Alzheimer amyloid-β- peptide disrupts membrane localization of glucose transporter 1 in astrocytes: implications for glucose levels in brain and blood

Abstract

Alzheimer’s disease (AD) is associated with disturbances in blood glucose regulation, and type-2 diabetes elevates the risk for dementia. A role for amyloid-β peptide (Aβ) in linking these age-related conditions has been proposed, tested primarily in transgenic mouse lines that overexpress mutated amyloid precursor protein (APP). Because APP has its own impacts on glucose regulation, we examined the BRI-Aβ42 line (“Aβ₄₂-tg”), which produces extracellular Aβ_1–42 in the CNS without elevation of APP. We also looked for interactions with diet-induced obesity (DIO) resulting from a high-fat, high-sucrose (“western”) diet. Aβ₄₂-tg mice were impaired in both spatial memory and glucose tolerance. Although DIO induced insulin resistance, Aβ_1–42 accumulation did not, and the impacts of DIO and Aβ on glucose tolerance were merely additive. Aβ₄₂-tg mice exhibited no significant differences from wild-type in insulin production, body weight, lipidemia, appetite, physical activity, respiratory quotient, an-/orexigenic factors, or inflammatory factors. These negative findings suggested that the phenotype in these mice arose from perturbation of glucose excursion in an insulin-independent tissue. To wit, cerebral cortex of Aβ₄₂-tg mice had reduced glucose utilization, similar to human patients with AD. This was associated with insufficient trafficking of glucose transporter 1 to the plasma membrane in parenchymal brain cells, a finding also documented in human AD tissue. Together, the lower cerebral metabolic rate of glucose and diminished function of parenchymal glucose transporter 1 indicate that aberrant regulation of blood glucose in AD likely reflects a central phenomenon, resulting from the effects of Aβ on cerebral parenchyma, rather than a generalized disruption of hypothalamic or peripheral endocrinology. The involvement of a specific glucose transporter in this deficit provides a new target for the design of AD therapies.

1. Introduction

Alzheimer’s disease (AD) is classically characterized by the histopathologic presence of amyloid-β plaques and tau tangles within the AD brain, but there is a growing appreciation for associated somatic pathologies (Folch et al., 2018). For instance, the prevalence of either impaired glucose tolerance or type-2 diabetes (T2D) is higher in patients with AD (Janson et al., 2004; Ohara et al., 2011), although this finding is not consistent across cohorts (Thambisetty et al., 2013). Despite this tendency toward elevated blood glucose concentration ([Glc]_b), patients with AD exhibit a decrease in cerebral metabolic rate of glucose (CMR_glc) (Gabel et al., 2010; Hoffman et al., 2000). Although T2D is also associated with diminished CMR_glc (Baker et al., 2011), this disorder does not appear to exacerbate accumulation of amyloid-β peptide (Aβ) (Peila et al., 2002; Roberts et al., 2014), suggesting that T2D does not predispose to AD.

The mechanism by which diabetic conditions cause a decline in CMR_glc has been investigated in animal models and a limited number of human studies. A considerable number of studies have documented diminished expression and/or activity of glucose transporters in brain endothelium in response to chronic hyperglycemia and/or diet-induced obesity (DIO) (Cornford et al., 1995; Gjedde, 1981; Jais et al., 2016; Lutz and Pardridge, 1993; Mooradian and Morin, 1991; Pardridge et al., 1990). In most cases, this phenomenon has been traced to a reduction in endothelial expression of glucose transporter 1 (GLUT1), the primary glucose transporter in cerebrovasculature. Whereas other tissues are bathed in high glucose under diabetic conditions, the brain—owing to the restrictions of the blood-brain barrier (BBB)—suffers a reduction in CMR_glc when there is a reflexive downregulation of glucose transporters in the cerebrovasculature (Baker et al., 2011; Boyle et al., 1995; Roberts et al., 2014). Attempts to characterize transporter levels or function in AD and its models have been less consistent and detailed.

To explore relationships between AD and somatic glucose regulation, we probed the physiological status of a mouse model that accumulates Aβ in the CNS. Previous studies of these parameters have relied primarily on models of AD that overexpress mutant forms amyloid precursor protein (APP) and other APP peptide fragments formed from proteolytic cleavage. Our studies (Kulas et al., 2019) and those of others (Botteri et al., 2018; Needham et al., 2008; Tu et al., 2012) indicate that APP has direct effects on insulin and glucose regulation independent of Aβ. Because the vast majority of AD cases are sporadic and derive from accumulations of Aβ without APP mutation or overexpression, we investigated glucose regulation in a model that accumulates Aβ_1–42 alone: the BRI-Aβ42 mouse line (McGowan et al., 2005). For comparison with circumstances in which metabolic syndrome prevails, we invoked DIO through feeding “western” diet. We find that overproduction of human Aβ_1–42 in mice is sufficient to produce reductions of peripheral glucose tolerance and CMR_glc. This glucose perturbation was independent of insulin resistance, insulin deficiency, obesity, an-/orexigenic hormones, and overt inflammatory events. Through a diverse, multifaceted set of experiments, we conclude that the elevation of blood glucose levels resulting from Aβ production is explained primarily by reduced transport of glucose into the brain due to aberrant subcellular localization of GLUT1 in brain parenchyma with no apparent involvement of vascular transporters.

2. Materials and methods

2.1. Human tissue

Specimens were obtained from a brain bank maintained locally which catalogs and characterizes autopsy material from the University of Arkansas for Medical Sciences UAMS Department of Pathology and the Central Arkansas Veterans Healthcare System Pathology Service. They were acquired from autopsies of individuals diagnosed as AD (without Parkinson’s disease or mixed Lewy body pathology) or age-matched controls (AMC), as per National Institute on Aging-Reagan guidelines (Montine et al., 2012), which incorporate consortium to establish a registry for Alzheimer’s disease evaluation of neuritic plaque density and Braak & Braak staging of neurofibrillary tangles. Samples were obtained by 3-D dissection of autopsy-provided material from left prefrontal cortex, which were flash frozen in N₂ (l) and stored at −80 °C. Specimens did not qualify as human subject research as per US Department of Health and Human Services Exemption 4. The N = 10 for each group. Average age of the AD cases was 81.1 year old, and they were 40%/60% female/male; the AMC cases were 83.0 year old and 30%/70% female/male; each group was 9:1 Caucasian:African-American. Postmortem intervals were 5 and 9 hours for AD and AMC, respectively.

2.2. Animals

Each protocol was approved by the Institutional Animal Care and Use Committee of the Central Arkansas Veterans Healthcare System (CAVHS) or of the UAMS. Mice were housed in a 12-h light/dark cycle at 23°C at the AAALAC-certified CAVHS vivarium unless otherwise noted. A normal diet (ND) and water was provided ad libitum (kcal: 22% protein, 16% fat, and 62% carbohydrates; LabDiet JL Rat and Mouse/6FOvals w/Hysil; 5k67-RHI-W 30). Mice were weaned at age 25 days or older and housed at a cage density of ≥82 cm/mouse of floor space. Both sexes were evaluated in some assessments, but females showed no effect of genotype or diet, so only data from male mice are presented here.

Hemizygous BRI-Aβ42 (Aβ₄₂-tg) transgenic mice [(McGowan et al., 2005); akin to MMRRC stock: #34842] were provided by Dr Todd Golde (University of Florida). The transgene is a fusion construct in which the Aβ_1–42 sequence replaces the C-terminal 23 amino acids of the BRI British dementia protein, directing a proteolytic liberation of Aβ_1–42 into the extracellular space, without overexpression of APP. The prion promoter construct used in this line results in transgene expression primarily in the CNS (Borchelt et al., 1996), and plasma levels of Aβ in the BRI-Aβ42 line are ~100-fold below those in an APP-transgenic line (Levites et al., 2006). All analyses were performed on mice hemizygous for the transgene. With the exception of the adrenalectomy study, all mice were backcrossed to C57BL/6N wild-type (WT, Harlan) to produce WT and Aβ₄₂-tg littermates that were Nnt^+/+ (as C57BL/6J are Nnt^−/−). Adrenalectomized mice were obtained from the Jackson Laboratory (Bar Harbor, ME) in a stock that had been maintained in C57BL/6J (JAX stock: #007182). Genotyping was performed by PCR for the transgene with the following primers: 5′- AAG GCT GGA ACC TAT TTG CC −3′ and ‘5- TAG TGG ATC CCT ACG CTA TG −3’ (307-bp product). A T-cell receptor gene was used in a positive-control reaction with the following primers: 5′- CAA ATG TTG CTT GTC TGG TG-3′ and 5′- GTC AGT CGA GTG CAC AGT TT −3’ (200 bp).

