Insulin Reverses the High-Fat Diet–Induced Increase in Brain Aβ and Improves Memory in an Animal Model of Alzheimer Disease

Mounting evidence suggests that Alzheimer disease (AD), diabetes, and obesity are linked by their common reliance on insulin signaling pathways in the central nervous system (1). Besides its broad actions in the periphery, insulin in the brain regulates not only energy homeostasis, but also synaptic plasticity and memory function (2–5). Epidemiological studies revealed that dysfunction in insulin production and signaling is associated with poorer cognitive performance (6–8). Conversely, intranasal administration of insulin improves memory function when administered to cognitively impaired (9) or normal adults (10). Furthermore, type 2 diabetes (T2D), a condition characterized by impaired insulin homeostasis, is now recognized as an important risk factor for AD, which is the leading cause of cognitive dysfunction in the elderly (11–15).

While the interaction between such apparently distinct diseases is puzzling, it implies that treatments targeted toward boosting insulin signaling could tackle both AD and T2D at the same time. Therefore, intense efforts have been made in recent years to decipher causal mechanistic links underlying this association, particularly in animal models. One of the first clues comes from the demonstration that the induction of insulin resistance and diabetes results in an increase in Aβ production, tau phosphorylation, and cerebrovascular inflammation in the brain of several transgenic models of AD (16–19). Accordingly, increased postmortem microvascular pathology is found in AD patients with metabolic comorbidities, including diabetes (20). In contrast, several lines of evidence suggest that the clinical expression of AD includes metabolic impairment as well. Experiments performed with human postmortem brain tissue show that insulin signaling and the number of insulin receptors in the hippocampus are reduced in AD patients compared with age-matched control subjects (21,22). Moderate to severe AD patients also have lower cerebrospinal fluid insulin levels and reduced insulin-mediated glucose disposal when compared with healthy control subjects (23,24). Furthermore, early T2D-like biomarkers such as insulin resistance and glucose intolerance can be detected in AD mouse models, suggesting that AD pathology could also affect peripheral metabolism (25–29). However, it has yet to be determined whether metabolic deficits accelerate AD progression or whether AD causes defects in peripheral metabolism, or both.

We report in this study the presence of impaired glucose tolerance in a mouse model of genetically induced AD-like neuropathology (3xTg-AD), which progressed to pancreatic β-cell degeneration and impaired insulin production after exposure to a high-energy diet. The high-fat diet (HFD) also massively increased brain amyloid pathology and worsened memory function in 3xTg-AD mice. A single acute injection of insulin in 3xTg-AD mice fed with HFD restored memory function and brought back soluble brain Aβ concentrations to control levels, while increasing plasma Aβ levels in plasma. Our data thus reveal tight links between AD and peripheral metabolic defects and suggest that insulin signaling in the brain plays a key role in the production and clearance of Aβ from the brain to the blood.

All animal experiments were approved by the Laval University ethics committee. The 3xTg-AD (APP_swe, PSI_M146V, tau_P301L) mouse model of AD was used for the experiments (30). 3xTg-AD mice were produced at our animal facilities and compared with nontransgenic (NonTg) littermates with the same genetic background (C57BL6/129SvJ). For the dietary experiments, female mice received either a control diet (CD; 12% kcal fat) or an HFD (60% kcal fat) for a 9-month period starting at the age of 6 months (see ref. 17 for a detailed description of the diets). Females were selected because they exhibit higher AD-like pathology than males (31–33). All behavioral and metabolic testing were started 1 month before sacrificing the animals to minimize the impact of the anxiety generated by the tests and allow enough resting and acclimation time between the experiments. Animals were fasted for 6 h and received an intravenous injection of insulin (3.8 units/kg of human insulin) or saline 5 min before sacrifice. This dose is known to induce a robust activation of insulin signaling in the muscle and the liver (34). They were killed by intracardiac perfusion under deep anesthesia with ketamine/xylazine. Brain, pancreas, liver, epididymal adipose tissue, and gastrocnemius muscle were rapidly dissected and excised. Brain, muscle, and liver were kept at −80°C until processing for Western blot and ELISA analysis. The number of animals in each group for all experiments is indicated directly in the graph.

