Annexin A6 membrane repair protein protects against amyloid-induced dystrophic neurites and tau phosphorylation in Alzheimer’s disease model mice

Two lines of 5XFAD mice were used in this study. 5XFAD B6/SJL F1 hybrid on mice was bred in-house by crossing 5XFAD transgenic males to B6/SJL F1 hybrid females and genotyped as described [86]. 5XFAD mice on a C57BL/6 J background were maintained by crossing 5XFAD transgenic males to C57BL/6 J females and genotyped using 5XFAD probes at Transnetyx. APP-NLGF mice on a C57BL/6 J background were obtained from Riken (RBRC06344) and the line was maintained by crossing homozygotes. Genotyping was performed using Riken probes through Transnetyx (Memphis, TN, USA). To obtain 5XFAD;Tau^-/- tissue, 5XFAD B6/SJL F1 hybrid males were crossed to Tau^-/- females from Jackson Laboratories (strain number 007251). The resulting 5XFAD;Tau^+/- males were crossed to Tau^-/- females to obtain 5XFAD;Tau^-/- animals that were harvested at 9 months of age. Genotyping for tau was performed as described on Jackson Laboratories website and in original reference [25]. Mice were sacrificed by a lethal dose of ketamine/xylazine, followed by transcardial perfusion with 10 ml ice cold 1xPBS containing protease and phosphatase inhibitors. One hemibrain was dissected into hippocampus and cortex which were snap-frozen separately. The other hemibrain was drop fixed in 10% buffered formalin overnight at 4 °C, transferred to 20% w/v sucrose in 1xPBS for 24 hours, and then stored in 30% w/v sucrose in 1xPBS with azide. All animal work was done with the approval of the Northwestern University IACUC, assurance number A3283-01. The number and sex of animals used per experiment is described in detail in the specific methods section of the experiment.

Synapsin-GCaMP6s-ER/K linker-TdTomato (VB230218-1013vfu) and calpain sensor synapsin-Cerulean-PLFAAR-Venus (VB231102-1456esn) were generated and packaged into AAV PHP.eB by Vectorbuilder. For retro-orbital injections, mice were anesthetized in 2.5% isoflurane and 1% oxygen, and then received 100 ul of virus into the retro-orbital sinus, using a 30 gauge needle.

For calcium imaging, five 5XFAD female mice and two non-Tg littermates received 5.5x10¹¹ viral genomes of syn-GCaMP6s-ERK-TdT at 5–6 months of age and were collected for live imaging 4 weeks later at 6–7 months of age. Twenty-four hours prior to live imaging, mice received intraperitoneal injection of 100 ul of 5 mg/ml MeX04 (HelloBio) in 15% DMSO, 45% propylene glycol, and 40% PBS pH 7.5 24 hours before harvest. Mice were anesthetized with ketamine/xylazine then transcardially perfused with superchilled oxygenated sucrose artificial cerebrospinal fluid (ACSF). One hemibrain was sliced on a vibratome in chilled oxygenated sucrose ACSF solution to 300μm coronal sections. Sections were warmed to 32 °C and equilibrated into oxygenated ACSF, and then cooled to room temperature for imaging. Slices were imaged at room temperature in ASCF for no more than 30 min. Images were collected with a 25x dipping lens on a Nikon A1R Multiphoton with Chameleon Vision titanium sapphire laser tuned to 920 nm for GCaMP6s/TdTomato and 760 nm for MethoxyX04. Images were collected every 2 μm for 50–70 μm z-stacks. In FIJI, maximum projections were generated from the stacks, Regions of interest (ROIs) were drawn around cell bodies and dystrophic neurites, and the ratio of GCaMP6s intensity to TdTomato intensity calculated for each ROI. For each mouse (n = 5), 81–131 cell bodies and 407–673 dystrophic neurites from 6 to 9 Z stacks from 3 individual slices were used for quantification. Data are presented as the average for each mouse.