2.3. Diet-induced obesity

To test for interactions between the glycemic effects of Aβ and those of DIO, we feed some mice western diet (WD; EnvigoTeklad TD.88137; kcal: 15% protein, 42% fat, 43% carbohydrate), an obesogenic nutritional source high in sucrose (34% by weight) and milk fat. Mice were split into four experimental groups: WT ND, WT WD, Aβ₄₂-tg ND, and Aβ₄₂-tg WD. At week 0, all mice were 6 week old ±3 days. Glucose tolerance tests (GTTs) and insulin tolerance tests (ITTs), both detailed below, were administered on alternating weeks to reduce handling stress effects. After the first ITT on week 1, half of each litter had the normal chow diet replaced with the WD. They were maintained on the diet until euthanasia at diet week 20. In addition to GTTs and ITTs, these mice were used for data on insulin production, hypothalamic gene expression, and circulating and tissue levels of hormones and cytokines.

2.4. Glucose, insulin, and pyruvate tolerance testing

Glucose tolerance was assessed through an intraperitoneal administration of glucose followed by intermittent monitoring of blood glucose concentration ([Glc]_b) for 120 minutes. Before glucose tolerance testing, mice were fasted (with drinking water provided) for 5 h. The fasting was conducted primarily during the mornings, typically starting between 8 am and 9 am, with GTTs beginning between 1 pm and 2 pm. Mice were weighed, and 2 g/kg of D-(+)-glucose (Sigma, St. Louis, MO) in sterile water (Baxter Healthcare Inc, Deerfield, IL) was administered (190–230 μL total volume) intraperitoneally. The glucose solution was filtered using a 0.24-μm syringe filter before use. An AlphaTRAK 2 glucometer (Zoetis Inc, Kalamazzo, MI) was used to sample blood from the tip of the tail just before injection and at 15, 30, 60, and 120 minutes after injection. Total area under the curve (tAUC) was calculated from a baseline of 0 mg/dL using the trapezoid rule. Incremental area under the curve (iAUC) was calculated by rendering the y-axis as the difference of each time point from the initial glucose reading for a given animal. Pyruvate tolerance testing was performed in a similar manner using 1 g/kg of 10% sodium pyruvate diluted in phosphate-buffered saline (PBS).

Insulin tolerance testing was performed in a similar manner as GTTs; however, mice were not fasted unless otherwise indicated. Insulin (Humulin R; Lily USA, Indianapolis, IN) was diluted in saline (Molecular Biologicals International, Irvine, CA) and administered ip at 0.35 U/kg. Because insulin evokes a downward deflection in [Glc]_b, the integral reported is area over the curve (AOC), calculated by rendering the y-axis as the absolute value of the difference of each time point from the initial glucose reading for a given animal.

2.5. Comprehensive lab animal monitoring system (CLAMS)

CLAMS cages (Oxymax; Columbus Instruments, Columbus OH) were used to assess several aspects of activity and metabolism, including food consumption and indirect calorimetry. A unique cohort of 13- to 17-week-old Aβ₄₂-tg mice and their WT littermates, distinct from those used in other assays, was evaluated in this paradigm. Mice were housed individually inside CLAMS cages at the UAMS facility with water and food ad libitum under a 14/10 hours light/dark cycle. The mice were acclimated for 21 hours before starting data collection, and data were collected for 48 hours. Body weights were obtained before and after the CLAMS session. The instrument simultaneously measures CO₂, O₂, food consumption, and 3-D physical movement. Energy expenditure and respiratory exchange ratio were calculated from the gas measurements. Activity was measured as the number of infrared beam breaks. The beams were stationed in the Z dimension to create two X-planes, one above the other, allowing inclusion of movement in the Z direction. Grooming and ambulation were recorded in aggregate based on the number beam breaks in the lower plane. Ambulation was differentiated by consecutive beam breaks in the lower plane and also recorded. Beam breaks in the upper plane (Z-axis) accounted for rearing and jumping movements. Sleep was defined by relatively prolonged periods of inactivity and has been validated with an 88%–94% agreement with electroencephalogram and video analysis (Pack et al., 2007).

2.6. Morris water maze (MWM)

A unique cohort of mice housed at the UAMS facility was assessed for spatial memory via MWM as described previously (Alexander et al., 2018). At 10 weeks of age, 14 mice per genotype were randomized into two groups: ND or WD (above); groups of Aβ-tg mice and WT mice were all from the same litters. At 16 weeks of age, the mice were trained to locate a clearly marked platform (visible-platform training, days 1 and 2), confirming that groups were similar regarding swimming and learning ability. Start locations were changed for each trial. Swimming paths are recorded with an EthoVision XT video tracking system (Noldus Information Technology). Mice were subsequently trained to locate the platform hidden beneath the surface of the opaque water (hidden-platform training, days 3–5). Mice were trained 4–8 times a day for 5 days. After training, a probe trial was conducted on day 6 comprising 60 seconds without a platform; the outcome measure was time spent in each quadrant of the pool. During the course of this testing, animals were maintained on their respective diets.

2.7. Assessment of adrenalectomized mice

To evaluate contributions of the hypothalamic-pituitary-adrenal (HPA) axis to the glycemic effects of Aβ, adrenalectomy was performed on BRI-Aβ42 mice [B6.Cg-tg(Prnp-ITM2B/APP695*42) A12Emcg/J; stock no. 007182] and WT littermates at 5.5 weeks of age by the Jackson Laboratory (Bar Harbor, ME); unoperated littermates were provided at the same time (6 per group). They were maintained on an ad libitum ND, and adrenalectomized mice were provided a 0.9% saline solution to counter the effects of mineralocorticoid deficiency on kidney function (Bristol and Drill, 1952). The animals were shipped to the CAVHS facility and allowed to acclimate for one week before testing. In one-week intervals, they underwent GTTs, ITTs, and a GTT in which blood was collected via retro-orbital eye bleed to measure insulin response to glucose challenge at 0 and 30 minutes. The following week all animals were euthanized. One Aβ₄₂-tg adrenalectomized and two WT adrenalectomized mice died during the second GTT. The DetectX Corticosterone Enzyme Immunoassay Kit (Arbor Assays) was used to compare circulating levels of glucocorticoids in serum of adrenalectomized mice at the time of euthanasia.

2.8. Tissue collection

Mice were rapidly anesthetized using intraperitoneal injection of 250 mg/kg sodium pentobarbital (Nembutal, Oak Pharmaceuticals or Diamondback Drugs). Blood was taken from the left and right ventricles of the heart in a heparinized syringe and placed on ice. A 50 μL aliquot was mixed with protease inhibitors. Blood was then centrifuged at 7500g for 15 minutes at 4 °C, and the resulting serum layer was carefully pipetted and stored at −80 °C. Mice were perfused with heparinized (50 U/mL) saline for approximately 5 minutes, when clear saline solution was leaving the body cavity and the liver appeared to blanch. The brain was extracted, and the hypothalamus, right cortex, and right hippocampus were flash frozen in liquid nitrogen. The left half brain was fixed by immersion in 10% formalin.

2.9. Comprehensive metabolic panel

The Piccolo Comprehensive Metabolic Panel and Piccolo Xpress chemistry analyzer were used to assess the concentration of alanine aminotransferase, albumin, alkaline phosphatase, amylase, calcium, creatinine, glucose, phosphorous, potassium, sodium, total bilirubin, total globulin, total protein, and blood urea nitrogen in heparinized whole blood. Blood was collected during euthanasia into a sodium heparinized tube, and 100 μL was added to each panel disc.