Insulin sensitivity and glucose tolerance were assessed using intraperitoneal insulin tolerance tests (ITTs) and glucose tolerance tests (GTTs), respectively (n = 7–12 animals/group). Because we used female mice, all tests were performed during the diestrus phase of the estrus cycle to minimize hormonal variations during the tests. After 6-h fasting, mice were injected with 1 unit/kg human insulin or 1 g/kg glucose. Glycemia was measured with a glucometer (OneTouch UltraMini; LifeScan, Milpitas, CA) with a blood drop from the saphenous vein before the injection and 15, 30, 45, 60, 90, and 120 min after the injection.

The detailed procedure for protein extraction is found in the Supplementary Material and is described elsewhere (35).

Aβ40 and Aβ42 were measured in cortex soluble and insoluble fractions (n = 10–13) as well as plasma (n = 6–12) using a human β-Amyloid ELISA (Wako, Osaka, Japan) according to the manufacturer’s instructions, and plates were read at 450 nm using a Synergy HT multidetection microplate reader (BioTek, Winooski, VT).

Epididymal fat and pancreas were dissected, weighed, and immediately postfixed in paraformaldehyde 4% (pH 7.4) for 48 h. Ten-micrometer sections were first hydrated in ethanol and stained in successive baths of hematoxylin (10 min), water (3 min), 70% alcohol, and 1% HCl solution and washed with running water. Lithium carbonate dips, running water, and eosin staining (30 s) were then performed prior to alcohol dehydration and coverslipping. Pancreatic islet area (n = 6 to 7 animals/group) was measured under bright-field illumination using Stereo Investigator software (MicroBrightField, Colchester, VT) integrated with an E800 Nikon microscope (Nikon Canada Inc., Mississauga, Ontario, Canada) from two sections of pancreas per mice. Approximately 20 islets per animal, from six to seven animals per group, were quantified with Neurolucida modeling software (MicroBrightField).

For Western immunoblotting, proteins from the parieto-temporal cortex were extracted with a Tris-buffered saline buffer for the soluble protein, a RIPA buffer for the membrane soluble protein, and a sarkosyl insoluble fraction for insoluble proteins, as described above. Protein quantification was done using bicinchoninic acid assays (Pierce, Rockford, IL). A total of 20 μg of protein from the cortex of 10–13 animals/group was loaded and separated by SDS-PAGE and then electroblotted onto a polyvinylidene difluoride membrane (Immobilon; Millipore). Membrane were blocked in 5% milk with 0.5% BSA for 1.5 h and immunoblotted with primary and then secondary antibodies followed by chemiluminescence reagents (KPL, Gaithersburg, MD). The list of primary antibodies that were used in our experiments is available in the Supplementary Table 3. Intensity of the bands was assessed with a Kodak Image Station 4000MM Digital Imaging System (Molecular Imaging Software version 4.0.5f7; Kodak, New Haven, CT).

Plasma insulin (n = 6–10 animals/group) and C-peptide (n = 5–8 animals/group) concentration were determined with an Ultrasensitive Insulin ELISA (Mercodia, Uppsala, Sweden) and C-peptide (Alpco Diagnostic, Salem, NH) according to the manufacturers’ instructions.