For calpain activity measurement, 5XFAD and non-Tg littermates (n = 3) received 1 x 10¹² viral genomes of Cerulean-PLFAAR-Venus (CFP–PLFAAR–YFP) at 7.5–9.5 months of age and were harvested 5 weeks later for immunofluorescence and biochemistry, as described above. Formalin-fixed brains were sectioned at 30 μm, and incubated with 1:15,000 of 1 mg/ml Thiazine red to label plaques. Imaging was performed on Nikon A1R confocal with a 20x NA 0.75 lens, and images captured in NIS-Elements software (Nikon). The CFP, YFP, and FRET (YFP emission wavelength with CFP excitation) intensities were quantified from single confocal images in FIJI. Three-to-four random cortical fields were used per mouse (n = 2) containing 2–11 neuronal soma and 9–64 dystrophic neurites per image for a total of 46 cell bodies and 245 dystrophic neurites.

Synapsin-A6GFP (VB200511-1403pbc), synapsin GFP (VB200603-1246qqd), and synapsin-A6 N32GFP (VB210122-1156evb) AAV plasmids were generated and packaged in AAV 8 serotype by VectorBuilder (Chicago, IL USA). Genomic titer was determined by quantitative PCR. 5XFAD transgene positive males were crossed to SJL/B6 hybrid females in timed matings to generate transgene negative and positive littermates. On P0, each pup in a litter (4 litters per virus, 22–28 pups) was cryo-anesthetized and injected with 2 µl/hemisphere containing 2 × 10¹¹ viral genomes of Syn-A6GFP, or 2 × 10¹⁰ viral genomes of Syn-GFP, to compensate for the higher expression of GFP alone. For syn-A6 N32-GFP, 2 × 10¹⁰ viral genomes/hemisphere were injected, due to known toxicity of the protein in muscle [108]. All injections were done using a using a 10 µl Hamilton syringe with a 30 gauge replaceable needle, as described in [71]. Pups were returned to the mother and aged to 4.5 months. All 5XFAD transgene positive males were used for analysis (4–6 per AAV).

Recombinant annexin A6-HIS tagged protein was generated by Evotec SE (Hamburg, Germany) contract research service for protein purification using E. coli as described in [29]. The AbVec2.0 vector was used to express A6-GFP intracellularly in HEK293 suspension cells with a 6X-Histidine tag to allow for purification via Ni-NTA chromatography. HEK293 suspension cells were grown with the FreeStyle Expression System (Gibco) and transfections performed using polyethylenimine and OptiPRO SFM (Gibco). Cells were pelleted and lysed in mild detergent M-PER (Thermofisher) containing protease inhibitor and benzonase nuclease. Protein from the lysates was purified via their His-tag using cOmplete Ni-NTA resin (Roche) and dialyzed against PBS, and then quantified on a NanoDrop (Thermofisher)

For stereotaxic injection of recombinant annexin A6-HIS and A6-GFP, 5XFAD and APP-NLGF mice were anesthetized with isoflurane (1.5–2.5%) and 1% oxygen in a stereotaxic apparatus. The lateral ventricle (B-L −0.5; M-L+/–1; D-V −2.5) or the dentate gyrus (B-L −2.5; M-L +/−2.0; D-V −2.2) and overlying cortex (B-L −2.5; M-L +/−2.0; D-V −2.0) were targeted. For A6-HIS, 5XFAD (n=5 males, 5–10 months old) or NLGF mice (n = 4, 2 males, 2 females, 9–11 months old) received 0.67 mg/kg into the hippocampus and cortex, or 6.6 mg/kg into the ventricle. Both injection targets produced similar subcellular localization of protein with different anatomic distribution, For A6-GFP, 5XFAD (n = 2 males, 11 months old) or NLGF mice (n = 3, 2 male, 1 female, 9–10 months old) received 0.45–0.9 mg/kg into the hippocampus and overlying cortex. Animals were sacrificed and tissue harvested 3–5 hours later, as described, and tissue sectioned and stained for imaging. 3–5 hours was chosen as a time frame that allowed protein to diffuse and bind to targets, but before it was degraded appreciably.

