Equipe Holzenberger - George C et al. Brain 2017.

06 - Novembre - 2017

The Alzheimer's disease transcriptome mimics the neuroprotective signature of IGF-1 receptor-deficient neurons. Brain. 2017.

The Alzheimer’s disease transcriptome mimics the neuroprotective signature of IGF-1 receptor-deficient neurons

Caroline George Géraldine Gontier Philippe Lacube Jean-Christophe François Martin Holzenberger Saba Aïd

Brain, Volume 140, Issue 7, 1 July 2017, Pages 2012–2027, https://doi.org/10.1093/brain/awx132

Abstract

Seminal studies using post-mortem brains of patients with Alzheimer’s disease evidenced aberrant insulin-like growth factor 1 receptor (IGF1R) signalling. Addressing causality, work in animal models recently demonstrated that long-term suppression of IGF1R signalling alleviates Alzheimer’s disease progression and promotes neuroprotection. However, the underlying mechanisms remain largely elusive. Here, we showed that genetically ablating IGF1R in neurons of the ageing brain efficiently protects from neuroinflammation, anxiety and memory impairments induced by intracerebroventricular injection of amyloid-β oligomers. In our mutant mice, the suppression of IGF1R signalling also invariably led to small neuronal soma size, indicative of profound changes in cellular homeodynamics. To gain insight into transcriptional signatures leading to Alzheimer’s disease-relevant neuronal defence, we performed genome-wide microarray analysis on laser-dissected hippocampal CA1 after neuronal IGF1R knockout, in the presence or absence of APP/PS1 transgenes. Functional analysis comparing neurons in early-stage Alzheimer’s disease with IGF1R knockout neurons revealed strongly convergent transcriptomic signatures, notably involving neurite growth, cytoskeleton organization, cellular stress response and neurotransmission. Moreover, in Alzheimer’s disease neurons, a high proportion of genes responding to Alzheimer’s disease showed a reversed differential expression when IGF1R was deleted. One of the genes consistently highlighted in genome-wide comparison was the neurofilament medium polypeptide Nefm. We found that NEFM accumulated in hippocampus in the presence of amyloid pathology, and decreased to control levels under IGF1R deletion, suggesting that reorganized cytoskeleton likely plays a role in neuroprotection. These findings demonstrated that significant resistance of the brain to amyloid-β can be achieved lifelong by suppressing neuronal IGF1R and identified IGF-dependent molecular pathways that coordinate an intrinsic program for neuroprotection against proteotoxicity. Our data also indicate that neuronal defences against Alzheimer’s disease rely on an endogenous gene expression profile similar to the neuroprotective response activated by genetic disruption of IGF1R signalling. This study highlights neuronal IGF1R signalling as a relevant target for developing Alzheimer’s disease prevention strategies.

Introduction

Increased life expectancy in the modern world exposes more and more humans to age-related diseases. Investigating the complex mechanisms of ageing provided strong evidence that blocking insulin-like growth factor (IGF) pathways prolongs lifespan in various species, including mammals (Holzenberger et al., 2003; Kappeler et al., 2008; Kenyon, 2010; Xu et al., 2014; Milman et al., 2016). Recently, inhibition of IGF1 and its principal regulator growth hormone has been proposed as a promising strategy to retard human ageing and extend healthy lifespan (Longo et al., 2015). Importantly, interventions slowing down ageing also postpone age-related pathologies, including Alzheimer’s disease (Cohen et al., 2006; Florez-McClure et al., 2007), the most common neurodegenerative pathology. Alzheimer’s disease is characterized by extracellular amyloid-β deposits and intracellular inclusions of tau aggregates that ultimately cause progressive, and so far irremediable decline of cognitive functions. Cohen et al. (2009) showed that lifespan-extending heterozygous IGF1R knockout (KO) confers neuroprotection and improves behaviour in Alzheimer’s disease mice (Cohen et al., 2009). In the brains of Alzheimer’s disease patients, abnormalities in insulin like growth factor 1 receptor (IGF1R) expression and downstream signalling molecules, coupled with insulin and IGF1 resistance have also been demonstrated, although causality remains unclear (Steen et al., 2005; Moloney et al., 2010; Bomfim et al., 2012; Talbot et al., 2012). Indeed, evidence is accumulating that long-term blockade rather than enhancement of IGF signalling supports neuronal function and neuroprotection (Freude et al., 2009; Gontier et al., 2015; Gazit et al., 2016; De Magalhaes Filho et al., 2017). We found recently that deleting IGF1R from adult neurons of young APP/PS1 mice protected them lifelong from Alzheimer’s pathology by clearing toxic amyloid-β via preserved autophagic compartment and enhanced systemic elimination (Gontier et al., 2015). Together, these findings suggested strong mechanistic involvement of neuronal IGF signalling in Alzheimer’s disease progression. However, interventional and systematic approaches to a better understanding of the interaction between Alzheimer’s disease progression and IGF signalling in neurons are missing. Here, we report that neuron-specific deletion of IGF1R, even when performed in old mice, still protected against amyloid-β oligomer (AβO)-induced memory impairment and neuroinflammation. Using microarray analysis we compared transcriptional responses of hippocampal neurons in the early stages of Alzheimer’s disease with neurons deprived of IGF signalling in vivo. This showed that highly overlapping molecular pathways were affected in the same direction in both conditions, especially those involved in neurotransmission, cellular stress response and cell growth. We also identified a set of IGF-dependent pathways that orchestrate neuroprotective mechanisms against amyloid-β proteotoxicity in adult Alzheimer’s disease neurons.

Materials and methods

Animal models

CaMKIIα-CreERT2+/0;IGF1Rflox/flox and APPswe/PS1dE9+/0;CaMKIIα-CreERT2+/0;IGF1Rflox/flox transgenic mice were generated as described (Gontier et al., 2015). Briefly, APPswe/PS1dE9 and CaMKIIα-CreERT2 transgenes were backcrossed to IGF1Rflox/flox knock-in mice to obtain APPswe/PS1dE9+/0;IGF1Rflox/flox and CaMKIIα-CreERT2+/0;IGF1Rflox/flox breeders. These two mutant genotypes were crossed to generate CaMKIIα-CreERT2+/0;IGF1Rflox/flox and APPswe/PS1dE9+/0;CaMKIIα-CreERT2+/0;IGF1Rflox/flox mice. Male and female mice were used. As no major differences prevailed between sexes, results are displayed together, except for amyloid-β enzyme-linked immunosorbent assay (ELISA; see below). Mice were housed in individually ventilated cages, enriched with a red polycarbonate shelter (Tecniplast) and a cotton pad. The maximum number of mice per cage was five males or six females. Mice were kept on a 12/12 h light/dark cycle at 22°C room temperature, and free access to water and standard mouse chow (LASQCdiet Rod16, Genobios). To induce neuron-specific IGF1R knockout, CaMKIIα-CreERT2+/0;IGF1Rflox/flox and APPswe/PS1dE9+/0;CaMKIIα-CreERT2+/0;IGF1Rflox/flox mice were injected intraperitoneally with tamoxifen twice per day for 5 consecutive days (42 mg/kg body weight per injection), resulting in inIGF1RKO and ADINKO conditional mutant mice, respectively. Tamoxifen (T5648, Sigma-Aldrich) was dissolved in sunflower seed oil/ethanol (10:1) at 10 mg/ml. Littermate control mice of the same genotypes were injected with vehicle alone. Efficiency of Cre-loxP recombination was systematically checked post-mortem in brain biopsies by PCR. This study was conducted in accordance with the guidelines for care and use of laboratory animals and European Community rules (86/609/EEC). All experiments were performed with the approval of Comité d’Éthique pour l’Expérimentation Animale “Charles Darwin” (specific approval Ce5/2012/043 and 01166.02).
Genotyping

DNA was isolated from tissue biopsies and subjected to multiplex PCR as described (Gontier et al., 2015).