2.10. ELISA for insulin and Aβ

Serum was collected from the tail vein of the cohort of mice used for DIO analysis, including the corresponding ND animals (above). Insulin levels were also assessed in eye bleeds from the cohort used for adrenalectomy, including the unoperated animals (above). Mice were fasted and injected with glucose as per the GTT assays. Blood was collected before and 30 minutes after injection, and it was allowed to clot before collection of serum by centrifugation. Insulin was measured using an Insulin Mouse Ultrasensitive ELISA (Crystal Chem Inc) as per manufacturer’s instructions for the wide range assay (0.1–12.8 ng/mL). Standards and samples were assayed in duplicate. Data points for individual samples in which a coefficient of variability between duplicates was greater than 20% were excluded. Aβ_1–42 was extracted from brain, plasma, and other tissues in three steps to yield fractions soluble in buffered saline, Triton X-100, and formic acid (McDonald et al., 2012). The peptide was measured in each fraction by Quantikine ELISA Kit (R&D Systems), as per manufacturer’s instructions.

2.11. Cerebral metabolic rate of glucose (CMR_glc) measurements

To test whether Aβ₄₂-tg mice showed deficits in CMR_glc as seen in other AD mouse models and AD itself, we assayed the accumulation of [³H]2-deoxyglucose (2-DG) and its primary metabolite in the cerebral cortex of Aβ₄₂-tg mice and WT littermates. Unanesthetized 15-week-old mice were acclimated to a restraint tube by 45-second restraint sessions conducted daily for one week. On the day of assay, 100 μCi/kg [³H]2-DG and 10 μCi/kg [¹⁴C]sucrose was injected into the lateral tail vein. Blood was collected from a separate (distal) tail site at 3, 7, 11, and 15 minutes after injection into heparinized tubes. Mice were euthanized via decapitation immediately after the final time point, trunk blood was collected in a tube containing heparin, and brain was quickly dissected and snap-frozen. Both hemispheres of the cerebral cortex were homogenized in radio-immunoprecipitation (RIPA) buffer (150 mM NaCl, 50 mM Tris HCl, 1 mM EDTA, 1% Triton X-100, 0.5% sodium deoxylcholate, 0.1% SDS, pH 7.4) and extracted with 65% ethanol. Liquid scintillation counting was used to measure cpm of ³H and ¹⁴C in plasma and in the soluble and insoluble fractions of tissue. CMR_glc was calculated based on the study by Vallerand et al. (1987). Specifically, the extracellular volume (in μL/mg) was calculated by dividing the amount of brain ¹⁴C (dpm per milligram of tissue) by the amount of ¹⁴C in blood (dpm per μL of plasma) at the time of death. The amount of extracellular [³H]2-DG (dpm per mg of tissue) was obtained by multiplying the extracellular volume (in μl per mg of tissue) by the blood concentration of [³H]2-DG (dpm per μl of plasma) at the time of death. This blood concentration of [³H]2-DG (dpm per mg of tissue) was subtracted from the total concentration to obtain the parenchymal concentration of [³H]2-DG (dpm per mg of tissue).

2.12. Quantitative real-time PCR

Tissue was homogenized in Qiagen RLT buffer using a Lysing Matrix D tube (MP Biomedicals; component of FastRNA Green Kit) and the FastPrep-24 instrument (MP Biomedicals) on setting “6” for 2 × 20 seconds. Lysate was centrifuged at 16,300g for 1 minute. Samples were transferred to a new tube and spun again for 3 minutes before carefully adding supernatant to a gDNA Eliminator spin column of the RNeasy Plus Mini Kit (Qiagen), and the RNA was extracted following the kit instructions. The quality of the RNA was assessed using the Agilent 2100 Bioanalyzer. All samples had an RNA integrity number greater than 7. A two-step RT-PCR protocol was used; RT (cDNA synthesis) was completed using the ImProm-II Reverse Transcription System (Promega) and random hexamer primers. A portion of each sample was pooled and serially diluted to form a standard curve for absolute quantification. Samples were diluted to fall within the center of the standard curve and combined with primers (Supplemental Table 1) and SYBR Green PCR Master Mix (Applied Biosystems) for amplification and detection using an ABI 7900HT Fast Real-Time PCR System. Rplp0, reported to be a stably expressing transcript in the hypothalamus under various diabetic conditions (Li et al., 2014), was used as a reference gene in all experiments. All reactions used the following three-stage protocol: Incubation/denaturation: 50 °C for 2:00/95 °C for 10:00; PCR amplification for 50 cycles: 95 °C for 0:15, 55 °C for 0:15, and 72 °C for 1:00; and melt curve: 95 °C for 0:15, 60 °C for 1:00, and 95 °C for 0:15.

2.13. GLUT1 determinations by SimpleWes

Glucose transport and utilization can be dramatically affected by changes in the subcellular distribution from intracellular stores to the plasma membrane. To determine if this contributed to differences in glucose uptake in AD or in Aβ₄₂-tg mice, we optimized conditions for separation of plasma membrane fractions and for resolution/detection of various GLUT isoforms. Tissue from cerebral cortex of human subjects and most mouse brains was pulverized in with mortar and pestle chilled in liquid nitrogen, and the powder was mixed to homogeneity.

For a subset of mice, brain microvessel isolates were prepared, making use of both cerebral hemispheres. After dissection from the skull and other brain regions, the hemispheres were carefully stripped of meningeal tissue while in ice-cold PBS, then Dounce-homogenized on ice in 0.6 mL Buffer D [2.7 mM KCl, 1.46 mM KH₂PO₄, 136.9 mM NaCl, 8.1 mM Na₂HPO₄, 5 mM D-glucose, 1 mM sodium pyruvate, 0.9 mM CaCl₂, 0.5 mM MgCl₂, and 5 mg/mL protease inhibitors (Pierce Mini Tablets, EDTA-free), pH 7.4]. The homogenate was mixed 1:1 with 30% Ficoll and centrifuged at 5800 × g for 15 minutes at 4 °C. The supernatant was discarded, and the pellet was resuspended in 1 mL of another 1:1 mixture of Buffer D and 30% Ficoll. After another centrifugation at 5800 × g for 15 minutes at 4 °C, the supernatant was discarded and the pellet was resuspended in Buffer D and centrifuged at 1000 × g for 3 minutes at 4 °C. The supernatant was discarded, and the pellet was stored at −80 °C.

Pulverized tissue, microvessel isolates, and cultured astrocyte pellets were processed via the Minute Plasma Membrane Protein Isolation and Cell Fractionation Kit as per the instructions from the manufacturer (Invent Biotechnologies, Plymouth MN) to separate plasma membrane from the remainder (cytosol plus organelle membranes). We confirmed identity of the fractions by exclusive detection of plasma membrane calcium ATPase, sarcoplasmic/endoplasmic calcium ATPase, and β-actin in their expected fractions.

Fractions were analyzed for GLUT1 protein using the SimpleWes capillary electrophoresis instrumentation (Protein Simple, San Jose CA). Fractions from mouse cortical preparations were loaded at 1.4 mg/mL; anti-GLUT1 antibody (rabbit monoclonal EPR3915; abcam, Cambridge MA) was loaded at a 1:50 dilution. Fractions from mouse microvessel isolates were loaded at 0.2 mg/mL, and remainder fractions were loaded at 0.47 mg/mL. For human samples, plasma membrane fractions were loaded at 1.8 mg/mL, and remainder fractions were loaded at 0.32 mg/mL; antibodies were loaded recognizing GLUT1 (1:50), −3 (1:100), or −4 (1:20). For samples from astrocyte cultures, plasma membrane fractions were loaded at 0.22 mg/mL, and remainder fractions were loaded at 0.41 mg/mL; anti-GLUT1 antibody was loaded at a 1:50 dilution. Gaussian peak integration was performed on the chemiluminescence values along the length of the capillaries, with peaks at ~45 kDa and ~55 kDa analyzed separately for GLUT1.

