BIN1 deficiency leads to DNA damage and neuronal insulin resistance through ATM dysregulation

The major neuronal isoform of the AD risk gene BIN1 is specifically reduced in patients. In the present work, we demonstrate that BIN1 is necessary for intact insulin signaling and its loss leads to cellular-level insulin resistance. Persistently activated mTORC1 feedback inhibits the insulin signaling; meanwhile, ATM-dependent DNA Damage Response decreases the expression of multiple insulin signaling pathway members. Unexpectedly, nuclear ATM activation is accompanied by autophagic degradation of the cytoplasmic ATM, rendering neurons susceptible to oxidative stress, which predisposes oxidative DNA damage, DNA strand breaks, and ATM-dependent D NA Damage Response. Treating BIN1-deficient mice with rapamycin or liraglutide to improve the insulin response, or with the natural antioxidant alpha-lipoic acid to attenuate oxidative stress, both effectively preserved the spatial cognitive ability. Lastly, we reanalyzed a single-cell transcriptomic dataset from human neurons and identified positive correlations between levels of BIN1 and the insulin signaling, as well as neuronal activity. Together, this work revealed an ATM-centered mechanism that globally damages neuronal health in BIN1-deficient pyramidal neurons, which could be a driving force in the AD disease course.

Alzheimer’s disease; BIN1; ATM; DNA damage response; insulin resistance. Abbreviations: AAV: adeno- associated virus; Aβ: amyloid -β; ACTB: β actin; AD: Alzheimer’s disease; AKT1: thymoma viral proto-oncogene 1; BIN1: bridging integrator 1; BIN1-iso1: BIN1, isoform 1; CA1: cornu Ammonis 1; DMEM: Dulbecco's modified Eagle medium; GWAS: genome-wide association study; MTOR: mechanistic target of rapamycin kinase; mTORC1: MTOR complex 1; RPS6KB1: ribosomal protein S6 kinase B1; qRT-PCR: real -time quantitative reverse transcription PCR; ROS: reactive oxygen species. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint

BIN1 is a multifunctional protein in both peripheral tissues and the central nervous system (CNS). For late- onset Alzheimer’s disease (AD), genome -wide association studies (GWAS) identified Bridging Integrator 1 ( BIN1) as the second most significant risk gene. 1, 2 In the CNS, BIN1 is broadly expressed in neurons, oligodendrocytes, and microglia. The major neuronal isoform of BIN1, BIN1 isoform 1 (BIN1-iso1), is significantly reduced in sporadic AD patients, 3-5, additionally supporting its participation in the disease course. The downstream consequence of this molecular event is unclear. Clinical studies revealed a close relationship between BIN1 and the microtubule- associated protein TAU. 6-8 BIN1 directly interacts with TAU, 9-11 and regulates intercellular propagation of TAU pathology;12 overexpression of BIN1 induces neuronal network hyperexcitability in a TAU-dependent manner. 13 Meanwhile, BIN1 also modulates neurodegeneration independently of TAU. Researchers found that loss of BIN1 impairs glutamate release at the presynapse and disturbs the membrane targeting of the AMPA receptor at the postsynapse. 14, 15 Employing the Drosophila larvae model, we previously reported that the ortholog of human BIN1, Amphiphysin, regulates dendritic branch dynamics and is required for synapse maturation. 16 Besides these synaptic functions, BIN1 genotypes were associated with the hippocampal volume in young healthy Chinese university students.17 In agreement with this clinical finding, we recently reported that BIN1 deficiency leads to decreased dendritic arbor size in cultured hippocampal neurons with regional volume loss that is visible in magnetic resonance imaging examination of mouse hippocampal CA1; we also proved that elevation of ULK3-initiated, the mammalian target of rapamycin complex 1 (mTORC1)-insensitive autophagic flux is beneath this dendritic atrophy.18 In BIN1-deficient neurons, decreased AKT activity facilitates the nuclear translocation of TFEB, which transcriptionally expresses autophagic and lysosomal genes. 18 As IRS1 – PI3K–AKT signaling is a main pathway downstream of insulin binding to its receptor, the reduction of AKT activity suggests the existence of neuronal insulin resistance, which is a common pathology in multiple neurodegenerative disorders, including AD. 19 Brain insulin resistance typically takes place without systemic insulin resistance in AD patients, and thus, some researchers proposed AD as a type 3 diabetes. 20 Neuronal insulin resistance is firmly related to cognitive decline through multiple mechanisms. 21 In the context of AD, Aβ oligomers are known to produce insulin resistance through competing with the insulin receptor, and activated glia-derived pro-inflammatory cytokines such as tumor necrosis factor-α also inhibit IRS1 function thro ugh the JNK signaling.22 As GSK- 3β is negatively regulated by AKT, tau hyperphosphorylation is generally regarded as a downstream event of neuronal insulin resistance. On the risk gene side, ApoE4 promotes the development of insulin resistance through increasing Aβ burden; 23 whether other risk genes are also linked to neuronal insulin resistance is insufficiently studied. Unlike AD, patients with another neurodegenerative disorder, ataxia telangiectasia (AT), are prone to develop systemic insulin resistance. 24, 25 AT is caused by loss-of-function mutations of the gene Ataxia telangiectasia mutated (ATM). As a phosphatidylinositol 3- kinase-like kinase (PIKK), ATM regulates diverse metabolic signaling pathways, and .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint ATM deficiency was firmly associated with impaired insulin signaling. 26, 27 In the nucleus, when DNA double-strand DNA breaks (DSBs) are detected, the Mre11-Rad50 - Nbs1 (MRN) complex recruits ATM and leads to its auto-activation; ATM then phosphorylates the histone H2AX to generate a docking platform (γH2AX foci) for the subsequent repair machinery, which is a critical step in DNA damage response (DDR). DDR contains extensive transcriptional and post-transcriptional regulations and affects a number of cellular processes. In a post-mortem study on the frozen frontal cortex of AD patients, high DDR levels were associated with downregulation of multiple important insulin signaling pathway (ISP) members, 28 supporting a correlation between these two cellular states, but whether DDR is a driver of cellular-level insulin resistance is unclear. In human fibroblasts and cancer cell lines, loss of BIN1 was associated with activation of ATM in an E2F1-dependent manner. 29 In the present work, we report that ATM activation and ATM-mediated DDR also occur in BIN1-deficient neurons. Interestingly, we found that this DDR is accompanied by a reduction of cytoplasmic ATM, and both of them impair insulin signaling th rough distinct routes. What’s more, cytoplasmic ATM reduction renders neurons hypersensitive to reactive oxygen species (ROS) stress, which elicits DNA strand breaks in the nucleus and subsequently nuclear ATM-mediated DDR. In mice with hippocampal neurona l Bin1 specifically reduced, we evidenced declined spatial cognitive performance, which was significantly attenuated by anti-insulin resistance therapy with the mTORC1 inhibitor rapamycin or the glucagon-like peptide- 1 receptor agonist liraglutide, or with a natural antioxidant alpha-lipoic acid targeting the upstream ROS-induced DNA damage. Based on these results, we propose that downstream of neuronal BIN1 deficiency, ATM-centered mechanisms produce intertwined insulin resistance, ROS hypersensitivity, and DNA damage.