10% formalin-fixed hemibrains were sectioned at 30 μm on a freezing sliding microtome and collected in cryopreserve (30% w/v sucrose, 30% ethylene glycol in 1X PBS) then stored at – 20 °C. If required, antigen retrieval was performed for 45 min at 80 °C in 0.1M sodium citrate, in TBS, and then allowed to cool for 15 min before rinsing well and proceeding with staining. Sections were blocked in 5% normal donkey serum in TBS with 0.1% Triton-X 100 (TBS-T) and then primary antibodies added in 1% BSA in TBS-T at 4 °C overnight. Primary antibody concentrations can be found in Table 1. Following washes, sections were incubated for 2 hours at room temperature with 1:750 dilutions of Alexa Fluor 405, 488, 568, or 647 conjugated donkey anti-mouse, goat, chicken, rat, or rabbit secondary antibodies, as appropriate, along with the following stains: 300 nM DAPI, and 1:15,000 dilution of 1 mg/ml Thiazine Red (ThR) or MethoxyX04 (MeX) (HelloBio). All secondary antibodies were obtained from ThermoFisher Scientific, except donkey anti-chicken 405, 488, and 647 were from Jackson Immunologicals (West Grove, PA, USA) and donkey anti-rat 568 from Abcam (Waltham, MA USA). Sections were mounted with Prolong Gold (Molecular Probes). For protein localization, images were captured from 2–3 sections per mouse, 2–3 mice per sex and genotype, comprising male and female mice between 6 and 9 months of age, on a Nikon A1R confocal microscope using 60x objective.

For quantification of LAMP1 and p-tau181 in dystrophic neurites, and Iba1 and GFAP quantification around plaques, three sections between Bregma −1.58 mm and Bregma −2.54 mm were stained and imaged per mouse. Images were acquired on a Nikon Ti2 Eclipse widefield microscope with a 10x objective, using the NIS-Elements software high content method to capture and tile whole sections. All image acquisition settings were maintained the same between treatment groups and genotypes. For image analysis, ROIs were drawn in cortex and hippocampus. Using the General Analysis tool, thresholding was set to distinguish Aβ42, LAMP1, MethoxyX04, Thiazine Red, and p-tau181 positive regions, and then, the percent area covered by each stain from the hippocampal or cortical ROI was calculated. Using the same sections, thresholding in the General Analysis tool was used to define sub-ROIs that correspond to individual plaques having both Aβ42 and LAMP1 or p-tau181 and Thiazine red positive pixels within cortical and hippocampal ROIs. The General Analysis tool was used to measure area covered by Aβ42 and by LAMP1, or p-tau181 and ThR or p-tau181 and MeX in a given sub-ROI (plaque), and the ratio between LAMP1:Aβ42, p-tau181:ThR or p-tau181:MeX was calculated in Excel. 615–1393 plaques were analyzed per mouse in cortex and 210–527 plaques in hippocampus. Four-to-five male mice per genotype/treatment group were analyzed, using three sections per mouse.

For measurement of area of GFP-positive (GFP+) dystrophic neurites, three sections from each mouse (5 A6-GFP injected and 5 GFP injected) were imaged on a Nikon A1R confocal with a 60x objective. The section of cortex directly above hippocampus to the top of layer 5 of the cortex was imaged, and a tiled imaged of multiple fields (8 × 4) was generated to create a single image 1474.36 by 753.23 microns. Within this region, Nikon NIS-Elements software was used to threshold GFP+ structures, excluding cell bodies based on size, and the size of each GFP+ structure was measured. 1761–4941 GFP+ dystrophic neurites were measured per mouse.

30 μm floating sections of hippocampal tissue from 3 individuals with autopsy confirmed AD pathology (1 male, 2 female, ADNC high—A3, B3, C3), and 3 non-cognitively impaired age-matched controls (2 male, 1 female) were obtained from the brain bank of the Mesulam Center for Cognitive Neurology and Alzheimer’s Disease at Northwestern University. Immunofluorescence was performed as described for mice, except that sections were blocked in 10% normal donkey serum, with 2% BSA in TBS with 0.1% TBS-T and primary antibodies added over night in 2% normal donkey serum with 2% BSA in TBS-T at 4 °C. As a final step before mounting, sections were incubated 7 min in TrueBlack (#23014 Biotium) diluted 1:40 in TBS.

