Faulty autolysosome acidification in Alzheimer’s disease mouse models induces autophagic build-up of Aβ in neurons, yielding senile plaques

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

A utophagy is the principal pathway for lysosomal degradation, maintaining cellular homeostasis by constitutively turning over obsolete proteins and organelles. It is induced further by disease and cell stress to eliminate abnormal proteins, aggregates and damaged organelles [1][2][3][4] . Autophagy encompasses several mechanisms for sequestering substrates and 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 and then closes to form an autophagosome (AP). APs mature to autolysosomes (ALs) upon fusion with LY or endolysosome, which introduces varied cathepsin proteases, other acid hydrolases and vATPase, the proton pump that acidifies AL lumens and activates the hydrolases. LYs are targets of causative gene products and risk factors for AD 5 , including the pathogenic amyloid precursor protein (APP) metabolites APP-βCTF and Aβ 6 that are actively generated within endosomal and autophagic pathways and are normally cleared by LYs 7 .
AD is defined neuropathologically by two lesions: intracellular tau aggregates (neurofibrillary tangles) and neuritic plaques composed of focally swollen (dystrophic) neurites (DNs) 8 , extracellular β-amyloid and many other proteins 9 . Additionally, AVs containing incompletely digested autophagy substrates accumulate progressively within affected neurons at the earliest disease stage 5,10-12 . The molecular basis for autophagy dysfunction in AD, its relationship to APP/amyloid pathology and its pathogenic implications are unclear due, in part, to technical challenges of monitoring ALP abnormalities in vivo in brain. To overcome these limitations, we generated transgenic mice (TRGL) with neuron-specific expression of tandem fluorescence-tagged LC3 (mRFP-eGFP-LC3 or tfLC3), an autophagy adaptor protein selectively associated with AP and AL 13 . The 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 and monitor AD-related ALP deficits, we crossed TRGL and AD model mice that develop either early-onset or late-onset disease pathology. In all five AD mouse models studied, we demonstrated early-appearing deficiencies of lysosomal vATPase activity, autophagy dysfunction in vulnerable neuron populations and accumulation of APP-βCTF and Aβ selectively within poorly acidified AL (pa-AL) well before extracellular β-amyloid deposition. Furthermore, we identified a unique autophagic stress response in more compromised neurons characterized by fulminant proliferation of AVs within perikarya and formation of large membrane blebs packed with Aβ/APP-βCTF-filled AVs. The strongly fluorescent petal-like blebs surrounding a DAPI-positive fluorescent nucleus generate flower-like profiles that we term 'PANTHOS' (poisonous flower). Notably, AV fusion with 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 neuron. Using an extensive Faulty autolysosome acidification in Alzheimer's disease mouse models induces autophagic build-up of Aβ in neurons, yielding senile plaques Articles NATuRE NEuROSCIENCE array of imaging and 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.