Persistent mTORC1 activation inhibits insulin signaling in BIN1-deficient neurons. In our recent study, we reported that BIN1 deficiency leads to reduced AKT activity and concurrent activation of mTORC1, 18 implying feedback inhibition of the IRS1 – PI3K– AKT signaling by mTORC1 itself and its downstream kinase RPS6KB1, which is a common route to cellular-level insulin resistance. To make clear whether this occurs, we dissected the cortex from neonatal mice and cultured the pyramidal neurons. On day in vitro (DIV) 5, we transduced neurons with a lentivirus expressing a shRNA targeting all isoforms of mouse Bin1 under the RNU6 promoter; On DIV11, cells were collected for western b lot (WB) analysis; in plko-shBin1 group, we observed a significant reduction in the Ser473 p hosphorylation of AKT; for cells receiving insulin treatment before harvest (100 nM for 10 minutes), p -AKT-Ser473 levels increased In the control group, but displayed no significant change in plko- shBin1 group ( Figure 1A ). Meanwhile , Bin1 knockdown substantial ly decreased the protein levels of insulin receptor substrate 1 (IRS1) ( Figure 1A ), additionally indicating impaired insulin signaling. To see whether this ISP disturbance is specific to developing neurons, we decreased Bin1 expression via RNAi in the HT22 hippocampal neuronal cell line. We found that a transient transfection of a shBin1-expressing plasmid (pGPU6-shBin1) is sufficient to produce a similar insulin resistance state ( Figure 1B); though insulin treatment still evoked a significant increase in p-AKT-Ser473 levels in the RNAi group, the magnitude is much smaller (Figure 1B). .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint As in primary neurons, Bin1 knockdown leads to increased mTORC1 activity in HT2 2 cells, as indicated by the significantly elevated p-RPS6KB1-Tyr389 levels ( Figure 1C). It is known that both mTORC1 and RPS6KB1 phosphorylate IRS1 at s erine residues and attenuate its activity as well as protein stability. We thus treated the cells with PF4708671, a RPS6KB1 inhibitor, and found that the AKT activity was partially restored (Figure 1D). Next, we treated the primary cortical neurons with the pan-mTOR inhibitor rapamycin (200 nM, 2 h), or a mTORC1-specific inhibitor hydroxyzine (10 μM, 2 h),30 and checked the ISP activity. As expected, two hours of rapamycin/hydroxyzine treatment effectively reduced p-RPS6KB1 -Tyr389 levels and successfully increased AKT activity (Figure 1E). Taken together, these results support that the persistent activated mTORC1 and RPS6KB1 feedback inhibits the upstream IRS1 –PI3K–AKT signaling, operating as a route of insulin resistance downstream of BIN1 deficiency. Rapamycin and liraglutide improve spatial cognitive performance in AAV-shBin1 mice. Neuronal insulin signaling is an important regulator of synaptic plasticity, and its state is tightly associated with cognitive functions. 31 As brain insulin resistance is common to AD patients, antidiabetic drugs are being investigated to be repurposed for AD treatment. Among them, the Glucagon-like peptide-1 (GLP-1) receptor agonist liraglutide is capable of crossing the BBB and attenuating insulin resistance both directly and indirectly. 32, 33 Therefore, we explored whether liraglutide and the mTOR inhibitor rapamycin could attenuate insulin resistance and improve cognitive performance in BIN1-deficient mouse models. As in our previous study, we engineered an adeno-associated virus (AAV9) carrying the identical Bin1 -targeting shRNA sequence used in our lentiviral constructs. This AAV also expresses a mCherry reporter driven by the Camk2a promoter, enabling convenient visualization of successfully transduced excitatory neurons (designated as AAV-Camk2a- mCherry-shBin1, hereafter abbreviated as AAV- shBin1; control groups received AAV expressing scrambled shRNA, denoted as AAV-scram. ). With stereotaxic intracranial injection, we precise ly delivered the AAVs to the hippocampal CA1 subregion in mice. (Figure 2A ). One month after the injection, mice received intraperitoneal injections of either liraglutide (200 μg/kg) or rapamycin (2 mg/kg) daily for four weeks. We then employed the Barnes maze test to assess the hippocampus-dependent spatial cognition. Mice were first habituated and spent 5 days in the acquisition training phase, at the end of which all mice could find the escape tunnel at a similar time scale. 3 days after the final training session, mice underwent the probe trial to assess their spatial reference memory. In this session, AAV- shBin1 mice exhibited significantly longer target hole latency and more incorrect explorations than controls. Results showed that liraglutide treatment significantly improved spatial memory performance, as evidenced by reduced latency to locate the target hole and fewer incorrect exploration attempts ( Figure 2B ); similarly, rapamycin treatment also reduced the latency, though it did not reduce the number of incorrect attempts ( Figure 2B ). In agreement with impaired spatial cognitive performance, CA1 slices from AAV-shBin1 mice displayed significantly reduced c-Fos levels (Figure 2C), indicating reduced neuronal activity. In HT22 cells, BIN1 deficiency also leads to a decrease in c-Fos protein ( Figure S1). Analyzing CA1 lysates with WB, .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint we observed significantly increased phosphorylation of eIF4E binding protein 1 (4EBP1 ) on Thr37/46 in the AAV- shBin1 group ( Figure 2D ), indicating elevated mTORC1 activity; meanwhile, AAV- shBin1 mice displayed significantly reduced IRS1 and p- AKT-Ser473 levels, with the latter partially restored by liraglutide or rapamycin treatment (Figure 2E ). In summary, these in vivo data agree with results from cellular models and support persistent mTORC1 activation as a conserved molecular mechanism driving insulin resistance in BIN1-deficient neurons. Reduced TSC1/TSC2 support persistent mTORC1 activation in BIN1-deficient neurons Why does BIN1 deficiency lead to persistent mTORC1 activation? mTORC1 is directly activated by the small GTPase Ras homolog enriched in brain (RHEB), which is inhibited by the tuberous sclerosis complex 1 (TSC1) –TSC2 complex. In patients with tuberous sclerosis, loss-of-function mutations of the TSC1 or TSC2 gene lead to persistent mTORC1 activation. Thus, we examined the protein levels of TSC1 and TSC2 with WB. In cultured cortical neurons, Bin1 knockdown specifically reduced TSC1 levels ( Figure 3A); in HT22 cells, Bin1 knockdown decreased both TSC1 and TSC2, while their phosphorylation (p-TSC1 -Ser511 and p-TSC2 -Thr1462) remained unchanged ( Figures 3B and 3C). At the transcript level, Bin1 knockdown significantly decreased TSC1, while TSC2 displayed a trend of reduction in the pGPU6- shBin1 group ( Figure 3 D). To investigate whether post-transcriptional mechanisms are also involved, we treated HT22 cells with the translation elongation inhibitor cycloheximide (CHX) (10 μM, 2 h or 4 h) and the proteasome inhibitor MG132 (10 μM, 2 h or 4 h). In the control group, MG132 treatment alone led to a time-dependent accumulation of TSC1 proteins, which was blocked by the additional CHX treatment, indicating that those proteins are newly synthesized; in the pGPU6- shBin1 group, however, no significant TSC1 accumulation was observed with the MG132 treatment alone ( Figure 3 E), suggesting impaired protein translation of TSC1. It thus appears that there are both transcriptional and translational mechanisms are responsible for the dysregulation of TSC1 when BIN1 is deficient. ATM-mediated DNA damage response leads to insulin resistance in BIN1-deficient neurons. Why do TSC1/TSC2 and IRS1 reduce when BIN1 is deficient? In addition, specifically in HT22 cells, total AKT appears to decrease in the RNAi group ( Figures 1B, 1D). How could Bin1 knockdown affect so many ISP proteins? The elevated mTORC1 activity p recludes an explanation of global translational repression. What’s more, though treatments with rapamycin or liraglutide attenuated insulin resistance, they primarily affect the phosphorylation of AKT but leave IRS1 levels unaltered ( Figures 1E, 2E ). Thus, there must be other mechanisms involved in the regulation of the insulin signaling following BIN1 knockdown. We recently reported that BIN1 deficiency induces ULK3- initiated autophagy elevation, which likely degrades some of the ISP members. 34 However, it turned out that neither attenuating ULK3 activity nor inhibiting lysosomal proteases rescued ISP protein levels ( Figure S2, S3 ), demonstrating that the autophagy- lysosome pathway is not responsible. In human fibroblasts and cancer cell lines, loss of BIN1 leads to ATM activation. 35 As ATM is an important mediator in DSBs-induced DDR, we hypothesize that BIN1 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint deficiency also elicits DDR in neurons. DDR activates a cascade of signaling events, including cell cycle arrest, DNA damage repair, and metabolic pathway remodeling, ultimately influencing the global cell state and cellular fate. 36 Noteworthy, DDR contains transcriptional and post-transcriptional alterations and could potentially alter ISP protein levels. Once activated, ATM phosphorylates H2AX to produce γH2AX foci, serving as a sensitive biomarker for detecting DSBs. In pGPU6- shBin1 cells, we noticed significantly increased γH2AX levels, which were attenuated by co -treating cells with the ATM inhibitor KU60019 (3 μM, 12 hours) ( Figure 4A ), demonstrating that BIN1 deficiency also leads to ATM-mediated DDR in HT22 cells. Interestingly, the KU60019 treatment almost completely restored protein levels of TSC1, TSC2, RPS6KB1, IRS1, and AKT in pGPU6-shBin1 cells ( Figures 4B and 4C ), and effectively restored cellular insulin response as evaluated by the increase of p- AKT-Ser473 levels following insulin treatment (Figure 4C). These dramatic and extensive effects strongly support DDR as a main driver of insulin resistance in BIN1-deficient cells. Similar to HT22 cells, BIN1 deficiency also increased γH2AX levels in primary cortical neurons ( Figure 4D ), and KU60019 treatment dramatically restored ISP protein levels in the plko -shBin1 group, though AKT is an exception ( Figure 4E). Unlike in proliferating cells, ATM in neurons is predominantly in the cytoplasm, where it acts as a PI3K family member that can directly phosphorylate AKT at Ser 473 to activate it,37 which explains why KU60019 did not improve AKT phosphorylation in cortical neurons as in HT22 cells ( Figure 4E). Collectively, these data support that ATM-mediated DDR is a main driver of neuronal insulin resistance when BIN1 is deficient. BIN1 deficiency decreases ATM protein levels, which also suppresses insulin signaling Following DSBs are sensed by the MRN complex, ATM is recruited and undergoes self- phosphorylation at Ser1981. 