The research in this study was performed in compliance with the Public Health Service Policy on Humane Care and Use of Laboratory Animals and all animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC) at the University of California, Davis (protocol code 20943 and date of approval: 04/18/2019). Male C57BL/6J mice (n = 21, 19 weeks of age) were purchased from The Jackson Laboratory (Bar Harbor, ME, United States) and housed (n = 1 per cage) in a 12-h light/dark cycle in a temperature- and humidity-controlled environment within the University of California, Davis Mouse Biology Program. Mice were fed a standard chow diet (Teklad Custom Diets #0915, Madison, WI, United States) during the acclimation period of 1 week and then were randomly assigned to three experimental dietary intervention groups (n = 7/group) for 12 weeks: low glycemic diet (LGD, Teklad Custom Diets #TD.08485; 67.9% kcal of carbohydrate, 19.1% protein, 13% fat, with 12% sucrose by weight), high glycemic diet (HGD, #TD.05230, 68.7% kcal of carbohydrate, 18.7% protein, 12.6% fat, with 34% sucrose/weight), and HGD+Curc (#TD.05230 + 0.2% curcumin in diet, equivalent to 1 g/day in humans, achieved by replacing 2.0 g/kg of corn starch content in the HGD formulation with isolated curcumin); the composition of the experimental diets is provided in Supplementary Table 13. The supplementation of 0.2% curcumin has been utilized in animal studies (50, 51) and the human equivalency of 1 g/day falls in the range for recommended supplementation (52, 53). Food and water were administered ad libitum and consumption of experimental diets were monitored by lab members.

Body weight was recorded at the conclusion of the 12-week dietary intervention period and terminal blood was collected via ventricular puncture following an 8-h fasting period and stored at −80°C. Total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), insulin, and glucose levels were measured from fasted serum samples, which were isolated by centrifuging the blood at 1,500 × g for 10 min at 4°C. Levels of TC, TG, HDL-C, and LDL-C were determined via enzymatic assays from Fisher Scientific (Hampton, NH, United States) and precipitation separation assays from Abcam (Waltham, MA, United States) that were adapted for a microplate format. Serum glucose and insulin levels were measured by an enzymatic assay from Fisher Scientific (Hampton, NH, United States) and electrochemiluminescence from Meso Scale Discovery (Rockville, MD, United States), respectively, in accordance with manufacturer’s instructions. All serum analyses were carried out by the UC Davis Mouse Metabolic Phenotyping Center (MMPC) Metabolic Core.

At the time of sacrifice following the 12-week study period, mice were anesthetized by intraperitoneal injection of xylazine/ketamine with dosing based on the amount required to achieve the surgical plane of anesthesia (24–25.5 mg/kg xylazine and 216–229.5 mg/kg ketamine) and euthanized by exsanguination. Intact brains were collected rapidly and the region containing the temporal lobe and hippocampus was isolated and embedded with HistoPrep Frozen Tissue Embedding Media (Fisher Scientific, Pittsburgh, PA, United States) under RNase-free conditions. Hematoxylin staining and microscopy visualization of the medial temporal lobe allowed for the identification of the hippocampus and hippocampal neurons, as previously reported (54). The verified hippocampal region was then coronally cryosectioned (8 μm, Leica Frigocut 2800 n Cryostat, Leica Biosystems, Buffalo Grove, IL, United States), bound to charged RNA-free PEN Membrane Glass slides, treated with RNAlater^®-ICE (Life Technologies, Grand Island, NY, United States) for RNA preservation, and stored at −80°C for further analysis. A flowchart portraying the workflow of subsequent analyses of the hippocampus is provided in Figure 1.

Hippocampal cryosections were immersed in nuclease-free water and dehydrated in desiccant in preparation for the laser capture microdissection (LCM) of endothelial microvessels (<20 μm), which were identified via alkaline phosphatase staining with 5-bromo-4-chloro-3-indolyl phosphate/nitro blue tetrazolium chloride (BCIP/NBT) substrate (55). Extraction of the microvascular endothelium of the cryosectioned hippocampus via LCM was performed with direct microscopic visualization of the entire vessel wall using a Leica LMD6000 Laser Microdissection Microscope (Leica Microsystems, Wetzlar, Germany). The isolated microvessels largely represented the cornu ammonis CA1 and CA3 regions in dorsal segments of the hippocampus, though further regional specification was not conducted.

Laser-captured hippocampal microvessels (300 microvessels/mouse, n = 3 mice/group) were further utilized, starting with total RNA extraction via Arcturus PicoPure™ RNA Isolation Kit (Thermo Fisher Scientific, Santa Clara, CA, United States) following the manufacturer’s instructions. The Affymetix (Santa Clara, CA) protocol for RNA quantification with SYBR Green I and ROX™ Passive Reference Dye was conducted and the RNA quality of the LCM-derived microvessels was assessed with a Nanodrop spectrophotometer.