HEK293 cells were obtained from ATCC and cultured in DMEM with glucose, 10% fetal bovine serum, and 1% penicillin/streptomycin all from ThermoFisher scientific. Calpain 1 small subunit expression construct was obtained from Origene (RC212322) and calpain 1 large subunit was obtained from Addgene (60941). Both plasmids, along with UBC-CFP–PLFAAR–YFP from VectorBuilder, were prepared with Machery Nagel MuceloBond Xtra maxi kits and transfections performed with Lipofectamine 2000 (Invitrogen). After 24 h, cells were lysed in RIPA buffer and sonicated, and protein concentration was quantified using BCA Assay (Pierce). 10 μg of protein was boiled 10 min in 1x LDS sample loading buffer, and immunoblotting performed as described below. For FRET imaging, cells were plated on coverslips, and then, 24 h after transfection were fixed in 4% paraformaldehyde, mounted on slides, and imaged, as described previously for FRET. SH-SY5Y cells were obtained from ATCC and cultured in DMEM with glucose, 10% fetal bovine serum, and 1% penicillin/streptomycin, all from ThermoFisher scientific. For imaging, cells were plated on coverslips coated with laminin in media containing retinoic acid to induce differentiation. GCaMPs-TdTomato was transfected into SH-SY5Y cells using Lipofectamine 2000. Cells were imaged on 25x dipping lens on a Nikon A1R Multiphoton with Chameleon Vision titanium sapphire laser tuned to 920nm for GCaMP6 s/TdTomato, in Ringer’s with calcium (155 mM NaCl, 4.5 mM KCl, 5 mM HEPES, 10 mM D-Glucose, 1 mM MgCl₂, and 2 mM CaCl₂). Images were captured every 5 s for 5 min, with 90 mM KCl in Ringer’s with calcium infused in after 1 min of imaging to depolarize cells. Calcium levels were quantified using FIJI as described above, taking the ratio of GCaMP6 s to TdTomato in each cell.

Snap frozen cortices and hippocampi were homogenized in 1000 μl or 300 μl, respectively, of T-PER Tissue Protein Extraction Reagent (ThermoFisher), supplemented with protease inhibitors (Calbiochem) and Halt Phosphatase Inhibitor Cocktail (Thermo Scientific). Protein concentration was quantified using BCA Assay (Pierce). 20 μg of protein was boiled 10 min in 1X LDS sample loading buffer, and then separated on 4–12% NuPage Bis-Tris Bolt gels (FisherScientific) in MES buffer (FisherScientific). Protein was transferred to nitrocellulose membrane using BioRad Trans_Blot Turbo Transfer System, then stained with 0.1% Ponceau in 5% w/v acetic acid, and imaged. Membrane was rinsed well, blocked, and then probed with the following primary antibodies: anti-APP (6E10, BioLegend 803001, 1:2000), anti-Annexin A6 (Abcam ab 1:5000), anti-β-tubulin (TuJ1, gift of Dr. Nicholas Kanaan, 1:10,000), anti-FLAG (ProteinTech, 80010-1-RR, 1:1000), anti-GFP (Abcam ab5450, 1:1000) followed by HRP-conjugated anti-mouse, anti-rabbit, or anti-goat secondary antibody (Vector Laboratories 1:10,000). 5% milk was used as a blocking agent. Blots were visualized using chemiluminescence (BioRad Clarity), and band intensities measured using a BioRad ChemiDoc Touch Imaging System and then quantified with the ImageLab software (BioRad). Signal intensities were normalized to that of tubulin. Statistical analyses were performed as described below.

Protein ribbon diagrams were generated using Swiss-PdbViewer using solved crystal structures of human annexin A6 with phosphorylation mimicking mutant T356D (1M9I), www.rcsb.org↗.

Student’s two-tailed t test and ANOVA were done using InStat software (GraphPad Software, Inc., San Diego, CA) to compare means of the various genotypes, genders, and treatment groups. * 0.05 > p > 0.01 ** 0.01 > p > 0.001 *** 0.001 > p > 0.0001; Error bars = S.E.M.

Since we hypothesized that contact with amyloid plaques disrupts the membrane of nearby axons leading to calcium influx and subsequent disruption of kinases and phosphatases, microtubule-based transport, and ultimately neuronal function, we wanted to assess steady-state resting calcium levels in DNs. To do so, we generated a sensor fusion protein, GCaMP6s-TdT, consisting of the highly sensitive calcium sensor GCaMP6s [16] fused to calcium insensitive TdTomato (TdT), which was included to normalize for sensor protein expression (Fig. 1A). Our calcium sensor allowed us to measure steady-state resting calcium, rather than calcium change over time, as is more commonly done using GCaMP variants alone. We inserted a stiff α-helical 30 residue ER/K linker between GCaMP6s and TdT to increase the distance between the two fluorescent proteins for minimizing FRET interactions [104, 109]. We verified the calcium responsiveness of GCaMP6s-TdT by measuring GCaMP6s:TdT fluorescence ratio during depolarization. SH-SY5Y cells were transfected with the neuron-specific human synapsin-GCaMP6s-ER/K-TdT (hSyn-GCaMP6s-ER/K-TdT) plasmid, differentiated with retinoic acid, and then imaged using multiphoton microscopy while being depolarized with 90 mM KCl. We observed that the ratio of GCaMP6s:TdT fluorescence was increased by about tenfold in depolarized cells compared to non-depolarized cells (Supplemental Fig. 1A, B).