38 In BIN1-deficient HT22 cells, p-ATM-Ser1981 levels are significantly elevated, confirming the activation of ATM; unexpectedly, ATM protein itself is significantly reduced in pGPU6- shBin1 cells (Figure 5A). Since ATM is a PI3K kinase, the change of its protein level provides another layer of regulation for the insulin signaling. Similar to HT22 cells, BIN1 deficiency also leads to a significant reduction of ATM proteins in cultured pyramidal neurons ( Figure 5B), and the decrease of ATM is also evidenced in hippocampal CA1 lysates extracted from AAV- shBin1 mice ( Figure 5C). This raises the question of why ATM reduction does not cancel ATM-mediated DDR in BIN1-deficient cells. One explanation is that BIN1 deficiency drives ATM redistribution to the nucleus to support the DDR. We performed immunofluorescence to evaluate ATM distribution in cultured hippocampal neurons; nevertheless, quantitative analysis revealed no significant difference in terms of ATM subcellular distribution between control and plko- shBin1 groups (Figure 5D). It thus appears that a small subset of total ATM protein is already sufficient for DDR activation in the nucleus, which operates in parallel with cytoplasmic ATM reduction and together drives the development of insulin resistance. In the cytoplasm, neuronal ATM deficiency was reported to cause nuclear accumulation of histone deacetylase 4 (HDAC4), 39 which may lead to insulin resistance through epigenetic silencing of GLUT4 and possibly also ISP members. 40 To make clear whether this route is involved, we treated primary neurons with a Class IIa HDAC-selective .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint inhibitor, LMK-235, and examined IRS1 protein levels with WB. In both control and shBin1 groups, we observed no significant alteration of IRS1 levels with the HDAC4 inhibition ( Figure 5E ), indicating the HDAC4-mediated epigenetic mechanism is not involved in IRS regulation. Next, we asked how ATM protein levels are regulated by BIN1 deficiency. As ATM was degraded through the autophagy pathway, 41 BIN1- deficiency induced autophagy elevation is a potential mechanism. 18 Thus, we attenuated lysosome acidification with the vacuolar H +-ATPase (V-ATPase) inhibitor Bafilomycin A1 (BafA1) in HT22 cells; as expected, we evidenced significant accumulation of ATM proteins in the sh Bin1 group ( Figure 5F ), supporting an autophagy route of ATM regulation by BIN1. ROS causes DNA damage in BIN1-deficient neurons In BIN1-deficient human fibroblasts and tumor cells, the activation of ATM is regarded to be E2F1-dependent, but is not caused by genuine DSBs. 29 DSBs have long been found in AD patients and have recently been identified as an important driver of the disease course,42, 43 It is important to determine whether BIN1-deficiency elicits DSBs. Firstly, we employed the comet assay to evaluate whether there are DNA strand breaks in BIN1 - deficient cells. In both cultured cortical neurons and HT22 cells, we observed significantly more “comet tails” in the RNAi groups (Figure 6 A), indicating the existence of true DNA damage. Cytoplasm ic ATM is a key ROS sensor that helps to maintain the redox homeostasis; 26 with reduced available ATM, BIN1-deficiency could compromise cells’ ability to confront ROS stress. To test this, we treated HT22 cells with a low dose of glutamate (5 μM, 24 h) to elicit a mild excitotoxicity. As expected, we observed a mild increase in the cellular ROS level in the control group; in the RNAi group, however, though their basal ROS level is not significantly different, the glutamate treatment elicited significantly higher ROS levels ( Figure 6B). It thus appears that BIN1 deficiency impairs cells’ ability to eliminate ROS. Excessive ROS may cause oxidative DNA damage. In agreement with the increased ROS susceptibility, a low concentration of H 2O2 challenge that could be effectively handled by normal HT22 cells significantly elevated the 8-oxo -deoxyguanosine (8-oxo-dG) levels in the RNAi group ( Figure 6C). Besides oxidative DNA damage, ROS could cause DNA strand breaks and mainly single- strand breaks (SSBs), which may also arise from the repair of the oxidized bases. Since the comet assay is insensitive to discriminate SSBs and DSBs, this raises the question whether BIN1 deficiency-induced DNA damage contains genuine DSBs. We employed immunofluorescence analysis of the γH2AX foci to address this conc ern. In cultured cortical neurons, we observed significantly more γH2AX foci in the plko -shBin1 group without any additional treatment ( Figure 6D); treating neurons with the ATM inhibitor KU60019 or an antioxidant N-acetyl-L-cysteine (NAC) both effectively eliminated them (Figure 6D). Therefore, BIN1 deficiency does induce γH2AX foci formation in an ATM- dependent and ROS-dependent manner, which is in agreement with a previous study that H2O2 elicits ATM-dependent formation of γH2AX foci.44 While γH2AX foci are a gold standard marker of DSBs in cells following exposure to γ -radiation, whether they indicate genuine DSBs in ROS-stressed cells is in debate. While even biologically relevant levels of oxidative stress can induce clustered DNA lesions that subsequently develop into DSBs,45 treating muscle cells with SSB-inducing agents also leads to ATM- dependent γH2AX formation and γH2AX foci, which are not accompanied by p53 phosphorylation. 46 This is .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint interesting as p53 is a canonical substrate of ATM in DSB-induced DDR. We thus examined the phosphorylation of p53. In HT22 cells, neither p-Ser15-p53 nor p-Ser46- p53 levels are elevated in the sh Bin1 group; in fact, p-Ser15-p53 levels are significantly decreased, and p-Ser46-p53 also displays a trend of reduction in the RNAi group ( Figure 6E). Next, we treated the primary neurons with Mirin, a potent MRN inhibitor, to see whether it works analog to ATM inhibition; we found Mirin did not affect γH2AX levels, nor did it alter the insulin signaling ( Figure 6F ), demonstrating that BIN1 deficiency- induced ATM activation does not involve the MRN recruitment step. These results support that BIN1 deficiency does not generate genuine DSBs. A recent study identified the base excision repair (BER) endonuclease AP-endonuclease 1 (APE1) as a direct ATM activator, 47 providing a molecular link between SSBs and ATM. Together, we believe that BIN1 deficiency leads to SSBs-induced ATM activation. Alpha Lipoic Acid improves spatial cognitive performance in AAV-shBin1 mice. ATM helps to maintain cellular redox homeostasis largely through enhancing the pentose phosphate pathway (PPP), which is a major source of the reduced form of nicotinamide adenine dinucleotide phosphate (NADPH). ATM archives this via increasing expression of the rate-limiting enzyme G6PD, 48 as well as stimulating its activity through promoting Hsp27 binding to and phosphorylating G6PD.49 Following neuronal Bin1 knockdown, we observed a significant decrease in G6PD proteins ( Figure 7A ) as well as its transcript G6pdx (Figure 7B). In agreement with the reduced G6PD levels, ELISA assays revealed markedly lower NADPH content in plko -shBin1 neurons ( Figure 7C ). The canonical Wnt/β-catenin signaling pathway regulates both glycolysis and PPP through promoting glucose uptake. Lithium chloride (LiCl) is known to enhance the Wnt/β -catenin signaling through inhibiting GSK- 3β-mediated β -catenin destruction. Treating HT22 cells with LiCl (20 mM, 24 h), we found that the reduced NADPH content in p GPU6-shBin1 cells was effectively restored (Figure 7D). Despite the effectiveness of ATM inhibition in preserving insulin response in BIN1- deficient cellular models, ATM itself is not a suitable target at the organism level because of its vital physiological roles, and ATM insufficiency in actuality promotes systemic insulin resistance. Instead, targeting the upstream DNA damage inducer is more rational. To this end, we aimed to eliminate excessive ROS with the Wnt/β -catenin activator LiCl or by applying a direct antioxidant, the alpha lipoic acid (α -LA). As described above, one month after injection of the AAVs, mice received intraperitoneal administration of either LiCl (2 mmol/kg) or α -LA (10 mg/kg) daily for four weeks. We then performed the Barnes Maze to evaluate their spatial cognitive ability. In a typical setting, α -LA but not LiCl treatment effectively reduced the latency reaching the target holes ( Figure 7E); the number of incorrect exploration attempts also displays a trend of reduction, but there is no significant difference ( Figure 7E ). Since both latency and exploration errors still appeared to be decreasing at the end of the training phase, we initially suspected that mice were insufficiently trained. However, retraining did not produce an appreciable additional improvement (Figure S4). Together, these results support the effectiveness of the anti-ROS strategy with α -LA, and the failure of LiCl might be from its insufficient strength or a cognition-impairing effect of the Lithium through other mechanisms. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint BIN1-high neurons are also more likely to express insulin signaling genes and neuronal activity genes. The above results draw a positive correlation between BIN1 protein levels and neuronal insulin signaling activity. This relationship is possibly conserved at the transcript level. To test this hypothesis, we imported a single-nucleus transcriptomic dataset (GSE157827) that contains an RNA-sequencing of the prefrontal cortex from AD patients as well as healthy control subjects. After clustering the cells into different groups, we further divided neurons into BIN1-low (cells with zero BIN1 count) and BIN1-high (cells with nonzero BIN1 count) subgroups. Evaluating the expression of ISP genes ( IRS1, TSC1, TSC2) and ATM, we found that they are all relatively highly expressed in the BIN1-high subgroup (Figure 8A); meanwhile, neurons expressing ISP genes or ATM are more likely to fall into the BIN1-high subgroup ( Figure 8A ). Similar to the ISP genes, we also analyzed four activity-related genes ( ARC, BDNF, EGR1, and NR4A1) specifically in excitatory neurons, with the results revealing a similar conserved relationship between BIN1 and them ( Figure 8B). FOS is excluded because of its more transient nature as an activity gene. It is worth noting that neuronal BIN1 levels are also positively related to the Base Excision Repair pathway that is responsible for SSB repair ( Figure S5). This is possibly because the chronic genotoxic stress produced by BIN1 deficiency eventually leads to defective DNA repair capability, as reported in neurons from AD patients.50

