The Medial Septum Is Insulin Resistant in the AD Presymptomatic Phase: Rescue by Nerve Growth Factor-Driven IRS1 Activation

Basal forebrain cholinergic neurons (BFCN) are key modulators of learning and memory and are high energy-demanding neurons. Impaired neuronal metabolism and reduced insulin signaling, known as insulin resistance, has been reported in the early phase of Alzheimer’s disease (AD), which has been suggested to be “Type 3 Diabetes.” We hypothesized that BFCN may develop insulin resistance and their consequent failure represents one of the earliest event in AD. We found that a condition reminiscent of insulin resistance occurs in the medial septum of 3 months old 3×Tg-AD mice, reported to develop typical AD histopathology and cognitive deficits in adulthood. Further, we obtained insulin resistant BFCN by culturing them with high insulin concentrations. By means of these paradigms, we observed that nerve growth factor (NGF) reduces insulin resistance in vitro and in vivo. NGF activates the insulin receptor substrate 1 (IRS1) and rescues c-Fos expression and glucose metabolism. This effect involves binding of activated IRS1 to the NGF receptor TrkA, and is lost in presence of the specific IRS inhibitor NT157. Overall, our findings indicate that, in a well-established animal model of AD, the medial septum develops insulin resistance several months before it is detectable in the neocortex and hippocampus. Remarkably, NGF counteracts molecular alterations downstream of insulin-resistant receptor and its nasal administration restores insulin signaling in 3×Tg-AD mice by TrkA/IRS1 activation. The cross-talk between NGF and insulin pathways downstream the insulin receptor suggests novel potential therapeutic targets to slow cognitive decline in AD and diabetes-related brain insulin resistance. Electronic supplementary material The online version of this article (10.1007/s12035-018-1038-4) contains supplementary material, which is available to authorized users.


Introduction
The basal forebrain cholinergic system (BFCS) modulates many important behaviors, through robust inputs to the cortex and hippocampus [1,2] and reciprocal feedback projections [3]. In particular, the cholinergic neurons of the medial septum, corresponding to the Ch1 subregion of the BFCS, modulate attention, reference and spatial memories, as well as learning in mice and humans [1,4]. Early synaptic failure and metabolic derangement in basal forebrain cholinergic neurons (BFCN) have been reported to parallel memory deficits in mild cognitive impairment (MCI) and in Alzheimer's disease (AD) [5][6][7]. For this reason, it has been proposed as a good predictor of MCI progression toward AD [8]. BFCN are considered an energy-demanding neuronal population because of their dependency upon acetyl-CoA for biosynthesis of both ATP and acetylcholine. In line with this, their high metabolic profile has been suggested to be the main cause of peculiar BFCN vulnerability in neurodegenerative diseases [9].
Neuronal metabolism and plasticity, and higher cognitive functions are regulated by the insulin pathway in the central nervous system [10,11]. Insulin is both produced locally in the brain [12] and transported from the circulation, concentrating in the cortex and forebrain tissues. Insulin receptors (IR) and Insulin receptor substrates (IRS) are widely distributed in the mammalian forebrain [10,13,14]. In particular, IRS 1 is the most studied isoform involved in the control of glucose homeostasis and brain physiopathology [15]. The docking of IRS to the IR upon insulin binding initiates a survival cascade mediated by PI3K/AKT and CREB, leading to elevation of c-Fos expression, translocation of the glucose transporters (Glut) to the plasma membrane [16,17], and consequent glucose uptake, boosting the neuronal activity [18]. On the other hand, neuronal metabolic impairment caused by blunted or deficient insulin signaling, called insulin resistance, affects neuronal functions and has been suggested to parallel the onset and progression of AD pathology [19,20]. High IRS 1 phosphorylation at inhibitory serine residues is a general consensus marker of brain insulin resistance [20]. Aberrant IRS 1 phosphorylation at S312, S616 and/or S636 (equivalent to rodent S307, S612 and S632, respectively) mainly exerted by GSK3beta (feedback control) and by JNK1/2 ad PKC zeta/lambda (feed-forward inhibition) [21,22] is the result of a chronic maladaptive mechanism that attenuates insulin signaling. Elevated serine phosphorylated IRS 1 level has been reported in the brain of a primate AD model as well as in AD patients, and is associated with early synaptic dysfunctions and late stage neurodegeneration in AD [23].
Of interest for our study, insulin has been demonstrated to regulate the expression of choline acetyltransferase (ChAT), the key acetylcholine (Ach) biosynthetic enzyme, in SH-SY5Y neurons [24] and in the retina [25]. Indeed, also other insulin pathway activators, like IGFI, IGFII and PPAR agonists, stimulate ChAT levels and sustain survival and differentiation of cholinergic neurons [26,27]. Accordingly, perturbed insulin signaling is associated with swelling of cholinergic neurites and reduced ChAT immunoreactivity in the medial septum of 3×Tg-AD mice, a common model of AD [28,29]. Impaired acetylcholine homeostasis has been also described in streptozotocin (STZ)-treated rats, a rodent model of diabetesrelated insulin resistance [30]. Taken together, all these experimental observations strongly indicate a cause-effect relationship between insulin resistance and cholinergic damage. Moreover, the insulin control of the cholinergic phenotype implicates that insulin pathway perturbations may contribute to cholinergic transmission deficits in AD. Hence, to ascertain whether BFCN develop insulin resistance in the AD brain is essential to understand the early molecular events underlying the AD pathology. While central insulin resistance has been recently observed in the neocortex and in the hippocampal region in 3×Tg-AD and Tg2576 mice [31], its potential impact on the BFCS remains unexplored.
To investigate the insulin pathway in healthy and AD affected BFCN, we nasally administered insulin to wild-type and 3×Tg-AD mice, and analyzed the activation of the insulin signaling in the medial septum, a BFCN enriched region of the BFCS. Our study shows that the classic insulin pathway is elicited by nasal insulin administration in the medial septum of wild-type (wt) mice but not of age and sex-matched 3×Tg-AD mice, suggesting a condition reminiscent of insulin resistance. Noteworthy, we set up and fully characterized an in vitro model of insulin resistance developed in cholinergic neurons, to study the underlying molecular events and assess candidate drug, like nerve growth factor (NGF), for counteracting the insulin-resistant state of BFCN.
This cellular model may be of interest for the design of novel therapeutic strategies in AD and brain-related metabolic disorders. It enabled us to show that NGF not only elicits the insulin pathway and controls glucose metabolism by Glut2 translocation to the plasmamembrane in healthy BFCN, but it also reduces BFCN insulin resistance by re-activating the IRS 1 -driven insulin pathway in vitro and in AD mice.