Amyloid-β oligomers

AβOs were prepared from synthetic peptide amyloid-β42 (Tocris Bioscience); phosphate-buffered saline (PBS) or reverse peptide amyloid-β42-1 were used as controls as described (Choi and Bosetti, 2009; Santos et al., 2012). Briefly, peptides were dissolved at 0.1 mg/ml in 99% hexafluoroisopropanol (HFIP) (105228, Sigma-Aldrich). HFIP was evaporated to obtain dry films that were resuspended in PBS at 1 μg/μl and incubated at 37°C for 48 h. AβOs were used within 12 h of preparation. AβOs were characterized by western blot using anti-amyloid-β 6E10 antibody (not shown) and by transmission electron microscopy.

Intracerebroventricular injections

Twenty-five-month-old inIGF1RKO and control mice were anesthetized by intraperitoneal injection of 100 mg ketamine and 10 mg xylazine per kg body weight. We administered 400 pmol of AβOs (aggregated from amyloid-β42) or vehicle (PBS) into the lateral ventricle (Aid et al., 2008). Stereotaxic coordinates were −0.5 mm antero-posterior, −1.0 mm medio-lateral, −2.3 mm dorso-ventral from bregma (Paxinos and Franklin, 2013). AβOs were injected with a 10 µl microsyringe fitted with 33-gauge needle and automated syringe pump (Stoelting) at 0.5 μl/min, in a final volume of 4 μl. After delivery, the needle was kept in place for 5 min to prevent backflow.

Barnes maze test

All behavioural experiments were performed in a dedicated room between 9 am and 6 pm, alternating experimental groups throughout this interval. The Barnes maze was used to assess spatial learning and memory as described (Gontier et al., 2015). Mice were trained four times per day for four consecutive days to escape from a brightly lit circular platform into a refuge box hidden beneath the target hole (one of the 20 holes positioned along the perimeter). Each trial lasted up to 5 min with an intertrial interval of at least 15 min. The maze was cleaned with 70% ethanol after each trial. On Day 5, long-term memory retention was evaluated during a 3-min session. Latency, errors (wrong hole) and distance travelled were recorded by automated video tracking (Viewer3, Biobserve). Individual search strategy was classified by the observer as ‘spatial’ (straight to target or adjacent hole), ‘serial’ (several holes visited in a sequential manner before reaching the target), or ‘random’ (not spatial nor serial). Mice ‘failed’ if the target was not located within 3 min.

Open-field test

Mice were placed in the middle of the 44 × 64 cm arena delimited by opaque walls, and allowed to explore for 10 min. Distance travelled and time spent in the centre zone (20 × 36 cm) versus surrounding periphery were recorded by automated video tracking. After each test, mice returned to their home cages, and the arena was thoroughly cleaned with 70% ethanol.

Antibodies

We used primary antibodies against Akt (9272, 1:1000; Cell Signaling Technology, CST), P-Akt (Ser473) (4058, 1:1000; CST), CREB (06-863, 1:1000; Millipore), P-CREB (Ser133) (06-519, 1:500; Millipore), drebrin (ab12350, 1:1000; Abcam), ERK (p44/42 MAPK) (4696, 1:1000; CST), P-ERK (P-p44/42 MAPK) (Thr202/Tyr204) (9106, 1:1000; CST), GAPDH (sc-20357, 1:1000; Santa Cruz Biotechnology), GFAP (Z0334, 1:2000; Dako), Iba1 (019–19741, 1:1000; Wako), IGF1Rβ (3027, 1:1000; CST), IRβ (sc-711, 1:1000; Santa Cruz Biotechnology), neurofilament medium polypeptide (NEFM) (ab64300, 1:1000; Abcam), NeuN (MAB377, 1:1000; Millipore), synapsin I (AB1543P, 1:1000; Millipore), S6 ribosomal protein (2317, 1:1000; CST), and P-S6 ribosomal protein (Ser235/236) (4858, 1:1000; CST). Secondary antibodies were Alexa Fluor® 488 anti-mouse antibody (A11001, 1:1000; Life Technologies), biotinylated anti-rabbit antibody (BA-1000, 1:200; Vector Laboratories), horseradish peroxidase (HRP)-conjugated anti-rabbit antibody (656120, 1:500; Life Technologies), HRP-conjugated anti-mouse antibody (626520, 1:500; Life technologies), and IRDye-conjugated anti-mouse (925-32210, 1:20 000; LI-COR) and anti-rabbit (926-68021, 1:20 000; LI-COR) antibodies.

Histology and immunohistochemistry

Mice were transcardially perfused with PBS 0.1 M followed by 4% paraformaldehyde (PFA) at 4°C under deep pentobarbital anaesthesia. Brains were incubated in 4% PFA overnight at 4°C, transferred to 30% sucrose for 48 h, and frozen in isopentane at −40°C. Serial coronal sections (30 μm) were chosen to include the lateral ventricle. Sections were stained with 1% cresyl violet (C-5042, Sigma-Aldrich), or immunolabelled using specific antibodies and diamino-benzidine (K3468, Dako) reaction. Sections were incubated overnight with primary antibody at 4°C, and then incubated for 1 h at room temperature with secondary antibody. Images were acquired using a DM5000B microscope (Leica) and analysed using ImageJ software (National Institutes of Health).

Morphology of neurons and microglia

We analysed neuronal morphology as described (Gontier et al., 2015). In brief, nucleus and soma size were determined by manually tracing the circumference of each NeuN-positive nucleus and each cresyl violet-stained neuronal soma using ImageJ. Four cortical brain sections (30-μm thick) per mouse were analysed. On average, 100 neurons of motor cortex layer V were measured per animal. To evaluate microglial morphology, we used four Iba1-stained sections per mouse (n = 4–6 mice/group). Total length per microglia was determined in motor cortex using AnalyzeSkeleton (ImageJ) (Leinenga and Götz, 2015).

Western blot

Mice were transcardially perfused with cold PBS under deep pentobarbital anaesthesia. Cortex and hippocampus were quickly dissected on ice and snap-frozen in isopentane. Brain tissue was homogenized in RIPA buffer (50 mM Tris, 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1% Triton™ X-100, 0.1% SDS, 0.5% sodium deoxycholate) containing phosphatase inhibitors (10 mM NaF, 1 mM NaVO3) and protease inhibitors (Roche) using a Polytron (Kinematica). Homogenates were centrifuged at 14 000 rpm for 20 min at 4°C. Supernatants were collected, and protein concentration determined using DC protein assay (Bio-Rad). For SDS-PAGE, 15, 40 or 60 μg of protein were electrophoresed through 4–20% gradient tris-glycine gels (Bio-Rad) and transferred onto polyvinylidene fluoride (PVDF) membranes. PVDF membranes were blocked for 1 h at room temperature with 5% non-fat dry milk in Tris-buffered saline containing 0.1% Tween 20. Membranes were incubated overnight at 4°C with primary antibodies followed by HRP- or IRDye-conjugated secondary antibodies for 1 h at room temperature. Blots were scanned using Odyssey CLx (LI-COR) or revealed with ECL (WP20005, Life Technologies) and bands visualized using ChemiDoc and Quantity One 4.2.1 (Bio-Rad). Digital scans of entire membranes are in the Supplementary material. All western blot experiments were performed at least twice. Signals were quantified using Image Lab software (Bio-Rad) or Image Studio (LI-COR), and were normalized to GAPDH or to total proteins determined from Ponceau S-stained membranes.