In the present work, we provide evidence to propose ATM as a key mediator downstream of neuronal BIN1 deficiency. On one side, BIN1 deficiency leads to activation of nuclear ATM and subsequently ATM-dependent DDR, which decreases several key ISP proteins. The surprising effectiveness of the ATM inhibitor supports that DDR is the main route to cellular-level insulin resistance. DDR and the subsequent DNA damage repair pathways modify chromosome accessibility for DNA-binding proteins, including transcription factors, leading to large-scale alterations of gene expression. 51 It is reported that this chromosome modification operates in a non-stochastic way, and p referentially activates stress-response transcriptional programs; 52 but the detailed molecular mechanism is currently unclear and awaits future investigations. On the other side, nuclear ATM- mediated DDR is accompanied by a significant decrease in cytoplasmic ATM, which also contributes to insulin resistance by directly reducing AKT phosphorylation. To reconcile this seeming conflict, we propose that a small fraction of the total ATM is sufficient to support the DDR. A similar scenario was captured by researchers before; an interesting study performed by Jiali et. al found that ATMKO mice express a new catalytically active ATM protein through alternative splicing, and it effectively responds to DNA damage despite the low protein level. 53 We also investigated the crosstalk between cytoplasmic ATM reduction and nuclear DDR. We found that cytoplasmic ATM deficiency-indu ced ROS hypersensitivity predisposes to oxidative DNA damage and the subsequent nuclear ATM-mediated DDR. It thus appears that autophagic degradation of the cytoplasmic ATM is a more upstream event, but it needs to be noted that the possibility of ROS- independent ATM activation is not completely excluded. As a nexus of this ATM-centered regulation, insulin resistance is a natural interfering target and has been proven to be effective in both cellular and animal models. Neuronal .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint insulin resistance is not new to AD researchers, and it is usually interpreted as a consequence of extracellular Aβ exposure, through binding and reducing membrane insulin receptors, or because of inflammatory factors such as TNFα and IL -1β from activated glial cells. In the present work, we report a new cell-autonomous mechanism in which DDR operates as a main driver through multi-layers regulation of the IRS1 – PI3K– AKT pathway. Though DDR was known to affect the insulin signaling, previous studies focused on activated ATM as a PI3K kinase and the inhibition of PI3K – AKT signaling by its downstream effector p53. The possibility of a global change in ISP protein levels was not sufficiently recognized before. Insulin signaling promotes catabolic metabolism to support cell growth, which is an energy-intensive process; on the other hand, DNA damage repair is also highly ATP-consuming and requires the cell to relocate energy from protein/lipid synthesis to it. Meanwhile, as DDR globally affects the cell state, it requires a different set of signals from the ones directing cellular growth, additionally raising the necessity to temporally suppress the insulin signaling. Considering the prime role of genome integrity in the cellular architecture, when DNA damage happens, it is rational to reduce the response to external signals, including insulin, and orchestrate signaling pathways to support DDR and the damage repair. The above argument suggests an overall positive correlation between DDR and insulin resistance, but whether this universally applies to other cells and/or in other settings requires future investigation, considering both DDR and insulin resistance are inherently heterogeneous. Clinical and preclinical studies have drawn a close relationship between neuronal insulin resistance and cognitive impairment. Upon binding of insulin to its receptor, the PI3K –- AKT–mTORC1 arm is activated to support those protein synthesis-required neuronal plasticity, with late-phase LTP (L-LTP) and metabotropic glutamate receptor-dependent long-term depression (mGluR-dependent LTD) in the hippocampus as two most well- known examples.54, 55 It is worth noting that the proper response to external signals other than a straight-lining activation of the relevant signaling pathway is critical for neuronal plasticity. In patients with loss-of-function mutations of TSC1/TSC2 , the persistent activation of mTORC1 in fact also leads to cognitive impairment, in which specific mTORC1 inhibition is a promising therapeutic strategy under intense investigation. 56 The other RAS –ERK1/2 arm also support neuronal plasticity especially for the above- mentioned mGluR-dependent LTD, 57 which requires ERK activation not only because ERK1/2 and AKT –mTORC1 converge on eIF4E to regulate protein synthesis, but also because ERK1/2 phosphorylates ELK1 and the transcription factor CREB that are important for the expression of neuronal plasticity genes. The pan-mTOR inhibitor rapamycin mainly attenuates the PI3K –AKT–AKT–mTORC1 arm, and its effectiveness in our behavior experiments supports the critical involvement of this arm. For liraglutide, except for its insulin secretion-promoting effect in the peripheral, it also directly binds to neuronal GLP-1 receptors, leading to increased intracellular cAMP levels that further activate the PKA –CREB signaling, thus supporting the cognitive improvement. Therefore, while liraglutide treatment did improve p-AKT levels in AAV- shBin1 mice, its cognitive improvement effect should not be interpreted as only from increased insulin sensitivity. If targeting insulin resistance is the main focus, one may also consider peroxisome proliferator- activated receptor γ (PPARγ) agonists that are represented by pioglitazone. These drugs are known to overcome insulin resistance at the cellular level, .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint but they typically penetrate the blood-brain barrier (BBB) poorly; 58 what’s more, though they attenuate cognitive decline in animal models, 59 this effect disappears in AD clinical trials . 60, 61 Leriglitazone is a recently developed brain penetrant PPARγ agonist, 62 and future studies may explore whether it suited neurodegenerative disorders with intact BBB better. Interestingly, a previous study reported that the GLP- 1 agonist increases PPARγ expression in endothelial cells. 63 It is thus possible that liraglutide works partly through activating the PPARγ signaling. This group of drugs is promising, and one of them, semaglutide, is currently under Phase III clinical evaluations for AD treatments. Though targeting insulin resistance is rational and has been proven to be effective, two concerns drove us to look for alternative interfering strategies. First, insulin resistance is too downstream for ATM-mediated DDR, which disturbs many other cellular processes; second, insulin resistance might not be all bad, as it also lowers energy consumption and ROS production. Indeed, some researchers believe insulin resistance at the cellular level is an evolutionarily conserved adaptive physiological mechanism aiming to control ROS, though it might produce systemic adverse effects. 64-66 Therefore, we turned to the upstream hopping to prevent DDR from the beginning and found that the antioxidants α - LA work as expected in terms of preserving cognitive ability. There is known crosstalk between ROS and insulin signaling; ROS by itself may impair insulin signaling through protein oxidation. The antioxidant α-LA is capable of increasing insulin sensitivity and is frequently prescribed for di abetes patients with peripheral neuropathy. In this sense, α - LA appears to simultaneously target the upstream DNA damage and the downstream insulin resistance. However, the half- life of an antioxidant (e.g., 30 min to 1 hour for α - LA) is typically short, which questions their suitability for confronting chronic oxidative stress; in addition, too many antioxidants may elicit reductive stress that disturbs proteostasis and also leads to insulin resistance. Therefore, the α -LA intervention in the present study should be viewed mainly as a proof-of-concept work. Further investigations are required to explore drugs that last long and more optimized interventional regimes. It is surprising that simply reducing the expression of an AD risk gene is sufficient to induce cellular-level insulin resistance, redox dyshomeostasis, genuine DNA damage, DDR, and autophagy elevation, which are all evidenced in AD patients and animal models. This coincidence strongly supports the loss of neuronal BIN1 as an important, and probably common, molecular event in the AD disease course, most likely in the later stage. Future studies may explore the value of neuronal BIN1 as a biomarker reflecting diverse aspects of the cell state. The present work also identifies ATM as a central mediator downstream of neuronal BIN1 deficiency, and we are thus facing a situation reminiscent of the neurodegenerative disorder AT. AT has no cue, but the antioxidant therapy displays some protective effects, 67 which is in agreement with our results. Attenuating ROS and especially ROS-induced DNA damage, alone or in combination with an intervention improving insulin sensitivity, might also be effective for AD treatment.

