Cerebrovascular insulin receptors are defective in Alzheimer’s disease

Abstract

Central response to insulin is suspected to be defective in Alzheimer’s disease. As most insulin is secreted in the bloodstream by the pancreas, its capacity to regulate brain functions must, at least partly, be mediated through the cerebral vasculature. However, how insulin interacts with the blood–brain barrier and whether alterations of this interaction could contribute to Alzheimer’s disease pathophysiology both remain poorly defined.

Here, we show that human and murine cerebral insulin receptors (INSRs), particularly the long isoform INSRα-B, are concentrated in microvessels rather than in the parenchyma. Vascular concentrations of INSRα-B were lower in the parietal cortex of subjects diagnosed with Alzheimer’s disease, positively correlating with cognitive scores, leading to a shift towards a higher INSRα-A/B ratio, consistent with cerebrovascular insulin resistance in the Alzheimer’s disease brain. Vascular INSRα was inversely correlated with amyloid-β plaques and β-site APP cleaving enzyme 1, but positively correlated with insulin-degrading enzyme, neprilysin and P-glycoprotein. Using brain cerebral intracarotid perfusion, we found that the transport rate of insulin across the blood–brain barrier remained very low (<0.03 µl/g·s) and was not inhibited by an insulin receptor antagonist. However, intracarotid perfusion of insulin induced the phosphorylation of INSRβ that was restricted to microvessels. Such an activation of vascular insulin receptor was blunted in 3xTg-AD mice, suggesting that Alzheimer’s disease neuropathology induces insulin resistance at the level of the blood–brain barrier.

Overall, the present data in post-mortem Alzheimer’s disease brains and an animal model of Alzheimer’s disease indicate that defects in the insulin receptor localized at the blood–brain barrier strongly contribute to brain insulin resistance in Alzheimer’s disease, in association with β-amyloid pathology.

Leclerc et al. report that circulating insulin activates cerebral insulin receptors (INSR) located on brain capillary endothelial cells forming the blood–brain barrier. Experiments with human brain samples and animal models provide evidence that INSR at the BBB are impaired in Alzheimer’s disease, thereby contributing to brain insulin resistance.

See Hughes and Craft (https://doi.org/10.1093/brain/awac433) for a scientific commentary on this article.

See Hughes and Craft (https://doi.org/10.1093/brain/awac433) for a scientific commentary on this article.

Introduction

The human brain is sensitive to insulin, a hormone essential to life that is also at the heart of the pathophysiology and treatment of diabetes.^1,2 The scientific literature is replete with studies in humans and other species reporting the effects of insulin on memory, cerebral blood flow, eating behaviour and regulation of whole-body metabolism, supporting a therapeutic potential in brain-dependent metabolic disorders.^3–6 Although local synthesis has been detected, most insulin exercising an effect on the brain circulates in the blood after being produced by the pancreas.^7,8 The blood–brain barrier (BBB) is a major interface between the blood and brain, controlling access to cerebral tissues. Therefore, to exert an effect in the CNS, circulating insulin must first interact with its insulin receptor (INSR) located on brain capillary endothelial cells (BCEC) forming the BBB.^8–11

INSR is a disulphide-linked homodimer (αβ)₂ that structurally and genetically belongs to the class II of receptor tyrosine kinases. Alternative splicing produces two isoforms of the α-chain: the short A isoform (INSRα-A) truncated by 12 amino acids (exon 11) and the long B isoform (INSRα-B).^12–15 The regulation of this alternative splicing is not fully elucidated but appears to be tissue-specific.¹² Binding of an agonist to the extracellular α-chain triggers autophosphorylation of the transmembrane β-chain, which contains multiple phosphorylation sites leading to the activation of INSR and downstream signalling pathways.

Beside classical amyloid-β (Aβ) and tau pathologies, Alzheimer’s disease is characterized by defective brain uptake of glucose and impaired response to insulin.^4,5,16–19 The most compelling evidence comes from post-mortem indexes of brain insulin resistance, such as changes in INSR or IRS1 phosphorylation status shown to be associated with cognitive impairment.^20,21 Insulin administration in animals is reported to improve memory function, including in mouse models of Alzheimer’s disease.^22,23 Several small clinical trials using intranasal delivery to avoid the hypoglycaemic effect of systemically injected insulin have reported benefits to memory scores in cognitively impaired adults.^6,24,25 However, recent larger clinical trials produced unclear results.^6,25–27 Still, increasing insulin sensitivity in the brain remains a promising avenue and a wide range of repurposed diabetes drugs are in ongoing clinical trials for Alzheimer’s disease, including metformin, thiazolidinediones and glucagon-like peptide 1 (GLP-1) analogues.^6,28,29

Whereas cerebral insulin resistance is often assumed to be restricted to neurons in Alzheimer’s disease,^30–32 it is supported by limited immunohistochemical (IHC) evidence.^33,34 Most initial studies on the distribution of brain INSR used macroscopic techniques.^35–39 However, recent single-cell transcriptomic analyses indicate that the mRNA transcript encoded by the INSR gene is found in higher concentrations in endothelial cells in the mouse and human brain.^40–42 Accordingly, a growing number of studies are showing that INSR located in the cerebral vasculature plays a key role in the action of insulin on the brain.^11,43,44 Based on the current understanding, INSR at the BBB binds circulating insulin to either act (i) as a classic receptor to trigger cell-signalling pathways within BCEC; or (ii) as a transporter to ferry an insulin molecule into the brain parenchyma.^45–49 Therefore, it remains unclear whether the same INSR protein can exert both roles.^8,50–52

Given this large set of clinical and preclinical data showing a primary role of insulin on brain function, we aimed to determine how circulating insulin interacts with the BBB INSR and whether the latter is defective in Alzheimer’s disease. We first sought to determine INSR levels in microvascular extracts from both mouse and human brain samples relative to brain parenchyma. To probe for changes in Alzheimer’s disease, we used microvessel extracts of parietal cortex from participants in the Religious Orders Study who underwent detailed clinical and neuropsychological evaluations.^53–55 Associations with clinical and biochemical data were assessed, including BBB transporters and receptors putatively involved in Aβ transport. To directly investigate the response of brain INSR to circulating insulin, we performed intracarotid insulin perfusion and found that INSR activation is localized at the BBB, where it is impaired by Aβ and tau pathologies in a mouse model of Alzheimer’s disease at different ages.