To assess resting calcium levels in vivo in DNs, we packaged syn-GCaMP6s-ER/K-TdT into AAV PHP.eB, which crosses the blood–brain barrier when injected intravenously [14, 55]. 6–7 month-old 5XFAD mice were retro-orbitally injected with syn-GCaMP6s-ERK-TdT AAV and brains collected 5 weeks later and sliced for live multiphoton microscopy imaging. To label plaques in live brain slices without relying on antibody staining, mice received a single IP injection of the brain-penetrant small molecule amyloid dye, Methoxy-X04, 1 day before harvest. DNs around dense core plaques labeled with Methoxy-X04 showed more green GCaMP6s on average than neuronal cell bodies (Fig. 1B). The GCaMP6s:TdT fluorescence ratio was quantified for individual DNs and neuronal soma in single 50–70 mm z-stack images (Fig. 1C) and as the average DN and soma ratios per mouse (5 mice, 6–9 DN and soma images each from 3 slices per mouse; Fig. 1D). Overall, DNs showed a significant increase in GCaMP6s:TdT ratio in compared to cell bodies, indicating increased resting calcium levels in DNs. Live multiphoton microscopy imaging of 5XFAD brain slices without syn-GCaMP6s-ER/K-TdT AAV injection demonstrated green- and red-channel autofluorescence in plaque cores and puncta in neuronal soma and surrounding neuropil (Supplemental Fig. 1C). We verified that the formation of DNs is not dependent upon the intraneuronal expression of human APP from the 5XFAD transgene by performing co-staining with a human-specific N-terminal APP antibody [112] and an APP antibody that recognizes both mouse and human APP (rabbit monoclonal Abcam ab32136), or an antibody recognizing BACE1, demonstrating that not all dystrophic neurites contain human APP (Supplemental Fig. 2). Rather, DNs appear to form as a response to amyloid plaques in close proximity.

Since calcium levels control the activity of the protease calpain, which is known to cleave tau potentially leading to the generation of aggregation-prone toxic fragments [11, 91], we also developed a calpain sensor to assess calpain activity in DNs. The calpain sensor construct (CFP–PLFAAR–YFP) consists of cerulean fluorescent protein (CFP) fused to mVenus yellow fluorescent protein (YFP) via a calpain cleavage site linker referred to by its amino-acid sequence, PLFAAR (Fig. 2A) [81]. To test the specificity of our calpain sensor, we expressed it in HEK293 cells with or without co-expression of the calpain 1 large and small subunits. Cleavage of CFP–PLFAAR–YFP was observed only when the calpain subunits were co-expressed, demonstrating specific detection of calpain activity (Supplemental Fig. 3A). A negative control calpain sensor with a GGGGS linker that cannot be cleaved by calpain, CFP-GGGGS-YFP, remained full length following co-expression of the calpain 1 large and small subunits (Supplemental Fig. 3A).

Next, we used our calpain sensor to assess calpain activity in vivo in DNs of 5XFAD mice. CFP–PLFAAR–YFP under control of the synapsin promoter (syn-CFP–PLFAAR–YFP) was packaged into AAV PHP.eB and retro-orbitally injected into 7-month-old 5XFAD mice and non-transgenic (non-Tg) littermates that were aged 5 weeks and brains harvested for immunoblot analysis of brain homogenates and FRET in brain sections. By immunoblot, we observed significantly increased cleavage of CFP–PLFAAR–YFP in 5XFAD compared to non-Tg mice, indicating greater calpain activity in the presence of amyloid pathology (Fig. 2B and C). To determine if increased CFP–PLFAAR–YFP cleavage occurred in DNs, we quantified FRET between CFP and YFP in neuronal soma and DNs of fixed brain sections from the same mice. First, we validated effective FRET between CFP and YFP in fixed HEK293 cells and fixed 5XFAD brains by photobleaching the acceptor (YFP) and observing an increase in fluorescence of the donor (CFP) and a decrease of FRET (Supplemental Fig. 3B and C, respectively). Finally, we determined FRET:YFP ratio (Fig. 2D, E) in neuronal soma and DNs of 5XFAD mice expressing CFP–PLFAAR–YFP, demonstrating that FRET was decreased in DNs compared to the cell body. Images depicting a scaled ratio of FRET:YFP intensity visualized differences in calpain activity between neuron cell bodies and DNs (Fig. 2D). CFP images, which showed better resolution of individual DNs and cell bodies, were used to select regions of interest for quantification. Thiazine red staining marked plaque locations, so only DNs in the proximity of plaques were selected (Supplemental Fig. 3D). Together, these results show that calpain activity is elevated in DNs, which may contribute to pathologic fragmentation of tau.