Rodent Strains
Triple transgenic AD (3xTg-AD) mice, harboring human APP Swedish, presenilin M146V and tauP301L mutations [32], and C57 Bl6/J mice were purchased (Jackson Lab), and housed in the Animal Facility of the Università Cattolica Medical School. Purchased Wistar rats (HARLAN Laboratories Ltd., Füllinsdorf, Switzerland) were housed at the Animal Facility of the "Institute of Cellular Biology and Neurobiology", National Research Council (IBCN-CNR; Rome, Italy) at the European Center of Brain Research (CERC). Animals were handled in compliance with the National (D.Lgs26/2014) and European Union legislation guidelines for animal welfare (2010/63/EU). All efforts were made to minimize the number of animals used and suffering.

Nasal Insulin Administration
The procedure involves the immobilization of the mouse and the direct administration of small volumes of the selected substances (2.5 μl per nostril) using a Gilson type P10 precision pipette. Awake mice were gripped by the skin of their necks and held gently, but firmly, upside-down in the palm of the hand. The tip is placed near the nostril so that the animal can inhale the drug directly from the tip. The operator makes a slight pressure on the pipette piston in order to facilitate the expulsion of the drug. The procedure is extremely rapid (about 1 min) and painless, and it does not require any anesthesia [34]. Mice were intranasal administered with bovine insulin (0.125 IU) or vehicle (saline) and sacrificed after 20′ or 40′, to assess the optimal treatment duration in order to achieve the activation of the medial septum. The septum was surgically removed from the brain and collected for further analysis. The activation of the insulin pathway (IR-IRS-AKT) was analyzed by western blotting (Suppl. Fig. 1 a-d) and showed that while IR and IRS1 were phosphorylated already after 20′ and are still elevated after 40′ insulin treatment, AKT phosphorylation required a longer treatment (40′). Based on these results, the duration of the nasal treatment was set at an intermediate time point (30′) allowing the concomitant detection of all the molecules of interest. For the rescue experiment, vehicle, NGF (40 μg), or insulin (0.125 IU) were nasally administered to 3×Tg-AD mice and, after 30 min, mice were sacrificed. Brain tissues were removed and frozen until further biochemical analysis. Mice were starved for 14-16 h before intranasal stimulation.