Results
Detecting in vivo ALP dysfunction. A tandem mRFP-eGFP-LC3 transgene (tfLC3) driven by the THY-1 promoter is postnatally expressed specifically in neurons. tfLC3 is expressed approximately one-fold higher than endogenous LC3 levels and has no detectable effects on the ALP 13 . Like endogenous LC3, tfLC3 binds to AP membranes and persists after AP-LY fusion as an internalized substrate degraded within AL, ultimately yielding non-fluorescent LYs. The tfLC3 on AP fluoresces yellow-green (eGFP/mRFP) at the neutral pH of AP, but AL maturation upon fusion with LY 14 acidifies the AL, causing fluorescence shifts from yellow to orange and then to red as eGFP fluorescence is quenched below pH 6.0 The sensor is composed of pH-resistant mRFP, pH-sensitive eGFP and LC3. An acidic environment triggers the quenching of the eGFP signal, resulting in the conversion of net yellow signal to red-only signal. In combination with LY marker (pseudo-blue), fully acidified AL (AL) or poorly acidified AL (pa-AL) produce purple or white color, respectively. b, tfLC3 fluorescence change in primary neurons. APs (double arrowheads) were seen at distal levels of axons, and pa-ALs (asterisk) were seen at more proximal levels, whereas fully acidified ALs (arrowhead) were predominantly located near or in the perikaryon. c, Representative fluorescence images from neocortical layer V neurons of TRGL mice co-labeled with the cytoskeleton marker MAP2. Arrowhead denotes fully acidified AL (AL). Scale bar, 10 μm. d, Representative fluorescence images of the tfLC3 fluorescence change under lysosomal acidification altered conditions (CQ) in TRGL mouse brain. Arrowheads denote AL or pa-AL. Scale bar, 20 μm. b-d, experiment was repeated three times independently with similar results.
AL acidification deficiency arises before β-amyloid deposits. We crossed TRGL mice 13 with Tg2576 mice 17 , an AD model that develops β-amyloid plaques starting at 10~12 months of age. ALP 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 and the lysosomal membrane protein LAMP1 (Extended Data Fig. 1b). Hue-angle-based assignment and quantification of AV subtypes in neocortex 13 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) and increased size of pa-ALs  Fig. 1c) and assayed their ATPase activity 18 . Consistent with 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) and 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 and APP51) (Extended Data Fig. 1d). The time course graph indicates age-dependent increased prevalence of pa-AL while vATPase activity declines (Fig. 2h).
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 early in AD model mice and progress with age. a, Representative fluorescence images of tfLC3, co-labeled with CTSD, in neocortical neurons of 5-month-old TRGL and Tg2576/TRGL mouse brains. ALs exhibit a red or purple color without or with CTSD co-localization, respectively, whereas pa-ALs exhibit a yellow or white signal depending on CTSD co-label, respectively. Scale bar, 20 μm. b, Number of pa-ALs in 5-month-old Tg2576/TRGL is elevated compared to neurons in TRGL littermates. n = 243 (TRGL) and n = 245 (Tg2576/TRGL) neurons from three mice. c, pa-AL size in 5-month-old Tg2576/TRGL are larger than neurons in TRGL littermates. n = 243 (TRGL) and n = 245 (Tg2576/TRGL) neurons from three mice. d, Lysosomal vATPase activity is decreased in 6-month-old male Tg2576 compared to WT littermate neocortex. n = 3 mice. e, Representative fluorescence images of 12-month-old TRGL and Tg2576/TRGL mouse brains. Scale bar, 20 μm. f, Number of pa-ALs in 12-month-old Tg2576/TRGL are elevated compared to TRGL littermate neocortical neurons and to 5-month-old Tg2576/TRGL. n = 202 (TRGL) and n = 213 (Tg2576/ TRGL) neurons from three mice. g, Lysosomal vATPase activity is decreased in 12-month-old male Tg2576 compared to WT littermates (and greater than in 6-month-old Tg2576). n = 5 mice. Violin plot colors correspond to the colors of the puncta (white: pa-AL; purple: AL). h, Time course analysis of vATPase activity and pa-AL number in Tg2576 mice. vATPase activity: n = 3 (1.6 months and 5 months) and n = 5 (12 months). pa-AL: n = 243 (1.6 months), n = 245 (5 months) and n = 213 (12 months). Quantitative data are presented as means ± s.e.m., unpaired t-test, two-tailed P value as indicated. a, e, experiment was repeated three times independently with similar results. See also extended Data Fig. 1. mo, month; rel., relative.

Articles
NATuRE NEuROSCIENCE (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 and expand perikaryal circumference. A 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 . Quantitative data are presented as means ± s.e.m., unpaired t-test, two-tailed P value as indicated. c, AV fractionation from 10-month-old Tg2576 mice. Fractions were obtained by pooling five mouse brains. The experiment was repeated two times independently with similar results. d, Schematic representation of the PLA performed using JRF/AβN/25 for APP-βCTF N-terminus and APPc for APP-βCTF C-terminus. e, Representative PLA fluorescence images from N2A-APPswe cells and 10-month-old Tg2576 mouse brain compared to WT controls. Arrowheads denote PLA signal for APP-βCTF. Scale bar, 20 μm. f, Representative PLA fluorescence images from Tg2576/TRGL mouse brain. PLA signals were co-localized with pa-AL, resulting in white puncta. Scale bar, 20 μm. a, c, e, f, The experiment was repeated three times independently with similar results. See also extended Data Fig. 2. IHC, immunohistochemistry; mo, month.

