Autophagy is markedly impaired in Alzheimer’s disease (AD)Here we reveal unique autophagy dysregulation within neurons in five AD mouse models in vivo identify its basis a neuron-specific transgenic mRFP-eGFP-LC3 probe of autophagy pH, multiplex confocal imaging correlative light electron microscopyAutolysosome acidification declines in neurons well before extracellular amyloid deposition, associated markedly lowered vATPase activity build-up of Aβ/APP-βCTF selectively within enlarged de-acidified autolysosomesIn more compromised yet still intact neurons, profuse Aβ-positive autophagic vacuoles (AVs) pack into large membrane blebs forming flower-like perikaryal rosettesThis unique pattern, termed PANTHOS (poisonous anthos (flower)), is also present in AD brainsAdditional AVs coalesce into peri-nuclear networks of membrane tubules where fibrillar β-amyloid accumulates intraluminallyLysosomal membrane permeabilization, cathepsin release lysosomal cell death ensue, accompanied by microglial invasionQuantitative analyses confirm that individual neurons exhibiting PANTHOS are the principal source of senile plaques in amyloid precursor protein AD models.

Autophagy is the principal pathway for lysosomal degradation, maintaining cellular homeostasis by constitutively turning over obsolete proteins organellesIt is induced further by disease cell stress to eliminate abnormal proteins, aggregates damaged organelles^1,2,3,4Autophagy encompasses several mechanisms for sequestering substrates their delivery to lysosomes (LYs)In the major autophagy–lysosomal pathway (ALP), macroautophagy, an elongating double membrane envelops cytoplasm or, via adaptor protein, engages specific targeted substrates then closes to form an autophagosome (AP)APs mature to autolysosomes (ALs) upon fusion LY or endolysosome, which introduces varied cathepsin proteases, other acid hydrolases vATPase, the proton pump that acidifies AL lumens activates the hydrolasesLYs are targets of causative gene products risk factors for AD⁵, including the pathogenic amyloid precursor protein (APP) metabolites APP-βCTF Aβ⁶ that are actively generated within endosomal autophagic pathways are normally cleared by LYs⁷.

AD is defined neuropathologically by two lesions: intracellular tau aggregates (neurofibrillary tangles) neuritic plaques composed of focally swollen (dystrophic) neurites (DNs)⁸, extracellular β-amyloid many other proteins⁹Additionally, AVs containing incompletely digested autophagy substrates accumulate progressively within affected neurons at the earliest disease stage^5,10,11,12The molecular basis for autophagy dysfunction in AD, its relationship to APP/amyloid pathology its pathogenic implications are unclear due, in part, to technical challenges of monitoring ALP abnormalities in vivo in brainTo overcome these limitations, we generated transgenic mice (TRGL) neuron-specific expression of tandem fluorescence-tagged LC3 (mRFP-eGFP-LC3 or tfLC3), an autophagy adaptor protein selectively associated AP AL¹³The tfLC3 probe enabled us to investigate individual vesicular components of the neuronal ALP in intact brain and, to our knowledge for the first time, assess AL acidification ratiometrically in neurons throughout disease progression in mouse AD models.

To identify monitor AD-related ALP deficits, we crossed TRGL AD model mice that develop either early-onset or late-onset disease pathologyIn all five AD mouse models studied, we demonstrated early-appearing deficiencies of lysosomal vATPase activity, autophagy dysfunction in vulnerable neuron populations accumulation of APP-βCTF Aβ selectively within poorly acidified AL (pa-AL) well before extracellular β-amyloid depositionFurthermore, we identified a unique autophagic stress response in more compromised neurons characterized by fulminant proliferation of AVs within perikarya formation of large membrane blebs packed Aβ/APP-βCTF-filled AVsThe strongly fluorescent petal-like blebs surrounding a DAPI-positive fluorescent nucleus generate flower-like profiles that we term ‘PANTHOS’ (poisonous flower)Notably, AV fusion endoplasmic reticulum (ER) yields intraluminal formation of β-amyloid fibrils in a tubular network surrounding the nucleus, yielding morphologic features of a cored amyloid plaque within the intact neuronUsing an extensive array of imaging histochemical techniques, we establish quantitatively that PANTHOS neurons are the origin of the vast majority of senile plaques in AD mouse models, thus prompting a reconsideration of the conventionally accepted sequence of events in plaque formation in AD.

Detecting in vivo ALP dysfunction

A tandem mRFP-eGFP-LC3 transgene (tfLC3) driven by the THY-1 promoter is postnatally expressed specifically in neuronstfLC3 is expressed approximately one-fold higher than endogenous LC3 levels has no detectable effects on the ALP¹³Like endogenous LC3, tfLC3 binds to AP membranes persists after AP–LY fusion as an internalized substrate degraded within AL, ultimately yielding non-fluorescent LYsThe tfLC3 on AP fluoresces yellow-green (eGFP/mRFP) at the neutral pH of AP, but AL maturation upon fusion LY¹⁴ acidifies the AL, cafluorescence shifts from yellow to orange then to red as eGFP fluorescence is quenched below pH 6.0 (ref. ¹⁵)LYs after autophagic clearance of fluorescent LC3 or after new LY biogenesis can be visualized by immunohistofluorescence (IHF) labeling LY markers (for example, cathepsin D (CTSD) or LAMP 2) tagged a third fluorophoreNotably, this third fluorophore also differentiates the yellow-fluorescing AP from an AP that fuses an LY is cathepsin-positive but fails to acidify adequately and, thus, fluoresces yellow by tfLC3 labeling alone (Fig. 1a)^13,14The latter profile is classified as a pa-AL.

AP maturation acidification are most easily appreciated when the transition from AP to AL is protracted during retrograde axonal transport in primary neuronal cultures of TRGL mice (Fig. 1b)AVs are much fewer in vivo in the intact mature brain¹⁶Fully acidified AL is concentrated within perikarya proximal dendrites in neurons (Fig. 1c, arrowhead)ALs fluoresce purple (combined red blue) in a three-fluorophore (RGB) analysis of neocortical perikarya, reflecting an efficient perikaryal acidification mechanism (Fig. 1d, top)To model an AL/LY acidification deficit in vivo validate the tfLC3 probe in intact brain in vivo, 6-month-old TRGL mice were administered the amphiphilic weak base chloroquine (CQ) or the vehicle alone (controls) by intraventricular infusion for 5 days, neurons in neocortical layers III–V were imaged (Fig. 1d)A rise in vesicle pH above 6.0 causes tfLC3-positive puncta to fluoresce yellowBased on a green/red channel merge alone, these puncta would be mis-identified as AP; however, IHF a CTSD antibody Alexa Fluor 647 (pseudo-blue) secondary antibody identifies these puncta as CTSD-positive and, therefore, as pa-ALIn a three-channel merge, they fluoresce white (green, red blue fluorescence) (Fig. 1d, RGB merge bottom), contrasting the purple acidified AL in normal neurons (Fig. 1d, RGB merge top)LYs remain blue after CQ, reflecting their pH-insensitive detection by IHF (Fig. 1d)A computer algorithm¹³ determines for each vesicle the relative contributions of the three fluorophores based on their hue angle saturation, which is a more precise objective representation of ‘color’ (vesicle identity) than achieved by visual perception.

