Abstract
The amyloid precursor protein (APP) is linked to the genetics and pathogenesis of Alzheimer's disease (AD). It is the parent protein of the β-amyloid (Aβ) peptide, the main constituent of the amyloid plaques found in an AD brain. The pathways from APP to Aβ are intensively studied, yet the normal functions of APP itself have generated less interest. We report here that glutamate stimulation of neuronal activity leads to a rapid increase in App gene expression. In mouse and human neurons, elevated APP protein changes the structure of the axon initial segment (AIS) where action potentials are initiated. The AIS is shortened in length and shifts away from the cell body. The GCaMP8f Ca2+ reporter confirms the predicted decrease in neuronal activity. NMDA antagonists or knockdown of App block the glutamate effects. The actions of APP on the AIS are cell-autonomous; exogenous Aβ, either fibrillar or oligomeric, has no effect. In culture, APPSwe (a familial AD mutation) induces larger AIS changes than wild type APP. Ankyrin G and βIV-spectrin, scaffolding proteins of the AIS, both physically associate with APP, more so in AD brains. Finally, in humans with sporadic AD or in the R1.40 AD mouse model, both females and males, neurons have elevated levels of APP protein that invade the AIS. In vivo as in vitro, this increased APP is associated with a significant shortening of the AIS. The findings outline a new role for the APP and encourage a reconsideration of its relationship to AD.
SIGNIFICANCE STATEMENT While the amyloid precursor protein (APP) has long been associated with Alzheimer's disease (AD), the normal functions of the full-length Type I membrane protein have been largely unexplored. We report here that the levels of APP protein increase with neuronal activity. In vivo and in vitro, modest amounts of excess APP alter the properties of the axon initial segment. The β-amyloid peptide derived from APP is without effect. Consistent with the observed changes in the axon initial segment which would be expected to decrease action potential firing, we show that APP expression depresses neuronal activity. In mouse AD models and human sporadic AD, APP physically associates with the scaffolding proteins of the axon initial segment, suggesting a relationship with AD dementia.
Introduction
Most models of Alzheimer's disease (AD) pathogenesis focus on a small peptide known as β-amyloid (Aβ). Aβ is the main constituent of the plaques found in the AD brain and is derived by proteolytic cleavage of a large Type I membrane protein known as the amyloid precursor protein (APP) (Goldgaber et al., 1987; Kang et al., 1987; Tanzi et al., 1987). The link to AD is compelling as mutations near the cleavage sites on APP that release Aβ have been identified as autosomal dominant AD disease genes. Further, as APP is located on human chromosome 21, individuals with Down syndrome make supranormal amounts of APP and show early-onset clinical and pathologic signs of Alzheimer's-like dementia (Ness et al., 2012).
Yet while much is known about the Aβ peptide, the normal function(s) of the full-length precursor protein remain little studied. APP is an evolutionarily ancient protein with analogs found, even in invertebrates such as nematodes and fruit flies (Zheng and Koo, 2006). It is expressed in tissues throughout the body (Uhlen et al., 2015), including brain. Its mRNA is found in astrocytes and neurons, but it is equally abundant in oligodendrocytes and endothelial cells (Zhang et al., 2014). APP expression begins at the pre-implantation stage of development (Fisher et al., 1991), and several developmental roles have been suggested. For example, during CNS development, mis-regulated expression of APP leads to defective cortical neuron migration (Young-Pearse et al., 2007). APP is found at the growth cone and mediates axon guidance (Kibbey et al., 1993; Rama et al., 2012; Wang et al., 2017). As neurons begin maturation, APP helps guide synapse formation (Priller et al., 2006; Klevanski et al., 2015), and promote dendritic spine formation (Lee et al., 2010). In the mature organism, APP mediates many fundamental cellular processes (Muller et al., 2017). The large extracellular domain of APP, known as secreted APP, has a variety of functions (Mattson et al., 1993; Furukawa et al., 1996; Mockett et al., 2017). The intracellular portion of APP (known as the APP intracellular domain), similar to the homologous portion of the Notch protein, translocates to the nucleus, although its function there is not well understood (Cao and Sudhof, 2001; Gao and Pimplikar, 2001). Finally, APP is a damage response protein. Multiple studies have shown that, in response to stress, the levels of APP in neurons increase, particularly at a site of axonal injury (Gentleman et al., 1993; Sherriff et al., 1994; Miyai et al., 2021; Muñoz et al., 2021).
We report here a new biological function for APP. We find that APP is a dynamic modulator of the axon initial segment (AIS), a specialized cellular compartment found at the junction between the neuronal soma and axon. The AIS is the site of action potential initiation as well as a barrier that helps maintain neuronal polarity (Colbert and Johnston, 1996; Jenkins and Bennett, 2001; Hedstrom et al., 2008; Kole et al., 2008; Ogawa and Rasband, 2008; Nelson and Jenkins, 2017; Huang and Rasband, 2018). Through changes in its length and cellular location, the AIS modulates firing probability and thus can fine-tune neuronal activity (Song et al., 2009; Grubb and Burrone, 2010a; Kuba et al., 2014). We show here that, after an excitotoxic challenge, APP levels rise in the AIS, concurrent with changes in its length and cellular location. We further show that APP overexpression alone is sufficient to alter AIS position and length. Finally, the AIS changes are observed in APP overexpressing transgenic mice, as well as in human sporadic AD, yet are independent of the presence of Aβ plaques. These findings offer a possible new connection between the symptoms of AD and the normal function of APP.
Materials and Methods
Human samples
Postmortem brain tissue was obtained from the National Institutes of Health NeuroBioBank at the University of Maryland, Baltimore, MD, the Human Brain and Spinal Fluid Resource Center, and the Mount Sinai/James J. Peters VA Medical Center, National Institutes of Health Brain and Tissue Repository, with approvals from Tissue Access Committee of NeuroBioBank. Subjects with AD and age-matched unaffected controls (UCs) were requested; any cases with >10 h postmortem interval were excluded. Both UC (Braak Stage 0-II) and AD (Braak Stage V-VI) female and male samples between the ages of 76 and 82 were used in this study. All tissues were from left frontal cortex (Brodmann area 9) that had been frozen without prior fixation and stored at −80°C. Tissues were cryosectioned at 10 μm and stored at −80°C until use.
