Neuroproteasomes regulate endogenous tau paired helical filament formation in an APOE genotype- and age-dependent manner

Abstract

In Alzheimer’s disease (AD), endogenous Tau undergoes a pathogenic transition to form paired helical filaments (PHFs), but the cellular mechanisms driving this process have been elusive. Here, we identify the neuron-specific plasma membrane proteasome ('neuroproteasome') as a critical determinant of Tau proteostasis. Selective inhibition of neuroproteasome function rapidly triggers the de novo formation of endogenous, sarkosyl-insoluble Tau PHFs in primary neurons and mouse brain, which share key biochemical and ultrastructural features with PHFs from human AD brains. The APOE gene has three isoforms (E2, E3, and E4), with APOE4 being the largest genetic risk factor for AD. Neuroproteasome abundance at the plasma membrane is differentially modulated by ApoE isoforms (E2 > E3 > E4) and declines with age. ApoE4 neurons accumulate Tau aggregates following modest neuroproteasome disruption, whereas ApoE2 neurons remain resistant. Our findings delineate a neuron-specific mechanism linking genetic and age-related risk factors to the formation of AD-relevant Tau pathology, and position neuroproteasome function as a potential target to preserve proteostasis.

Introduction

The conversion of normally soluble proteins into aggregated macromolecular assemblies is a pathological hallmark common to many neurodegenerative diseases^1,2. A defining feature of Alzheimer’s disease and a subset of Tauopathies is the transformation of endogenous soluble Tau into aggregated paired helical filaments (PHFs) that accumulate in defined brain regions and correlate with disease progression^3,4. While rare familial forms of tauopathies are driven by Tau mutations, this pathological transition in AD occurs in the absence of Tau mutation or overexpression^3,5. The fundamental cellular mechanisms which can lead to the formation of endogenous PHF-Tau aggregates in AD have remained elusive.

Neurons rely on intricate protein homeostasis (proteostasis) networks to maintain the proteome and prevent aberrant aggregation. However, proteostatic control weakens during aging, the primary risk factor for AD⁶. Furthermore, the strongest genetic risk factor for sporadic AD, the ε4 allele of apolipoprotein E (APOE), markedly exacerbates Tau pathology compared to the common ε3 and protective ε2 alleles^7–9. Despite these strong correlations, little is known mechanistically about how AD risk factors converge to disrupt neuronal proteostasis and promote the formation of Tau filaments. Considering ApoE is primarily a glia-derived secreted extracellular factor¹⁰, we hypothesized that specialized proteostasis machinery at the neuronal plasma membrane might provide a critical link to intracellular Tau stability. We focused on a neuron-specific proteasome complex localized to the neuronal plasma membrane (‘neuroproteasome’)¹¹. While neuroproteasomes are conserved in vertebrates¹², their physiological role and relevance to disease remain undefined.

Here, we demonstrate that neuroproteasome dysfunction actively drives the formation of AD-relevant PHF Tau aggregates. Using selective membrane-impermeant inhibitors, we show that disrupting this plasma membrane-bound proteasome induces de novo aggregation of endogenous Tau in primary neurons and in vivo (wild-type mice and mice expressing humanized Tau). The resulting sarkosyl-insoluble aggregates exhibit key features of human AD PHFs, including characteristic ultrastructure and AD-relevant phosphorylation patterns. We also identify ApoE and its receptor Lrp1 as neuroproteasome interactors and show that neuroproteasome activity and membrane localization are modulated by ApoE isoforms and decline with age—two major risk factors for AD. We provide evidence from human patient tissues supporting the hypothesis that neuroproteasomes link ApoE genotype to Tau proteostasis. Reducing neuroproteasome function lowers the threshold for aggregation in a genotype-specific manner, with ApoE4 neurons exhibiting heightened susceptibility and ApoE2 neurons being relatively protected. Together, our findings define neuroproteasomes as a distinct node of the proteostasis network that integrates genetic and aging-related risk to govern the pathological transition of Tau in Alzheimer’s disease.

Results

Generating selective neuroproteasome inhibitors

Neuroproteasomes share the same essential subunits as cytoplasmic proteasomes, complicating genetic strategies to selectively perturb their localization or function. Instead, to reveal neuroproteasome function, we developed neuroproteasome-specific inhibitors based on Epoxomicin, a covalent proteasome inhibitor with no measured off-target effects. The specificity of Epoxomicin is based on the unique chemistry of the catalytic threonines of proteasome subunits β1, β2, and β5 – the α-amino group of the threonine opens the C-terminal epoxide of Epoxomicin which forms the final inhibitory covalent morpholino adduct¹³. The catalytic residues of Serine and Cysteine proteases do not have α-amino groups, thereby rendering Epoxomicin specific. Numerous N-terminally modified Epoxomicin analogs retain proteasomal specificity without reported off-target effects^14,15, and crucially, Epoxomicin only inhibits active proteasomes.

Previous work used N-terminal biotinylated Epoxomicin (Biotin-Epox) to target proteasome subunits¹⁶. Reasoning that neuroproteasomes are exposed extracellularly¹¹, we hypothesized that introducing a hydrophilic PEG linker between biotin and Epoxomicin's N-terminus would drastically reduce cell permeability¹⁷, thereby rendering the inhibitor selective for neuroproteasomes. Initial screening identified Biotin-PEG24-Epoxomicin (iBEp) (Fig. 1a, Extended Data Fig. 1a-g, 2a) as a lead candidate. iBEp selectively targets catalytic proteasome subunits, confirmed by fluorescent labeling with a Fluor-conjugated PEG24-Epoxomicin and LC/MS-based identification of iBEp-bound targets after streptavidin enrichment in primary mouse neurons and mouse brain (Fig. 1b, Extended Data Fig. 2b, Table S1).

Figure 1: Development of selective neuroproteasome inhibitors.

a, Structure of neuroproteasome-specific inhibitor impermeant Biotin-PEG24-Epoxomicin (iBEp), b, Lysates treated with indicated compounds. Note only three subunits bound by JF-Epox corresponding to covalent modification of proteasome subunits. Note lack of off targets, highlighting specificity retained by N-terminal PEGylated Epoxomicin derivatives, supported by MS analyses in Table S1. (c-h) Orthogonal measures of permeability of iBEp compared to Epoxomicin. c, PAMPA assay to test physiochemical permeability of molecules. N=3 biological replicates. d, Proteasome catalytic activity of membrane and cytosolic fractions of primary neurons treated with iBEp. N=3 biological replicates. e, GFPu degradation assay to measure turnover of unstable GFP. DIV14 primary neurons transfected with ubiquitin-proteasome reporter GFPu treated with indicated drugs. Number of GFP+ cells per 100 cells quantified, each biological replicate plotted. n=6 regions imaged/replicate, N=3 biological replicates, Scale bars=25μm. f, LC3-GFP analysis to visualize autophagosome formation. DIV14 primary hippocampal neurons from LC3-GFP transgenic mice treated with indicated drugs. LC3-GFP signal normalized to MAP2 area plotted, individual biological replicates plottedn=6 regions imaged/replicate, N=3 biological replicates. Scale bar = 25 μm. g, Quantitative PCR-based measurement of proteasome bounceback following iBEp or Epoxomicin (Epox) treatment. Primary DIV14 neurons treated with indicated drugs. Relative Gene Expression plotted, N=2 biological replicates h, Immunoblots of primary WT neurons treated with DMSO, Biotin-PEG24-Azide linker, iBEp, or Epoxomicin. i, Structure of Sulfo-MG132 (SulfoMG). j, PAMPA assay of validated compounds as well as test compounds SulfoMG and MG132 over 24 hours, N=3 biological replicates. k, MDCK assay with indicated compounds over 24 hours. N=2 biological replicates. l, Micrographs from DIV14 primary WT hippocampal neurons transfected with ubiquitin-proteasome reporter GFPu and treated with indicated drugs, quantification to right, N=3 biological replicates. Scale bar = 100 μm. m, Immunoblots of lysates from DIV14 hTau-KI primary neurons treated with indicated compounds for 12 hours and immunoblotted using indicated antibodies. Li-Cor based quantification of p62 and Ubiquitin intensities normalized to Actin, plotted relative to DMSO. N=3 biological replicates. For all plots, p values are indicated, data are mean +/− SEM, and were analyzed by One-Way ANOVA Tukey’s Multiple Comparison Test. Source data and uncropped blots available.

We extensively validated iBEp's cell impermeability using multiple assays. iBEp showed no detectable permeability in the highly sensitive PAMPA-BBB assay¹⁸, being at least 2500-fold less permeable than Epoxomicin (Fig. 1c, Extended Data Fig. 2c). We also observed no iBEp permeability in the MDCK-MDR1 assay¹⁹ (Extended Data Fig. 2d) or into primary mouse neurons assessed by both biotin detection (Extended Data Fig. 2e,f) or LC/MS-based metabolite profiling (1 pmol and 1 pg limit of detection, respectively) (Extended Data Fig. 2e-i). Importantly, iBEp added to primary mouse neurons selectively inhibited proteasomes in the membrane fraction but not the cytosol, confirming its activity at the target location (Fig. 1d, Extended Data Fig 2g-j). LC/MS analyses also confirm that iBEp is stable and is not cleaved in mouse brain or in neurons (Extended Data Fig. 2h-k).

Consistent with its impermeability, iBEp did not elicit conventional cellular responses to cytosolic proteasome inhibition, despite being an active inhibitor. Unlike Epoxomicin, iBEp treatment did not cause accumulation of the ubiquitin-proteasome reporter GFPu²⁰ (Fig. 1e), induce an increase in LC3-GFP levels²¹ (an indicator of autophagosome formation)(Fig. 1f), trigger the proteasome bounceback response²²(Fig. 1g), or increase total intracellular ubiquitin conjugates or p62 levels²³ in primary neurons (Fig. 1h, Extended Fig. 2a). Altogether, our extensive characterization of iBEp shows that it is a specific, stable, active, and cell-impermeable neuroproteasome inhibitor.

To corroborate these findings with an independent tool, we developed Sulfo-MG132 (SulfoMG), modifying the reversible inhibitor MG132¹⁴, with a membrane-impermeant sulfonated biotin linker (Fig. 1i, Extended Data Fig 3a-d). Physiochemical and cellular assays confirmed SulfoMG is also membrane-impermeant, stable, and active (Fig. 1j-m, Extended Data Fig. 3e-g). Biotin pull-downs to reveal SulfoMG-bound proteins revealed the catalytic proteasome subunits as well as some off-targets of MG132 (eg. Calpains²⁴) (Table S1). However, since these are all cytosolic targets, and SulfoMG is cell-impermeable and less potent than MG132 (Extended Data Fig. 3g,h), we suggest that there are no off-targets of SulfoMG in intact cells.

Neuroproteasome inhibition induces endogenous sarkosyl-insoluble Tau aggregates

To investigate the functions of neuroproteasomes in proteostasis, we treated primary mouse neurons with iBEp and identified proteins in detergent-soluble and -insoluble fractions, representing soluble and aggregated species, by quantitative TMT-proteomics (Fig. 2a). Neuroproteasome inhibition caused only 52 proteins to become significantly enriched in the sarkosyl-insoluble fraction, compared to over 250 proteins increasing in the soluble fraction (Fig. 2b,c, Extended Data Fig. 4a, Table S2). Because of the potential importance of failed protein degradation resulting in protein aggregate formation without changing total protein levels, we identified proteins that showed significant and greater than log₂-fold increases in the insoluble fraction with no change in the soluble fraction. This revealed only four proteins: Ckmt1, Tau, Taf15, and Hspb1 (Fig. 2d). Notably, aggregated Tau is a primary hallmark of AD, while Taf15 and Ckmt1 have been linked to Tau tangles^25,26.

Figure 2: Neuroproteasome inhibition drives formation of sarkosyl-insoluble Tau in primary neurons.

a, Schematic to analyze changes in sarkosyl-soluble and insoluble proteomes in response to neuroproteasome inhibition. b, Differential enrichment of proteins in the sarkosyl-soluble fraction of neurons treated with iBEp versus DMSO was analyzed and plotted as log2-fold change vs –log10(P-value). Green dots represent selected proteins enriched in the iBEp treatment compared to DMSO treatment. c, Differential enrichment of proteins in sarkosyl-insoluble fraction of primary neurons treated with iBEp versus DMSO. Green dots represent selected proteins enriched in iBEp treatment. d, Identification of proteins which are exclusively enriched in the insoluble fraction with no corresponding change in the soluble fraction (red). Proteins plotted as differences in log2 fold changes between iBEp insoluble and iBEp soluble. e, Sarkosyl fractionation of primary mouse WT neurons treated with iBEp or Epox. Sarkosyl-soluble and -insoluble fractions immunoblotted using indicated antibodies. LiCor-based quantification of relative Tau levels calculated as follows: Soluble and insoluble Tau intensities were normalized to respective soluble GAPDH intensity. All Tau levels plotted relative to DMSO in either soluble or insoluble fraction. N=4 biological replicates, f, Sarkosyl fractionation of primary hTau-knock-in (hTau-KI) neurons treated with iBEp or Epox. Sarkosyl-soluble and -insoluble fractions immunoblotted using indicated antibodies. LiCor-based quantification of relative hTau levels. N=7 biological replicates, g, Sarkosyl fractionation of primary WT neurons treated with alternative inhibitor IgG-Epox. (Top) Schematic of coupling Epoxomicin or PEG24 linker to IgG. (Bottom) Sarkosyl-soluble and -insoluble fractions immunoblotted using indicated antibodies. LiCor-based quantification of relative Tau levels. N=3 biological replicates h, Sarkosyl fractionation of primary hTau-KI neurons treated with alternative reversible inhibitor SulfoMG or MG132. Sarkosyl-soluble and -insoluble fractions immunoblotted using indicated antibodies. LiCor-based quantification of relative Tau levels. Data (right) are mean ± SEM, N=4 biological replicates. i, Immunoblots of sarkosyl fractionations from DIV14 primary neurons treated with indicated compounds for 12 hours. Sarkosyl-soluble and insoluble fractions immunoblotted using indicated antibodies. LiCor-based quantification of relative Tau levels. N=3 biological replicates. All data are mean ± SEM, One-Way ANOVA Tukey’s Multiple Comparison Test. Source data and uncropped blots available.

Based on these data, we sought to test whether neuroproteasome inhibition indeed drives formation of endogenous sarkosyl-insoluble Tau aggregates. We first confirmed a significant increase in sarkosyl-insoluble Tau upon iBEp treatment in primary mouse neurons by immunoblot (Fig. 2e). No such signal was seen in neurons obtained from Tau-KO mice (Extended Data Fig. 4b) or following treatment with the inactive iBEp analog Biotin-PEG24-IleIleThr (identical to iBEp but lacking the active site epoxyketone warhead) (Extended Data Fig. 4c), confirming specificity. Neuroproteasome inhibition similarly induced sarkosyl-insoluble aggregation of human Tau in primary neurons from hTau-KI mice (Fig. 2f, Extended Data Fig. 4d), which express all six human Tau isoforms at endogenous levels without mutations²⁷. As an orthogonal means of testing whether neuroproteasome inhibition could induce the accumulation of sarkosyl-insoluble Tau, we coupled PEG24-Epoxomicin onto normal mouse immunoglobulin (IgG), which is too large and hydrophilic to penetrate across the plasma membrane (Fig. 2g). Similar to iBEp, we found that IgG-Epox induces the accumulation of sarkosyl-insoluble Tau in primary wild-type (WT) mouse neurons, compared to IgG-PEG24-IleIleThr (IgG-Epox without epoxyketone warhead) alone (Fig. 2g). Finally, we found that SulfoMG induces similar partitioning of hTau into the sarkosyl-insoluble fraction as iBEp in primary hTau-KI neurons (Fig. 2h).