Western Blotting
Tissues samples and cultured neurons were digested in a RIPA buffer with "complete protease and phosphatase inhibitory cocktail" (Roche) and centrifuged (10,000 rpm, 20′). The supernatants were collected, and the amount of total protein was determined by Quick Start Bradford Dye Reagent. Each sample (40 μg) was separated by SDS-PAGE in precast 4-12% Bis-Tris Plus gels (Bolt, Invitrogen), transferred to nitrocellulose membranes (0.45 μM, GE Healthcare), and incubated for 1 h at room temperature with 10% non-fat dry milk in TBS-T (10 mM Tris, pH 7.5, 100 mM NaCl, and 0.1% Tween-20). The overnight incubation with primary antibody (4°C) was followed by incubation with the appropriate HRP-conjugated secondary antibody (1:2000, Pierce, 1 h, RT) and the ECL substrate (32106; Pierce-Thermo Scientific). The films were digitized using a professional scanner (HP Scanjet 4050) and quantified by gel densitometry using the ImageJ software (NIH). Measurements were standardized between the experimental groups using the same calibration system and a fixed threshold over the background. Data are expressed as percentage optical density relative to control group, and presented as means ± SEM.

Immunofluorescence Labeling and Microscopy
Primary cultures were fixed for 20 min in PBS containing 4% paraformaldehyde, permeabilised with PBS plus 0.3% Tween, and quenched by ammonium chloride (50 mM, 30′, RT). Aspecific staining by the secondary antibody was blocked by incubation with normal donkey serum (10%, 1 h, RT). The overnight incubation (4°C) with primary antibodies was followed by the appropriate secondary antibodies (1:2000, 1 h, RT).

Nuclear c-Fos Staining and Analysis
c-Fos staining was performed by overnight incubation of control and insulin resistant cholinergic neurons with rabbit antic-Fos antibody (1:500), followed by secondary anti-rabbit Alexa Fluor-488 antibody (1:2000, 1 h, RT). Nuclei were counterstained by DAPI (1:1000; 15 min, RT). c-Fos immunofluorescence was acquired with an epifluorescent microscope (Leica CTR5500; Leica Microsystems) equipped with a CCD camera (Leica). Images for direct comparison were collected using the same settings. The number of c-Fos positive over the total number of DAPI stained nuclei per field (field area = 0.366 μm 2 ; ×20 objective) was measured in 5 different fields per coverslip, in at least three coverslips per experimental group. The analysis was performed after calibrating for particle pixels size (50-400 pixels) and applying a fixed threshold over the background. Nuclei were counted both by manual and automated counting methods (ImageJ software, NIH) with comparable results. Data are expressed as percentage to control group (CTR) and presented as mean ± sem.

Glucose Uptake Assay
Primary cholinergic neurons were seeded at 2-5x10 4 cells/ well on a glass coverslips. Neurons were starved (90′) by incubation with Neurobasal medium without B27 supplement (37°C, 5% CO 2 ). The glucose uptake measurement was performed by means of the "GluTracker Glucose Uptake cell-based Kit" following manufacturer's instructions (K681; Biovision Inc.). The bright green fluorescence generated by the fluorescent 2-deoxy-glucose conjugate (GluTracker) is proportional to the amount of glucose taken up by cells and can be used as a direct measure of the fluorescent glucose analogue (GluTracker) uptake. Briefly, the GluTracker mix was prepared as follow and added to each well: 376 μl of neurobasal medium, 4 μl GluTracker Reagent, and 20 μl of GluTracker Enhancer. Cholinergic neurons were incubated at 37°C with 5% CO 2 for 30 min. After incubation, the cells were washed once with ice-cold 1X Analysis Buffer (500 μl), and then replaced with fresh 1X Analysis Buffer (200 μl). Then, neurons were fixed for 20 min in PBS containing 4% paraformaldehyde, counterstained with DAPI (Life Technologies), and mounted on coverslips with Prolong Gold (Life Technologies). Images were acquired with an epifluorescent microscope (Leica CTR5500; Leica Microsystems) using a blue excitation fluorescence filter (excitation range: 420-495 nm) and a × 20 objective. The total immunofluorescence intensity per field (field area = 0.366 μm 2 ) was measured in 5 different fields per coverslips, in at least 4 coverslips per experimental group. Data reported in the graphs are expressed as percentage relative to control group and presented as mean ± sem.