Animals
A colony of R1.40 transgenic mice (B6.129-Tg (APPSw)40Btla/Mmjax) was established from mice originally purchased from The Jackson Laboratory. The R1.40 colony is maintained on the C57BL/6J background. Wild type littermates served as controls. Both genders of the mice were used in this study. Two females and 2 males of each genotype were examined. Initial experiments were conducted in Hong Kong. These were conducted with under regulations established by the Government of Hong Kong SAR and were approved by the Animal Ethics Committee of the Hong Kong University of Science and Technology. Colonies were maintained and bred in the Animal and Plant Care Facility at Hong Kong University of Science and Technology, and all animal procedures complied with university guidelines and the Government of Hong Kong Department of Health.
Additional animals used in this study were derived from stock maintained and bred in Pittsburgh in the facility maintained by the Division of Laboratory Animal Resources, University of Pittsburgh, School of Medicine. For animals used in Pittsburgh, all protocols were approved by the Institutional Animal Care and Use Committee. Animals were treated in compliance with the Institute for Laboratory Animal Research of the National Academy of Sciences' Guide for the care and use of laboratory animals.
Genotyping for the R1.40 transgene locus was done with the recommended PCR primers using the PCR Ready Mix kit (E3004; Sigma-Aldrich). The sequences of the genotyping primers were as follows: R1.40 transgene forward: 5′-CTT CAC TCG TTC TCA TTC TCT TCC A-3′; R1.40 transgene reverse: 5′-GCG TTT TTA TCC GCA TTT CGT TTT T-3′; internal control forward: 5′-CAA ATG TTG CTT GTC TGG TG-3′; internal control reverse: 5′-GTC AGT CGA GTG CAC AGT TT-3′.
Primary culture
The culture surface of each well was coated with poly-L-lysine (0.5 mg/ml, diluted in boric acid buffer, Sigma). Wild type E16.5 C57BL/6J embryos were harvested from gravid females. After decapitation, their cerebral cortices were isolated and stored at 4°C in PBS with 1 mg/ml glucose. Cortices were cut into small pieces using forceps and treated with 0.25% trypsin-EDTA (Sigma) for 10 min at 37°C. The tissue was then washed in 10% FBS in DMEM to inactivate the trypsin, then washed in Neurobasal medium (Invitrogen), and triturated to produce a single-cell suspension. Cells were plated onto poly-L-lysine-coated coverslips in 24-well plates at a density of 48,000 cells/well. The medium to maintain the neurons was Neurobasal medium supplemented with 2% B27 (Invitrogen), 1% Glutamax (Invitrogen), and 10,000 U/ml penicillin/streptomycin (Invitrogen). Cultures were maintained at 37°C in a humidified atmosphere of 5% CO2. Every 5 d, the culture medium was refreshed by replacing half of the old medium.
Human neuron induction
An iPS line was prepared from a healthy, 23-year-old female subject of European ancestry. Cryopreserved lymphocytes were reprogrammed to iPSC using Sendai-viral-expressed transcription factors (Oct3/4, Sox2, Klf4, and c-Myc; CytotuneTM-iPS Reprogramming Kit, Fisher Scientific) by RUCDR Infinite Biologics (now Sampled). Pluripotent cells were selected by colony morphology, alkaline phosphatase positivity, and immunocytochemical expression of Oct4 and TRA-1-60. The pluripotency of resulting iPSC was subsequently confirmed by colony morphology and immunocytochemistry for Oct4 and Tra-1-60 markers (Popova et al., 2023). To generate glutamatergic neurons, iPS cultures were infected with lentiviruses carrying the neuronal transcription factor Neurogenin2 (Ngn2) and a reverse tetracycline-controlled transactivator, in mTeSR medium (Stem Cell Technologies) with 1.5 μm ROCK inhibitor (Y27632, Tocris), as described previously (Zhang et al., 2013). The following day, the medium was replaced with Neurobasal containing B27 supplement (Fisher Scientific), GlutaMAX, 2 μg/ml doxycycline, and 5 μm ROCK inhibitor. Twenty-four hours following induction, transduced cells were selected with 1 μg/ml puromycin with doxycycline. Four days later, on day 5, cells were dissociated with Accutase (Stem Cell Technologies) and replated onto a monolayer of primary mouse glial cells on glass coverslips. The medium was transitioned to Neurobasal Plus medium containing B27 Plus supplement and CultureOne supplement (Fisher Scientific), with GlutaMAX and 200 mm ascorbic acid and replenished every 3-4 d. To inhibit glial overgrowth, 2 μm cytosine arabinoside was used as needed. Treatment and morphologic analyses were conducted 4-5 weeks following initial induction with doxycycline.
Plasmids
A construct encoding human APP bearing the Swedish mutation (APPSwe) was kindly provided by Amy Fu of Hong Kong University of Science and Technology. The APP-GFP construct with human APP and C-terminal GFP tag was provided by Hei-man Chow at Hong Kong University of Science and Technology. APP shRNA (pENTR APP shRNA 2) was acquired from Addgene (Plasmid #30135) and recloned into a lentiviral plasmid. FUGW-APPSwe virus (Addgene 190804) and synapsin-driven GCaMP8f lentivirus plasmid were provided by Ronald Hart.
Transfection
Transfection was performed using Lipofectamine LTX with plus reagent kit (Fisher Scientific A12621). Each plasmid was diluted to 50 ng/µl in Opti-MEM medium and incubated for 5 min at room temperature. Then 2 µl of lipofectamine-LTX was added to a separate 50 µl of Opti-MEM. The two portions of Opti-MEM were mixed for a final volume of 100 µl and incubated for 20 min at room temperature before adding to the neurons. After 5 h of incubation, the medium was changed back to conditioned medium. Neurons were cultured in vitro for 7 d (DIV7) in a 5% CO2 atmosphere at a density of 24,000 cells per well of a 24-well plate with medium half-replaced once with fresh NeuroBasal medium at DIV5. The medium was changed from NeuroBasal complete to NeuroBasal without antibiotics at a final volume of 400 µl. The conditioned medium was collected and stored at 4°C after which the 100 µl mixture of DNA/lipofectamine-LTX was then added without disruption into one well of the 24-well plates. The neurons were incubated for 4 h and then changed back to their previously conditioned medium after it had been prewarmed. The neurons were cultured at 37°C in a 5% CO2 atmosphere for 2 more days before fixation or initiation of other treatments.