Neuroproteasome inhibition by iBEp and SulfoMG induces Tau aggregation while total proteasome inhibition with the cell-permeant parent compounds Epoxomicin and MG132 does not^28,29, despite Epoxomicin and MG132 inhibiting both cytosolic and neuroproteasomes. We hypothesized that cytosolic proteasome inhibition induces a compensatory response which may clear inclusions. Consistent with this, co-treatment of iBEp and Epoxomicin together does not induce insoluble Tau, compared to iBEp alone (Extended Data Fig. 4e). We considered that cytosolic proteasome inhibition, and not neuroproteasome inhibition, resulted in compensation by the autophagy/lysosomal degradation system^28–30 (Fig. 1f,m). Supporting our hypothesis, we find sarkosyl-insoluble Tau from neurons co-treated with the lysosomal protease inhibitor Leupeptin together with Epoxomicin (Fig. 2i). We interpret these findings to mean that the lysosome is involved in clearing aggregates when cytosolic proteasomes are inhibited, thereby explaining why Epoxomicin masks neuroproteasome inhibition-induced Tau aggregate formation.

Neuroproteasome inhibition-induced Tau aggregates are Paired Helical Filaments

Next, we examined whether neuroproteasome inhibition would induce the formation of sarkosyl-insoluble endogenous Tau aggregates in vivo. Here, we stereotactically injected iBEp into the mouse hippocampus, a region that displays robust Tau pathology in human AD patients and shows some of the earliest pathology in AD³¹. Based on membrane fractionation of hippocampal tissues, we do not detect permeability of iBEp in vivo (Extended Data Fig. 5a). iBEp injection resulted in a significant increase of sarkosyl-insoluble Tau in hippocampi from hTau-KI mice (Fig. 3a). There was no detectable signal in iBEp-injected hippocampi from Tau-KO mice (Extended Data Fig. 5b).

Figure 3: Neuroproteasome inhibition drives formation of endogenous Tau PHFs in vivo.

a, Immunoblots of sarkosyl-soluble and insoluble fractions from hTau-KI mice stereotactically injected with iBEp ipsilaterally and IIT-PEG24-Biotin contralaterally into CA1 of hippocampus. Relative Tau levels calculated as follows: Soluble and insoluble Tau intensities were normalized to respective soluble GAPDH intensity. All Tau levels plotted relative to inactive scaffold in either soluble or insoluble fraction. Data (right) are mean ± SEM, N= 6 biological replicates; (3M, 3F) analyzed by two tailed paired T-Test. b, Sarkosyl-insoluble fractions of hippocampi injected from iBEp-injected mice analyzed by negative stain EM. Negative staining effect was achieved on left, positive staining on right four panels. c, Fractions from (b) stained using anti Tau antibodies conjugated to 6nm gold beads. d,e, Sarkosyl-insoluble fractions of hippocampi injected with iBEp from either hTau-KI (d) or Tau-KO (e) mice. Low magnification image scale bar is 5 μm, high magnification is 200 nm. Despite extensive searching, no fibrils were observed in Tau-KO mice even after 3-fold concentration of samples. All fibre-like densities in Tau-KO samples lack helical characteristics at high magnification. Similar results were obtained from three separate experiments with N=4 animals each. Source data and uncropped blots available.

AD-relevant Tau aggregates are sarkosyl-insoluble Tau paired helical filaments as determined by electron microscopy³². To characterize the insoluble Tau formed upon neuroproteasome inhibition, we analyzed sarkosyl-insoluble fractions of hippocampi from mice following 72 hours of iBEp injection by electron microscopy. Depending on the staining protocol, we obtained both positively and negatively stained samples (Fig. 3b). Twisted filaments which resembled paired helical filaments (PHFs) were readily apparent by negative stain EM analyses (Fig. 3b). These were confirmed to be Tau PHFs by immuno-gold EM (Fig. 3c). We did not detect any straight filaments in our preparations. No twisted filaments could be detected in samples obtained from Tau-KO animals, despite exhaustive searching at low and high magnification (Fig. 3d, e, Extended Data Fig 5c, d). The maximal width of the Tau PHFs formed by neuroproteasome inhibition was 25.9 +/− 3.3 nm, while the crossover spacing was 87 +/− 8.8 nm (Extended Data Fig 5e). Together, these results establish that neuroproteasome inhibition is sufficient to trigger the formation of PHFs from endogenous Tau in the absence of genetic lesions or overexpression.

Neuroproteasome inhibition-induced Tau aggregates have AD-like features

Neuroproteasome inhibition induces a shift of sarkosyl-insoluble Tau that migrates at a higher molecular weight on an SDS-PAGE gel, at approximately 64kDa (Fig. 2e-i, Extended Data Fig. 4c,d, 6a). Similar shifts in molecular weight of Tau are observed in the AD brain and are a hallmark phenotype of multiple neurodegenerative disorders³³, likely reflecting Tau phosphorylation. Using quantitative phosphoproteomics we find a time-dependent increase in Tau phosphorylation after iBEp treatment of primary mouse neurons at 4 of 26 identified sites which correspond to human S202, T205, T217, and S404 (Fig. 4a, Table S3). These sites are established markers of phosphorylated Tau in the human AD brain⁴.

Figure 4: Neuroproteasome inhibition-induced aggregates have AD-like features.

a, (Top) Schematic of experimental setup for phosphoproteomics. (Bottom) Quantitative phosphoproteomics from primary WT neurons treated with iBEp. Heatmap displays all identified Tau phosphopeptides and their relative depletion (pink) or enrichment (green) compared to DMSO controls at indicated timepoints. Phosphopeptides enriched in iBEp treated neurons in green. b, Immunohistochemical analysis of mice stereotactically injected with iBEp to measure Tau phosphorylation in vivo. hTau-KI mice stereotactically injected into CA1 of hippocampus with iBEp ipsilaterally and DMSO injected contralaterally. Mice were collected 72 hours post injection and sections were stained using indicated antibodies: NeuN (blue) and AT8 (pTau,green). Quantification of total AT8 signal intensity in hippocampus per section normalized to NeuN counts in that section. Analysis was done blinded to experimental condition. N=7 (3M, 4F) independent animals, n=2–3 sections/animal. Scale bars=500 μm (Top), scale bar=25 μm (Bottom). c, Immunohistochemical analysis of mice stereotactically injected with orthogonal inhibitor IgG-Epox to measure Tau phosphorylation in vivo. Experiment conducted identically to (b) but with stereotactic injection of IgG-Epox ipsilaterally and IgG-PEG24 contralaterally. N=6 (2M, 4F) independent animals, n=2–3 sections/animal, d, ThioS staining (to mark β-sheet containing aggregates) of sections from mice stereotactically injected with iBEp. Conditions identical to (c), but stained with ThioS and NeuN. N=7 (3M, 4F) independent animals. Inset: representative example of flame-like ThioS positive inclusion in iBEp-treated hippocampus. Data are quantification of number of flame-like ThioS positive inclusions in CA1 per section, N=7 (3M, 4F) animals, n=2–3 sections per animal. Counting and analysis was blinded to experimental condition. Scale bars=100 μm. e, Higher magnification image of flame-like ThioS-positive inclusions (magenta), NeuN (blue). Top, ThioS alone, Bottom, merge with NeuN. Scale bar=20 μm. f, g, Single Z-plane micrographs of sections from iBEp-injected mice co-stained with ThioS (magenta) and AT8 (Green), overlap appears white. (f) Contains two representative examples of flame-like ThioS+ inclusions while (g) displays representative example of thread-like inclusions. Scale bar=20 μm. All data are mean +/− SEM analyzed by paired two tailed T-Test, Source data and uncropped blots available. Source data available.

To validate these data, we performed immunostaining for phosphorylated Tau in either hTau-KI neurons treated with iBEp or sections from hTau-KI mice stereotactically injected with iBEp. hTau-KI mice were stereotactically injected with iBEp and contralaterally injected with vehicle control. We validated the specificity of our antibodies and approach in both immunocytochemistry (Extended Data Fig. 6b,c) and immunohistochemistry (Extended Data Fig. 6d,e). Tau phosphorylation at S202/T205 was significantly elevated in iBEP-treated primary hTau-KI neurons (Extended Data Fig. 6f) and in hippocampi from mice injected with iBEp compared to the contralateral DMSO controls (Fig. 4b). We observed AT8+ (phospho- S202/T205) signal mislocalized to somatodendritic compartments in iBEp-injected hippocampi, which extend into the stratum radiatum and stratum moleculare (Fig. 4b). No Tau S202/T205 phosphorylation was detectable in Epoxomicin-treated primary neurons (Extended Data Fig. 6g) or in hTau-KI mice injected with Epoxomicin compared to contralateral controls (Extended Data Fig. 6h). We reproduce the increase in mislocalized AT8+ dendrites in hTau-KI mice stereotactically injected with IgG-Epox compared to contralateral IgG-PEG24 controls (Fig. 4c).

Thioflavin S (ThioS) stains β-pleated sheets, including aggregated Tau tangles and β-amyloid plaques. We conducted ThioS staining of sections from hTau-KI mice stereotactically injected with iBEp and contralaterally injected with vehicle control. After 72 hours, we find a strong increase in ThioS staining from the CA1 region of hippocampi exposed to iBEp relative to contralateral controls (Fig. 4d). ThioS-positive aggregates appear as flame-like inclusions in both the CA1 pyramidal layer and the granule cell layer of the Dentate Gyrus (Fig. 4d,e, Extended Data Fig. 6i-k). Neuroproteasome inhibition-induced flame-like ThioS-positive aggregates co-localize with phosphorylated Tau (AT8) in single Z-plane images from neurons in the CA1 and DG of hippocampus (Fig. 4f,g). We also observed thread-like ThioS+ aggregates in the hippocampal neuropil (Fig. 4g, Extended Data Fig. 6i-m). Taken together with the biochemical and EM analyses, these results establish that neuroproteasome inhibition rapidly induces aggregation of endogenous Tau into assemblies that resemble those observed in patients with sporadic AD.

Neuroproteasomes associate with ApoE

To understand whether neuroproteasomes are part of the mechanism for Tau aggregation in neurodegenerative diseases such as AD, we sought to gain insights into how neuroproteasomes are regulated. After extensive screening (Extended Data Fig. 7a-c), we generated a conditional knock-in 20S-FLAG transgenic mouse line into the endogenous Psma3 locus (Fig. 5a). We generated mice that expressed FLAG-tagged proteasomes selectively in neurons in a Cre-dependent manner (Extended Data Fig. 7d, e). The FLAG tag does not affect the catalytic activity of the 26S or 20S proteasome relative to controls (Fig. 5b, Extended Data Fig 7f). Moreover, we find that mice expressing 20S-FLAG appear indistinguishable from wild-type littermates, confirming that the endogenous FLAG tag on Psma3 is inert.

Figure 5: Neuroproteasomes associate with ApoE.

a, Schematic for endogenous and conditional tagging of the 20S proteasome with FLAG tag. (Top) Schematic of edited Psma3 gene locus, endogenous Exon 11 flanked with loxP sites, and an insert containing 3x-FLAG tagged Exon 11 was inserted into the 3' UTR of the gene. Transgene referred to as 20S-FLAG. (Right) Schematics of proteasome complexes: 20S and 26S (20S + 19S regulatory particle). b, Catalytic activity of proteasomes isolated from 20S-FLAG transgenic mice. Proteasomes isolated using 26S Immunoprecipitation (IP) (left) or FLAG IPs (right) from BAF53b-Cre mice (black) or 20S-FLAG/BAF53b-Cre mice (blue). AMC fluorescence generated from proteasome-dependent cleavage of Suc-LLVY-AMC plotted. Data are mean ± SEM from N=3 biological replicates, analyzed by Two-Way ANOVA Fisher’s LSD. c, Immunoblots from surface biotinylation of hippocampi from 20S-FLAG mice alone or with BAF53b-Cre. Lysates (Total) and Streptavidin pulldowns (Surface) immunoblotted using indicated antibodies. Merge indicates overlay (yellow) between FLAG (green) and α1–7 blots (red). Similar results obtained from at least N=3 independent animals. d, Co-IP and mass spectrometry (MS) analysis of 20S-FLAG proteasomes from cytosolic and membrane-enriched fractions. Cytosolic and membrane fractions were prepared from whole brains of 20S-FLAG/BAF53b-Cre and immunoprecipitates were isolated using anti-FLAG affinity beads. Samples were subjected to tandem mass tag (TMT) labeling and quantitative mass spectrometry. Differential levels of proteins enriched from FLAG pulldowns from membrane versus cytosol were analyzed and plotted as log2-fold change vs –log10(P-value). Green dots represent proteins enriched in the membrane FLAG IP, blue dots represent 20S core subunits, and gray dots represent proteins in 19S cap. e, Immunoblots of 20S-FLAG Co-IP from cytosolic and membrane fractions of 20S-FLAG alone (Cre -) or 20S-FLAG/BAF53b-Cre (Cre +) mouse brain tissue (inputs). IgG alone (IgG IP) or FLAG antibody (FLAG IP) used for IP indicated. Source data and uncropped blots available.

The 20S-FLAG was efficiently incorporated into the neuroproteasome, as determined by surface biotinylation experiments in primary neurons and hippocampal tissue from 20S-FLAG transgenic mice (Fig. 5c, Extended Data Fig. 7g). To identify neuroproteasome interactors, we extracted membrane proteins from plasma membrane-enriched fractions (Extended Data Fig. 7h) from 20S-FLAG/Baf53b-Cre³⁴ mouse brains followed by FLAG affinity isolation to extract neuroproteasomes and TMT-based IP-MS (Fig. 5d). 64 proteins were enriched in the neuroproteasome fraction and depleted in the cytosolic fraction, including enrichment of ApoE and its receptor Lrp1 (Fig. 5d, Table S4). We focused on ApoE given that the ApoE allele influences the risk for Tau aggregation^8,9,35. We do not observe other ApoE receptors³⁵, such as ApoER2, VLDLR, and LDLR in our proteomics analysis. We validated these proteomic data by affinity isolating FLAG-neuroproteasomes as well as cytosolic FLAG-proteasomes and observed co-purification of ApoE and Lrp1 with the FLAG-neuroproteasome (Fig. 5e). We only observe ApoE and Lrp1 after immunoprecipitating against FLAG from Cre-positive membrane fractions and not from membrane fractions of transgene controls with no Cre expression, cytosolic controls from Cre+ or Cre- brains, or in mock IPs (Fig. 5e).

ApoE isoforms differentially modulate neuroproteasome localization

The most significant genetic risk factor for sporadic AD is the ApoE4 isoform of the APOE gene, while the ApoE3 isoform is neutral and the ApoE2 isoform is considered protective¹⁰. Given the connection between Tau aggregation and ApoE carrier status, we examined whether ApoE isoforms could regulate neuroproteasomes. We obtained human ApoE isoform knock-in mice that each express one of three fully humanized ApoE isoforms³⁶. Surface biotinylation and immunoblotting revealed a strong reduction of neuroproteasome surface localization in the hippocampus of ApoE4-KI mice, compared to ApoE3-KI and ApoE2-KI mice (Fig. 6a).