Reagents and antibodies .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint Leupeptin (SG2012 ), bafilomycin A1 (SF2730) were purchased from Beyotime Biotechnology. Chloroquine was purchased from Beijing Vokai Biotechnology (A31639) and Macklin (C843545). SU6668 (N872395), Glutamate (#L810495) and metformin (#M813341) were purchased from Macklin. Rapamycin (GC15031) , liraglutide (GC10311), and insulin (GP20673) were purchased from GlpBio . PF -4708671 (HY- 15773), Hydroxyzine (HY-B0548A), KU60019 (HY-12061), LMK-235 (HY-18998), Z - Mirin (HY-117693) and NAC( HY-B0215), Cycloheximide (HY-12320) were purchased from MedChemExpress. Lithium chloride (#A100416-0025) was purchased from Sangon Biotech. The primary antibodies used for WB and immunofluorescence were: anti-BIN1 (Proteintech, 14647-1 -AP), anti-ACTB (Proteintech, HRP-60008), anti-MTOR (7C10) (Cell Signaling Technology, 2983), anti-phospho-p70S6 Kinase (Thr389) (108D2) (Cell Signaling Technology, 9234), anti-p70S6 Kinase (Cell Signaling Technology, 9202), anti-AKT (Proteintech, 10176-2 -AP), anti-phospho-AKT (S473) (Proteintech, 66444-1 - lg), anti-phospho-AKT (T308) (Santa Cruz, sc-271966), anti-TSC2 (Proteintech, 24601 - 1-AP), anti-phospho-TSC1 (S511) (Proteintech, 80340-1 -RR), anti-TSC1 (Proteintech, 29906-1-AP), anti-γH2AX (Proteintech, 29380-1-AP), anti-H2AX (Proteintech, 10856-1- AP), anti-Tau (Proteintech, 10274-1- AP); anti-phospho-Tau (Ser396) (abcam, ab32057), anti-phospho-ATM (S1981) (ProMab, P30249); anti-ATM (Abways, CY7229); anti-c- Fos (Affinity, AF0132); anti-G6PD (Proteintech, 66373-1-Ig); anti-EIF4EBP1 (Beyotime, AF5159); anti-phospho-EIF4EBP1 (Beyotime; AF5806). The secondary antibodies used for WB and immunofluorescence were: HRP-conjugated Goat anti-Mouse IgG (Proteintech, SA00001-1), HRP-conjugated Goat anti-Rabbit IgG (Proteintech, SA00001- 2), CoraLite488-conjugated Goat anti-Mouse IgG (Proteintech, SA00013-1), CoraLite 488-conjugated Goat anti-Rabbit IgG (Proteintech, SA00013-2), Rhodamine (TRITC)- conjugated Goat anti-Mouse IgG (Proteintech, SA00007-1), Rhodamine (TRITC)- conjugated Goat anti-Rabbit IgG (Proteintech, SA00007- 2), and Alexa Fluor™ Plus 405 - conjugated Goat anti-Rabbit IgG (Invitrogen, A48254). RNA interference, lentivirus, and AAVs The siRNAs were all synthesized by Genepharma with sequences shown below: siBin1: 5'-GCTCAATCAGAACCTCAATGAT-3'; siUlk3: 5'-GGUUAUUUCUAAAGUUAGA-3'; To reduce Bin1 expression in cultured primary neurons, we constructed a lentivirus vector expressing an shRNA targeting all isoforms of mouse Bin1 under the Rnu6 promoter as well as an EGFP reporter under a separate hPGK1 promoter for convenient visualization of successfully transduced cells under a microscope (plko- U6-shBin1- hPGK-EGFP); the control virus expresses a scrambled sequence of sh Bin1 (plko- U6- scramble-hPGK-EGFP). Both lentiviruses were packaged and produced by Hanbio. To reduce neuronal Bin1 expression in the hippocampal CA1, we employed a rAAV2/9 Mir30a-based vector expressing the same shRNA targeting all isoforms of mouse Bin1, as well as a mCherry reporter, both under the Camk2a promoter for specifical expression in excitatory neurons (rAAV- Camk2a-mCherry-5’MiR30a-shBIN1-3’MiR30a-WPREs, abbreviated as AA V-Camk2a-mCherry-shBin1 or AA V -shBin1); rAAV- Camk2a- .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint mCherry-5’MiR30a-scramble-3’MiR30a-WPREs (abbreviated as AA V-Camk2a- mCherry-scramble or AA V -scram.) was used as the control . The two AA Vs were packaged by BrainVTA. Culture of primary hippocampal neurons The primary hippocampal and cortical neurons were isolated from newborn (P0) mice with a standard procedure reported before 68. Briefly, mice were disinfected with 75% alcohol before sacrifice and dissection; in cold DMEM (Thermo Scientific, C11995500BT), the meninges were carefully and gently peeled off with forceps, and the hippocampus and the cortex were identified and dissected out; the hippocampus and cortex were then cut up and digested in a solution containing 1 mg/ml DNase (Merck, D4513) and 2 mg/ml papain (Sangon Biotech, 9001-73 -4), filtered through a Falcon® 40 µm cell strainer (Corning, 352340); cells were then transferred to Neurobasal™ -A (Thermo Scientific, 10888022 ) containing 2% B- 27™ (Thermo Scientific, 17504044), 1% penicillin/streptomycin (Thermo Scientific, 15140122), and 2 mM L-glutamine (Thermo Scientific, 35050061 ). Neurons were cultured in Dishes or plates coated with 0.01 mg/ml poly-D-lysine (Merck, P6407) for 5 days before transfection or lentivirus transduction. Lentivirus transduction The lentivirus was introduced into the primary hippocampal neurons at 5 days in vitro (DIV5). 24 h after the transduction, the old medium containing the virus was removed and replaced with fresh medium. On DIV11, neurons were harvested for the subsequent immunofluorescence, real-time quantitative reverse transcription PCR (qRT-PCR), or WB analyses. Mouse maintenance, separation of brain tissue The 8-week -old wild-type C57BL/6J mice were purchased from the Animal Core Facility of Nanjing Medical University. All mice underwent a 12-hour light/dark cycle under appropriate ambient temperature and humidity, with good ventilation, and ad libitum access to food and water. Mice were injected with AAV- Camk2a-mCherry-shBin1 or AAV-Camk2a-mCherry-scramble in bilateral hippocampal CA1 and fed for 8 weeks before the sacrifice. Mice were anesthetized by intraperitoneal injection of 30 ml/kg phenobarbital (Merck, P3761). After perfusion with cold physiological saline, the skull was opened, and the whole brain tissue was collected. All tissues were frozen in liquid nitrogen and stored at -80℃ before use. Stereotaxic intracranial injection Mice were anesthetized and fixed in a stereotaxic frame; the hair near the injection site was cut off and disinfected twice with betadine. Carefully cut the skin along the sagittal slit with disinfected scissors to expose the skull surface. For the introduction of AAVs into the bilateral hippocampal CA1, 0.5 µL of virus for each side was injected at the following coordinates: 2.5 mm behind bregma, 1.0 mm off, and 1.9 mm deep. The infusion rate was 0.1 µL /min with a Hamilton 33G syringe needle, which remained at the injection site for 5 minutes after the infusion to prevent virus spillover. After the injection, mice were placed in an isolated environment to facilitate their recovery. All experiments were performed 8-10 weeks after the injection. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint Intraperitoneal injection Mice were intraperitoneally injected with 2 mg/kg rapamycin or 200 µg/kg liraglutide for 4 weeks after AAV infection for 8-10 weeks. The control group received an equal amount of PBS by intraperitoneal injection. Behavioral tests Mice's behavioral images were captured with a computer-connected digital camera (TopScan CleverSys, USA). The test mice were acclimated to the room for 2 hours before a behavioral test. Learning and memory abilities were assessed with the Barnes Maze test. Barnes maze Barnes maze is used to evaluate long-term spatial memory in mice 69. The behavioral test consists of five days of training and one day of testing. The experimental device is a disk with a diameter of 90 cm. There are 22 holes evenly distributed at 5 cm from the edge of the disk, of which only one hole is equipped with an escape box. A bright light source was placed above the platform, forcing the mice to explore and hide in the box. During the training period of 5 consecutive days, the mice were trained once a day. If the mice did not enter the escape box within 300 s, they were guided into it, and the mice were removed and put back into the cage 30 s later. The tests were conducted three days apart aft er the training phase. During the test period, each mouse was active for 300 s. Software was used to record data in real time during behavioral processes. Finally, the time when mice entered the escape box and the number of times they sniffed the wrong hole were counted. ROS detection ROS levels were evaluated with a DCFHDA-based ROS detection kit (Beyotime, S0033S) according to the manufacturer’s manual. Briefly, after removing the cell culture medium, an appropriate volume of diluted DCFHDA was added, and cells were then incubated in the incubator at 37℃ for 20 min. Cells were washed three times with serum -free culture medium to fully remove the DCFHDA that did not enter cells before digestion with 0.25% pancreatin. Cells were then collected and detected with a Varioskan Flash microplate reader (Thermo Scientific, USA). The excitation wavelength of 488 nm and the emission wavelength of 525 nm were used for the detection. NADPH detection The kit (No. BYSH-0208W) was purchased from Boyan Biotechnology. According to the instructions, NADPH was extracted, and a given amount of extract B was used to lyse the cells. The lysed cells were collected into an EP tube using a cell scraper on ice at 4℃ for 30 min. After incubation, they were immediately placed on ice for 5 min. The samples were then transferred to a new EP tube after 4℃ pre -cooling centrifugation, with a speed of 12000 rpm for 10 min. Half of the supernatant was aspirated and transferred to a new EP tube. Extract A was added in small amounts multiple times, and the solution was tested with a pH paper for color development. The pH of the resulting solution was adjusted to neutral, and the volume of extract A added to each sample was recorded. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint Under 4℃ conditions, the samples were centrifuged at 12000 rpm for 5 min, and the supernatant was transferred to a new EP tube and placed in an ice bath before testing. According to the instructions, reagents one, two, and three (powders should be centrifuged before dissolution in distilled water) were added to the 96-well plate in the corresponding amounts. The samples were incubated in the dark at 37℃ for 10 min, followed by the addition of reagent four. The microplate reader was set to 37℃ and 450 nm wavelength, and the absorbance values of each well were measured and recorded as A1. Thirty minutes later, the absorbance values of each well were measured again using the same parameters on the microplate reader and recorded as A2. The final content was calculated using the formula provided in the manual. 