C) and were group-housed. WT and AKO mice (n = 5 per genotype) were placed on a high fat diet (HFD; Research Diets D12492; 60% total fat and 32% saturated fat) for 12 weeks starting at 15 weeks of age and water was provided ad libitum. The experimental protocol was approved by the Institutional Animal Care and Use Committee at the Minneapolis VAHCS.

Brain Dissection and RNA Isolation
Hippocampal tissue was rapidly dissected from WT and AKO mice sacrificed during the light phase, 5-8 h after lights on [131]. Total RNA was extracted from hippocampal tissue with the aid of Trizol (Invitrogen; Carlsbad, CA, USA), purified using the RNeasy Mini Kit (Qiagen, Hilden, Germany) and both procedures were performed according to the manufacturer's specification [131,132].

RNA-Seq cDNA Library Synthesis
Total RNA (3 µg) samples were sent to Novogene (Davis, CA, USA) for RNA-seq analysis. Briefly, RNA purity (OD260/OD280) was quantified using a Nanodrop spectrophotometer (Thermo Scientific; Waltham, MA, USA), RNA integrity and potential contamination was analyzed using agarose gel electrophoresis, and RNA integrity was further analyzed using the Agilent 2100 bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). mRNA was purified from total RNA using poly-T oligo-attached magnetic beads and mRNA was then fragmented randomly. All sequencing was performed in one batch. First strand cDNA was synthesized using random hexamer primer and M-MuLV Reverse Transcriptase Minus (RNase H−). Second strand cDNA synthesis was then performed using DNA Polymerase I and RNase H with dTTP replaced by dUTP. Double-stranded cDNA was purified using AMPure XP beads. Remaining overhangs of the purified doublestranded cDNA were converted into blunt ends via exonuclease/polymerase activities. After adenylation of 3 ends of DNA fragments, NEBNext Adaptor with hairpin loop structure was ligated to prepare for hybridization. The second strand cDNA was then digested by USER enzyme. The final library was obtained by PCR amplification and purification of PCR products by AMPure XP beads. Library quality was ensured using the Agilent 2100 bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). Clustering was performed on a cBot cluster generation system using the TruSeq PE Cluster Kit v3-cBot-HS (Illumina; San Diego, CA, USA). Library sequencing was performed using the Illumina HiSeq Platform (Illumina; San Diego, CA, USA).

RNA-Seq Analysis
Raw data (raw reads) were stored in FASTQ (fq) format. Raw reads containing adaptors, over 10% indeterminate bases, or over 50% low-quality bases were removed. Clean reads were mapped to a reference genome using STAR software. Differential expression analysis was performed using the DESeq2 R package. P-values were adjusted using Benjamini and Hochberg's procedure. Enrichment analysis of differentially expressed genes was performed using clusterProfiler software (version 3.16), including GO, DO, KEGG and Reactome database enrichment [133].

Reactome Analysis
Transcript quantification was performed with Salmon 1.0 [134] packaged in clus-terProfiler 3.14.2 [135]. It was run with mapping mode selective alignment, minimum count of three, and relative abundance calculated in transcripts per million (TPM) relative to reference RNA transcripts from GENCODE VM23. Overrepresentation of transcripts and pathways and all other expression analysis was conducted with Reactome gene set analysis (ReactomeGSA) with Reactome pathway browser 3.7, database release 74 using the gene set enrichment analysis (GSEA) method and including disease pathways [136]. Within ReactomeGSA, normalization was performed with edgeR's calcNorm-Factors function, transformed using limma's voom and normalizeBetweenArrays functions, and pathway analysis using limma's camera function as implemented in the respective Bioconductor R package. Reactome identifiers are shown in square brackets and can be accessed at https://reactome.org. Unless otherwise noted, statistical significance was evaluated with a cutoff of ≤ 0.01 using the false discovery rate (FDR) as calculated using the Benjamini-Hochberg procedure for pathways and multiple correction adjusted p-values for transcripts (or genes). Transcript identifiers were translated to Ensembl gene IDs (names) with g:Profiler [137]. Additional ad hoc pathway analysis was conducted with GO, KEGG databases. Estimate surrogate cell type proportions variables were calculated with BRETIGEA [138]; input gene-level expression calculated by summing transcript TPMs per gene, and output with mouse species and separate runs with 20, 50, and 200 markers for all six cell types (astrocytes, endothelial cells, microglia, neurons, oligodendrocytes, and oligodendrocyte precursor cells). Senescence was evaluated with a list of 88 gene markers from the Tabula Muris Consortium [139].

Informed Consent Statement: Not applicable.
Data Availability Statement: Datasets generated during this study are the property of the U.S. Department of Veterans Affairs and will be made available upon request.