Tissue processing
R1.40 mice and littermate controls at 18 months of age were used for histologic analysis. Animals were anesthetized by a 1:1 mixture of 10 mg/ml ketamine (Covetrus) and 2 mg/ml xylazine (Covetrus) at a dose of 0.02 ml/g. The mice were then perfused with 2 ml 0.9% saline at a flow rate of 10 ml/min. The brains were dissected free of the skull, and the cortex was bisected on the midline. The left hemisphere was frozen on dry ice and stored at −80°C until used for immunoblotting. To prepare tissue for histology, the right hemisphere was postfixed by immersion in 4% paraformaldehyde for 24 h. After postfixation, the brains were cryoprotected by incubation in 30% sucrose in PBS for 24 h. The brains were then embedded into OCT blocks for cryosectioning at 10 µm, in either the coronal or sagittal plane. Sections were stored in −80°C until processed for immunohistochemistry or immunofluorescence.
Immunochemistry
For tissue culture cells, after fixation and washing with PBS, cells were permeabilized with 0.5% Triton X-100 in PBS (PBST). Cells were blocked for 1 h at room temperature with PBST containing 5% donkey serum then incubated overnight at 4°C with primary antibody in PBST at the dilutions listed in Table 1. The next day, cells were washed in PBS and incubated with secondary antibody diluted at 1:500 in PBST for 1 h at room temperature, followed by staining with DAPI for 5 min. The coverslips were then mounted in Hydromount and viewed on a fluorescence microscope (Olympus DP80) or on a confocal microscope (Nikon Eclipse Ti2).
For both human and mouse brain tissue, cryosections were immersed in citrate buffer (pH 6.0) and heated at 95°C for 10 min for antigen retrieval. After cooling and rinsing, the sections were incubated in 10% donkey serum at room temperature for 1 h to reduce nonspecific antibody binding followed by overnight incubation at 4°C with the primary antibody. The primary antibodies used, along with their sources, are listed in Table 1. After rinsing, the sections were incubated in AlexaFluor-conjugated secondary antibody for 1 h at room temperature. Cell nuclei were counterstained with DAPI. For each biological repeat, at least three sections were processed using the above-mentioned method and three ROIs per section were imaged for quantification. The ROIs were chosen by first identifying the region where the measurements were to be taken at low power using only the DAPI channel. Once selected, a higher magnification was chosen at random from the low-power field, an image captured with the camera in all fluorescent channels, and the resulting image used for the quantifications described. Fluorescent imaging was done on a Nikon Eclipse Ti2 Confocal microscope with a 60× (1.4 Numerical Aperture) oil objective lens and A1R (Andor, 512 µm × 512 µm) camera, X-Cite 120 LED illumination system, and a DAPI (400-418 nm), FITC (450-490 nm), TRITC (554-570 nm), and Cy5 (640-670 nm) filter sets controlled by NIS-Elements AR software.
RNA isolation and qPCR
DIV14 neurons treated with 1 μm glutamate for the times indicated in the text then harvested using the QIAGEN RNeasy Mini kit (QIAGEN, 74104) according to the manufacturer's instructions. For each sample, an equal amount (500 ng) of RNA was reverse transcribed to cDNA using iScript reverse-transcription kit (Bio-Rad, 1708841). qRT-PCR (45 cycles: LightCycler480 software) was performed using fast SYBR Green Master Mix (Bio-Rad 1725122). GAPDH served as a loading control. The sequence of App primers are as follows: forward, ACTCTGTGCCAGCCAATACC; reverse, GAACCTGGTCGAGTGGTCAG.
Stochastic optical reconstruction microscopy (STORM)
For STORM imaging, cells were incubated with the primary antibody at 4°C overnight, and then with Alexa647-conjugated and CF568-conjugated goat-anti-rabbit/mouse secondary antibody for 2 h at room temperature. STORM imaging was done on a custom-built STORM imaging system in the Liu laboratory based on an Olympus IX71 inverted microscope frame with a 100×, NA = 1.4 oil immersion objective (UPLSAPO 100XO, Olympus), as described previously (Xu et al., 2018). For two-color STORM imaging, 30,000 frames were acquired at an exposure time of 20 ms per frame for each color (642 and 561 nm). Online drift correction was independently performed every 200 frames (∼4 s) with fluorescent beads (Fisher Scientific, F8803) as fiduciary markers (excited with 488 nm laser) throughout the image acquisition process, based on our established method (Ma et al., 2017). In contrast to the periodic appearance of βIV-spectrin in super-resolution images of the AIS (Xu et al., 2013), the pattern we observed using our AnkG antibody is more uniform. This is to be expected as the antibody was raised against the C-terminal amino acids of the ankyrin G protein. This portion of AnkG is explicitly not periodic (Leterrier et al., 2015).
Image analysis and fluorescence intensity quantification
Fiji software and MATLAB were used to quantify the cellular and AIS fluorescence intensity. For cellular fluorescence intensity, freehand line tool was used to select the outline of a cell body with MAP2 staining as counterstained ROI. An average of three background regions with the same area as a given cell body stain was used as control for background staining and subtracted from each image. Raw Integrated Density was used as measurement for the cellular intensity.