2.14. Primary cultures of astrocytes

Mixed-glia cultures were established from neonatal Sprague-Dawley rats as described previously (McMullan et al., 2012). The cultures were maintained in the Minimal Essential Medium (Earle’s salts) supplemented to 10% with fetal bovine serum for 10–14 days. Microglia were removed by physical dislodgement, including aggressive lavage. Astrocytes were then trypsinized, subcultured in 15-cm plates in the Minimal Essential Medium /fetal bovine serum. Characterization of these cultures with cell type-specific antibodies demonstrated them to be ≥95% astrocytes (Supplement Fig. 1). The cultures were treated after they had reached 90% confluency. Lysates were prepared with the Minute Plasma Membrane Protein Isolation and Cell Fractionation Kit as per the instructions from the manufacturer (Invent Biotechnologies, Plymouth MN) to produce plasma membrane and cytosolic fractions.

Aβ_1–42 (Anaspec) was dissolved at 1 mM 4 hours in hexafluoroisopropanol, aliquoted, and evaporated to dry; it was stored under desiccant at −80 °C. For aggregation, an aliquot was dissolved at 2 mM in anhydrous dimethylsulfoxide, and then diluted to 150 μM in PBS at 4 °C. The peptide was incubated in PBS at 4 °C for 24 hours before use.

2.15. Immunofluorescence

Primary astrocyte cultures were plated onto glass coverslips coated with rat tail collagen. After treatment, the cultures were fixed for 15 minutes in ice-cold PBS containing 4% paraformaldehyde, followed by three washes with PBS. Some cultures were permeabilized by application of PBS containing 0.3% Tween-20 at room temperature for 30 minutes, followed by three washes with PBS; other cultures were subjected immediately to blocking without permeabilization. Blocking comprised exposure for 1 hour to blocking solution: PBS containing 1% fatty acid-free bovine serum albumin. Primary antibody against GLUT1 (1:300; GT14-A, Alpha Diagnostic International, San Antonio TX), glial fibrillary acidic protein (1:1000; ab53554, Abcam), Iba1 (1:1000; 013-27,691, FUJIFILM Wako), pan-neuronal marker (1:25; MAB2300, Millipore-Sigma), phosphotyrosine (1:100; 05-947, Millipore-Sigma), S100B (1:200; NCL-S100p, Novocastra), or VE-cadherin (1:25; #36-1900, ThermoFisher) was applied for 2 hours at room temperature in blocking buffer, followed by three washes with PBS. Secondary antibodies were Alexa Fluor 488 donkey anti-rabbit (for GLUT1, S100, and VE-cadherin) and Alexa Fluor 594 donkey anti-goat (for GFAP) or anti-mouse (for Iba1, pan-neuronal, and phosphotyrosine), applied for 1 hour at room temperature in blocking buffer (1:500), followed by three washes with PBS. Coverslips were mounted onto microscope slides with ProLong Gold Antifade Mountant containing DAPI (ThermoFisher). Images were acquired with a Nikon Eclipse Ti using an Illuminator Sola and excitation/emission (nm) pairs of 360/460 (DAPI), 480/535 (AF488), and 560/630 (AF594). The camera was an Andor Zyla 5.5 sCMOS. For quantitative comparisons of GLUT1 across treatment conditions, consistent illumination intensity and acquisition times were used for all images.

2.16. Statistical analysis

For comparisons between two groups, Student’s t-test with Welch’s correction was used. One-way ANOVA with Bonferroni post hoc was used in assessing groups of 3 or more. Two-way ANOVA with Bonferroni post hoc was used in assessing multiple comparisons over time. Grubb’s test was used to assess for outliers using GraphPad’s online assessment tool (https://www.graphpad.com/quickcalcs/Grubbs1.cfm). All statistics were carried out using GraphPad Prism 7 software. Enzyme-linked immunosorbent assay (ELISA) curves were assessed using a four-parameter logistic regression curve (https://www.myassays.com/four-parameter-logistic-curve.assay).

3. Results

3.1. Spatial memory is impaired in Aβ₄₂-tg mice

Mouse models of Aβ accumulation in the CNS have shown impairments in spatial memory consistent with hippocampal deficits; several such models have also shown aberrations in regulation of peripheral glucose. However, many of these models rely on overexpression of APP, which appears to have effects on insulin and glucose dynamics itself. To exclude such contributions, we examined glucose regulation in the BRI-Aβ42 mouse (“Aβ₄₂-tg”), which releases Aβ_1–42 into CNS interstitial fluid without an APP transgene. This line also avoids any Aβ-independent effects of APP mutation; the latter may be more relevant for modeling familial AD than sporadic AD. Aβ₄₂-tg showed a deficit in spatial memory. Specifically, WT mice trained in the MWM showed a preference for the target quadrant during a probe trial (removal of escape platform after training), whereas their Aβ₄₂-tg littermates did not (Fig. 1).

Fig. 1.

Cerebral Aβ accumulation and western diet impaired spatial memory. Aβ₄₂-tg mice (“Aβ-tg”) and wild-type littermates (“WT”) were maintained on normal diet (ND); additional mice of each genotype were also subjected to 6 weeks of DIO. The mice were trained in the MWM task and tested in a probe trial on day 6. Values represent the percentage of time spent in each quadrant of the pool during the probe trial. WT ND males spent significantly more time in the target quadrant than in any other quadrant, indicating retention of training (*p ≤ 0.05, **p ≤ 0.01).

3.2. Aβ overproduction influences systemic glucose homeostasis

Compared with WT littermates, Aβ₄₂-tg mice displayed impairments in an intraperitoneal GTT. Blood glucose levels ([Glc]_b) in Aβ₄₂-tg mice spiked to higher levels and were slower to recover, resulting in an increase in tAUC (Fig. 2A, B). After a 5-h fast, Aβ₄₂-tg mice also had a higher basal [Glc]_b and trended toward a higher basal [Glc]_b in the nonfasted (fed) state (Fig. 2C, D). The dynamic range of the response was calculated by subtracting the basal [Glc]_b for each mouse individually to arrive at an iAUC. This value was also elevated significantly in Aβ₄₂-tg mice (Supplement Fig. 2). The following week, an ITT was performed to measure insulin sensitivity in the same cohort of mice. Because an ITT detects a decrease from baseline, integration of this parameter was performed as AOC. Responses in ITTs were equivalent in both genotypes (Fig. 2E, F). Likewise, no differences were detected in pyruvate tolerance tests, a measure of hepatic gluconeogenesis (Supplement Fig. 3).

Fig. 2.

Peripheral glucose tolerance was impaired in Aβ₄₂-tg mice. Glucose tolerance testing (GTT) and insulin tolerance testing (ITT) were performed on Aβ₄₂-tg mice (“Aβ-tg”) and WT littermates on a ND or western diet (DIO) for 9 weeks. A: On both diets, Aβ₄₂-tg mice had significantly higher [Glc]_b than WT at 30 and 60 minutes (*p < 0.05, **p < 0.01). DIO elevated [Glc]_b levels within both genotypes (WT Normal vs. WT DIO: ^⧧⧧P < 0.01, ^⧧⧧⧧⧧P < 0.0001; Aβ-tg Normal vs. Aβ-tg DIO: ^##p < 0.01, ^####p < 0.0001). B: DIO increased total area under the curve (tAUC) in the GTT (***p < 0.001). Aβ₄₂-tg mice had greater tAUC than WT on the same diet (*p < 0.05). C, D: Fasting and nonfasting basal [Glc]_b were higher in Aβ-tg DIO than in either WT DIO or Aβ-tg ND (*p < 0.05, ***p < 0.001, ****p < 0.0001). E: ITT showed that the Aβ-tg DIO group had a significantly greater impairment than WT DIO (*p < 0.05, **p < 0.01, ***p < 0.001), but Aβ-tg ND was not different from WT ND. F: AOC in the ITT indicated that the WT DIO group had impaired responses to insulin compared with WT ND, and Aβ-tg DIO also showed impairment (*p < 0.05). All values are mean ± SEM.