Once extracted, RNA from the hippocampal microvessels was further processed for transcriptomic analysis with Clariom D Mouse Arrays (one per mouse) that contained over 7 million probes for protein-coding and protein non-coding genes such as microRNAs, long non-coding RNAs, and small nucleolar RNAs (Thermo Fisher, Santa Clara, CA, United States). Extracted RNA (122.3 pg/mouse) was utilized for preparation of complimentary RNA (cRNA) and single-stranded cDNA (sscDNA) with the GeneChip^®WT Pico Kit (Thermo Fisher, Santa Clara, CA, United States). Subsequently, the sscDNA (5.5 μg) obtained from 20 μg cRNA was fragmented via uracil-DNA glycosylase (UDG) and apurinic/apyrimidinic endonuclease 1 (APE 1) and labeled by terminal deoxynucleotidyl transferase (TdT) with the biotin-linked DNA Labeling Reagent. The UC Davis Genome Center performed the hybridization, staining, and scanning of the arrays in accordance with the Thermo Fisher Scientific WT array hybridization protocol. The fragmented and labeled sscDNA samples were hybridized in the GeneChip™ Hybridization Oven 645 for 16 h at 45°C and then washed and stained with the GeneChip™ Fluidics Station 450. Finally, the microarrays were scanned with the GeneChip™ Scanner 3000 7G (Thermo Fisher Scientific, Santa Clara, CA, United States) and the Thermo Fisher Scientific Transcriptome Analysis Console software was utilized for quality control of the microarrays and data analysis. The gene expression data of the low glycemic diet (LGD) and high glycemic diet (HGD) can be found in Gene Expression Omnibus (GEO) dataset GSE185057 while the genomic data from the HGD+Curc microarrays is deposited under accession number GSE314833.

For microarrays, statistical analysis of microvessel transcriptomes was conducted using ANOVA ebayes (Thermo Fisher Scientific Transcriptome Analysis Console software, Santa Clara, CA) with false discovery rate (FDR) correction. Differentially expressed genes (DEGs) from the microarray with significant p < 0.05 were considered as significantly differentially expressed. Diet intervention effects on body weight, lipid levels, glucose, and insulin were expressed as means ± standard error of the mean (SEM). Statistical significance (p ≤ 0.05) was assessed using unpaired t-tests (GraphPad software, La Jolla, CA, United States), or the Mann–Whitney test for non-normally distributed data.

At the conclusion of the 12-week diet intervention period, overall body weight amongst the LGD and HGD controls did not differ and the HGD+Curc treatment did not have a significant effect (Figure 2A). Similarly, serum levels of total cholesterol (Figure 2B) and triglycerides (Figure 2C) were unchanged across the three study groups. Notably, HGD+Curc significantly increased high-density lipoprotein cholesterol (HDL-C) levels (91.39 ± 26.35 vs. 11.83 ± 14.43 mg/dL, p < 0.05) (Figure 2D) and decreased low-density lipoprotein cholesterol (LDL-C) (9.89 ± 5.86 vs. 1.98 ± 0.76 mg/dL, p < 0.05) compared to HGD alone (Figure 2E). In addition, HGD+Curc elevated circulatory insulin (108.84 ± 61.14 vs. 291.43 ± 112.59 mg/dL, p < 0.01) significantly (Figure 2F) but did not affect glucose levels (Figure 2G) in relation to LGD and HGD.

Statistical analysis of the microarray data revealed that the HGD treatment compared to LGD modulated 786 differentially expressed genes (DEGs) in male murine hippocampal microvessels (Supplementary Table 1). More specifically, 201 protein-coding genes and 65 non-coding genes were regulated by the HGD treatment; the latter category included 19 microRNAs (miRNAs), 16 long non-coding RNAs (lncRNAs), and 30 small nucleolar RNAs (snoRNAs). Of these characterized genes, 190 were upregulated and 76 were down-regulated by HGD intervention compared to LGD (Figure 3A). Furthermore, the remaining 520 DEGs were categorized as pseudogenes, multi-complex genes, or unassigned genes with symbols (known) or without symbols (unidentified).

Amongst those classified, a total of 201 protein-coding genes were differentially expressed by HGD relative to LGD, which were mostly upregulated (n = 185; fold change range 1.5–11.4), while a few underwent downregulation (n = 16; fold change range of −4.38 to −1.5) (Supplementary Table 2). Bioinformatic analysis to uncover gene and pathway ontology indicated that the coding DEGs were involved in pathway regulation of major neurodegenerative diseases like Huntington’s, Parkinson’s, Alzheimer’s, and prion diseases and amyotrophic lateral sclerosis (ALS) as well as cellular metabolism (e.g., mitochondrial complex assembly, oxidative phosphorylation, and thermogenesis) (Figure 3B).