Western blots
Cultured cells were harvested and homogenized in ice-cold RIPA buffer (EMD Millipore) with 1× PhosSTOP phosphatase inhibitor mixture (Roche Applied Science) and 1× complete protease inhibitor cocktail (Roche Applied Science). The homogenate was then centrifuged at 4°C for 20 min at 15,000 rpm. The supernatant containing protein was collected. Human brain samples (200 mg from Brodmann area 9 [BA9]) were acquired from Mount Sinai and the NIH NeuroBioBank. A fraction of each specimen was homogenized using the protocol of Nelson et al. (2020). Briefly, the tissue was homogenized in buffer containing 8 m urea, 5% SDS, and 5 mm N-ethylmaleimide in a 1.5 ml microcentrifuge tube then hand-dounced in 10 volumes/weight of 65°C homogenization buffer. The homogenate was next incubated at 65°C for 20 min. After the incubation, the sample was mixed 1:1 with 5× PAGE buffer (5% (w/v) SDS, 25% (w/v) sucrose, 50 mm Tris, pH 8, 5 mm EDTA, bromophenol blue). The lysates were stored at −80°C until use. Protein concentration was determined using the Bradford assay (Bio-Rad). A total of 10 μg protein was then separated by SDS-PAGE and transferred to Immuno-Blot PVDF membranes (Bio-Rad). Membranes were incubated in 5% nonfat milk for 1 h at room temperature to block nonspecific binding. Primary antibodies were applied at room temperature overnight. After rinsing 3 times with TBST, membranes were incubated with secondary antibodies for 1 h at room temperature. Signals were visualized with SuperSignal West Pico, Dura, or Femto chemiluminescent substrate (Fisher Scientific). For the 480 kDa AnkG immunoblotting gel, a high molecular weight protein standard was used (Fisher Scientific LC5699). For all other gels, the protein standard used was a normal wide range from 5 to 250 kDa, Precision Plus Protein Dual Color Standards (Bio-Rad 1610374).
Coimmunoprecipitation
Wild type and R1.40 brains were homogenized in a Dounce homogenizer. The lysates were then incubated in coimmunoprecipitation buffer (20 mm Tris HCl, pH 8.0, 137 mm NaCl, 1% Nonidet P-40 (v/v) and 2 mm EDTA) on ice for 30 min. The brain lysates were incubated with primary antibody and protein G magnetic beads overnight at 4°C with moderate shaking. Then the beads were washed with coimmunoprecipitation buffer 3 times and eluted with 1× SDS-PAGE loading buffer. The eluted proteins were then run on SDS/polyacrylamide gels and subjected to Western blotting.
Aβ fibril and oligomer preparation
Beta amyloid (Aβ1-42) aggregation kit was purchased from rPeptide (A-1170-2). To produce fibrillar Aβ, the Aβ powder was dissolved in NeuroBasal medium to a final concentration of 220 μm, followed by incubation for 5 d at 37°C. To prepare Aβ oligomers, 22.2 μl DMSO was added to 0.5 mg of HFIP-treated Aβ1-42 (A-1163-1) to a final concentration of 5 mm. The Aβ and DMSO mixture was then vortexed for 30 s to make sure all the Aβ powder was dissolved in DMSO followed by a short centrifugation. The solution was sonicated for 10 min in a bath sonicator followed by the addition of 1087 μl cold 4°C phenol-red-free F-12 medium to achieve a final concentration of 100 μm. The Aβ solution was then vortexed for 15 s and incubated at 4°C for 24 h. Afterward, Aβ was centrifuged at 14,000 × g for 10 min at room temperature. The supernatant was then collected for use.
Statistical analysis
Data are expressed as mean ± SEM for each group. The statistical significance of changes in different groups was evaluated by unpaired Student's t test using GraphPad Prism software. p values <0.05 were considered to be significant.
Results
Activity increases APP levels and causes AIS length and position to change
APP is recognized as a stress response protein. Following oxidative stress, excitotoxicity, or exposure to free radicals, APP protein levels increase (Frederikse et al., 1996; Masliah et al., 1997; White et al., 1999). This is relevant to the situation in AD as excitotoxicity is cited as an important risk factor (Greenamyre and Young, 1989). Indeed, blunting the effects of excitotoxicity is the theoretical mechanism of action of the FDA-approved AD drug, memantine (Johnson and Kotermanski, 2006). To test this association of APP with excitotoxicity in vitro, we used E16.5 mouse cortical neurons cultured for DIV14 and exposed them to a relatively low concentration of glutamate (1 μm). We found that, in treated but not untreated cells, the intensity of neuronal APP immunostaining increased noticeably 24 h following glutamate addition, although MAP2 immunostaining revealed no major structural changes in dendritic integrity (Fig. 1A–D). The increased APP fluorescent signal was obvious in the cell body as well as in the dendrites and was blocked by preincubating the neurons with 20 μm dCPP, a selective NMDAR antagonist (Monnet et al., 1995; Costa et al., 2009).
While the quantitative increase in staining was evident throughout the cell, we also observed a qualitative change in the axon and AIS. We identified the axonal process and the location of the AIS by immunostaining with the scaffolding protein, ankyrin G (AnkG, red signal, Fig. 1A–D, white arrows). In untreated cells, the axon had low, but detectable, amounts of APP immunoreactivity (Fig. 1A, green signal). After a 24 h treatment with 1 μm glutamate, however, APP infiltrated the axon hillock and the AIS (Fig. 1C). Using ImageJ software, we found that the average APP fluorescent signal intensity increased more than twofold in the AIS of treated neurons (Fig. 1E). This expansion of the domain of APP staining was found in virtually every neuron in our cultures. Although the initial assays were performed after 24 h, the APP response to the glutamate challenge was far more rapid. We exposed DIV14 neuronal cultures to 1 μm glutamate, harvested the cells after varying lengths of time, and performed qPCR on the cell lysates. After 30 min of stimulus, the level of App mRNA started to rise and within 1 h was increased threefold to fivefold (Fig. 1J). This is consistent with the documented role of APP as a stress response protein.
In addition to the increased level of APP in glutamate-treated neurons, the properties of the AIS, as revealed by the pattern of Ankyrin G staining, also changed. Its length was significantly reduced, and the distance between the cell body and the proximal end of the AIS was significantly increased (Fig. 1F,G). These changes were mediated by glutamate directly, acting through the NMDAR. A 30 min pretreatment of the cultures with dCPP completely blocked the AIS changes as well as the increase in APP protein (Fig. 1H,I). One plausible interpretation of these findings is that the activity caused by glutamate led to an increase in APP and that it was the increased APP that led to the changes in the AIS. The AIS changes in turn would be expected to increase the threshold for initiating an action potential (Grubb and Burrone, 2010a; Evans et al., 2013; Muir and Kittler, 2014).