Amyloid-β ELISA

Amyloid-β42 and amyloid-β40 peptides were extracted from cortical tissue, separated into formic acid-treated insoluble and diethylamine (DEA)-soluble fractions, and quantified using human specific-ELISA (Invitrogen) as described (Dansokho et al., 2016). Results in Fig. 1H–J are from females. Males (not shown) revealed slightly lower amyloid-β levels (Gontier et al., 2015), but no difference between genotypes.


Figure 1



Blocking neuronal IGF signalling in old mice during advanced stages of Alzheimer’s disease does no longer protect from amyloid pathology. (A) We induced neuronal IGF1R knockout at age 17 months by intraperitoneal tamoxifen injection in APPswe/PS1dE9;CaMKIIα-CreERT2;IGF1Rflox/flox mice that exhibit late-stage amyloid pathology (ADINKO). Alzheimer’s disease mice (AD) received vehicle and served as control. Mice were submitted to cognitive tests 5 months later and amyloid pathology evaluated thereafter. (B–G) Learning memory was assessed by training in a Barnes maze, four times per day for 4 days, and all groups of mice succeeded in learning (repeated measures ANOVA for time P < 0.0001) (B–D). Long-term memory retention was assessed on Day 5 (E–G). n = 15–20 mice. Males and females performed similarly and groups were combined. (H and I) Amyloid-β40 and amyloid-β42 abundance in cortical homogenates of old ADINKO and AD mice. Soluble (H) and insoluble fractions (I) were measured by human specific ELISA. n = 9–11 mice. (J) Plasma amyloid-β40 and amyloid-β42 were assayed in old ADINKO and AD mice. n = 9–11 mice. (K) Wet weight of ADINKO brain diminished. Mann-Whitney U-test, ***P < 0.001, n = 20 mice. Data are mean ± SEM.

Blocking neuronal IGF signalling in old mice during advanced stages of Alzheimer’s disease does no longer protect from amyloid pathology. (A) We induced neuronal IGF1R knockout at age 17 months by intraperitoneal tamoxifen injection in APPswe/PS1dE9;CaMKIIα-CreERT2;IGF1Rflox/flox mice that exhibit late-stage amyloid pathology (ADINKO). Alzheimer’s disease mice (AD) received vehicle and served as control. Mice were submitted to cognitive tests 5 months later and amyloid pathology evaluated thereafter. (B–G) Learning memory was assessed by training in a Barnes maze, four times per day for 4 days, and all groups of mice succeeded in learning (repeated measures ANOVA for time P < 0.0001) (B–D). Long-term memory retention was assessed on Day 5 (E–G). n = 15–20 mice. Males and females performed similarly and groups were combined. (H and I) Amyloid-β40 and amyloid-β42 abundance in cortical homogenates of old ADINKO and AD mice. Soluble (H) and insoluble fractions (I) were measured by human specific ELISA. n = 9–11 mice. (J) Plasma amyloid-β40 and amyloid-β42 were assayed in old ADINKO and AD mice. n = 9–11 mice. (K) Wet weight of ADINKO brain diminished. Mann-Whitney U-test, ***P < 0.001, n = 20 mice. Data are mean ± SEM.

Blocking neuronal IGF signalling in old mice during advanced stages of Alzheimer’s disease does no longer protect from amyloid pathology. (A) We induced neuronal IGF1R knockout at age 17 months by intraperitoneal tamoxifen injection in APPswe/PS1dE9;CaMKIIα-CreERT2;IGF1Rflox/flox mice that exhibit late-stage amyloid pathology (ADINKO). Alzheimer’s disease mice (AD) received vehicle and served as control. Mice were submitted to cognitive tests 5 months later and amyloid pathology evaluated thereafter. (B–G) Learning memory was assessed by training in a Barnes maze, four times per day for 4 days, and all groups of mice succeeded in learning (repeated measures ANOVA for time P < 0.0001) (B–D). Long-term memory retention was assessed on Day 5 (E–G). n = 15–20 mice. Males and females performed similarly and groups were combined. (H and I) Amyloid-β40 and amyloid-β42 abundance in cortical homogenates of old ADINKO and AD mice. Soluble (H) and insoluble fractions (I) were measured by human specific ELISA. n = 9–11 mice. (J) Plasma amyloid-β40 and amyloid-β42 were assayed in old ADINKO and AD mice. n = 9–11 mice. (K) Wet weight of ADINKO brain diminished. Mann-Whitney U-test, ***P < 0.001, n = 20 mice. Data are mean ± SEM.

Transmission electron microscopy

Ten microlitres of amyloid-β aggregated solution were adsorbed onto 200-mesh carbon and Formvar-coated grids (CF200-Cu-50, Euromedex), and negatively stained with 2.5% uranyl acetate for 45 s. Grids were examined using an electron microscope (EM 912 OMEGA, Zeiss) at 80 kV, and images captured with a digital camera (2 k × 2 k side-mounted TEM CCD, Veleta).

Laser microdissection and RNA extraction

Brains from 6-month-old female mice were quickly dissected, snap-frozen in isopentane at −40°C and stored at −80°C until use. We collected 30 coronal sections of 50-μm thickness per animal using a freezing microtome. Frozen sections were mounted on PEN-membrane 1 mm glass slides (#415101-4401-600, Zeiss), fixed for 2 min at 4°C in 70% ethanol, and stained in 1% cresyl violet (C-5042, Sigma-Aldrich). Sections were then dehydrated in 50% (30 s), 70% (30 s) and 100% ethanol (1 min) at 4°C, and air-dried. All solutions and materials were RNase-free to prevent RNA degradation. Laser assisted-microdissection was performed using a P.A.L.M. MicroBeam system with RoboSoftware (Zeiss). Total RNA was extracted from microdissected hippocampal CA1 by RNeasy® Micro Kit (74004, Qiagen) according to manufacturer’s instructions and eluted with 14 μl of RNase-free water. RNA concentrations were determined using NanoDrop 1000 (Thermo Scientific).

Microarray processing and data analysis

After validation of RNA quality using Agilent RNA6000 nano chips and Bioanalyzer 2100, we reverse transcribed 25 ng of total RNA using Ovation Pico WTA System V2 (NuGEN). The resulting double strand cDNA was used for amplification based on SPIA technology. After purification according to the NuGEN protocol, 5.2 μg of sense target DNA were fragmented and biotin-labelled using Encore Biotin Module kit (NuGEN). After control of fragmentation using Bioanalyzer 2100, cDNA was hybridized to GeneChip Mouse Transcriptome Assay (MTA) 1.0 (Affymetrix) at 45°C for 17 h. After overnight hybridization, chips were washed on fluidic station FS450 following Affymetrix protocols and scanned using GCS3000 7 G. Scanned images were analysed using Expression Console software (Affymetrix) to obtain raw data and metrics for quality control. Observations of metrics and distributions of raw data revealed no experimental outlier. Data were normalized using robust multiarray averaging (RMA, Bioconductor software). We first controlled and analysed data by principal component analysis (PCA) and used one-way ANOVA to extract differentially expressed genes with Genomics Suite (Partek). We performed expression analysis to identify single genes, but also to discover differentially activated key cell functions. Enrichment analyses were performed using Ingenuity Pathway Analysis (IPA, Qiagen) and Pathway Studio (PWS, Insilicogen) software, with threshold fixed to P < 0.02.