The hippocampal microarray analysis indicated that the HGD treatment also differentially expressed non-coding RNAs such as miRNAs, lncRNAs, and snoRNAs in comparison to LGD. Firstly, the 19 modulated miRNAs were primarily downregulated (n = 18; fold change (fc) range of −3.93 to −1.5) while only one (mmu-miR-692) was upregulated with a fold change of 4.97 (Supplementary Table 3). Targets of the modulated miRNAs totaled 527 genes that were involved in phosphoinositide 3-kinase (PI3K)/v-akt murine thymoma viral oncogene homolog 1 (AKT), mitogen-activated protein kinase (MAPK), and Janus kinase/signal transducer and activator of transcription (JAK/STAT) signaling pathways and extracellular matrix (ECM) maintenance such as ECM-receptor interactions and cytoskeletal focal adhesion (Figure 3C). Furthermore, all 16 differentially expressed lncRNAs were downregulated by HGD relative to LGD with a fold change range of −6.23 to −1.51 (Supplementary Table 4). Identification of 542 genes targeted by differentially expressed lncRNAs were involved in neuronal function pathways like neural crest differentiation, nitric oxide (NO)/cyclic guanosine monophosphate (cGMP)/protein kinase G (PKG) mediated neuroprotection, and phosphodiesterases in addition to synaptic signaling (Wingless-related integration site (Wnt) and glutamatergic synapse regulation) (Figure 3D). All pathways significantly regulated (p < 0.05) by HGD intervention relative to LGD, organized by identification with the Kyoto Encyclopedia of Genes and Genomes (KEGG) and Wiki Pathways databases, have been provided in Supplementary Figure 1 for coding genes (1A) as well as gene targets for miRNAs (1B) and lncRNAs (1C). Finally, relative expressions of the 30 snoRNAs were primarily downregulated (n = 26; fc −3.27 to −1.51) with some upregulation present (n = 4; fc 1.92–19.32) (Supplementary Table 5). Relevant target genes and pathways of these snoRNAs were not unveiled with literature searches and bioinformatic tools.

Microarray analysis was also conducted with the curcumin supplemented group to uncover how this dietary polyphenol influenced the genome of the murine hippocampal microvessels that were exposed to a high glycemic load. Statistical analysis of the microarray data demonstrated that 1,887 DEGs were affected by HGD+Curc treatment relative to the HGD control group (Supplementary Table 6). The DEGs identified with gene symbols included 560 protein coding and 146 non-coding genes (42 miRNAs, 40 lncRNAs, and 64 snoRNAs). Within those classified into these categories, 168 genes were upregulated and 538 downregulated by HGD+Curc in comparison to HGD (Figure 4A). The remaining 1,181 DEGs were categorized as pseudogenes, multi-complex genes, immunoglobulin (Ig) variable chain genes, or unassigned genes with gene symbols (known) or without symbols (unidentified).

In this study following 12 weeks of intervention, HGD induced multi-level genomic changes in the hippocampal microvasculature independently from significant differences in body weight and levels of total cholesterol, HDL-C, LDL-C, triglycerides, insulin, and glucose compared to LGD, which is largely consistent with our previously published study in female mice (20). Under these circumstances, this high glycemic load likely triggered early, tissue-specific cerebrovascular changes that either preceded or occurred independently of systemic alterations in insulin, glucose, and other biochemical markers. Coding genes were primarily upregulated by HGD, mostly for neurodegenerative diseases such as Alzheimer’s, Parkinson’s, Huntington’s, and ALS in addition to cellular metabolic processes of oxidative phosphorylation and thermogenesis and other diabetic issues like cardiomyopathy and NAFLD. The onset of hyperglycemic complications, such as NAFLD, have been closely linked to cognitive impairment, mainly due to the chronic-low grade inflammatory environment as neuroinflammation can be induced via chemokine infiltration across the BBB and activation of microglia (79). Furthermore, development of atherosclerosis and cardiomyopathy can lead to microvascular dysfunction and hippocampal impairment due to decreased cerebral blood flow (80, 81). In line with the reported literature regarding neurological complications brought on by a high-glycemic dietary challenge (10), this study showed that these pathways were linked through genes involved in the assembly of mitochondrial complexes I (Ndufa13, Ndufb6, Ndufb7, Ndufc1), II (Sdhb), III (Cox7a2l, Uqcr11, Uqcrc1), and IV (Cox5a, Cox5b, Cox6a1). The upregulation of these mitochondrial complex genes induced by HGD suggests possible accelerated production of ROS via oxidative phosphorylation, though not directly studied here, that has been linked to neurodegeneration such as aberrant oxidative damage, protein aggregation (i.e., protein folding), and promotion of neuroinflammatory and apoptotic pathways (82).