In Situ Proximity Ligation Assay (PLA)
The interaction of TrkA with pIRS 1 Y608 and IRS 1 were detected by PLA assay, a sensitive technique allowing detection of direct binding (<40 nM distance) between endogenous proteins [37]. Specific primary antibodies (working dilution: 1:100) and species-specific secondary antibodies (PLA probes), each conjugated to a unique short DNA strand (MINUS and PLUS) were used. Briefly, once MINUS and PLUS PLA probes are in close proximity, the DNA strands interact and the subsequent addition of two other circle-forming DNA oligonucleotides takes place, leading to DNA amplification reaction, detected by Texas red fluorescence. As previously reported [38], DIV10 cholinergic neurons were fixed in 4% PFA for 15 min and subjected to in situ PLA according to the manufacture's instruction (Duolink In Situ Detection Reagents Red kit, Sigma Aldrich, DUO92008). To stain the nuclei dried coverslips were mounted with mounting medium (DUO82040). At the end of the procedure, every single event of PLA generated a fluorescent red spot. Fluorescence images were acquired on a TCS-SP5 confocal laser scanning microscope (Leica Microsystems GmbH Wetzlar, Germany) using 63 × 1.35 NA oil immersion objective. High resolution images were acquired as zstack with a 0.5 μm z-interval (at least 10 planes), and converted to maximum projection images (LASAF software platform; Leica Microsystems). High-resolution images were analyzed by ImageJ and the number and intensity of PLA dots per neuron were measured. A manual threshold was set up and applied to all images to eliminate the background fluorescence. Objects larger than 5 μm2 were rejected, thereby effectively removing nuclei. The remaining objects were counted as PLA puncta. At least 20 neurons in five non-overlapping fields, and from 3 independent experiments, were randomly chosen by a blind observer. As a control for primary antibodies specificity, double immunolabelings with anti-TrkA and anti-pIRS 1 Y608 or anti-TrkA and anti-IRS 1 antibodies (working dilution: 1:100) were run in parallel to each set of PLA experiments (Suppl. Fig. 3a and Suppl. Fig. 3b, respectively). PLA dots were undetectable following omission of the primary antibodies (Suppl. Fig. 4).

Statistical Analysis
Tissues from at least n = 3 animals per experimental group were analyzed. All experiments using primary neurons were performed at least three times independently, each in triplicate. The graphs were generated using PRISM (GraphPad Software, Inc., San Diego, CA, USA), and the data are presented as the mean ± sem. ANOVA followed by Student's T test or Tukey-Kramer post hoc was used to analyze the data, depending on the number of variables and groups (Statview-SAS, Cary, NC, USA). A p value ≤ 0.05 was considered statistically significant.

Data Availability
The authors declare that all the data supporting the findings of this study are available within the article, and from the corresponding author upon reasonable request.