Statistical analysis

Data are presented as mean ± standard error of the mean (SEM). Sample size for each experiment was estimated based on previous experiments using these models and/or pilot studies (Cohen et al., 2009; Gontier et al., 2015). Mice matched by gender, age, genotype and weight were randomly assigned to experimental groups. Investigators were blinded to group allocation (AβOs, genotype) during experiments, data collection and analysis. No mice were excluded from analysis except in behavioural studies based on the preset criterion: animal does not move during test. Statistical tests were performed using Prism (GraphPad). F-test was conducted to test normal distribution and homogeneity of variance. For normally distributed data, groups were compared using two-tailed unpaired Student’s t-test or ANOVA. Otherwise we used Mann-Whitney U-test or Kruskal-Wallis test. Data from Barnes maze experiments were analysed by repeated measures (RM) two-way ANOVA and Fisher’s exact test. P ≤ 0.05 was considered significant. Microarray data were analysed by one-way ANOVA using a threshold of P < 0.02 to select differentially expressed genes.

Results

Inactivation of neuronal IGF signalling during advanced-stage Alzheimer’s disease does not mitigate amyloid pathology

We previously showed that blocking adult neuronal IGF signalling during presymptomatic disease phase alleviates subsequent amyloid pathology and prevents cognitive deficits in APP/PS1 (AD) mice, mainly by favouring amyloid-β clearance (Gontier et al., 2015). To gain further insight, also into the therapeutic potential of interfering with IGF signalling, we here asked whether disrupting IGF1R during symptomatic Alzheimer’s disease stages could still affect a full-blown disease phenotype. We therefore knocked out IGF1R in CaMKIIα-positive forebrain neurons of old (age 17 month) APP/PS1 mice (a mutant that we called ADINKO), when APP/PS1 mice exhibit advanced amyloid pathology and multiple Alzheimer’s disease symptoms (Fig. 1A). To find out whether IGF1R suppression still improved Alzheimer’s disease-related behavioural deficits, we analysed spatial memory using the Barnes maze 5 months later. Both ADINKO and AD mice succeeded in learning the maze through repeated training sessions, as evidenced by decreased distance travelled, shorter latency, and fewer cumulative errors over days (Fig. 1B–D). Similar to this unchanged learning ability, memory retention was also indistinguishable between ADINKO and AD mice (Fig. 1E–G). To directly assess the effects of IGF1R inactivation in old mice on advanced amyloid pathology, we measured cortical abundance of amyloid-β peptides in soluble and insoluble fractions. Again, and in clear contrast to receptor inactivation performed in young mice (Gontier et al., 2015), we found that after inactivation at advanced age, cortical amyloid-β40 and amyloid-β42 levels were similarly elevated in ADINKO and AD mice (Fig. 1H and I), and the same was true for plasma amyloid-β40 and amyloid-β42 (Fig. 1J). These results indicated that suppressing neuronal IGF1R at advanced-stage Alzheimer’s disease could not mitigate advanced brain amyloid pathology, nor did it improve systemic amyloid-β clearance. Yet, our neuron-specific IGF1R knockout induced at old age was not lacking efficiency: Cre-lox gene excision was as wide-spread as in brains induced in young mice, and loss of IGF1R led to the same marked changes in cell homeostasis and typical reduction in neuronal soma size, that were also evidenced by slightly, but significantly diminished brain weight (Fig. 1K; −7%, P < 0.0001) and reduced cerebral cortical mass (hemi-cortex: ADINKO 72.1 ± 1.5 mg versus AD 87.4 ± 1.2 mg, n = 20–21, −17%, P < 0.0001).

To selectively address the predicted neuroprotective effects of neuronal IGF1R knockout induced in old brains, we switched to an in vivo mouse model of acute amyloid-β proteotoxicity, consisting of AβO intracerebroventricular injection in mice with inducible neuronal IGF1R knockout (inIGF1RKO). Inducing IGF1R knockout alone at old age produced efficient hippocampal Cre-lox excision of floxed IGF1R alleles (Fig. 2A and B) and a marked (55%) local decrease in IGF1R protein (P < 0.001), while insulin receptor levels remained unchanged (Fig. 2C). In accordance with the relative proportion of neurons versus other cell types (Herculano-Houzel et al., 2006), results indicated that IGF1R was ablated in almost all CaMKIIα-positive hippocampal neurons. We reported that neuronal IGF1R knockout at young age reduced neuronal soma size in an Alzheimer’s disease context (Gontier et al., 2015). Here we showed that IGF1R ablation in old mice similarly leads to significant decrease in neuronal soma size (−17%, P < 0.05), while neuron density and cell nucleus size remained unchanged (Fig. 2D and E). Diminished neuronal cytoplasm also provoked reduced brain weight in inIGF1RKO mice (Fig. 2F; −5%, P < 0.05), demonstrating that IGF signalling exerts robust and lifelong control of soma size in post-mitotic neurons.


Figure 2



Deleting IGF1R from aged neurons entails histomorphological changes in the forebrain. (A) Inducible neuron-specific IGF1R knockout (inIGF1RKO) was achieved combining CaMKIIα-CreERT2 transgene with floxed Igf1r alleles (IGF1Rflox/flox). CaMKIIα-CreERT2;IGF1Rflox/flox mice were intraperitoneally injected with tamoxifen at age 17 month; controls received vehicle alone. (B) Gene deletion (IGF1RKO) monitored by PCR in hippocampus (n = 10–12 mice per group). (C) Western blot analysis of hippocampus. Mann-Whitney U-test, ***P < 0.001, n = 10 mice per group. (D) Representative micrographs of cresyl violet-stained cortical neurons. Neuronal soma size was reduced in inIGF1RKO cortex while neuron density remained unchanged. Two-tailed unpaired Student’s t-test, *P ≤ 0.05, n = 7–8 mice per group. Four brain sections were analysed per mouse. (E) Cell nucleus size was measured from NeuN-stained sections, n = 10–12 mice per group. (F) Wet weight of inIGF1RKO and control brain. Mann-Whitney U-test, *P ≤ 0.05, n = 7 mice per group. Bar graphs represent mean ± SEM.

Deleting IGF1R from aged neurons entails histomorphological changes in the forebrain. (A) Inducible neuron-specific IGF1R knockout (inIGF1RKO) was achieved combining CaMKIIα-CreERT2 transgene with floxed Igf1r alleles (IGF1Rflox/flox). CaMKIIα-CreERT2;IGF1Rflox/flox mice were intraperitoneally injected with tamoxifen at age 17 month; controls received vehicle alone. (B) Gene deletion (IGF1RKO) monitored by PCR in hippocampus (n = 10–12 mice per group). (C) Western blot analysis of hippocampus. Mann-Whitney U-test, ***P < 0.001, n = 10 mice per group. (D) Representative micrographs of cresyl violet-stained cortical neurons. Neuronal soma size was reduced in inIGF1RKO cortex while neuron density remained unchanged. Two-tailed unpaired Student’s t-test, *P ≤ 0.05, n = 7–8 mice per group. Four brain sections were analysed per mouse. (E) Cell nucleus size was measured from NeuN-stained sections, n = 10–12 mice per group. (F) Wet weight of inIGF1RKO and control brain. Mann-Whitney U-test, *P ≤ 0.05, n = 7 mice per group. Bar graphs represent mean ± SEM.
Deleting IGF1R from aged neurons entails histomorphological changes in the forebrain. (A) Inducible neuron-specific IGF1R knockout (inIGF1RKO) was achieved combining CaMKIIα-CreERT2 transgene with floxed Igf1r alleles (IGF1Rflox/flox). CaMKIIα-CreERT2;IGF1Rflox/flox mice were intraperitoneally injected with tamoxifen at age 17 month; controls received vehicle alone. (B) Gene deletion (IGF1RKO) monitored by PCR in hippocampus (n = 10–12 mice per group). (C) Western blot analysis of hippocampus. Mann-Whitney U-test, ***P < 0.001, n = 10 mice per group. (D) Representative micrographs of cresyl violet-stained cortical neurons. Neuronal soma size was reduced in inIGF1RKO cortex while neuron density remained unchanged. Two-tailed unpaired Student’s t-test, *P ≤ 0.05, n = 7–8 mice per group. Four brain sections were analysed per mouse. (E) Cell nucleus size was measured from NeuN-stained sections, n = 10–12 mice per group. (F) Wet weight of inIGF1RKO and control brain. Mann-Whitney U-test, *P ≤ 0.05, n = 7 mice per group. Bar graphs represent mean ± SEM.