Materials and methods

Human samples: Religious Orders Study

Parietal cortex samples were obtained from participants in the Religious Orders Study (ROS), a longitudinal clinical and pathological study of ageing and dementia.^53–55 Each participant enrolled without known dementia and agreed to an annual detailed clinical evaluation and brain donation at death. The study was approved by an Institutional Review Board of Rush University Medical Center. All participants signed an informed consent, an Anatomic Gift Act for brain donation and a repository consent allowing their data and biospecimens to be shared. A total of 21 cognitive performance tests were administered, of which 19 were used to create a global measure of cognition and five cognitive domains: episodic, semantic, working memory, perceptual speed and visuospatial ability.^56,57 Participants received a clinical diagnosis of Alzheimer’s dementia, mild cognitive impairment (MCI) or no cognitive impairment (NCI) (n = 20 for each group) at the time of death by a neurologist, blinded to all post-mortem data, as previously described.^56,58,59 In addition, current prescription medication usage in the last 2 weeks before each evaluation were available, such as antihypertensive and diabetes medications.^60,61 The neuropathological diagnosis is based on the ABC scoring method found in the revised National Institute of Aging–Alzheimer’s Association (NIA-AA) guidelines for the neuropathological diagnosis of Alzheimer’s disease.⁶² Four levels of Alzheimer’s disease neuropathological changes (not, low, intermediate or high) were determined using: (i) Thal score assessing phases of Aβ plaque accumulation⁶³; (ii) Braak score assessing neurofibrillary tangle pathology⁶⁴; and (iii) CERAD (Consortium to Establish a Registry for Alzheimer’s Disease) score assessing neuritic plaque pathology.⁶⁵ Individuals classified as ‘Controls’ had no or a low level of Alzheimer’s disease neuropathological changes while those classified as ‘AD’ included participants with intermediate or high levels of Alzheimer’s disease neuropathological changes.⁵⁵ Neuritic plaques and neurofibrillary tangles in the parietal cortex were counted following Bielschowsky silver impregnation, as previously described.⁶⁶ As previously described,^55,67,68 soluble and insoluble levels of Aβ₄₀ and Aβ₄₂ were also assessed in homogenates and microvascular extracts of parietal cortex. See Supplementary Table 1 for the characteristics of the cohort.

Animals

All animals had free access to laboratory food and water and were maintained under a 12-h light–dark cycle at 22°C. The 3xTg-AD mouse model of genetically induced Alzheimer’s dementia-like neuropathology is a widely used model that progressively develops both Aβ deposits and neurofibrillary tangles (τ pathology) as well as cognitive deficits, due to the expression of three mutant genes: β-amyloid precursors protein (APP_swe), presenilin-1 (PS1_MA146V) and tau (tau_P301L).^69,70 It fully develops neuropathological and behavioural changes ∼12 months of age.^71–73 Although the 3xTg-AD mouse is considered a model of modest overproduction of Aβ and tau, signs of defects in Aβ clearance have been detected at older ages.^22,74,75 Previous studies showed that 3xTg-AD mice also display metabolic disorders, such as glucose intolerance or impaired insulin sensitivity, aggravated by old age and high-fat diets.^{22,23,70,73,76–78} Brain Aβ concentrations in this model decrease after insulin administration.^22,79 Here, they were compared with non-transgenic (non-Tg) littermates with the same genetic background (C57Bl6/129SvJ). All groups of 3xTg-AD and non-Tg mice were randomly assigned to experimental conditions. For all experiments 3xTg-AD mice between 6 and 18 months were used for microvessel comparisons (42 males and 34 females) and for intracarotid in situ cerebral perfusion (ICSP, 35 males and 30 females). In addition, 6-month-old C57Bl/6 mice (11 males and 13 females) were used for insulin stimulation experiments (Supplementary Fig. 1), and for insulin transport quantification 15-week-old males Balb/c (n = 22), C57Bl/6 mice (n = 7), 13.5-month-old non-Tg (n = 20) and 3xTg-AD (n = 21) female mice were used.

Isolation of brain microvessels

Human brain microvessels were extracted following the procedure described previously.⁵⁵ Following a series of centrifugation steps, including a density gradient centrifugation with dextran and a filtration (20-µm nylon filter), two fractions were obtained: one enriched in cerebral microvessels, the other consisting of microvessel-depleted parenchymal cell populations. Examples of the efficiency of the separation are shown in Fig. 1, using immunodetection of endothelial and neuronal markers.⁵⁵ Proteins of both fractions were extracted using a lysis buffer (150 mM NaCl, 10 mM NaH₂PO₄, 1 mM EDTA, 1% Triton X-100, 0.5% SDS and 0.5% deoxycholate) containing the same protease and phosphatase inhibitors cocktails (Bimake), and supernatants were kept for western immunoblotting analyses. Brain microvessels from mice were generated with a similar procedure reported in previous work^54,55,80 after an intracardially perfusion with phosphate saline buffer containing a cocktail of protease inhibitors (SIGMAFAST™ Protease Inhibitor Tablets #S8820) and phosphatase inhibitors (1 mM sodium pyrophosphate and 50 mM sodium fluoride) under deep anaesthesia with ketamine/xylazine intraperitoneal (300/30 mg/kg). A scheme is shown in Supplementary Fig. 1.

Figure 1.

INSRα-B levels are reduced in Alzheimer’s disease, correlating with cognitive dysfunction. (A) Pro-INSR, INSRα and INSRβ are enriched in vascular fractions from the parietal cortex, along with endothelial markers claudin-5, ABCB1(P-gp) and CD31, whereas NeuN, a neuronal marker, is rather concentrated in the microvessel-depleted parenchymal fraction. (B) Schematic representation of the transmembrane INSR: the extracellular α chain binds circulating insulin and the intracellular β chain acts as a primary effector of insulin signaling pathway through auto-phosphorylation. INSRα is expressed as isoforms A and B due to alternative splicing (exon 11 is spliced for isoform B). (C) Human INSR is colocalized with claudin-5 and collagen IV in brain microvessels and hippocampal section, whereas no colocalization was observed with NeuN-labelled neurons in microvessel-depleted fractions (magnification ×20, scale bar = 20 μm). (D–M) Dot plots of the concentrations of INSR forms in microvessels comparing participants based on neuropathological diagnosis following the ABC criteria (D–H) or clinical diagnosis (I–M). Unpaired t-test ($P < 0.05) or one-way ANOVA followed by Tukey’s post hoc test (**P < 0.01). (N) Correlation between the levels of vascular INSRα-B and the global cognitive score. Linear regression analyses were controlled for educational level, age at death, sex and apoE genotype. F ratio and P-value are shown. (O and P) Dot plots of the concentrations of pro-INSR and INSRα-B in microvessels comparing APOE4 carriers based on the ABC neuropathological diagnosis. Welch-ANOVA followed by Dunnett’s post hoc test (¤P < 0.05). (Q) Representative western blots of consecutive bands are shown. Data were log transformed for statistical analysis and are represented as scatter plots with a logarithmic scale. Horizontal bars indicate mean ± SEM. ABC = Dx Neuropathological Diagnosis; A-AD = Alzheimer’s disease; ColIV = Collagen-IV; C = Control; Clinical Dx = Clinical Diagnosis; CypB = Cyclophilin B; EC = Extracellular; IC = Intracellular; O.D. = optical density; P = Microvessel-depleted parenchymal fraction; T = Total homogenate; Va = Vascular fraction enriched in microvessels.

