Tau, synapse loss and gliosis progress in an Alzheimer’s mouse model after amyloid-β immunotherapy

Abstract

Preclinical studies assessing drugs for Alzheimer’s disease (AD) are conducted in animal models that usually display only one neuropathological feature of AD, whereas patients present with a complex combination of comorbidities and neuropathologies. Importantly, it is well established that amyloid-β (Aβ) plaque and tau tangle accumulation interact in a phase-dependent manner, making it difficult to predict how targeting one might influence the other, as well as downstream degeneration.

We developed a transgenic mouse model, APP/PS1xTau22, with progressive cortical Aβ deposition and hippocampal tau neurofibrillary inclusions, to investigate how both neuropathologies act jointly to bring about neural degeneration, synapse loss and glial phenotypes. We then assessed whether applying murine chimeric aducanumab, an anti-amyloid immunotherapy, could impact the synergistic relationship between amyloid and tau.

Drug treatment resulted in a ∼70% reduction in Aβ deposition in hippocampal and cortical areas and produced a robust peri-plaque microglial and astrocytic response. Removing amyloid from the brain did not reverse or slow tau pathology or alter synapse loss.

Our findings suggest that, once the interaction between amyloid and tau is set in motion, reducing plaque burden by Aβ immunotherapy may stimulate glial responses, but is insufficient to curb degenerative phenotypes in this model.

Introduction

The course of Alzheimer’s disease (AD) is characterized by the neocortical deposition of amyloid-β (Aβ) plaques and tau neurofibrillary tangles originating in the medial temporal lobe.^1,2 Genetic and biochemical evidence provides strong support for the amyloid hypothesis, which positions Aβ as the primary initiator of the AD pathogenesis.^3,4 The expansion of neurofibrillary lesions into higher cortical areas also appears to be accelerated by established plaque deposition, suggesting that amyloid creates a permissive environment for tau propagation.^5,6 Moreover, the local microenvironment of Aβ deposits recruits tau-containing dystrophic neurites, likely disrupting neural pathways. This is of particular importance, since the advancement of tau inclusions across brain networks is a closer neuropathological correlate of neuron loss and symptom severity in patients than plaque deposition.^7,8

Amyloid-targeting antibodies have been shown to be remarkably effective at clearing amyloid plaques from the brain, with three immunotherapies recently receiving regulatory approval for the treatment of AD.^9-11 Despite dramatic reductions in brain amyloid load, as measured by PET, patients continue to show cognitive decline, albeit at a statistically significant slower rate. One possible explanation for the disconnect between the magnitude of amyloid biomarker changes and symptomatic improvement is that preclinical studies use animal models that develop only amyloid pathology, whereas AD patients present with ample tau pathology, neuroinflammation and other concurrent neuropathologies. Deciphering how amyloid and tau interact with immune mechanisms to drive neurodegeneration will be crucial in determining which secondary disease processes can be slowed, interrupted or reversed by amyloid-targeting immunotherapies and will help to inform future approaches, including the use of combination therapies.

Both active and passive Aβ immunotherapy lead to robust clearance of Aβ deposits and, in some instances, even improve neural system integrity in mouse models of amyloidosis.^9,12-14 To explore whether immunotherapy can produce similar benefits when both amyloid and tau are present, we bred the Thy-Tau22 (Tau22) transgenic mouse model of tauopathy¹⁵ against a well-characterized APP/PS1 transgenic line. Although Aβ and tau deposition are produced by transgene expression, the resultant APP/PS1xTau22 progeny recapitulate the stereotypical distribution of cortical Aβ plaques and hippocampal tau tangles seen in humans. Here, we examine how Aβ immunotherapy with aducanumab impacts glial, synaptic and tau-related phenotypes in a more complex neuropathological environment.

Materials and methods

Animals and drug treatment

Animal care, housing and procedures were performed in compliance with guidelines established by the Massachusetts General Hospital (MGH) Institutional Animal Care and Use Committee, and in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals. Mice were housed under a 12 h light/dark cycle and provided with food and water ad libitum. Hemizygote APP/PS1 mice purchased from Jackson Laboratory [B6;C3-Tg(APPswe, PSEN1dE9)85Dbo/Mmjax; Stock 034829-JAX] were bred with hemizygote Thy-Tau22 (Tau22) mice¹⁵ provided by Luc Buée (Université de Lille, France) to generate transgenic mice hemizygous for mutant amyloid precursor protein (APP) and presenilin 1 (PS1) and/or tau (MAPT). Wild-type (WT) littermates were also used. All groups were sex balanced. Genotyping was performed by Transnetyx Inc. Equal concentrations of aducanumab and isotype control IgG were administered weekly for 12 weeks by intraperitoneal injection at a dose of 30 mg/kg beginning at 6 months of age. Nine-month-old mice were sacrificed 1 week after receiving the last dose.