Neuronal IGF1R knockout in old mice alleviates cognitive deficits induced by amyloid-β oligomers

AβOs are considered key mediators of synaptic and cognitive impairments in Alzheimer’s disease (Lesné et al., 2013; Ferreira et al., 2015). Intracerebroventricular injection of AβOs has been successfully used as proteotoxic insult causing neuronal dysfunction, synapse failure, cell metabolic and cognitive dysfunctions (Clarke et al., 2015). We generated AβOs by in vitro aggregation and checked quality by transmission electron microscopy. Amyloid-β incubation for 48 h at 37°C yielded a homogeneous product of 10–30 nm AβOs (Supplementary Fig. 1). This product was intracerebroventricularly injected into the brains of old inIGF1RKO and control mice in which the knockout had been induced several months earlier (Fig. 3A). To find out whether inactivation of IGF signalling protected from AβO-induced spatial memory deficits, we tested the mice in a Barnes maze. Time to reach the refuge box (P < 0.0001), number of errors (P < 0.01), and distance travelled (P < 0.001) were significantly reduced over days in all groups, indicating effective learning in the absence of IGF1R and/or after AβO challenge (Fig. 3B). However, during the memory-retention test, AβO-treated mice searched twice as long for the refuge box than control mice (Fig. 3C; P < 0.05), revealing AβO-induced long-term spatial memory deficits. Importantly, these deficits were no longer observed in inIGF1RKO + AβO mice. The better memory performance of inIGF1RKO + AβO mice compared with AβO mice was all the more evident since they preferentially used a spatial search strategy (Fig. 3D) (AβO, 33%; inIGF1RKO + AβO, 67%; Fisher’s exact test, P < 0.001). In addition, more AβO-treated mice failed the memory-retention test than control mice (failure: control 20%; AβO 48%; Fisher’s exact test, P < 0.001), and we observed a rescue of the proteotoxic phenotype in inIGF1RKO + AβO mice compared with AβO mice (Fig. 3D) (failure: inIGF1RKO + AβO 10%; Fisher’s exact test, P < 0.001). Besides memory impairments, Alzheimer’s disease also involves anxiety and sensory motor perturbations, which can be assessed by the open-field test. Results revealed no difference in speed and distance travelled between groups (data not shown), suggesting AβOs did not induce motor deficits. After a single intracerebroventricular injection of AβOs, mice avoided the anxiogenic centre (Fig. 3E; −34%, P < 0.05). In contrast, the behavioural pattern of inIGF1RKO mice remained unaffected by AβO injection, indicating that blocking neuronal IGF1R protected against AβO-induced anxiety. Collectively, behavioural data demonstrated that IGF1R suppression in old mice, prior to AβO proteotoxic challenge protected from cognitive and anxiety-like defects.


Figure 3



Suppression of neuronal IGF1R in aged mice prevents amyloid-β oligomers-induced cognitive deficits. (A) Control and inIGF1RKO mice, tamoxifen-induced at age 17 months, were injected intracerebroventricular with AβOs or vehicle at age 25 months. Behaviour was subsequently tested in Barnes maze and open-field. (B) Control and inIGF1RKO mice started training 72 h after AβO injection in a Barnes maze, four times per day for 4 days. In all groups, latency (P < 0.0001), errors (P = 0.01) and distance travelled (P < 0.001) diminished, indicating effective learning. Ability to learn remained unaffected by suppression of IGF1R and/or administration of AβOs (repeated measures ANOVA). (C and D) Reference memory was tested 24 h after the last training. Latency (C) doubled in AβO-injected controls and normalized in the absence of neuronal IGF1R (Mann-Whitney U-test, *P ≤ 0.05, **P < 0.01). Similarly, failure in AβO-injected controls to find the target hole was rescued in inIGF1RKO mice (D) (Fisher’s exact test, P < 0.001). n = 14–21 mice per group. (E) Open-field test on Day 11 post-intracerebroventricular injection revealed decreased anxiety-like behaviour of inIGF1RKO mice (Mann-Whitney U-test, *P ≤ 0.05, **P < 0.01), n = 14–20 mice per group. Data are mean ± SEM.

Suppression of neuronal IGF1R in aged mice prevents amyloid-β oligomers-induced cognitive deficits. (A) Control and inIGF1RKO mice, tamoxifen-induced at age 17 months, were injected intracerebroventricular with AβOs or vehicle at age 25 months. Behaviour was subsequently tested in Barnes maze and open-field. (B) Control and inIGF1RKO mice started training 72 h after AβO injection in a Barnes maze, four times per day for 4 days. In all groups, latency (P < 0.0001), errors (P = 0.01) and distance travelled (P < 0.001) diminished, indicating effective learning. Ability to learn remained unaffected by suppression of IGF1R and/or administration of AβOs (repeated measures ANOVA). (C and D) Reference memory was tested 24 h after the last training. Latency (C) doubled in AβO-injected controls and normalized in the absence of neuronal IGF1R (Mann-Whitney U-test, *P ≤ 0.05, **P < 0.01). Similarly, failure in AβO-injected controls to find the target hole was rescued in inIGF1RKO mice (D) (Fisher’s exact test, P < 0.001). n = 14–21 mice per group. (E) Open-field test on Day 11 post-intracerebroventricular injection revealed decreased anxiety-like behaviour of inIGF1RKO mice (Mann-Whitney U-test, *P ≤ 0.05, **P < 0.01), n = 14–20 mice per group. Data are mean ± SEM.

Suppression of neuronal IGF1R in aged mice prevents amyloid-β oligomers-induced cognitive deficits. (A) Control and inIGF1RKO mice, tamoxifen-induced at age 17 months, were injected intracerebroventricular with AβOs or vehicle at age 25 months. Behaviour was subsequently tested in Barnes maze and open-field. (B) Control and inIGF1RKO mice started training 72 h after AβO injection in a Barnes maze, four times per day for 4 days. In all groups, latency (P < 0.0001), errors (P = 0.01) and distance travelled (P < 0.001) diminished, indicating effective learning. Ability to learn remained unaffected by suppression of IGF1R and/or administration of AβOs (repeated measures ANOVA). (C and D) Reference memory was tested 24 h after the last training. Latency (C) doubled in AβO-injected controls and normalized in the absence of neuronal IGF1R (Mann-Whitney U-test, *P ≤ 0.05, **P < 0.01). Similarly, failure in AβO-injected controls to find the target hole was rescued in inIGF1RKO mice (D) (Fisher’s exact test, P < 0.001). n = 14–21 mice per group. (E) Open-field test on Day 11 post-intracerebroventricular injection revealed decreased anxiety-like behaviour of inIGF1RKO mice (Mann-Whitney U-test, *P ≤ 0.05, **P < 0.01), n = 14–20 mice per group. Data are mean ± SEM.

