Electrical stimulation of the vagus nerve improves amyloid pathology in delirium superimposed on dementia

Abstract

Background

Delirium and delirium superimposed on dementia (DSD) are common complications affecting patients suffering from ongoing neurodegenerative pathologies. Peripheral surgical trauma can trigger neuroinflammation and ensuing DSD via mechanisms that remain poorly understood. Given the multifactorial therapeutic effects of neuromodulation, including vagal nerve stimulation, we have tested a minimally invasive approach to combat DSD following orthopedic surgery.

Methods

We performed orthopedic surgery on 5xFAD and CVN-AD mice and tested the efficacy of minimally invasive percutaneous vagus nerve stimulation (pVNS). We applied immunohistochemical, biochemical, and behavioral assays to evaluate the impact of surgery on postoperative delirium on DSD pathology in Alzheimer’s disease-like mice. To confirm the role of systemic factors in neuroinflammation and amyloid-β dyshomeostasis, we conducted experiments using interleukin-6 (IL-6), a cytokine commonly upregulated in postoperative delirium and in vitro co-culture assays for validation.

Results

In AD-like mice surgery induced acute changes in amyloid-β; perioperative treatment with pVNS effectively reduced amyloid-β load, plaque sphericity, and neuronal loss. The rescue of these pathological hallmarks led to improved delirium-like behavior, as demonstrated by the 5-choice serial reaction time task on postoperative days 1 and 2. pVNS improved microglial morphology, particularly near amyloid-β plaques. Acute isolation of microglial cells from 5xFAD mice after surgery indicated that pVNS partially enhanced key Disease-Associated Microglia (DAM) markers. The contribution of pro-inflammatory cytokines to amyloid-β aggregation was validated using an in vitro transwell culture model following Cytomix exposure, which also caused endothelial barrier disruption. Finally, we isolated IL-6 as a well-established biomarker of postoperative delirium and described its role in DSD pathology following systemic administration.

Conclusion

These findings establish a role for neuromodulation after pVNS in regulating perioperative immunity and advance a new paradigm for perioperative interventions in patients at risk for DSD.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s42234-025-00194-5.

Background

Delirium and dementia are two common causes of cognitive decline in older adults (Fong and Inouye 2022). Delirium is an acute confusional state diagnosed by inattentiveness, altered consciousness, and cognitive deficits that often develop suddenly after surgery and are especially prevalent in patients with pre-existing dementia. Despite the increased incidence of delirium in Alzheimer’s disease patients (Arabidze, et al., 2023), approaching 31% in patients over 60 years old (Avelino-Silva et al. 2017) with nearly two-fold greater odds of one-year mortality for patients with delirium superimposed on dementia (DSD) following orthopedic surgery (Lee et al. 2017), the mechanistic interrelationship between these two neurologic conditions is still poorly understood.

Dysregulated immunity has been highlighted as a cardinal feature of AD, with growing evidence describing changes in neuroimmune communication during aging and neurodegeneration (Bettcher et al. 2021). Immune dysregulation has been posited as a critical driver of neurodegeneration with epidemiologic studies correlating elevated systemic cytokines with AD (Lai et al. 2017; Swardfager et al. 2010) and with key mutations in immune-related genes been critically implicated in the central nervous system (CNS) inflammation (Chen and Colonna 2021). Indeed, elevated systemic cytokines can shape multiple CNS outcomes, including glial activation, neuronal dysfunction, and behavioral changes (Salvador et al. 2021). Trauma is a prototypical trigger of the innate immune response and an inseparable component of any surgical procedure that can exert profound effects on behavior and cognitive outcomes (Yang et al. 2020). Neuroinflammation is a common feature of postoperative neurocognitive disorders, with growing evidence of elevated pro-inflammatory cytokines in older adults after surgery (Wang et al. 2021). In addition, the neuroimmune and vascular interface is further impaired in DSD as a synergistic effect from surgical trauma on a vulnerable brain, which further advances barrier pathology, neuroinflammation and microglial activation with aging and neurodegeneration (Knox et al. 2022; Wang et al. 2020).

Strategies to effectively curtail inflammation in the perioperative setting are limited and hindered by multiple safety considerations. Bioelectronic medicine (the use of devices to interface with the nervous system) provides a novel means to harness neuromodulatory approaches to resolve inflammation more effectively. Indeed, the vagus nerve can engage neuronal mechanisms able to control inflammation (Jin et al. 2024; Tracey 2002). Typically, electrical stimulation of the vagus nerve (VNS) requires surgical implantation of a cuff electrode onto the cervical vagus. This approach can be challenging to use in an acute neurological complication, such as postoperative delirium. To overcome this issue, we have established a minimally invasive VNS method to selectively target the nerve via ultrasound-guided needle electrode placement in anesthetized mice (Huffman et al. 2019). Using this percutaneous vagal nerve stimulation (pVNS) method, we have identified biomarkers signifying activation of parasympathetic efferent fibers and have demonstrated engagement of the anti-inflammatory reflex, resulting in a reduction of TNFα in blood as first demonstrated by Tracey and colleagues (Borovikova et al. 2000). In the present work, we applied a common orthopedic surgical model, which often results in postoperative delirium, and describe a novel application of pVNS to prevent surgery-induced acute amyloid beta (Aβ) elevation and rescue delirium-like behavior in AD-vulnerable mice.

Methods

Animal experiment approval

All experiments were performed in strict compliance with animal protocols approved by the Institutional Animal Care and Use Committees (IACUC) at Duke University A153-23-07.

Mice

5xFAD (MMRRC stock # 34848) were purchased from The Jackson Laboratory and experimentally tested at 6-months old. APPSwDI/mNos2^-/-AD mice (CVN-AD) were kindly provided by Dr. Carol Colton (Department of Neurology, Duke University) and aged up to 12-months for experiments. All mice were on a C57 background and maintained in-house under standard housing conditions (12-hour light/dark cycle and fed ad libitum). All transgenic mice were genotyped by Transnetyx; both males and females were used (unless otherwise stated) and randomized within each experimental cohort.

Orthopedic surgery

Orthopedic surgery was performed as described (Wang et al. 2020) under sevoflurane anesthesia (Patterson Veterinary, Greeley, CO, USA) and analgesia (buprenorphine, 0.1 mg/kg subcutaneously; ZooPharm, Laramie, WY, USA). Briefly, a small incision was performed on the shaft of the tibia followed by muscle dissection and stripping of the periosteum. Pinning of the tibia was performed by inserting a 0.38-mm stainless steel rod, and osteotomy was performed on the upper crest of the bone. Surgeries were performed between 9 AM and 1 PM, and all mice were included in the study.