Insulin Resistance Occurs in the Medial Septum of 3 Months Old 3×Tg-AD Mice
Brain insulin resistance is known to affect cognition and occurs at early stage in forebrain of different AD mouse models [19,40]. Within the basal forebrain, the medial septum is well-known for being a primary target of AD neuropathology [4,7,8,29]. However, insulin responsivity in the presymptomatic septum from AD mice has not been investigated so far.
Cholinergic Neurons Express IR and IRS 1 , Are Insulin Sensitive, and Respond to NGF by Activating the Insulin Pathway Medial septum contains different type of neurons, including cholinergic, GABAergic and glutamatergic neurons [4]. Thus, to specifically investigate insulin responsivity in the cholinergic component of the medial septum, we resorted to a primary culture of cholinergic neurons.
First, we studied the expression of IR and IRS 1 in vitro, in rodent primary cholinergic neurons (E17, DIV10) by triple immunofluorescence labeling, using specific antibodies against the IR-β subunit, the IRS 1 and ChAT, a marker for cholinergic neurons. We found that ChAT-positive neurons Fig. 1 Basal forebrain system responsivity to insulin in wild-type and AD mice (a-j) Lysates of medial septum from nasal insulin and vehicletreated 3 months old 3xTg-AD (3xTg + INS and 3xTg + veh, respectively; 30′) and age-matched wild-type (wt + INS and wt + veh, respectively; 30′) mice were analyzed by WB for the level of phosphorylated and total IR (f, g and a, b; respectively), IRS 1 Y608 (f, h and a, c; respectively) IRS 1 S307 (f, i and a, d; respectively) and AKT (f, j and a, e; respectively). The level of pIR Y1150/1151 , pIRS 1 Y608 and pAKT in insulin treated mice were measured by WB using specific antibodies and β-actin, as loading control. The respective ratios over total IR, IRS 1 , and AKT levels were calculated and expressed as percentage of the vehicle treated mice. (a-e) Representative WB (a) and bar graphs (b-e) showing the rapid activation of the canonical insulin pathway in the medial septum of wt mice upon insulin administration, with increased phosphorylation of IR Y1150/1151 (182.2 ± 15.6% of wt + veh; **p < 0.01; Fig. 1b), IRS 1 Y608 (166.9 ± 59.4% of wt + veh; *p < 0.05; Fig. 1c) and AKT (172.6 ± 6.2% of wt + veh; **p < 0.01; Fig. 1e). (f-j) Representative WB (f) and bar graphs (g-j) indicate a blunted insulin response in the medial septum of 3xTg AD mice as a response to nasal stimulation. In fact, level of pIR Y1150/1151 (86.4 ± 8.8% of 3xTg + veh, p = 0.194; Fig. 1g), pIRS 1 Y608 (86.9 ± 9.0% of 3xTg + veh; p = 0.283; Fig. 1h) and pAKT (100.4 ± 6.1% of 3xTg + veh, p = 0.954; Fig. 1j) were unchanged following insulin treatment, a molecular state reminiscent of insulin resistance. On the other hand, insulin administration was able to downregulate the level of inhibitory pIRS 1 S307 in wt mice (83.9 ± 4.9% of wt + veh, *p < 0.05; Fig. 1a, d), but failed to reduce it in AD mice (109.9 ± 10.6% of 3xTg + veh, p = 0.402; Fig. 1f, i) express both IR and IRS 1 (Fig. 2a, a 1 ) in neuronal culture. To assess their responsivity to insulin, cholinergic neurons (E17; DIV10) were incubated with insulin (10 nM, 30′) and the activation of IR, IRS 1 , and AKT were analyzed by WB (Fig. 2b-e). Upon insulin treatment (INS) we observed increased levels of pIR Y1150/1151 (531.1 ± 112.4% of CTR; **p < 0.01; Fig. 2c), pIRS 1 Y608 (296.7 ± 35.5% of CTR; **p < 0.01; Fig. 2d), and pAKT (344.1 ± 38.8% of CTR; **p < 0.01; Fig. 2e) compared to unstimulated neurons (CTR). Cholinergic neurons are dependent upon NGF supply for their specification and postnatal development [42,43]. Of interest, NGF has been demonstrated to stimulate the insulin signaling in PC12-differentiated neurons [44] as well as in primary sympathetic and sensory neurons [45]. For this reason, we investigated whether NGF was able to activate the insulin pathway in cholinergic neurons. We found that the levels of pIR Y1150/1151 (274.7 ± 56.5% of CTR; *p < 0.05; Then, to further characterize insulin effects on cholinergic neurons, we assayed neuronal metabolism by mean of 2dglucose incorporation and nuclear expression of c-Fos, an Intermediate Early Gene transiently activated in response to neurotrophic stimuli in cholinergic neurons [46,47]. Both insulin and NGF have been reported to induce c-Fos protein expression in neurons [48,49]. As expected ( Fig. 2f-g), we observed that stimulation of cholinergic neurons with both insulin and NGF increased the number of c-Fos positive nuclei per field (INS, 222.7 ± 20.1% of CTR, **p < 0.01; and NGF, 383.3 ± 74.1% of CTR; **p < 0.01, respectively). Insulin has been previously shown to promote glucose uptake in different type of CNS cells [50,51]. We assayed glucose uptake by incubation of neurons with a bright green fluorescent 2-deoxy-glucose conjugate (GluTracker), using a cell-based commercial kit (BioVision) allowing a non-radioactive analysis of glucose metabolism by fluorescence microscopy. Both insulin and NGF significantly enhanced glucose uptake in cholinergic neurons (INS, 280.3 ± 6.7% of CTR, **p < 0.01; and NGF, 183.2 ± 33.1% of CTR, *p < 0.05; respectively; Fig. 2h Glucose is taken up by glucose transporters (Glut) following stimulus-dependent translocation of Glut from the cytosol to the plasma membrane. In particular, Glut2 is a glucose sensor and is widely expressed in the brain, including the medial septum [52]. Glut4 is the insulin-responsive Glut isoform in the peripheral organs and is present in several brain tissues [53]. However, their role in glucose uptake in cholinergic neurons has not been established. In order to assess whether Glut2 and/or Glut4 are implicated in glucose uptake in cholinergic neurons, we immunolabeled them with specific anti-Glut2 and anti-Glut4 antibodies. Glut2 staining was found to be dispersed in the cytosol in starved neurons (CTR), while treatment with NGF induced a strong Glut2 re-localization at the plasma membrane (Fig. 2j) at both cell bodies (arrows) and neurites (arrowheads). By contrast, neither insulin nor NGF were able to induce translocation of Glut-4 to the plasma membrane, despite this glucose transporter is expressed in primary cholinergic neurons (Suppl. Fig. 1e).
Insulin Resistance Can Be Induced in Primary Cholinergic Neurons by Chronic High Insulin and It Is Ameliorated by NGF Brain insulin resistance is defined as a state of reduced brain responsivity to insulin stimulation. It is characterized by decreased tyrosine phosphorylation of IR and IRS 1 and the concomitant increase in serine phosphorylation of IRS 1 , especially at S307 [20]. Chronic high insulin has been already reported to induce a state mimicking insulin resistance in cortical neurons in vitro [54]. Thus, in order to study insulin resistance in primary cholinergic neurons, the latter were treated with 2μM insulin for 72 h (from DIV10 to DIV13). Acute stimulation with 10 nM insulin (30′) was performed to assess insulin responsivity following chronic high insulin exposure. The experimental scheme of the chronic high insulin protocol is reported in Fig. 3m.