Blocking neuronal IGF signalling alleviates amyloid-β oligomer-associated inflammation

Intracerebroventricular-injected AβOs reliably induce brain inflammation (Choi and Bosetti, 2009; Clarke et al., 2015). We therefore asked whether the rescue in cognitive functions was accompanied by improved inflammatory status. In control mice, AβO administration resulted in a robust astrogliosis, with a >2-fold increase in density of cortical astrocytes (Fig. 4A; P < 0.01). In contrast, the AβO-induced increase in astrocyte density was limited to 19% in inIGF1RKO mice (P < 0.05). Thus, AβOs provoked significantly less astrogliosis in inIGF1RKO cortex than in controls (P < 0.01). AβOs also caused a 21% increase in density of Iba1-positive microglia in control mice, but failed to induce microgliosis in inIGF1RKO mice (Fig. 4B; P < 0.05). Depending on activation, morphology of microglia ranges from highly ramified to amoeboid-like. To explore microglia activation, we quantified branching. Total branch length per microglia was significantly reduced in AβO-injected controls (−24%, P < 0.05; Fig. 4C), which is typical of enhanced activation (Liaury et al., 2012; Andreasson et al., 2016), whereas microglia in inIGF1RKO + AβO mice was similar in phenotype to control and inIGF1RKO mice (Fig. 4C; P < 0.05). Thus, while microglia in AβO mice were strongly activated, they appeared less reactive in inIGF1RKO + AβO mice, suggesting that neuroinflammatory response was limited, most likely due to more efficient elimination of toxic peptides in inIGF1RKO + AβO brains. We also tried to evidence AβO-induced damage at the synaptic level by measuring drebrin and synapsin. However, we found no changes in synaptic proteins, possibly because 2 weeks after the proteotoxic challenge, and following multiple behavioural tasks, synapses had mostly recovered (Supplementary Fig. 2). Together, these results strongly indicated that neuronal IGF1R knockout induced in old mice protected against AβO-induced neuroinflammation.


Figure 4



inIGF1RKO brains are protected from AβO-induced reactive neuroinflammation. Immunohistochemistry was performed 13 days after intracerebroventricular injection. (A) Density of astrocytes in cortex was determined from GFAP immunostaining. (B) Anti-Iba1 IHC revealed that density of microglia was normalized in cortex of inIGF1RKO mice. (C) Total branch length per microglia was determined using AnalyzeSkeleton (ImageJ). Mann-Whitney U-test, *P ≤ 0.05, **P < 0.01, n = 4–6 mice per group; four brain sections were analysed per mouse. Data are mean ± SEM.

inIGF1RKO brains are protected from AβO-induced reactive neuroinflammation. Immunohistochemistry was performed 13 days after intracerebroventricular injection. (A) Density of astrocytes in cortex was determined from GFAP immunostaining. (B) Anti-Iba1 IHC revealed that density of microglia was normalized in cortex of inIGF1RKO mice. (C) Total branch length per microglia was determined using AnalyzeSkeleton (ImageJ). Mann-Whitney U-test, *P ≤ 0.05, **P < 0.01, n = 4–6 mice per group; four brain sections were analysed per mouse. Data are mean ± SEM.
inIGF1RKO brains are protected from AβO-induced reactive neuroinflammation. Immunohistochemistry was performed 13 days after intracerebroventricular injection. (A) Density of astrocytes in cortex was determined from GFAP immunostaining. (B) Anti-Iba1 IHC revealed that density of microglia was normalized in cortex of inIGF1RKO mice. (C) Total branch length per microglia was determined using AnalyzeSkeleton (ImageJ). Mann-Whitney U-test, *P ≤ 0.05, **P < 0.01, n = 4–6 mice per group; four brain sections were analysed per mouse. Data are mean ± SEM.

Neurons in early-stage Alzheimer’s disease and IGF1R knockout neurons exhibit convergent transcriptomic signatures

To explore the gene expression signature that protects IGF1R knockout neurons from amyloid-β proteotoxicity, we performed a genome-wide microarray analysis of microdissected CA1 from young control, inIGF1RKO, AD and ADINKO mice (Fig. 5A). We chose the hippocampal CA1 because of its compact population of pyramidal neurons (Supplementary Fig. 3), and its known vulnerability to degeneration (Ginsberg et al., 2012; Landel et al., 2014). Overall gene expression levels among the four groups were the same, as evidenced by principal component analysis and normalized intensities of all probes (Supplementary Fig. 4). The highest number of differentially expressed genes was observed when comparing inIGF1RKO mice with controls, revealing 623 up- and 513 downregulated genes. Comparing ADINKO with AD mice, we identified 178 up- and 107 downregulated genes (Fig. 5B). The proportion of molecular categories encoded by differentially expressed genes were conserved between experimental groups and include predominantly enzymes, transcription regulators, transporters, and G protein-coupled receptors (Supplementary Fig. 5). The top five altered gene networks for each pairwise comparison highlighted functional annotations mostly related to nervous system and neurodegenerative diseases, reflective of the homogenous population of CA1 pyramidal neurons (Supplementary Fig. 6). We next analysed effects of Alzheimer’s disease on neuronal integrated functions using PWS software, and found that Alzheimer’s disease-regulated genes involved in six main cell functions, namely neurotransmission, neuronal growth and differentiation, signal transduction, metabolism, stress response and proteostasis (Fig. 5C and Supplementary Table 1). Consistently, Alzheimer’s disease also affected these same functions in IGF1R knockout neurons (Supplementary Fig. 7A, Supplementary Tables 2 and 3). Interestingly, neuronal IGF1R deletion impacted essentially the same six main cell functions (Fig. 5D and Supplementary Table 4). Using IPA, we confirmed that early stage Alzheimer’s disease and neuronal IGF1R ablation affected strongly overlapping biological pathways (Supplementary Fig. 7B–F, Supplementary Tables 5 and 6). Within these integrated neuronal functions, Alzheimer’s disease and IGF1R deletion affected 29 biological processes, of which 25 changed in the same direction: these processes mainly involved cytoskeleton organization, neurite growth, differentiation, cellular stress response and synaptic transmission (Fig. 5E). Moreover, our analysis revealed that 102 of 103 changes in gene expression common to Alzheimer’s disease and IGF1R ablation occurred in the same direction, only the expression of neurofilament medium polypeptide Nefm changed in the opposite direction (Fig. 5F and G). Overall, these results clearly indicated that transcriptomic signatures of neurons from AD and inIGF1RKO mice exhibit strong and extended similarities. These findings suggest that early stage Alzheimer’s disease neurons activate a neuroprotective program resulting in cellular changes very similar to those occurring after specifically neuronal IGF1R ablation. Note that when we extracted all transcription factors with significantly altered gene expression from the microarray data (Supplementary Table 7), filtering them for known implication in Alzheimer’s disease pathogenesis and comparing AD with inIGF1RKO groups, we obtained several candidate transcriptional regulators, including Arntl, Ezh2, Hdac5 and Xbp1, which all behaved identically in both groups. It is likely that these transcription factors contributed to the similitude of AD and inIGF1RKO neuronal behaviour.


Figure 5



Transcriptional signatures of neurons from early-stage Alzheimer’s disease and of neurons with adult-onset IGF1R knockout reveal similar pattern of affected biological pathways and share many differentially expressed genes. (A) CaMKIIα-CreERT2;IGF1Rflox/flox and APPswe/PS1dE9;CaMKIIα-CreERT2;IGF1Rflox/flox mice were intraperitoneally injected with tamoxifen or with vehicle alone at 3 months of age. Hippocampus CA1 was laser-microdissected from brain sections of 6-month-old control, inIGF1RKO, AD and ADINKO mice. We used GeneChip Mouse Transcriptome Assay (MTA) for genome-wide microarray, and performed functional analyses by Pathway Studio (PWS) and Ingenuity Pathway Analysis (IPA) software. Scale bar = 600 μm. (B) Number of significantly (i.e. P < 0.02) up- (red) or downregulated (blue) genes in all four pairwise comparisons. (C and D) All significantly affected integrated functions were identified using PWS and ranked, with the most affected on top. Altered integrated functions in AD neurons (C) and inIGF1RKO neurons (D) compared with controls. (E) IPA revealed that 25 of 29 neuronal biological processes affected in both AD neurons and inIGF1RKO neurons were altered in the same direction. (F) Venn diagram representing the numbers of differentially expressed genes from AD mice versus control comparison and from inIGF1RKO mice versus control comparison, and their overlap. (G) Of 103 significant changes in gene expression common to AD and inIGF1RKO, 102 occurred in the same direction (blue). Solely the neurofilament medium polypeptide Nefm was significantly regulated in the opposite direction (red). One-way ANOVA, P < 0.02, n = 5 mice per group.