Percutaneous vagus nerve stimulation

pVNS was performed immediately after orthopedic surgery as previously described (Huffman et al. 2019) as part of the same anesthetic regimen presented above. Briefly, mice were maintained under 2–5% sevoflurane and placed supine on a heated platform. The fur in the ventral cervical region was removed using hair removal cream (Nair; Church & Dwight, Trenton, NJ). Under ultrasound guidance (Vevo 3100 Fujifilm, NY), a concentric bipolar needle electrode (TECA Elite Disposable Concentric Needle Electrode; Natus Neurology Incorporated, Madison, WI, USA) was positioned at the right cervical branch of the vagus nerve by visualizing the needle electrode in relation to key anatomic landmarks such as the carotid artery. In most cases, approximately 10 min were required to anesthetize the animal, prepare the area, position the needle, and identify a bradycardic response; a small number of outliers required longer (up to ~ 30 min). The vagus nerve was electrically stimulated with biphasic, charge-balanced pulses delivered at 20 Hz and 300-µs pulse width (Pulsar 6 bp-as; FHC, Bowdoin, ME). Pulse-oximetry (MouseSTAT Jr.; Kent Scientific Corporation, Torrington, CT) was applied to monitor the real-time heart rate. Bradycardia was used as a biomarker for successful vagus nerve stimulation, and stimulation amplitude was titrated to produce a 10% reduction in baseline heart rate. Animals without sufficient bradycardia were excluded from the study (< 10% were excluded). Following amplitude titration, the frequency was reduced to 10 Hz to eliminate bradycardia effects and the stimulation was maintained for 30 min as previously described in (Huffman et al. 2019). Following stimulation, the needle was retracted, and the animal was removed from anesthesia to recover. Sham mice underwent the same procedure, including needle placement, just without electrical stimulation.

Delirium-like behavior

Assessment of a delirium-like phenotype in mice with 5-CSRTT has previously been described (Yang et al. 2023). Briefly, mice were food-restricted to 90% of their free-feed weight and trained to nose-poke for a 20 mg sucrose pellet reward. Once performance was stable, mice were trained to nose-poke to a light-funnel (1 funnel out of 5) that was illuminated for 5 s and to meet a criterion of at least an 80% success rate over 3 consecutive days. Upon meeting the criterion, the illumination interval was reduced in 0.5-second increments until reaching the criterion of a 0.5-second time of illumination. Subsequently, individual mice were randomly assigned to control or pVNS treatment conditions. Once mice attained a minimum 80% success rate for 3 consecutive days, orthopedic surgery was performed the following day, and mice were randomly assigned to VNS or no treatment groups. Mice were returned to 5-CSRTT and assessed for performance over 5 consecutive days after surgery. In addition to the percent trials with a correct response (success), the latency (s) to nose poke in each trial, the percent trials where the mouse failed to respond (% Omitted Responses), and the percent trials where a mouse responded but made an incorrect choice (% Trials with Nosepoke Errors) were examined.

IL-6 injection

A separate cohort of 5xFAD mice received IL-6 50 µg kg^− 1 (Peprotech, Rocky Hill, New Jersey, USA) intraperitoneal (i.p.) injection. Animals receiving the same volume of the 0.9% saline solution served as controls. Immediately following the IL-6 injection, mice underwent 30 min of pVNS stimulation under sevoflurane anesthesia using the same stimulation parameters described above. Brain tissues were harvested 24 h after IL-6 administration for subsequent analyses.

Immunohistochemistry

One day after orthopedic surgery, mice were euthanized under deep isoflurane anesthesia, and perfused with ~ 40 mL PBS and 4% PFA. Brains were isolated and post-fixed in 4% PFA for 24 h at 4 °C, washed with PBS, and then transferred to 30% sucrose for OCT embedding. Serial sections were collected in 45-µm slices. Double- or triple-labelled staining was performed using the following protocol. Briefly, sections were washed with PBS 3 times for 5 min each, and blocked with blocking buffer solution (2.5% BSA; 0.5% Triton) for 1 h at room temperature (O'Brien, et al., 2019). Sections were subsequently incubated with primary antibodies (IBA-1, AB178846, Abcam; CD68 MCA1957, Bio-Rad; and Amyloid beta, 8243, Cell Signaling) overnight at 4 °C. The next day, sections were washed with PBS 3 times for 10 min each, and then incubated with Alexa-488- or 594- or 647-conjugated secondary antibodies (1:500; all from Invitrogen, Carlsbad, CA, USA) for 1 h in room temperature. After washing in PBS, sections were cover-slipped using Fluoroshield with DAPI (Sigma). Z-stack images were obtained using a confocal laser microscope (LSM880, Carl Zeiss, Oberkochen, Germany). X-34 labeling: Free floating sections were washed 3 times with PBS (15 min each), permeabilized for 30 min in PBS-Triton 0.5%, covered from light, incubated for 20 min in 0.01mM of X-34 in 40% Ethanol/60% PBS, washed with 40% Ethanol/60% PBS solution for 5 min, rewashed with PBS for 15 min, blocked for 1 h in blocking buffer solution, and incubated overnight at 4 °C with other primary antibodies. The next day, sections were washed 3 times with the PBS-Triton 0.5% (15 min each) and subsequently incubated for 1 h at RT with the secondary antibody diluted in the blocking buffer solution. Lastly, sections were washed 3 times with PBS (10 min each) and finally mounted on a with DAPI-free mounting media (#F6182, Sigma).

Terminal Deoxynucleotidyl transferase dUTP nick-end labeling (TUNEL)

To investigate cell death, sections were stained by TUNEL (Millipore Sigma, S7110) following the manufacturer’s protocol. Z-stack images were obtained using a confocal laser microscope (LSM880, Carl Zeiss, Oberkochen, Germany). Quantification of images was performed using Imaris (10.1.1).

Image analyses

Maximum intensity projected images were converted to 8-bit images, and the background was subtracted using ImageJ software (NIH, Bethesda, MD). Images were subjected to threshold analysis, and the percentage of area was analyzed using the “Measure” tool in Image J. For 3D-rendered images, Z-stack images were imported into Imaris to create 3D surface model. A voxel filter was applied to remove non-specific staining, and minimum thresholds were used for all groups. The surface volume and surface-to-surface colocalization volume were calculated by Imaris.