Stimulation of the Insulin Pathway and Neuronal Metabolism by NGF Is Achieved Through IRS 1 Activation and Binding to TrkA
We demonstrated the ability of NGF to improve insulin resistance, mainly stimulating IRS 1 in primary cholinergic neurons. This is in line with the findings that several members of the receptor tyrosine kinases (RTK) superfamily like IGFI/II receptors, the BDNF receptor TrkB, and ErbB can trans-activate the insulin pathway by enhancing IRS 1 and the downstream insulin pathway (e.g., AKT, GSK3β) [56]. Accordingly, we asked whether the rescue effect of NGF on survival and metabolism in insulin resistant cholinergic neurons depends on IRS 1 . To address this question, we incubated cholinergic neurons with NGF with and without previous treatment with the selective IRS inhibitor NT157. First, NT157 (10 μM, 2 h) was found to effectively inhibits IRS 1 (Suppl. Fig. 2a) and AKT (Suppl. Fig. 2b) phosphorylation under insulin stimulation. Then, we analyzed AKT activation, c-Fos expression and glucose uptake in control (Fig. 4a, c, e) and insulin resistant (Fig.  4b, d, f) cholinergic neurons. We found that IRS inhibition by NT157 abolished the NGF effect on pAKT levels in control (NGF + NT157, 68.6 ± 9.3% of DIV10 CTR; NGF + NT157 vs NGF, **p < 0.01, Fig. 4a and Suppl. Fig. 2g) and in insulin resistant (RN + NT157, 93.1 ± 15.1% of DIV13 CTR; RN + NT157 vs RN; **p < 0.01, Fig. 4b and Suppl. Fig. 2h) neurons. A similar trend was observed for c-Fos expression, indicating that NGF effect was lost both in control (NGF + NT157, 96.1 ± 35.5% of DIV10 CTR; NGF + NT157 vs NGF, **p < 0.01; Fig. 4c) and insulin resistant conditions (RN + NT157, 96.0 ± 35.7% of DIV13 CTR; RN + NT157 vs RN, **p < 0.01; Fig. 4d). Further, following incubation with the specific IRS inhibitor NT157, NGF was unable to rescue glucose uptake both in control (NGF + NT157, 35.3 ± 7.5% of DIV10 CTR; NGF + NT157 vs NGF, **p < 0.01; Fig. 4e) and insulin resistant (RN + NT157, 11.3 ± 6.5% of DIV13 CTR; RN + NT157 vs RN, **p < 0.01; Fig. 4f) cholinergic neurons. Of note, NT157 treatment not only abolished NGF effect on glucose uptake in insulin resistance, but it also further reduced glucose metabolism, as compared to control (NGF + NT157 vs DIV10 CTR, **p < 0.01; Fig.  4e) and insulin resistant conditions (RN + NT157 vs RI; *p < 0.05; Fig. 4f), suggesting that most of the glucose uptake likely occurs under the IRS control in cholinergic neurons. Of note, NT157 incubation did not affect neuronal viability, as assessed by the MTT test (p = 0.16; Suppl. . We found that NT157 abolished the observed NGF effect on Glut2 translocation to the plasma membrane (Figs. 2j and 4g), while no effect was observed for Glut4 (Suppl. Fig. 2i). These data demonstrate that the NGF effect on neuronal metabolism in control and insulin resistant cholinergic neurons is abolished by the IRS inhibitor NT157, and it thus requires IRS activation.
NGF stimulation of insulin pathway via IRS 1 transactivation has been previously reported in cancer cells, and occurs upon IRS 1 recruitment by the NGF receptor TrkA [19,28]. For this reason, we asked whether TrkA is able to bind and recruit IRS 1 also in our experimental conditions. To investigate the existence of a TrkA/IRS 1 complex in cholinergic neurons we resorted to the Proximity Ligation Assay (PLA) approach. PLA is a welldescribed technique providing enough sensitivity to evaluate endogenous protein's close proximity in native conditions. PLA allows visualization of single events of protein interaction by generating single immunofluorescent dots.