8 -oxo-dG detection Pre-treatment of the samples and ELISA On DIV11, neurons were exposed to 20 μM H2O2 for 1 h, and then collected in PBS. Cellular DNA was extracted through the freeze-thaw cycle: samples were frozen to −80°C and then immediately transferred to a 37°C water bath for thawing for a total of three cycles. Samples were then centrifuged at 12000 rpm at 4°C for 15 mins, with the supernatant collected for the subsequent ELISA steps and the precipitate for protein extraction. For ELISA, the mouse 8-oxo -dG ELISA Kit (Huabo Deyi, Beijing, China) was employed. Samples were analyzed in triplicate (50 µL sample/well) according to the manufacturer's instructions. Concentration of the 8-oxo -dG (ng/ml) was then normalized to the protein concentration (nmol 8-oxo-dG/mmol protein). Cell line culture and transfection Immortalized mouse hippocampal HT22 cells (Wuhan Pricella Biotechnology, CL-0697) were cultured in DMEM high-glucose medium containing 10% fetal bovine serum with penicillin and streptomycin. Transfections we re performed with Lipofectamine™ 3000 (Invitrogen, L3000075) when cells reached ~70% confluence. Transfected cells were cultured in the incubator (37℃, 5% CO2) for an additional 72 h before harvest. Comet electrophoresis Also known as single-cell gel electrophoresis, the collected cell suspension is evenly spread on a microscope slide previously coated with 1% normal melting point agarose in a 37℃, 75 μL, 0.7% low-melting-point agarose solution, and incubated at 4°C for 10 min to solidify. The slide is then incubated in the lysis solution at 4℃ , 1-2 h. After lysis, the slide is incubated at room temperature in an alkaline electrophoresis buffer for 20-60 min to unwind the DNA, followed by electrophoresis (25 V) for 20-30 min. The slide is then placed in a neutralizing buffer (0.4 M Tris, pH 7.5) at 4°C for 5-10 min, after which the buffer is discarded and the slide is incubated in the dark with 20 μL Propidium Iodide Solution for 15 min. The slide is washed three times with deionized water, covered with a coverslip, and finally observed under a fluorescence microscope to obtain images. ELISA .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint Collect the supernatants from primary neuronal cultures of each group. According to the instructions for the mouse A β1-42 and Aβ 1-40 ELISA kit, add the prepared supernatants and standards to the reaction wells, incubate at 37℃ for 30 min. After incubation, wash the wells 5 times, then add the enzyme- labeled reagent and react at 37℃ for another 30 min. Following the washing steps, add the chromogenic solution and allow a 10-minute color development reaction. Finally, add the stop solution to terminate the reaction, and read the absorbance values of each well within 15 min using an enzyme-labeled microplate reader. Immunofluorescence Tissues were fixed with 4% paraformaldehyde for 24 h, dehydrated with 20% sucrose in Phosphate-buffered saline (PBS, Sangon Biotech, E607008) for 24 h, and then 30% sucrose in PBS for ~24 h until they sank to the bottom. Tissues were embedded in OCT glue (S akura Finetek, 4583), sliced at a thickness of 20 μm with a CM1950 cryostat microtome (Leica Biosystems, Germany), and lastly mounted on gelatin-coated slides. 0.1% Triton X-100 (Sangon Biotech, A600198) for 1 h was applied for tissue permeabilization, and then tissues were blocked in 5% BSA (Sangon Biotech, A600332). For cultured cells, the permeabilization time is 10 min. For primary antibodies, samples were incubated at 4℃ overnight; for secondary antibodies, samples were incubated at room temperature for 2 h. WB RIPA buffer (Beyotime, P0013B) with a protease inhibitor cocktail (Beyotime, st506) and a phosphatase inhibitor cocktail (Roche, 04906837001) was used for cell lysis and protein extraction. For tissue homogenization, cells or tissue lysate were placed in an ice shaker at 4℃ for 30 min to fully lyse. With 1 0000 x g centrifugation, supernatant was collected and protein quantification was performed using Enhanced BCA Protein Assay Kit (Beyotime, P0010S). The SDS-PAGE Sample Loading Buffer 5X (Beyotime, P0015) was added to the remaining protein lysate, and the protein was heated at 95℃ for 5 min. Thirty-μg protein samples were separated by electrophoresis at 80 v, 30 min, 120 v, 2 h, and then transferred to Immobilon®-PSQ PVDF Membrane (Merck, ISEQ08100), blocked with 5% dry skim milk, incubated with different primary antibodies in a shaking bed at 4℃ overnight, and washed with TBST for 3 times, 10 min each time. After incubation for 1 h, the corresponding HRP-conjugated secondary antibody was washed with TBST 3 times, 10 min each time. When there are too many samples to fit in a single gel but analyzing in one batch is necessary, we run two gels and the following procedures alongside to enable a cross-comparison between groups on different gels. qRT-PCR RNAiso Plus (Takara Bio, 9109) was used to extract total RNA from cells or tissues. 4x EZscript Reverse Transcription Mix II (with gDNA Remover) kit (EZBioscience, EZB- RT2GQ) and 2x SYBR Green qRT-PCR Master Mix (EZBioscience, A0001- R1) were used for q RT-PCR with the following settings: 95℃ for 5 min, 60℃ for 30 s for 40 cycles, 95℃ for 10 s; Actb was employed as an internal reference; the delta-delta Ct

was used to calculate the relative fold of target genes. All primers were synthesized by Sangon Biotech. Sequences of all primers were listed in Table S1. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint Image analysis A DM4000 B fluorescence microscope (Leica Biosystems, Germany) and an LSM 710 laser scanning confocal microscope (Zeiss, Germany) were employed to image all cell slides or tissue sections. For each group, at least 3 slides were randomly imaged. For immunostaining or fluorescence intensity analysis, ~3 regions of interest (ROIs) with the fixed area were randomly selected for each neuron from high-magnification images; averaged intensities of each ROI were employed for comparison. Single-cell transcriptomics Data Source and Analysis The GSE157827 dataset, which is from single-nucleus RNA -sequencing of prefrontal cortex from AD patients and healthy control subjects, was re-analyzed in the present work. Briefly, after standard quality control and preprocessing, a total of 166,685 cells were identified, of which 88,363 were annotated as neurons. Neurons were then stratified into BIN1-high (non-zero BIN1 expression) and BIN1-low (zero expression) groups based on BIN1 transcript levels. To eliminate the influence of different group sizes, 43993 cells in each group were employed (randomly selected when necessary) for the plotting analysis. Expression of selected ISP genes ( IRS1, TSC1, TSC2, ATM ) was compared between the two groups. Data visualization was performed using grouped bar plots and histograms. Employing the same analytical framework, we also analyzed expression of neuronal activity-related genes (ARC, BDNF, EGR1, and NR4A1) and base excision repair (BER)– related genes ( OGG1, NEIL1, NEIL2, APEX1, XRCC1, PARP1 ) between BIN1-defined subgroups in excitatory neurons (n = 62229 in total; n = 29679 in each subgroup). Excitatory neurons were defined by simultaneous high expression of SLC17A7 (encoding VGLUT1) and CAMK2A (encoding calcium/calmodulin-dependent protein kinase II alpha) in the annotated neurons. Study approval All animal experiments were conducted following the international standards on animal welfare and the guidelines of the Institute for Laboratory Animal Research of Nanjing Medical University (approval no. IACUC: 2201039 and IACUC:1812054-4). Statistical analyses All data were statistically analyzed by GraphPad Prism. Data were expressed as Mean + SEM unless otherwise noted. Two -tailed Student’s t -test was performed for comparing two groups of data, and two- way ANOVA followed by Tukey’s post hoc test was employed for comparisons among 4 or more groups. P < 0.05 was considered statistically significant. Data availability statement The data used and analyzed during the current study are available from the corresponding author up on request. Disclosure statement .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint The authors declare that they have no conflict of interest.