Microglial isolation

Microglia were isolated from the hippocampus using a well-established method as described in Bordt et al. (Bordt et al. 2020) for RNA extraction and qPCR. Briefly, following perfusion with ice cold saline, brains were removed and hippocampal tissue from both hemispheres was dissected from whole brains on ice using sterile forceps and minced with a razor blade. Homogenate was then submerged in Hank’s Buffered Salt Solution (HBSS; Thermofisher Scientific) containing collagenase A (Roche, 1.5 mg/ml) and DNAse 1 (Roche, 0.4 mg/ml). This was then incubated in a water bath at 37 °C for 15 min. Next, mechanical digestion of tissue was performed by sequentially passing samples through successively smaller glass Pasteur pipettes. Once a single cell suspension was obtained, samples were filtered, rinsed in HBSS, and centrifuged at 2400 rpm for 10 min at 4 °C. Samples were then incubated for 15 min with CD11b antibody conjugated magnetic beads (Miltenyi Biotec) before passage through a magnetic bead column (Quadro MACS Separator and LS columns, Miltenyi Biotec) to separate CD11b + cells (microglia) from CD11b-cells. Both CD11b + and CD11b-cell populations were washed in 1X PBS and used for RNA extraction.

RNA Extraction, cDNA synthesis, and real-time qPCR (RT-qPCR)

Isolated microglia were homogenized in TRIzol (Thermo-Fisher Scientific) and then vortexed for 10 min at 2000 rpm. After 15 min resting RT, chloroform was added (chloroform: Trizol: 1:5) and samples were vortexed for 2 min at 2000 rpm, rested at RT for 3 min, and then centrifuged (15 min at 11,800 rpm; 4 °C). From the resulting gradient, the aqueous phase was separated, and isopropanol was added to precipitate RNA (1:1 with aqueous phase). Samples were again vortexed, rested at RT for 10 min, and then centrifuged (15 min at 11,800 rpm; 4 °C). The RNA pellets were rinsed twice in ice-cold 75% Ethanol, resuspended in nuclease-free water and frozen at −80 °C until cDNA synthesis. 200ng RNA was used for cDNA synthesis using QuantiTect Reverse Transcription Kit (Qiagen) following the manufacturer’s instructions. RT-qPCR was run on a Mastercycler ep realplex (Eppendorf) using the SYBR Green PCR Kit (Qiagen). All PCR primers were designed in the lab from the Harvard PrimerBank (https://pga.mgh.harvard.edu/primerbank/) and purchased from Integrated DNA Technologies. Relative gene expression was calculated using the 2-ΔΔCT method, relative to the housekeeping gene (18 S). Samples were removed before unblinding if they failed to amplify or if there was a secondary peak in the melting temperature plot indicating contamination.

Transwell Co-Culture model

The in vitro model of systemic inflammatory factors activating the BBB and promoting microglia-mediated assembly of Aβ42 was performed in a transwell co-culture system. The bEnd.3 cells (ATCC; cat:) were cultured on ThinCert^® cell culture inserts (transparent, 0.4 μm pore diameter, 2E6 pore density, 24-well format; Greiner Bio-One GmbH; cat: 662641) above the BV-2 cells (gift of Dr. Sanjay B. Maggirwar), which were cultured on 12-mm coverslips (Neuvitro; cat: GG-12–15 H) coated with 0.1 mg/mL poly-D-lysine (PDL; Sigma; cat: P1149). Specifically, bEnd.3 cells were cultured at a density of 70k cells/insert in complete endothelial media comprised of Dulbecco’s Modified Eagle Medium (DMEM, high glucose, pyruvate, no glutamine; ThermoFisher; cat: 10313-021) supplemented with 10% v/v fetal bovine serum (FBS; Atlas Biologicals; cat: F-0500-D) and 1% v/v GlutaMAX (ThermoFisher; cat: 35050079). Barrier formation by the bEnd.3 cells was measured via TEER to ensure they approached previously reported steady state conditions of ~30Ω*cm² (Bordt et al. 2020; Hung et al. 2022). The same day that bEnd.3 cells demonstrate monolayer conditions, BV-2 cells were seeded onto the PDL-coated coverslips at a density of 45k cells/coverslip in complete microglial media comprised of the same DMEM supplemented with 10% v/v FBS, 1% v/v GlutaMAX, and 1% v/v antibiotics (penicillin + streptomycin; ThermoFisher; cat: 15070063). After 2 h, the BV-2 are replenished with fresh complete microglial media, and the next morning the bEnd.3 transwells are assembled above the BV-2 coverslips and allowed to acclimate for 3 h in reduced-serum conditions (replenished with complete media containing 1% v/v FBS instead of 10% v/v).

Systemic cytokine profile induction of endothelium in transwell Co-Culture model

After the 3-hr acclimation period for the bEnd.3 monolayers and BV-2 cultures were treated according to the conditions outlined in Table 1. Specifically, bEnd.3 monolayers were either given a cytokine cocktail (Cytomix) or saline vehicle (veh_cyto) while BV-2 cells were either exposed to amyloid monomers or a DMSO vehicle (veh_Aβ). A concentrated stock solution of Cytomix was prepared using recombinant murine cytokines for IL-1β (PeproTech; cat: 211-11B), TNF-α (PeproTech; cat: 315–01 A), and IFN-γ (ThermoFisher; cat: 315-05) in 0.9% NaCl saline solution at a density of 0.5 µg/mL for each cytokine. This stock was then diluted in reduced serum complete endothelial medium to 3 ng/mL. The corresponding vehicle treatment (veh_cyto) was prepared using an equal volume dilution (0.6% v/v) of saline in reduced serum complete endothelial medium. The amyloid monomer treatment is diluted from a stock solution of 1 mM Aβ42 (AnaSpec; cat: AS-24224) reconstituted in DMSO to a final concentration of 2.5 µM in reduced-serum complete microglial media. The corresponding vehicle treatment (veh_Aβ) was prepared by diluting DMSO (0.25% v/v) into reduced-serum complete microglial media.

Table 1.