AL acidification deficiency arises before β-amyloid deposits

We crossed TRGL mice¹³ Tg2576 mice¹⁷, an AD model that develops β-amyloid plaques starting at 10~12 months of ageALP patterns in 1.6-month-old Tg2576/TRGL crosses were indistinguishable from single-TRGL littermates (Extended Data Fig. 1a); however, by 5 months of age, more than 90% of neocortical layer III–V perikarya had acquired yellow fluorescent AVs in addition to acidified ALs (Extended Data Fig. 1a)CTSD co-labeling revealed that the yellow AVs are exclusively CTSD-positive and, therefore, pa-ALs (Fig. 2a, bottom panels)pa-AL was also positive for CTSB the lysosomal membrane protein LAMP1 (Extended Data Fig. 1b)Hue-angle-based assignment quantification of AV subtypes in neocortex¹³ revealed four-fold more pa-ALs in Tg2576/TRGL (9.0 ± 0.5 per neuronal cross-section) than in TRGL (2.1 ± 0.3), significantly fewer mature ALs (4.4 ± 0.4 versus 6.6 ± 0.3 per neuronal cross-section) (Fig. 2b) increased size of pa-ALs ALs (1.3 ± 0.04 versus 0.48 ± 0.03 1.75 ± 0.09 versus 0.74 ± 0.05, respectively) (Fig. 2c)By 12 months, perikaryal pa-ALs further increased in Tg2576/TRGL (17.2 ± 0.7 per neuronal cross-section) (Fig. 2e,f)To further document AL acidification deficits in Tg2576 brain, we isolated AL/LY-enriched fractions by OptiPrep density centrifugation (Extended Data Fig. 1c) assayed their ATPase activity¹⁸Consistent observed pH deficits, vATPase activity in LY/AL of 6-month-old Tg2576 was decreased (65.6 ± 4.1%) compared to that of age-matched wild-type (WT) littermates (Fig. 2d) decreased further by 12 months in Tg2576 mouse brain (45.3 ± 3.7% relative to WT) (Fig. 2g)ATPase activity was similarly reduced in brains from two other mouse models of AD (5xFAD APP51) (Extended Data Fig. 1d)The time course graph indicates age-dependent increased prevalence of pa-AL while vATPase activity declines (Fig. 2h).

APP-βCTF/Aβ accumulate in pa-AL at early stages of disease

APP-βCTF Aβ accumulate intracellularly before β-amyloid is deposited extracellularly in AD, the endosomal–lysosomal system representing the main subcellular site for their generation^19,20,21To relate APP-βCTF/Aβ intracellular accumulation to early AL acidification deficits in Tg2576 mice, we localized APP metabolites within AV subtypes a monoclonal antibody (JRF/AβN/25) that detects APP-βCTF Aβ²²By 5 months, 40% of layer III–V neocortical perikarya in Tg2576/TRGL mice contained Aβ/APP-βCTF-positive puncta (Fig. 3a), which were almost exclusively pa-AL (88.6 ± 2.4%) based on CTSD co-immunolabeling imaging of four fluorophores (Fig. 3a, arrows, Fig. 3b)Immunoblot analyses on subcellular fractions from Tg2576 brains confirmed that LC3-II enriched AV fractions contain abundant APP-βCTF as well as γ-secretase components (presenilin 1 nicastrin) (Fig. 3c) Aβ (Extended Data Fig. 2a)Aβ localization in AVs was further validated by Aβ1-42 antibody (JRF/cAβ42/26) (Extended Data Fig. 2b, arrowhead)Also, APP-βCTF localization in AVs was further validated by an in situ proximity ligation assay (PLA) a modified Duolink technology (Methods) involving two primary antibodies directed against different epitopes (N-terminus or C-terminus) on APP-βCTF (Fig. 3d)PLA fluorescence (red) detected APP-βCTF in APPswe-overexpressing N2A cells Tg2576 neurons at considerably higher levels than in controls (Fig. 3e, arrowheads, Extended Data Fig. 2c,d)Notably, PLA signal (blue) revealed that APP-βCTF selectively accumulated in ALs that were poorly acidified in Tg2576/TRGL perikarya (92.9 ± 1.3%, n = 50 neurons) (Fig. 3f Extended Data Fig. 2e–g).

Progressively compromised neurons massively accumulate pa-AL

In 10-month-old Tg2576/TRGL mice, a subpopulation of neocortical neurons (layer III–V) began to accumulate substantially enlarged pa-ALs, which bulge the plasma membrane outward (Fig. 4a, enlarged right panel, arrowhead)The further massive proliferation of LC3-positive vesicles was accompanied by formation of large strongly fluorescent membrane blebs that project from the plasma membrane expperikaryal circumferenceA central nuclear region devoid of LC3 fluorescence (Fig. 4a) could be labeled by nuclear markers, including DAPI, histone H3 or lamin A/C (Fig. 4b,c)The absence of autofluorescence in this nuclear area excluded the possibility that DAPI signal was non-specific autofluorescence due to amyloid (Extended Data Fig. 3a)Most AVs in affected perikarya were LY-marker-positive by IHF, indicating that they were pa-ALs (Fig. 4d Extended Data 3b), which reflects a severe deficit of AL maturation acidification.

We observed an identical autophagic neurodegenerative pattern in five different mouse models of AD, including models accelerated neuropathology onset (5xFAD, TgCRND8 PSAPP) or delayed onset (Tg2576 APP51—an exceptionally late-onset model expressing hAPPwt)²³ (Fig. 4e)5xFAD/TRGL mice develop robust ALP disruption neuronal degeneration at an early age (starting after 2 months depending on sex)^24,25,26 (Extended Data Fig. 3c–e) in a more reproducible pattern than in Tg2576 or APP51 mice (Extended Data Fig. 3e)We used this model in further investigations on the relationship between the development of LC3-positive membrane blebs disease progression, including quantitative amyloid plaque pathologyTo our knowledge, similar huge AV-filled perikaryal membrane protrusions, as further defined ultrastructurally (Fig. 5), have not been previously described in a neurodegenerative state²⁷Because these rosettes of large fluorescent blebs surrounding a central DAPI-positive nucleus resemble petals of a flower, we have termed this unique degenerative process PANTHOS refer to the affected cells as PANTHOS neurons.