Meanwhile non-coding (miRNAs, lncRNAs, and snoRNAs) were primarily downregulated by HGD compared to LGD. Targets genes of HGD-regulated miRNAs were involved in the interplay between kinase phosphorylation (JAK/STAT, MAPK, PI3K-Akt) pathways responsible for influencing the neuroinflammatory and oxidative stress environment of neurological diseases (83). More specifically, JAK/STAT and MAPK signaling has been linked to abnormal accumulation of tau-protein and amyloid-beta (Aβ) characteristic of Alzheimer’s disease (83, 84) while disrupted PI3K/Akt activity impairs neuroplasticity (83, 85). Targets of DE miRNAs were also associated with extracellular matrix (ECM) maintenance like ECM-receptor interactions and cytoskeletal focal adhesion, mainly linked through targeting of collagen subunits. These ECM-interactions, including the perineuronal nets (PNNs) of the central nervous system, are disrupted by increased degradation activity via matrix metalloproteinases (MMPs) (86, 87) and binding of the cell-to-ECM connective integrins to Aβ (86, 87) in contribution to neurodegenerative conditions as these interactions are crucial for long-term potentiation (LTP) in hippocampal neurons (86, 88). Furthermore, all DE lncRNA were downregulated whose gene targets were associated with neuroprotection and neural crest differentiation mediated through NO-cGMP-PKG and Wnt signaling pathways as well as glutamatergic synapse regulation. NO-cGMP-PKG signaling is essential for LTP as NO aids in regulation of neuroplasticity, memory, and hypothalamic responsibilities, but Aβ accumulation impedes NO production and cGMP/PKG downstream activity (89) while Wnt signaling can protect against mitochondrial dysfunction (90). Finally, the structurally modifying ribosomal 2’-O-methylation activity of C/D box snoRNAs has been linked to CNS disorders and neurodegeneration (91), though the Snord82 downregulated by HGD in this study has not yet been directly implicated. Altogether, the neurodegenerative effect in the hippocampal microvasculature by HGD was once again seen in this study, irrespective of whole-body diabetic markers.

Curcumin has gained traction in studies of its dietary bioactivity for neuroprotection in the central nervous system due to increasing evidence that it can act as a genomic and epigenetic modulator in multiple diseases like cancer, diabetes, and even neurodegenerative diseases (48, 92–96). Several studies have revealed that curcumin can impact large number of genes simultaneously, such as endothelial cells in vitro (97) or within the aorta of ApoE-/- mice (98), however, the multigenomic impact remains largely unknown. These epigenetic mechanisms by curcumin have included regulation of DNA methylation, histone modifications, and expression of non-coding RNAs like microRNAs, lncRNAs, and circular RNAs (92–96, 99–101). While the influence of curcumin has been documented in the brain within models of neurodegeneration (92, 100, 101) and diabetes (42, 92, 102, 103), even specifically in the hippocampus (104–106), they have focused on the temporal and/or hippocampal region as a whole and overall cognitive function. Therefore, these studies have not addressed the effect of curcumin on hippocampal microvessels, especially regarding epigenomic regulation. Notably, the differential expression of snoRNAs by curcumin in the cerebral endothelium has not been previously documented, thus this study highlights a potential new non-coding RNA-related level of epigenomic neuroprotection. Altogether, this current study emphasized the nutrigenomic influence of curcumin on hippocampal endothelial microvasculature subjected to a high glycemic dietary load within a wild-type model through multi-level modulation of coding mRNAs, putative transcription factors, and non-coding RNAs as summarized in Figure 11.

The first indication of the curcumin’s bioactivity in this study was depicted in the circulation as HGD+Curc raised HDL-C and lowered LDL-C level compared to HGD, suggesting a mild corrective effect on dyslipidemia, though there was no overall effect on body weight, TC, and TG. Also, HGD+Curc significantly elevated circulatory insulin compared to HGD, which indicated an insulinotropic (i.e., promotion of insulin secretion) response as curcumin has been reportedly capable of protecting pancreatic β–cell damage under hyperglycemic conditions and diabetic patients (107–110). Thus, the curcumin-induced elevation in serum insulin was likely due to preservation of pancreatic β-cell production/secretory capacity of insulin and reduction of lipotoxic stress on β-cells via improved lipid homeostasis. However, more direct measurements of pancreatic β-cell health, such as the homeostatic model assessment for β-cell function (HOMA-β), need to be performed to further investigate the impact of curcumin consumption on circulatory insulin under the conditions of this study.

The HGD+Curc-induced regulatory pathways of coding genes were predominately linked through the downregulation of oxidative phosphorylation with implications for curcumin-mediated inhibition excessive mitochondrial ROS formation. Structural characteristics of curcumin can directly quench reactive free radicals as the phenolic hydroxyl groups can act as hydrogen or electron donors to form phenoxyl radicals and the keto-enol moiety can promote antioxidant activity via chelation with redox-active metal ions (i.e., Cu^{2 +}, Zn^{2 +}, and Fe^{3 +}) (111, 112). Furthermore, curcumin can indirectly affect ROS levels by promoting the activity of endogenous antioxidant enzymes like superoxide dismutase (SOD), catalase (CAT), and glutathione peroxidase (GPx) via Nrf2 (111); these targets were not assessed in this study but would be important work for future investigations.