NATuRE NEuROSCIENCE
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 and Extended Data 3b), which reflects a severe deficit of AL maturation and acidification. We observed an identical autophagic neurodegenerative pattern in five different mouse models of AD, including models with accelerated neuropathology onset (5xFAD, TgCRND8 and PSAPP) or delayed onset (Tg2576 and APP51-an exceptionally late-onset model expressing hAPPwt) 23 (Fig. 4e). 5xFAD/TRGL mice develop robust ALP disruption and neuronal degeneration at an early age (starting after 2 months depending on sex) [24][25][26] (Extended Data Fig. 3c-e) and in a more reproducible pattern

NATuRE NEuROSCIENCE
been previously described in a neurodegenerative state 27 . Because 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 and 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

NATuRE NEuROSCIENCE
of the perikaryon ( Fig. 5a and Extended Data Fig. 4). Electron microscopy (EM) analysis on brains of 5xFAD/TRGL mice confirmed the continuity of blebs with the perikaryal cytoplasm and 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 with 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 and fully incorporated AVs (Fig. 5c, box ii: yellow arrows, and Extended Data Fig. 5a, inset: yellow arrowheads) as well as bundles of 6-nm fibers (Fig. 5c, box ii, and Extended Data Fig. 5a, red arrowheads) that are strongly Aβ immunoreactive (Fig. 6d, box ii). In other EM images, AVs and Aβ-positive fiber-containing tubular extensions are seen to be in the process of fusing ( Fig. 5d and 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 and their evident cytoplasmic origin distinguish them from DNs, blebs were further distinguished from DNs, which are enriched with neurofilaments, exhibit weak signal for lysosomal markers (CTSD and LAMP2) as shown by IHF (Extended Data Fig. 5c, arrow) and 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 with serial block-face scanning EM imaging using an Apreo scanning electron microscope. Reconstruction of more than 500 z-plane images recreated the entire PANTHOS neuron in three dimensions (Supplementary Fig. 1 and 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 with 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 expand into 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 and 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 immunoreactivity. Co-labeling of these PANTHOS neurons with DAPI and 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 with 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 and 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, with widths of around 10 nm, approximated the known diameters of fibrillar β-amyloid 28 (Fig. 6d, box ii). Resembling the PANTHOS morphologies in Fig. 5b, Aβ IEM of a PANTHOS neuron with 3D6 additionally detected 3D6-positive AVs packed into perikaryal blebs. Perikaryal AVs were also shown to be continuous with, and incorporated into, the central Aβ-positive network of membrane tubular structures ( Fig. 6e and Extended Data Fig. 6e (3D6 and 4G8 IEM)). IEM with 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 high 29 . However, as accumulating AVs deplete sources of available membrane, APP-rich ER and Golgi membranes join endosomes as major sources of APP-βCTF/Aβ generation. Therefore, ER and Golgi are likely key contributors to the expansion of the amyloid fibril network, supporting AP/AL formation by contributing both membrane and β-amyloid precursor.

Articles
NATuRE NEuROSCIENCE brain was detectable in PANTHOS lesions (n = 3 mice, 105 neurons and 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)  Articles NATuRE NEuROSCIENCE invasion and advanced neurodegeneration. This percentage is likely an underestimate because immunoEM analyses with the nuclear marker KDM1/LSD1 revealed nuclear remnants even after loss of nuclear integrity (Extended Data Fig. 5e). The temporal and 1:1 spatial relationship among PANTHOS, intracellular perinuclear Aβ accretion and amyloid plaque formation, therefore, indicates that the vast majority of amyloid plaques originate from a corresponding individual PANTHOS neuron. The transition from intact nucleated PANTHOS neurons to the more advanced stage of DAPI disappearance with glial invasion of the cell likely represents the loss of cellular integrity and conversion to an extracellular plaque.