Figure 6: ApoE isoforms differentially modulate neuroproteasome localization.

a, Surface biotinylation of hippocampi from hApoe-KI mice. Lysates (Total) and Streptavidin pulldowns (Surface) were immunoblotted using indicated antibodies. Experimenters were blinded to genotype. LiCor-based quantification of surface β5 intensity was normalized to total β5 intensity. N=6 biological replicates per genotype (3M, 3F independent mice). Data normalized to ApoE3-KI intensity, analyzed by Two-Way ANOVA Fisher’s LSD. b, c Surface biotinylation (b) and antibody feeding (c) of 20S-FLAG neurons treated with exogenous ApoE lipoparticles. DIV16 primary neurons treated with recombinant ApoE isoforms (rE2, rE3, rE4) or POPC/Cholesterol(PC)-conjugated ApoE isoforms (PC-rE2, rE3, and rE4). (b) Lysates (Total) and Streptavidin pulldowns (Surface) were immunoblotted using indicated antibodies. LiCor-based quantification of surface FLAG and GluN1 intensities were normalized to corresponding total intensities. Data normalized to corresponding vehicle control. N=3 biological replicates. One-Way ANOVA Tukey’s Multiple Comparisons Test. (c) Neurons fed live with anti-FLAG antibodies (magenta) and subsequently stained with MAP2 (green). LiCor-based quantification of surface FLAG fluorescence intensity was normalized to total MAP2 surface area. Data plotted relative to PC alone. N=3 biological replicates, n=21 quantified regions, One-Way ANOVA Tukey’s Multiple Comparisons Test, Scale bar=2 μm. All data are mean ± SEM, source data and uncropped blots available.

We next tested the role of ApoE receptors in modulating neuroproteasome localization using the well-established pan-ApoE receptor antagonist, Receptor-associated protein (RAP)³⁷. (Extended Fig. 8a). We incubated primary mouse neurons with purified RAP and measured surface protein levels using surface biotinylation and immunoblotting. Treatment with RAP rapidly decreased the surface localization of Lrp1, without modifying another membrane protein GluR1 (Extended Data Fig. 8b). We also find that RAP reduced the surface abundance of neuroproteasomes (Extended Data Fig. 8b).

ApoE is a lipoprotein that is released in a lipid-bound form by glia in the brain³⁵. We therefore determined whether extracellular ApoE could modulate neuroproteasome membrane localization and if ApoE lipidation was important for this modulation. We conjugated recombinant ApoE isoforms with a mixture of POPC and Cholesterol (PC) and purified ApoE lipoproteins, referred to as PC-rApoE2, E3, and E4 (Extended Data Fig. 8c) which resemble the endogenous ApoE lipoproteins observed in the human brain³⁸. In primary mouse neurons from 20S-FLAG mice, we find that PC-ApoE4 reduces neuroproteasome membrane localization compared to PC-ApoE3, whereas PC-ApoE2 increase neuroproteasome membrane localization (Fig. 6b). We found no change in neuroproteasome levels with delipidated ApoE particles, lipoparticles alone, nor do we find PC-ApoE-dependent changes in other membrane proteins such as GluN1 (Fig. 6b). We find similar results in primary WT mouse neurons, suggesting that the observed ApoE-isoform dependent effect on localization is not an artifact of the FLAG epitope tag on neuroproteasomes (Extended Data Fig. 8d).

We next performed surface labeling experiments to determine whether ApoE isoforms modified the subcellular distribution and localization of neuroproteasomes. We observed punctate neuroproteasome localization when feeding anti-FLAG antibodies in primary hippocampal neurons (Fig. 6c, Extended Fig. 8e). No staining was observed when feeding with antibodies against MAP2, secondary antibodies alone (Extended Data Fig. 8e-h), or from neurons not transduced with Cre (Extended Data Fig. 8i). We observe that PC-ApoE2 increases neuroproteasome surface localization by nearly twofold and that PC-ApoE4 reduces neuroproteasome surface localization by nearly threefold compared to PC vehicle or PC-ApoE3 controls (Fig. 6c). Similar to our biochemical analyses, delipidated ApoE did not modify neuroproteasome surface localization. These data recapitulate the observations from the hApoE-KI mice and indicate that extracellular lipidated ApoE isoforms can influence neuroproteasome localization.

Neuroproteasomes link ApoE and aging to Tau proteostasis

Having established that neuroproteasome function is both critical for Tau proteostasis and is directly modulated by ApoE isoforms, we next determined whether these ApoE-dependent effects dictate neuronal vulnerability to Tau aggregation. We began by testing whether ApoE-dependent mislocalization of neuroproteasomes is observable in postmortem AD patient tissues. We observed no disease-, genotype-, or region-dependent changes in the cytosolic proteasome population (Extended Data Fig. 9a). We next monitored surface localization of neuroproteasomes by immunoblotting plasma membrane-enriched fractions from two different brain regions of human brains, BA7 and BA4, which were obtained from the same patient (Fig. 7a-d). BA7 has substantial amyloid plaque and Tau tangle pathology as well as neuronal loss, whereas BA4 has little Tau pathology or neurodegeneration, but still has amyloid plaques. The difference between high and low Tau pathology regions across ApoE genotypes provides an opportunity to analyze the relationships between ApoE isoform, Tau, and neuroproteasome levels. We controlled for sample loading both by total protein and then by normalizing against Kv1.2, a protein whose expression has been reproducibly demonstrated as unchanged in the AD brain^39–41. We observed that individuals with the APOE ε4/4 genotype exhibited significantly reduced neuroproteasome levels compared to APOE ε3/3 carriers, regardless of region (Fig. 7b). Neuroproteasome levels were lower in APOE ε3/3 individuals with AD compared APOE ε3/3 controls, consistent with a disease-associated effect (Fig. 7b). When comparing all three groups — unaffected APOE ε3/3, AD APOE ε3/3, and AD APOE ε4/4 — neuroproteasome levels were unchanged in BA4 (low Tau pathology) between controls and AD APOE ε3/3, but were lower in BA7 (high Tau pathology), consistent with an inverse relationship between neuroproteasome levels and Tau pathology (Fig. 7b).

Figure 7: Neuroproteasomes link ApoE and aging to Tau proteostasis.

a, Immunoblots of postmortem human brain tissue of patients with indicated genotypes, with and without AD, from BA4 and BA7 regions. All AD tissues were Braak V/VI. b-d, Quantification of all human patient tissues. 14 AD ApoE3/3, 17 Control ApoE3/3, and 21 AD ApoE4/4 from BA4 (low Tau pathology) and paired samples from BA7 (high Tau pathology). LiCor-based quantification of membrane β5 intensities were normalized to intensities of membrane loading control K_v1.2. Box plots show the median (center line), the interquartile range (25th-75th percentiles plotted as the box), with whiskers representing the 10th and 90th percentiles. b, (left) Effect of ApoE isoform alone; (middle) Effect of AD comparing only ApoE3/3 tissues; (right) Effect of genotype and pathology compared across all samples. Analyzed by (left and middle) T-Test, (right) Two-way ANOVA with Tukey’s Multiple Comparisons Test. c, Quantification of relative membrane β5 to determine interaction between ApoE genotype and Tau pathology, analyzed by two-way ANOVA with Tukey’s Multiple Comparisons Test d, Differences of differences of means of relative β5 intensities between AD BA4 and AD BA7 to isolate ApoE genotype effect independent of pathology, analyzed by paired T-Test. e, Surface biotinylation of hippocampi from WT mice at indicated ages. Total and surface fractions indicated, surface over total/protein plotted relative to 1month levels quantified to right, N=3 biological replicates, analyzed by One-Way ANOVA Tukey’s Multiple Comparisons Test. f, Immunoblots of sarkosyl-insoluble fractions of primary neurons from hTau/hApoE double knock-in (KI) treated with a dose curve of iBEp. LiCor-based quantification of sarkosyl-insoluble Tau intensity. Insoluble Tau intensities were normalized to respective soluble GAPDH intensity, plotted relative to inactive scaffold. Data (middle) are mean ± SEM, N= 4 biological replicates. (right) Proteasome activity assays of membrane fractions of neurons from ApoE/hTau-KI dKI mice plotted against Tau aggregation (LiCor-based quantification of insoluble Tau intensity). Data are mean ± SEM normalized to respective inactive scaffold controls. N=3 biological replicates, analyzed by Two-Way ANOVA Tukey’s Multiple Comparisons Test. Source data and uncropped blots available.

To determine the effects of genotype and Tau pathology, we performed interaction analyses between region and genotype. These analyses pool all APOE ε3/3 carriers, regardless of disease status. This revealed that neuroproteasome reductions in APOE ε4/4 were present regardless of brain region compared to APOE ε3/3, but were stronger in regions with high Tau pathology (Fig. 7c). To more specifically isolate the genotype from the disease-associated effects, we performed a difference of differences analysis between regions only from patients with AD but between genotypes. We found the difference in neuroproteasomes between BA4 and BA7 was significantly greater in APOE ε4/4 individuals (Fig. 7d), indicating a genotype-driven amplification of regional vulnerability. Since BA4 exhibits minimal Tau pathology yet shows a genotype-dependent decrease, these findings collectively support that neuroproteasome levels are reduced by APOE ε4/4 independent of local Tau burden, but are significantly lower in more pathologically affected regions.

While these findings underscore the impact of APOE genotype, aging represents the largest overall risk factor for AD⁶. We therefore next examined the effect of aging on neuroproteasome levels. We observed a progressive age-dependent reduction in membrane-localized neuroproteasomes in wild-type mice starting around 12 months (Fig. 7e). While aging thus represents a major risk factor impacting neuroproteasome levels, the striking differential modulation by ApoE isoforms (Figs. 6, 7a-d) points to a more direct mechanism influencing genotype-specific risk. To determine if this ApoE modulation functionally alters susceptibility to Tau pathology, we quantified Tau aggregation as a function of neuroproteasome inhibition across a range of iBEp doses in primary neurons from mice expressing human Tau alongside ApoE2, ApoE3, or ApoE4. We found no sarkosyl-insoluble Tau aggregates without neuroproteasome inhibition in any line (Fig. 7f, Extended Data Fig. 9b). However, after iBEp treatment, Tau aggregation exhibited a sharply genotype-dependent relationship to neuroproteasome activity. ApoE4 neurons exhibited insoluble Tau accumulation when only ~20% of neuroproteasome activity is lost, whereas ApoE3 and ApoE2 neurons do not cross this threshold until ~60% and ~85% of activity are inhibited, respectively (Fig. 7f, Extended Data Fig. 9c,d). Consistent with this, we find that RAP, which reduces neuroproteasome localization at the plasma membrane similar to ApoE4, also is not sufficient but renders neurons more susceptible to Tau aggregation (Extended Data Fig. 9e). These findings imply that ApoE4 neurons operate with reduced reserve capacity to adapt to proteotoxic insults or age-related decline, whereas ApoE2 neurons possess higher reserves, relative to ApoE3s. The neuroproteasome may therefore serve as a critical determinant of neuronal susceptibility to Tau pathology in an ApoE genotype–dependent manner.

Discussion

In this study, we identify the neuron-specific plasma membrane proteasome (‘neuroproteasome’) as a critical regulator of Tau proteostasis, mechanistically linking it to the major AD risk factors, APOE genotype and aging. We demonstrate that selective neuroproteasome inhibition rapidly drives the de novo formation of endogenous Tau into PHF aggregates. Furthermore, we show that ApoE isoforms differentially modulate neuroproteasome function and localization, establishing genotype-specific thresholds for Tau aggregation. These findings carry significant implications for understanding both Tau pathogenesis and ApoE biology in the context of AD.

A central question in AD research is how PHF Tau pathology forms. Our results offer one potential insight by identifying neuroproteasome disruption as sufficient to trigger pathological conversion from endogenous, wild-type Tau. The rapid formation of PHFs, together with AD-relevant phosphorylation, suggests neuroproteasome impairment could be an early event rendering neurons vulnerable to subsequent pathogenic cascades. Identifying the relationships between neuroproteasomes and these mechanisms should be a central focus of future work. While the speed at which Tau aggregates develop following neuroproteasome inhibition may be surprising given that Tau pathology emerges after months in Tau overexpression and genetic Tauopathy models, there is compelling evidence that individual tangles in these models can form within 24 hours in vivo^42,43. While acute in vivo neuroproteasome inhibition does not fully replicate the chronic, widespread pathology of end-stage AD, it provides critical proof-of-principle for a direct link between neuroproteasome dysfunction and the formation of PHFs, a key feature of AD-relevant tau pathology. The appearance of pathology in specific neuronal populations might reflect differential underlying vulnerability or cell-to-cell variability in neuroproteasome function in vivo, which is not accurately quantifiable with the current tools available. How neuroproteasome inhibition-induced PHF formation relates to other important aspects of tau pathology, such as hyperphosphorylation, secretion, spreading, seeding, and neurodegeneration, are important future directions.

Delineating the pathways by which ApoE isoforms confer dramatically different risks for AD is crucial for developing a mechanistic explanation for AD pathogenesis. Our findings provide a novel mechanistic explanation for how APOE genotype influences Tau pathology, a link that has been poorly understood. ApoE4-induced reduction of neuroproteasome levels is not sufficient to drive Tau aggregation but requires a “second hit” to sufficiently compromise neuroproteasome function under this threshold level (e.g. by aging or potentially by other insults blunting neuroproteasome function). In contrast, ApoE2 neurons are afforded protection not only by higher neuroproteasome expression but also by a greater tolerance to its loss — a dual mechanism of resilience. Based on these data, we suggest ApoE isoforms help establish a neuron's proteostatic reserve capacity by modulating neuroproteasomes. These observations hold in human tissues, where neuroproteasome levels are reduced in APOE ε4/4 patients in multiple brain regions, and show even further reduction in regions with high Tau pathology than those with low Tau pathology. Future work should focus on how ApoE influences neuroproteasome localization through lipidation or other mechanisms, as well as the precise structural mechanism by which neuroproteasome membrane localization is modulated.

ApoE is likely one of multiple yet-to-be-identified genetic mechanisms that can modulate neuroproteasome localization, and consequently, the neuronal capacity for Tau proteostasis. The neuroproteasome interactors we found here will be an informative starting place for identifying and studying relevant mechanisms. These studies will likely gain insights into if and how neuron recognize and respond to neuroproteasome perturbations. Determining why neurons do not trigger canonical protective responses to neuroproteasome inhibition, and what if any responses are enacted, will likely be valuable. Overall, our findings highlight the importance of understanding genetic and cell biological networks that regulate neuroproteasomes, and other neuronal-specific principles of proteostasis, to achieve long-term neuronal health. The neuroproteasome thus emerges as a key molecular node integrating both genetic and age-related risk factors and helps explain differential susceptibility to AD pathology. Consequently, strategies aimed at preserving or enhancing neuroproteasome function, or mitigating the negative impact of ApoE4 on this pathway, represent potential new therapeutic avenues for AD and perhaps related tauopathies.