Differentially expressed genes related to distinct neurodegenerative diseases in this study leaned toward Alzheimer’s related neurofibril aggregation as HGD+Curc downregulated Aβ precursor protein (App), Aβ precursor-like proteins (Aplp1 and Aplp2) and Aβ precursor protein-binding, family B, member 1 (Apbb1). Interruption of Aβ aggregation was potentially facilitated by the decreased expression of Rho-associated coiled-coil containing protein kinase 1 (Rock1) by HGD+Curc and subsequent phosphorylation of App. A more direct inhibition of amyloidogenesis by curcumin was also possibly achieved here as the keto–enol ring and aromatic hydroxyl groups of curcumin reportedly can react covalently with the aromatic rings or form hydrogen bonds with amino acid residues with polar pockets of Aβ, respectively (113). Furthermore, tau-phosphorylation was potentially inhibited due to the decreased expression of calcium/calmodulin-dependent protein kinase IV (Camk4), glycogen synthase kinase 3 alpha (Gsk3a) as well as the increased expression of miR-142a-5p that targeted cyclin-dependent kinase 5 (Cdk5); while the CNS-associated Gsk3β isoform (114) was not identified by as a coding DEG, it was targeted by Gm20675. Gsk3 activity is also tied to insulin signaling in a negative-feedback loop as Gsk3-overexpression diminishes insulin-mediated glycogen synthesis and glucose homeostasis via pancreatic β-cell dysfunction, leading to peripheral insulin resistance, while insulin-receptor binding inhibits Gsk3 activation (115). Thus, this downregulation of Gsk3a and indirect action on Gsk3β may be tied to the potential insulinotropic capability of curcumin in this study. Targeting of caspase signaling, which has been linked to the neuronal apoptosis aspect of dementia (116, 117), was also observed in this study as Casp3 and Casp9 were targeted by lncRNA Gm20675 and miR-142a-5p/-199a-3p, respectively. Along with mitochondrial dysfunction, ALS pathology has been mainly linked to aggregation of misfolded SOD1 and transactive response DNA binding protein-43 (TDP-43, encoded by Tardbp) (118, 119), which were both downregulated by HGD+Curc.

Regulation of multiple pathways by HGD+Curc in this study indicated potential BBB preservation. Pathological angiogenesis (i.e., formation of new blood vessels), or neovascularization in the context of microvascular networks, has been implicated with the neurodegenerative disease progression and is heavily regulated by VEGF/VEGFR interactions and downstream signaling (120–122). Binding of VEGF to its receptor VEGFR can activate PI3K and Rho GTPases like Ras homolog gene family, member A (Rhoa) and cell division cycle 42 (Cdc42), which can stimulate actin cytoskeleton/ECM-remodeling in the endothelium by Rock1 and cofilin-1 (Cfl1) (123, 124). This angiogenic process appears to be partially inhibited as genes involved in PI3K signaling (Pik3ca, Pik3r3, Pik3c2a), Rho GTPases (Rhoa, Cdc42/Cdc42ep1), and cytoskeleton migration (Rock1, Cfl1) were downregulated by HGD+Curc. As mentioned previously, permeability of the vascular endothelium/BBB can also be affected by ECM-interactions such as focal adhesion degradation by MMPs and transendothelial migration of leukocytes facilitated by adhesion molecules (86, 125). The miRNA-mediated targeting of Mmp11 (let-7a-5p, let-7k, miR-98-5p, miR-6394) and Mmp24 (miR-142a-5p, -199a-3p, -665-3p) along with reduction of neural cell adhesion molecule 1 (Ncam1) and aforementioned Rhoa, Rock1, and PI3K players by HGD+Curc appear to moderately affect these interactions. Angiogenesis can also be influenced by hypoxia, seen in cases like ischemic stroke (126) and is facilitated by hypoxia-inducible factor-1 (HIF-1) and VEGF/PI3K signaling that all can contribute to a neuroinflammatory environment (125, 127). Hyperglycemia has been linked to upregulation of HIF-1 and loss of endothelial tight junctions in brain microvascular endothelium due to VEGF-overexpression (128). The HIF-1 complex (heterodimer composed of α and β subunits) can be stabilized by heat shock protein 90 (Hsp90ab1, Hsp90b) (129), which were reduced by HGD+Curc. Hypoxic signaling can also contribute to Aβ formation and tau hyperphosphorylation (130). Taken together, observed changes in the expression of genes are suggestive of a decrease in BBB permeability, a key factor in neurodegenerative disease development.

Downregulation of these VEGF-mediated angiogenic and BBB-remodeling pathways by HGD+Curc may also indicate a moderate anti-cancer functionality of curcumin by restricting blood flow to tumor cells and reducing cell migration via ECM-degradation related to metastasis (131, 132). PI3k signaling governs cell cycle progression/proliferation, adherence and migration (133, 134) while blunting the cell cycle arrest and apoptotic activities of forkhead box O (FoxO) signaling (134). Therefore, downregulation of players in PI3K pathway (Pik3ca, Pik3r3, Pik3c2a) and cell cycle mediators like cyclin D2 (Ccnd2, G1 to S), cyclin B1 (Ccnb1, G2 to M), and Ras homolog enriched in brain (Rheb) by HGD+Curc may indicate an inhibitory effect on the HGD-induced influence toward tumor cell proliferation. Reduction of genes related to glucose transporter type 1 (Glut1, encoded by Slc2a1), ATP synthase (Atp5a1, Atp5b, Atp5g2, Atp5g3), sodium/potassium (Na⁺/K⁺) ATPase (Atp1a1, Atp1a2, Atp1a3, Atp1b1, Atp1b2), and the vacuolar ATPase (Atp6v0a1, Atp6v0b, Atp6v0c, Atp6v0e2, Atp6v1c1, Atp6v1d) as well as non-coding targeting of ATP-binding cassette (ABC) subunits (Abcc3, Abcc5, Abcc10) by curcumin in this study may further indicate anti-cancer capabilities due to inhibition by limiting the energy needs of tumorous cells (135–138), though further investigation is warranted.