Tissue isolation

At 6 and 9 months of age, mice were euthanized with CO₂ and transcardially perfused with ice-cold phosphate buffered saline (PBS) for 3.5 min at a rate of 10 ml/min. After the brain was removed, one hemisphere was microdissected and immediately frozen over dry ice, and the other hemisphere was drop-fixed in 4% paraformaldehyde for 24 h before being transferred to a large volume of PBS. Frozen tissue was stored at −80°C until processed. Fixed hemispheres were shipped at room temperature to NeuroScience Associates Inc. and embedded in a non-infiltrating gelatin matrix before sectioning on a sliding microtome. Free-floating coronal sections, 30-μm thick, were shipped back to MGH in cryoprotectant antigen preserve solution and stored at −20°C.

Preparation of brain homogenates

Frozen cortical tissue was quickly thawed and immediately homogenized in ice-cold PBS with 1× protease inhibitor cocktail (Cat. No. 5871, Cell Signaling). After centrifugation at 3000g for 10 min at 4°C, the PBS-soluble supernatant was separated from the pellet and used for western blotting and biochemical characterization.

To isolate and measure soluble Aβ species by ELISA, cortical tissue was homogenized in 1× Tris-buffered saline (TBS) with 1× protease inhibitor. Samples were then ultracentrifugated at 100 000g for 12 min at 4°C using an MLA-130 ultracentrifuge rotor with 1 ml tubes. The supernatant was diluted to a concentration of 1:50 for ELISA measurements. All samples were stored at −80°C until use.

Western blotting

Total protein content of PBS-soluble lysates was measured by bicinchoninic acid (BCA) assay (Cat. No. ab102536, Abcam). Equal quantities of protein were mixed with loading and reducing buffers and heated to 95°C for 5 min before loading onto a NuPAGE 4%–12% Bis-Tris gel (Invitrogen). Proteins were transferred to a nitrocellulose membrane using an iBlot transfer machine for 7 min at 25 V. Membranes were treated with LI-COR blocking buffer (LI-COR Biosciences) for 1 h at room temperature and incubated with primary antibodies against human tau (1:5000; Cat. No. A0024, DAKO) and human APP (1:5000 Cat. No. 29765, Cell Signaling), with GAPDH as a standardizing control (1:5000; Cat. No. 2118, Cell Signaling), overnight at 4°C. After applying IRDye 680- and 800RD-conjugated secondary antibodies, membranes were imaged on the Odyssey CLV scanner (LI-COR Biosciences) and quantified using Image Studio Lite Quantification software (version 5.2; LI-COR Biosciences).

Semi-denaturing detergent agarose gel electrophoresis

Semi-denaturing detergent agarose gel electrophoresis (SDD-AGE) was carried out as described.^16,17 Agarose (1.5%) was carefully mixed with boiling 1× Buffer G (20 mM Tris-Base, 200 mM glycine) until dissolved. Once cooled, sodium dodecyl sulphate (SDS) was added to a final concentration of 0.02%. Cortical lysates were incubated with sample buffer for 7 min at room temperature prior to loading. The SDD gel was run with cold Laemmli buffer (Buffer G with 0.1% SDS) at 30 V for 22 h at 4°C. Resolved proteins were then transferred to a polyvinylidene difluoride membrane between 20 pieces of thick Whatman paper and eight pieces of thin Whatman paper for 16 h at 4°C. Membranes were blocked as described and incubated with anti-total tau antibody (1:5000; Cat. No. A0024, DAKO) in LI-COR antibody diluent for 2 h at room temperature. After applying IRDye 680RD-conjugated secondary antibody, membranes were scanned on the Odyssey CLV scanner. Initially, the signal intensity of the entire lane was quantified, representing the abundance of total tau in each sample. Each lane was then separated into 12 bins of equal size, such that the highest molecular weight tau could be detected at the top of the gel, and the lowest molecular weight peptides could be found at the bottom of the gel. The relative abundance of different molecular weight tau species is expressed as the quantity of tau within a bin as a function of total tau in a sample.