Transcriptional signatures of neurons from early-stage Alzheimer’s disease and of neurons with adult-onset IGF1R knockout reveal similar pattern of affected biological pathways and share many differentially expressed genes. (A) CaMKIIα-CreERT2;IGF1Rflox/flox and APPswe/PS1dE9;CaMKIIα-CreERT2;IGF1Rflox/flox mice were intraperitoneally injected with tamoxifen or with vehicle alone at 3 months of age. Hippocampus CA1 was laser-microdissected from brain sections of 6-month-old control, inIGF1RKO, AD and ADINKO mice. We used GeneChip Mouse Transcriptome Assay (MTA) for genome-wide microarray, and performed functional analyses by Pathway Studio (PWS) and Ingenuity Pathway Analysis (IPA) software. Scale bar = 600 μm. (B) Number of significantly (i.e. P < 0.02) up- (red) or downregulated (blue) genes in all four pairwise comparisons. (C and D) All significantly affected integrated functions were identified using PWS and ranked, with the most affected on top. Altered integrated functions in AD neurons (C) and inIGF1RKO neurons (D) compared with controls. (E) IPA revealed that 25 of 29 neuronal biological processes affected in both AD neurons and inIGF1RKO neurons were altered in the same direction. (F) Venn diagram representing the numbers of differentially expressed genes from AD mice versus control comparison and from inIGF1RKO mice versus control comparison, and their overlap. (G) Of 103 significant changes in gene expression common to AD and inIGF1RKO, 102 occurred in the same direction (blue). Solely the neurofilament medium polypeptide Nefm was significantly regulated in the opposite direction (red). One-way ANOVA, P < 0.02, n = 5 mice per group.
Transcriptional signatures of neurons from early-stage Alzheimer’s disease and of neurons with adult-onset IGF1R knockout reveal similar pattern of affected biological pathways and share many differentially expressed genes. (A) CaMKIIα-CreERT2;IGF1Rflox/flox and APPswe/PS1dE9;CaMKIIα-CreERT2;IGF1Rflox/flox mice were intraperitoneally injected with tamoxifen or with vehicle alone at 3 months of age. Hippocampus CA1 was laser-microdissected from brain sections of 6-month-old control, inIGF1RKO, AD and ADINKO mice. We used GeneChip Mouse Transcriptome Assay (MTA) for genome-wide microarray, and performed functional analyses by Pathway Studio (PWS) and Ingenuity Pathway Analysis (IPA) software. Scale bar = 600 μm. (B) Number of significantly (i.e. P < 0.02) up- (red) or downregulated (blue) genes in all four pairwise comparisons. (C and D) All significantly affected integrated functions were identified using PWS and ranked, with the most affected on top. Altered integrated functions in AD neurons (C) and inIGF1RKO neurons (D) compared with controls. (E) IPA revealed that 25 of 29 neuronal biological processes affected in both AD neurons and inIGF1RKO neurons were altered in the same direction. (F) Venn diagram representing the numbers of differentially expressed genes from AD mice versus control comparison and from inIGF1RKO mice versus control comparison, and their overlap. (G) Of 103 significant changes in gene expression common to AD and inIGF1RKO, 102 occurred in the same direction (blue). Solely the neurofilament medium polypeptide Nefm was significantly regulated in the opposite direction (red). One-way ANOVA, P < 0.02, n = 5 mice per group.

IGF1R suppression in adult Alzheimer’s disease neurons reverses specific cell functions

By comparing the transcriptome signatures of ADINKO with AD neurons using PWS software, we found that IGF1R deletion primarily regulated genes involved in signal transduction, stress response, neurotransmission, growth and differentiation (Fig. 6A and Supplementary Table 8). Interestingly, neuronal IGF1R inactivation affected markedly fewer biological functions in AD neurons than in control neurons (cf.Fig. 6A and Fig. 5D). Consistent with that result, we obtained similar results using IPA (Supplementary Fig. 7, Supplementary Tables 9 and 10). Changes in gene expression common to AD and to ADINKO neurons impacted essentially six major cell biological processes, of which five were regulated in the opposite direction (Fig. 6B). This was closely linked to the fact that 52 of 53 gene changes common to AD and ADINKO neurons were reversed (Fig. 6C and D). Genes with significantly altered expression included several that are known to be affected by Alzheimer’s disease, namely transthyretin (Ttr), 24-dehydrocholesterol reductase (Dhcr24), and the membrane G protein-coupled receptor kinase 5 (Grk5). Of note, Nefm was one of the three genes that were differentially regulated in all comparisons (Supplementary Fig. 8C and D). Consequently, 12.6% (52 of 414) of all genes deregulated in AD were significantly reversed in ADINKO mice. Taken together, these data clearly demonstrated that a conspicuous number of Alzheimer’s disease-associated transcriptional and functional changes are counteracted by preceding neuronal IGF1R deletion. Neurons knockout for IGF1R that subsequently develop Alzheimer’s disease show restrained expression of a significant number of genes that are all involved in the endogenous reaction of neurons to amyloid-β proteotoxicity. This appears as a coordinated prevention of gene overexpression entailing benefits in case of neurodegenerative disease. The limitation of inflammatory response, previously demonstrated in ADINKO mutants, may be a typical example for these beneficial effects. Finally, we used differentially expressed genes common to both comparisons to assemble a gene network depicting intramodular connections and hub genes (Fig. 6E). This molecular network clustered around key genes coding for Akt (Ak murine thymoma proto-oncogene protein family), APP (amyloid precursor protein), CREB1 (cAMP responsive element binding protein 1), HSPA8 (heat shock 70 kDa protein 8), and UBC (ubiquitin C). Significantly, biological functions represented by this network included amino acid and lipid metabolism, cell morphology, synaptic transmission, dementia and learning (Supplementary Table 11). We next selected key elements of this network and investigated how their differential expression translated into protein. For this, we used hippocampi at 6 and 17 months of age in which IGF1R had been deleted at 3 months. Similar to the findings in old mice (Fig. 2D), ablation of IGF1R at 3 months led to swift reduction of neuronal soma size in inIGF1RKO and in ADINKO mice (Fig. 7A). Through IPA and PWS analysis, we had identified cytoskeleton organization as an integrated function that was significantly impacted by Alzheimer’s disease and IGF1R deletion (Supplementary Tables 1 and 4–6). Here, we focused on NEFM. At 6 months, hippocampal NEFM was diminished in inIGF1RKO and ADINKO (Fig. 7B). At 17 months, when major Alzheimer’s disease symptoms were present, neuropathology markedly increased NEFM levels, while ablation of neuronal IGF1R reduced NEFM levels, in inIGF1RKO and in ADINKO brains. Thus, complete suppression of neuronal IGF signalling in AD mice restored hippocampal NEFM to control levels. Network analysis (Fig. 6E) suggested that Akt and CREB could play key roles in regulating interactions. Since Akt and CREB were not differentially expressed, we explored their hippocampal phospho-activation. We also examined phosphorylation of ERK1/2, a canonical pathway involved in IGF signalling, and S6 ribosomal protein (S6) that functions downstream of Akt. Total levels of Akt, ERK1/2, S6 and CREB were not changed among the four experimental groups. However, P-Akt (Ser473), P-ERK1/2 (Thr202/Tyr204) and P-S6 (Ser235/236) were increased in AD mice (two-way ANOVA, effect of AD: P < 0.001, Fig. 7C). This increase was 2- to 4-fold in P-Akt, P-S6 and P-ERK1/2. IGF1R deletion tended to reduce phospho-activation of Akt, S6 and ERK1/2 comparing ADINKO with AD mice, though this did not reach statistical significance (for P-S6, ADINKO versus AD: P = 0.09, Fig. 7C). By contrast, P-CREB (Ser133) increased in the absence of IGF1R, both in physiological and pathological context, while Alzheimer’s disease had no effect on P-CREB (Fig. 7C). Together, this suggested differential roles for Akt pathways and CREB in the interactions between IGF signalling and Alzheimer’s disease. Interestingly, at 17 months, the situation had again much evolved and we observed strong effects of Alzheimer’s disease, namely conspicuous decrease in P-Akt and P-S6 (two-way ANOVA, effect of AD: P < 0.0001, Fig. 7D). Taken together, gene network analysis clearly identified several key factors linking inactivation of IGF signalling to functions determining resistance to neurodegeneration.