Acute exposition of the BBB to insulin using intracarotid in situ cerebral perfusion

The i n situ cerebral perfusion (ISCP) technique consists in a direct cerebral infusion of a labelled compound at the BBB through the right common carotid artery.⁸¹ Briefly, mice were anaesthetized by intraperitoneal injection of a mixture of ketamine/xylazine intraperitoneally (140/8 mg/kg). The right external and the right common carotid artery were ligated, and the common carotid artery was catheterized with polyethylene tubing (0.30 × 0.70 mm) filled with heparin (25 IU/ml). The syringe containing the perfusion fluid was placed in an infusion pump and connected to the catheter. The thorax was opened, the heart cut and perfusion started immediately. The perfusion fluid contained bicarbonate buffered physiologic saline containing 128 mM NaCl, 24 mM NaHCO₃, 4.2 mM KCl, 2.4 mM NaH₂PO₄, 1.5 mM CaCl₂, 0.9 mM MgCl₂ and 9 mM D-glucose. The solution was gassed with 95% O₂–5% CO₂ for pH control (7.4), filtrated (0.20 µm) and used at 37°C. For radiolabelled insulin transport experiment mice were perfused with 0.1 nM of ¹²⁵I-insulin, 0.3 µCi/ml of ³H-sucrose as a vascular marker and 20 nM S961 for competition. The physiological buffer was supplemented with 0.25% bovine serum albumin (BSA) and the perfusion lasted 120 s at 1.25 ml/min. For insulin stimulation studies, 6-month-old C57Bl6 and 16-month-old non-Tg and 3xTg-AD mice were perfused for 120 s, at 2.5 ml/min, with insulin (100 or 350 nM) or saline.

Immunofluorescence analysis of isolated human microvessels

The method was similar to previous publications.^54,55 Both vascular and microvessel-depleted extracts on glass side were fixed using a 4% paraformaldehyde solution in phosphate buffer saline (PBS) for 20 min at room temperature (RT) and then blocked with a 10% normal horse serum and 0.1% Triton X-100 solution in PBS or commercial TrueBlack® Background Suppressor and Blocking Buffer (Biotium) for 1 h at RT. For Aβ immunolabeling, an additional pretreatment with 90% formic acid during 10 min was performed between the fixation and blocking steps. Following an incubation overnight at 4°C with primary antibodies diluted in a 1% NHS and 0.05% Triton X-100 solution in PBS, or TrueBlack^® Blocking Buffer (Biotium), vascular and microvessel-depleted extracts were incubated with secondary antibodies diluted in the same solution as the primary antibodies during 1 h at RT. Primary and secondary antibodies are listed in Supplementary Table 2. Cell nuclei were counterstained with DAPI (Thermo Fisher Scientific, 0.02% in PBS) and slides were mounted with Mowiol mounting medium or Prolong™ Diamond antifade (Thermo Fisher Scientific). If needed, slides were incubated with TrueBlack® Plus (Biotium) and TrueView® (Vector Laboratories) to decrease autofluorescence and background. Between each step, three washes of 5 min in PBS were performed.

Images were taken using a fluorescence microscope (EVOS Fl Autoimaging system, Thermo Fisher Scientific) at magnification ×20–40 or a laser scanning confocal microscope (Olympus IX81, FV1000) with sequential scanning acquisition using optimal z-separation at a magnification of ×20.

Western blot

For western immunoblotting, proteins from the vascular fraction were extracted and a protein quantification was done using bicinchoninic acid assays (Thermo Fisher Scientific). Protein homogenates from human and murine brain microvascular extracts were added to Laemmli’s loading buffer and samples were heated for 10 minutes at 70°C (human and 3xTg-AD samples) or not (samples from insulin stimulation studies). Equal amounts of proteins per sample (4–8 µg) were resolved on a sodium dodecyl sulphate–polyacrylamide gel electrophoresis (SDS–PAGE) 8% acrylamide. All samples, loaded in a random order, were run on the same immunoblot experiment for quantification.

Proteins were electroblotted on PVDF membranes, which were then blocked during 1 h at RT with Superblock™ in PBS blocking buffer (Thermo Fisher Scientific) or PBS containing 0.1% Tween 20 and 3% BSA. If needed, membranes were stained with No-Stain™ protein labelling reagent (Thermo Fisher Scientific) to assess a similar protein load in the samples. Membranes were then incubated overnight at 4°C with primary antibodies (all listed in Supplementary Table 2). Membranes were then washed three times with PBS containing 0.1% Tween 20 and incubated during 1 h at RT with the secondary antibody in PBS containing 0.1% Tween 20 and 1% BSA. Membranes were probed with chemiluminescence reagent (Luminata Forte Western horseradish peroxidase substrate, Millipore) and imaged using the myECL imager system (Thermo Fisher Scientific) or the Amersham Imager 680 (GE Healthcare Bio-Sciences). Densitometric analysis was performed using the Image Lab™ Software. Uncropped gels of immunoblotting experiments conducted with human samples are shown in Supplementary Fig. 2.

ELISA for human Aβ and APP-βCTF

Concentrations of β-secretase-derived βAPP fragment (APP-βCTF or C99) in detergent-soluble extracts of brain microvessels (parietal cortex) were determined using Human APP-βCTF ELISA Assay kit (IBL) according to the manufacturer’s instructions.

Statistical analysis

Statistical analyses were performed using GraphPad Prism 9.0 and JMP 15 software packages. The threshold for statistical significance was set to P < 0.05. Homogeneity of variance and normality were determined for all data sets using D’Agostino and Pearson’s or Shapiro–Wilk normality tests. When normality was verified, unpaired Student’s t-tests were used to identify significant differences between two groups. Otherwise, the Welch correction or a Mann–Whitney test was performed. When more than two groups were compared, parametric one-way ANOVA followed by Tukey’s multiple comparison tests were performed unless variances were different, in which case Welch-ANOVA followed by Dunnett’s multiple comparison tests were used. When both normality and variance equality were not confirmed, a non-parametric Kruskall–Wallis test followed by Dunn’s multiple comparison was performed. Relative optical density values obtained were log transformed when needed to reduce variances and provide more normally distributed data when appropriate. For insulin stimulation experiments, data were presented as a percentage of the control group to facilitate the comparison of INSR activation between groups. For ISCP experiments, a ROUT test (Q = 1%) available on GraphPad Prism was performed to determine outliers. Multivariate analyses were used to assess the association between continuous variables. Significance of correlations with ante-mortem clinical scores were adjusted for the following covariates: education level, sex, age at death and Apolipoprotein E (APOE) genotype.

Study approval

All procedures performed with volunteers included in this study were in accordance with the ethical standards of the institutional ethics committees and with the 1964 Helsinki Declaration. Written informed consent was obtained from all individual participants included in this study. All procedures relating to mouse care and experimental treatments were approved by the Laval University animal research committee (CPAUL) in accordance with the standards of the Canadian Council on Animal Care.