Proteinase K digestion

Cortical homogenates were incubated with 0.25 μg/ml proteinase K (Cat. No. AM246, ThermoFisher) in 10 mM Tris-HCl pH 7.4 for 30 min at room temperature, as described.¹⁸ The enzymatic reaction was stopped by denaturing at 95°C for 10 min in loading and reducing buffers. Samples were subsequently loaded on a NuPAGE 4%–12% Bis-Tris gel with 10× MES running buffer (ThermoFisher). All tau fragments were revealed using a mixture of anti-tau primary antibodies targeting several regions of the tau protein (Cat. Nos. 43894S, 30328, Cell Signaling, MAB2241, Millipore), and fragments smaller than 28 kDa were quantified as described.

In vitro tau seeding assay

Tau seeding activity was measured in vitro with the widely used fluorescence resonance energy transfer (FRET) biosensor assay.^19,20 HEK293 cells expressing the repeat domain of P301S-mutant tau fused with cyan or yellow fluorescent protein (CFP and YFP, respectively) were plated in poly-D-lysine-coated 96-well plates at 40 000 cells per well. Cells were cultured for 24 h at 37°C with 5% CO₂ in Dulbecco’s modified Eagle's medium (DMEM), 10% fetal bovine serum and 1% penicillin–streptomycin. The following day, 8 ng of total protein from cortical extracts were incubated with 1% lipofectamine 2000 transfection reagent (ThermoFisher) in opti-MEM for 20 min at room temperature, then applied to HEK293 biosensor cells. Each sample was loaded in triplicate. After 24 h, cell medium was removed, and 50 μL of 1× trypsin was added for 5 min at 37°C before being replaced again with fresh medium. To measure FRET signal intensity by flow cytometry, cells were centrifuged at 1000g for 10 min, resuspended in 2% paraformaldehyde for 20 min at room temperature and pelleted again. After resuspending in PBS, the cells were immediately run on the MACSQuant VYB flow cytometer (Miltenyi Biotech, Inc.). Tau seeding for each well was calculated by multiplying the percentage of FRET⁺ cells by the median fluorescence intensity of the FRET⁺ population. Cells treated with lipofectamine alone were used to establish the FRET⁻ population.

ELISA

A human/rat Aβ₄₂ ELISA kit was run according to the manufacturer’s recommendations (Cat. No. 290-62601; Fujifilm, WAKO Chemicals). Plates were developed using a Wallac plate reader (Perkins Elmer) at 450 nm absorbance. Samples were fitted to a six-point standard curve. Each ELISA measurement was standardized by the total protein concentration in the sample, as determined by BCA assay, and expressed in micrograms per microlitre.

Immunohistochemistry

For fluorescent immunolabelling, free-floating, gelatin-embedded tissue sections were washed of cryoprotectant with PBS, treated with 50% ethanol in water for 20 min at room temperature and blocked with 20% normal goat serum in PBS with 0.3% Triton-X for 1 h at room temperature. Sections were then incubated with the following primary antibodies in 10% normal goat serum in PBS with 0.3% Triton-X overnight at 4°C: AlexaFluor 488- or 647-conjugated rabbit anti-β-amyloid (1:200; Cat. No. 51374 and 42284, respectively, Cell Signaling); biotinylated mouse anti-phospho-tau (AT8; 1:1000; Cat. No. MN1020B, ThermoFisher); mouse anti-phospho-tau (AT100; 1:1000; Cat. No. MN1060, ThermoFisher); Cy3-conjugated mouse anti-GFAP (1:500; Cat. No. C9205, Millipore); rabbit anti-Iba1 (1:500; Cat. No. 019-19741, Wako Chemicals USA Inc.); mouse anti-neurofilament (SMI312; 1:500; Cat. No. 837904, BioLegend); mouse anti-tau paired helical filaments (PHF1; 1:1000; Peter Davies, Albert Einstein College of Medicine, USA). Tissue was subsequently washed with PBS and incubated with AlexaFluor-conjugated secondary antibodies (1:500; ThermoFisher) for 1 h at room temperature before being washed, mounted onto 50.8 mm x 76.2 mm subbed glass slides, and coverslipped with DAPI Fluoromount G (Cat. No. 0100-20, Southern Biotech).