3.3. DIO and Aβ have only additive effects on glucose intolerance

To test for potential interactions of Aβ with the diabetes-inducing aspects of obesity, WT and Aβ₄₂-tg mice were subjected to DIO through feeding a “WD.” DIO was manifest as significant weight gain in all mice fed WD, regardless of genotype (Fig. 3A). DIO was also accompanied by hyperlipidemia and hypercholesteremia; total blood cholesterol levels were 61.5 ± 8.5 mg/dL in ND mice and 185.5 ± 32.9 mg/dL in DIO mice (mean ± SEM). Impairments in GTTs and ITTs were observed in all DIO mice (Fig. 2A, B). Glucose intolerance developed over several weeks in DIO mice but was consistently present in Aβ₄₂-tg mice (Fig. 3B). Glucose tolerance was even further disrupted in DIO Aβ₄₂-tg mice, generally appearing as an additive effect of DIO and Aβ expression. Glucose levels in Aβ₄₂-tg mice were significantly different from WT at 60 minutes on either diet, but DIO had a stronger influence on glycemic rise overall. Basal [Glc]_b was elevated in DIO Aβ₄₂-tg mice in both fasting and fed states (Fig. 2C, D), consistent with metabolic syndrome and incipient diabetes. DIO did not alter the levels of Aβ in any tissue assayed (mean: 8.9 ng/mg protein in cerebral cortex; 0.45 nM in plasma). DIO produced insulin resistance in both WT and Aβ₄₂-tg, reflected in the small AOC (Fig. 2E, F). By contrast, overexpression of Aβ_1–42 did not impair insulin tolerance (AOC) in either dietary condition (Fig. 2F).

Fig. 3.

Progressive changes in DIO paradigm. Aβ₄₂-tg and WT littermates were fed a ND or western diet for 12 weeks (reflected in the abscissa label). A: Mice were weighed biweekly without fasting (****p < 0.0001 each DIO group vs. its ND counterpart). B: GTT was performed on weeks alternating with the body weight measurements. (*p≤0.03 vs. WT Normal, ^##p < 0.002 Aβ-tg Normal, ^†††p = 0.0005 vs. WT DIO).

We also tested the effects of DIO on memory performance. Consistent with prior reports (Jeon et al., 2012; Kasper et al., 2018; Lu et al., 2011; Mi et al., 2017), impaired performance was observed in DIO mice (Fig. 1). A combined effect of DIO and Aβ expression could not be detected because either alone removed essentially all preference for the target quadrant. Swim speed and distance swum were not impacted by diet or genotype. It is noteworthy that the mice began training at 16 weeks of age, at a time when soluble Aβ_1–42 is abundant in the CNS but approximately 7 months before the development of cerebral plaques.

3.4. Insulin production is not impaired in Aβ₄₂-tg mice

Blood was collected during the course of GTT on week 6 of the DIO study. Samples were taken immediately before (i.e., after a 5-h fast) and 30 minutes after an i.p. glucose administration, and insulin levels were measured in serum by ELISA. Mice fed an ND had similar fasting (T₀) insulin levels and responded to glucose with an elevation (Table 1); no significant differences were observed between genotypes. DIO mice of both genotypes showed hyperinsulinemia and poor responses to glucose. There was no significant difference between genotypes.

Table 1.

Insulin levels impacted by diet but not genotype

Mouse group		[Insulin] (ng/mL)
Genotype	Diet (6 wk)	T0		T30a		%Δ
WT (littermates)	normal	0.474	±0.043	1.07	±0.179b	125.8	±27.94
	western	2.66	±0.547	2.80	±0.789	22.26	±31.43
Aβ42-tg	normal	0.553	±0.107	1.10	±0.173b	147.4	±53.75
	western	5.78	±2.03	3.74	±1.42	−17.69	±24.21

^a30 min after challenge with glucose (2 g/kg body wt.).

^bp ≤ 0.01 versus T₀.

3.5. Hypothalamic control of glucose homeostasis appeared unaltered in Aβ₄₂-tg mice

The hypothalamus plays a key role in the coordination of circulating glucose, appetite, energy utilization, and fat metabolism. It is acted upon by both neurologic signals and humoral/endocrine factors; its output is similarly diverse. Based on the alterations in glucose handling revealed by the GTT, we chose to assess feeding behavior, activity cycles, sleep patterns, and energy metabolism using CLAMS cages.

Mice were acclimated in CLAMS cages for 20 h before 48 h of data collection during a 14/10-h light/dark period. Thirteen-week-old Aβ₄₂-tg mice consumed the same amount of food and were not different in body weight compared with their WT counterparts (Fig. 4A). Physical activity (motility) was equivalent between WT and Aβ₄₂-tg (Fig. 4B, a–b), and extrapolation of the beam breaks used for this assessment indicated no differences in time sleeping (Fig. 4B, C). The WT and transgenic mice did not differ by the amount of oxygen consumed or carbon dioxide expelled (Fig. 4C, a–d). Energy expenditure and respiratory exchange ratio (reliance on carbohydrates for energy) did not differ by genotype (Fig. 4C, e–f).

Fig. 4.

Aβ₄₂-tg mice show no aberrations in food consumption, body weight, physical activity, or respiration. Aβ₄₂-tg mice and WT littermates were acclimated to CLAMS cages and then monitored for 48 hours. A: Aβ₄₂-tg mice consumed an equivalent mass of food (a) during light and dark hours; they also achieved similar body weights (b). B: Monitoring of activity with infrared beams during light and dark hours indicated no difference between genotypes in ambulation, as indicated by movement in the x-axis (a), or in jumping/rearing, as indicated by movement in the z-axis (b); sleep time (c) was derived from the movement data and also showed no differences. C: Consumption of O₂ (a, b) and production of CO₂ (c, d) was monitored, indicating no genotypic differences in these primary measures or in the calculations of energy expenditure (e) or respiratory exchange ratio (f).

A biochemical assessment of hypothalamic function may be achieved through measures of expression of relevant neuropeptides. The peptides themselves were below the level of reliable measurement without pooling tissue from multiple animals, so we analyzed mRNA in hypothalamic homogenates using qRT-PCR. The expression levels of Agrp, Pomc, and Crh and Ptgs2 in 28-week-old mice were equivalent across genotype (Fig. 5A). Transcripts of receptors and signaling proteins Insr, Irs1, and Mc4r, all important to melanocortin system function, were also unaltered in Aβ₄₂-tg mice (Fig. 5B).

Fig. 5.

Hypothalamic gene expression in 28-week-old WT and Aβ_42-overexpressing mice. Hypothalamus tissues from fasted mice that were 28 weeks of age were homogenized, and mRNA was extracted. A: Using qRT-PCR (see “Methods”); Agrp, Pomc, Crh; and Ptgs2 gene expression were normalized to the stable transcript Rplp0. B: Insr, Irs1, Socs3, and Mc4r expression was normalized using Rplp0 and showed no differences. Comparisons were evaluated using Student’s t-test with Welch’s correction. All values are mean ± SEM.

3.6. HPA axis contributions to Aβ-induced glucose intolerance

The HPA axis is one mechanism through which neurologic events can impact peripheral glucose levels. Because the hyperglycemic tendency of Aβ₄₂-tg did not appear to be insulin dependent, we tested whether a hyperactive HPA axis might be producing excessive glucocorticoids and thereby stimulating hepatic glucose production. Aβ₄₂-tg and WT mice were adrenalectomized (“Aβ_42-ADX” and “WT-ADX”) and compared with unoperated littermates as controls (“Aβ₄₂-tg” and “WT”). Adrenalectomy decreased corticosterone levels in both genotypes (Fig. 6A). Because glucocorticoids can impact insulin production, we also measured circulating insulin levels in the adrenalectomized mice. Neither fasting insulin nor insulin level in response to glucose challenge differed between the groups (Supplement Fig. 4A). GTTs and ITTs were performed on these groups as described previously. Replicating our previous findings, Aβ₄₂-tg had had a slower decline in [Glc]_b than WT, and this effect of Aβ expression was also apparent between genotypes within the ADX groups (Fig. 6B). Unoperated Aβ₄₂-tg mice had an impairment compared with unoperated WT in the tAUC, and iAUC showed a significant impairment in Aβ₄₂-tg mice compared with WT in both intact and ADX mice (Fig. 6C). ADX of both genotypes showed higher in [Glc]_b than intact controls during ITTs, but Aβ expression again produced no effect on ITTs (Fig. 6D, Supplement Fig. 4B). After fasting, mice had no difference in basal [Glc]_b (Fig. 6D).

Fig. 6.