PANTHOS—a unique pattern of neurodegeneration in AD

The greater resolution of autophagic profiles afforded by the tfLC3 probe allowed us to visualize by confocal imaging the AV-filled blebs extending directly from the perikaryal cytoplasm of PANTHOS neurons via necks that taper toward the center of the perikaryon (Fig. 5a Extended Data Fig. 4)Electron microscopy (EM) analysis on brains of 5xFAD/TRGL mice confirmed the continuity of blebs the perikaryal cytoplasm identified AVs as the principal constituents within blebs (Fig. 5b)Perikaryal blebs exhibit long membrane-bound necks extending from the soma of the PANTHOS neuron (Fig. 5c, box i, outline arrowheads)Additional features of PANTHOS neurons at higher EM resolution include a centrally located electron-dense network of radiating membrane-bound tubular extensions containing partially fused fully incorporated AVs (Fig. 5c, box ii: yellow arrows, Extended Data Fig. 5a, inset: yellow arrowheads) as well as bundles of 6-nm fibers (Fig. 5c, box ii, Extended Data Fig. 5a, red arrowheads) that are strongly Aβ immunoreactive (Fig. 6d, box ii)In other EM images, AVs Aβ-positive fiber-containing tubular extensions are seen to be in the process of f(Fig. 5d Extended Data Fig. 5b, light-blue arrowheads).

EM analysis of brain sections labeled histochemically for the lysosomal enzyme acid phosphatase (ACPase) further confirmed the identity of most AVs in blebs as strongly ACPase-positive ALs, including those within the tapered bleb necks connecting blebs to the cytoplasm of the degenerating perikaryon (Fig. 5e, inset: yellow arrowheads)Although the asymmetric morphology of perikaryal blebs their evident cytoplasmic origin distinguish them from DNs, blebs were further distinguished from DNs, which are enriched neurofilaments, exhibit weak signal for lysosomal markers (CTSD LAMP2) as shown by IHF (Extended Data Fig. 5c, arrow) are infrequent compared to perikaryal blebs around PANTHOS neurons (Extended Data Fig. 5d).

To further establish the perikaryal origin of the many AV-filled blebbing profiles, we performed correlative light electron microscopy (CLEM) together serial block-face scanning EM imaging an Apreo scanning electron microscopeReconstruction of more than 500 z-plane images recreated the entire PANTHOS neuron in three dimensions (Supplementary Fig. 1 Movie 1)A stacked EM image sequence from 370 to 430 of region of interest (ROI) area (Fig. 5f) confirmed that the sizes of early-stage PANTHOS profiles approximate the size of normal neurons (Fig. 5g), but these profiles have expanded circumference as perikaryal blebbing becomes more extensive (Fig. 5h)The DAPI-positive center area of PANTHOS neurons approximates the sizes of the electron-dense centrally located areas in the stacked EM image (Fig. 5g)ImmunoEM analyses the nuclear marker KDM1/LSD1 confirmed the existence of nuclear remnants in the central area by detecting strong immunoreactivity in the same central area even after nuclear integrity was extensively disrupted (Extended Data Fig. 5e)A movie sequence through these serial sections clearly visualized dozens of AV-filled membrane blebs arising within the cytoplasm from tapered necks that expinto large bulbous projections (blebs) from the perikaryon (Supplementary Movie 1), as shown in a colorized section from the full set (Fig. 5h)A 3D reconstruction modeling illustrates the extensive blebbing of the perikaryon (Fig. 5i).

PANTHOS neurons are the principal origin of amyloid plaques

In 5xFAD/TRGL mice, Aβ APP-βCTF accumulate selectively within pa-ALs before β-amyloid plaques appear (Fig. 6a, arrowheads), as in Tg2576 mice (Fig. 3)Transition of neurons to a PANTHOS pattern is accompanied by robust accretion of perinuclear Aβ/APP-βCTF immunoreactivityCo-labeling of these PANTHOS neurons DAPI anti-β-amyloid antibody (4G8) identified a 4G8-positive corona surrounding a DAPI-positive nucleus remnant at the center of most affected perikarya (Fig. 6b)The progression of PANTHOS formation respect to β-amyloid accretion was further confirmed in the late-onset AD mouse model APP51 (Extended Data Fig. 6a–c).

In 5xFAD/TRGL mice, quantitative spectral analysis of the PANTHOS neuron’s central area discriminated DAPI fluorescence from fluorescence due to 4G8 immunolabeling (Fig. 6c)At more advanced stages of PANTHOS, DAPI fluorescence gradually disappears as more β-amyloid accumulates centrally (Extended Data Fig. 6d)Ultrastructural 3D6 immunoelectron microscopy (IEM) analyses localized this central accretion of Aβ immunoreactivity (Fig. 6d) within intraneuronal membranous tubular profiles (Fig. 6d, box i)Within many of these same profiles, 3D6-positive bundles of fibrils, widths of around 10 nm, approximated the known diameters of fibrillar β-amyloid²⁸ (Fig. 6d, box ii)Resembling the PANTHOS morphologies in Fig. 5b, Aβ IEM of a PANTHOS neuron 3D6 additionally detected 3D6-positive AVs packed into perikaryal blebsPerikaryal AVs were also shown to be continuous with, incorporated into, the central Aβ-positive network of membrane tubular structures (Fig. 6e Extended Data Fig. 6e (3D6 4G8 IEM))IEM antibodies to either LC3 or CTSD confirmed that these vacuoles are AVs (Extended Data Fig. 6e)ER, a key source for AP membrane components, is increasingly mobilized to supply membrane for new APs as autophagy induction in AD brain remains high²⁹However, as accumulating AVs deplete sources of available membrane, APP-rich ER Golgi membranes join endosomes as major sources of APP-βCTF/Aβ generationTherefore, ER Golgi are likely key contributors to the expansion of the amyloid fibril network, supporting AP/AL formation by contributing both membrane β-amyloid precursor.