Lysosomal permeabilization promotes neuronal cell death.
Lysosomal alkalinization is reported to promote lysosomal membrane permeabilization and cathepsin release into cytosol 30 . Cytosolic and 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 affected. We further examined the association of PANTHOS with lysosomal membrane permeabilization using CTSD IHF. Compared 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 with CTSD. We 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 with Thioflavin S (Thio-S) for the detection of dense-cored senile plaques ( Fig. 8a and 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 and microglia. Neither glial cell type was frequently associated initially with PANTHOS neurons, and, therefore, these cells were unlikely to be a major triggering factor in PANTHOS development. In 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 and 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 and 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 and 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 and A4, respectively), as confirmed by z-stack confocal imaging (Extended Data Fig. 8b).

Discussion
Our transgenic dual-fluorescence probe, identifying autophagic compartments and 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 acidification. Furthermore, 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/ Aβ. The advance of PANTHOS generates an intraneuronal perinuclear 'core' of β-amyloid within membrane tubules. Preliminary 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 forming. Progressive failure of an initially neuroprotective autophagy response in neurons 29 is accompanied by an evolution of PANTHOS toward neuronal cell death involving lysosomal membrane permeabilization, cathepsin release and, ultimately, glial invasion and extracellular plaque formation and 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 extracellularly. The emergence of pH deficits in AL coincided with lowered vATPase activity in brain LYs-the likely molecular basis for the acidification deficit. Declining AL acidification was accompanied by selective build-up of APP-βCTF and Aβ within enlarged pa-ALs. These APP metabolites are known to be both generated and degraded in ALs 12,31 and amphisomes 32 . We also cannot exclude additional delivery of APP-βCTF and Aβ to AL/LY through microautophagy and chaperone-mediated autophagy 33 .

NATuRE NEuROSCIENCE
with neurofilaments, only weakly cathepsin-immunoreactive and LAMP2-immunoreactive and frequently myelinated.
Autophagic stress develops in many congenital lysosomal storage disorders (LSDs). In several of these disorders, Niemann-Pick type C (NPC1) 34 and mucopolysaccharidosis type III (MPS-III), pH has been shown to be elevated 35 . Among LSDs, NPC1 has con-siderable phenotypic overlap with AD (for example, paired-helical filaments, cholinergic neurodegeneration, endosome anomalies, disease acceleration by ApoE4, intracellular Aβ/βCTF elevation and modest amyloid deposition [36][37][38] ); tauopathy has been reported in mouse MPS-III models and intracellular synuclein, and Aβ accumulations are detected in MPS-III brain 39  with these disorders usually do not survive to mid-adult ages may partly explain the infrequency of amyloid plaques. Even individuals with AD due to PSEN1 mutations generally do not develop amyloid plaque pathology until the fourth decade of life, when aging factors may contribute 10,40 . Also, neuronal ALP induction continues to increase in AD brain even as substrate clearance declines 29,41 , compounding autophagic stress and likely increasing βCTF/Aβ generation 12 . PANTHOS neurons account quantitatively for the overwhelming majority of plaques that developed in five different AD models at the ages studied. In young 5xFAD mice, β-amyloid cored plaques, detected with β-amyloid antibodies, showed nearly 1:1 coincidence with a single PANTHOS neuron exhibiting a central nucleus. Even 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 ALs. This 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 advanced. Subsequent microglial and 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 and neuronal cell death. By 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 plaque. This 'inside-out' process accords with and substantiates hypotheses from many investigators 44,45 . In versions of this hypothesis, Aβ and 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 models 46 and, in AD brain, the frequent presence of amyloid surrounding DAPI-positive nuclei 47,48 and neuronal lysosomal hydrolase abundance within extracellular β-amyloid 49 .
Our findings add to mounting evidence that lysosomal acidification and the dysregulation of the vATPase complex are common targets of genetic and metabolic disruptions associated with neurodegenerative disease 50 . Coupled with previous evidence 10,20 , our findings strongly support a pathogenic link between APP metabolites and LY dysfunction in AD. Notably, remediating PSEN1-related lysosomal pH deficits by various means ameliorates autophagy failure and other AD-related pathology in AD models 7,24 . Additional 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 deficit. Beyond 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 and to facilitate assessment of autophagy/lysosome modulators as therapeutic agents.

online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41593-022-01084-8. provided lightness is less than 1, which is the usual case for our data. Hue angle was converted to color using the hue color wheel.