Data availability

The data that support findings of this study are available from the corresponding author on reasonable request. Data from the ROS can be requested at https://www.radc.rush.edu. Database of gene expression in adult brain and perivascular cells are available at http://betsholtzlab.org/VascularSingleCells/database.html (Betsholtz laboratory), and for human (control and AD) at https://twc-stanford.shinyapps.io/human_bbb/ (Tony Wyss-Coray laboratory).

Results

INSRs are enriched in human brain microvessels: levels of the isoform INSRα-B are reduced in Alzheimer’s disease

To determine the post-mortem localization of the INSR in the human brain, we first extracted microvessels from parietal cortex samples collected from ROS participants. We opted for a soft procedure using sequential centrifugation and filtration steps, adapted for frozen samples, as previously described.^54,55 We observed that INSR were present in higher concentrations in vascular fractions (Va) containing brain microvessels, compared to microvessel-depleted parenchymal fractions (P) and whole brain homogenates (T) (Fig. 1A). The validity of the method was confirmed by the coinciding enrichment in endothelial markers claudin-5, ATP binding cassette protein B1 (ABCB1) (P-gp) and platelet endothelial cell adhesion molecule (CD31) in the same vascular fractions, whereas neuronal nuclei (NeuN) staining was more prominent in microvessel-depleted fractions (Fig. 1A), as shown in our previous work.^54,55

INSR is a heterotetrametric receptor combining two dimers composed of both α (extracellular) and β (intracellular) chains, recognizable by different antibodies (Fig. 1B and Supplementary Table 2) but all coded from the same INSR gene. INSRα incorporates the binding site for circulating insulin and exists in two isoforms depending on the absence or presence of exon 11, depicted as A and B, respectively. The full-size precursor of the INSR, pro-INSR and both INSRα and INSRβ isomers were all enriched in vascular fractions (Fig. 1A).

The localization of INSRα on microvessels was confirmed by confocal immunofluorescence where the signal of human INSRα antibody overlapped with claudin-5, a tight-junction protein of brain endothelial cells, but was almost absent in microvessel-depleted fractions containing NeuN-positive cells (Fig. 1C and Supplementary Fig. 3). In addition, immunosignals for INSRβ and collagen IV, a basal membrane protein, colocalized in human microvessels from microvascular-enriched fractions and brain sections (Fig. 1C).

Western blot analyses revealed that INSRα-B levels were lower in vascular fractions from participants with a neuropathological (−48%) or clinical (−40% and −59% compared to NCI and MCI, respectively) diagnosis of Alzheimer’s disease, but not significant versus Controls (Fig. 1G and L). Whereas levels of pro-INSR, INSRβ or INSRα-A did not differ (Fig. 1D–F and I–K). A lower INSRα-B was also found in association with Thal, Braak and CERAD ratings (Supplementary Fig. 4). In the parenchymal fractions, where the INSRα B isoform was absent, levels of the INSRα A isoform were not different between groups (Supplementary Fig. 2). Interestingly, participants with a neuropathological or clinical diagnosis of Alzheimer’s disease had a higher vascular INSRα-A/B ratio (Fig. 1H and M), a molecular index of insulin resistance in the liver and adipose tissue.^12,82–85 In addition, ante-mortem global cognition was positively correlated with the cerebrovascular content in INSRα-B (Fig. 1N), but not to other INSR isoforms (Supplementary Fig. 5). No significant association between INSR levels and age of death (Supplementary Table 1) was observed. Finally, participants with Alzheimer’s disease carrying one apoE4 allele had lower concentrations of pro-INSR (−83% and −83%) and INSRα-B (−40% and −56%) compared to Alzheimer’s disease non-carriers and control participants, respectively (Fig. 1O and P). Immunoblots are shown in Fig. 1Q and in Supplementary Fig. 2.

Levels of vascular INSR correlate with Aβ pathology or BBB proteins involved in Aβ production and clearance

We next investigated the association between vascular INSR and several markers of Aβ pathology previously assessed in whole homogenates and/or in microvascular extracts from the parietal cortex from the same series of ROS participants (Fig. 2A).^55,67,68 Significant inverse correlations were found between vascular levels of INSRα-B, neuritic plaques, as well as vascular β-secretase (BACE1, β-secretase-derived βAPP fragment), an enzyme implicated in APP cleavage and Aβ formation (Fig. 2B and C). By contrast, INSRα-B levels in vascular extracts were positively correlated with vascular neprilysin and insulin-degrading enzyme (IDE), two enzymes involved in Aβ degradation (Fig. 2A and D). In addition, strong associations were detected between vascular INSRα-B, ABCB1 (P-gp) and low-density lipoprotein receptor-related protein 1 (LRP1), implicated in Aβ efflux,^86,87 suggesting that more INSRα-B at the BBB is associated with a higher clearance of Aβ from brain to blood (Fig. 2A and E). However, soluble and insoluble _­levels of Aβ₄₀ or Aβ₄₂ in parietal cortex proteins homogenates did not show any association with vascular INSRα-B. A positive association between INSRβ and total soluble tau [F(1,50) = 7.92; P = 0.0079] and an inverse association with neurofibrillary tangles [F(1,50) = −7.10; P = 0.0141] were observed, but not with total insoluble tau or other forms of INSR (Fig. 2A). Of note, total soluble tau does not differentiate Alzheimer’s disease versus control in this cohort, whereas insoluble total tau and neurofibrillary tangles are higher in Alzheimer’s disease (Supplementary Table 1).^67,68 Finally, we noted that INSRα-B and most of the other INSR isoforms were positively associated with caveolin-1 and endothelial nitric oxide synthase (eNOS), two proteins involved in receptor recycling and endothelial generation of nitric oxide, respectively (Fig. 2A, F and G).

Figure 2.

Cerebrovascular levels of INSRα-B correlate with neuritic plaques and Aβ-related proteins located on the BBB. (A–G) Correlations between cerebrovascular INSRs and neurofibrillary tangle counts, phosphorylated-tau T231/S235 (AT180), S396/404 (AD2, confirmed with PHF1) (Tau neuropathology), cortex neuritic plaque counts and Aβ concentrations (Aβ-neuropathology) in brain homogenates. Correlations were also established with other proteins assessed in microvessel-enriched fractions. Soluble and insoluble proteins are found in TBS-soluble and formic acid-soluble fractions, respectively. Linear regressions analyses were controlled for educational level, age at death, sex and ApoE genotype, and were performed to generate F-ratios and P-values in the heatmap (•P < 0.05; ••P < 0.01; •••P < 0.001; ••••P < 0.0001). Red and blue highlighted cells, respectively, indicate significant positive and negative correlations (A). Graphical representation of noteworthy correlations between INSRα-B and Aβ-related markers (B–D) and transporters (E), as well as microvessel markers (F and G). APP = amyloid precursor protein; O.D. = optical density; RAGE = receptor for advanced glycation end products.