Consistent PANTHOS being the principal source of amyloid plaques, immunolabeling of β-amyloid 3D6 in 5xFAD/TRGL mice revealed an exclusive co-incidence a one-to-one quantitative relationship between individual PANTHOS neurons individual amyloid plaques (Fig. 7a)All PANTHOS neurons were 3D6-positive, 91.7 ± 0.01% of the total 3D6 signal in brain was detectable in PANTHOS lesions (n = 3 mice, 105 neurons 94 lesions counted) (Fig. 7b)Moreover, a DAPI-positive nuclear signal, including condensed or fragmented/diffuse signals in the perikaryal center (Fig. 7c), was detectable in 91.4 ± 1.29% (n = 6, two sections per mouse) of PANTHOS lesions in cortex from 2.7-month-old 5xFAD/TRGL mice (Fig. 7d, top graph)In older mice (6 months), 67.8% of the PANTHOS neurons still displayed DAPI nuclear signal (Fig. 7d, bottom graph) despite glial invasion advanced neurodegenerationThis percentage is likely an underestimate because immunoEM analyses the nuclear marker KDM1/LSD1 revealed nuclear remnants even after loss of nuclear integrity (Extended Data Fig. 5e)The temporal 1:1 spatial relationship among PANTHOS, intracellular perinuclear Aβ accretion amyloid plaque formation, therefore, indicates that the vast majority of amyloid plaques originate from a corresponding individual PANTHOS neuronThe transition from intact nucleated PANTHOS neurons to the more advanced stage of DAPI disappearance glial invasion of the cell likely represents the loss of cellular integrity conversion to an extracellular plaque.

Lysosomal permeabilization promotes neuronal cell death

Lysosomal alkalinization is reported to promote lysosomal membrane permeabilization cathepsin release into cytosol³⁰Cytosolic membrane/vesicle fractionation analyses markedly increased levels of lysosomal enzymes in the cytosol of brain from 6-month-old 5xFAD mice compared to brains from WT littermates (Fig. 7e)Lysosomal enzyme leakage was detectable at 6-month, but not young (2.7-month), brains, when many fewer neurons are affectedWe further examined the association of PANTHOS lysosomal membrane permeabilization CTSD IHFCompared to an adjusted normal neuron (Fig. 7f, arrow), a PANTHOS neuron (Fig. 7f, arrowhead) displayed diffuse CTSD immunoreactivity in a 5xFAD/TRGL mouse brain co-labeled CTSDWe ruled out the involvement of a caspase-3-mediated apoptotic cell death, because PANTHOS neurons were caspase-3-negative (Extended Data Fig. 7a).

PANTHOS neurons evolve into senile plaques in AD models

To characterize the evolution of PANTHOS neuron lesions into mature plaques, we immunolabeled PANTHOS Thioflavin S (Thio-S) for the detection of dense-cored senile plaques (Fig. 8a Extended Data Fig. 7b,c)In quantitative analyses of 5xFAD/TRGL at 2.2 months of age, half of the PANTHOS profiles were Thio-S-positive, whereas, in 6-month-old 5xFAD/TRGL mice, more than 95% were Thio-S-positive (Fig. 8a, graph)To further characterize the evolution of PANTHOS neuron lesions into mature plaques, we immunolabeled reactive astrocytes microgliaNeither glial cell type was frequently associated initially PANTHOS neurons, and, therefore, these cells were unlikely to be a major triggering factor in PANTHOS developmentIn quantitative analyses of 5xFAD/TRGL at 2.7 months of age, most PANTHOS neurons were unengaged by microglia or astrocytes (Fig. 8b)In older 5xFAD/TRGL mice (6 months), when greater numbers of PANTHOS neurons exhibited advanced loss of structural integrity, relatively few affected neurons were unengaged by microglia astrocytes (Fig. 8b).

In older 5xFAD mice, PANTHOS lesions frequently expanded into larger senile plaques when adjacent PANTHOS neurons merged into a single larger structure (Fig. 8c, A1 A2, respectively) that comprised multiple Thio-S-positive dense cores (Fig. 8d)Within these growing lesions, newly recruited PANTHOS neurons could still be recognized (Extended Data Fig. 8a, arrowheads), but loss of integrity of the original PANTHOS neurons its adjacent neighbors created an expanding central core of persisting β-amyloid as other cellular debris is cleared, yielding, finally, an enlarged extracellular dense-cored senile plaque (Fig. 8b, A3 A4, respectively), as confirmed by z-stack confocal imaging (Extended Data Fig. 8b).

Our transgenic dual-fluorescence probe, identifying autophagic compartments associated changes in their pH in vivo, established that autophagy failure in five different APP-AD mouse models originates from an early decline of AL/LY acidificationFurthermore, we uncovered a previously undescribed pattern of extreme autophagic stress, termed PANTHOS, in individual neuronal perikarya, which is characterized by massive perikaryal accumulations of poorly acidified AVs containing APP-βCTF/AβThe advance of PANTHOS generates an intraneuronal perinuclear ‘core’ of β-amyloid within membrane tubulesPreliminary analyses of human AD brain (Extended Data Fig. 9) revealed a similar PANTHOS in selected neocortical neurons, which is most easily appreciated immunocytochemically at the Braak II pathologic stage as the first β-amyloid plaques are formingProgressive failure of an initially neuroprotective autophagy response in neurons²⁹ is accompanied by an evolution of PANTHOS toward neuronal cell death involving lysosomal membrane permeabilization, cathepsin release and, ultimately, glial invasion extracellular plaque formation expansion (Extended Data Fig. 10, diagram).

AL acidification deficits in Tg2576 mice were detectable by 5 months of age—more than 4 months before β-amyloid deposited extracellularlyThe emergence of pH deficits in AL coincided lowered vATPase activity in brain LYs—the likely molecular basis for the acidification deficitDeclining AL acidification was accompanied by selective build-up of APP-βCTF Aβ within enlarged pa-ALsThese APP metabolites are known to be both generated degraded in ALs^12,31 amphisomes³²We also cannot exclude additional delivery of APP-βCTF Aβ to AL/LY through microautophagy chaperone-mediated autophagy³³.

The exceptional resolution of our tfLC3 autophagy probe, combined advanced ultrastructural multiplex confocal imaging methods, enabled the recognition of a unique pathobiologic process (‘PANTHOS’) in intact neurons within vulnerable cell populationsThis morphologic pattern, not previously reported, to our knowledge, includes AV accumulation (‘autophagic stress’) so extreme that it induced huge AV-filled plasma membrane blebs accelerated perinuclear accretion of Aβ β-amyloid fibrils within tubulo-vesicular structures created, in part, through AV fusionLarge AV-filled blebs were shown by CLEM 3D serial ultrastructural analyses to be formed by plasma membrane evagination to originate from the perikaryal cytoplasmTheir asymmetric morphology, high hydrolase content extensive distribution encircling an affected perikaryon far outnumbered DNsUnlike perikaryal blebs, DNs were commonly enriched neurofilaments, only weakly cathepsin-immunoreactive LAMP2-immunoreactive frequently myelinated.