1. Seshadri, S. et al. Genome-wide analysis of genetic loci associated with Alzheimer disease. JAMA 303, 1832-1840 (2010). 2. Kunkle, B.W. et al. Genetic meta-analysis of diagnosed Alzheimer's disease identifies new risk loci and implicates Aβ, tau, immunity and lipid pr ocessing. Nat Genet 51, 414-430 (2019). 3. Glennon, E.B. et al. BIN1 is decreased in sporadic but not familial Alzheimer's disease or in aging. PLoS One 8, e78806 (2013). 4. Holler, C.J. et al. Bridging integrator 1 (BIN1) protein expression increases in the Alzheimer's disease brain and correlates with neurofibrillary tangle pathology. J Alzheimers Dis 42, 1221-1227 (2014). 5. De Rossi, P. et al. Predominant expression of Alzheimer's disease-associated BIN1 in mature oligodendrocytes and localization to white matter tracts. Mol Neurodegener 11, 59 (2016). 6. Franzmeier, N., Rubinski, A., Neitzel, J. & Ewers, M. The BIN1 rs744373 SNP is associated with increased tau-PET levels and impaired memory. Nat Commun 10, 1766 (2019). 7. Hu, H. et al. Association between methylation of BIN1 promoter in peripheral blood and preclinical Alzheimer’s disease. Translational Psychiatry 11, 89 (2021). 8. Guo, F. et al. sTREM2 Mediates the Correlation Between BIN1 Gene Polymorphism and Tau Pathology in Alzheimer's Disease. J Alzheimers Dis 101, 693-704 (2024). 9. Sottejeau, Y. et al. Tau phosphorylation regulates the interaction between BIN1’s SH3 domain and Tau’s proline -rich domain. Acta Neuropathologica Communications 3, 58 (2015). 10. Chapuis, J. et al. Increased expression of BIN1 mediates Alzheimer genetic risk by modulating tau pathology. Mol Psychiatry 18, 1225-1234 (2013). 11. Sartori, M. et al. BIN1 recovers tauopathy-induced long-term memory deficits in mice and interacts with Tau through Thr(348) phosphorylation. Acta Neuropathol 138, 631-652 (2019). 12. Calafate, S., Flavin, W., Verstreken, P. & Moechars, D. Loss of Bin1 Promotes the Propagation of Tau Pathology. Cell Rep 17, 931-940 (2016). 13. Voskobiynyk, Y. et al. Alzheimer's disease risk gene BIN1 induces Tau- dependent network hyperexcitability. Elife 9 (2020). 14. De Rossi, P. et al. Neuronal BIN1 Regulates Presynaptic Neurotransmitter Release and Memory Consolidation. Cell Rep 30, 3520-3535 e3527 (2020). 15. Schürmann, B. et al. A novel role for the late-onset Alzheimer’s disease (LOAD)- associated protein Bin1 in regulating postsynaptic trafficking and glutamatergic signaling. Molecular Psychiatry 25, 2000-2016 (2020). 16. Sheng, C. et al. Experience-dependent structural plasticity targets dynamic filopodia in regulating dendrite maturation and synaptogenesis. Nat Commun 9, 3362 (2018). .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint 17. Zhang, X. et al. Bridging Integrator 1 (BIN1) Genotype Effects on Working Memory, Hippocampal Volume, and Functional Connectivity in Young Healthy Individuals. Neuropsychopharmacology 40, 1794-1803 (2015). 18. Jin, Y. et al. BIN1 deficiency enhances ULK3-dependent autophagic flux and reduces dendritic size in mouse hippocampal neurons. Autophagy, 1-20 (2024). 19. Talbot, K. et al. Demonstrated brain insulin resistance in Alzhe imer’s disease patients is associated with IGF-1 resistance, IRS-1 dysregulation, and cognitive decline. The Journal of Clinical Investigation 122, 1316-1338 (2012). 20. Steen, E. et al. Impaired insulin and insulin-like growth factor expression and signaling mechanisms in Alzheimer's disease --is this type 3 diabetes? J Alzheimers Dis 7, 63-80 (2005). 21. Banks, W.A., Owen, J.B. & Erickson, M.A. Insulin in the brain: There and back again. Pharmacology & Therapeutics 136, 82-93 (2012). 22. De Felice, F.G., Lourenco, M.V. & Ferreira, S.T. How does brain insulin resistance develop in Alzheimer's disease? Alzheimer's & Dementia 10, S26-S32 (2014). 23. Huang, Y.-W.A., Zhou, B., Wernig, M. & Südhof, T.C. ApoE2, ApoE3, and ApoE4 Differentially Stimulate APP Transcription and Aβ Secretion. Cell 168, 427-441.e421 (2017). 24. Bar, R.S. et al. Extreme Insulin Resistance in Ataxia Telangiectasia. New England Journal of Medicine 298, 1164-1171 (1978). 25. Schalch, D.S., McFarlin, D.E. & Barlow, M.H. An Unusual Form of Diabetes Mellitus in Ataxia Telangiectasia. New England Journal of Medicine 282, 1396- 1402 (1970). 26. Ditch, S. & Paull, T.T. The ATM protein kinase and cellular redox signaling: beyond the DNA damage response. Trends in Biochemical Sciences 37, 15-22 (2012). 27. Ambrose, M. & Gatti, R.A. Pathogenesis of ataxia-telangiectasia: the next generation of ATM functions. Blood 121, 4036-4045 (2013). 28. Simpson, J.E. et al. Neuronal DNA damage response-associated dysregulation of signalling pathways and cholesterol metabolism at the earliest stages of Alzheimer-type pathology. Neuropathology and Applied Neurobiology 42 , 167- 179 (2016). 29. Folk, W.P. et al. Loss of the tumor suppressor BIN1 enables ATM Ser/Thr kinase activation by the nuclear protein E2F1 and renders cancer cells resistant to cisplatin. Journal of Biological Chemistry 294, 5700-5719 (2019). 30. Allen, S.A., Tomilov, A. & Cortopassi, G.A. Small molecules bind human mTOR protein and inhibit mTORC1 specifically. Biochem Pharmacol 155, 298-304 (2018). 31. Spinelli, M., Fusco, S. & Grassi, C. Brain Insulin Resistance and Hippocampal Plasticity: Mechanisms and Biomarkers of Cognitive Decline. Frontiers in Neuroscience Volume 13 - 2019 (2019). 32. Hunter, K. & Hölscher, C. Drugs developed to treat diabetes, liraglutide and lixisenatide, cross the blood brain barrier and enhance neurogenesis. BMC Neuroscience 13, 33 (2012). .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint 33. Yaribeygi, H., Sathyapalan, T. & Sahebkar, A. Molecular mechanisms by which GLP-1 RA and DPP-4i induce insulin sensitivity. Life Sciences 234, 116776 (2019). 34. Jin, Y. et al. BIN1 deficiency enhances ULK3-dependent autophagic flux and reduces dendritic size in mouse hippocampal neurons. Autophagy 21, 223-242 (2025). 35. Folk, W.P. et al. Loss of the tumor suppressor BIN1 enables ATM Ser/Thr kinase activation by the nuclear protein E2F1 and renders cancer cells resistant to cisplatin. J Biol Chem 294, 5700-5719 (2019). 36. Shimizu, I., Yoshida, Y., Suda, M. & Minamino, T. DNA damage response and metabolic disease. Cell Metab 20, 967-977 (2014). 37. Alexander, A. et al. ATM signals to TSC2 in the cytoplasm to regulate mTORC1 in response to ROS. Proceedings of the National Academy of Sciences 107, 4153- 4158 (2010). 38. Lavin, M.F., Kozlov, S., Gatei, M. & Kijas, A.W. ATM-Dependent Phosphorylation of All Three Members of the MRN Complex: From Sensor to Adaptor. Biomolecules 5 , 2877-2902 (2015). 39. Li, J. et al. Nuclear accumulation of HDAC4 in ATM deficiency promotes neurodegeneration in ataxia telangiectasia. Nature Medicine 18, 783-790 (2012). 40. Lin, Y. et al. HDAC4: an emerging target in diabetes mellitus and diabetic complications. European Journal of Medical Research 30, 429 (2025). 41. Cheng, A. et al. ATM loss disrupts the autophagy-lysosomal pathway. Autophagy 17, 1998-2010 (2021). 42. Adamec, E., Vonsattel, J.P. & Nixon, R.A. DNA strand breaks in Alzheimer's disease. Brain Research 849, 67-77 (1999). 43. Dileep, V. et al. Neuronal DNA double-strand breaks lead to genome structural variations and 3D genome disruption in neurodegeneration. Cell 186, 4404- 4421.e4420 (2023). 44. Narciso, L. et al. The Response to Oxidative DNA Damage in Neurons: Mechanisms and Disease. Neural Plasticity 2016, 3619274 (2016). 45. Sharma, V. et al. Oxidative stress at low levels can induce clustered DNA lesions leading to NHEJ mediated mutations. Oncotarget 7, 25377-25390 (2016). 46. Fortini, P. et al. DNA damage response by single-strand breaks in terminally differentiated muscle cells and the control of muscle integrity. Cell Death & Differentiation 19, 1741-1749 (2012). 47. Zhao, H., Li, J., You, Z., Lindsay, H.D. & Yan, S. Distinct regulation of ATM signaling by DNA single-strand breaks and APE1. Nature Communications 15, 6517 (2024). 48. Zhang, Y. et al. Mitochondrial redox sensing by the kinase ATM maintains cellular antioxidant capacity. Sci Signal 11 (2018). 49. Cosentino, C., Grieco, D. & Costanzo, V. ATM activates the pentose phosphate pathway promoting anti-oxidant defence and DNA repair. Embo j 30, 546-555 (2011). 50. Weissman, L. et al. Defective DNA base excision repair in brain from individuals with Alzheimer's disease and amnestic mild cognitive impairment. Nucleic Acids Research 35, 5545-5555 (2007). .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint 51. Stanic, M. & Mekhail, K. Integration of DNA damage responses with dynamic spatial genome organization. Trends in Genetics 38, 290-304 (2022). 52. Silva, E. & Ideker, T. Transcriptional responses to DNA damage. DNA Repair 79, 40-49 (2019). 53. Li, J., Chen, J., Vinters, H.V., Gatti, R.A. & Herrup, K. Stable Brain ATM Message and Residual Kinase-Active ATM Protein in Ataxia- Telangiectasia. The Journal of Neuroscience 31, 7568-7577 (2011). 54. Levenga, J. et al. AKT isoforms have distinct hippocampal expression and roles in synaptic plasticity. eLife 6, e30640 (2017). 55. Hou, L. & Klann, E. Activation of the Phosphoinositide 3-Kinase-Akt- Mammalian Target of Rapamycin Signaling Pathway Is Required for Metabotropic Glutamate Receptor-Dependent Long-Term Depression. The Journal of Neuroscience 24 , 6352-6361 (2004). 56. Curatolo, P., Specchio, N. & Aronica, E. Advances in the genetics and neuropathology of tuberous sclerosis complex: edging closer to targeted therapy. The Lancet Neurology 21, 843-856 (2022). 57. Gallagher, S.M., Daly, C.A., Bear, M.F. & Huber, K.M. Extracellular Signal- Regulated Protein Kinase Activation Is Required for Metabotropic Glutamate Receptor-Dependent Long-Term Depression in Hippocampal Area CA1. The Journal of Neuroscience 24 , 4859-4864 (2004). 58. Cai, W. et al. Peroxisome proliferator- activated receptor γ (PPARγ): A mast er gatekeeper in CNS injury and repair. Progress in Neurobiology 163-164, 27-58 (2018). 59. Alhowail, A., Alsikhan, R., Alsaud, M., Aldubayan, M. & Rabbani, S.I. Protective Effects of Pioglitazone on Cognitive Impairment and the Underlying Mechanisms: A Review of Literature. Drug Des Devel Ther 16, 2919-2931 (2022). 60. Maeshiba, Y. et al. Disposition of the new antidiabetic agent pioglitazone in rats, dogs, and monkeys. Arzneimittelforschung 47, 29-35 (1997). 61. Abyadeh, M. et al. Comparative Analysis of Aducanumab, Zagotenemab and Pioglitazone as Targeted Treatment Strategies for Alzheimer's Disease. Aging Dis 12, 1964-1976 (2021). 62. Pizcueta, P. et al. Development of PPARγ Agonists for the Treatment of Neuroinflammatory and Neurodegenerative Diseases: Leriglitazone as a Promising Candidate. International Journal of Molecular Sciences 24, 3201 (2023). 63. Onuma, H. et al. The glucagon-like peptide 1 receptor agonist enhances intrinsic peroxisome proliferator- activated receptor γ activity in endothelial cells. Biochemical and Biophysical Research Communications 451, 339-344 (2014). 64. Erol, A. Insulin resistance is an evolutionarily conserved physiological mechanism at the cellular level for protection against increased oxidative stress. BioEssays 29, 811-818 (2007). 65. Barzilai, N. & Ferrucci, L. Insulin Resistance and Aging: A Cause or a Protective Response? The Journals of Gerontology: Series A 67, 1329-1331 (2012). 66. Hoehn, K.L. et al. Insulin resistance is a cellular antioxidant defense mechanism. Proceedings of the National Academy of Sciences 106, 17787-17792 (2009). .