Subcellular fractionation, gel electrophoresis and western blotting. AV prep:
Procedures were performed as previously described 58 . For each mouse genotype, cerebral cortices from five or more brains were pooled. The samples were homogenized and subjected to differential centrifugation to separate a fraction enriched in AVs, LYs and mitochondria as previously described. The different organelles in this fraction were isolated by floatation in a discontinuous gradient of metrizamide (50%, 26%, 24%, 20% and 10%), and the LY-enriched fraction was recovered in the 24-16% interface. A fraction enriched in ER resealed vesicles (microsomes), and the cytosolic fraction was obtained in the pellet and supernatant, respectively, after centrifugation of the supernatant at 100,000g for 1 hour. Cytosol and membrane/vesicle prep: Cerebral cortices from male 5xFAD and WT mouse brain were homogenized with buffer (20 mM Tris-Cl, pH 7.4 with 250 mM sucrose, 1 mM EGTA, 1 mM EDTA, 1 mM MgCl 2 and protease and phosphatase inhibitor (Roche)). The post-nuclear homogenates obtained by centrifugation (1,000g, 10 minutes) were further fractionated into cytosolic and membrane/vesicle fractions by high-speed centrifugation (150,000g, 50 minutes), and equal proteins were loaded on a gel. Samples were mixed with 2× SDS sample buffer and incubated for 5 minutes at 100 °C. After 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 and then incubated overnight in primary antibody. HRP-conjugated secondary antibody was added the next morning and incubated for 1 hour at room temperature. The blot was developed using an Invitrogen ECL kit.
vATPase activity assay. AD transgenic mice were studied at the indicated age point together with WT or TRGL littermates as a control. Mouse hemi-brain was homogenized in 10× volume of homogenization buffer by 40 strokes in a Teflon-coated pestle. Lysates 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) with 50% OptiPrep in HB. The resulting mixture, 2 ml in 25% OptiPrep, was placed at the bottom of a clear ultracentrifuge tube (14 × 95 mm, Beckman Coulter) and was overlaid successively with 1.5 ml each of 20%, 15%, 14%, 12.5%, 10% and 5% OptiPrep in cold HB. The 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 and analyzed by WB analysis. LY-enriched fractions (mixture of 20 μl of each OptiPrep fraction from 15 to 18) were mixed with 0.052% NaN 3 for blocking the mitochondrial ATPase activity. The vATPase activity was measured using the ATPase Assay Kit (Innova Biosciences, 601-0120) according to the manufacturer's protocol. Control samples were measured in the presence of the vATPase inhibitor concanamycin A (1 μM) (Sigma-Aldrich, C9705), and the experimental values were subtracted accordingly. Absorbance was measured at 650 nm, and solutions of P i were used to generate a standard curve.
Ultrastructural EM analyses. Mice were perfused with 2.5% glutaraldehyde and 2% PFA in 0.1 M sodium cacodylate buffer, pH 7.4 (Electron Microscopy Sciences). Brains were removed and sectioned using a vibratome into 50-µm or 100-µm sections and placed in fixative solution and stored at 4 °C. Samples were then treated with 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) and then treated with 2% aqueous uranyl acetate overnight at 4 °C in the dark. Samples were then washed and sequentially dehydrated with increasing concentrations of ethanol (20%, 30%, 50%, 70%, 90% and 100%) for 30 minutes each, followed by three additional treatments with 100% ethanol for 20 minutes each. Samples were then infiltrated with increasing concentrations of Spurr's resin (25% for 1 hour, 50% for 1 hour, 75% for 1 hour, 100% for 1 hour and 100% overnight at room temperature) and then incubated overnight at 70 °C in a resin mold. For transmission electron microscopy ultrastructural analysis, 70-nm sections were cut using a Leica Reichert Ultracut S ultramicrotome and a Diatome diamond knife, placed onto grids and then post-stained with 2% uranyl acetate and lead citrate. Images were taken using 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 tissue. The trimmed block was then glued to a VolumeScope specific SEM stub (Agar Scientific, AGG1092450) using 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 and 400 µm deep using an ultramicrotome. Only the slides of the block were sputter coated with a 30-nm-thick layer of gold, as the bottom of the block was already mounted to the stub with the silver conductive epoxy before gold coating, and the top of the block was covered during the coating process.
The final prepared sample was imaged using an Apreo scanning electron microscope (Thermo Fisher Scientific) equipped with a VolumeScope module for serial block-face imaging operating in low vacuum mode at 50 Pa using a pole piece mounted backscatter detector, VS-DBS. The brain tissue (M-A) dataset was acquired using an accelerating voltage of 2 kV and a beam current of 100 pA.