Aβ₄₂-tg mice retain impaired glucose tolerance after adrenalectomy. GTT was performed on mice 19 days after adrenalectomy (ADX), and insulin tolerance testing ITT was performed the following week. WT adrenalectomized (WT-ADX) and Aβ₄₂-tg adrenalectomized (Aβ_42-ADX) mice were compared with WT and Aβ₄₂-tg unoperated littermates. A: After a 5-h fast, corticosterone was measured in serum. Adrenalectomy decreased corticosterone levels in WT animals (***p < 0.001) and Aβ₄₂-tg animals (****p < 0.0001); unoperated Aβ₄₂-tg mice had increased corticosterone relative to unoperated WT (*p < 0.05). B: GTT indicated that [Glc]_b was elevated in Aβ₄₂-tg more than in WT at 30 and 60 minutes (^††p < 0.01, ^††††p < 0.0001). Aβ_42-ADX had significantly higher [Glc]_b than WT-ADX at 30 minutes (****p < 0.0001). C: Aβ₄₂-tg had higher AUC values than WT in the unoperated group (****p < 0.0001) and in the ADX group (*p < 0.05). D: Adrenalectomized mice trended toward greater insulin sensitivity (Aβ₄₂ vs. Aβ₄₂-ADX: ^##p < 0.01, WT vs. WT-ADX: ^‡‡p < 0.01). All values are mean ± SEM.

3.7. Cerebral metabolic rate of glucose (CMR_glc) was diminished in Aβ₄₂-tg mice

The absence of any aberration in insulin production or sensitivity suggested that the impaired glucose tolerance in Aβ₄₂-tg mice may have resulted from aberrant utilization in an insulin-independent compartment, such as the CNS. AD is associated with reductions in CMR_glc, typically assessed in humans with fluorodeoxyglucose positron emission tomography (PET) imaging. To perform an equivalent assessment in mice, we performed [³H]2-deoxyglucose uptake assays in 15-week-old WT and Aβ₄₂-tg mice. [¹⁴C]Sucrose was coinjected to account for blood volume contributions. Blood samples from the tail vein were taken to measure the component of transport calculations dependent upon blood concentration. These measurements revealed that net [³H]2-deoxyglucose uptake was lower in Aβ₄₂-tg mice than controls (WT K_i: 0.2259 ± 0.0354; Aβ₄₂-tg K_i: 0.1334 ± 0.0219; p = 0.043).

3.8. Aβ impairs translocation of GLUT1 to plasma membrane in astrocytes

Assessment of CMR_glc depends partially on the activity of hexokinase to trap the 2-DG intracellularly, but it occurs at concentrations of glucose below saturation of that enzyme. Therefore, CMR_glc appears to be dictated by rates of both phosphorylation and transport of glucose, the latter being a multiple step process dependent upon saturable transporters (Barros et al., 2005). Conceivably, differences in mobilization by other transporters within the brain parenchyma could also affect this process through shifting the glucose equilibrium. We assessed the state of GLUT1 and related transporters in cerebral cortex to determine whether impairment in one or more transporters might explain both the lower CMR_glc and the surfeit of [Glc]_b in Aβ₄₂-tg mice. This venture was informed by proteomic analysis we recently performed on Aβ₄₂-tg and WT cerebral tissue that had been fractionated for membrane versus cytosolic fractions (Ayyadevara et al., 2019). Among those data, the membrane:cytoplasm ratio of GLUT1 was found to be reduced by 22% in Aβ₄₂-tg brain. Thus, in addition to total expression levels, we determined the fraction of transporter that was functionally localized by fractionating plasma membrane and comparing it with the remainder of cellular fractions. Steady-state levels of transporters in fractionated cerebral cortex of WT and Aβ₄₂-tg mice were assessed by a process of capillary electrophoresis coupled to immunodetection (“SimpleWes”) because of its sensitivity, quantitative validity, and commendable size resolution within the range of 40–60 kDa. Endothelial cells express a GLUT1 moiety migrating at 55 kDa, whereas CNS parenchymal cells express a 45-kDa GLUT1, considered to be primarily astrocytic. The proportion of 45-kDa GLUT1 at the plasma membrane was significantly lower in Aβ₄₂-tg than in WT cortical homogenates (Fig. 7A, C). No significant differences in 55-kDa GLUT1 were detected between WT and Aβ₄₂-tg mice (Fig. 7A, C). Attempts were made to perform similar analyses of other glucose transporters. GLUT2 could not be subjected to fractional calculation because plasma membrane levels in most specimens were below the level of detection, but total GLUT2 levels were reduced by 45% in the cortex of Aβ₄₂-tg mice (Fig. 7B). No difference was apparent in the fractional distribution of GLUT3 or GLUT4 (Fig. 7C). Total expression levels of all glucose transporters other than GLUT2 were similar across the genotypes (Supplement Fig. 5). GLUT1 and GLUT3 total levels were also unaltered in AD (Supplement Fig. 6).

Fig. 7.

GLUT1 fractional distribution to the plasma membrane was perturbed in Aβ₄₂-tg mice and AD cortex. Homogenates from cerebral cortex were fractionated to separate plasma membrane (PM) from the other cellular fractions (“remainder”), then samples normalized for protein content were resolved by capillary electrophoresis, and GLUT proteins were detected by antibody-dependent chemiluminescence. A: Pseudogel depiction of the scans of 45-kDa and 55-kDa GLUT1 detection in whole cerebrum from WT and Aβ₄₂-tg mice. (Small discrepancies in vertical position reflect idiosyncracies among the capillaries and are common with this technology.) B: Pseudogel depiction of GLUT2 PM and remainder fractions in whole cerebrum from WT and Aβ₄₂-tg mice. C: Quantification of the PM fractional value (“total” = PM + remainder) for the indicated proteins in whole cerebrum from WT and Aβ₄₂-tg mice. D: Quantification of total GLUT1 (left ordinate; normalized to total protein) and PM fractional value of GLUT1 (right ordinate) from cerebral microvessel isolates. D: Quantification of the PM fractional value (“total” = PM + remainder) for the indicated proteins from human subjects diagnosed with AD or no neurological disease (age-matched controls, AMC). All values are mean ± SEM.

The difference detected between 45-kDa and 55-kDa GLUT1 suggested that Aβ accumulation impacted GLUT1 trafficking in parenchymal cells but not in the vascular endothelial cells. To test this more directly, we performed microvessel isolation on the cortical tissue of WT and Aβ₄₂-tg mice; microvessels were then subjected to fractionation into plasma membrane and remainder. Similar to the results with total cortical homogenates, the membrane trafficking of the 55-kDa GLUT1 in this preparation was not different between WT and Aβ₄₂-tg (Fig. 7D).

Differences in total levels of GLUT1 have been reported in AD for some brain regions, but this does not appear to be a regionally consistent phenomenon (Vannucci et al., 1997). To determine if AD patients could suffer deficiencies in CMR_glc as a consequence of GLUT1 trafficking, we prepared membrane fractions of tissue homogenates from cerebral cortex of AD and AMC. As seen in the Aβ₄₂-tg mice, human AD cortical tissue showed a lower distribution of 45-kDa GLUT1 to the plasma membrane (Fig. 7E). GLUT3 appeared to be unaffected, either in subcellular distribution or total levels.

To determine whether the lower plasma membrane fraction of GLUT1 could be a direct effect of Aβ, we treated primary cultures of astrocytes with oligomeric preparations of Aβ_1–42. Fractionated samples were assayed by SimpleWes as with cortical homogenates. After 6 hours of exposure to Aβ (5 μM), the 45-kDa GLUT1 present in these cultures showed a lower fractional partitioning to the plasma membrane than seen in control cultures (Fig. 8A). This effect was replicated by treatment with proinflammatory cytokines. Attenuated delivery of GLUT1 to the surface of Aβ-treated primary astrocytes was also detected by immunocytochemistry. Cultures were subjected to a mild fixation and immediately processed in either a permeabilized or unpermeabilized state, confirmed by differential detection of GFAP, an intracellular marker. Aβ treatment was associated with lower levels of GLUT1 on the surface or unpermeabilized cultures (Fig. 8B).

Fig. 8.