Autophagic stress develops in many congenital lysosomal storage disorders (LSDs)In several of these disorders, Niemann–Pick type C (NPC1)³⁴ mucopolysaccharidosis type III (MPS-III), pH has been shown to be elevated³⁵Among LSDs, NPC1 has considerable phenotypic overlap AD (for example, paired-helical filaments, cholinergic neurodegeneration, endosome anomalies, disease acceleration by ApoE4, intracellular Aβ/βCTF elevation modest amyloid deposition^36,37,38); tauopathy has been reported in mouse MPS-III models intracellular synuclein, Aβ accumulations are detected in MPS-III brain³⁹That individuals these disorders usually do not survive to mid-adult ages may partly explain the infrequency of amyloid plaquesEven individuals AD due to PSEN1 mutations generally do not develop amyloid plaque pathology until the fourth decade of life, when aging factors may contribute^10,40Also, neuronal ALP induction continues to increase in AD brain even as substrate clearance declines^29,41, compounding autophagic stress likely increasing βCTF/Aβ generation¹².

PANTHOS neurons account quantitatively for the overwhelming majority of plaques that developed in five different AD models at the ages studiedIn young 5xFAD mice, β-amyloid cored plaques, detected β-amyloid antibodies, showed nearly 1:1 coincidence a single PANTHOS neuron exhibiting a central nucleusEven at a relatively late stage of compromise, intracellular Aβ-immunoreactive fibrils forming a perinuclear core are mainly contained within membrane-bound tubular structures derived from fusion of Aβ-positive ALsThis stage can be reached without appreciable microglial or astrocytic invasion that would reflect ‘eat me’ signaling from dying neurons^42,43, implying, therefore, that neuronal structural integrity is prolonged even as PANTHOS is quite advancedSubsequent microglial astrocytic invasion of the PANTHOS neuron heralds the eventual cell death that converts this amyloid lesion within an intact neuron into an extracellular amyloid plaque.

β-amyloid plaque formation in AD has commonly been considered to originate from extracellular deposition of β-amyloid derived from secreted Aβ, which then triggers secondary neuritic dystrophy neuronal cell deathBy contrast, our evidence in diverse AD models supports the opposite sequence—namely, extracellular plaques mainly evolve from intraneuronal build-up of β-amyloid within membrane tubules, forming a centralized amyloid ‘core’ within single intact PANTHOS neurons that subsequently degenerate to give rise to the classical senile plaqueThis ‘inside-out’ process accords substantiates hypotheses from many investigators^44,45In versions of this hypothesis, Aβ its oligomeric species generated intracellularly within ALP compartments can gain access to the extracellular space by neurodegeneration, local membrane damage or unconventional secretion (exocytosis)Importantly, a few investigators have described intracellular membrane-enclosed amyloid fibrils in AD mouse models⁴⁶ and, in AD brain, the frequent presence of amyloid surrounding DAPI-positive nuclei^47,48 neuronal lysosomal hydrolase abundance within extracellular β-amyloid⁴⁹.

Our findings add to mounting evidence that lysosomal acidification the dysregulation of the vATPase complex are common targets of genetic metabolic disruptions associated neurodegenerative disease⁵⁰Coupled previous evidence^10,20, our findings strongly support a pathogenic link between APP metabolites LY dysfunction in ADNotably, remediating PSEN1-related lysosomal pH deficits by various means ameliorates autophagy failure other AD-related pathology in AD models^7,24Additional supporting evidence from our group shows that the PANTHOS cascade in APP-based AD models described in this report can be significantly alleviated by pharmacologically targeting the lysosomal pH deficitBeyond the significance of findings revealed here, we anticipate broad potential of our transgenic dual-fluorescence tfLC3 autophagy probe to characterize ALP changes sensitively over time in other neurodegenerative disease models to facilitate assessment of autophagy/lysosome modulators as therapeutic agents.

Cell lines reagents

WT APPswe stably expressed murine neuroblastoma (N2a) cells were maintained in DMEM penicillin–streptomycin 10% FBS at 37 °C 5% CO₂ (ref. ⁵¹).

Mouse lines animal care

We used the Tg2576 mouse line (B6;SJL-Tg(APPSWE)2576Kha), which expresses mutant human APP (Swedish K670N/M671L) is maintained on a B6;Dba/2F1;SW backgroundFor TRGL (Thy-1 mRFP-eGFP- LC3) mouse generation, targeting vector for tfLC3 was constructed by insertion of tfLC3 into Thy1.1 expression cassette^13,52Tg2576/TRGL mice were studied at 1.6, 5, 10 12 months together TRGL littermates as a controlThe tfLC3 was crossed 5xFAD (Tg6799, C57BL/6NTAC), which expresses mutant human APP and PSEN1 (APP KM670/671NL: Swedish, I716V: Florida, V717I: London, PSEN1 M146L, L286V)⁵³, then tfLC3/5xFAD mice were studied at 1.6, 2.7, 4 6 months together age-matched controlsTgCRND8 mice, which express mutant human APP (Swedish K670N/M671L Indiana V717F)⁵⁴, were crossed TRGL, 1.9-month-old males were usedPS/APP mice⁵⁵, which express mutant human APP (Swedish K670N/M671L) mutant PS1 (PS1M146L), were crossed TRGL, 3.1-month-old males were usedAPP51 mice²³, which express WT human APP751, were crossed TRGL, females were usedDetailed mouse age sex information are in the figure legendsThe mice were maintained in the Nathan Kline Institute (NKI) animal facility housed at ~22.8 °C room temperature a humidity level of ~55% on a 12-hour light/dark cycleAll animal experiments were performed according to the ‘Principles of Animal Care’⁵⁶ approved by the Institutional Animal Care Use Committee at the NKI.

Human brain

Paraformaldehyde (PFA)-fixed tissue blocks obtained from prefrontal cortex (Brodmann area 9/10) were kindly provided from Emory Alzheimer’s Disease, from Marla Gearing (Alzheimerʼs Disease Research Centers/Center for Neurodegenerative Disease), demographic information outlined.

We used Braak stage II brains (E05-57: 86 years old, black female postmortem interval (PMI) of 6 hours; E05-54: 85 years old, white female PMI of 7 hours; OS96-08: 65 years old, white male PMI of 4 hours).