To evaluate recognition memory, the object-recognition task was performed (n = 9 to 10 animals/group). Mice were introduced in a clear box (29.2 cm × 619 cm × 612.7 cm) with two different objects for 5 min. One hour later, they were put in the same box with a new and a familiar object for 5 min. To verify the effect of insulin on recognition memory, a single insulin injection (1 unit/kg) was administered intraperitoneally, and the tests were performed 2 h later to allow glycemia to return to baseline (n = 6–13 animals/group). Recognition index = (time exploring the new object − time exploring the old object)/time exploring the old object. The time exploring the object was defined as the time the mouse spent smelling or exploring an object (36–38). Spatial memory evaluation was performed using the Barnes maze (San Diego Instruments, San Diego, CA) as previously described (39). The test lasted for 9 days, 4 training days (4 trials per day, 3 min each), and 2 test days (1 trial of 90 s). Mice were placed on a circular platform with 19 holes and 1 escape zone and exposed to two aversive stimuli: a bright light and a strong noise to stimulate the need to find the escape zone. The time to find the escape zone was used to evaluate the spatial memory of the mice.

For cerebral immunofluorescence experiments, 25-μm–thick slices were used and stained with 6E10 antibody to label amyloid plaques. Plaque size was measured in the hippocampus at bregma −2.7 mm using the Stereo Investigator software (MicroBrightField) integrated with an E800 Nikon microscope (Nikon Canada Inc.) (n = 3–6 animals/group). For pancreatic immunofluorescence, paraffin-embedded 10-μm pancreas sections from mice perfused with paraformaldehyde (4%, pH 7.4) were used. Detection of amyloid pathology of the pancreatic tissue was done using 6E10 and 6C3 antibodies. To be able to quantify the number of 6E10-positive β-cells, a double staining was performed using an insulin antibody (n = 6 and 7 animals/group). The number of apoptotic β-cells in pancreatic islets was assessed with an antibody against cleaved caspase-3 (n = 5–9 animals/group). Anti-mouse or anti-rabbit secondary antibodies conjugated with either Alexa Fluor 488 or 568 were used. Immunofluorescence was examined using an epifluorescence microscope (Olympus Provis AX70; Olympus, Melville, NY) and photographs were taken using a SPOT digital camera (Diagnostic Instruments, Sterling Heights, MI). All images were prepared for illustration in Adobe Photoshop 7.0.

Data are presented as mean ± SEM. Statistical analysis and number of mice per group are specified in each figure. Equality of the variances between the groups was determined using the Bartlett test. When more than two groups were compared, one-way or two-way ANOVA were used. For one-way ANOVA with equal variances groups, Tukey post hoc analysis was performed. For unequal variances, groups were compared with Dunnett post hoc analysis. When we had two groups to compare, an unpaired Student t test was performed, with a Welch correction included if variances were not equal. Coefficients of correlation were determined using the Pearson correlation test. Statistical significance was set at P < 0.05. All statistical analyses were performed with Prism 4 (GraphPad, San Diego, CA) or JMP (version 9.0.2; SAS Institute Inc., Cary, NC) softwares.

To induce obesity and insulin resistance in 3xTg-AD mice, we fed them an HFD, from 6 months of age, known to trigger these metabolic perturbations and aggravate AD-like pathology (16,17). As expected, HFD induced a significant weight gain (32 and 42% vs. CD, NonTg and 3xTg-AD, respectively; Fig. 1A), higher calorie intake (17 and 13% vs. CD, NonTg and 3xTg-AD, respectively; Fig. 1B) and visceral fat accumulation (212 and 275% vs. CD, NonTg and 3xTg-AD, respectively; Fig. 1C and D), and a modest decrease in insulin sensitivity as evidenced by the ITT (82 and 70% ITT area under the curve [AUC]; Fig. 1E and F) and Western blot assessment of phosphorylated (p-)AKT (p-Ser⁴⁷³) levels in the muscle (−5 and −24%) and liver (−36 and −55%) (Supplementary Fig. 1). However, HFD had a similar impact in both 3xTg-AD and NonTg mice on these parameters. Interestingly, 3xTg-AD transgenes alone were sufficient to induce glucose intolerance (120% compared with NonTg mice fed the CD).