INSR is altered in 3xTg-AD mouse microvessels

To assess the localization of INSR in the mouse brain, we performed the same extraction procedures used with human samples. We observed a vascular enrichment for all isoforms, but stronger for pro-INSR and INSRα-B, compared to total homogenates or microvessel-depleted parenchymal fractions (Fig. 3A), corroborating human data. The INSRα-A/B ratio in brain microvessels was lower in mice compared to humans (Figs 1A and 3A). The concomitant enrichment of endothelial markers claudin-5, CD31 and ABCB1 (P-gp) in vascular fractions validated the method.

Figure 3.

Lower vascular INSR in old 3xTg-AD mice. (A) Representative western blots showing enriched content in pro-INSR, INSRα and INSRβ in vascular fractions compared to parenchymal fractions (rich in neuronal marker NeuN) and total homogenates from the mouse brain, along with known endothelial markers claudin-5, CD31 and ABCB1(P-gp). (B–F) Vascular levels of pro-INSR, INSRβ, INSRα-A, INSRα-B and BACE1 in 3xTg-AD and Non-Tg mice at 6, 12 and 18 months of age (42 males and 34 females). Two-way ANOVA followed by Tukey’s post hoc test (*P < 0.05; ns = non-significant), and linear trend model. Data are presented as mean ± SEM. (G) Representative western blots of consecutive bands are shown. 3xTg-AD = tri-transgenic mice; CypB = Cyclophilin B; EC = Extracellular; IC = Intracellular; NTg-Non-Tg = non-Tg mice; O.D. = optical density; P = Microvessel-depleted parenchymal fraction; T = Total homogenate; Va = Vascular fraction enriched in microvessels.

To determine whether changes in INSR are a consequence of classical Alzheimer’s disease neuropathology, we applied the same methodology to brain samples from 3xTg-AD mice at different ages (6, 12 and 18 months). Multivariate analyses showed that the 3xTg-AD genotype was associated with lower INSRα-A or pro-INSR levels, although differences at each age remained non-significant (Fig. 3B, D and G). Similar to the findings in human microvessel extracts, no difference between groups was observed for INSRβ (Fig. 3C and G). Interestingly, INSRα-B levels in microvessel-enriched fractions were lower in 3xTg-AD mice at 18 months of age compared to non-Tg mice (-51.1%), in accordance with human data (Fig. 3E and G). A significant linear trend from 6 to 18 months (r² = −0.1828, P = 0.0116) and a genotype–age interaction were detected, consistent with an age-induced decrease of INSRα-B only in 3xTg-AD animals (Fig. 3E and G). By contrast, we also found a significant genotype–age interaction for vascular BACE1, associated with a linear upward trend with age in 3xTg-AD mice (r² = 0.2130, P = 0.0076) (Fig. 3F and G). These results unveil a pattern of changes of INSR, particularly INSRα-B, induced by age and Aβ/tau pathologies, in agreement with human data.

The insulin-induced activation of INSR at the BBB is blunted in 3xTg-AD mice

Our results so far indicate that cerebral INSR are concentrated in microvessels, but with levels decreasing along with Alzheimer’s disease pathology. Previous published data indicate that Alzheimer’s disease is associated with reduced activation of INSR and its downstream intracellular signalling.^4,20,21 However, it remains to be shown where blood-borne insulin activates INSRs in the brain, and whether this activation is disturbed by Alzheimer’s disease pathology. Like other members of the tyrosine kinase family of receptors, the binding of an agonist triggers autophosphorylation of the INSR. However, its implication as a transporter to ferry circulating insulin into the brain parenchyma has been recently disputed.⁸ We thus directly assessed the transport of insulin through the BBB using intracarotid ISCP, with or without an INSR-specific competitive antagonist S961 (Fig. 4A). Perfused ¹²⁵I-insulin was set at 0.1 nM, close to the physiological concentration of circulating insulin in the mouse.^88,89 We found that the brain uptake coefficient of ¹²⁵I-insulin was very low (0.028 ± 0.007 µl/g·s) compared to a highly diffusible compound such as diazepam (45.6 ± 2.6 µl/g·s). In addition, S961 (20 nM) did not block¹²⁵I-insulin transport. We also confirmed that neither insulin nor S961 had an impact on BBB permeability using ³H-sucrose (Fig. 4A and B). Finally, 3xTg-AD transgenes had no significant impact on ¹²⁵I-insulin transport or BBB integrity in 13.5-month-old mice (Fig. 4C). These data indicate that INSR at the BBB does not act as the main transporter for insulin, in accordance with published in vitro and in vivo data.^50,51,90

Figure 4.

The activation of INSR after intracarotid injection of insulin is restricted to brain microvessels and is blunted in 3xTg-AD mice. (A) Illustration of the ISCP technique. (B) Brain uptake coefficient of ¹²⁵I-insulin (0.01 nM) perfused alone or with S961 (20 nM), a selective INSR antagonist competing with insulin, in Balb/c mice aged 15 weeks. ³H-Diazepam (0.3 µCi/µl) was perfused in age-matched C57Bl6 mice to emphasize the difference between a highly diffusible drug and insulin (left). No significant change in cerebrovascular volume due to the treatment was observed by coperfusing the vascular marker ³H-sucrose (0.3 µCi/ml) (right). Unpaired t-test or parametric one-way analysis of variance (ns, non-significant). Data are presented as mean ± SEM. (C) Brain uptake coefficient of ¹²⁵I-insulin (0.01 nM) comparing non-Tg and 3xTg-AD mice aged 13.5 months (left). No significant change in cerebrovascular volume due to the treatment was observed by coperfusing the vascular marker ³H-sucrose (0.3 µCi/ml) (right). Unpaired t-test (ns, non-significant). Data are presented as mean ± SEM. (D) Representative western blots showing phosphorylated INSRβ in vascular fractions following perfusion of insulin (Aspart or Toronto) by ISCP in 6-month-old C57Bl6 mice, compared to parenchymal fractions. Endothelial markers CD31 and ABCB1 (P-gp) as well as the neuronal markers β-tubulin III, NeuN and GAP43 and were immunoblotted on the same membranes. (E–L) Dot plots of the concentrations of INSR isoforms in microvessels comparing mice based on genotype and insulin perfusion (E–I), and cerebrovascular proteins BACE1, eNOS and caveolin-1 (J–L). One-way or two-way analysis of variance followed by Tukey’s post hoc test (*P < 0.05; ****P < 0.0001). Data were log transformed for statistical analysis. Outliers were removed from statistical analyses as described in the 'Methods section'. Horizontal bars indicate mean ± SEM. (M) Representative western blots of consecutive bands are shown, from the same samples, but from a different randomization. 3xTg-AD = tri-transgenic mice; As = insulin Aspart (NovoRapid^®); CypB = Cyclophilin B; Diaz = Diazepam; Ins = insulin perfusion; GAP43 = Growth Associated Protein 43; NTg-Non-Tg = non-Tg mice; P = Microvessel-depleted parenchymal fraction; S961 = INSR antagonist; S-Sal = saline perfusion; To = regular insulin Toronto (Novolin^® ge); Va = Vascular fraction enriched in microvessels.