Antibodies reagents

Anti-PS1 loop mouse monoclonal antibody (MAB5232:clone PS1-loop, 1:1,000) anti-nicastrin mouse monoclonal antibody (MAB5556: clone 9C3, 1:1,000) were purchased from ChemiconRabbit anti-CTSD (Rudy4, 1:2,000) antibody NFL (21.4, 1/250) were produced in-house¹⁸CTSB was from Neuromics (GT15047, 1:250)LAMP2 was from the Developmental Studies Hybridoma Bank (ABL-93, 1:200)LIMP2 was from Novus (NB400-129, 1:200)Antibodies directed against APP, Aβ and/or other APP proteolytic species included APPc (Sigma-Aldrich, A8717, 1:250); 4G8 (BioLegend: clone 4G8, 800701, 1:250); C1/6.1 monoclonal antibody against the C-terminal 20 residues of APP (made in-house, 1:400, NKI)Additional mouse monoclonal antibodies were generous gifts from Marc Mercken (Janssen Pharmaceuticals/Johnson & Johnson): JRF/AβN/25 (specific to Aβ1-7, 1:200); 3D6 (specific to Aβ1-5, 1:250); JRF/cAb42/26 (specific to Aβ42, 1:200)⁵⁷; MAP2 (Sigma-Aldrich, M9942: clone HM-2, 1:250); NSE (Dako, M0873: clone BBS/NC/VI-H14, 1:250); histone H3 (4499, 1:200)Lamin A/C (4777: clone 4C11, 1:200) Tom20 (42406, 1:2,000) were from Cell Signaling TechnologyKDM1/LSD1 (Abcam, ab129195: clone EPR6825), GFAP (Sigma-Aldrich, AB5804, 1:250), IbaI (Wako, 019-19741, 1:250), ATP6 V1A (GeneTex, GTX110815, 1:1,000), ATP6 V0a1 (Abcam, ab176858, 1:2,000) Rab5 (Abcam, ab218624: clone EPR21801, 1:1,000)Rab7 (Cell Signaling Technology, 9367: clone D95F2, 1:1,000), PDI (BD Biosciences, 610946: clone 34, 1:1,000), STX6 (Cell Signaling Technology, 2869: clone C34B2, 1:2,000), Tubulin (Sigma-Aldrich, T8535:clone JDR.3B8, 1:5,000), Actin (Sigma-Aldrich, A1978: clone AC-15, 1:5,000) anti-p62 (ProGen Biotech, GP62-C, 1:500)Anti-SEC61B rabbit pAb (15087-1-AP, 1:1,000) was from ProteintechHRP-linked rabbit IgG (711-035-152, 1:5,000), mouse IgG (711-035-150, 1:5,000), rat IgG (712-035-150) goat IgG (705-035-003) secondary antibodies were purchased from Jackson ImmunoResearchProlong Diamond Antifade Mount (P36961), goat anti-mouse Alexa Fluor 647 (A21235), goat anti-rat Alexa Fluor 647 (A21247), goat anti-rabbit Alexa Fluor 647 (A21245) donkey anti-rabbit Alexa Fluor 405 (A48254) secondary antibodies were from Thermo Fisher ScientificMouse on Mouse (M.O.M) detection kit (BMK-2201), normal-donkey (S-2000-20) normal-goat (S-100) serum blocking solution were from Vector LaboratoriesThio-S (T1892) was from Sigma-Aldrich.

Ratiometric analysis of AL AP acidity

Procedures were performed as previously described¹³Confocal images were analyzed the Zen Blue Image Analysis Module from Carl Zeiss MicroscopyThe R, G B intensity values of each vesicle were calculated the profile function of ZenThe RGB ratio of each vesicle was converted into a hue angle saturation range by entering the values of R, G B for a given puncta into the formula as follows: Hue° = IF(180/PI()×ATAN2(2×R-G-B,SQRT(3)×(G-B)) < 0,180/PI()×ATAN2(2×R-G-B,SQRT(3)×(G-B)) + 360,180/PI()×ATAN2 (2×R-G-B,SQRT (3) × (G-B)))Saturation percent of the hue angle was calculated by entering the values of R, G B for a given puncta into the following formula = (MAX(RGB) − MIN(RGB)) / SUM(MAX(RGB) + MIN(RGB))×100, provided lightness is less than 1, which is the usual case for our dataHue angle was converted to color the hue color wheel.

Subcellular fractionation, gel electrophoresis western blotting

AV prep: Procedures were performed as previously described⁵⁸For each mouse genotype, cerebral cortices from five or more brains were pooledThe samples were homogenized subjected to differential centrifugation to separate a fraction enriched in AVs, LYs mitochondria as previously describedThe different organelles in this fraction were isolated by floatation in a discontinuous gradient of metrizamide (50%, 26%, 24%, 20% 10%), the LY-enriched fraction was recovered in the 24–16% interfaceA fraction enriched in ER resealed vesicles (microsomes), the cytosolic fraction was obtained in the pellet supernatant, respectively, after centrifugation of the supernatant at 100,000g for 1 hourCytosol membrane/vesicle prep: Cerebral cortices from male 5xFAD WT mouse brain were homogenized buffer (20 mM Tris-Cl, pH 7.4 250 mM sucrose, 1 mM EGTA, 1 mM EDTA, 1 mM MgCl₂ protease phosphatase inhibitor (Roche))The post-nuclear homogenates obtained by centrifugation (1,000g, 10 minutes) were further fractionated into cytosolic membrane/vesicle fractions by high-speed centrifugation (150,000g, 50 minutes), equal proteins were loaded on a gelSamples were mixed 2× SDS sample buffer incubated for 5 minutes at 100 °CAfter electrophoresis on a 4–20% Tris-glycine gradient gel (Invitrogen), proteins were transferred onto 0.45-µm PVDF membranes (Millipore) for detection of all other proteins then incubated overnight in primary antibodyHRP-conjugated secondary antibody was added the next morning incubated for 1 hour at room temperatureThe blot was developed an Invitrogen ECL kit.

vATPase activity assay

AD transgenic mice were studied at the indicated age point together WT or TRGL littermates as a controlMouse hemi-brain was homogenized in 10× volume of homogenization buffer by 40 strokes in a Teflon-coated pestleLysates were centrifuged at 1,000g for 20 minutes to generate the post-nuclear supernatant (PNS)The PNS was then adjusted to 25% OptiPrep (Sigma-Aldrich, D1556) 50% OptiPrep in HBThe resulting mixture, 2 ml in 25% OptiPrep, was placed at the bottom of a clear ultracentrifuge tube (14 × 95 mm, Beckman Coulter) was overlaid successively 1.5 ml each of 20%, 15%, 14%, 12.5%, 10% 5% OptiPrep in cold HBThe gradients were centrifuged for 18 hours at 100,000g at 4 °C in an SW 40 rotor (Beckman Coulter)Next, 500-μl fractions were collected from the top of the ultracentrifuge tubes analyzed by WB analysisLY-enriched fractions (mixture of 20 μl of each OptiPrep fraction from 15 to 18) were mixed 0.052% NaN₃ for blocking the mitochondrial ATPase activityThe vATPase activity was measured the ATPase Assay Kit (Innova Biosciences, 601-0120) according to the manufacturer’s protocolControl samples were measured in the presence of the vATPase inhibitor concanamycin A (1 μM) (Sigma-Aldrich, C9705), the experimental values were subtracted accordinglyAbsorbance was measured at 650 nm, solutions of P_i were used to generate a standard curve.