Discussion
To the best of our knowledge, this is the first study addressing medial septum sensitivity to insulin stimulation in wt and AD mice. Here we present in vivo evidence of medial septum sensitivity to insulin upon nasal administration, with the activation of the canonical insulin pathway in wild-type mice (Fig. 1a-e).
Although the brain has been long time considered an insulin independent organ, compelling evidences show that insulin not only is locally produced, but it also induces glucose uptake, has neuroregulatory properties on nutrient intake in the brain, modulating learning and memory, and controlling body energy intake [12,41,57]. Accordingly, IR and IRS mRNA and protein are co-expressed in several brain tissues, like the hippocampus, the cortex, the hypothalamus, including the basal forebrain [13,58,59].
Conversely, insulin treatment was unable to elicit the activation of the key insulin signaling molecules (IR-IRS 1 -AKT) in the basal forebrain of 3×Tg-AD mice (Fig. 1f-j). Nasal administration is a well characterized and extensively used method of brain delivery [60], and nasal insulin stimulation has been shown to be the paradigm of choice to highlight brain insulin resistance [20]. Thus, insulin failure to activate the insulin pathway in the medial septum of 3×Tg-AD mice suggests that these mice developed a condition reminiscent of brain insulin resistance. Moreover, insulin resistance has been observed at 3 months of age, while synaptic deficits and cognitive impairments are not detectable before 5-6 months of age in 3×Tg-AD mice [32,61,62]. Neocortical insulin resistance and systemic deregulation of glucose metabolism can be detected only 2-3 months later, before the appearance of blood hyperglicemia and peripheral insulin resistance [31,62], pinpointing that insulin resistance in the medial septum develops during the AD presymptomatic phase of 3×Tg-AD mice [63,64].
The perturbation of the insulin pathway has been shown to affect cholinergic phenotype and synaptic functions in cultured neurons [24]. Thus, the effects of insulin and insulin resistance on the BFCS and related cognition is of great relevance for ameliorating insulin resistance-dependent impairment of cognition observed in both neurodegenerative disorders (e.g. AD) and metabolic diseases (e.g. Type 2 Diabetes, T2D).
In order to investigate a crosstalk between the insulin signaling and neuronal metabolism in BFCN, we cultured neurons from the rodent septum following a well established and characterized in vitro model culture [35,36]. We observed that cultured cholinergic neurons express both IR and IRS (Fig.  2a), as already observed in vivo in rodents [14]. Moreover, they activate the canonical insulin pathway in response to insulin stimulation by increasing levels of pIR Y1151/1150 (Fig.  2b-c; 500% of CTR), pIRS 1 Y608 (Fig. 2b, d; 300% of CTR), and pAKT (Fig. 2b, e; 350% of CTR). In line with this, glucose uptake was significantly augmented (Fig. 2h-i; 275% of CTR) and c-Fos expression increased in response to insulin stimulation ( Fig. 2f-g; 200% of CTR). These findings show that BFCN are insulin sensitive and their neuronal metabolism and activity are under insulin control.
Next, we set up a novel chronic high insulin-based cellular paradigm to specifically address insulin resistance in cholinergic neurons and found that the IR-IRS-AKT signaling pathway is significantly downregulated in cholinergic neurons (Fig. 3). In turn, the activity of the main IRS 1 serine kinases, GSK3β and JNK1/2, are upregulated ( Fig. 3f-h), while c-Fos expression and glucose uptake are significantly reduced (Fig. 3i-l). These findings indicate that this in vitro paradigm does recapitulate the main features of insulin resistance at the cellular level, and may represent a valuable tool for the screening of insulin sensitizers and potential disease-modifying drugs in cholinergic neurons, of relevance for therapy of AD and insulin-related brain diseases.