Methods

Antibodies

Antibody	Vendor	Catalog	Application	Dilution
M anti Actin	Abcam	ab8226	WB	1:1000
R anti Actin	Abcam	ab8227	WB	1:1000
R anti PSMB5	Bethyl	A303–847A	WB	1:1000
R anti PSMA3	Proteintech	11887–1-AP	WB	1:1000
M anti alpha 1–7	Enzo	MCP231	WB	1:1000
R anti GluR1	Cell Signaling	13185S	WB Ab feeding	1:1000 1:100
R anti pan-ApoE	Cell Signaling	13366	WB	1:1000
M anti human-ApoE	D.Holtzman	HJ15.7	WB	1:1000
M anti TFR	Abcam	ab1086	WB	1:1000
R anti TFR	Abcam	ab84036	WB	1:1000
M anti GAPDH	Santa Cruz	sc-32233	WB	1:2500
M anti FLAG M2	Sigma	F1804	WB Ab feeding	1:1000 1:100
M anti His tag	Genetech	A00186	WB	1:1000
M anti StrepII tag	Genetech	A01732	WB	1:1000
R anti Rpt5	Bethyl	A303–538A	WB	1:1000
Ch anti MAP2	Thermo	PA-10005	ICC Ab feeding	1:1000 1:100
R anti Tau (R2295)	Virginia Lee	N/A	WB ICC	1:1000 1:1000
M anti pTau (CP13)	Peter Davies	N/A	ICC	1:2500
M anti pTau (AT8)	ThermoFisher	MN1020	IHC ICC	1:1000 1:2000
R anti NeuN	Sigma	ABN78	IHC	1:1000
M anti KV2.1	DSHB	K14/16	WB	1:1000
R anti PSMB1	ProteinTech	11749–1-AP	WB	1:1000
R anti Lrp1	Abcam	ab92544	WB	1:1000
M anti Tau (DA9)	Peter Davies	DA9	IHC	1:1000
R anti Puromycin	Kerafast	Kf-ab02366–23.0	ICC	1:1000
R anti p62	Abcam	ab109012	WB	1:1000
M anti Ubiquitin	CST	3936S	WB	1:1000
R anti Synaptophysin	Synaptic Systems	101203	IHC	1:1000
GFP-Booster AF488	Proteintech	gb2AF488	ICC	1:1000
Anti mouse IgG 680 LT	LiCor	926–68022	WB	1:10,000
Anti Rat IgG 680 RD	LiCor	925–68076	WB	1:10,000
Anti Rabbit IgG 800 CW	LiCor	926–32211	WB	1:10,000
Streptavidin 680 LT	LiCor	926–68031	WB	1:10,000
Anti rabbit IgG AF647	ThermoFisher	A21244	ICC	1:1000
Anti mouse IgG AF488	ThermoFisher	A11001	ICC	1:1000
Anti mouse IgG AF647	ThermoFisher	A21240	ICC	1:1000
Anti chicken IgY AF488	ThermoFisher	A32931	ICC	1:1000
Anti chicken IgY AF568	ThermoFisher	A78950	ICC	1:1000
Anti chicken Alexa Fluor 488 Plus	ThermoFisher	A32931	ICC	1:1000

Mice

All animal procedures were performed under protocols compliant and approved by the Institutional Animal Care and Use Committees of Columbia University Medical Center. Wildtype mice (C57BL/6J, stock number 027 from Charles River Laboratories) and mice transgenic for 20S-FLAG^flox/flox, BAF53b Cre⁴⁴ (RRID: IMSR_JAX:027826), Tau knockout⁴⁵ (RRID: IMSR_JAX:007251), LC3-GFP, hTau KI⁴⁶, ApoE2 KI, ApoE3 KI, and ApoE4 KI⁴⁷ were used in this study. All mice were used in the C57/B6J background. Non-sibling 20S-FLAG^flox/WT mice were crossed to generate a homozygous 20S-FLAG^flox/flox line. 20S-FLAG^flox/flox were bred with neuronal Cre driver line BAF53b Cre to generate a 20S-FLAG^flox/flox BAF53b-Cre line which expresses 20S-FLAG in neurons. hTau KI mice were crossed with ApoE2 KI, ApoE3 KI, and ApoE4 KI mice and homozygosed to generate homozygous hTau/ApoE2 KI, hTau/ApoE3 KI, and hTau/ApoE4 KI lines. Subsequent generations were confirmed by genotyping. Mice were genotyped using ear punches from weaning-age mice and tail snips from P0–1 pups post-mortem for primary neuron cultures. For the majority of experiments, mice were euthanized with carbon dioxide-induced anoxia and decapitated as a secondary method of euthanasia.

Synthesis of iBEp

Synthesis of Azido-PEG(n)-Epox

Azido-PEG(n)-Epox (where n = 1, 4, 8, 12, or 24) was synthesized using Fmoc solid phase peptide synthesis (SPPS), followed by couping of the epoxy group in solution. All reactions were carried out at rt. SPPS was carried out in a capped 6 mL solid phase extraction tube with a frit and a metal stopcock. The tube was shaken by hand for 1 minute washes and rotated on a tube rotator during incubation periods longer than 10 minutes. A vacuum trap was used to release the solution phase after each deprotecting, coupling, and washing step. Washing steps were carried out between all deprotecting and coupling steps.

100–200 mg of Fmoc-Thr(tBu)-Wang resin (ChemPep Cat# 161702) was swelled in 3 mL of dimethylformamide (DMF) for minimum 1 hour, then Fmoc deprotection was done by incubating the resin in 3 mL of 20% piperidine in DMF for 3 × 10 minutes, followed by 5 × 1 minute washes. Next, 3 eq of Fmoc-Ile-OH (ChemPep Cat# 101001) and 2.9 eq of Hexafluorophosphate Benzotriazole Tetramethyl Uronium (HBTU) was dissolved in 3 mL of DMF and added to the resin, followed by addition of 9 eq of N,N-Diisopropylethylamine (DIEA). The coupling reaction was carried out for minimum 3 hours, followed by 5 × 1 minute washes. The Fmoc group was deprotected and the second Fmoc-Ile-OH was coupled using the same procedure as described above. 1.5 eq of Azido-PEG X-acid (X = 1, 4, 24) and 1.5 eq of HBTU was then dissolved in 3 mL of DMF and added to the resin, followed by addition of 4.5 eq of DIEA. After an overnight reaction period, the resin was washed with 3 mL DMF for 3 × 1 minute, then with 2 mL dichloromethane (DCM) for 2 × 1 minute and left to dry on the vacuum trap for at least 30 minutes. The resin was then moved to a capped glass vial, and a cleavage solution containing 95% trifluoroacetic acid (TFA), 2.5% triisopropylsilane (TIPS), and 2.5% DCM was added on top of the resin. The cleavage reaction occurred for 2 hours with continuous stirring using a magnetic stir bar. Finally, the entire cleavage solution was ran through glass wool to filter out the resin, and the cleaved peptide was precipitated into cold diethyl ether. The precipitate was centrifuged, decanted, reconstituted in 3 mL of water and 100 μL of acetonitrile (ACN), then lyophilized.

The epoxy group was coupled to the Azido-PEG X-Ile-Ile-Thr-OH peptide in solution using the procedure described in⁴⁸. In one glass vial (vial A), Azido-PEG X-Ile-Ile-Thr-OH (compound A) was dissolved in 1 mL of THF. The solution was purged with argon gas. Then, 2 eq of tert-butyldimethylsilyl chloride (TBSCl) and 2.2 eq of imidazole was added to compound A in tetrahydrofuran (THF). The solution was purged with argon gas and left to run overnight. Next, 1.05 eq of hydroxybenzotriazole (HOBt), 1.05 eq of HBTU, and 3 eq of DIEA was added to vial A. The solution was purged with argon gas and left to react for 30 minutes. In a separate vial (vial B), 1.05 eq of (S)-2-Amino-4-methyl-1-((r)-2-methyloxiran-2-yl)pentan-1-one 2,2,2-trifluoroacetate (Combi-Blocks) and 3 eq of DIEA was dissolved in 1 mL THF. The solution was purged with argon gas and left to react for 5–10 minutes. Next, the content of vial B was transfered to vial A under argon gas, and the mixture was allowed to react for 4–5 hours. The reaction was then concentrated using the SpeedVac vacuum concentrator. The concentrated mixture was reconstituted in 1 mL of methanol, then injected directly onto the HPLC for purification, then lyophilized.

Synthesis of Biotin-PEG X-Epox and JF647-Epox

Biotin-PEG X-Epox was synthesized by conjugating Biotin-PEG X1-Epox (where X1 = 0, 4, 8) (BroadPharm) to Azido-PEG X2-Epox (where X2 = 1, 4, 8, 12, 24) using copper-catalyzed azide-alkyne cycloaddition (CuAAc Click reaction). The Click reaction typically consisted of 300 μM CuSO4, 200 μM BTTES, 1500 μM sodium ascorbate, 200 μM azide moeity, and 200 μM alkyne moeity in 1 mL of PBS with 10% isopropyl alcohol. The reaction occurred at rt for minimum 4 hours, or at 37ºC for 2 hours. The crude reaction was injected directly onto HPLC for purification, then lyophilized.

JF647-Epox was also synthesized via CuAAc Click reaction between Janelia Fluor 647-azide and Epox-alkyne (received from Trader group), using the same ratios of reagents as described above. The reaction mixture was used directly to treat HEK cell lysate.

High Performance Liquid Chromatography (HPLC)

The compounds were purified using the Agilent 1290 Infinity II reverse-phase HPLC series in combination with the Agilent ZORBAX Eclipse SB-C18 column, 4.6 × 250 mm, 5 μm. The HPLC method for Azido-PEG X-Epox is a gradient that starts from 95:5 water+0.1% TFA:ACN+0.1% TFA and goes to 5:95 water+0.1% TFA:ACN+0.1% TFA within 30 minutes, flow rate 2 mL/min. The HPLC method for Biotin-PEG X-Epox and SMG-132 is a gradient that goes from 90:10 water:ACN to 25:75 water:ACN within 10 minutes, flow rate 2 mL/min.

Liquid Chromatography-Mass Spectrometry (LC-MS)

The molecular weights of the compounds were confirmed using the UPLC-H-class combined with a XEVO G2 XS mass spectrometer at the mass spectrometry core facility at the chemistry department at Columbia University. The MS column is the Acquity UPLC BEH C8 column, 130Å, 2.1 mm × 50 mm, 1.7 μm. The LC gradient goes from 95:5 water+0.1% formic acid:ACN+0.1% formic acid to 5:95 water+0.1% formic acid:ACN+0.1% formic acid within 5 minutes, with a flow rate of 0.2 mL/min. The LC-MS data was analyzed using the software MassLynx 4.2.

PAMPA and MDCK assays

We contracted Enamine, a commercial CRO with extensive expertise in small molecule synthesis and characterization, to perform cell permeability assays as well as stability assays in mouse brain lysate.

PAMPA conditions

All steps of the PAMPA were carried out according to the pION Inc. PAMPA Explorer™ Manual. The main principle of the assay is the incubation of the compound in a donor chamber (a well in the Donor Plate) with an aqueous buffer, which is separated from the acceptor chamber (a well in the Acceptor Plate) with another buffer by a phospholipid or hydrocarbon membrane fixed on filter support. After the test, concentrations in the corresponding donor and acceptor wells are measured and permeability is calculated.

BBB model was simulated using BBB-1 phospholipid mix. Clozapine, chlorpromazine (high permeability), and ranitidine (low permeability) were used as reference compounds. All compounds were tested in triplicates. Prisma HT buffer, (pH 7.4) containing 50 μM test compounds (0.5% DMSO) was added into the Donor Plate wells. Brain Sink Buffer was added into each well of the Acceptor Plate. Incubation was done at room temperature for 4 hours without stirring. After incubation, samples of the tested compounds were diluted 10-fold with acetonitrile containing internal standard and were analyzed using the HPLC system coupled with a tandem mass spectrometer. Then the apparent permeability coefficient was calculated.

All test articles were detected using Shimadzu HPLC system including vacuum degasser, gradient pumps, reverse phase HPLC column, column oven, and autosampler. The HPLC system was coupled with a tandem mass spectrometer API 3000 (PE Sciex Cat# LC/MS-1SX-1165). The TurboIonSpray ion source was used in positive ion mode. Acquisition and analysis of the data were performed using Analyst 1.6.3 software (PE Sciex).

The formula for calculating logPapp was as follows:

logP_app=log10(−2.303×VdA×(t−τss)×εa×log10(11−Rm100%×C_donC_ref

VD – volume of transport buffer in donor compartment (0.3 ml);

A – surface area of the lipids in the insert (effective growth area of the insert - 0.3 sq.cm);

t – time of the assay, seconds;

C_don – final concentration of test compound in the donor compartment;

C_ref – starting concentration of test compound in the donor compartment;

εa – apparent filter porosity, equals to 0.76;

RM – membrane mass retention, calculated as Rm=(1−C_don+C_accC_ref)×100%;

C_acc – final concentration of test compound in the donor compartment;

τss – steady state lag time, estimated as (54𝑅𝑀+1)×60𝑠.

Values of C_don/C_ref and C_acc/C_ref are practically calculated using optical absorbance data.

MDCK conditions

Canine MDR1 Knockout, Human MDR1 Knockin MDCKII cells (MDR1-MDCKII) (Sigma-Aldrich Cat#MTOX1303) were cultured in 75 cm2 flasks to 80–90% confluence according to the manufacturer's recommendations in a humidified atmosphere at 37°C and 5% CO₂. Cells were detached with Trypsin/EDTA solution and resuspended in the cell culture medium to a final concentration of 4 × 10₅ cells/ml. Then, 400 μl of the cell suspension was added to each well of the HTS Multiwell Insert System and 25 mL of prewarmed complete medium was added to the feeder tray. Confluent MDR1-MDCKII monolayers expressing P-gp were obtained after 4 – 5 days post-seeding. Integrity of the cell monolayers was determined by measuring the trans-epithelial electrical resistance using an epithelial voltammeter (Millipore, Cat # MERS 000 01). Mature MDR1-MDCKII cell monolayer exhibited the TEER values > 100 Ω × cm2. The 24-well insert plate was removed from its feeder tray and placed in a new sterile 24-well transport analysis plate. The inserts were washed with PBS after medium aspiration.

To determine the rate of compounds transport in apical (A)-to-basolateral (B) direction, 300 μL of the test compound dissolved in transport buffer (9.5 g/L Hanks’ BSS and 0.35 g/L NaHCO3 with 0.81 mM MgSO4, 1.26 mM CaCl₂, 25 mM HEPES, pH adjusted to 7.4) was added into the filter wells; 1000 μL of transport buffer was added to transport analysis plate wells.

Ketoprofen, Atenolol, Quinidine, and Digoxin were used as reference compounds.

To determine transport rates in the basolateral (B) to apical (A) direction, 1000 μl of the test compound solutions was added into the wells of the transport analysis plate, the wells in the filter plate were filled with 300 μl of buffer (apical compartment). The final concentration of the test compounds was 1 μM.

The effect of the inhibitor on the P-gp-mediated transport of the tested compounds was assessed by determining the bidirectional transport in the presence or absence of cyclosporine A. The MDR1-MDCKII cells were preincubated for 30 minutes at 37°C with 10 μM of cyclosporine A in both apical and basolateral compartments. After removal of the preincubation medium, the test compounds (final concentration 1 μM) with cyclosporine A (10 μM) in transport buffer were added in donor wells, while the receiver wells were filled with the appropriate volume of transport buffer with 10 μM cyclosporine A respectively.

The plates were incubated for 90 minutes at 37°C under continuous shaking at 100 xg. After this, 75 μl aliquots were taken from the donor and receiver compartments for LC-MS/MS analysis. All samples were mixed with 2 volumes of acetonitrile followed by protein sedimentation by centrifuging at 10000 xg for 10 minutes. Supernatants were analyzed using an HPLC system coupled with a tandem mass spectrometer.

All solutions of test and reference compounds were prepared manually, and further manipulations with the solutions were performed with automation using Opentrons.