NATuRE NEuROSCIENCE
A total of 509 images were collected with a slice cutting thickness of 100 nm. The final image dimension was 8,855 × 9,500 with a pixel resolution of 15 nm in x and y and a dwell time of 5 µs.
ImmunoEM and acid phosphatase histochemistry. Tissue was processed as described above. Sections of 70 nm were cut on a Leica ultramicrotome with a diamond knife. The sections were placed onto carbon formvar 75 mesh nickel grids and etched using 4% sodium metaperidotate for 10 minutes before being washed twice in distilled water and then blocked for 1 hour. Grids were incubated with 3D6, KDM1/LSD1, LC3 or CTSD antibodies (1:2 dilution) at 4 °C overnight. The next day, grids underwent seven washes in 1× PBS and were then incubated in anti-mouse or anti rabbit 10-nm gold secondary (1:50 dilution) for 1 hour. After this, the grid was washed seven times in 1× PBS and twice in distilled water. Grids were then silver enhanced for 5 minutes (Nanoprobes). Grids were finally post-stained with 1% uranyl acetate for 5 minutes, followed by two washes in water and then stained with lead citrate for 5 minutes, followed by a final two washes in distilled water. Samples were then imaged on a Thermo Fisher Scientific Talos L120C operating at 120 kV. Acid phosphatase histochemistry: PS/APP mouse brains were transcardially perfused with fixative (4% PFA, 1% glutaraldehyde in 0.1 M sodium cacodylate buffer, pH 7.4, containing 0.025% calcium chloride, 5% sucrose and 0.075% cytidine 5′-monophosphate (CMP)). The brains were removed and further immersion-fixed in 4% PFA for 4 hours at 4 °C. Vibratome sections (50 µm) were cut, rinsed in 0.1 M sodium cacodylate buffer containing 5% sucrose and 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 with 5% sucrose, 5 ml of 0.025 M manganese chloride, 3 ml of 1% lead nitrate, pH 5.0, filtered with #50 paper) for 1 hour at 37 °C. After washing in Tris-maleate buffer and then sodium cacodylate buffer containing 5% sucrose, sections were briefly treated with 1% sodium sulfide in sodium cacodylate buffer containing 5% sucrose and rinsed well in sodium cacodylate buffer containing 5% sucrose. The sections were then post-fixed in 1% osmium tetroxide and processed for EM embedding.
Confocal laser scanning microscopy. Immunocytochemistry was performed as previously described 58 . Animals were anesthetized and perfused with Perfusion Fixative Super Reagent (Electron Microscopy Sciences, 1223SK) after being washed with Perfusion Wash Super Reagent (Electron Microscopy Sciences, 1222SK). Brains were dissected and immersed in the same fixative for 24 hours, and then 40-µm sagital sections were made using a vibratome. Brain sections were further stained with indicated antibody overnight and then visualized with Alexa Fluor-conjugated secondary antibody. Imaging was performed using a Plan Apochromat ×20 or ×40/1.4 oil objective lens on a LSM880 laser scanning confocal microscope with the following parameters: eGFP (ex: 488, em: 490-560 with MBS 488), mRFP (ex: 561, em: 582-640 with MBS 458/561), Alexa Fluor 647 (ex: 633, em: 640-710 with MBS 488/561/633) and DAPI (ex: 405, em: 410-483) with best signal scanning model to exclude crosstalk between each wavelength; image acquisition with frame (1,024 × 1,024) scanning mode with averaging 4 line-scan, speed 6. Thio-S staining: Confocal imaged sections were dehydrated and incubated with 1% aqueous Thio-S for 8 minutes. Wash with 80% ethanol (2 × 3 minutes), 95% ethanol (3 minutes) and ddH 2 0 (three times). Analyze slide with the combination of DAPI/eGFP/mRFP filter set. Human AD brain staining: 40-µm free-floating sections cut on a vibratome from fixed tissue blocks were washed once in 1× PBS and rinsed twice in ddH 2 O, followed by incubation in 70% (v:v) formic acid for 12 minutes at 27 °C. Sections were washed 3 × 5 minutes in ddH 2 O and incubated for 4 minutes at 105 °C in 1.0 mM EDTA, pH 8.0, to unmask antigens and allowed to cool to room temperature on a bench, followed by 3 × 5-minute rinse in ddH 2 O. Sections were blocked for 60 minutes in 5% normal horse serum (v:v) and 0.2% Triton X-100 (blocking buffer) and incubated with primary antibodies for 18 hours at 4 °C in blocking buffer, followed by washing 3 × 5 minutes in 1× PBS. Incubation 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, and autofluorescence was blocked by autofluorescence blocker (TrueBlack, Biotium) following the manufacturerʼs protocol. Sections were washed 3 × 5 minutes at room temperature and mounted with 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 control. High-resolution images were acquired on a Zeiss LSM880 confocal microscope with Airyscan using a Plan Apochromat ×40/1.4 oil DIC M27 objective. Vesicle quantification analysis was performed as previously described 13 .