We hypothesized that DNs form due to membrane damage from contact with plaque-associated Aβ, which leads to increased calcium influx and calpain and kinase activation, as well as microtubule depolymerization, impaired axonal transport, and pathologic tau generation. If true, then enhanced membrane repair should reduce Aβ-associated DNs and pathologic tau. Annexin A6 functions to repair damage to plasma membranes in cells, especially those that undergo mechanical stress, such as muscle cells, or large cells with a high surface area, like neurons. We therefore investigated the possibility of enhancing membrane repair with annexin A6 to decrease the formation of dystrophic neurites.

First, we determined the cellular localization of endogenous annexin A6 in the brain using a specific antibody (rabbit monoclonal Abcam ab199422). We immunostained mouse and human brain sections and found annexin A6 localization in neurons, but not microglia or astrocytes, of wild-type mice (Fig. 3C) and cognitively normal aged humans (Fig. 3D). Most striking is the strong annexin A6 localization at the plasma membrane in both human and murine neurons.

Annexins A1 and A2 play secondary roles in membrane repair in muscle, adding to the repair cap after the recruitment of annexin A6 (Fig. 3B), and therefore may participate in membrane repair in neurons. To investigate this, we investigated single-cell RNA sequencing data from the Human Protein Atlas (https://www.proteinatlas.org/↗), which indicates high levels of annexin A6 in neurons. Conversely, annexins A1 and A2 are found in endothelial cells and pericytes in the brain, with very low levels in some excitatory neurons (A1) and inhibitory neurons (A2) (Supplemental Fig. 4A–C). Our immunostaining confirmed predominant localization of annexin A2 in vasculature and choroid plexus with no visible staining in neurons (Supplemental Fig. 4D, E), as previously reported [122]. Together, our results demonstrate that annexin A6 strongly localizes to neuronal membranes, and is a prime candidate for neuronal membrane repair in the brain.

The strong prevalence of annexin A6 over A1 and A2 in neurons, along with the observation that A6 localizes to sites of injury in neurons [29] and is the key factor for initiation of membrane repair in myofibers [30, 31], led us to focus on annexin A6 expression as a potential approach for reducing Aβ-associated membrane damage, DN formation, and pathologic tau generation in AD. To test whether annexin A6 could reduce DNs, we used a somatic brain transgenesis approach [71] to express syn-GFP AAV or syn-A6-GFP in the neurons of mice throughout their lifetime. On P0, we performed intracerebroventricular injections of pups with either syn-GFP AAV or syn-A6-GFP AAV8, then aged them to 4.5 months and collected brain tissue. Immunoblot analysis showed that A6-GFP was expressed ~8 fold higher than endogenous A6 (Suppl. Fig. 5 A, B), which mice tolerated well with no obvious ill effects, and protein expression was widespread through the cortex and hippocampus (Suppl. Fig. 5C). GFP fluorescence revealed A6-GFP expression in neurons and DNs (Fig. 4A). A6-GFP colocalized with BACE1 in DNs, indicating A6-GFP reached the correct location to repair potential membrane damage in DNs. AAV-mediated A6-GFP overexpression recapitulated the localization of endogenous A6, which was also found in membranes of neurons and DN (marked by BACE1 or APP) in 5XFAD transgenic (Fig. 4B) and APP-NLGF knock-in (Fig. 4C) amyloid model mice, and brains from AD patients (Fig. 4D, E) where it could play a role in amyloid-induced membrane repair.