The apparent permeability (Papp) was calculated for MDR1-MDCKII permeability assay using the following equation: 𝑷𝒂𝒑𝒑=_𝑽𝑨𝑨𝒓𝒆𝒂×𝑻𝒊𝒎𝒆×[_𝒅𝒓𝒖𝒈]_𝒂𝒄𝒄[_𝒅𝒓𝒖𝒈]_𝒊𝒏𝒊𝒕𝒊𝒂𝒍,_𝒅

VA – volume of transport buffer in acceptor well

Area – surface area of the insert (equals to effective growth area of the insert - 0.7 sq.cm),

Time – time of the assay

[drug]acc – peak area of test compound in acceptor well

[drug]initial,d – initial amount of the test compound in a donor well

Papp is expressed in 10–6cm/sec.

Efflux ratio (Papp(BA)/Papp(AB)) reveals the difference in Papp as a result of active transport. If the efflux ratio is greater than 2, this indicates the occurred active efflux.

To identify the P-gp substrate, the P-gp inhibitor was added to the incubation medium. A decrease of the efflux ratio in the presence of an inhibitor indicates the compound is a P-gp substrate.

The % recovery was calculated using the following equation:

%__recovery=__C_acc×Vacc+_Cd×VdCinitial,_d×Vd×100,

Vacc – volume of compound solution in acceptor well (cm3),

Vd – volume of compound solution in donor well (cm3),

C_acc – peak area of test compound in acceptor well,

Cd – peak area of test compound in donor well,

Cinitial,d – initial amount of the test compound in a donor well.

Stability assays

Mouse brain lysate was prepared from pooled brains of male Balb/c mice (n=10). Mice brains were fragmented into small pieces and homogenized in four volumes of lysis buffer (1% Triton X-100, 100 mM NaCl, 1 mM MgCl₂, 0.1 mM CaCl₂, 50 mM Tris, adjusted to pH=7.4) using SPEX SamplePrep 1600 MiniG. Samples were centrifuged at 6000 g for 20 minutes. Supernatants were decanted and collected. The obtained mouse brain lysate was flash-frozen in liquid nitrogen. Aliquots were stored at −70°C until use.

The brain lysate incubations were carried out in 96-well plates in 5 aliquots of 60 μL each (one for each time point), in duplicates. Test compounds (1 μM, final DMSO and acetonitrile concentration was 0.005% and 1% respectively) were incubated at 37 °C with shaking at 200 xg. Five time points over 120 minutes have been analyzed. The reactions were stopped by adding 240 μL 90% acetonitrile containing internal standard (verapamil) with subsequent brain proteins sedimentation by centrifuging at 5000 xg for 5 minutes. Supernatants were analyzed by the HPLC system coupled with a tandem mass spectrometer. The percentage of the test compounds remaining after incubation in mouse brain lysate and their half-lives (T_1/2) were calculated.

Proteasome bounceback assays

DIV14 cortical neurons were treated with either 1 μM Epoxomicin (Epox) or 1 μM Neuroproteasome Inhibitor (iBEp) for 6 hours. Following treatment, RNA was isolated using the RNeasy Mini Kit (Qiagen Cat#74624) and eluted in 10 μL RNAse free water. cDNA was prepared using the iScript cDNA Synthesis Kit (Bio-Rad Cat# 1708890) according to the manufacturer’s recommendation (all 10 μL RNA was used for the reaction). Mouse-specific qPCR primers from Radhakrishnan et al. 2010 were used (primer sequences listed below). The qPCR reaction was performed in a 384-well plate using the PowerUp SYBR Green Master Mix (ThermoFisher Cat#A25742) and 0.75 μL cDNA per reaction.

Gene	Forward Primer	Reverse Primer
18S	5’-AGGGGAGAGCGGGTAAGAGA-3’	5’-GGACAGGACTAGGCGGAACA-3’
PSMA7	5’-AACGTCTGTATGGCCTTTGC-3’	5’-GTCACTGGGTCCTCCACTGT-3’
PSMB4	5’-TTCACTGGCCACTGGTTATG-3’	5’-CGAACGGGCATCTCTGTAGT-3’
PSMB7	5’-CTGTCTTGGAAGCGGA TTTC-3’	5’-GCAACAACCA TCCCTTCAGT-3’
PSMD12	5’-TCACAGACCTGCCAGTCAAG-3’	5’-AGGTTTTAGTCAGCCGAGCA-3’

Proteasome subunit β5 activity assay

Chymotrypsin-like catalytic activity was assessed using a Suc-Leu-Leu-Val-Tyr-AMC activity assay. HEK cell lysate was diluted in activity assay buffer (25 mM Tris pH 7.4, 2.5 mM MgCl₂, and 10 mM NaCl) supplemented with 10 mM ATP. 20 μM Suc-LLVY-AMC (Enzo Life Sciences Cat# BML-P802) was added to the diluted lysate, and proteasomal activity was assessed in the absence of inhibitors and with serially diluted concentrations of inhibitors (25, 12.5, 6.25, 3.125, and 1.5625 μM). The assay was performed using the Corning^® 96-well Black Flat Bottom Polystyrene NBS Microplate, and read at excitation/emission wavelengths 360 nm/460 nm using the Molecular Devices SpectraMax iD5 Microplate Reader.

Intracellular metabolite extractions

Primary DIV14 cortical neurons were treated with compounds for 24 hours (at concentrations indicated in drug treatments section) and metabolites were extracted precisely according to a detailed published protocol^49,50 and analyzed by LC/MS.

JF647-Epox treatment

0.1 μM JF647-Epox was added to 100 μL of mouse brain lysates (whole mouse brain homogenized in 25mM Tris 7.4, 100mM NaCl, 0.1mM CaCl₂, 2mM MgCl₂, 0.1% NP-40, supernatant after removing nuclei in 500g spin) and left to react for 1 hour at 37ºC. The reaction was quenched by adding 100 μL of SDS loading buffer and ran on a 10% acrylamide gel. The gel was imaged at 680 nm fluorescence before being subjected to immunoblotting as described above.

Drug treatments

Epoxomicin (Selleckchem Cat#S7038) and iBEp were used to treat cultured neurons at 1 μM, IgG-Epox or IgG-PEG24 were used at 0.25 μM, and SulfoMG and MG132 (Selleckchem Cat#S2619) were used to treat cultured neurons at 1 μM. Leupeptin (20 μM, Thermo Cat# 78435) was added every two hours over a 12 hour timecourse based on previous studies^51,52. 2ul of drugs were stereotactically injected at the following concentrations in vivo: 6.4 μM iBEp, 1.2 μM IgG-Epox, 10 μM Epoxomicin. For in vivo hippocampal injections, iBEp concentrations were optimized. Initial dose-ranging studies indicated that concentrations exceeding 12 μM induced seizure activity. Therefore, a concentration of 6.4 μM was selected for all subsequent experiments, as this dose achieved detectable inhibition of membrane-fraction proteasomes without causing observable adverse physiological effects.

GFPu cell penetration assay

Primary hippocampal neurons on coverslips were co-transfected with the Ub-G76V-GFP construct and the pCAGGS-mCherry construct (Addgene, Plasmid #41583) at DIV7 or DIV8. The plasmids (0.5 μg/coverslip for Ub-GFP and 0.1 μg/coverslip for mCherry) and lipofectamine 2000 transfection reagent (0.75 μL/coverslip) were reconstituted in neurobasal media (-phenol red) and incubated together at 37ºC for 30 minutes. 600 μL of NM0 per well was removed and 100 μL of the plasmid/lipofectamine mixture was added. The cells were incubated for 4 hours at 37ºC, then 500 μL of NM0 was added back into the well. At DIV14, the cells were treated with 1 μM Epoxomicin, MG132, iBEp, or SulfoMG132. After a 12-hour incubation, the cells were processed as described in the immunocytochemistry section. The cells were incubated with chicken anti-MAP2 and rabbit anti-mCherry primary antibodies (GeneTex Cat#GTX128508), followed by goat anti chicken 647 and goat anti-mouse 568 secondaries, and GFP Booster Alexa Fluor 488 (ChromoTek Cat#gb2AF488). The coverslips were then mounted and imaged as described in the immunocytochemistry section.

Synthesis of SulfoMG-132

SulfoMG-132 was synthesized via strain-promoted alkyne-azide cycloaddition between DBCO-Sulfo-Link-Biotin (BroadPharm, Cat# 22296) and MG-132-PEG24-azide (Affinity Chemistry Cat#S2619). 25 mM DBCO-Sulfo-Link-Biotin was reacted 1:1 with 20 mM MG-132-PEG24-azide, both resuspended in MilliQ water. MilliQ water was added to make up to the final volume. The reaction was run overnight with continuous stirring at RT and purified using the same method as iBEp. Product fractions from the HPLC were dried by vacuum evaporation and sent for LC-MS analysis. Dried fractions were resuspended in MilliQ water and stored at 4°C.

Calpain/Cathepsin L activity assay

The effective concentrations of MG132 and SulfoMG132 against calpain and cathepsin L was measured using an Ac-Leu-Leu-Tyr-AFC assay and a Z-Phe-Arg-AFC assay (both from Echelon Biosciences Cat#875–20 and 65147–22-0, respectively. The activity assay was carried out using the same procedure as the Suc-LLVY-AMC activity assay, except the plate was read at excitation/emission wavelengths 395 nm/495 nm.

Synthesis of IgG-PEG24-Azide and IgG-PEG24-Epox

Normal mouse IgG was reacted 10 eq of Azido-PEG24-NHS ester (BroadPharm Cat# BP-23542) in PBS to make IgG-PEG24-Azide or reacted with 10 eq of DBCO-NHS ester (BroadPharm Cat# BP-22231) and 10 eq of Azido-PEG24-Epox in PBS to make IgG-PEG24-Epox. The reactions occurred for 2 hours at 37ºC, then filtered using the Amicon filter with 10 kDa MWCO to remove unreacted products.

Immunocytochemistry

Neurons were treated with iBEp, Epox, SulfoMG, MG132, and/or CHX for 12 hours as described in drug treatments section. After the treatment, neurons were washed with cold PBS supplemented with 0.1 mM CaCl₂ and 2 mM MgCl₂ (PBSCM) and fixed in 4% paraformaldehyde (Electron Microscopy Sciences Cat#15710)/4% sucrose diluted in PBS for 8 minutes. Following fixation, neurons were washed 3 times in PBS. Coverslips were blocked in 1X GDB (0.2 mM NaPO₄, 0.2% gelatin, 0.8 M NaCl, and 0.1% TritonX-100). Primary antibodies were diluted as noted in 1X GDB. After incubation, the coverslips were washed in PBS and incubated with secondary antibodies diluted 1:1000 in 1X GDB. Coverslips were washed in PBS and mounted and imaged using a CSU-W1 SoRa confocal microscope. Images are representative maximal Z-projections of multiple optical sections. Images were quantified using ImageJ(CITATION) by integrated density analysis. For all analyses, analyzers were blinded to experimental conditions.

Maximum intensity projections were made for each channel. Maximum intensity projections in ND2 format were imported into Fiji using Bio-Formats Importer plugin. GFPu: GFP+ cells were manually counted and normalized to total number of cells, as counted by nuclei stained by DAPI. To count nuclei, the DAPI channel was thresholded using Huang and nuclei were counted using Particle Analyzer with a particle size of 8–15 μm. LC3-GFP: LC3-GFP fluorescence intensity was measured using Fiji integrated density and normalized to MAP2 surface area. To obtain surface area, the MAP2 channel was converted from a 16-bit image to 8-bit and converted to binary using the Default method. Area fraction was measured from the binary MAP2 image to calculate surface area for normalization. All conditions were normalized to average DMSO value. 3 images were analyzed per replicate. CP13: CP13 and R2295 fluorescence intensity was measured using FIJI integrated density. Background CP13 signal for each image was measured in an area with no neurons or projections. CP13 fluorescence intensity was background subtracted and normalized to R2295 fluorescence intensity or MAP2 surface area and all conditions were normalized to average DMSO value. Antibody feeding: total FLAG surface signal was quantified by FIJI integrated density and normalized to MAP2 surface area as quantified above.

Sarkosyl extraction

For in vitro sarkosyl extractions, primary neurons were treated for 12 hours and lysed in sarkosyl lysis buffer (250 mM NaCl, 2 mM MgCl₂, 0.2 mM CaCl₂, 10% sucrose, 20 mM Tris pH 7.4, 1.5% n-laurylsarcosine (Sigma, 30% stock Cat#61747), protease inhibitors (Roche Cat#11697498001), and phosphatase inhibitors (Sigma Cat#P004)) and collected. Samples were spun at 10,000 × g and debris was removed. Samples were pelleted at 186,000 × g at 10°C for 60 minutes. The supernatants were collected as part of the sarkosyl-soluble fraction and pellets were resuspended in the sarkosyl lysis buffer. Samples were pelleted at 186,000 × g at 10°C for 60 minutes and supernatants were collected and pooled with the sarkosyl-soluble fraction from the first spin. The pellets were taken as the sarkosyl-insoluble fractions, and resuspended in 6M urea, 2M thiourea. For in vivo sarkosyl extractions, mice were euthanized 72 hours following stereotactic intrahippocampal injections, the hippocampi were dissected out on ice and homogenized in a dounce homogenizer in sarkosyl lysis buffer (800 mM NaCl, 10% sucrose, 1 mM EGTA, 25 mM Tris pH 7.4). The lysate was then brought up to 1% n-laurylsarcosine, homogenized, and shaken at room temperature for 1 hour. Supernatants were spun at 186,000 × g at 10°C for 60 minutes and supernatants were collected as the sarkosyl-soluble fraction. Pellets were washed and resuspended with the same buffer without sucrose and 0.2% sarkosyl. Samples were pelleted at 186,000 × g at 10°C for 60 minutes and this process was repeated twice. The pellets were taken as the sarkosyl-insoluble fractions, and resuspended in 6M urea, 2M thiourea for SDS-PAGE analyses and PBS for EM analysis.

Stereotaxic Surgery

Mice were anesthetized with Avertin by weight through an intraperitoneal injection, given local and general painkillers Marcaine and Carprofen along with optical lube to prevent dryness. On the stereotax, the surgical site was shaved and prepared with 3 alternating rounds of betadine and ethanol washes. A vertical incision was made and the open skull area was swabbed with hydrogen peroxide to visualize bregma. The stereotactic equipment was centered on bregma and then adjusted via manipulators to ((+/−)2.5 mm laterally and 2.7 mm posteriorly or (+/−)1.8 mm laterally and 1.9 mm posteriorly. A drill was used to open the holes in the skull on either hemisphere and a needle was lowered 2.2 mm or 1.3mm in order to inject into the hippocampus. The microinjector pump then delivered a dose of 2 μL of the drug/saline solution and retracted after a minute after delivery was complete. The wound was closed and sealed with VetBond.

Electron Microscopy of Tau fibrils, negative stain, and Immuno-EM

4 hTauKI mice (2M, 2F) that were ~5 months old were bilaterally injected with iBEp and collected 72 hours post-surgery. 8 hippocampi were collected and processed for sarkosyl-insoluble fractions as above, but final pellets were resuspended in PBS and applied to glow-discharged carbon-film-coated copper 200 mesh grids (Electron Microscopy Sciences, CF200-CU) for 1 minute. Excess sample was removed by blotting with filter paper. The grids were washed with drops of water for three times, and stained in 2% uranyl acetate for 60 seconds. Following the staining, the grids were lightly blotted to remove the bulk of the stain, and let air dry. To achieve a positive staining effect, grids were more thoroughly blotted before air-drying to reduce staining of the background.