Treatment groups for transwell co-culture system

Group	bEnd.3 Treatment	BV-2 Treatment
A	vehcyto (0.6% v/v saline)	vehAβ (0.25% v/v DMSO)
B	vehcyto (0.6% v/v saline)	2.5 µM Aβ42 monomers
C	3 ng/mL Cytomix	vehAβ (0.25% v/v DMSO)
D	3 ng/mL Cytomix	2.5 µM Aβ42 monomers

Transwell Co-Culture model immunocytochemistry

After the 24-hr incubation period, the transwell co-culture system was briefly washed with ice-cold 1x DPBS, followed by fixation with 4% PFA in 1x PBS solution for 12 min at room temperature with gentle agitation on an orbital shaker. This was followed by a wash with 100 mM glycine solution and two subsequent washes with 1x PBS. The transwell inserts and coverslips were then separated for individual processing for immunostaining and imaging. Both were permeabilized with 0.15% v/v Triton-X in 1x PBS for 15 min at room temperature under gentle agitation, followed by three washes with 1x PBS. The cells are then blocked with 5% bovine serum albumin (BSA; Sigma; cat: A8412) and 0.05% Triton X-100 in 1x PBS for 1 h at room temperature under gentle agitation. The cells were immunolabeled with the following antibodies in Table 2. Due to overlapping primary hosts, the BV-2 cells underwent an initial tertiary staining protocol, followed by a standard secondary staining protocol. For the tertiary staining, the BV-2 cells were first labeled with the amyloid fibril (OC) antibody in staining solution (2% BSA with 0.025% Triton-X in 1x PBS) at 4°C overnight with gentle agitation. This was followed by washes with 1x PBS, a 1-hr incubation at room temperature with gentle agitation in the mono-Fc secondary antibody in staining solution, another set of washes with PBST (1x PBS + 0.1% Tween), then final labeling with the complementary tertiary antibody containing the fluorophore in staining solution for 1 h at room temperature with gentle agitation. After washing with PBST, the BV-2 cells were briefly fixed again with 4% PFA for 3 min, followed by the same glycine and 1x PBS wash steps above. The remaining targets were immunolabeled following a standard secondary protocol where a primary antibody cocktail with all the remaining antibodies combined in staining solution is incubated at 4 °C overnight with gentle agitation. The same washing and secondary staining conditions are followed as previously outlined, except the secondary staining solution now contains all the fluorophore-conjugated antibodies. The bEnd.3 monolayers are immunolabeled with the same secondary protocol, followed by removal of the membranes from the transwell insert for mounting and imaging. The coverslips and membranes are mounted using Prolong™ Diamond Antifade Mountant with DAPI (for bEnd.3; ThermoFisher; cat: P36962) or without DAPI (for BV-2; ThermoFisher; cat: P36961) and allowed to cure overnight before imaging on a structured illumination microscope system.

Table 2.

Antibody panel for immunocytochemical analysis of transwell co-culture model

Panel for bEnd.3 Staining
Primaries					Secondaries
Host	Target	Dilution	Source	Catalog	Host	Fluor	Dilution	Source	Catalog
Hm	PECAM-1	1:200	Thermo	MA3105	Gt	AF488	1:750	Jackson	127–545-099
Rt	VCAM-1	1:200	BD Bio	553,330	Gt	AF647	1:750	Thermo	A-21,247
Panel for BV-2 Staining
Primaries					Secondaries
Host	Target	Dilution	Source	Catalog	Host	Fluor	Dilution	Source	Catalog
Ms	6E10	1:200	Biolegend	803,014	Gt	AF488	1:750	Thermo	A-11,001
Rbt	A11 Oligomer	1:200	StressMarq	SPC-506	Gt	AF568	1:750	Thermo	A-11,036
Chk	IBA-1	1:200	Synaptic Sys	234,009	Gt	AF405	1:750	Thermo	A-48,260
Rbt*	OC Fibril	1:200	StressMarq	SPC-507	*Tertiary Stained, see next line for next two stains
Gt	Rbt IgG	1:40	Jackson	111-007-003	Dk	AF647	1:500	Thermo	A-21,447

Structured illumination imaging

A previously reported structured illumination scheme used by the Gelbard lab was applied to acquire immunofluorescent data on the transwell co-culture model. An Olympus BX51 microscope is equipped with a Prior Lumen 200 Hg lamp illumination system (Prior cat: LM200B1-A) and an OptiGrid structured illumination optical element (Qioptiq). Emission from the immunolabeled samples was detected with a Hamamatsu ORCA-ER scientific camera. The following pairs of excitation and emission filters were used: 350 nm/DAPI (Semrock; cat: FF02-447/60 − 25); 405 nm/FITC (Semrock; cat: FF01-524/24–25); 488 nm/TRITC (Semrock; cat: FF01-593/40 − 25); 568 nm/Cy5 (Semrock; cat: FF01-692/40 − 25). An infinity-corrected 20x UPlanApo 0.70 NA objective (Olympus) was used to capture z-stacks at 1 μm step sizes, which were compressed into the extended focus view shown in the representative images used in this manuscript. The exposure time and gain for each channel were optimized and kept constant across samples.

Transwell Co-Culture model image analysis

The acquired z-stacks were analyzed using Volocity 3DM Quantitation Software (Quorum Technologies). Objects were selected for “true” immunoreactivity for the labeled target using the “Find Objects” function to remove background artifacts that do not meet the threshold criteria of n standard deviations above mean fluorescence intensity, where n ranges from 1.5 to 3 standard deviations depending on the channel. For amyloid analysis, true amyloid was determined as 6E10 + objects that were co-incident with IBA-1 + objects. These true amyloid populations were then screened for double positivity with either protofibril formation (OC+) or oligomer assembly (A11+). For all confirmed objects, the total intensities and total volumes for the positive objects were obtained for each image stack and exported for statistical analysis. For DAPI-stained samples (BV-2 cells), the total number of unique DAPI objects was counted as a proxy for cell count.

Statistics

Investigators were blind to the experimental group until analyses were completed. Behavioral data were analyzed using SPSS 29 (IBM SPSS Statistics, Chicago, IL). Baseline measures were examined with t-tests to compare treatment effects and to determine if a difference between groups exists, if the percent change in response post-surgery relative to baseline was required. For behaviors measured across sequential days, Repeated Measures Analysis of Variance (RMANOVA) was used, with test days as the within-subjects effect, and the two treatment conditions as the between-subjects effect; Bonferroni-corrected pairwise comparisons were used to determine differences in treatment effects at each time point. In all cases, homogeneity of variance was intact as assessed by Levene’s Test for Equality of Error Variances. No violations of homogeneity were detected within the data, and no corrections were warranted. For all other data, GraphPad Prism 10 (GraphPad Software, San Diego, CA, USA) was used for statistical analysis. For these data, comparisons between groups were assessed using an ordinary one-way analysis of variance (ANOVA) followed by Tukey’s multiple-comparisons post hoc tests. For the ICC and TEER data from the transwell co-culture samples, a full-interaction two-way ANOVA was performed to determine group effects due to Cytomix induction and Aβ42 monomer presence, but not the cross-term effects of the combo treatments. As such, a Fisher’s test was used to determine statistical significance. In all cases, data are presented as mean ± SEM, and statistical significance was defined as p < 0.05.