Ultrastructural EM analyses

Mice were perfused 2.5% glutaraldehyde 2% PFA in 0.1 M sodium cacodylate buffer, pH 7.4 (Electron Microscopy Sciences)Brains were removed sectioned a vibratome into 50-µm or 100-µm sections placed in fixative solution stored at 4 °CSamples were then treated 1% osmium tetroxide in 100 mM sodium cacodylate buffer pH 7.4 for 30 minutes, washed in distilled water four times (10 minutes per wash) then treated 2% aqueous uranyl acetate overnight at 4 °C in the darkSamples were then washed sequentially dehydrated increasing concentrations of ethanol (20%, 30%, 50%, 70%, 90% 100%) for 30 minutes each, followed by three additional treatments 100% ethanol for 20 minutes eachSamples were then infiltrated increasing concentrations of Spurr’s resin (25% for 1 hour, 50% for 1 hour, 75% for 1 hour, 100% for 1 hour 100% overnight at room temperature) then incubated overnight at 70 °C in a resin moldFor transmission electron microscopy ultrastructural analysis, 70-nm sections were cut a Leica Reichert Ultracut S ultramicrotome a Diatome diamond knife, placed onto grids then post-stained 2% uranyl acetate lead citrateImages were taken a Ceta camera on a Thermo Fisher Scientific Talos L120C transmission electron microscope operating at 120 kV.

For the 100-µm-thick embedded samples for serial block-face scanning electron microscopy (SEM), a diamond wire saw was used to remove excess resin around the embedded tissueThe trimmed block was then glued to a VolumeScope specific SEM stub (Agar Scientific, AGG1092450) a two-part silver conductive epoxy (Ted Pella, H20E EPO-TEK)The sample was further trimmed down to a block face of 1,000 µm × 900 µm 400 µm deep an ultramicrotomeOnly the slides of the block were sputter coated a 30-nm-thick layer of gold, as the bottom of the block was already mounted to the stub the silver conductive epoxy before gold coating, the top of the block was covered during the coating process.

The final prepared sample was imaged an Apreo scanning electron microscope (Thermo Fisher Scientific) equipped a VolumeScope module for serial block-face imaging operating in low vacuum mode at 50 Pa a pole piece mounted backscatter detector, VS-DBSThe brain tissue (M-A) dataset was acquired an accelerating voltage of 2 kV a beam current of 100 pAA total of 509 images were collected a slice cutting thickness of 100 nmThe final image dimension was 8,855 × 9,500 a pixel resolution of 15 nm in x and y a dwell time of 5 µs.

ImmunoEM acid phosphatase histochemistry

Tissue was processed as described aboveSections of 70 nm were cut on a Leica ultramicrotome a diamond knifeThe sections were placed onto carbon formvar 75 mesh nickel grids etched 4% sodium metaperidotate for 10 minutes before being washed twice in distilled water then blocked for 1 hourGrids were incubated 3D6, KDM1/LSD1, LC3 or CTSD antibodies (1:2 dilution) at 4 °C overnightThe next day, grids underwent seven washes in 1× PBS were then incubated in anti-mouse or anti rabbit 10-nm gold secondary (1:50 dilution) for 1 hourAfter this, the grid was washed seven times in 1× PBS twice in distilled waterGrids were then silver enhanced for 5 minutes (Nanoprobes)Grids were finally post-stained 1% uranyl acetate for 5 minutes, followed by two washes in water then stained lead citrate for 5 minutes, followed by a final two washes in distilled waterSamples were then imaged on a Thermo Fisher Scientific Talos L120C operating at 120 kVAcid phosphatase histochemistry: PS/APP mouse brains were transcardially perfused fixative (4% PFA, 1% glutaraldehyde in 0.1 M sodium cacodylate buffer, pH 7.4, containing 0.025% calcium chloride, 5% sucrose 0.075% cytidine 5′-monophosphate (CMP))The brains were removed further immersion-fixed in 4% PFA for 4 hours at 4 °CVibratome sections (50 µm) were cut, rinsed in 0.1 M sodium cacodylate buffer containing 5% sucrose then in 0.05 M Tris-maleate buffer containing 5% sucrose, followed by incubation in the reaction medium (25 mg of CMP, 7 ml of distilled water, 10 ml of 0.05 M Tris-maleate buffer 5% sucrose, 5 ml of 0.025 M manganese chloride, 3 ml of 1% lead nitrate, pH 5.0, filtered #50 paper) for 1 hour at 37 °CAfter washing in Tris-maleate buffer then sodium cacodylate buffer containing 5% sucrose, sections were briefly treated 1% sodium sulfide in sodium cacodylate buffer containing 5% sucrose rinsed well in sodium cacodylate buffer containing 5% sucroseThe sections were then post-fixed in 1% osmium tetroxide processed for EM embedding.