For immunogold labeling, the sample solution was applied to the glow-discharged grid and incubated for 2 minutes. The grid was then incubated on a drop of the blocking solution (0.1% gelatin in PBS) for 10 minutes. Excess blocking solution was blotted, and the grid was incubated in R2295 primary antibody at 1:50 dilution in the blocking solution for 1 hour at room temperature. The grid was then washed three times in the blocking solution, and incubated in the secondary antibody (6nm Au-conjugated Donkey anti-Rabbit) at 1:20 dilution in the blocking solution for 1 hour at room temperature. After blotting the excess antibody solution, the grid was washed three times in water and stained in 2% uranyl acetate for 60 seconds.

The grid was examined with Hitachi HT7800 TEM at 100 kV, and micrographs were routinely acquired at 5000x and 120,000x magnification with a TVIPS TemCam-XF416 CMOS camera.

Immunohistochemistry

For AT8 and Thioflavin S staining under PBS and iBEp conditions, 7 hTauKI mice (4 male and 3 female) at ~5 months of age were used. For AT8 staining under IgG-PEG24 and IgG-Epox conditions, 6 hTauKI mice (2 male and 4 female) at ~4–5 months of age were used. For AT8 staining under PBS and Epox conditions, 3 female hTauKI mice were used at ~4–5 months of age. 72 hours following stereotactic intrahippocampal injections, mice were anesthetized with ketamine/xylazine and perfused with PBS before being perfused with 4% PFA (Electron Microscopy Sciences Cat#15710)) diluted in PBS. Mice were decapitated and brains were removed and fixed in 4% PFA in PBS overnight. Brains were cryopreserved in sequential incubations with sucrose. Brains were embedded in OCT (Sakura Cat# 4583) on dry ice, sectioned on a cryostat and collected on super frost glass slides (VWR Cat# 48311–703). Sections were permeabilized in blocking buffer (10% normal goat serum (Vector Laboratories Cat#S-1000–20), 1% BSA in PBS) supplemented with 0.1% TritonX-100 and incubated in blocking buffer. Sections were incubated with primary antibodies Mouse anti-AT8 and Rabbit anti-NeuN diluted 1:1000 and with secondary antibodies diluted 1:1000. Sections were also stained with DAPI (ThermoFisher Cat# D1306) diluted 1:1000 and submerged in 0.1% Sudan Black B (Sigma Cat#199664) in 70% ethanol to reduce autofluorescence. Thioflavin S (Sigma Cat#T1892) staining done by incubating NeuN or AT8 stained slices in 500 μM Thioflavin S in 50% ethanol for 8 minutes, washed in 50% ethanol, and then submerged in Sudan Black and mounted. They were then mounted and imaged using a CSU-W1 SoRa confocal microscope. Images are representative maximal Z-projections of multiple optical sections with the notable exception of AT8/Thioflavin S staining which are single Z-planes from confocal sections. Images were quantified using ImageJ by integrated density analysis.

Quantification: ROIs of the same size were defined for each hemisphere to include the dentate gyrus and CA1, and CA3 hippocampal regions and for background AT8 signal, which was sampled from the hypothalamus where AT8 signal was minimal. AT8 and NeuN were measured from ROIs using FIJI integrated density. AT8 fluorescence intensity was background subtracted and normalized to NeuN intensity. NeuN intensities did not vary across any measured conditions, NeuN signal intensities also corresponded to the number of NeuN+ nuclei. All conditions were normalized to average DMSO value. ThioS: Sections between 0–50um from the injection site were selected for quantification. ThioS positive signal that resembled a flame-like appearance in the CA1 region of the hippocampus were quantified. ThioS counts are shown as total number of flame-like signal per section. ThioS+ cells were only found in CA1 and DG neuronal cell body layers and not in the molecular layer - only CA1 was counted. Thread-like ThioS+ inclusions were found in the molecular layer but not counted since they could not be consistently assigned to a cell body since the associated cell body may have been in a different section. We could not detect ThioS in non-neuronal cells. ThioS images were quantified by an experimenter blinded to experimental condition and manually counted.

Mass Spectrometry

All proteomic experiments were performed with isobaric tandem mass tagging followed by LC-MS/MS quantitative mass spectrometry.

For quantitative total and phosphoproteomics following iBEp treatment: 3 biological replicates of DIV14 neurons were treated with drugs (as described in drug treatments) and then fractionated (as described in sarkosyl extraction section) using sarkosyl-containing lysis buffer. Insoluble fractions were lysed in 6 M Urea and 2 M Thiourea. Proteins were crashed out of the soluble fraction with high methanol to remove sarkosyl buffer. All samples were digested with Trypsin and peptides were labeled with 16-plex TMTpro reagents. Combined samples were resuspended in Buffer A and offline fractionated. The 96 well plate was combined in checkerboard fashion for a total of 24 final fractions. 12 fractions were dried by speedvac, cleaned on a C18 packed stage tip and eluted. Samples were then resuspended in 5% ACN, 5% formic acid and analyzed by LC-MS3 using a real time search, data dependent method. 12 fractions from the total proteome HPRP set were analyzed on an Orbitrap Lumos mass spectrometer using a 150 minutes method with real-time search. Peptides were detected (MS1) and quantified (MS3) in the Orbitrap and sequenced (MS2) in the ion trap. The MS2 spectra were searched using the COMET algorithm against a Uniprot composite database derived from the mouse proteome containing its reversed complement and known contaminants. The proteins from the 12 runs were filtered to a 1% false discovery rate (FDR) using the target-decoy strategy combined with linear discriminant analysis and were filtered to a <1% FDR. The proteins were quantified only from peptides with a summed SN threshold of >150.

For quantitative phospho-proteomics, MS2 spectra were searched using the COMET algorithm against a Uniprot composite database derived from the mouse proteome containing its reversed complement and known contaminants. Phosphorylation of Ser, Thr and Tyr residues was set as a differential modification. Peptide spectral matches of phosphorylated peptides were filtered to a 1% false discovery rate (FDR) using the target-decoy strategy combined with linear discriminant analysis and the proteins were filtered to a <1% FDR. The proteins were quantified only from peptides with a summed SN threshold of >150. Isolation specificity was set to be greater than 0.5. The combined sample was resuspended in Buffer A (5%ACN, 50 mM AmBic, pH 8.0) and Basic pH reverse phase (bRP) phos sample offline fractionated into a 96-well plate. This plate was run by columns for a total of 12 final fractions, which were dried by speedvac, cleaned on a C18 packed stage tip and then eluted and resuspended in 5% ACN, 5% formic acid. These samples were analyzed by LC-MS2 on an Orbitrap Eclipse mass Spectrometer using a 180 minute high-rez MS2 method. Peptides were detected (MS1) and quantified (MS2).

For co-IP experiments, 4 independent brains from 2M and 2F 20S-FLAG/BAF53b-Cre mice were fractionated into cytosolic and membrane fractions (as described above in Cellular fractionation section) and incubated with anti-FLAG nanobodies (as described above in immunoprecipitation by affinity beads section). Samples were reduced on bead (10 mM TCEP, 55°C for 1 hour), alkylated (18.75 mM iodoacetamide, room temperature for 30 minute) and then digested from the beads with trypsin (2.5 μg trypsin; 37 °C). Following digestion, the resulting peptides were labelled with Tandem Mass Tag (TMT) 10-plex reagents and pooled. The combined sample was resuspended in Buffer A (5%ACN, 50 mM Ammonium Bicarbonate, pH 8.0) and Basic pH reverse phase (bRP) sample offline fractionated into a 96-well plate. This plate was run by columns for a total of 12 final fractions, which were dried by speedvac, cleaned on a C18 packed stage tip and then eluted and resuspended in 5% ACN, 5% formic acid. These samples were analyzed by LC-MS2 on an Orbitrap Eclipse mass Spectrometer using a 180 minute high-rez MS2 method. Peptides were detected (MS1) and quantified (MS2).

The details of the high resolution MS2 run were with Comet Search Parameters of 50 ppm for Peptide Mass Tolerance, 0.02 for Fragment Ion Tolerance, and 2 as the limit for Maximum Internal Cleavage site and Maximum differential per sites. The Reported Quant parameters were set to 0.003 for tolerance, 0 for ms3. The peak picking was set to max, number of isotopes were limited to 2 and MS2 isolation width was set to 0.9. The MS2 spectra were searched using the SEQUEST algorithm against a Uniprot composite database derived from the human proteasome containing its reversed complement and known contaminants. Peptide spectral matches were filtered to a 1% false discovery rate (FDR) using the target-decoy strategy combined with linear discriminant analysis and the proteins were filtered to a <1% FDR. The proteins were quantified only from peptides with a summed SN threshold of >100. Isolation specificity was set to be greater than 0.7.

Mass spectrometry analysis

Peptide-spectrum matches (PSMs) were adjusted to a 1% false discovery rate. PSM filtering was performed using a linear discriminant analysis. To quantify the TMT-based reporter ions in the datasets, the summed signal-to-noise (S/N) ratio for each TMT channel was obtained and found the closest matching centroid to the expected mass of the TMT reporter ion (integration tolerance of 0.003 Da). Proteins were quantified by summing reporter ion counts across all matching PSMs. PSMs with poor quality, or isolation specificity less than 0.7, or with TMT reporter summed signal-to-noise ratio that were less than 100 or 150 or had no MS³ spectra were excluded from quantification.

The list of quantified proteins exported from Proteome Discoverer 2.1 was utilized as the input for our differential expression analysis. The raw values were organized in a matrix where each column represented a sample and each row a protein. To normalize the raw expression values, we began by log₂ transforming the matrix with a +1 for computation. Then we median polished the log-transformed values by subtracting the row median from each row, followed by the subtraction of the column median from each column. The resulting normalized expression values for each sample appeared normally distributed and was comparable across samples.

For the detection of differential regulation, we followed the recommendation outline in ⁵³. An empirical Bayes method was employed on the normalized matrix to detect differences between the iBEp-treated group compared to DMSO. The empirical Bayes method shrinks individual protein’s sample variance towards a pooled estimate, and creates a more stable and powerful inference in differential protein abundance detection.

For the selection of the colors in the heatmap, we carried out feature-scaling of the normalized expression values on a peptide-by-peptide basis. For each gene, this process assigns the largest expression a value of 1, and the smallest expression a value of 0. The remaining values are scaled between 0 and 1 based on where they are relative to the largest and smallest expression values. These were then normalized to DMSO. For instance, a feature-scaled value of 0.5 represents an expression level that is halfway between the lowest expression and the highest expression observed for a gene. In other words, this sample’s expression is 50% of the maximum fold change away from the lowest and the highest expression values at this gene.

Computational analysis of substrate properties

From quantitative MS results, proteins were selected for bioinformatic comparisons based on the relative changes in quantities between iBEp and DMSO control treatments. Proteins with greater than 25% increase in quantities under iBEp treatment over the control were categorized as “increased”. Proteins with changes smaller than ±10% under iBEp treatment were considered “unaffected” by iBEp. The list of proteins was mapped to entries in UniProtKB/Swiss-Prot to obtain the sequences for all subsequence analyses. Aggregation propensity calculation was performed on full-length protein sequences using TANGO version 2.3.1 (Fernadez-Escamilla et al., 2004). The grand average of hydropathy was calculated based on the Kyte and Doolittle hydropathy scale (Kyte and Doolittle, 1982), using codes written with Biopython (Cock et al., 2009). The percentage of disordered regions within the full-length protein was calculated using Metapredict V2 (Emenecker, Griffith and Holehouse, 2021).

Statistics

No statistical methods were used to predetermine sample size. Mice were assigned to experimental groups based on genotype and no additional randomization was required. In many in vivo experiments, mice were bilaterally injected and processed identically so where therefore internally controlled but blinded to analyzers by the technician generating sections. The following experiments were randomized to experimenters (Figures refer to main and include associated supplements): ApoE-KI surface biotinylation, cellular fractionation with postmortem patient tissues, as well as stereotactic injections (tubes were blinded to experimenter). For human tissues, we obtained as many samples as reasonable from two ADRCs based on norms in the field. All statistical analyses were performed using Origin Prism and GraphPad software except for mass spectrometry analysis which was analyzed using Bioconductor DEqMS⁵⁴, accounting for appropriate distribution and variance to ensure proper statistical parameters were applied. Experimental sample sizes were chosen according to norms within the field. Experiments were excluded only if positive or negative controls failed, but no other exclusions were made. The observed magnitude of differences, together with the low replicate variance, permits high power of analysis based on the sample size chosen. Statistical analysis using Student’s t tests, ANOVAs, paired t-tests and the appropriate post hoc tests were performed as described in each figure legend. Data distribution was assumed to be normal but this was not formally tested unless specifically states. Data collection and analysis were not performed blind to the conditions of the experiments unless otherwise indicated (for example, in many of the imaging and in vivo experiments, analyzers were blinded to experimental condition).

Microscopes

All images were acquired with a Nikon Ti2-E stand and CSU-W1 or CSU-W1 SoRa spinning disks with ORCA-fusionBT Hamamatsu cameras. Images were maximum intensity projected except for the colocalization experiments between ThioS and AT8 which are presented as single Z-planes.

Generation of tagged 20S DNA constructs

Plasmids containing full length mouse PSMA1, 2, 3, 7, and PSMB1, 2, 3, 4, 5, 6, and 7 were acquired from Dharmacon Inc. 6XHis and StrepII constructs were designed as c-term tags with GGS and GGA linkers, respectively and were cloned into pCAGEN (Addgene Plasmid #11160) for transfection and pLenti hSynapsin Cre WPRE (Addgene Plasmid #86641) for lentivirus production. Q5 DNA polymerase (NEB Cat#M0491) was used in PCRs to amplify proteasome subunit sequences with respective tags using PCR conditions recommended by the manufacturer. Amplified fragments were run on a 1% agarose gel and visualized to confirm PCR and purified with a commercial kit (Machery Nagel Cat#740609.50). pCAGEN was digested with EcoRV (NEB Cat# R0195) and pLenti hSynapsin Cre WPRE was digested with AgeI and EcoRI (NEB Cat# R3552 and R0101, respectively). Digested plasmids were run on a 1% agarose gel and gel extracted using a standard kit (Machery Nagel Cat#740609.50). Purified vectors and fragments were assembled using Gibson assembly (NEB Cat#E2611) and constructs were transformed into competent cells, NEB5α competent cells (NEB Cat#C2987) for pCAGEN constructs, and NEBStable competent cells (NEB Cat#C3040) for pLenti. Single colonies were sequence verified using Sanger sequencing and preserved in glycerol stocks. All plasmids used in this study were grown up and purified using standard kits from commercial vendors.

Lentiviral production

LentiX293T cells were transfected using the standard protocol in normal culture medium with polyethylenimine (PEI). Cells were transfected with PEI, envelope pMD2.G (Addgene Plasmid #12259), packaging psPAX2 (Addgene Plasmid #12260) and transfer plasmid. As per the standard protocol, the media was aspirated and replaced with OptiMEM (Gibco Cat# 31985062) supplemented with 10 mM sodium butyrate (Sigma Cat# B5887) and 1 mM sodium pyruvate (Sigma Cat# P2256) 24 hours post transfection. Virus-containing media was collected 72 after transfection, concentrated with LentiX Concentrator (Takara Cat#PT4421–2), and resuspended in Neurobasal (Gibco Cat# 21103049).