Figure 6



Blocking IGF signalling in AD neurons reverses biological networks involved in lipid and amino acid metabolism, cellular stress response and proteostasis. (A) Integrated functions that are significantly impacted in ADINKO neurons compared with AD neurons, identified using PWS. (B) IPA revealed that five of six neurobiological processes affected in both AD and ADINKO neurons were reversed in the latter. (C) Venn diagram representing the number of differentially expressed genes affected in AD versus control comparison and in ADINKO versus AD comparison. (D) Of 53 gene expression-changes common to AD and ADINKO neurons, 52 were reversed in the latter (red). Additional Venn analyses are in Supplementary Fig. 8. (E) Transcriptome network integrating 43 of 53 differentially expressed genes identified in both AD versus control and ADINKO versus AD comparison. Overlay was from ADINKO versus AD comparison. Main clusters were genes coding for Akt, APP, CREB1, UBC and HSPA8. IPA, one-way ANOVA, P < 0.02, n = 5 mice per group.

Blocking IGF signalling in AD neurons reverses biological networks involved in lipid and amino acid metabolism, cellular stress response and proteostasis. (A) Integrated functions that are significantly impacted in ADINKO neurons compared with AD neurons, identified using PWS. (B) IPA revealed that five of six neurobiological processes affected in both AD and ADINKO neurons were reversed in the latter. (C) Venn diagram representing the number of differentially expressed genes affected in AD versus control comparison and in ADINKO versus AD comparison. (D) Of 53 gene expression-changes common to AD and ADINKO neurons, 52 were reversed in the latter (red). Additional Venn analyses are in Supplementary Fig. 8. (E) Transcriptome network integrating 43 of 53 differentially expressed genes identified in both AD versus control and ADINKO versus AD comparison. Overlay was from ADINKO versus AD comparison. Main clusters were genes coding for Akt, APP, CREB1, UBC and HSPA8. IPA, one-way ANOVA, P < 0.02, n = 5 mice per group.
Blocking IGF signalling in AD neurons reverses biological networks involved in lipid and amino acid metabolism, cellular stress response and proteostasis. (A) Integrated functions that are significantly impacted in ADINKO neurons compared with AD neurons, identified using PWS. (B) IPA revealed that five of six neurobiological processes affected in both AD and ADINKO neurons were reversed in the latter. (C) Venn diagram representing the number of differentially expressed genes affected in AD versus control comparison and in ADINKO versus AD comparison. (D) Of 53 gene expression-changes common to AD and ADINKO neurons, 52 were reversed in the latter (red). Additional Venn analyses are in Supplementary Fig. 8. (E) Transcriptome network integrating 43 of 53 differentially expressed genes identified in both AD versus control and ADINKO versus AD comparison. Overlay was from ADINKO versus AD comparison. Main clusters were genes coding for Akt, APP, CREB1, UBC and HSPA8. IPA, one-way ANOVA, P < 0.02, n = 5 mice per group.

Figure 7



Neuronal cell size, cytoskeleton organization and signal transduction affected by IGF1R knockout and Alzheimer’s disease. (A) Size of cell soma in cortical neurons from 6-month-old mice. Representative brain sections and morphometric quantification. Mann-Whitney U-test, *P ≤ 0.05, **P < 0.01, n = 5–6 mice per group. (B) Western blot of NEFM in hippocampus relative to NeuN, at 6 and 17 months of age. GAPDH served as loading control. (C) Western blot of P-Akt, P-S6, total P-ERK, P-CREB and IGF1R, together with representative scans from immunoblots, all at age 6 months. Note that the order of groups in scans differs between panels. (D) Western blot of P-Akt and P-S6 in 17-month-old hippocampus. Two-way ANOVA was used to test IGF1R knockout or Alzheimer’s disease effect, followed by Mann-Whitney U-test; *P ≤ 0.05, **P < 0.01, ***P < 0.001, n = 5–7 mice per group. Data are mean ± SEM.

Neuronal cell size, cytoskeleton organization and signal transduction affected by IGF1R knockout and Alzheimer’s disease. (A) Size of cell soma in cortical neurons from 6-month-old mice. Representative brain sections and morphometric quantification. Mann-Whitney U-test, *P ≤ 0.05, **P < 0.01, n = 5–6 mice per group. (B) Western blot of NEFM in hippocampus relative to NeuN, at 6 and 17 months of age. GAPDH served as loading control. (C) Western blot of P-Akt, P-S6, total P-ERK, P-CREB and IGF1R, together with representative scans from immunoblots, all at age 6 months. Note that the order of groups in scans differs between panels. (D) Western blot of P-Akt and P-S6 in 17-month-old hippocampus. Two-way ANOVA was used to test IGF1R knockout or Alzheimer’s disease effect, followed by Mann-Whitney U-test; *P ≤ 0.05, **P < 0.01, ***P < 0.001, n = 5–7 mice per group. Data are mean ± SEM.

Neuronal cell size, cytoskeleton organization and signal transduction affected by IGF1R knockout and Alzheimer’s disease. (A) Size of cell soma in cortical neurons from 6-month-old mice. Representative brain sections and morphometric quantification. Mann-Whitney U-test, *P ≤ 0.05, **P < 0.01, n = 5–6 mice per group. (B) Western blot of NEFM in hippocampus relative to NeuN, at 6 and 17 months of age. GAPDH served as loading control. (C) Western blot of P-Akt, P-S6, total P-ERK, P-CREB and IGF1R, together with representative scans from immunoblots, all at age 6 months. Note that the order of groups in scans differs between panels. (D) Western blot of P-Akt and P-S6 in 17-month-old hippocampus. Two-way ANOVA was used to test IGF1R knockout or Alzheimer’s disease effect, followed by Mann-Whitney U-test; *P ≤ 0.05, **P < 0.01, ***P < 0.001, n = 5–7 mice per group. Data are mean ± SEM.

Discussion

Neuronal IGF1R blockade promotes long-term neuroprotection against amyloid-β proteotoxicity

Our results demonstrate that IGF1R-def

Centre de Recherche
Saint-Antoine
UMR_S 938

Hôpital St-Antoine

Contact

Tél : +(33) 1 49 28 46 89