Results

pVNS prevents acute Aβ aggregation after surgery

We tested the effects of pVNS following orthopedic surgery as a translational model relevant to postoperative delirium and AD patients since they are at greater risk for experiencing falls and fractures (Buchner and Larson 1987; Curtis et al. 2024; Ruggiero et al. 2024; Zhou et al. 2023a). Stimulation of the vagus nerve (bipolar, biphasic, symmetrical rectangular pulses, 0.3 ms pulse-width, 10 Hz) was performed over 30 min immediately following orthopedic surgery (Fig. 1A). Using 5xFAD mice, we found that orthopedic surgery induced a rapid increase of both Aβ (pan-marker) and X-34 (core plaque marker) at 24 h, which was prevented by perioperative vagus nerve stimulation (Fig. 1B, C, and D). Further analyses and 3D reconstructions of the X-34 plaques (a lipophilic fluorescent derivative of Congo red) revealed morphological changes in plaque sphericity following surgery. These changes in volume and sphericity of the X-34 plaques were rescued by pVNS treatment not only in the hippocampus (DG area) but also in the prefrontal cortex (Fig. 1E and F).

Fig. 1.

Effects of surgery and pVNS on amyloid pathology. A Schematic illustration of pVNS in the tibia fracture model. pVNS was delivered for 30 min immediately following orthopedic surgery. B Representative images of Aβ (magenta) and X34 (blue) immunostaining in the hippocampus 24 h after surgery. Scale bar: 200 μm and 50 μm (high mag). C-D Quantification of amyloid (% area) in the hippocampus. E Representative images of X34 (blue) and 3D reconstruction by Imaris. F quantification of X34 sphericity. Scale bar: 50 μm and 7 μm (3D). Data are presented as mean ± SEM and analyzed by ordinary one-way analysis of variance with Tukey’s multiple comparisons post-hoc tests, *p < 0.05; **p < 0.01; n = 5–7/group. Abbreviations: C: control, Sx: tibia surgery, pVNS: percutaneous vagus nerve stimulation

pVNS rescues DSD and surgery-induced neuronal apoptosis in AD-like mice

Next, we used the 5-choice serial reaction time task (5-CSRTT) to assess inattention and vigilance in mice (Robbins 2002), which are key features of patients with delirium (Marra et al. 2018). Before surgery, all mice averaged 29 test trials (29.2 ± 1.2 trials), and there was no difference between the two treatment conditions. Although 5xFAD male mice had achieved ≥ 80% success rate in the 5-CSRTT before surgery and were randomly assigned to either pVNS treatment or a surgical control group (without pVNS) on the day of surgery, the final examination of baseline performance (averaged success rate over 3 days) before surgery showed a small, yet significant difference between the control and pVNS treatment groups (surgery: 80.67 ± 1.11%, surgery + pVNS: 87.58 ± 2.53%, t_(1,14) = 2.949, p = 0.011). The success rates of each mouse in the 5-CSRTT test during postoperative days 1–5 were normalized to their individual baseline rates. We found that mice receiving pVNS following surgery showed significantly less decline in 5-CSRTT performance on postoperative days 1 and 2 compared to mice that received surgery alone (Fig. 2A). Across the 5 days post-surgery the mice averaged 26–27 trials each day, (26.3 ± 1.0–27.4 ± 0.9 trials/day). RMANOVA found no differences between the two treatment conditions in the number of trials administered. In addition, the latency(s) to nosepoke during testing was delayed in surgery-controls relative to the pVNS mice (Sup Fig. 1A). Treatment with pVNS significantly reduced errors of omission in surgical animals (Sup Fig. 1B). There was also a trend for the pVNS mice to commit fewer errors during testing (Sup Fig. 1C). Together, these data indicate that pVNS treatment protects 5xFAD mice from a delirium-like phenotype in 5-CSRTT.

Fig. 2.

Effects of surgery and pVNS on delirium-like behavior and acute neuronal apoptosis. A Schematic of the 5-CSRTT apparatus to study inattentiveness. On postoperative days 1 and 2, mice in the surgery alone group showed significantly impaired performance in the 5-CSRTT test (percent successful choices over days normalized to baseline performance) compared to pVNS-treated mice. Data was analyzed by RMANOVA, *p < 0.05 compared to Sx + pVNS group at the same time point; n = 5–11/group. B Representative images of TUNEL assay (green) with NeuN (magenta) in the dentate gyrus (DG). Scale bar: 20 μm and 5 μm (high mag). C Quantification of volume of TUNEL⁺ in NeuN⁺ nuclei at 24 h after surgery. Data are presented as mean ± SEM and analyzed by ordinary one-way analysis of variance with Tukey’s multiple comparisons post-hoc tests, *p < 0.05; ***p < 0.001; n = 6/group. Abbreviations: C: control, Sx: tibia surgery, pVNS: percutaneous vagus nerve stimulation

Using a separate cohort of 5xFAD mice, we evaluated cell death by TUNEL staining. We found increased DNA fragments in neurons at 24 h, suggesting an increased acute neuronal apoptosis following orthopedic surgery, and this acute apoptosis was reduced by pVNS (Fig. 2B and C). Although delirium usually resolves within days following surgery, we investigated whether this acute event could further drive neurodegeneration in the 5xFAD model. At 1 month after surgery, we observed no differences between groups for markers of neurodegeneration (Aβ and X-34 plaques) and neuronal apoptosis (TUNEL, Sup Fig. 2).