Overexpression of APP is sufficient to change the AIS configuration
To test the idea that APP regulates the configuration of the AIS, we asked whether, in the absence of a glutamate stimulus, increased APP alone would be sufficient to produce similar effects. We cotransfected APP-encoding and GFP-encoding plasmids into DIV7 cultured neurons. Control cultures were transfected with only a GFP plasmid. Two days later, the cells were fixed and immunostained for MAP2 and either βIV spectrin or AnkG (Fig. 2A–C). Transfected neurons had elevated APP that infiltrated all processes, including the axon. Consistently, in APP-expressing neurons, the length of the AIS was reduced by almost one-half compared with control. The distance from the cell body to the proximal end of the AIS also increased (Fig. 2B). Thus, APP overexpression alone, in the absence of glutamate stimulation, is sufficient to cause the AIS to shift away from the cell body and reduce in length.
The AIS in AD: in vitro data
That APP overexpression can result in significant changes in the AIS raises the question of whether the elevated levels of APP that are found in AD have similar effects and thus possibly contribute to the symptoms of dementia. We tested this idea first by comparing the properties of the AIS in neurons transfected with either the WT human APP (hAPP) or the well-characterized Swedish variant, K595N/M596L (hAPPSwe), an autosomal dominant disease gene that causes early-onset AD. We found that cells transfected with hAPPSwe underwent a shrinkage of the AIS that was nearly 50% greater than that found with wild type (WT) hAPP (Fig. 2A–C). Neither WT hAPP nor hAPPSwe overexpression caused visible damage to the transfected neurons. At 2 d after transfection, the appearance of the MAP2-immunostained dendritic structure was not notably different from neurons expressing only the control GFP vector.
We were concerned that the levels of APP protein after transfection might be artificially high, so much so that they would create a cellular state that did not reflect normal physiological conditions. We therefore cultured cortical neurons from brains of the R1.40 mouse AD model, whose hAPPSwe transgene is expressed at only 2-3 times the levels of the endogenous mouse App gene (Lamb et al., 1993, 1997). This is close to the increase in mouse App expression observed following a glutamate challenge (Fig. 1J). We found that the response of the AIS, even with this more modest increase in APP expression, was still significant. In R1.40 neurons, axonal APP immunostaining invaded the axon hillock as well as the AIS, and the length of the AIS was reduced compared with neurons cultured from WT animals of the same litter (Fig. 2D,E). Compared with the APPSwe-transfected neurons, the reduction of the length of the R1.40 AIS was less dramatic (Fig. 2E). This suggests that the APP-driven shortening of the AIS is both dose- and genotype-dependent. Intriguingly, the APP effect is seen even during neuronal maturation. We used βIV spectrin to track the appearance of the AIS as a function of time in culture. In WT neurons, the AIS begins to form in the growing axons after DIV4. In R1.40 neurons, by contrast, the AIS does not appear until DIV6 (Fig. 2F).
To obtain a more fine-grained picture of the interplay of APP with the components of the AIS, we used STORM. The resulting high-resolution images showed that, in WT mouse cultures, endogenous APP (stained with an APP N-terminal antibody) gave a weak signal in the AIS but was clearly visible in the cell body (Fig. 3A). In R1.40 neurons, however, APP staining was stronger and could be found entering the axon and the region of the AIS. The impression gained when comparing this pattern with that of the AnkG scaffold protein was that APP was “pushing” the AIS away from the cell body. Quantification of the relative signal intensity of the two antibodies was consistent with this idea (Fig. 3B).
We next asked whether the AIS scaffold proteins physically interacted with APP. We prepared lysates from frozen brain samples of UC and AD patients (Brodmann area 9). We performed immunoprecipitations with AnkG or βIV spectrin antibodies, or with a KO-validated APP N-terminus antibody (APP-N′), and then probed Western blots of the precipitates with an AnkG antibody (Fig. 3C). Both AnkG and βIV spectrin antibodies pulled down the expected three species of AnkG (Fig. 3C, top and bottom gel, lanes 2 and 3). Immunoprecipitation of Alzheimer's brain lysates sample using the 6E10 and APP antibody also pulled down AnkG (Fig. 3C, bottom gel lane 4). A similar result was found with lysates from UC tissue, but the strength of the AnkG band was considerably reduced (Fig. 3C, top gel, lane 4). To further explore the details of this interaction, we quantified the amounts of AnkG brought down by APP immunoprecipitation, normalizing the results to the amount brought down by AnkG itself. We found that the APP pulled down nearly twice as much AnkG from the AD brain as from UC tissue. This was true for both the 480 and 270 kDa AnkG (Fig. 3D). Curiously, while the 190 AnkG isoform was present in the APP immunoprecipitates, its levels were largely unchanged between the AD and UC samples. Thus, in human AD brain lysates, the physical interaction of AnkG and APP is more extensive than in control lysates. The same pattern of association was found with lysates from WT and R1.40 mouse brains (data not shown). Together, these results demonstrate that APP, a membrane protein implicated in AD, is colocalized and physically associated with an important scaffolding protein of the AIS.
The data thus far show that upregulation of APP, induced either genetically or by a glutamate stimulus, is sufficient to change the length and location of the AIS. We next asked what were the effects of reducing the level of APP on the structure of the AIS. We cotransfected cultured cortical neurons with a plasmid encoding APP shRNA plus a GFP plasmid (to identify transfected cells) 48 h before treating them with 1 μm glutamate. We monitored the effectiveness of the shRNA knockdown in two ways. First, we measured the intensity of APP immunostaining in transfected (GFP-positive) compared with nontransfected cells and found it to be reduced by >80% (Fig. 3E,F). Second, we infected sister cultures with shAPP or scrambled control, lysed the cells, and ran Western blots for total APP protein (Fig. 3I). The intensity of the APP band was reduced relative to the loading control (Fig. 3I).