Curcumin and its derivative demethoxycurcumin as well as related phase I and II metabolites (dihydrocurcumin, hexahydrocurcumin, tetrahydrocurcumin, curcumin glucuronide, and curcumin sulfate) had significant binding interactions with one or more TFs potentially involved in regulation of coding DEGs identified in this study. The largest number of coding DEGs modulated by HGD+Curc interacted with the cAMP-responsive element binding protein (CREB) transcription factor, which has widely reported to be involved with neuroprotection and neuroplasticity (139). Inhibitory phosphorylation of CREB (serine-129) (140) by Gsk3β can lead to decreased hippocampal neurogenesis and activity of pro-survival genes like brain-derived neurotrophic factor (BDNF) (141, 142). On the other hand, CREB can be activated by PI3K/Akt at a different site (serine-133) (143, 144) and bind to the promoter region of VEGF (145) and thus is related to tumorigenic vascularization and proliferation. Notably, TFAP2A had the lowest, most significant binding energies with curcumin and related metabolites, except for tetrahydrocurcumin. TFAP2A has a dual role as it is involved in differentiation of cranial neural crest cells and inhibition of this TF can lead to the development of facial clefts, specifically branchio-oculo-facial syndrome (146–148), though it can also influence ECM remodeling via MMPs and promote angiogenesis via VEGF and HIF-1α in oncogenic conditions (149). In another study, curcumin inhibited the oncogenic TFAP2A-induced ECM remodeling in colorectal cancer via downregulation of genes in the ECM-receptor pathway (150). All curcumin-related compounds had significant binding efficacy with nuclear respiratory factor-1 (NRF1), not to be confused with nuclear factor erythroid 2-related factor 1 (Nrf1, encoded by Nfe2l1). NRF1 activity reportedly mediates mitochondrial biogenesis and alleviates Aβ-induced degenerative mitochondrial dysfunction (151), which reiterates the potential influence curcumin had on mitochondrial function in the hippocampal microvasculature under high glycemic exposure.

SP1 is another transcription factor that has been seen as a driver of angiogenesis in microvascular endothelial cells under hyperglycemic (152) or hypoxic conditions (153) while another SP family member (SP3) has been associated with regulation of BBB players like transferrin receptor and occludin (152). Though SP1 may be related to neuronal survival and synaptogenesis (154), it is also involved in Alzheimer’s disease through binding to promoter regions of App and its cleaving enzyme β-secretase (155). Histone deacetylases (HDACs) are responsible for chromatin compacting that represses gene transcription thus HDAC3, which is the predominantly expressed class I HDAC in the brain and highly expressed in the hippocampus, is involved in the silencing of genes essential for neuronal survival and plasticity (156, 157). Though HDAC3 can be involved in normal brain development by regulating neural progenitor cells, it has been implicated in the progression of neurodegeneration and neurotoxicity largely due to phosphorylation by Gsk3β and interactions with HDAC1 (156, 157). HDAC3 can be overexpressed within the hippocampus in diabetic conditions (158), which can promote BBB transendothelial permeability. As curcumin has been reportedly can inhibit activity of multiple classes of HDACs (159), HGD+Curc dietary intervention may have facilitated neuroprotective functionality through the inhibition of HDAC3. TCF12, is a member of the basic helix-loop-helix (bHLH) protein family that has been linked to the promotion of neurogenesis, primarily mesodiencephalic dopaminergic neurons, and heterodimerization with other bHLHs can promote neuronal differentiation during cortical development (160, 161). In an endogenous antioxidant mechanism that is commonly seen with curcumin bioactivity (162), small MAF proteins form heterodimers with nuclear factor erythroid 2-related factor 1 (Nrf2) in order to bind antioxidant response elements (AREs) in the promoters of target genes (163). Finally, FOXF1 is reportedly involved with embryonic development of gut-derived organs such as the intestine, stomach, liver, gallbladder, and lung (164) and has been linked to microvascular and endothelial health, but this was primarily in the context of lung development (165, 166). These coding DEGs regulated by this transcription factor may be involved in modulation of neurodevelopmental disorders like other members of the forkhead box family (167), but it has not yet been directly linked to neuronal function.