Regulation of microglial dyshomeostasis following surgery and pVNS in 6-months old 5xFAD and 12-months old CVN-AD mice

Neuroinflammation is a key hallmark of AD, with microglia playing active roles both in neurodegeneration as well as delirium pathology (Hansen et al. 2018). Here we focused on changes in microglial morphology (Iba-1, Fig. 3A, B) and activation to a phagocytic phenotype (CD68, Fig. 3A, C). Treatment with pVNS reversed the surgery-induced microglial reactivity, especially near Aβ plaques, as noted in the 3D image reconstructions (Fig. 3A, D). To further interrogate the functional state of the microglia after surgery and pVNS, we used beads-selected CD11b + microglia extracted from 5xFAD mice and evaluated the expression of disease-associated microglial (DAM)-like markers from hippocampal tissue (Sup Fig. 3). The expression of several transcripts, including transcription factor EB (TFEB), TMEM119, MERTK, Clec7a, and TREM2, was reduced following surgery, suggesting alterations in the profile of these cells, which in turn may contribute to the changes in plaque compaction and clearance (Sup Fig. 3). Treatment with pVNS restored some of these transcripts to near-control levels, especially TMEM119, a key gene also found in human microglia, and TFEB, a master transcriptional regulator of lysosomal biogenesis and autophagy implicated in Aβ clearance (Gu et al. 2022).

Fig. 3.

pVNS regulates microglia activation after surgery in 5xFAD. A Representative images of X34 (blue), Iba-1 (red), and Cd68 (green) immunostaining in the hippocampus 24 h after surgery, with 3D image reconstructions. Scale bar: 50 μm and 10 μm (arrows indicate the 3D reconstructed images inset in the last panel). B-C Quantifications of CD68 and Iba-1 are presented as percentage of the area. D Quantification of the volume of Iba1 in contact with X-34 from Imaris. Data are presented as mean ± SEM and analyzed by ordinary one-way analysis of variance with Tukey’s multiple comparisons post-hoc tests, *p < 0.05; **p < 0.01; ***p < 0.001; n = 5–7/group. Abbreviations: C: control, Sx: tibia surgery, pVNS: percutaneous vagus nerve stimulation

Similar to the 5xFAD mice, we also used CVN-AD mice, a slow-progressing AD-like mouse model that exhibits DSD at 12 months of age (Wang et al. 2020), and found that pVNS treatment was also effective in rescuing surgery-induced Aβ accumulation and microglial morphological changes, including in close proximity of blood vessels (Fig. 4A-C). Taken together, pVNS impacts microglial function in AD-vulnerable mice, which may contribute to the compaction and clearance of aberrant amyloid deposition after surgery.

Fig. 4.

Regulation of postoperative Aβ and microglia in older CVN-AD mice. A Representative images of Aβ (red) and Iba-1 (green) in the CVN hippocampus after surgery. Scale bar: 50 μm. B-C Quantification of the percentage of area in hippocampus. Data are presented as Mean ± SEM, and analyzed by ordinary one-way analysis of variance with Tukey’s multiple comparisons post-hoc tests, *p < 0.05; **p < 0.01; *** p < 0.001; n = 8/group. Abbreviations: C: control, Sx: tibia surgery, pVNS: percutaneous vagus nerve stimulation

Peripheral IL-6 is sufficient to trigger acute Aβ dyshomeostasis, which pVNS can rescue

To validate the influence of systemic factors as putative drivers of Aβ pathology in DSD, we exposed 5xFAD mice to IL-6. IL-6 is a prototypical cytokine elevated in patients with delirium (Taylor et al. 2023) and also increased in DSD following orthopedic surgery (Wang et al. 2020). We injected IL-6 at a dosage comparable to the circulating levels detected in mice with postoperative delirium (Hu et al. 2022). IL-6 was able to induce Aβ-labeled plaque deposition, measured by X-34, 24 h following intraperitoneal administration (Fig. 5A – C). This effect is comparable to tibia fracture surgery-induced Aβ deposition (Fig. 5B, C). Treatment with pVNS, as performed in the surgical model, significantly reduced levels of Aβ and Iba-1 in the hippocampus compared to the non-treated group (Fig. 5D-F). These findings indicate that peripheral IL-6 is sufficient to drive amyloid accumulation and microglial activation previously observed following orthopedic surgery, and treatment with pVNS was effective in ameliorating the CNS pathology.

Fig. 5.

pVNS prevents Aβ changes following systemic IL-6 exposure. A Schematic of the IL-6 intraperitoneal (i.p.) injection in 5xFAD mice. B-C Representative images of X34 in the hippocampus and quantification showing comparable levels of X34 following IL-6 intraperitoneal (i.p.) injection or orthopedic surgery. Scale bar: 20 μm. D Representative images of Aβ (red) and Iba-1 (green) in 5xFAD hippocampus after IL-6 injection and pVNS treatment (30 min of pVNS stimulation immediately after intraperitoneal injection). E-F Quantification of Iba-1 and Aβ in the hippocampus. Scale bar: 50 μm. Data are presented as Mean ± SEM, and analyzed by ordinary one-way analysis of variance with Tukey’s multiple comparisons post-hoc tests, *p < 0.05; **p < 0.01; *** p < 0.001; n = 4–6/group. Abbreviations: C: control, Sx: tibia surgery, pVNS: percutaneous vagus nerve stimulation, IL6 i.p.: interleukin-6 intraperitoneal injection

Systemic inflammatory factors lead to amyloid aggregation, microglia activation, and brain endothelial cells disruption

To reverse translate the impact of systemic inflammation to the AD vulnerable brain we used an in vitro co-culture model in a transwell system involving immortalized murine cell lines for brain microvascular endothelial cells (bEnd.3) and microglia (BV-2) (Fig. 6A). To mimic a systemic inflammatory profile, a cytokine cocktail (Cytomix) containing matching dosages of IL-1β, TNF-α, and IFN-γ (3 ng/mL) was applied to the apical surface of the transwell system and incubated for 24 h. Given the appreciable levels Aβ in control conditions of 6-month-old 5xFAD mice, 2.5 µM of Aβ₄₂ monomers were added contemporaneously to the basolateral compartment containing the BV-2 cells (Fig. 6B). These conditions were chosen based on previous work done demonstrating the role of each cytokine in linking delirium to neuroinflammation in a systemic inflammatory model (Cibelli et al. 2010; Terrando et al. 2010; Zhou et al. 2023b); the selected cytokine dosages of the combined treatment are reflective of previously reported additive, synergistic effects between the cytokines to exacerbate BBB dysfunction (Anastasiou et al. 2021; McHale et al. 1999). In fact, exposure to Cytomix and Aβ₄₂ was able to disrupt bEnd.3 endothelial monolayer as measured by transendothelial electrical resistance (TEER) 24 h following incubation (Fig. 6C).

Fig. 6.