There is general consensus on the cross-talk between insulin and Insulin like Growth Factor 1 (IGF1) pathways, resulting in common IRS-mediated PI3K/Akt and Ras/Raf/ MAPK signaling initiated by hybrid receptors. Of note, IGF1 has been shown to promote the cholinergic phenotype of septal neurons and deregulation of IGF1 pathway has been implicated in neurodegenerative diseases [19,20,27]. Interestingly, our data indicate that NGF may counteract the detrimental effects of brain insulin resistance by acting on the IRS 1 stimulation, the early, main and common effector of insulin and IGF1 receptors signalings. Thus, the involvement of IGF1 deserves future investigations and it is reasonable to hypothesize the co-occurrence of insulin and IGF1 resistances in our chronic high insulin model and a beneficial effect of NGF on IGF1 resistance in AD.
Furthermore, we deeper characterized the glucose metabolism addressing the involvement of the glucose transport protein (Glut) in these events. In particular, we focused on Glut2, a glucose sensor and Glut4, the prototypical insulin dependent Glut in peripheral organs [53]. While Glut2 is insulinindependent in the peripheral organs, its role in the CNS is not well understood. Indeed, Glut2 mRNA and protein have been reported within specific brain nuclei included the rodent and human basal forebrain [52,65,66], and have been implicated in neurotrophic control of glucose homeostasis in the rodent CNS [52,67], as well as in zebrafish brain development [49]. Here, we observed that Glut2 massively translocates to the plasma membrane in the cell body, axon and dendrites in cholinergic neurons upon NGF treatment, and to a minor extent, following insulin stimulation (Figs. 2j and 4g).
As opposite, neuronal Glut4 localization is unaffected by insulin and NGF treatments (Suppl. Fig. 1e), neither by the IRS inhibitor NT157 (Suppl. Fig. 2i). These findings indicate that NGF-and insulin-mediated glucose uptake is mainly under Glut2 control in cultured cholinergic neurons (see also Fig. 2j), and suggest a minor role for Glut4 in our experimental conditions. In line with our observations, insulin failure in stimulating Glut4-dependent glucose uptake has been also reported in the human brain [20], supporting the idea that glucose metabolism regulation by Gluts is tissue and cell type specific.
Further, we hypothesized that NGF induction of IRS 1 relies on a direct interaction between IRS 1 and the neurotrophin receptor TrkA. In fact, the NGF receptor TrkA and IRS 1 can directly interact, as shown with the two hybrid system in yeast and induce downstream genes, like c-Fos [68]. In order to investigate whether TrkA and IRS 1 endogenous molecules directly interact, we resorted to the PLA and found that NGF treatment does stimulate the binding of its receptor TrkA to IRS 1 (Fig. 4k-m). Interaction between TrkA and IRS 1 is even more pronounced after tyrosine phosphorylation of IRS 1 (Fig.  4h-j), which is promoted by NGF (Fig. 2b, d). On the other hand, the IRS inhibitor NT157 completely abolished NGF effect on TrkA-IRS 1 complex formation, suggesting the requirement of an active IRS 1 for the TrkA-IRS 1 interaction occurring under NGF stimulation.
Finally, we investigated the in vivo effect of NGF on the cholinergic system by nasal delivery to 3xTg-AD mice. Indeed, our findings indicate that NGF improves insulin resistance in the septum and induces the insulin pathway through tyrosine phosphorylation of IRS 1 , leading to the AKT-mediated pro-survival signaling. Noteworthy, NGF sustained the level of ChAT, a cholinergic marker affected by insulin resistance, possibly by ChAT stabilization, exerting further neuroprotective action for BFCN in AD-like neurodegeneration.

Conclusion
Overall, these data indicate that NGF may elicit IRS 1 stimulation in cholinergic neurons by-passing IR activation through IRS binding to the NGF receptor TrkA, as already demonstrated with other insulin sensitizers [56]. NGFdriven re-activation of the insulin pathway through disinhibition and tyrosine phosphorylation of IRS 1 , a primary gatekeeper in insulin signaling, is a potential novel strategy to slow cognitive decline in AD and diabetes-related brain insulin resistance. Ongoing collaborative studies are aimed at testing the in vivo potential of intranasal NGF administration on neuronal insulin resistance in AD and T2D mouse models. Given the growing demand for safe and effective disease-modifying drugs in AD, the design of novel NGF-based approach during the AD prodromal phase is of foremost clinical importance for future AD therapeutic strategies.