The HFD acted synergistically with 3xTg-AD transgenes to deteriorate glucose tolerance (28% compared with 3xTg-AD mice fed CD; Fig. 1G and H). As blood glucose levels are controlled through pancreatic insulin production, we measured insulin production (blood insulin and C-peptide) during the GTT. Insulin secretion after glucose administration was completely blunted in HFD-fed 3xTg-AD mice (−100% for insulin, Fig. 1I, and −55% for C-peptide, Fig. 1J and K), providing further evidence of an additive effect between HFD and AD transgenes on insulin production.

We finally evaluated the impact of HFD on memory function in 10-month-old mice using two different tests: the object-recognition task (Fig. 1L) and the Barnes maze (Fig. 1M and N). Similar to what was observed in previous studies (16), both tests revealed further memory impairment in 3xTg-AD mice fed the HFD (−15% recognition index and 183% time to find the exit zone) without changes in voluntary locomotor activity or anxiety-like behavior (Supplementary Fig. 1).

Insulin is produced by pancreatic β-cells. A shared pathological hallmark of T2D and AD is the presence of amyloid pathology (40), which, in T2D, is associated with apoptosis and loss of β-cells (41). The 3xTg-AD mouse model expresses an amyloid precursor protein (APP) bearing the Swedish mutation under the control of the Thy 1.2 promoter, restricting transgene expression and human Aβ production in cerebral tissue (30,42). We thus sought to determine whether Aβ accumulated in the pancreas as well using an immunofluorescence technique with the 6E10 antibody raised against a protein sequence found in both human APP and Aβ but not in murine orthologs. Indeed, human APP/Aβ staining was found in the pancreas of 3xTg-AD mice, where it was localized in the islets (Fig. 2A). A semiquantitative analysis revealed that ∼90% of islets were 6E10 positive in 15-month-old 3xTg-AD mice, with the prevalence of 6E10-positive islets similar between control and HFD-fed mice (Fig. 2B). As the 6E10 antibody targets, both Aβ and full-length APP of human origin, we used another antibody to bind specifically to human Aβ peptides. Consistently, the Aβ-specific staining was apparent only in 3xTg-AD mice (Fig. 2C). The absence of pancreatic expression of APP or tau transgenic mRNA confirmed that the Aβ peptide found in the pancreas was not derived from APP expressed locally (Fig. 2D).

As the Aβ peptide has been shown to induce apoptosis markers in brain cells (43,44), we examined pancreatic islets for caspase-3 activation. 3xTg-AD mice fed the HFD had 116% more islets that stained positive for cleaved caspase-3 than HFD-fed NonTg mice (Fig. 2E and F), suggesting that Aβ might potentiate the effect of HFD on β-cell death (45). These observations were corroborated by islet areas 58% smaller in the pancreas of 3xTg-AD mice fed the HFD compared with NonTg mice on the same diet (Fig. 2G and H). Taken together, these results suggest that Aβ peptides produced in the brain were transported to the periphery and accumulated in pancreatic β-cells, triggering islet degeneration and impairing insulin production.

The induction of obesity in the 3xTg-AD mouse not only caused a deterioration of memory function, but also had peripheral pathological consequences including impaired insulin production. We first determined the extent by which the HFD would increase AD neuropathology in our experimental paradigm. While a 4-month exposure to the HFD had overall no effect on amyloid or tau protein levels in 10-month-old mice (Supplementary Fig. 2), prolonging the diet to 9 months led to a massive increase in soluble Aβ40 and Aβ42 in the cerebral cortex of 15-month-old 3xTg-AD mice (536%) (Fig. 3A), in line with previous studies (16,17).