Transfection and Transduction

HEK293T and LentiX293T cells were transfected using a polyethylenimine (PEI) protocol. Primary neurons were transfected using lipofectamine 2000 (Invitrogen Cat#11668019) according to the manufacturer's protocol. iSyn Cre AAV was obtained from the Columbia University Zuckerman Institute Virology Core.

Cell Culture

HEK293T-LentiX cells (Takara Cat#632180) were cultured in DMEM supplemented with 10% fetal bovine serum (Corning Cat#35–015-CV), 2 mM glutamine, 1 mM sodium pyruvate (Sigma Cat#P2256) and penicillin/streptomycin (100 U/mL and 100 μg/mL, respectively). For wildtype primary cultures, E17 C57Bl/6 mouse cortical and hippocampal neurons were maintained in Neurobasal Medium (Gibco Cat# 21103049) supplemented with 2% B-27 (Gibco), penicillin/streptomycin (100 U/mL and 100 μg/mL respectively), and 2 mM glutamine. For transgenic primary cultures, P0–1 postnatal mouse cortical or hippocampal neurons were maintained in Neurobasal-A Medium (Gibco Cat#10888022) supplemented with 2% B-27, penicillin/streptomycin (100 U/mL and 100 μg/mL, respectively), and 2 mM glutamine. Standard neuronal culture protocols were followed to harvest and grow the cell and neuronal cultures. Whole litters were used to generate neuronal cultures and sex of the pups was not monitored.

Generation of 20S-FLAG mice

20S-FLAG mice transgenic genomic sequence was designed in house, contracted out to Biocytogen Pharmaceuticals and generated using CRISPR/Cas9 editing. Zygotes were microinjected with Cas9-RNP complexes, selected for in vitro, and founders were identified by genotyping. F0 founders were crossed to produce F1s which were confirmed by genotyping and by southern blot analysis to confirm the lack of random integration and correct insertion of the target sequence into the endogenous locus. Subsequent generations were confirmed by genotyping.

Western blotting

To ensure that equal amounts of protein were loaded, samples were analyzed by the RC DC™ assay (Bio-Rad Cat#5000121). The samples were prepared following the RC DC™ user manual and the absorbance was read at 750 nm on the Molecular Devices SpectraMax iD5 Microplate Reader. Gels were run using the standard western blot protocol either on 10% Criterion TGX gels (Bio-Rad Cat#5671034) or on 10% gels made in the laboratory. All antibodies were made up in a blocking buffer (5% BSA in 0.1% TBST). Western blots were imaged on a Licor Odyssey M Imager. Images were quantified using ImageStudio by standard densitometry analysis, which accounts for a large range of signal intensities using quantitative IR secondaries. A rectangle of the same area was drawn for each band for a given gel and signal intensities were output by ImageStudio.

Cellular fractionation for plasma membrane enrichment

Cellular fractionation experiments were performed as previously described⁵⁵. For cellular fractionation experiments with primary neurons, cultured neurons were lysed in a hypotonic lysis buffer (20 mM Tris pH 7.4, 5 mM NaCl, 2 mM MgCl₂, 0.2 mM CaCl₂, complete protease inhibitor cocktail tablet (Roche Cat#11697498001)) supplemented with 0.1% GDN and adjusted to 100 mM NaCl and collected. Nuclei were pelleted at 400 × g, heavy membranes were pelleted at 5000 × g, and the supernatant containing plasma membrane enriched samples was pelleted at 75,000 × g for 30 minutes. Pelleted membranes were washed twice by homogenizing in lysis buffer and re-pelleted. Following two washes, membranes were collected and prepared for SDS PAGE. For cellular fractionation with brain tissue, whole tissue from mouse brain was homogenized in lysis buffer (20 mM Tris pH 7.4, 100 mM NaCl, 2 mM MgCl₂, 0.2 mM CaCl₂, complete protease inhibitor cocktail tablet). Nuclei were pelleted at 400 × g, and the pellet was washed in lysis buffer and re-pelleted. The supernatant containing membranes was brought up to 0.1% GDN, heavy membranes were pelleted at 5000 × g, and the supernatant containing plasma membrane enriched samples was pelleted at 75,000 × g for 30 minutes. Pelleted membranes were rinsed and then resuspended, pelleted, and this was repeated three times.

Surface Biotinylation and Streptavidin Pulldown

Surface biotin-labeling was performed as previously described⁵⁵. For surface biotinylation with hApoE-KI mice, brains were obtained from 6 ApoE2-KI mice (3 male and 3 female), 6 ApoE3-KI mice (3 male and 3 female), and ApoE4-KI mice (3 male and 3 female) at ~12 months of age. For surface biotinylation with 20S FLAG mice, 4 mice were used. Mouse brain tissue was recovered in cutting ACSF (195 mM sucrose, 10 mM NaCl, 15 mM glucose, 26 mM NaHCO₃, 2.5 mM KCl, 1.25 mM NaH₂PO₄, 0.5 mM CaCl₂, and 1 mM MgCl₂, bubbled with carbogen) at 4°C with light agitation for 30 minutes prior to biotin labeling. Mouse brain tissue and cultured neurons were washed in pH 8.0 cold PBSCM (Gibco Cat#10010023) and incubated in 0.5 mg/mL Sulfo-NHS-LC-Biotin (Thermo Fisher) dissolved in PBSCM at 4°C for 10 minutes. The labeling reaction was quenched in 50 mM glycine in two sequential 10-minute incubations at 4°C. Mouse brain tissue and cultured neurons were collected and homogenized in RIPA buffer (50 mM Tris pH 7.4, 250 mM NaCl, 1% Triton X-100, 0.5% Sodium Deoxycholate, 0.1% SDS, 5 mM EDTA, complete protease inhibitor cocktail tablet (Roche Cat#10010023)). Samples were pelleted at 20,000 × g for 10 minutes at 4°C and the supernatant was taken for streptavidin pulldown. For streptavidin pulldowns, lysate was incubated with high-capacity streptavidin agarose beads (ThermoFisher Cat#20353) overnight and washed 5 times with RIPA buffer before elution in SDS sample buffer supplemented with 25 mM biotin.

LLVY-based proteasome activity assay and ATPase assay

Chymotrypsin-like catalytic activity was assessed using a Suc-LLVY-AMC activity assay. Lysate was collected and diluted in activity assay buffer (25 mM Tris pH 7.4, 2.5 mM MgCl₂, and 10 mM NaCl) supplemented with 2 mM ATP. 10 mM Suc-LLVY-AMC (Enzo Life Sciences Cat#BML-P802) was added to the diluted lysate, and proteasomal activity was assessed +/− 10 μM Epoxomicin by measuring fluorescence intensity at 460 nm. ATPase assays were performed using the ATPase Assay Kit (Abcam Cat#ab234055), according to manufacturer instructions.

Immunoprecipitation by affinity beads

Immunoprecipitation (IP) was performed in a batch format. For FLAG IP from the membrane protein extract, ChromoTek DYKDDDDK Fab-Trap™ Agarose resin (Proteintech Cat#hfa) was equilibrated with LSB buffer. The extract from each brain along with about 75ul of beads was added to equilibrated beads and incubated with end-to-end rotation. The beads were washed and proteins were eluted with 2x SDS sample loading buffer.

FLAG IP from cytosolic fractions were performed similarly with the following modifications. About 50 μl beads slurry was equilibrated in the PD buffer (20 mM Tris-HCl, 100 mM NaCl, 2 mM MgCl₂, 0.5 mM ATP) for each IP reaction. The IP reaction was then incubated and proteins were eluted with a 2x SDS sample loading buffer. UBL IP from cytosolic fractions were performed using UBPBio Rapid 26S proteasome Purification kit (UBPBio Cat#J4310). The affinity beads were prepared by combining GST-Ubl with equilibrated glutathione agarose beads slurry for each IP reaction.

RAP Purification

Stable RAP expression plasmid was obtained from (Kerafast Cat#EMD008) and was transformed into T7 Express lysY Competent E. coli (NEB Cat# C3010I). E. coli containing the plasmid were cultured, induced with Isopropyl β-D-1-thiogalactopyranoside (Hello Bio Cat# HB3941), and incubated⁵⁶. Cultures were pelleted and resuspended in Resuspension Buffer (50 mM Tris HCl, 50 mM EDTA, 15% sucrose, pH 8.0). Cells were lysed on ice in 1 mg/mL lysozyme (ThermoFisher Cat#89833) and 1:1000 DNase (Worthington Biochemical Cat# 9003–98-9). 0.02% TritonX-100 (v/v) and complete protease inhibitor cocktail tablets (Roche Cat# 04693116001) were mixed into the lysate. Lysate was passed consecutively through 18 and 25 gauge needles and centrifuged. The supernatant was removed and incubated with Glutathione Sepharose resin (Cytiva Cat# 17075601). Beads were packed into a gravity column and washed with 5 column volumes in Buffer A (20 mM HEPES, 100 mM KCl, 0.2 mM EDTA, 5% glycerol, and complete protease inhibitor cocktail tablets). RAP-GST was eluted in Buffer A with 25 mM glutathione (Fisher Scientific Cat#AAJ6216606). Fractions confirmed by Coomassie were dialyzed in a dialysis buffer (20 mM HEPES, 50 mM NaCl, and 5% glycerol at pH 7.4) and incubated with Thrombin beads (Innovative Research Cat# IHUTHRB100UG). Beads were removed and RAP was purified using an AKTA pure with a Superdex 200 increase column (Cytiva Cat#28990944) in SEC buffer (150 mM NaCl, 0.2 mM EDTA pH 8, and 10% glycerol). Fractions confirmed by Coomassie were concentrated using a 30 kDa MWCO filter and concentration was assessed using a BCA assay (ThermoFisher Cat#23225). RAP was stored in single use aliquots at −80°C.

Lipidated ApoE particles

Lipidated ApoE 2, 3, and 4 (PC-ApoE) were prepared from recombinant ApoE2, 3 and 4 (AlexoTech Cat#AE-100–50, AE-101–50, and AEN-102–10, respectively) as previously described⁵⁷. Samples were purified after lipidation using superose 6 increase GL 10/300 columns using Buffer A (20 mM Phosphate buffer, 50 mM NaCl, pH 7.4) at a flow rate of 0.5 mL/minute. To show negative stain images of the ApoE isoforms lipidated POPC and cholesterol, copper grids with carbon film were glow discharged and incubated with 30 μL of sample. Samples were then washed 3 times with water and then stained with uranyl formate. Grids were blotted with a filter paper and imaged with a JOEL TEM microscope at 50k magnification (0.212492 nm/pixel).

Antibody feeding

Antibody feeding was performed as previously described⁵⁸. On DIV13, neurons were treated with 400 nM PC-ApoE. 24 hours later, hippocampal neurons were washed with PBSCM. On ice, hippocampal neurons were treated with Mouse anti-FLAG M2 antibodies (1:100) for treatment conditions, and additionally with Chicken anti-MAP2 antibodies (1:100) and Rabbit anti-GluR1 antibodies (1:100) for controls. Critically, all antibody feeding experiments are always performed with negative controls each time, where neurons are fed with anti-MAP2 antibodies to ensure no intracellular labeling is observed. This ensures that cultures are healthy – if signal is ever observed using negative controls, samples are no longer considered reliable and are discarded. Antibodies were diluted in PBSCM supplemented with 1% BSA and incubated on live neurons. Antibodies were washed and neurons were rinsed in PBSCM. Neurons were fixed in 4% PFA (Electron Microscopy Sciences Cat#15710)/4% sucrose and samples were blocked in 1X GDB (0.2 mM NaPO₄, 0.2% gelatin, 0.8 M NaCl, and 0.1% TritonX-100). Chicken anti-MAP2 primary antibody was diluted as noted in 1X GDB and coverslips treated with PC-ApoE were incubated in primary dilutions. Coverslips were washed, incubated in secondary antibodies diluted 1:1000 in 1X GDB and washed again. Neurons were mounted on glass slices and imaged using a CSU-W1 SoRa confocal microscope. Images are representative maximal Z-projections of multiple optical sections. Images were quantified using ImageJ by integrated density analysis.

Human Subjects

Human samples were obtained from MassGeneral Institute for Neurodegenerative Disease. The use of post-mortem brain tissue was approved by the Harvard University institutional review board (IRB) and the Columbia University ADRC brain bank and associated IRB. Informed consent was obtained from all subjects. The authors did not have access to any identifying personal information. Alzheimer's disease neuropathologic change rating was made using three criteria. Blind assessment for density of neuritic beta amyloid, distribution of beta amyloid plaques and Braak Tau stage were conducted using the scoring system recommended by the Consortium to Establish a Registry for Alzheimer's disease⁵⁹, Thal Score⁶⁰ and Bielchowsky's silver stain⁶¹ along with total Tau immunostaining respectively. According to the Braak Tau pathway, the following brain regions were collected from either unaffected ApoE3/3 controls or Braak V ApoE3/3 or Braak V ApoE4/4 patients. A full list of deidentified information about the patient list is presented in Extended Table 5.

We obtained material from the Brodmann Area (BA4 – primary motor area) and BA7 (parietal associative area). We analyzed these regions as a means of untangling how neuroproteasome localization changes in a limited survey of severe Tau pathology (BA7) versus less severe Tau pathology (BA4) containing regions of the brain. Tissue from BA7 contained severe tangle and amyloid pathology as well as neuronal loss whereas tissue from BA4 had moderate amyloid pathology but not tangle accumulation or neuronal loss. We fractionated these samples to remove nuclei, mitochondria, and cytosol, enriching for plasma membranes and detergent-insoluble proteins. We monitored the success of our fractionation using LLVY-based proteasome activity assays out of the supernatant of our washes and continued washing our plasma membrane preparations until cytosolic proteasomes were undetectable.

The brains were sliced, flash frozen and dissection from BA4 and BA7 to be homogenized in lysis buffer (50 mM HEPES, 0.1 mM CaCl₂ and 2 mM MgCl₂, 100 mM NaCl) supplemented with fresh 1 mM ATP and 0.1% GDN (Anatrace Cat# GDN101 - GDN). Nuclei, mitochondria, and heavy membranes enriched in ER and Golgi membranes were pelleted at 5,000 and then 10,000 × g and washed several times. The supernatant containing the membranes was pelleted at 75,000 × g, for 30 minutes and the supernatant was taken as cytosol. Pelleted membranes were washed by homogenizing in lysis buffer and re-pelleted. After each wash, the supernatant was analyzed by LLVY-based proteasome activity assay to assess cleanliness of the prep. Following washes, plasma membrane-enriched and cytosolic fractions were prepared for mass spectrometry and SDS PAGE.

Code Availability Statement

No custom software codes were used in the data analysis for this paper.

Extended Data

Extended Data Figure 1: Validation for synthesis of PEGylated Epoxomicin derivatives.

a, Structure of Azide-PEG(n)-Epoxomicin handle to enable rapid conjugation of linkers to global proteasome inhibitor Epoxomicin using Click chemistry, b-f, HPLC analysis (above) and LC/MS spectra (below) for Biotin-PEG1-Epox (b), Biotin-PEG4-Epox (c), Biotin-PEG8-Epox (d), Biotin-PEG12-Epox (e), Biotin-PEG24-Epox (f), g, HPLC analysis of iBEp (Biotin-PEG24-Epox) after purification.