To see the effect of APP knockdown, we immunostained the AIS with AnkG. In control cells (transfected with a scrambled shRNA), after exposure to 1 μm glutamate, the length of the AIS was reduced and its distance from cell soma increased (Fig. 3E,G,H) just as was seen in nontransfected cells (not shown). By contrast, in cells transfected with APP shRNA, exposure to 1 μm glutamate did not significantly change either the length of the AIS (Fig. 3H) or its distance from the cell soma (Fig. 3G). APP shRNA alone induced a modest, but nonsignificant, increase in the length of the AIS and had no effect on the soma to AIS distance. These data suggest that APP is required for the glutamate-dependent changes in AIS structure.
Enhanced APP expression would be expected to increase APP proteolytic processing and hence to increase levels of the Aβ peptide. This raised the possibility that the effects of APP expression were indirect and mediated through its breakdown product, Aβ. To test this idea, fibrillar Aβ [Aβ (f)] and Aβ oligomers [Aβ (o)] were prepared from commercially available peptide by standard protocols (Mehta et al., 1998; Zhu et al., 2019). The level of aggregation was verified by Western blot using the 6E10 antibody as a probe (Fig. 4A). Both types of Aβ preparations were then used to treat DIV14 neuronal cultures. After 24 h, the cells were fixed and immunostained for AIS scaffold proteins and synaptic protein markers. Neuronal number remained unchanged after Aβ treatment, even while the density of synapses was reduced (Fig. 4B). Despite the loss of synapses, however, the length of the AIS remained unaffected in both the Aβ (f) and Aβ (o) treated cultures (Fig. 4C,D). The distance of the proximal end of the AIS from the cell soma likewise showed no significant change (Fig. 4E).
To determine whether the mouse neurons we were using for our Aβ assays accurately reflected the situation in the human brain, we used human iPSC-derived neuronal cultures to test the effect of Aβ on the dimensions of the AIS of human neurons. Four to 5 weeks after neuronal induction (see Materials and Methods), the mature human iPSC-derived neurons were treated with Aβ (o) for 24 h followed by immunostaining for the neuronal marker, TuJ1, and the AIS marker, AnkG (Fig. 4F). In untreated cultures, the average length of the AIS (∼30 µm) was similar to the length of the AIS in cultured mouse neurons (∼35 µm). The distance from the cell soma to the most proximal portion of the AIS was slightly greater in human neurons (∼15 µm) compared with mouse primary neurons (∼11 µm), but remarkably similar given the differences in the two preparations. The similarities carried over into the Aβ response. Neither the length of the AIS (Fig. 4G) nor its distance from the soma (Fig. 4H) was altered in the human neurons by the application of Aβ. These data suggest that the alterations of the AIS are caused by the overexpression of APP directly and not indirectly through increased levels of Aβ.
The AD brain has evidence of increased calpain protease activity (Saito et al., 1993; Sheehan et al., 1997; Leissring et al., 2000), as do AD mouse models (Kurbatskaya et al., 2016; Ahmad et al., 2018). Both AnkG and βIV spectrin are calpain substrates, and their proteolytic cleavage leads to the dispersion of the ion channels normally clustered at the AIS (Schafer et al., 2009; Greer et al., 2013). Glutamate stimulation would be expected to activate NMDA channels, increase intracellular calcium, and thus alter calcium dynamics, although there are questions as to how this happens (Woods and Padmanabhan, 2012). To test whether the reduction of AIS length we observed was because of the activation of calpain, we treated primary cultures of WT mouse neurons with the calpain inhibitor, PD150606 (1 mm), then transfected the neurons with APPSwe. After calpain inhibition, we found that the reduction in AIS length was significantly less than in cells without inhibitor (Fig. 4I,J). Immunocytochemistry with 6E10 showed that the same amount of APP was expressed, and its distribution within the cells was comparable in the presence and absence of PD150606. It should be noted, however, that the length of the AIS was only partially restored by this treatment (Fig. 4I). The suggestion is that, while calpain plays a role in the APP effects, there are likely other mechanisms involved.
The APP-induced changes in AIS structure are reflected in neuronal activity
Previous studies suggest that, by decreasing AIS length and increasing its distance from the cell body, the elevated levels of APP should raise the threshold for action potential generation (Grubb and Burrone, 2010a; Kuba et al., 2014). This is predicted to decrease the activity of the neuronal networks that establish themselves among cultured dissociated neurons. To test this prediction, we infected neuronal cultures on DIV10 with a lentivirus expressing GCaMP8f, a calcium sensor that is widely used to monitor activity in living nerve cells (Zhang et al., 2021). Sister cultures were coinfected with GCaMP8f plus an APP expressing virus. At DIV14, APP-infected cells (e. g., the cell in the white box and the arrow in Fig. 5B) showed noticeably higher mCherry signal than that in the nearby non–APP-transfected cells in the same culture. This was true in both the cell body and the processes. The chemograph showed the change of the GCaMP8f signal changes in the green and red boxes from Figure 5B. We then monitored either the total GCaMP8f signal (ΔF/F) over a 15 s time frame (Fig. 5C–E) or the total number of GCaMP8f flashes over the same time period (Fig. 5D,E). Both measurements were reduced showing that, as predicted, APP overexpression reduces the activity of neuronal cultures.
The AIS in AD: in vivo data
The data described thus far have come from in vitro assays. To ask whether similar results would be found in vivo, we compared the AIS in cryostat sections from the brains of WT and R1.40 mice. R1.40 mice carry a human APPSwe transgene in addition to the endogenous (WT) mouse App gene. In vivo, the measured length of the WT AIS was shorter (10-15 µm) than it was in vitro (30-35 µm). Even with this difference, the relative decrease in the length of the AIS of R1.40 neurons was readily apparent (Fig. 6A). The reduction was seen in all regions of cortex but failed to reach statistical significance in more posterior samples (Fig. 6B). To rule out the possibility that a change in orientation was the cause of this difference, we also took AIS measurements of cortical neurons in the coronal plane and obtained similar results (Fig. 6C). As an additional control, we compared the apparent length of the AIS measured in cryostat sections of different thicknesses. While we observed a modest 20%-25% increase in the measured AIS length in 20 µm sections compared with the 10 µm used above, there was no change in the relative difference (or its statistical significance) between the AIS of WT and R1.40 neurons (data not shown). The changes in AIS length in the R1.40 mouse were found, although the density of the AIS did not differ between the two genotypes (Fig. 6D,E). We also quantified the AIS lengths of R1.40 pyramidal cells of the CA fields of the hippocampus (Fig. 6F,H) and found them to be significantly shorter than WT in both the sagittal (Fig. 6F) and coronal planes (not shown). Unlike the neocortex, however, the density of AIS was reduced in the CA fields of R1.40 mice (Fig. 6G).