To determine whether such an accumulation of soluble Aβ was caused by a deficit in brain insulin signaling (26,46,47), groups of animals were challenged with an acute insulin intravenous administration 5 min before their death. We first measured tau protein levels and found no major effect of insulin of soluble phosphorylated tau in control or HFD-fed mice (Supplementary Fig. 3). Nevertheless, we found that the level of sarkosyl-insoluble tau was reduced following insulin injection (Supplementary Fig. 3, −20 and −27%). Yet, insulin increased cortical p-AKT (p-Ser⁴⁷³) by 30% in 3xTg-AD mice cortex (Supplementary Fig. 3), indicating a detectable effect on the phosphoinositide 3-kinase–AKT cellular cascade. No effect of insulin was seen on synaptic protein concentrations (Supplementary Fig. 3 and Supplementary Table 1). Strikingly, insulin injection completely reversed the HFD-induced rise in cortical soluble Aβ levels back to control levels in the cortex of 3xTg-AD mice (−86%) (Fig. 3A and B). Similar results were observed in the hippocampus (Supplementary Fig. 2). No detectable changes were observed in insoluble Aβ40 (Fig. 3C) and Aβ42 (Fig. 3D) or plaque formation (Fig. 3F and G). Accordingly, a single insulin injection decreased the soluble/insoluble Aβ42 ratio (Fig. 3E).

Because of the known effects of insulin on synapse formation and synaptic remodeling (48,49), we measured the behavioral consequences of a single intraperitoneal insulin injection. In line with previous data generated from intracerebroventricular insulin in rats (50), we found that memory function, as measured by the object-recognition index, was improved by 38% 2 h after a single intraperitoneal insulin injection in 16-month-old mice fed the HFD for over 10 months (Fig. 3H and I; P < 0.05). Therefore, the rise in soluble Aβ concentrations occurring after several months of exposure to HFD in 3xTg-AD mice could be rescued to control levels by a single administration of insulin, which also restored memory impairment.

We next sought to find plausible mechanisms underlying the effects of insulin on amyloid pathology and memory in the 3xTg-AD mouse. We first evaluated Aβ production by indexing the amyloidogenic cleavage of APP. Despite no difference in full-length APP (Fig. 4A) or β–C-terminal fragment (Supplementary Fig. 4), intravenous insulin administration increased soluble α-APP by 69% in mice fed the CD and by 154% in mice fed the HFD within 5 min (Fig. 4B). We then identified three molecular correlates, which could underlie the insulin-induced decrease in Aβ levels and the increase in α-APP. First, a reduction in β-secretase 1 (BACE1) protein was detected in insulin-injected 3xTg-AD mice (−29% CD and −53% HFD) (Fig. 4D). BACE1 cleaves the extracellular fragment of APP, which is later cleaved by the γ-secretase complex generating at the same time Aβ peptide (51). This observation is supported by previous in vitro studies showing a decrease BACE concentration following addition of insulin to cell culture (52). Second, insulin administration increased X11α (14% CD and 37% HFD), an adaptator protein that binds to APP and reduces its amyloidogenic cleavage (53,54), which was inversely correlated with insoluble Aβ42 (r² = 0.13) (Fig. 4E and F). Third, insulin reduced by 50% the LC3I/LC3II ratio in mice fed the HFD (Fig. 4G–J), consistent with a reduction of autophagy. As β-secretase cleavage can occur in the autophagosome, reduction of autophagy could be another mechanism by which insulin decreases Aβ production (55). Because recent research revealed that amyloid pathology is rapidly regulated accordingly to brain metabolic status (56,57), we determined the concentration of subunit 1 of cytochrome oxidase (CO1), which is considered a reliable index of metabolic status (58). Yet, we found a 42 and 33% reduction of the protein level of CO1 in the cerebral cortex of mice fed either the CD or HFD, respectively, following insulin injection (Fig. 4K). Interestingly, cortical CO1 was positively correlated with soluble Aβ42 concentrations (r² = 0.13) (Fig. 4L). Altogether, these data are consistent with an insulin-induced reduction in Aβ production. However, it is unlikely that such a metabolic process alone explains the massive reduction of soluble amyloid protein observed in 3xTg-AD mice within 5 min after insulin injection.