Aβ and cytokines evoke aberrant trafficking of GLUT1 in cultured astrocytes Primary cultures of astrocytes were treated 6 hours with aggregated Aβ_1–42 (2.5 μM) or 4 hours with a combination of IL1β and TNF (10 and 30 ng/mL, respectively). A: PM fractional value of 45-kDa GLUT1 detected by capillary-gel immunodetection as in Fig. 7. B: Immunofluorescence of GLUT1 and GFAP performed on cultures that had been permeabilized or not before application of primary antibodies. Absence of GFAP staining in unpermeabilized cells validates that GLUT1 staining in those cultures is likely restricted to cell-surface protein. Scale bar = 25 μm.

4. Discussion

An intriguing question arising from the dysregulation of glucose in AD is the extent to which this disorder’s brain pathology and clinical symptoms arise secondary to peripheral events. A key aspect of human AD pathogenesis is a decrease in CMR_glc. Similar declines are seen in diabetics (both type-1 and type-2) and in animal models of chronic hyperglycemia. Other tissues experience a surfeit of glucose under such conditions, but the brain does not receive glucose passively; rather, it relies on saturable transporters at the BBB. A reactive downregulation of these transporters in response to hyperglycemia appears to contribute to the reduced CMR_glc seen in diabetic conditions. Accordingly, most evidence indicates that the cognitive impairment seen in T2D is attributable to vascular dementia rather than AD per se (Hassing et al., 2002; Matioli et al., 2017; Mortel et al., 1993; Pruzin et al., 2018; Raffaitin et al., 2009). This study explored the metabolic profile of mice that express human Aβ_1–42, a highly aggregative form of Aβ implicated in the pathogenesis of AD. We chose to avoid models relyingon overexpression and mutation of APP and presenilin, as each of these can have confounding effects on glucose metabolism (Botteri et al., 2018; Kulas et al., 2019; Lee et al., 2013; Needham et al., 2008; Tu et al., 2012). In contrast to those models, the Aβ_42-transgenic mice did not present with a generalized metabolic syndrome as might be expected if there were functional abnormalities in hypothalamic circuitry, peripheral hormones, adiposity, peripheral inflammation, or pancreatic insufficiency. They did, however, exhibit a cognitive deficit that was correlated with impaired glucose tolerance. Evaluation of cerebral glucose utilization suggested that the peripheral glucose phenotype of Aβ₄₂-tg mice—and perhaps their human counterparts—results from a CMR_glc impairment, itself due to an Aβ-induced irregularity in the translocation of GLUT1 to the plasma membrane in cells of the cerebral parenchyma. The perturbation of cerebral glucose was evident at ages much younger than that required for the observation of amyloid plaques in this line, indicating that it was driven by soluble Aβ.

Transgenic models of AD that overexpress APP have been reported to exhibit perturbations of general aspects of glucose and energy homeostasis, including impaired glucose tolerance, insulin resistance, and attenuated responses to leptin (Lyra E Silva et al., 2019). By contrast, the peripheral phenotype of Aβ₄₂-tg was much more limited. While they had impaired glucose tolerance, Aβ₄₂-tg mice were not significantly heavier than WT. In T2D, impaired glucose tolerance is explained by a reduction in glucose uptake by peripheral tissues, initially as a consequence of insulin resistance in these tissues; this often progresses to a state of insufficient insulin production by pancreatic β-cells. Aβ₄₂-tg mice had a response to insulin equivalent to that of WT mice, indicating that their aberrant GTT results were not due to peripheral insulin resistance; insulin production was intact, as well. Moreover, our comprehensive metabolic panel demonstrated elevated glucose in the absence of biomarkers for hepatic or renal disorder. By these metabolic indices, Aβ₄₂-tg appear similar to humans with AD, who do not show differences from controls in body weight, insulin production, insulin sensitivity, or general hypothalamic function. Hepatic gluconeogenesis in response to pyruvate administration was similar in Aβ₄₂-tg and wild-type, suggesting that impaired GTTs did not arise from hepatic dysfunction. However, a small portion of the glucose intolerance was attenuated by adrenalectomy, indicating that elevated hepatic glucose production (perhaps by glycogenolysis) may have contributed weakly to the phenotype. AD, risk for AD, and cognitive impairment are all associated with elevations in cortisol levels (Ouanes and Popp, 2019).

We tested for hypothalamic contributions to the glucose phenotype of Aβ₄₂-tg mice through multiple approaches. Metabolic (CLAMS) cages were used to examine food consumption and respiratory quotient, the latter of which can detect differences in the degree to which energy production depends on carbohydrates versus other nutritional sources of energy. Infrared beam crossings in the cages permitted monitoring of physical activity, attributes of which were also extrapolated to inferences about sleep/wake behavior. These assessments indicated that the glucose tolerance of Aβ₄₂-tg mice was not associated with differences in diet, metabolism, activity, or sleep. This conclusion is supported by the failure to detect differences in transcriptional expression of melanocortin system neuropeptides that influence satiety and feeding.

Our 2-deoxyglucose measurements demonstrated that Aβ₄₂-tg mice phenocopy the decrease in CMR_glc seen in humans with AD. In the absence of other explanations, this diminished utilization by the most glucose-demanding organ is a plausible explanation for the exaggerated glucose excursions in Aβ₄₂-tg and the human disorder they model. While the human brain is only 2% of body mass, it may account for as much as 60% of our glucose utilization at rest (Berg et al., 2002). This percentage is smaller in mice, but rodent brains still create an outsized demand on glucose disposition (Sokoloff, 1977). Saturable glucose transporters limit access of this carbohydrate to the CNS, and diminutions in endothelial GLUT1 are well-documented in conditions of chronic hyperglycemia (Gjedde and Crone, 1981; McCall et al., 1982; Pardridge et al., 1990). AD and mouse models thereof do not universally exhibit hyperglycemia to a degree and consistency sufficient to downregulate endothelial GLUT1 expression. Indeed, we find this moiety unaltered. Instead, an isoform of GLUT1 specific to parenchymal cells appeared to be significantly lower in its functional localization. Astrocytic end-feet make canonical contributions to the neurovascular unit, and astroglia serve as a conduit for energy substrates from the vasculature to neurons, evidently supplying glucose and its derivatives to neurons (Dienel, 2012). Interruptions in this glucose flux are likely to 1) diminish neuronal function and 2) shift the equilibrium for central import of glucose from blood. Together, the lower CMR_glc and mislocalization of parenchymal GLUT1 that we have detected suggest that elevations in blood glucose in AD reflect a central phenomenon, resulting from the effects of Aβ on cerebral parenchyma, rather than a generalized disruption of hypothalamic or peripheral endocrinology. Feasibility of this explanation is supported by the impaired glucose tolerance exhibited in mice lacking GLUT4 in neural tissues (Reno et al., 2017); however, that phenotype also included insulin resistance and other attributes suggesting that it involved impaired glucose sensing, likely in the hypothalamus.

It is clear that CMR_glc is correlated with cognitive performance. Decades of imaging studies have demonstrated that CMR_glc declines in patients with AD (Friedland et al., 1983; Herholz et al., 2007; Laforce et al., 2014; Leuzy et al., 2018). The phenomenon is also observed in essentially every syndrome of age-related dementia (Wilson et al., 2019), including frontotemporal dementia, dementia with Lewy bodies, and Creutzfeldt-Jakob disease (Kim et al., 2012). It is perhaps the universality of the phenomenon to various dementing disorders that makes FDG-PET a more accurate predictor of MCI-to-dementia conversion than are CSF measurements of Aβ and tau (Caminiti et al., 2018). In fact, the longitudinal decline in CMR_glc is associated with concurrent cognitive decline so intimately that it exceeds psychological examinationss in statistical power for monitoring longitudinal change (Landau et al., 2011). CMR_glc decline has been characterized as an effect of synaptic pathology, but it seems just as likely that restricted glucose delivery would be the cause of diminished mental performance, as is clearly demonstrated during acute hypoglycemic bouts (Strachan et al., 2001) and in higher performing MCI subjects (Ossenkoppele et al., 2014). Some reports indicate that CMR_glc is elevated before its decline in certain regions (Benzinger et al., 2013), but it is speculated that this reflects compensatory activity that is responsible for the higher cognitive performance in MCI versus AD (Ashraf et al., 2015; Ossenkoppele et al., 2014). This is not exclusive of the possibility that this elevated neurophysiological activity is, in fact, driving subsequent Aβ deposition (Cohen et al., 2009). Mice of the BRI-Aβ42 line used here have been reported to be free of cognitive impairment (Kim et al., 2013). Compared with the present study, the mice used in that report were considerably older (15 months); even the WT performed quite poorly, producing a swim-path percentage in the target quadrant scarcely better than chance (28.30 ± 2.0). It might also be noted that the mice in that study were on a mixed strain background (C57BL/6 × C3H) which may have introduced variability.