Extended Data Figure 2: iBEp is a cell impermeant proteasome inhibitor.

a, Immunoblot of lysates from primary WT neurons treated with varying Biotin-PEG lengths coupled to Epoxomicin. Increases in ubiquitin conjugates are a hallmark of cytosolic proteasomal inhibition. p62 is a marker for autophagosomes which is transiently induced by proteasome inhibition. Note lack of both markers in Biotin-PEG24-Epox or Biotin-PEG12-Epox. b, Three exposures (Low, Medium, High) of immunoblot of lysates from primary neurons treated with DMSO or iBEp for 12 hours. Black arrowheads denote endogenous biotinylated proteins, Red arrowheads denote covalent modification of proteasome subunits with biotinylated epoxomicin (iBEp). Note there are only three bands differentially labeled between iBEp treatment and DMSO which correspond to correct MW of catalytic proteasome subunits. c, Schematic of assays to measure physiochemical properties of membrane permeability of compounds. Parallel artificial membrane permeability assay (PAMPA) and Madin-Darby Canine Kidney (MDCK) assay. LC/MS sensitivity of assay is 1pg. Data are mean +/− SEM. N=3 biological replicates. d, MDCK assay with indicated compounds for 24 hours. Data are mean +/− SEM. N=3 biological replicates ***p<0.001 by One-Way ANOVA e, Schematic for intracellular metabolite extractions from neurons treated with indicated inhibitors. f, Biotin quantification by HABA of extracted intracellular metabolites after compound treatment. Spike in indicated where compound is spiked into lysate prior to processing to establish positive control. Data are mean +/− SEM. ****p<0.0001 Paired two tailed T-test. g, LC chromatogram of metabolites extracted from DIV14 cortical neurons from WT mice treated with Epoxomicin (black) or iBEp (red) for 12 hours. Note peak in Epox treated cells corresponds to Epoxomicin, no iBEp peak is identified. h, LC/MS spectra of metabolites extracted from DIV14 cortical neurons from WT mice treated with Epoxomicin, note mass corresponding to Epoxomicin identified in spectrum. i, LC/MS spectra of metabolites extracted from DIV14 cortical neurons from WT mice treated with iBEp, experimental spectra from iBEp treated neurons in red, expected spectra of indicated compounds overlaid in Black onto iBEp spectrum to emphasize lack of breakdown products or intact iBEp inside neurons. Epox from iBEp is expected cleavage product if iBEp were to be cleaved at amide bond linking Epoxomicin peptide to PEG linker. j, Proteasome catalytic activity assays of membrane (red) and cytosolic (black) fractions treated with a dose curve of iBEp. Catalytic activity was assessed by monitoring degradation of Suc-LLVY-AMC. N=3 replicates. k, Indicated compounds incubated with mouse brain extracts and measured over time by LC/MS for degradation and stability.

Extended Data Figure 3: SulfoMG is a reversible and cell-impermeant proteasome inhibitor.

a, HPLC analysis of MG132-PEG24-Azide. b, Overlay of HPLC chromatograms of MG132-PEG24-Azide (black) and the overnight reaction between MG132-PEG24-Azide and DBCO-Sulfo-Link-Biotin (red). Note peaks emerging at 5.5 and 6 minutes corresponding to the formation of Sulfo-MG132 (SulfoMG). c, HPLC analysis of SulfoMG after purification. d, LC/MS spectrum of purified SulfoMG. Expected mass of SulfoMG (2254.0525) is observed. Other peaks correspond to adducts resulting from LC/MS ionization in the spectrometer. e, Biotin quantification by HABA of extracted intracellular metabolites after compound treatment. Spike in indicated where compound is spiked into lysate prior to processing to establish positive control. Data are mean +/− SEM. N=3 biological replicates ****p<0.0001 two tailed Student T-test. f, Indicated compounds incubated with mouse brain extracts and measured over time by LC/MS for degradation and stability. Data are mean +/− SEM. N=2 biological replicates. g, Proteasome catalytic activity assays comparing inhibitory effects of a dose curve of SulfoMG (red) and MG132 (black). Data are mean +/− SEM. h, Calpain and Cathepsin L catalytic activity assays comparing inhibitory effects of a dose curve of SulfoMG (red) and MG132 (dark gray). Data are mean +/− SEM. N=3 biological replicates.

Extended Data Figure 4: Neuroproteasome inhibition induces accumulation of endogenous sarkosyl-insoluble Tau in primary neurons.

a, Analysis of biophysical characteristics of soluble or insoluble proteome from neurons treated with iBEp relative to controls. Proteins which are unaffected between iBEp and DMSO in gray and increased in iBEp in red. Box plots show the median (center line), the interquartile range (25th-75th percentiles plotted as the box), with whiskers representing the 10th and 90th percentiles. Data analyzed by two-tailed Unpaired T-test. b, Sarkosyl fractionation of primary neurons from Tau knockout (Tau-KO) immunoblotted using R2295 antibody against total Tau. Neurons treated with indicated compounds and subjected to sarkosyl fractionation. We do not observe any signal at the expected molecular weight for Tau as indicated by the gray arrowhead. c, Immunoblots of sarkosyl soluble and insoluble fractions from primary WT neurons treated with DMSO, Biotin-PEG24-Azide linker, iBEp, or Epoxomicin. Relative Tau levels calculated as follows: Soluble and insoluble Tau intensities were normalized to respective soluble GAPDH intensity. All Tau levels plotted relative to DMSO in either soluble or insoluble fraction. Data are mean ± SEM, N=4 biological replicates, ****p<0.0001 by One-Way ANOVA Tukey’s Multiple Comparison Test. d, Full-length blot from Figure 2e to demonstrate molecular weight of sarkosyl-insoluble Tau species formed following neuroproteasome inhibition with iBEp (red), compared to Epoxomicin (Epox) or DMSO controls. e, Sarkosyl fractionation of primary neurons treated with indicated compounds. Epox + iBEp co-treatement indicated. Relative Tau levels calculated and plotted compared to inactive scaffold. N=3 biological replicates, ****p<0.0001 by One-Way ANOVA Tukey’s Multiple Comparison Test.

Extended Data Figure 5: Neuroproteasome inhibition induces formation of endogenous PHFs in vivo.

a, Immunoblots of membrane and cytosolic fractions of hippocampi from mice stereotactically injected with iBEp ipsilaterally or DMSO controls contralaterally. b, Immunoblots of sarkosyl-soluble and insoluble fractions of hippocampi from Tau-KO mice stereotactically injected with iBEp ipsilaterally or DMSO controls contralaterally. N=3 (2M, 1F) c,d, Sarkosyl-insoluble fractions of hippocampi injected with iBEp from either hTau-KI (c) or Tau-KO (d) mice. Low magnification image scale bar is 5μm, high magnification is 200 nm. Despite extensive searching, no fibrils were observed in Tau-KO mice even after 3-fold concentration of samples. All fibre-like densities when examined at high magnification appeared as protein clusters lacking helical characteristics. Similar results obtained from at least N=3 independent animals.

Extended Data Figure 6: Neuroproteasome inhibition-induced Tau aggregates have AD-like features.

a, Full-length immunoblot demonstrating high molecular species of sarkosyl-insoluble Tau species formed following neuroproteasome inhibition with iBEp (red), compared to Epoxomicin (Epox) or DMSO controls. N=7 biological replicates. b, Micrographs of DIV14 primary hippocampal neurons obtained from Tau knockout (Tau-KO) mice stained using indicated antibodies and DAPI. Scale bars=20 μm. c, Micrographs of DIV14 primary hippocampal neurons from hTau-KI (top) and Tau-KO (bottom) mice stained using only secondary antibodies as indicated. Scale bars=20μm. d, Micrographs of hippocampal sections obtained from Tau-KO mice stained using indicated antibodies. Scale bars=1mm. e, (top) Micrographs of hippocampal sections obtained hTau-KI mice stereotactically injected bilaterally and stained with secondary antibodies alone. Note staining of injection site at cortex, indicating damage, but no signal in hippocampus. (bottom) Micrographs of hippocampal sections from Tau-KO mice stained using indicated secondary antibodies. Scale bars=1mm. (b-e) replicated in at least three independent cultures or mice f, Micrographs of DIV14 primary hippocampal neurons obtained from hTau-knock-in (hTau-KI) mice treated with indicated drugs for 12 hours and stained using indicated antibodies. Quantification of pTau signal intensity was normalized to total Tau. Data are mean ± SEM normalized to DMSO. N= 3 biological replicates, n=4 images/replicate, analyzed by two-tailed Paired T-test. Scale bars=15μm. g, Experiment as in (f) but with Epoxomicin. N=2 biological replicates, n=4 images/replicate), data are mean +/− SEM. h, Micrographs of sections from hTau-KI mice stereotactically injected Epoxomicin ipsilaterally and DMSO contralaterally. Quantification of pTau signal intensity was normalized to NeuN counts. Analysis was done blinded to experimental condition. Data are mean ± SEM normalized to DMSO. N=3 independent animals, n=2 sections/animal, Scale bars=500μm. p > 0.05 by two-tailed Paired T-test. i-m, Representative micrographs of flame-like and thread-like Thioflavin-S positive aggregates from experiment in fig 4d, seen in each of N=7 independent animals. Scale bar=20μm.

Extended Data Figure 7: Validation of endogenous Cre-dependent FLAG-tagged 20S proteasome transgenic line.

a, Proteasome catalytic activity assay from lysates of HEK293 cells transfected with indicated genes and tags. ns, not significant from control, One-way ANOVA. b, Candidate hits from (a) subjected to ATPase assay. ns indicates p>0.5, not significant from control, One-way ANOVA. N=3 biological replicates. c, Candidates from (b) transfected into HEK293 cells and doubling time was measured and plotted, ns indicates p>0.5, not significant from control, One-way ANOVA. Data are mean +/− SEM. N=12 biological replicates. d, Immunoblots of proteasome immunopurifications from the cytosol of brains from 20S-FLAG/BAF53b-Cre mice. Immunoprecipitation (IP) using FLAG beads isolate FLAG-tagged 20S core particle as well as the 20S-containing 26S particle, whereas IP against the 19S only isolates 26S particles. FLAG and 26S IP from mouse brain cytosolic fractions immunoblotted using indicated antibodies. e, Proteasome purifications from the cytosol of livers from 20S-FLAG/BAF53b-Cre mice. Note, compare (d) which was done from brain cytosol. f, Catalytic activity of proteasomes isolated by 26S IP (squares) and FLAG IP (circles) from whole liver cytosol. Data are mean ± SEM from three replicates. ****p<0.0001 by Two-Way ANOVA Tukey’s Multiple Comparison Test. N=3 biological replicates. g, Immunoblots from surface biotinylation of primary neurons from 20S-FLAG mice transduced with Cre AAVs. Lysates (Total) and Streptavidin pulldowns (Surface) were immunoblotted using the indicated antibodies. h, Immunoblots of membrane fractionation of brains from 20S-FLAG/BAF53b-Cre mice. Lysates, cytosolic fraction, and membrane fraction immunoblotted using indicated antibodies.

Extended Data Figure 8: ApoE isoforms differentially modulate neuroproteasome localization.

a, Schematic of surface biotinylation of Receptor-Associated Protein (RAP)-treated neurons. Influence of RAP (blue) on surface ApoE receptors (light red) is depicted. Gray circles represent endosomes. b, Immunoblots from surface biotinylation of DIV14 primary neurons obtained from WT mice treated with RAP for indicated time. Lysates (Total) and Streptavidin pulldowns (Surface) were immunoblotted using indicated antibodies. LiCor-based quantification of surface Lrp1 and β5 intensity was normalized to corresponding total signal. Data (right) are mean ± SEM normalized to 0 hr condition. N=3 biological replicates analyzed by One-Way ANOVA Tukey’s Multiple Comparisons Test. c, Representative images from negative stain EM (top) of hApoE lipoproteins PC-rE2 (green), PC-rE3 (gray), and PC-rE4 (purple) and their corresponding chromatograms from size-exclusion chromatography. Coomassie G250 stained gel comparing PC-ApoE isoforms and non-lipidated recombinant ApoE isoforms. Similar results were obtained from at least N=3 preparations. d, Surface biotinylation of WT neurons treated with exogenous ApoE lipoparticles. DIV16 primary neurons treated with recombinant ApoE isoforms (rE2, rE3, rE4) or POPC/Cholesterol(PC)-conjugated ApoE isoforms (PC-rE2, rE3, and rE4). Lysates (Total) and Streptavidin pulldowns (Surface) were immunoblotted using indicated antibodies. LiCor-based quantification of surface 20S and GluN1 intensities were normalized to corresponding total intensities. Data are mean ± SEM normalized to corresponding vehicle control. N=3 biological replicates, analyzed by One-Way ANOVA Tukey’s Multiple Comparisons Test. e, Illustrative schematic for antibody feeding experiments. f, Micrographs of anti-MAP2 and anti-FLAG antibody feeding of live DIV14 primary hippocampal neurons obtained from 20S-FLAG mice. Note lack of MAP2 signal. Scale bars=10 μm. N=3 biological replicates. g, Micrographs of anti-GluR1 and anti-FLAG antibody feeding of live DIV14 primary hippocampal neurons obtained from 20S-FLAG mice. Scale bars=10 μm. N=3 biological replicates. h, Micrographs from DIV14 primary hippocampal neurons obtained from 20S-FLAG mice which received the same treatment as antibody feeding samples as in Fig 6c, but only secondary antibodies were used. Scale bars=10 μm. N=3 biological replicates. i, DIV14 primary neurons from 20S-FLAG mice transduced with control AAVs without Cre and anti-FLAG antibodies were fed in Fig 6c and above in f,g. Samples were stained with MAP2 antibodies following fixation. N=3 biological replicates. Scale bars=10 μm.

Extended Data Figure 9: Neuroproteasomes link ApoE and aging to Tau proteostasis.

a, Immunoblots of cytosolic fractions from BA7 and BA4 from Apoε3/3 patients with AD and without AD (unaffected) and Apoε4/4 patients with AD were immunoblotted using indicated antibodies. Quantification of all human patient tissues. 14 AD ApoE3/3, 17 Control ApoE3/3, and 21 AD ApoE4/4 from BA4 (low Tau pathology) and paired samples from BA7 (high Tau pathology). Box plots show the median (center line), the interquartile range (25th-75th percentiles plotted as the box), with whiskers representing the 10th and 90th percentiles. Effect of ApoE isoform alone, Effect of AD comparing only ApoE3/3 tissues, Effect of genotype and pathology compared three way, Quantification of tissues to determine interaction between ApoE genotype and Tau pathology, Differences of differences of means between AD BA4 and AD BA7 to isolate ApoE genotype effect independent of pathology. No significant different were detected across any comparison. b, GAPDH loading control and mTau levels for ApoE-KI/hTau dose curve. c, Data from Fig 7f, (left) initial proteasome activity plotted. (right) unnormalized raw values from proteasome activity assay plotted, y-axis is insoluble Tau levels from Fig 7f. d Data from Fig 7f, neuroproteasome activity at each dose of iBEp normalized against uninhibited basal activity from each genotype. e, Immunoblots of sarkosyl-insoluble fractions of primary neurons from WT mice treated with RAP protein and a dose curve of iBEp. Red indicates RAP-treated neurons, Black indicated PBS-treated neurons. LiCor-based quantification of sarkosyl-insoluble Tau intensity was normalized to soluble GAPDH. All data in c-e are data are mean +/− SEM.

Data Availability

All data in the manuscript are available within the manuscript as source data or supplemental data files. Reagents, including the inhibitors and mouse lines, can be requested from the contact investigator and will be distributed upon reasonable request and availability to fill all requests. Mouse lines will be deposited at Jackson Labs after publication and inhibitors will hopefully be commercially available shortly.