Another key factor underlying the pathogenic accumulation of Aβ in the brain is its clearance from cerebral tissue (59,60). We first found no change in levels of insulin-degrading enzyme (Fig. 4B and Supplementary Table 2), the main Aβ-degrading enzyme (61–63). Secondly, we studied two transporter proteins implicated in the influx and efflux of Aβ across the blood–brain barrier (64–66): LDL receptor-related protein 1 and receptor for advanced glycation end products (Fig. 4M and Supplementary Table 2). However, no effect of insulin was detected on the concentrations of these proteins in the cortex of 3xTg-AD mice. Since part of brain Aβ is cleared toward the blood (60,67), we measured plasma Aβ42 following i.v. insulin injection. Interestingly, a 2.4-fold increase in Aβ42 was detected 15 min following insulin injection in mice fed with HFD for 10 months (Fig. 4N). This is in agreement with data in human AD patients in which insulin injection results in increased plasma Aβ within 120 min (68,69). Interestingly, plasma Aβ42 concentrations were positively correlated with the performance of 3xTg-AD mice in the object-recognition task (r² = 0.56) (Fig. 4O). In brief, our results suggest that the rapid reversing effect of insulin on soluble Aβ in the brain results from of a combination of complementary mechanisms acting on central Aβ production and clearance toward the blood.

We used the 3xTg-AD mouse model of AD to study the interrelation of two highly prevalent diseases in the elderly, AD and T2D. This strategy allowed us to demonstrate that the genetic induction of AD-like pathology in the brain was sufficient to induce peripheral glucose intolerance in mice, but not insulin resistance. Our data pinpointed the accumulation of toxic Aβ in the islets as a possible cause of a defect in pancreatic insulin production. The introduction of an obesogenic diet induced insulin resistance in both NonTg and 3xTg-AD mice, but in the latter, it exacerbated not only amyloid pathology but also diabetic signs such as pancreatic β-cell death and insulin secretion failure. Such defects in insulin production and signaling had major consequences on AD-like phenotype, as a single injection of insulin restored soluble Aβ concentrations to control level, coinciding with increased plasma Aβ concentrations and improved memory in the object-recognition task.

The glucose intolerance observed in the 3xTg-AD mouse model of AD was particularly striking and is in general agreement with previous observations made in other models of cerebral amyloidosis (26,28,70). Feeding 3xTg-AD mice with an obesogenic diet further decreased glucose tolerance and blunted the secretion of insulin and C-peptide in plasma after acute glucose administration. This observation suggests that glucose intolerance in the 3xTg-AD mouse results from a defect in insulin production in the pancreas. In the same animals, we observed a massive accumulation of human Aβ in pancreatic β-cells associated with a notable shrinkage of islets and signs of caspase activation. Because APP/tau transgenes are expressed under a neuronal promoter (30), consistent with our PCR analyses, the most probable source for Aβ found in this study in pancreatic β-cells is the brain. This raises the intriguing possibility that Aβ peptides found in pancreatic cells derive from circulating Aβ originating from the brain. As Aβ and amylin share a common pathological pathway (71), the presence of Aβ peptide in the pancreas of 3xTg-AD may have damaged pancreatic β-cells. Such a hypothesis is fueled by the fact that amyloid pathology is associated with degeneration of islet cells in the pancreas of diabetic patients (71). Index of caspase activation and hypomorphic pancreatic islets found in 3xTg-AD mice fed the HFD also support this idea. Thus, consequent defects in pancreatic β-cell function could account, at least in part, for the reduction of insulin production in 3xTg-AD mice fed the HFD.

A particularly surprising result presented in this study is the rapid ability of insulin to decrease soluble brain Aβ levels. Within minutes, intravenous insulin administration cancelled the long-term effect of the HFD on soluble Aβ concentrations in the cortex and hippocampus of 3xTg-AD mice. To our knowledge, this is the first evidence for such a rapid effect of a treatment on soluble Aβ. Our subsequent investigations identified changes in molecular markers implicated in Aβ production, all altered by a single insulin injection, including increased α-APP, increased X11α, decreased BACE, and decreased autophagy-related proteins (53,55,72). Brain soluble Aβ concentration is known to depend on neuronal activity (56,68,73), possibly through changes in amyloidogenic APP processing (68). Such an effect can be rapid, as microdialysis data show that lowering neuronal activity decreases Aβ levels within a 1–4-h time frame in vivo (56). The present data suggest that insulin may act through these metabolic pathways since it acutely induced a reduction of CO1, a well-established marker of neuronal activity, which was correlated with Aβ42 levels. Together and in line with in vitro data (52), our results support an effect of insulin on Aβ production.