Next, we used ISCP to infuse insulin in the carotid, to assess the activation of cerebral INSR in vivo and compare microvessel-enriched versus parenchymal fractions. We first showed that, in 6-month-old C57Bl6 mice, intracarotid administration of insulin (0.4 mg/kg or 350 nM) triggers phosphorylation of tyrosine residues (Y1150/1151) on INSRβ within 120 seconds compared to the same physiological buffer without insulin (Fig. 4D). Faster acting Aspart insulin NovoRapid^® and regular insulin Novolin^®ge Toronto (350 nM) were compared and both led to the phosphorylation of INSR (Fig. 4D and Supplementary Fig. 6). Strikingly, this activation of INSRβ was only observed in microvascular extracts, while no trace was detected in the parenchymal fraction despite low but detectable levels of INSRs (Fig. 4D). This result indicates that INSR at the BBB is activated by circulating insulin and is consistent with a brain response to circulating insulin occurring predominantly at the BBB level, not in parenchymal cells.

To assess whether the activation of INSR is altered by Aβ and tau neuropathologies, we performed the same intracarotid insulin injection experiments in 3xTg-AD mice followed by microvessel isolation and immunoblotting. While insulin-induced phosphorylation of INSRβ in microvessels of 16-month-old non-Tg mice, such activation was greatly attenuated in 3xTg-AD mice of the same age (Fig. 4E and M), with both types of insulin used (hexameric Novolin^®ge Toronto or monomeric Aspart insulin NovoRapid^®). In contrast, levels of INSRβ or pro-INSR were similar between groups (Fig. 4F, G and M). Replicating the data shown in Fig. 3E in 18-month-old 3xTg-AD mice, INSRα-B concentrations in vascular extracts were lower in this group of 15-month-old 3xTg-AD mice (Fig. 4H and M). The vascular INSRα-A/B ratio was higher in 3xTg-AD mice, further supporting insulin resistance at the level of the BBB (Fig. 4I and M).

We also assessed the effect of intracarotid insulin on other vascular proteins in non-Tg and 3xTg-AD mice and no difference was found in the levels of BACE1, neprilysin, IDE and other Aβ transporters (P-gp/ABCB1, LRP1 or receptor for advanced glycation endproducts) (Fig. 4J and M and Supplementary Fig. 6). This is consistent with the 3xTg-AD mouse being a model of Aβ overproduction, rather than a model of reduced Aβ clearance. However, a lower concentration of caveolin-1 was observed in 3xTg-AD compared to non-Tg mice (Fig. 4K–M).

Results of these experiments show that Alzheimer’s disease-like neuropathology impairs the activation of vascular INSRs in the 3xTg-AD mouse model. This collection of data indicates that circulating insulin primarily interacts with INSRs exposed on the luminal side of the BBB in both human and mice and that the INSR-mediated response is blunted in Alzheimer’s disease.

The reduction of INSR is associated with BACE1 activity in the BBB

The strong inverse correlation between BACE1 and INSR in the neurovascular compartment led us to explore whether a decrease in INSR levels could be a consequence of BACE1 protease activity. The capacity of BACE1 to cleave INSR has been demonstrated in the muscle and liver, therein contributing to insulin resistance.⁹¹ Higher BACE1 cleavage activity of the INSR β-chain has been recently associated with cognitive impairment and type-2 diabetes in a clinical cohort study.⁹² BACE1 levels in the cerebrovasculature are higher in Alzheimer’s disease and associated with cognitive impairment.⁵⁵

To further assess BACE1 activity in microvessels, we measured the concentration of APPβ-CTF, a marker of BACE1 cleaving activity. Microvascular levels of APPβ-CTF were significantly higher in Alzheimer’s disease compared to controls on the basis of the ABC neuropathological diagnosis (Fig. 5A), whereas a similar upwards trend was noted in Alzheimer’s disease compared to MCI on the basis of the clinical diagnosis (P = 0.053) (Fig. 5B). We first observed a strong positive association between vascular APPβ-CTF and INSRα-A/B ratio [F(1,40) = 12.6; P = 0.002], and a significant inverse correlation with INSRα-B [F(1,40) = −10.9; P = 0.003] (Fig. 5C and D). In contrast, vascular APPβ-CTF showed a significant positive correlation with BACE1 [F(1,42) = 45.7; P < 0.0001] (Fig. 5E). Immunofluorescence experiments confirmed the localization of both BACE1 and Aβ₄₂ on capillaries (Fig. 5F and G). These combined data highlight the importance of BACE1 in Alzheimer’s disease for both cerebrovascular Aβ pathology and insulin resistance, and suggest a role for BACE1 in cleaving vascular INSR (Fig. 5F–H).

Figure 5.

BACE1 cleavage product APPβ-CTF is negatively associated with human neurovascular levels of INSRα-B. (A and B) Dot plots of the concentrations of APPβ-CTF in microvessels comparing participants based on neuropathological diagnosis following the ABC criteria (A) or clinical diagnosis (B). Unpaired t-test ($P < 0.05) or parametric one-way analysis of variance followed by Tukey’s post hoc test. Data were log transformed for statistical analysis and are represented as scatter plots with a logarithmic scale. Horizontal bars indicate mean ± SEM. (C–E) Correlations between the levels of vascular APPβ-CTF, INSRα-A/B ratio, INSRα-B and BACE1, in human brain vascular fractions. Linear regression analyses were controlled for educational level, age at death, sex and ApoE genotype. F-ratio and P-values are shown. (F) Representative immunofluorescence labelling of human INSRα showing the colocalization with Aβ₄₂ and collagen IV in brain microvessels of an Alzheimer’s disease patient with stage 4 cerebral amyloid angiopathy (CAA) pathology (magnification ×40, scale bar = 20 μm). (G) Representative immunofluorescence labelling of BACE1 showing its colocalization with collagen IV in brain microvessels of an Alzheimer’s disease patient (magnification ×20, scale bar = 20 μm). (H) Illustration of the role of β-secretase in cleaving APP and, hypothetically, INSR in the brain vasculature of Alzheimer’s disease patients. ABC Dx = neuropathological diagnosis; AD = Alzheimer’s disease; APP = amyloid precursor protein; APPβ-CTF = β-secretase-derived βAPP C-terminal fragment; AICD = amyloid precursor protein intracellular domain; BACE1 = β-site APP cleaving enzyme; CAA = cerebral amyloid angiopathy; Clinical Dx = clinical diagnosis; ColIV = collagen IV; EC = extracellular; IC = intracellular; O.D. = optical density; sAPPβ = β-secretase-derived βAPP soluble fragment.

Discussion

The present study provides evidence that INSR is enriched at BBB and could underlie the brain insulin resistance observed in Alzheimer’s disease. We specifically show that: (i) vascular INSR defects are linked to Alzheimer’s disease pathology and symptoms in humans and that the activation of INSR is reduced by Alzheimer’s disease neuropathology in a mouse model; and (ii) INSR located on the BBB is involved in the rapid response to circulating insulin, where it acts as a tyrosine kinase receptor not a transporter. Our data suggest that pancreas-produced insulin interacts primarily with the INSR on the luminal side of the brain vasculature and that its response becomes defective in Alzheimer’s disease, possibly following cleavage by BACE1. This highlights a previously unrecognized role of INSR in the cerebral vasculature as a systemically accessible drug target in Alzheimer’s disease.