The in vitro data demonstrated that exposure to Aβ peptides alone does not affect the properties of the AIS. Qualitative observations of β4 spectrin/APP (Clone: 6E10) double-immunolabeled material supported this view as they revealed no obvious bending of the AIS in or near cortical amyloid plaques (Fig. 7A). R1.40 mice start to build up plaques in the frontal cortex by 13.5 months. By 18 months, plaque density has increased significantly (Lamb et al., 1997; Kulnane and Lamb, 2001; Lehman et al., 2003). We thus measured the density and length of the AIS in 18 month R1.40 mouse frontal cortex. We computed both values in areas defined by three concentric rings centered on 6E10-positive Aβ plaques (Fig. 7B), a method established by Marin et al. (2016). We found no statistically significant association between AIS length (Fig. 7C) or density (Fig. 7D) and the distance of the AIS from the center of a plaque. Thus, consistent with our in vitro findings, the reduction of AIS length was not sensitive to the presence of amyloid.
To test whether the situation in human AD was similar to the R1.40 mouse AD model, we turned to human postmortem brain samples. We chose Brodmann Area 9 (BA9) as our test region. We immunostained 10 µm cryostat sections for MAP2, APP (N-terminus), and AnkG. In material from UCs (Fig. 7A), the AIS was well stained, and many (though not all) could be clearly associated with their neuron of origin (Fig. 8Ai). While in control specimens the levels of intracellular APP immunoreactivity were low, a few neurons displayed elevated staining (Fig. 8A, yellow arrow). In cells with low levels of APP, an associated AIS could often be identified (Fig. 8Aii). For neurons with elevated APP, however, we could rarely identify its associated AIS. To be sure, the thickness of the cryostat sections was less than the average diameter of a neuron. Thus, for any given neuron, the AIS could well have been located in an adjacent section. The consistency of the finding, however, was striking and it was amplified by the situation in the AD material.
In all 6 AD cases examined, the presence of amyloid plaques was readily observed. Though less often discussed in the literature, we found an equally impressive increase in the level of background staining. This background can be clearly seen in Figure 8Bi, 8Bii (insets). As each of the AD cases was sporadic in nature, the elevated levels of APP immunostaining were most probably a function of age and environmental factors. In addition to the higher APP background, virtually every neuron in cortex presented with strong intracellular APP immunoreactivity (Fig. 8B, yellow arrows). Plus, as would be predicted from the few high APP-expressing neurons in the UCs, we could almost never define a cell of origin for any of the AIS present in an AD brain section.
The AnkG channel from the merged images (Fig. 8A,B) was used to quantify the dimensions of the AIS in UCs (Fig. 8C) and in AD cases (Fig. 8D). It was notable that, although the size of human brain is much larger than that of the mouse brain, the length of the AIS of human prefrontal neurons (10-18 µm) was only slightly longer than that found in the mouse (10-15 µm). By contrast, the average density of AIS in the human material was about one-fifth of that found in mouse (Fig. 8E). Further, just as we found in the R1.40 transgenic mouse, the average length of the AIS was decreased by about one-third in the AD brain compared with UCs (Fig. 7F). Double-staining with AnkG and N-term APP plainly showed that, in all AD cases, the APP protein had infiltrated the AIS region (Fig. 8Bii). An important distinction with the mouse, however, was that there was little evidence of a reciprocal gradient of APP and AnkG. In the R1.40 mouse, the excess APP appeared to shift the peak of AnkG nearly 10 µm distally from that seen in WT. In the human, although the intensity and total length of AnkG staining region were reduced, the shift was less apparent (Fig. 8G, compare with green symbols in Fig. 3C).
To obtain a more global picture of these changes, we performed Western blots of APP and Ankyrin G using lysates prepared from cryostat sections of the same blocks of human BA9 tissue used for immunocytochemistry (Fig. 8H). Not unexpectedly, we found that the levels of APP protein were considerably higher on average in the AD brain than in UCs (Fig. 8I). The increase was statistically significant and quantitatively similar to the increase seen in APP levels in the R1.40 mouse brain. By contrast, the overall levels of Ankyrin G protein were significantly decreased in AD brain compared with UCs, significantly so for both the 480 and 270 kDa bands (Fig. 8I).
Discussion
Our findings reveal several new and unexpected roles of the full-length APP protein in brain function. First, we found that a sustained increase in neuronal activity leads to a rapid increase in App gene expression. Second, we demonstrated that elevated levels of APP protein, whether caused by glutamate, APP overexpression, or an APP transgene, led to two important changes in the AIS: it shortens in length and shifts in position away from the cell body. Third, we showed that these changes in AIS position and length are associated with the predicted decrease in neuronal network activity. Fourth, we showed that the effect on AIS length is observed in neurons both in vivo and in vitro. Finally, we confirmed that in humans with sporadic AD cortical neurons have elevated levels of APP protein, and that portions of this excess APP invade the AIS domain. This increased APP is associated with a significant shortening of the AIS compared with UCs.