Duolink in situ detection. APP-βCTF was assessed using Duolink II detection reagents orange (Sigma-Aldrich, DUO92013), as instructed by the manufacturer 59 . In brief, cell or brain tissues were incubated overnight at 4 °C in primary APPc (Sigma-Aldrich, A8717, 1:250) and JRF/AβN/25 antibody solution and then washed and incubated in PLA probe plus and minus solution for 1 hour at 37 °C. Tissue sections were washed and incubated with Ligation-Ligase solution for 30 minutes at 37 °C and then incubated with amplification-polymerase solution for 100 minutes at 37 °C. Cell or sections were then mounted using Duolink II and DAPI and viewed using a Zeiss LSM880 confocal microscope.
Statistics and reproducibility. Statistical parameters, including the definitions and value of sample size (n), deviations and P values, are reported in the figures and corresponding figure legends. Statistical analyses using Prism 8 (GraphPad) were conducted on data originating from at least three independent experimental replicates. Statistical analyses between two groups were performed by a two-tailed unpaired Student's t-test. Data are expressed as mean ± s.e.m. Differences were considered significant with P < 0.05. Depending on the data analysis, sample size related to number of animals was determined by the standards accepted in the field. No specific statistical methods were used to predetermine sample sizes, but sample size was determined based on experience from previous studies 13,60 . Data distribution was assumed to be normal, but this was not formally tested. The samples were not blinded during initial planning because we wanted to ensure that the number of WT and AD mouse models was balanced and age and sex were matched. The mice were then randomly assorted for the studies, and the investigators were blinded when doing the experiments and running data analyses.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

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

Code availability
No custom software code was used.
1 nature portfolio | reporting summary

March 2021
Corresponding author(s): JU-HYUN LEE and RALPH A. NIXON Last updated by author(s): Apr 19, 2022 Reporting Summary Nature Portfolio wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Portfolio policies, see our Editorial Policies and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.

n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted Field-specific reporting Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.

Life sciences
Behavioural & social sciences Ecological, evolutionary & environmental sciences For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Life sciences study design
All studies must disclose on these points even when the disclosure is negative.

Sample size
Depending on the data analysis, sample size related to number of animals was determined by the standards accepted in the field. No specific statistical methods were used to pre-determine sample sizes but sample size was determined based on experience from previous studies (

Replication
The number of biological or technical replicates for each experimental group is listed in the corresponding figure legends. All image analyses were sampled across at least 3 animals. vATPase analysis were sampled across 4 replicates from at least 3 animals. AV fractions were sampled across 2 replicates from the pool of at least 5 animals. Biochemical analyses were sampled across 2 replicates from at least 3 animals. Human AD brain analyses were sampled from 3 individuals. Analysis was performed independently by multiple investigators to ensure reproducibility. All attempts at replication were successful.