INSR localization in brain microvessels

So far, the exact cellular localization of INSR in the brain has remained uncertain due to conflicting evidence. Autoradiography studies in rat brain sections reported prominent binding levels of ¹²⁵I-insulin in the hippocampal formation,³⁶ choroid plexus³⁸ and hypothalamus,⁹³ as well as in the olfactory bulb, cerebral cortex and cerebellum.³⁷ A similar distribution pattern of INSRα was found using in situ hybridization in rat brain.³⁹ Western blot analyses detected INSRα in the human cerebral cortex and hippocampus but not in the white matter.³⁵ These studies, however, did not provide insights on the cellular location of the receptor. Initial studies performed by a single group using IHC in paraformaldehyde-fixed brain sections reported a localization of INSRβ in different areas of the rat forebrain (olfactory bulb, hypothalamus and hippocampus), largely exhibiting localization in cells resembling neurons.^33,34 However, more recent studies with RNA sequencing in mouse and human brain cells showed that INSR is largely expressed in endothelial and glial cells.^40–42 An older study with autoradiograms of ¹²⁵I-insulin hinted toward a 130-kDa doublet corresponding to isoforms -B and -A of INSRα located in cerebral microvessels.¹⁰ These discrepancies may be explained by a recurring challenge in the difficulty of finding appropriate antibodies for the INSR, and its isoforms, particularly when using human samples.

The present set of data indicates that most INSR isoforms in the human and mouse brain are predominantly localized in microvessels, and this was particularly the case for the isoform INSRα-B, which was virtually absent from the brain parenchyma. Beside this preferential localization of INSR in microvessels, which has been underestimated in previous years, a significant fraction of INSR is also present in other cells within the parenchyma, where they can interact with low levels of insulin in the brain interstitial fluid. Indeed, cultured astrocytes and neurons respond to insulin, and cell-specific knockouts demonstrate that removing INSR from these cells affects insulin signalling in vivo.^94–99 Thus, our results do not rule out a key role of INSR located on neurons or astrocytes. Nonetheless, they indicate that BBB INSR is probably the primary cerebral target that responds to insulin circulating in the blood after secretion by the pancreas.

Microvascular INSR at the BBB as the site of brain insulin resistance in Alzheimer’s disease

The lower levels of vascular INSRα-B in subjects with either a neuropathological or clinical diagnosis of Alzheimer’s disease, which were inversely associated with cognitive scores, can be interpreted as a sign of brain insulin resistance in Alzheimer’s disease. Although the INSRα-B isoform was reduced, pro-INSR, INSRβ and INSRα-A did not differ significantly between groups. No association between cerebrovascular INSR and age was found in the ROS cohort, contrasting with previous studies that reported dysregulation of circulating insulin and reduced transport, binding and signalling in the brain with ageing.^44,100–106 We found corroborating data in old 3xTg-AD mice, where INSRα-B was decreased at 18 months of age compared to non-Tg controls. This shift towards a higher INSRα-A/B isoform ratio at the BBB in Alzheimer’s disease may have mechanistic implications. INSR isoforms have been the subject of more scrutiny in peripheral organs (muscle, adipose tissue and liver), in cancers and in metabolic diseases, such as obesity and diabetes.^{83–85,107–109} Whereas the cellular implication of a change in INSRα-A/B ratios remains unclear, the short INSRα-A is preferentially expressed in foetal tissues containing proliferating cells (growth signal) and INSRα-B is predominant in adult tissue in differentiated cells (metabolic signal).¹² INSR formed of INSRα-A or -B isoforms would show a dissimilar affinity for insulin or tyrosine kinase activity according to some but not all studies.^{12–14,110–112} Higher INSRα-A/B ratios are observed in adipocytes from obese patients and in the liver from patients with type-2 diabetes, both returning to normal after weight loss and diabetes remission.^83,84 Therefore, a higher INSRα-A/B ratio in insulin sensitive organs has been proposed as a molecular index of insulin resistance.¹² In sum, the observed higher vascular INSRα-A/B ratio can be interpreted as a sign of insulin resistance in the cerebrovasculature, possibly hindering the capacity of the INSR to trigger intracellular events/signalling in Alzheimer’s disease.

Circulating insulin activates INSR in microvessels but not in the parenchyma: this activation is blunted in animal models of Alzheimer’s disease

One of the main observations reported here is that the phosphorylation of INSRβ after direct intracarotid infusion of insulin was detected at the BBB level and not in the parenchyma. Despite the presence of INSRs and neurons in the parenchymal fractions, no evidence of increased phosphorylation of INSRβ was detected even at a very high dose. This suggests that the physiological short-lived peak of insulin in the blood, typically observed after a meal, exerts most of its action on INSR located on the BBB and less in neurons. Whereas most cells in microvessel extracts are BCEC^54,55 containing INSR mRNA transcripts,^40–42 it is important to note that pericytes and astrocyte feet are also present in smaller amounts. It is notable that this insulin-induced INSRβ phosphorylation, observed in all control mice, was greatly attenuated in brain microvessels from 3xTg-AD mice, indicating that Alzheimer’s disease neuropathology impairs INSR signalling. This defect in INSR signalling was observed even with the high concentration of insulin used. Previous studies using post-mortem ex vivo stimulation of whole brain human slices with insulin have reported lower phosphorylation of INSRβ (Y1150/1151 and Y960) in the cerebellar cortex and the hippocampal formation in Alzheimer’s disease patients.²¹ However, similar studies performed in the middle frontal gyrus cortex did not find associations between p-INSRβ and diabetes, Aβ burden, tau tangle density or cognitive scores, although pS⁴⁷³ AKT1 was associated with less angiopathy.^20,113 Although post-mortem human data cannot be used to infer any causal relationship, which may be bidirectional, changes in INSR levels and signalling observed in old 3xTg-AD mice are likely to be a consequence of Alzheimer’s disease-like pathology and/or transgenes, and perhaps related to the other metabolic defects reported in this model.^{22,73,76,78,114} The present results suggest that INSRs located on microvessels contribute to the previously reported evidence of central insulin resistance in Alzheimer’s disease.

Of note, INSR isoforms can form hybrids with IGF-1 receptors (IGF-1R),^{36,37,115,116} and lower IGF-1R levels and impaired signalling have been associated with Aβ and tau neuropathology in the brain of Alzheimer’s disease patients.^117–121 Along with INSR, lower IGF-1R phosphorylation was observed following ex vivo stimulation with insulin in hippocampal sections from Alzheimer’s individuals.²¹ Interestingly, insulin displays a greater affinity for INSRα-B homodimers compared to INSRα-A homodimers and hybrid receptors.^110,122 Further studies are thus warranted to determine the contribution of IGF-1R alone or within INSR/IGF-1R complexes, in the response of the brain to insulin, particularly in Alzheimer’s disease.