Cytokine activation of bEnd.3 monolayers drive microglia-mediated assembly of amyloid in transwell co-culture model of BBB-microglia crosstalk. A Schematic illustration of transwell model system and associated treatments and assays. Monolayers of bEnd.3 cells were formed on transwell inserts and assembled over BV-2 cells cultured on coverslips with a 3 h acclimation period in reduced serum conditions. Co-cultures were treated overnight with a combination of Cytomix with Aβ42 monomers and corresponding vehicle control combinations (as outlined in Table 1). Changes in barrier health were assayed via TEER B with complementary immunocytochemical analysis of microglial activation and assembly of Aβ42 into higher-order assemblies (C-E). (B) TEER measurements of barrier integrity and activation are represented as % change compared to steady-state monolayer values. C Representative immunofluorescence images of BV-2 cells with or without Aβ42 monomers after bEnd.3 treatment with or without Cytomix. ΙΒΑ−1 (blue), 6E10 (green), OC (cyan), A11 (magenta). Scale bar 15 μm. D Quantification of microgliosis based on IBA-1 reactivity. E-H Quantification of Aβ42 species based on total amyloid content (6E10 reactivity) and respective unaggregated (6E10 solo reactivity), fibrillar (6E10 OC double reactivity), and oligomeric (6E10 A11 double reactivity) fractions. I Representative images and quantification of Aβ (magenta), Iba1 (green), and Vcam1 (red) immunoreactivity in the hippocampus of 5xFAD mice 24 h after orthopedic surgery vs. controls. Scale bar: 200 μm and 20 μm (high mag). J-L Quantification of Aβ, Iba1 and Vcam1 presented as percentage of the area. Data presented as mean ± SEM and all quantifications analyzed for statistical significance using a two-way ANOVA on a full-effects model and Fisher’s LSD. *p < 0.05; **p < 0.01; ***p < 0.001; n = 3/treatment or unpaired t-test (J-L). **p < 0.01; n = 6/group. Abbreviations: Veh: vehicle, C: control, Sx: tibia surgery

We next applied immunocytochemistry to corroborate our in vivo findings of systemic inflammation driving microglia activation (Fig. 6D) and evaluated in more details the corresponding conversion of amyloid monomers into higher order assemblies (Fig. 6E – H). While the presence of Aβ₄₂ monomers demonstrates a marginal increase in Iba-1 immunoreactivity in veh_cyto treated bEnd.3 monolayers, we only observed morphologic evidence of microgliosis in BV-2 cells cultured basolateral to bEnd.3 cells exposed to Cytomix (Fig. 6D). This correlates well with the observed diminution of bEnd.3 monolayer health in Cytomix treated groups as well as previous reports of Aβ₄₂ monomers exhibiting minimal inflammatory activation of microglia compared to amyloid aggregates (Gouwens et al. 2016). Correspondingly, when comparing the BV-2 cultures that were exposed to Aβ₄₂ monomers, only those that were cultured basolateral to bEnd.3 monolayers with Cytomix were able to drive assembly of higher order amyloid fibrils (6E10 + OC + double immunoreactivity) and oligomers (6E10 + A11 + double immunoreactivity) that result in reduced expression of unaggregated fractions (6E10 + solo reactivity) (Fig. 6E – H). These differences are unlikely to be attributed to differences in Aβ₄₂ content in the cultures given the equimolar treatments, as well as no differences in total amyloid reactivity (6E10 reactivity, Fig. 6E). We corroborated these results by extracting proteins from the basolateral BV-2 cultures and performed pulldown enrichment of amyloid and downstream analysis via western blot and dot blot assays (Sup Fig. 4). Of note Cytomix induced ~ 20% increase of amyloid fibrils in BV-2 cells cultured basolateral to veh_cyto treated bEnd.3 monolayers. This may be due to amyloid aggregation in solution (Paranjape et al. 2013) that favors protofibril formation in the brain vulnerable to DSD (O’Nuallain et al. 2010). Despite the constraints of our model, we nevertheless found a 75% increase in induction of fibril-like assembly in the Cytomix-induced bEnd.3 and amyloid-exposed BV-2 transwell co-cultures.

Finally, using 6-month-old 5xFAD mice we found similar changes in Aβ and Iba1 expression following surgery-induced systemic inflammation (as already shown in Fig. 1), but these features were also paralleled by induction of vascular cell adhesion molecule 1 (VCAM1) (Fig. 6I – J) suggesting a putative role for endothelial cell activation and amyloid aggregation both in vitro and in vivo.

Discussion

Characterization of key molecular and functional features of delirium, especially in vulnerable subjects with ongoing neurodegeneration, remains elusive, thus limiting effective interventions. Here, we demonstrated a novel application for minimally invasive vagus nerve stimulation in postoperative delirium and ensuing DSD. We confirmed that acute Aβ accumulation, microglial dyshomeostasis, and neuronal loss coincide with the transient nature of DSD in AD-like mice, and pVNS was effective in preventing the onset of this brain pathology.

VNS has established clinical efficacy and important clinical implications that can extend beyond the treatment of chronic autoimmune disorders and epilepsy (Koopman et al. 2016; Tynan et al. 2022). These potential therapeutic pathways of neuromodulation may translate well to clinical treatment for neurodegenerative conditions, including DSD and AD (Murdock et al. 2024). Using preclinical models of DSD we found that surgery induced acute changes in Aβ aggregation, which was effectively prevented by perioperative pVNS. This acute increase in postoperative amyloid burden may have direct translational implications as Aβ burden was found to correlate with postoperative delirium severity, albeit this report was a small case-control study (Torres-Velazquez et al. 2022). An independent association between cerebrospinal fluid (CSF) Aβ₄₂ levels and delirium incidence was reported in patients who underwent primary elective hip or knee arthroplasty (Cunningham et al. 2019), suggesting that targeting Aβ may relieve secondary CNS complications in patients at risk for developing delirium after surgery. Indeed, other acute stressors, including traumatic brain injury and sepsis, were also shown to increase Aβ and impair neuronal functions (Basak et al. 2021; Danielson et al. 2021; Giridharan et al. 2023; Scott et al. 2016). The features of Aβ changes following acute stressors, including surgery, are likely distinct from the development of classical Alzheimer’s dementia that occurs over decades. In fact, we did not observe long-lasting effects on neurodegeneration 1 month after surgery, suggesting the changes in Aβ aggregation are acute and likely immune-driven, which is also supported by our in-vitro assays. Cytokine-like effects of Aβ have been described where low concentration of amyloid can synergize with pro-inflammatory cytokines to trigger neurotoxicity (LaRocca et al. 2021) as well as further cytokine production, including IL-6 and IL-8 (Gitter et al. 1995). Indeed, low-grade systemic inflammation has also been shown to influence the onset of AD pathology, including elevating Aβ accumulation and DAM transcript expression (Guerrero-Carrasco et al. 2024). These bidirectional mechanisms still require further elucidation, both for AD and DSD.