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint 67. Reliene, R. & Schiestl, R.H. Experimental antioxidant therapy in ataxia telangiectasia. Clin Med Oncol 2, 431-436 (2008). 68. Beaudoin, G.M.J. et al. Culturing pyramidal neurons from the early postnatal mouse hippocampus and cortex. Nature Protocols 7, 1741-1754 (2012). 69. Pitts, M.W. Barnes Maze Procedure for Spatial Learning and Memory in Mice. Bio Protoc 8 (2018). Figure 1. ( A) Representative WB images showing the expression of IRS1, p- AKT- Ser473, AKT, and BIN1 protein in primary mouse cortex neurons in scramble and plko- shBin1 groups with or without insulin treatment (100 nM, 10 min). Right, a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing the .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint corresponding quantification of IRS1 and p-AKT-Ser473 bands intensity. n = 3 per group. (B) Representative WB images showing the expression of IRS1, p-AKT -Ser473, AKT, and BIN1 protein of HT22 cells in scramble and pGPU6- shBin1 groups with or without insulin treatment (100 nM, 10 min). Right, a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing the corresponding quantification of IRS1 and p- AKT-Ser473 bands intensity. n = 3 per group. ( C) Representative WB images and a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing p- RPS6KB1 and RPS6KB1 protein expression in different groups of HT22 cells. n = 3 per group. (D) Representative WB images showing protein levels of IRS1, p- AKT-Ser473, and AKT in scramble and pGPU6- shBin1 groups with or without PF4708671 treatment (10 μM, 1 h); n = 3 per group. ( E) Representative WB images showing protein levels of IRS1, p-AKT -Ser473, AKT, p-RPS6KB1, RPS6KB1, and BIN1 in scramble and plko- shBin1 groups with or without hydroxyzine (10 μM, 2 h) and rapamycin (200 nM, 2 h) treatment; n = 4 per group. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint Figure 2. ( A) An illustration showing stereotaxic intracranial injection into the hippocampal CA1 region of 2-month-old mice. ( B) Heat maps showing representative trajectories of mice in the Barnes maze during the probe phase; bottom, line graphs showing the latencies and mistakes made (primary errors) before reaching the escape tunnel during the acquisition training phase; bar graphs (Mean + SEM) overlaid with dot plots (individual data points) showing the latencies and primary errors before reaching the previous escape tunnel location during the probe phase. n = 12 mice per group. ( C) Representative immunofluorescence images showing c-Fos + cells in hippocampal CA1; .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint right, a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing the numbers of c-Fos + cells in different groups. Scale bar: 100 μm. n = 6 per group. (D) Representative WB images and a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing p-4EBP1(T37/46) and 4EBP1 protein expression in scramble and AAV-shBin1 groups. n = 4 per group. ( E) Representative WB strips showing protein levels of IRS1, p- AKT-Ser473, AKT, and Bin1 from CA1 lysates in different groups; right, a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing the corresponding quantification of IRS1, p- AKT-Ser473, AKT , and Bin1 bands intensity. n = 8 per group. Figure 3. (A) Representative WB images and a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing TSC1, TSC2, and BIN1 protein expression in scramble and plko-shBin1 groups. n = 3 per group. ( B) Representative WB images and a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing p - TSC1-Ser511 and TSC1 protein expression in different groups of HT22 cells. n = 3 per group. (C) Representative WB images and a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing p -TSC2-Thr1462 and TS C2 protein expression in different groups of HT22 cells. n = 3 per group. ( D) A bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing the relative transcript levels of Bin1, Tsc1, and Tsc2 in different groups of HT22 cells. Transcripts were examined by qRT-PCR with Actb as an internal reference. n = 3 per group. ( E) Representative WB images showing the expression of TSC1, TSC2, and BIN1 protein of HT22 cells in different groups with or without cycloheximide (10 μM, 2 h or 4 h) or MG132 (10 μM, 2 h or 4 h) treatment . .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint Right, a line chart 1 (Mean ± SEM) showing the TSC1 protein expression in different groups. n = 3 per group. Figure 4. (A) Representative WB images showing protein levels of γ H2AX, H2AX, and Bin1 in scramble and pGPU6-shBin1 groups with or without KU60019 treatment (3 μM, 72 h); n = 3 per group. ( B) Representative WB images showing protein levels of p - RPS6KB1, RPS6KB1, TSC1, TSC2, and BIN1 in scramble and pGPU6- shBin1 groups with or without KU60019 treatment (3 μM, 72 h); n = 3 per group. ( C) Representative WB images showing protein levels of IRS1, p- AKT-Ser473 and AKT in scramble and pGPU6-shBin1 groups with or without KU60019 (3 μM, 72 h) and insulin (100 nM, 10 min) treatment; n = 3 per group. (D) Representative WB images and a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing γ H2AX, H2AX and BIN1 protein expression in different groups of primary cortex neurons. n = 3 per group. ( E) Representative WB images showing protein levels of p -RPS6KB1, RPS6KB1, TSC1, .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint TSC2, IRS1, p-AKT-Ser473, AKT and BIN1 in scramble and plko-shBin1 groups with or without KU60019 treatment (3 μM, 72 h); n = 3 per group. Figure 5. (A) Representative WB images showing protein levels of p -ATM and ATM in scramble and pGPU6- shBin1 groups; n = 3 per group. ( B) Representative WB images and a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing ATM and BIN1 protein expression in different groups of primary cortex neurons and p rimary hippocampal neurons. n = 3 per group. ( C) Representative WB images and a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing ATM and BIN1 protein expression in the scramble and AAV-shBin1 groups. n = 6 per group. ( D) .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint Representative immunofluorescence images showing the distribution of p -ATM and ATM of hippocampal neurons in scramble and plko- shBin1 groups; bottom, a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing the percentage of area occupied by p-ATM and ATM in the nucleus or cytoplasm in different groups; Scale bar: 10 μm. n = 11-12 per group. (E) Representative WB images and a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing IRS-1 and BIN1 protein expression in scramble and plko-shBin1 groups with or without LMK-235 treatment (200 nM, 24 h); n = 3 per group. ( F) Representative WB images and a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing IRS-1 and ATM protein expression in scramble and pGPU6- shBin1 groups with or without BafA1 treatment (0.1 μM, 24 h); n = 3 per group. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint Figure 6. (A) Representative comet assay images show the DNA damage in the different groups of primary hippocampal neurons and HT22 cells. (B) A bar graph (Mean + SEM) .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint overlaid with dot plots (individual data points) showing the ROS levels in scramble and pGPU6-shBin1 groups with or without Glutamate treatment (5 μM, 24 h); n = 3 per group. (C) A bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing the 8-oxo -dG levels in scramble and pGPU6- shBin1 groups with or without H 2O2 treatment (20 μM, 1 h); n = 3 per group. (D) Representative immunofluorescence images showing the γH2AX+ cells in the scramble and plko- shBin1 groups with or without KU60019 (3 μM, 72 h) and NAC (3 mM, 1 2 h) treatment. Scale bar: 20 μm. n = 15 per group. (E) Representative WB images showing protein levels of p -53-Ser46, p-53-Ser15, and P53 in scramble and pGPU6- shBin1 groups with or without KU60019 treatment (3 μM, 72 h); n = 3 per group. ( F) Representative WB images and a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing IRS1, p-A KT-Ser473, AKT, γH2AX, H2AX, and BIN1 protein expression in scramble and plko- shBin1 groups with or without Z-Mirin treatment (50 μM, 6 h); n = 3 per group. Figure 7. (A) Representative WB images and a bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing G6PD and BIN1 protein expression in different groups of primary hippocampal neurons. n = 3 per group. (B) A bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing the relative transcript levels of G6pdx in different groups of primary hippocampal neurons. Transcripts were examined by qRT-PCR with Actb as an internal reference. n = 6 per group. (C) A bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing the NADPH levels in scramble and plko- shBin1 groups; n = 6 per group. ( D) A bar graph (Mean + SEM) overlaid with dot plots (individual data points) showing the NADPH levels in scramble and pGPU6- shBin1 groups with or without LiCl treatment (20 mM, 24 h); n = 3 per group. ( E) Heat maps showing representative trajectories of mice in the Barnes maze during the probe phase; bottom, line graphs showing the latencies and mistakes made .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint (primary errors) before reaching the escape tunnel during the acquisition training phase; bar graphs (Mean + SEM) overlaid with dot plots (individual data points) showing the latencies and primary errors before reaching the previous escape tunnel location during the probe phase. n = 13 mice per group. Figure 8. ( A) Left, bar graph showing mean expression levels of ISP genes including IRS1, TSC1, TSC2, and ATM in BIN1-low (counts is zero) and BIN1-high (counts is non- zero) groups; right, histograms showing distribution of cell counts across the expression levels of the corresponding genes (log2 normalized counts for IRS1, TSC1, TSC2 , and ATM respectively in each cell). ( B) similar to ( A) but plotting neuronal activity genes including ARC, BDNF, EGR1, and NR4A1. .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 21, 2025. ; https://doi.org/10.1101/2025.08.20.671271doi: bioRxiv preprint