One must also keep in mind that the ISCP paradigm used here detects the acute effect of insulin that reaches the BBB interface. Longer exposure times to insulin may be necessary to transmit an effect to neurons deeper in the brain parenchyma. However, a report using hyperglycaemia clamp and microdialysis showed that a longer peripheral exposure to insulin did not result in increased insulin levels in brain interstitial fluid, nor in higher hippocampal Akt phosphorylation.¹²³

The loss of INSR is associated with a pattern of protein changes implying lower clearance and higher production of Aβ

Our report highlights the possible association between vascular INSR levels and an altered equilibrium between production and clearance of Aβ at the level of the BBB. On the one hand, INSRα-B was found to be positively associated with ABCB1 (P-gp) and LRP1, two endothelial proteins involved in Aβ efflux from the parenchyma through the BBB to the blood.^55,86,87 Similar relationships were found with neprilysin and IDE, two key enzymes involved in Aβ degradation. Neprilysin levels were previously found to be reduced in Alzheimer’s disease microvessels,^55,124 while Aβ can be a substrate of IDE competing with insulin.^125,126 While cerebrovascular INSRα-B was associated with neuritic plaque counts in the cortex and Aβ clearance proteins in the BBB, no such relationship was detected with Aβ₄₂ concentrations. This may suggest that other variables are in play in the regulation of brain Aβ levels, such as alternative clearance pathways.^55,127,128 Published mechanistic data show that Aβ (i) can promote lysosomal degradation of INSR in 3xTg-AD murine and porcine brain capillaries⁴³; and (ii) acts as a competitive antagonist of insulin to the INSR, thereby contributing to insulin resistance.^16,129 On the other hand, INSRα-B was negatively correlated with BACE1, a β-secretase initiating the cleavage of amyloid precursor protein (APP) and formation of APPβ-CTF and Aβ¹³⁰: both of which are found in higher concentrations in Alzheimer’s disease brain or microvessels.^55,131–133 Using induced pluripotent stem cell (iPSC) neurons carrying APP and/or PSEN1 familial Alzheimer’s disease mutations, a recent study reported that BACE1-produced APPβ-CTF alter endosomal and intracellular trafficking, which might contribute to BBB defects.¹³⁴ Finally, BBB INSRα-B was also associated with levels of endothelial nitric oxide synthase (eNOS or NOS3) and caveolin-1, two Alzheimer’s disease-relevant vascular proteins involved in insulin uptake, INSR trafficking and signalling.^135,136 Altogether, these observations indicate that the preservation of INSR signalling in the BBB is associated with increased clearance of Aβ, lower BACE1 activity, neurovascular integrity and sustained cognitive performance.

What is the cause of INSR decrease in Alzheimer’s disease?

Overall, these series of observations are consistent with the hypothesis that Alzheimer’s disease is accompanied by a reduction in binding sites available to circulating insulin at the level of the BBB. Several mechanisms can explain the observed decrease in INSRα-B. First, we found corroborating data in old 3xTg-AD mice, where INSRα-B was decreased at 18 months of age with no changes in INSRβ. This suggests that age and Alzheimer’s disease neuropathology are involved. Second, it could be proposed that microvascular cells expressing INSRs are specifically lost in Alzheimer’s disease. This hypothesis is, however, unlikely, as we did not find a significant reduction in endothelial cells in this cohort as assessed with claudin-5, CD31 or CypB.⁵⁵ A third explanation involving a change at the gene transcription level is also unlikely since the decrease was specific to the isoform INSRα-B, whereas INSRα-A, INSRβ and pro-INSR were mostly unaffected. Fourth, preferential alternative splicing of INSR or isoform-specific conversion could be at play, but underlying mechanisms are not fully understood and little studied in the field of neurology.^12,137,138 Candidate splicing factors and protein convertases are expressed at low level in BCEC and their activity could underlie the specific reduction in INSRα-B observed here.^40,41,139 Overall, we postulate that post-translation modifications are more likely to underlie the observed decrease of the long isoform INSRα-B.

The strength of the correlation between vascular levels of BACE1 and INSRα-B, as well as the vascular INSRα-A/B ratio, was remarkable. A strong association was also established between INSRα-B and the concentration of APPβ-CTF, which is considered as an index of BACE1 activity.^130,134,140 BACE1 is probably the most studied enzyme in Alzheimer’s disease, due to its role in the production of Aβ and subsequent accumulation in the brain.^132,133 BACE1 is also expressed in microvascular endothelial cells,^{40,41,139,141} in higher concentrations in persons with Alzheimer’s disease, correlating with cognitive impairment.⁵⁵ However, BACE1 is not only involved in the cleavage of APP, but it also has multiple other substrates, including INSR.¹⁴⁰ Indeed, reports indicate that INSR is the target of proteases, such as calpain 2 and presenilin 1 (γ-secretase), but also BACE1, which has been previously proposed as a mechanism of regulation of INSR signalling.^91,142,143 Notably, studies in the liver show that BACE1 cleaves INSR between amino acids 933–956 of the β-chain, disturbing insulin signalling and generating a soluble fragment (sINSRα, insulin receptor soluble fragment).^91,143 In the liver of db/db mice, higher BACE1 protein and mRNA levels, combined with the generation of sINSRα was observed.⁹¹ Conversely, BACE1^−/− mice have significantly less plasma sINSRα and higher liver INSRβ levels.⁹¹ More recent data suggest that BACE1 cleaving activity of INSR may be related to cognitive impairment and type-2 diabetes. The elevated plasma levels of BACE1 were found to correlate with the generation of sINSRα and glycemia in diabetics patients (Type I and II).^92,144 In addition, increased plasma BACE1 activity was proposed as a biomarker for predicting the conversion from MCI to Alzheimer’s disease.¹⁴⁵ Keeping these published data in mind, the series of strong correlations between BACE1 levels, BACE activity index (APPβ-CTF) and INSR levels, all measured in cerebral microvessel extracts, suggest a mechanistic scheme in which BACE1-mediated cleavage of INSR contributes to insulin resistance in Alzheimer’s disease. A favourable clinical implication is that drugs may not need to fully cross the BBB to alter INSR posttranslational modifications or to mimic the action of pancreas-secreted insulin on the brain.

Funding

This work was supported the Canadian Institutes of Health Research (CIHR) to F.C. (grant number PJT 168927). The study was supported in part by P30AG10161, P30AG72975 and R01AG15819 (D.A.B). F.C is a Fonds de recherche du Québec-Santé (FRQ-S) senior research scholar. M.L. was supported by a scholarship from the Fondation CHU de Québec. P.B held scholarships from the Fondation du CHU de Québec, a joint scholarship from the FRQ-S and the Alzheimer Society of Canada (ASC) and a scholarship from the CIHR. V.C. was supported by a scholarship from Fonds d’Enseignement et de la Recherche (FER) from the Faculty of Pharmacy, Laval University.

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.