To determine the effects of annexin A6-GFP on amyloid pathology and DN formation, brain sections from A6-GFP or GFP-alone transduced mice were stained with antibodies against Aβ42 and LAMP1, the latter recognizing lysosomal and late endosomal compartments as well as autophagic organelles that accumulate to high levels in DNs, making it a commonly used DN marker [20, 21, 46] (Fig. 5A). We used Nikon NIS-Elements software to define each plaque and its associated DN halo and determine the ratio of the respective areas stained for LAMP1 and Aβ42 (Fig. 5B). Smaller amyloid plaques grow at faster rates compared to larger plaques and have proportionally greater amounts of neuritic dystrophy [20], so we stratified plaques according to size, based on staining with an anti-Aβ42 antibody, before averaging LAMP1:Aβ42 ratios. As expected, smaller plaques had a higher LAMP1:Aβ42 ratios than larger plaques (Fig. 5B). Importantly, we observed a significant decrease of the LAMP1:Aβ42 ratio of smaller plaques in the hippocampus and cortex of 5XFAD brains injected with A6-GFP AAV compared to those injected with GFP AAV (Fig. 5B). Smaller plaques exhibit the most rapid growth and cause the greatest neurotoxicity [20], so reducing DNs around smaller plaques is likely to have beneficial effects. The distribution of plaque sizes was equivalent in A6-GFP and GFP expressing animals (Suppl. Fig. 5D). The total percent area positive for LAMP1 in the cortex was significantly decreased in A6-GFP AAV-injected 5XFAD brains, while that of Aβ42 was unchanged (Fig. 5C), indicating that A6-GFP specifically reduced DNs with minimal effects on Aβ deposits.

Using higher magnification confocal microscopy images, we observed that the area of individual GFP-positive DNs was significantly reduced in A6-GFP AAV-injected mice compared to those injected with GFP AAV (Fig. 5D). A6-GFP did not affect microglia or astrocytes around plaques, as the ratio of microglial marker Iba1 and astrocyte marker GFAP to the amyloid stain MeX04 in the area within 15μm of the plaque core was unchanged (Fig. 5E). Together, these results strongly suggest that annexin A6 overexpression reduces the formation of DNs, likely via its membrane repair function, but does not affect Aβ pathology.

Previous work has suggested that plaques and associated neuritic dystrophy play a key role in pathologic AD tau seeding and spreading [53, 72, 89]. When tau seeds purified from human AD brain are injected into amyloid mouse model brains, tau spreading, as detected by antibody AT8, positively correlates with the amount of plaques and the degree of neuritic dystrophy around plaques, with AT8 positivity appearing specifically in DNs rather than cell bodies [53, 72, 89]. CSF and plasma biomarkers p-tau181 and p-tau231 correlate significantly with Aβ pathology and predict AD onset years in advance [6, 7, 79, 107]. We therefore hypothesized that these early markers of Aβ pathology would be present in DNs. Indeed, using specific antibodies for p-tau181 (Fig. 6A, C) and p-tau231 (Fig. 6D), we observed that p-tau181 and p-tau231 accumulated in DNs of 5XFAD and APP-NLGF brains, as shown by colocalization with LAMP1 and/or BACE1. No signal for p-tau181 or p-tau231 was detected in 5XFAD;Tau^-/- brain (Fig. 6A, D). As we show here, and others have reported [92, 96], amyloid mouse models still have robust DN formation in the absence of tau, clearly indicating that tau is not required for generating DNs. Interestingly, p-tau231 and p-tau181 had different localization patterns. P-tau181 appeared in mossy fibers and DNs in both wild-type and 5XFAD mice (Suppl. Fig. 6A, B), while p-tau231 localization was more somato-dendritic (Suppl. Fig. 6C, D). These observations suggest that p-tau181 is predominantly found in axonal compartments compared to p-tau231. However, p-tau231 does appear in DNs, but less frequently than p-tau181. Moreover, these results strongly support the hypothesis that amyloid deposition leads to aberrant tau phosphorylation via the formation of DNs.