Diminished CMR_glc (Baker et al., 2011; Hwang et al., 2017; Roberts et al., 2014) and cognitive impairment (Biessels and Despa, 2018) are also seen in T2D. It is notable that neuron loss has not been documented in T2D. While the two conditions share some attributes, the evidence linking AD to T2D is weak. The Rush longitudinal cohort study of aging found that neither HbA1c nor diabetes was correlated with plaque, tau, or infarct pathology (Pruzin et al., 2017); cognitive and synaptic deficits in a model of T2D are independent of tau (Trujillo-Estrada et al., 2019). T2D confers no elevation in the accumulation of Aβ or neurofibrillary pathology (Beeri et al., 2005; Heitner and Dickson, 1997; Peila et al., 2002; Roberts et al., 2014). A wealth of data indicates that the greater risk for dementia among individuals with T2D can be explained adequately by the incidence of vascular dementia (above). Likewise, AD may not make a significant impact on propensity for developing diabetes. Our study finds no convincing perturbation of fundamental endocrinology or peripheral glucose physiology in response to Aβ accumulation. This is consistent with the fact that the age-related accumulation of senile plaques in the hypothalamus does not correlate with an AD diagnosis (van de Nes et al., 1998; van de Nes et al., 2006).

Together, our findings indicate that trafficking of GLUT1 to the plasma membrane in parenchymal cells of the brain may be responsible for depressed CMR_glc in AD. We did not detect significant differences in steady-state levels of the total glucose transporter proteins in either Aβ₄₂-tg mice or human AD. Diminished overall levels of glucose transporters have been reported in other mouse models (Liu et al., 2017; Merlini et al., 2011); some evidence indicates this may represent merely a reduction in cerebrovascular mass (Ahn et al., 2018; KuznetsovaSchliebs, 2013). Increased expression of GLUT1 and −3 has also been noted in an AD model (Chua et al., 2012), potentially as a compensatory reaction to glucose starvation. Winkler et al. (2015) reported that global haploinsufficiency of GLUT1 resulted in age-related degenerative changes in the BBB, reductions in cerebral blood flow and CMR_glc, neurodegeneration, and memory deficits. The alterations in BBB were also noted after conditional haploinsufficiency was restricted to the vascular epithelium.

In the human brain, regional differences are more readily appreciated and may be discrepant with regard to the effects on glucose transporters by a disease process manifest in a regionally progressive pattern. Kalaria and Harik (1989) first reported that AD was accompanied by lower levels of cytochalasin B-binding sites, interpreted to be hexose transporters, in the microvessels of hippocampus; Horwood and Davies (1994) used immunohistochemistry to confirm that this likely reflected an impact on GLUT1. Vannucci et al. (1997) reported that the 45-kDa form of GLUT1 was specifically lower in AD, albeit this was limited to the caudate. Both the 45-kDa and the 55-kDa forms of the protein were found to be depleted in unspecified regions of the AD cerebral cortex by Simpson et al. (1994). Liu et al. (2008) found lower levels of both GLUT1 and GLUT3 in AD frontal cortex. We chose to analyze frontal lobe as well, hoping to avoid a high degree of late-stage neuronal loss that would be significant in the hippocampus and the temporal lobe in general. Similar to our findings in relatively young mice, human prefrontal cortex showed a reduction in plasma membrane distribution without a change in total GLUT1 levels. It is tempting to speculate that this may be an early event, occurring before the loss of overall transporter levels during the course of AD pathogenesis in a given brain region.

We also noted a reduction in total levels of GLUT2 in Aβ₄₂-tg mice. It is difficult to interpret this finding because there is no clear consensus on the role of GLUT2 in the brain, where it seems localized primarily to astrocytes. The K_M of GLUT2 for glucose is 15–20 mM, compared with 1–5 mM for GLUT1. Zetterling et al. (2011) reported that the concentration of glucose in human brain interstitial fluid is 0.1–5.5 mM and 1.7 ± 0.9 mM in frontal lobe. Consistent with its low affinity, GLUT2 is important for export of glucose produced by gluconeogenesis and glycogenolysis in hepatocytes. As astrocytes also store glycogen, GLUT2 could play a similar role in these cells. The fact that plasma membrane levels of GLUT2 were below the limit of detection further obscures interpretation of this transporter’s role. GLUT2 may be present only at discrete points on the astrocyte membrane, perhaps releasing glucose from astrocytes at synapses. If this is the case, reductions in its overall levels in Aβ₄₂-tg brains may reflect a generalized failure in the ability of astrocytes to serve as a glucose conduit between capillaries and neurons.

Discerning the true impact of AD pathogenesis on glucose transporters may depend on a mechanistic understanding of the molecular and biochemical chain of events. Our experiments with primary cultures of astrocytes indicate that the deficit in GLUT1 localization may be a direct result of Aβ. However, proinflammatory cytokines replicated the effect of Aβ on cultured astrocytes, so it is possible that Aβ exerts an indirect effect mediated by microglial activation. Although steps were taken to diminish microglia in our primary astrocyte cultures, a few remain. Decreases in FDG-PET signal have been interpreted as a correlate of inflammatory gliosis (Schroeter et al., 2009). Intracellular trafficking of GLUT1 is not understood as well as it is for GLUT4. In many cell types, GLUT1 appears to transit through the secretory pathway in a constitutive and unregulated manner. However, there is a clear difference in the structural composition of GLUT1 in parenchymal cells, a size difference we exploited to detect differences in parenchymal transporter against the backdrop of prominent endothelial expression. It is plausible that this difference provides a mechanistic distinction in GLUT1 trafficking. It is also intriguing to speculate about the relationships between trafficking in the secretory pathway, ER stress, and metabolic syndrome, along with disturbances in autophagy. Trafficking of GLUT1 to the plasma membrane can be influenced by mammalian target of rapamycin (mTOR) (Makinoshima et al., 2015; Olsen et al., 2014), which is a major node for regulation of autophagy. APOE ε4, the most significant genetic factor for sporadic AD, has been demonstrated to inhibit autophagy in astrocytes (Simonovitch et al., 2016). Building on this phenomenon, we recently demonstrated a mechanistic explanation by which the gene’s protein product (ApoE4) binds autophagy-gene promoters at their “CLEAR” enhancer sequences, thus competitively inhibiting the key autophagic transcription factor transcription factor EB (Parcon et al., 2018). If the resultant tendency toward protein aggregation and other forms of proteinopathy interferes with the secretory pathway, this may contribute to disturbances in localization of GLUT1 in astrocytes. A correlation between compromised CMR_glc and generalized astrocytic pathology has been documented in AD (Carter et al., 2019), and there is an increasing appreciation for glial contributions to such imaging modalities (Zimmer et al., 2017).

In conclusion, we found that the perturbation of glucose tolerance in Aβ₄₂-tg mice occurred in a manner that was independent of peripheral metabolic syndrome, hepatic glucose production, and insulin delivery or responses. This suggested an aberration in an insulin-independent tissue, such as the brain. Indeed, these mice showed a reduction in CMR_glc similar to human patients with AD. We further demonstrated inadequate trafficking of GLUT1 to the plasma membrane of parenchymal brain cells, likely astrocytes. Together, these data suggest that the anomalous glucose intolerance in Aβ₄₂-tg mice—and, potentially, humans with AD—reflects cerebrocentric effects that have little to do with endocrinological or physiological output of the hypothalamus. These findings further emphasize the growing evidence for involvement of astrocyte biology in the pathogenesis of AD. It will be important to determine where these events fit in the etiological chain of events that includes neurofibrillary pathology, tissue atrophy, and functional dementia.