Multiple miRNAs differentially expressed by HGD+Curc compared to HGD have been reportedly involved in endothelial health in terms of BBB integrity, angiogenesis, and vascular inflammation. As stated previously, the let-7 network node of mmu-let-7a-5p, -let-7k, and -miR-98-5p connected to the largest subset of target genes which was notable as several members of the let-7 family have been linked to regulation of cerebrovascular inflammation and angiogenesis (168). Of these, miR-let-7a and miR-98-5p have been reported to help preserve BBB integrity via prevention of tight junction loss as well as inhibition of proinflammatory cytokine release and immune cell infiltration, even under the case of hyperglycemic stress with miR-let-7a (169). Another study found an anti-angiogenic influence of miR-let-7a by targeting the TGFβ pathway, particularly Tgfb3 (170) which was downregulated by HGD+Curc. An additional group of miRNAs involved in angiogenic regulation is the miR-181 family, of which miR-181d-5p was upregulated by HGD+Curc. Particularly, miR-181d-5p has been found to negatively regulate hyperglycemia-induced VEGF-mediated angiogenesis in human retinal microvascular endothelial cells (171) and promote blood-tumor barrier permeability in glioma endothelial cells, which may aid in delivery of chemotherapeutic drugs (172). Retinal neovasularization related to diabetic retinopathy was also targeted by miR-384-3p, which inhibited this angiogenic process by targeting hexokinase 2 (173). MiR-384-3p activity has also shown relevance in the context of Alzheimer’s disease as it has been reported to target App and its cleaving enzyme β-secretase (174). Regarding neuroinflammation, NF-κB signaling in the vascular endothelium has been targeted in other studies by miR-193b-3p in a direct manner through promoting NF-κB p65 acetylation and inhibition of HDAC3 (175), a TF identified in this study, in addition to an indirect manner through miR-199a-3p-mediated targeting of mTOR signaling that reduced NF-κB p65 phosphorylation and adhesion molecule expression related to leukocyte adherence (176). Several other miRNAs outside of this study have also been associated with neurodegeneration (177) as well as endothelial homeostasis and inflammation (178, 179).

In the scope of this study, pathways for targets of DE miRNAs by HGD+Curc participated in conjunction with coding genes for synaptic signaling while exclusive pathways were involved in endothelial ECM maintenance in terms of biosynthesis of N-glycans and glycoaminoglycans (GAGs). BDNF, a type of neurotrophin synthesized in high concentrations within neuronal cell bodies and glia of the hippocampus, is important for synaptogenesis, synaptic plasticity/LTP, and neurotransmitter release (180). Multiple integrative downstream pathways are modulated by BDNF following cleavage and bondage to tropomyosin receptor kinase B (TrkB, Ntrk2) receptors, which can recruit SHC-transforming protein 3 (Shc3), that are all involved in neurotransmitter release such as PI3K-Akt/mTOR (neuronal survival) and MAPK/ERK (phosphorylation of synaptic vesicles) (180). Additionally, neuroprotective BDNF activation can inhibit Gsk3β through phosphorylation by dedicator of cytokinesis 3 (Dock3) (181). These BDNF-related processes were influenced by HGD+Curc through non-coding targeting of Ntrk2/TrkB by Gm16121, Shc3 by miR-142a-5p, -384-3p, -665-3p, and Dock3 by let-7a-5p/let-7k/miR-98-5p, miR-6394, miR-665-3p. However, BDNF is upregulated in a tumor environment due to its influence of pro-survivability cascades like PI3K/Akt/mTOR and MAPK (182).

GAGs and N-glycans are both essential components of the neurovascular ECM that regulate vascular homeostasis and cellular communication as GAGs primarily form the protective endothelial glycocalyx layer (EGL) (183), while N-glycans are attached to transmembrane proteins (i.e., N-glycosylation) to modulate protein folding, cell signaling, and adhesion (184). GAGs involved in the regulation of the PNNs of central nervous system ECM influenced by HGD+Curc were heparan sulfate and chondroitin sulfate as miRNAs (let-7a-5p, let-7k, miR-98-5p, -142a-5p, -199a-3p, -6394) targeted genes related to synthesis like chondroitin sulfate synthase 3 (Chsy3) and sulfation enzymes such as N-deacetylase/N-sulfotransferase 2 (Ndst2), heparan sulfate-glucosamine 3-O-sulfotransferase 3A1 (Hs3st3a1) and carbohydrate sulfotransferase 3 (Chst3) which may indicate an influence of curcumin on the glycocalyx aspect of the BBB. Sulfation of GAGs to generate heparan sulfate and chondroitin sulfate contribute to the protective negative charge of the EGL that helps regulate permeability of charged molecules (185), but dysregulation of sulfation patterns are linked to tumor migration and amyloid aggregation (186, 187). Furthermore, miRNAs (let-7a-5p, let-7k, miR-98-5p, -384-3p, -142a-5p) targeted N-glycosylation facilitators like dolichyl-diphosphooligosaccharide protein glycosyltransferase (Ddost), phosphomannomutase 2 (Pmm2), and dolichyl-phosphate beta-glucosyltransferase (Alg5). More to the point of regulating cell signaling of glycoproteins, DE miRNAs (let-7a-5p/let-7k/miR-98-5p, miR-6394) targeted β-galactoside alpha-2,6-sialyltransferase 1 (ST6Gal1) that influences tumorigenic cell adherence by adding sialic acid to glycoproteins (188). Altogether, targets of miRNA differentially expressed by HGD+Curc portrayed a potential versatile regulation of curcumin on aspects like BBB permeability, ECM-regulation and synaptic signaling.