Confocal laser scanning microscopy

Immunocytochemistry was performed as previously described⁵⁸Animals were anesthetized perfused Perfusion Fixative Super Reagent (Electron Microscopy Sciences, 1223SK) after being washed Perfusion Wash Super Reagent (Electron Microscopy Sciences, 1222SK)Brains were dissected immersed in the same fixative for 24 hours, then 40-µm sagital sections were made a vibratomeBrain sections were further stained indicated antibody overnight then visualized Alexa Fluor-conjugated secondary antibodyImaging was performed a Plan Apochromat ×20 or ×40/1.4 oil objective lens on a LSM880 laser scanning confocal microscope the following parameters: eGFP (ex: 488, em: 490–560 MBS 488), mRFP (ex: 561, em: 582–640 MBS 458/561), Alexa Fluor 647 (ex: 633, em: 640–710 MBS 488/561/633) DAPI (ex: 405, em: 410–483) best signal scanning model to exclude crosstalk between each wavelength; image acquisition frame (1,024 × 1,024) scanning mode averaging 4 line-scan, speed 6Thio-S staining: Confocal imaged sections were dehydrated incubated 1% aqueous Thio-S for 8 minutesWash 80% ethanol (2 × 3 minutes), 95% ethanol (3 minutes) ddH₂0 (three times)Analyze slide the combination of DAPI/eGFP/mRFP filter setHuman AD brain staining: 40-µm free-floating sections cut on a vibratome from fixed tissue blocks were washed once in 1× PBS rinsed twice in ddH₂O, followed by incubation in 70% (v:v) formic acid for 12 minutes at 27 °CSections were washed 3 × 5 minutes in ddH₂O incubated for 4 minutes at 105 °C in 1.0 mM EDTA, pH 8.0, to unmask antigens allowed to cool to room temperature on a bench, followed by 3 × 5-minute rinse in ddH₂OSections were blocked for 60 minutes in 5% normal horse serum (v:v) 0.2% Triton X-100 (blocking buffer) incubated primary antibodies for 18 hours at 4 °C in blocking buffer, followed by washing 3 × 5 minutes in 1× PBSIncubation in appropriate secondary antibodies (Invitrogen Alexa Fluor), diluted 1:500 in blocking buffer for 2 hours at 27 °C, was followed by washing 3 × 5 minutes in PBS, autofluorescence was blocked by autofluorescence blocker (TrueBlack, Biotium) following the manufacturerʼs protocolSections were washed 3 × 5 minutes at room temperature mounted aqueous medium VectaShield containing DAPI as a nuclear counterstain (Vector Laboratories).

Vesicle quantification

The same neuronal populations of TRGL single littermate were used as a controlHigh-resolution images were acquired on a Zeiss LSM880 confocal microscope Airyscan a Plan Apochromat ×40/1.4 oil DIC M27 objectiveVesicle quantification analysis was performed as previously described¹³.

Duolink in situ detection

APP-βCTF was assessed Duolink II detection reagents orange (Sigma-Aldrich, DUO92013), as instructed by the manufacturer⁵⁹In brief, cell or brain tissues were incubated overnight at 4 °C in primary APPc (Sigma-Aldrich, A8717, 1:250) JRF/AβN/25 antibody solution then washed incubated in PLA probe plus minus solution for 1 hour at 37 °CTissue sections were washed incubated Ligation-Ligase solution for 30 minutes at 37 °C then incubated amplification-polymerase solution for 100 minutes at 37 °CCell or sections were then mounted Duolink II DAPI viewed a Zeiss LSM880 confocal microscope.

Statistics reproducibility

Statistical parameters, including the definitions value of sample size (n), deviations and P values, are reported in the figures corresponding figure legendsStatistical analyses Prism 8 (GraphPad) were conducted on data originating from at least three independent experimental replicatesStatistical analyses between two groups were performed by a two-tailed unpaired Student’s t-testData are expressed as mean ± s.e.mDifferences were considered significant with P < 0.05Depending on the data analysis, sample size related to number of animals was determined by the standards accepted in the fieldNo specific statistical methods were used to predetermine sample sizes, but sample size was determined based on experience from previous studies^13,60Data distribution was assumed to be normal, but this was not formally testedThe samples were not blinded during initial planning because we wanted to ensure that the number of WT AD mouse models was balanced age sex were matchedThe mice were then randomly assorted for the studies, the investigators were blinded when doing the experiments running data analyses.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Unprocessed scans of all immunoblots statistical source data in the paper are included as Source Data Figs. 1 and 2, respectivelyCorrelative light serial block-face scanning electron microscopy data that support the findings of this study are included as Supplementary Fig. 1 Movie 1Other information that supports the findings of this study is available from the corresponding author upon reasonable request. Source data are provided this paper.

No custom software code was used.

This work was supported by NIH P01AG017617 R01AG062376 to R.A.NWe are very grateful to TYoshimori (Osaka University) for the mRFP-eEGFP-LC3 construct used in transgenic mice to AMCuervo (Albert Einstein College of Medicine) for AV fractionationN2a APPswe was a generous gift from GThinakaran (Morsani College of Medicine, University of South Florida), human AD brain was kindly provided by MGearing (Center for Neurodegenerative Disease), support from an Alzheimerʼs Disease Research Centers grant (P50 AG025688)Also, we thank DrMMercken (Janssen Research Development, Belgium) for generously providing JRF/AβN/25, 3D6, JRF/cAb42/26 antibody.

Authors Affiliations

1. Center for Dementia Research, Nathan SKline Institute, Orangeburg, NY, USA

   Ju-Hyun Lee, Dun-Sheng Yang, Chris NGoulbourne, Eunju Im, Philip Stavrides, Anna Pensalfini, Cynthia Bleiwas, Martin JBerg, Chunfeng Huo, James Peddy, Monika Pawlik, Efrat Levy, Mala Rao & Ralph ANixon

2. Department of Psychiatry, New York University Langone Health, New York, NY, USA

   Ju-Hyun Lee, Dun-Sheng Yang, Eunju Im, Efrat Levy, Mala Rao & Ralph ANixon

3. Thermo Fisher Scientific, Hillsboro, OR, USA

   Han Chan & Cedric Bouchet-Marquis

4. Departments of Biochemistry & Molecular Pharmacology, New York University Langone Health, New York, NY, USA

   Efrat Levy

5. NYU Neuroscience Institute, New York University Langone Health, New York, NY, USA

   Efrat Levy & Ralph ANixon

6. Department of Cellular Neurology, Hertie Institute for Clinical Brain Research, University of Tübingen, Tübingen, Germany

   Mathias Staufenbiel

7. Department of Cell Biology, New York University Langone Health, New York, NY, USA

   Ralph ANixon

Contributions

J.-H.LR.A.Nwere equally responsible for experimental design data interpretation mainly contributed to writing revising the manuscriptJ.-H.L., P.SC.Bconducted the experimentsD.YC.Gperformed EM/IEMC.B.-MH.Ccarried out block-face serial SEM 3D reconstruction analysesE.Iconducted vATPase assaysP.SJ.Pconducted tissue processingD.Y., P.S., A.P., M.B., C.GM.Scontributed to data interpretationM.Rprovided TRGL miceE.Lcritically read the manuscript supervised animal breedingC.HM.Pmaintained animals carried out genotyping.

Corresponding authors

Correspondence to Ju-Hyun Lee or Ralph ANixon.

Competing interests

All authors declare no competing interests.

Peer review information

Nature Neuroscience thanks Louise Serpell the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Publisher’s note Springer Nature remains neutral regard to jurisdictional claims in published maps institutional affiliations.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution reproduction in any medium or format, as long as you give appropriate credit to the original author(s) the source, provide a link to the Creative Commons license, indicate if changes were madeThe images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the materialIf material is not included in the article’s Creative Commons license your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holderTo view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

Reprints Permissions