VNS has well-established anti-inflammatory effects via cholinergic regulation of pro-inflammatory signaling in splenic macrophages (Andersson and Tracey 2012; Rosas-Ballina et al. 2011; Wang et al. 2003) and neuroimmune modulation, including cell trafficking in the spinal cord (Natarajan et al. 2024). Afferent fibers of the vagus nerve detect pro-inflammatory signals such as IL-6, IL-1β and TNF-α that are released from injured or infected tissues. These signals are relayed to the nucleus tractus solitarius in the brainstem, which integrates the input and initiates a reflexive efferent response. Even though the vagus nerve does not directly innervate the spleen, efferent fibers pass the signals through the celiac ganglia and the superior mesenteric ganglion to the spleen and activate the splenic sympathetic nerves (Andalib et al., 2023). These nerves release norepinephrine, which binds to β2-adrenergic receptors on CD4⁺ T cells in the spleen. Suppression of inflammation following activation of the “cholinergic anti-inflammatory pathway” is mediated by this vagus-splenic signaling, which links norepinephrine binding to β2-adrenergic receptors of CD4 + T cells with acetylcholine release on macrophages, thereby inhibiting TNFα production in the spleen (Rosas-Ballina et al. 2011). Using this surgical model, we previously described the contributions of inflammatory cytokines, including TNFα and IL-6, in driving BBB disruption (Terrando et al. 2011; Yang et al. 2019), suggesting that inflammatory cytokines can directly contribute to barrier impairment, endothelial cell activation, and microglial reactivity (Miller-Rhodes et al. 2022). Importantly, the vagus nerve can counter-regulate inflammatory responses and initiate resolution through engagement of several neuroimmune axes (Jin et al. 2024). We previously reported that pVNS engages the “cholinergic anti-inflammatory pathway”, thereby reducing systemic levels of TNF⍺ and rescuing microglial activation after endotoxemia (Huffman et al. 2019). Indeed, elevation of systemic inflammatory biomarkers is well described in patients with delirium, including a robust association with plasma levels of IL-6 (Taylor et al. 2023; Wang et al. 2021). Targeting body-wide, systemic factors can be challenging due to the increased risks of off-target effects, especially in postoperative recovery, where cytokines are crucial for host defense and tissue healing outside the CNS (Yang et al. 2020). It is possible that by engaging the “cholinergic anti-inflammatory pathway”, as previously established with this method (Huffman et al. 2019), circulating levels of pro-inflammatory cytokines, including IL-6, are insufficient to activate endothelial cells (i.e., VCAM-1) and impair the BBB after surgery. Indeed, impairments in barrier properties are often described in the AD brain and may contribute to the dysfunctional neuroimmune crosstalk in neurodegenerative conditions (Da Mesquita et al. 2018; Sweeney et al. 2018). Alternatively, pVNS may also shift the microglial activation state through afferent pathways to brainstem nuclei, allowing for better compaction of Aβ plaques and reduced neurotoxicity ( Condello et al. 2015), thus conferring protection from DSD onset. Microglia regulation of plaque morphology and deposition has been described in AD-like mice, mediating both CNS pathological features and progression of cognitive deficits (Ennerfelt et al. 2022, 2023; Huang et al., 2021). When microglia are dysfunctional, plaque equilibrium is adversely affected, and the amount of dense-core plaques is reduced in APP/PS1 AD mice (Huang et al. 2021b). The specific mechanisms for pVNS-mediated Aβ reduction and microglial regulation require further investigation, and its multimodal effects on multiple biological functions are likely contributing to the neuroprotection in DSD.

Overall, the opportunity to regulate CNS outcomes by bioelectronic approaches emphasizes a novel treatment strategy for postoperative complications like delirium and lays the groundwork for future studies in aging and neurodegeneration.

Conclusion

Postoperative neurologic complications are common, especially in patients with pre-existing neurodegeneration. Here, we demonstrated a novel protective effect from vagal nerve stimulation in preventing delirium superimposed on dementia and highlighted a key contribution from peripheral inflammation in driving acute changes in amyloid-β deposition and neuroinflammation in AD-vulnerable mice.

Abbreviations

5-CSRTT

5-choice serial reaction time task

AD

Alzheimer’s disease

Aβ

amyloid beta

BBB

blood-brain barrier

C

Control

CNS

central nervous system

CSF

cerebrospinal fluid

DAM

disease-associated microglia

DSD

delirium superimposed on dementia

i.p.

intraperitoneal

IL-6

interleukin-6

Neun

neuronal nuclear antigen

pVNS

percutaneous vagus nerve stimulation

Sx

surgery (tibia fracture)

TFEB

transcription factor EB

TNF⍺

tumor necrosis factor-alpha

TUNEL

terminal deoxynucleotidyl transferase dUTP nick end labeling

veh

vehicle

Authors’ contributions

CS, PYW, WJH, JD, RV performed the main experiments. PYW and WJH performed the pVNS. CS, JD, AB processed, imaged and analyzed the pVNS experiments in the 5xFAD mice. TY supervised analyses on the 5xFAD. RMR, AN and WCW designed, performed, and analyzed the behavioral assay. DR and SDB designed, performed and analyzed the microglial isolation experiments. WC and JO performed the in vitro assays. NT conceived the idea and supervised the study with input from HAG, TY, and WMG. NT wrote the manuscript with WJH, JD, RMR and input from CS, SB, HAG, TY, and WMG. All authors read and approved the manuscript.

Funding

National Institutes of Health grants R01-AG057525, R01-AG083979A1, RF1-AG079138 and R21-AG055877, Alzheimer’s Association (2019-AARG-643070 and AARF-24-1313412).

Data availability

The dataset supporting the conclusions of this article is included within the article as additional files.

Declarations

Ethics approval and consent to participate

All experiments were performed in strict compliance with animal protocols approved by the Institutional Animal Care and Use Committees (IACUC) at Duke University A153-23-07.

Consent for publication

Not applicable.

Competing interests

WJH, WMG, NT are inventors on US Patent 11,083,892. No other disclosures were reported.