Through regulation of Aβ steady-state levels in the brain, Aβ clearance from the brain to cerebrospinal fluid and blood is believed to play a key role in the progression of late onset AD (74). We found that the decrease in Aβ levels in the brain coincided with a sharp rise in Aβ in plasma of 3xTg-AD mice following insulin injection. This suggests that the potentiation of Aβ clearance is the mechanism most likely to explain such a rapid insulin-induced reduction in brain Aβ concentrations. Therefore, our results argue for pleiotropic effects of insulin on complementary pathways, leading to decreased Aβ production and increased Aβ clearance within minutes.

The rapid effect of insulin on soluble Aβ levels observed has several implications. First, such quick variation induced by an endogenous peptide hormone calls for caution when using soluble Aβ concentrations as long-term surrogate markers of AD progression, in neuroimaging or as a biomarker in biological fluids (75). Still, that soluble Aβ varies so quickly in the brain is not so surprising based on recent data in humans indicating that soluble Aβ concentrations oscillate through the day and according to brain metabolic status (56,72,75,76). Nevertheless, the effect of insulin on brain and plasma Aβ was accompanied by improved memory performance in object recognition, supporting a therapeutic relevance of insulin action. Despite the possibility that insulin may have acted through mechanisms unrelated to Aβ, the coincidence is intriguing. Although such a rapid effect of insulin on Aβ concentrations needs first to be demonstrated in humans, our data strengthen the case already made by several authors (1,24,77) of using insulin as a therapeutic tool in AD. Because of the risk of hypoglycemia and the possible induction or exacerbation of peripheral insulin resistance, caution is warranted as the effects observed in this study follow an acute injection and may not be sustainable over a chronic treatment. However, intranasal administration could be an interesting alternative to deliver insulin to the brain, in the light of encouraging preliminary results in AD patients (77).

Taken together, our results suggest that in the 3xTg-AD mouse model of AD, centrally expressed Aβ can contribute to pancreatic cell degeneration in both AD and T2D, leading to defects in insulin production. Exaggerated dietary intake of fat-derived energy could trigger a vicious cycle for which the main consequences are further accumulation of Aβ in the brain and impaired cognition, which can be reversed by a single injection of insulin. If replicable in humans, these findings highlight the potential of correcting insulin signaling defects as a promising therapeutic tool to modulate cerebral concentrations of Aβ and treat Aβ-related symptoms of AD.

Funding. This study was made possible by funding from the Canadian Institutes of Health Research (MOP 102532 and IAO 74443), the Alzheimer Society of Canada, and the Canada Foundation for Innovation. M.V. was supported by a Canadian Institutes of Health Research Scholarship. The work by F.C. is supported by a salary award from the Fonds de la Recherche en Santé du Québec.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. M.V. designed experiments; performed animal experiments, Western blot, insulin ELISA, and immunofluorescence; and wrote the first versions of the manuscript. P.J.W. contributed to experiment design and manuscript writing. C.T. performed the pancreas RT-PCR and the ELISA for APP–C-terminal fragment and Aβ and helped with mice perfusion. I.S.-A. performed the ELISA for Aβ and helped with mice perfusion. G.C. performed the ELISA for C-peptide. V.E. contributed to experiment design. D.L. did the characterization of pancreatic islet size. J.V. performed the hippocampus Western blot. E.P. provided an expertise for tau analysis. Y.G. performed and provided expertise for plasma lipid analyses. A.M. performed the experimental design. F.C. conceived the experimental design and wrote the manuscript. F.C. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.