Since annexin A6-GFP overexpression reduced DNs in 5XFAD brain (Fig. 5), we hypothesized that A6-GFP would also decrease phosphorylated tau associated with DNs. Indeed, we observed a significant reduction in the ratio of p-tau181 to amyloid in 5XFAD mice injected with syn-A6-GFP AAV compared to those injected with syn-GFP AAV (Fig. 6B). We hypothesized that increased kinase activity in DNs could be responsible for p-tau accumulation. Activated, phosphorylated forms of two kinases, CaMKII and JNK, known to phosphorylate p-tau181, p-tau217, and p-tau231 [78, 117], colocalized with p-tau181 in 5XFAD DNs (Fig. 6E–G). In human AD brain, p-tau181 colocalized with total tau, BACE1, and APP in DNs around plaques (Fig. 6H–J) and p-tau 231 colocalized with p-JNK and the DN marker APP (Fig. 6K). Co-occurrence of p-tau with p-JNK and p-CaMKII in DNs does not demonstrate a physical interaction in these structures, but does suggest that DNs could be sites of early tau phosphorylation, especially p-tau181, which is used as an early CSF and plasma biomarker that indicates amyloid pathology and predicts AD [19, 65, 107].

To determine the effects of reduced annexin A6 function on DNs, we utilized a naturally occurring annexin A6 truncation (p.N32*) generated by altered splicing that introduces a premature stop codon, which results in a protein lacking the last four of the eight annexin repeats [108]. Even at low levels, A6 N32 protein exerts strong dominant-negative effects, blocking the assembly of the repair cap by preventing the recruitment of full-length annexin A6 and other repair cap components to the site of membrane damage [31, 108]. Using the P0 intracerebroventricular injections of AAV8 described previously, we overexpressed an N32 A6-GFP fusion protein in neurons of 5XFAD mice and found a loss of membrane localization compared to A6-GFP (Fig. 7A), which was also observed in non-transgenic mice (Suppl. Fig. 7). The average LAMP1:Aβ42 ratio was significantly increased in mice overexpressing N32 A6-GFP compared to those expressing A6-GFP (Fig. 7B). Additionally, the ratio of p-tau181 to amyloid as measured by MeX04 staining was significantly elevated by N32 A6-GFP (Fig. 7C). Together, the increase and decrease of DNs and p-tau181 caused by expression of N32 A6-GFP and A6-GFP, respectively, strongly suggest an important role for annexin A6 in the repair of Aβ-induced membrane damage and DN formation.

Previous work in skeletal muscle demonstrates that both intracellular annexin A6 overexpression and extracellular rA6 can enhance membrane repair [30]. Intracellular and extracellular annexin A6 promote myofiber membrane resealing after laser injury and reduce dye influx to equivalent amounts. Additionally, rA6 is also effective in acute (lytic damage from cardiotoxin injection) and chronic (muscular dystrophy due to sarcoglycan mutation) muscle injury models [30]. These results strongly suggest that rA6 could be an effective therapeutic agent, since it can home to sites of membrane damage from the extracellular space allowing exogenous administration of rA6. We previously published that exogenous rA6 localized to sites of laser damage on primary neurons [29], but we wanted to test whether extracellular rA6 would home to DNs in vivo. To do so, we administered a single stereotaxic injection of recombinant A6-HIS (Fig. 8A, B) or A6-GFP (Fig. 8C, D) into the lateral ventricle of 5XFAD (Fig. 8A,C, E) or APP-NLGF (Fig. 8B, D) mice and harvested brains 3 hours later. In the immediate vicinity of the needle track (Fig. 8E), A6-HIS strongly bound to damaged membranes resulting from the injection. Further from the injection site away from needle injury (boxed regions in Fig. 8E), immunofluorescence confocal microscopy revealed that both extracellular A6-HIS and A6-GFP localized to plasma membranes of BACE1- and LAMP1-positive DNs (Fig. 8A, B). In some cases (top panel Fig. 8A), rA6 completely coated the DN membrane, suggesting extensive externalization of phosphatidylserine, which could target the DN for engulfment and clearance by microglia or indicate cell death [101]. However, in most cases, we observed rA6 puncta more indicative of localized regions of membrane disruption. Although single stereotaxic injections of rA6 are unlikely to have long-term therapeutic effects, they demonstrate that extracellular rA6 can in principle target damaged membranes on dystrophic neurites near amyloid plaques as a proof of concept for an AD therapy. Together, our results support further investigation of extracellular administration of rA6 as a potential novel therapeutic approach to limit Aβ-induced membrane damage for the prevention of dystrophic neurites in AD, which could have beneficial effects on tau pathology.

Northwestern University filed a US provisional patent application on behalf of KRS, ARD, EMM, and RV related to the use of annexins in treating neuronal cell membrane injury in September 2022.