The AIS is an important yet highly dynamic region of a neuron. Alterations of the length and position of the AIS are activity-dependent and, in turn, affect neuronal activity (Grubb and Burrone, 2010b; Huang and Rasband, 2018; Leterrier, 2018). As the site of action potential initiation, even subtle changes in the AIS can have measurable effects on neuronal excitability (Grubb and Burrone, 2010a; Kuba et al., 2010; Kuba, 2012; Atapour and Rosa, 2017; Kneynsberg and Kanaan, 2017). A longer AIS, or one more proximal to the cell body, will result in a more excitable neuron because of a lower action potential firing threshold. The reverse situation is also true. When the length of the AIS decreases or it shifts to a more distal position along the axon, neuronal activity is decreased (Grubb and Burrone, 2010a; Rasband, 2010). While this correlation between AIS structure and electrophysiological function has been clear, the mechanisms that link changes in AIS structure with activity have remained largely unknown. Correlations with altered calcium levels (Evans et al., 2013) and the resulting activation of calpain, a protease capable of cleaving both Ankyrin G and βIV spectrin, offer one possible explanation (Schafer et al., 2009). Yet the studies described here using calpain inhibition suggest that calpain itself can provide only a partial explanation of the observed phenomena.
The actions of APP we have documented offer a compelling, additional mechanism to explain the correlation between neuronal activity and AIS behavior. With immunocytochemistry in both mouse and human, we showed that, in a mature AIS, APP is in close apposition to both ankyrin G and βIV spectrin. The results of the coimmunoprecipitation experiments confirm this by showing that APP physically interacts with both major scaffolding proteins. The association is found with human as well as with mouse APP proteins. Manipulating the levels of APP either genetically or through glutamate stimulation shows that this physical interaction is coupled with structural consequences for the AIS. With increased neuronal activity, App gene transcription is rapidly upregulated, and protein levels follow rather closely. The transcriptional response is mostly likely the key to this regulation as the APP protein has a relatively short half-life (Xiao et al., 2015). Coupled with the electrophysiological consequences of changes in AIS structure, the dynamic nature of the transcriptional response suggests that, under normal conditions, APP acts as a modulator on neuronal excitability. When neuronal activity increases, APP levels rise causing the distance between the AIS and the soma to increase and the length of the AIS to decrease. The overall effect is to reduce the neuron's firing threshold and thus serve as a protective mechanism.
The data from human AD neurons and mouse AD models suggest that, beyond its contribution to normal neuronal activity, the AIS may also play a role in neurodegenerative disease. Elevated APP levels, such as those found in AD, drive the same changes in AIS position and length, as does increased neuronal activity. The relevance of this finding to AD is enhanced by data showing that hAPPSwe has effects on the AIS that are more dramatic than those of WT hAPP. The effects are cell autonomous and mediated by endogenous full-length APP protein, not by exogenous Aβ. Addition of Aβ to the culture medium (oligomers or fibrils) has no effect on the AIS of either human or mouse neurons. Plus, the length of the AIS in the neurons of the mouse brain is unaffected by its proximity to an amyloid deposit. The immunoprecipitation data further support the connection between the dynamics of the AIS and the symptoms of AD. Antibody to APP pulls down ankyrin G and βIV spectrin from both human and mouse brain lysates (Fig. 3C). The human observation is particularly informative as the AD cases were all sporadic; the increased levels of APP would be exclusively WT in structure. The retarded development of the AIS in R1.40 cultures may have additional implications for the full clinical picture of AD. AD has long been recognized as having a developmental component, with intellectual performance in youth inversely related to the risk of dementia in old age (Snowdon et al., 1996; Whalley et al., 2000). In this light, the slow development and shorter length of the AIS in cultured R1.40 neurons suggest that elevated APP expression may be one explanation for this connection.
In sporadic AD, we speculate that the aging process leads to a relentless increase in the damage experienced by neurons. As APP is a damage response protein, this results in a chronic elevation in the levels of APP. We speculate that this lifetime accumulation of APP, worsened by hyperexcitability, irritation because of inflammation, vascular problems, or a traumatic injury, eventually tips the system into a disease state. The levels of APP never return to normal leading to a sustained shift in the position of the AIS away from the cell body and a reduction in its length. Both changes would have the effect of lowering neuronal excitability. We found that the reduction of AIS length compared with control animals was uniformly distributed throughout the R1.40 frontal cortex, which emphasizes that the changes are APP-dependent, not Aβ-dependent. The effects on the AIS can be assumed to occur throughout the nervous system, not only in proximity to amyloid plaques. Our model further predicts that familial AD mutations in APP exacerbate the normal age-associated vulnerabilities, leading to an earlier age of onset, as is clinically observed.
One important observation that does not fit our proposed model is that hyperexcitability and epileptic activity are features of both AD and its mouse models (Di Lazzaro et al., 2004a, 2004b; Palop et al., 2007; Minkeviciene et al., 2009; Pennisi et al., 2011), yet one of the phenotypes of APP-overexpressing neurons is a reduced probability of firing. Although only speculation, we would propose that the APP overexpression and its effects on the AIS are not experienced equally by all brain regions and that those regions, or neurons, that are spared attempt to compensate for reduced network activity increasing their firing frequency. This is consistent with the heterogeneity we observed in our human material in both the APP intensity in different neurons and the infiltration of APP in different individual AIS.
We note that, while mutations of APP leading to familial AD are consistent with an Aβ-dependent disease mechanism, the Alzheimer's genetic data formally identify only APP; the involvement of Aβ is inferential. The PSEN1 and PSEN2 familial AD genes fortify the Aβ argument, but only indirectly as the γ-secretase has dozens of substrates apart from APP. We propose that consideration be given to the idea that the actions of the full-length APP protein on the AIS and the resultant disruption of network activity provide an alternative explanation for the APP genetic data. This would revise our disease models and offer novel therapeutic approaches to halt the advance of AD.
Footnotes
This work was supported by Hong Kong University of Science and Technology Grant R9321; University of Pittsburgh School of Medicine start-up support and a Pitt Momentum award to K.H. and R.B.; and National Institutes of Health Grants R01 CA254112 to Y.L. and 5U10 AA008401 to R.P.H. We thank the Neuropathology Core of the Alzheimer's Disease Research Center at University of Pittsburgh for providing human samples. The Alzheimer's Disease Research Center is supported by National Institute on Aging Grants P30 AG066468 and P50 AG005133.
The authors declare no competing financial interests.
- Correspondence should be addressed to Karl Herrup at herrup{at}pitt.edu