Pathways unique to the gene targets of lncRNAs differentially expressed by HGD+Curc were predominantly involved in pathways regarding neurodevelopmental disorders, rather than neurodegenerative. Such disorders like autism as well as fragile X and Rett syndromes can be associated with mitochondrial dysfunction (189), but findings from this study highlight a link through glutamatergic synaptic signaling as subunits of N-methyl-D-aspartate (NMDA) receptors (Grin2A, Grin2B, Grin2D) were targeted by 6–14 lncRNAs differentially expressed by HGD+Curc. Important for synaptic plasticity and LTP for memory, glutamatergic NMDA receptor interactions can be regulated through the NO/cGMP pathway (190) but can be impaired by MTHFR deficiency, which is linked to increased phosphorylation of hippocampal Aβ precursor protein due to dysfunctional folate metabolism (191). Craniofacial development was also associated with DE lncRNAs through TFAP2A, the TF that significantly interacted with all curcumin-related metabolites, which was targeted by Gm16084. Again, lncRNA targets complementarity participated in pathways of neurodegeneration through the inclusion of these NMDA receptor subunits and the aforementioned targeting of Gsk3β and Casp3 by Gm20675. Regarding the C/D box snoRNAs, Snord82 has been reported to be a potential tumor suppressor of prostate cancer (192) was upregulated by our HGD+Curc group, which was a reversal of HGD regulation. Additionally, Snord16a is a potential biomarker for colon cancer (193) and Snord59a is reported to be a tumor immune infiltration-associated snoRNA (194), which were downregulated and upregulated, respectively, by HGD+Curc. However, further studies for relevancy of these Snords in the context of brain cancer would be necessary.

This study showed that curcumin supplementation within a high glycemic diet (HGD+Curc) may have a dual role of moderate neuroprotection and potential anti-tumorigenicity in our model of hippocampal microvasculature, though these often involve regulation of the same pathways in opposing directions. Since consequences of high glycemic diet consumption are complex, regulation of curcumin may have acted in a multi-faceted manner and could be elucidated in direct models of neurodegeneration and brain cancer, which has been reviewed previously (38, 43–45, 195–200). Studies investigating curcumin have included multiple models of aging, Alzheimer’s, Huntington’s Parkinson’s, multiple sclerosis, and ischemic/hemorrhagic stroke (43–45, 195–197). Additionally, these studies have included direct measurements of Aβ aggregation, mitochondrial function, and antioxidant/anti-inflammatory markers as well as cognitive tests (43–45, 195–197). Furthermore, some clinical studies have been conducted to assess the effect curcumin on cognitive function in healthy/non-demented older adults (201) as well as Alzheimer’s and Parkinson’s patients (43, 201). Studies of neuroprotection have also been conducted and included in these reviews that involved curcumin nanoparticles (45) and other curcumin related metabolites mentioned in this study (202) with an emphasis on tetrahydrocurcumin (203), demethoxycurcumin (202, 204), and hexahydrocurcumin (205). Finally, the anti-diabetic effect of turmeric and curcumin have been widely reviewed in multiple models (38, 47, 107).

Some limitations should be addressed as the results displayed are representative of the bioinformatics resources utilized and available at the time of the study and subsequent identification of target genes and pathways may vary with the use of other databases and gene ontology analysis tools. Large amounts of DEGs found in the hippocampal microarrays were either miscellaneous (i.e., pseudogenes, multi-complex, immunoglobulin (Ig) variable chain genes) or unidentified. Some of the DEGs common to both HGD/LGD and HGD+Curc comparisons represented in Figure 10 were unidentified, as indicated by only their Affymetrix IDs, leaving pathway enrichment analysis to be done with the remaining identified genes. It is important to note that conclusions about full regulation of the identified pathways cannot be made definitively as the DEGs and non-coding targets characterized within this study via microarray analysis and bioinformatic tools were not totally comprehensive in their respective pathways. Regarding the overall model, the HGD in this study did not appear to induce a hyperglycemic status systemically as changes in body weight and serum TG, TC/HDL-C/LDL-C, TG, insulin and glucose levels compared to LGD were not observed. Studies with endpoints of hyperglycemia and insulin resistance in mice have utilized a high fat diet alone or combined high-fat, high-sucrose diets for short-term studies (e.g., < 16 weeks) (206–209) or high-sucrose diets alone for extended study periods (e.g., 55 weeks) (210). As the diets in this study were isocaloric (∼3.6–3.7 kcal/g) with similar fat content (12.6–13.0 % kcals), the lack of changes in serum metabolic markers may indicate that the consumption of a high-sucrose diet alone at the starting age of 20 weeks and a duration of 12 weeks without the additional stressor of excess calories/fat were not sufficient to induce systemic hyperglycemia and insulin resistance. Additionally, the study was conducted with only male mice so further experimentation to account for sex differences in the analysis for bioactivity of curcumin alongside a high glycemic diet would be needed. Lastly, no direct measurements of abnormal protein aggregation associated with neurological diseases like Aβ, SOD1, α-synuclein were measured.