For immunolabelling and super-resolution imaging of synaptic puncta, ethanol pretreatment was omitted, and sections were incubated with the following primary antibodies overnight at room temperature: guinea pig anti-Bassoon (1:1500; Cat. No. 141004, Synaptic Systems) and rabbit anti-PSD95 (1:1000; Cat. No. 51-6900, ThermoFisher). Following incubation with AlexaFluor-conjugated secondary antibodies for 4 h at room temperature, tissue was mounted onto charged glass slides and coverslipped with ProLong Glass Antifade mounting medium (Cat. No. P36980; Invitrogen) using a 170 ± 5 μm No. 1.5H high-precision coverglass for super-resolution imaging.

For immunoenzymatic histochemistry, washed sections were quenched of endogenous peroxidase activity using 3% hydrogen peroxide and 10% methanol in water for 30 min at room temperature before blocking with 10% normal goat serum in PBS with 0.3% Triton-X. Following overnight incubation with rabbit anti-NeuN (1:1000; Cat. No. 24307, Cell Signaling), tissue was treated with biotinylated anti-rabbit IgG (1:400). Signal amplification was performed using the Vectastain Elite ABC-HRP kit (Cat. No. PK-6100, Vector Laboratories), and chromogenic staining was developed using a DAB substrate kit (Cat. No. SK-4100, Vector Laboratories).

Microhaemorrhage analysis

Four coronal sections were washed in PBS with agitation and incubated in a 1:1 mixture of freshly prepared 5% potassium ferrocyanide (LabChem) and 5% hydrochloric acid (Sigma Aldrich) for 30 min. After being washed in distilled water for 10 min, sections were mounted and allowed to dry before being dehydrated in a graded ethanol series, delipidated in xylene and coverslipped. All tissue sections were surveyed manually, and each haemosiderin-positive microhaemorrhage was imaged using a 20× objective. Microhaemorrhages were traced manually and their size measured using QuPATH software.²¹

Image acquisition and analysis

Whole-section widefield images were acquired at ×40 magnification using a NanoZoomer (Hamamatsu). Acquisition settings were held constant for all images being directly compared by quantitative analysis. The entire hippocampus, CA1 pyramidal cell layer and cortex were segmented manually, and intensity thresholding, manual cell counting and length measurements were performed using QuPath software. For analyses denoting ‘hippocampus’ and ‘cortex’, the entire visible brain area was quantified.

Neuritic plaques were captured at ×40 magnification using an FV3000 confocal laser scanning microscope (Olympus). To quantify AT8- and SMI312-immunopositive dystrophies within and immediately surrounding cortical dense-core plaques, the DAPI⁺ core of each Aβ plaque was delineated, and the borders were extended by 10 μm using QuPath software. The number of AT8⁺ and SMI312⁺ puncta within the boundaries of each dense-core plaque was quantified and normalized by plaque size. Similar analyses were performed to quantify the percentage area of Iba1 and GFAP immunoreactivity within the plaque corona; however, for these measurements, images were captured at ×40 magnification using a NanoZoomer, and the borders of each dense-core plaque were extended by 25 and 75 μm, respectively, again using QuPath software. Each measurement was standardized by individual plaque size.

Pre- and postsynaptic terminals immunolabelled with anti-Bassoon and -PSD95, respectively, were captured in the stratum radiatum using the Zeiss Elyra 7 super-resolution microscope with Lattice SIM2. Z-stacks, 3.91 μm thick, were imaged at a step size of 0.126 μm using a 63× oil objective. All images were acquired with laser power 0.9% and 90 ms exposure and processed using SIM2 ‘Low contrast’ reconstruction settings. The total number of Bassoon and PSD95 puncta was determined by thresholding using Imaris Software (Oxford Instruments) and normalized per millimetre cubed of tissue.

Statistical analysis

An independent-samples t-test was performed for parametric analyses of two groups. When the F-statistic was significant, Welch’s t-test was applied. The Mann–Whitney U-test was performed for non-parametric analyses, as determined using the Shapiro–Wilk normality test. One-way ANOVA was performed for parametric analyses of three or more groups. The Holm–Sidak multiple comparisons test was performed between all genotypes, except between APP/PS1 and Tau22 groups. Two-way ANOVA was used to measure main effects of sex and genotype and interactions. Extreme values identified by the ROUT outlier test (P < 0.01) were excluded and are not shown.

Parametric data are presented in column bar graphs, where error bars represent the mean ± standard error of the mean (SEM). Non-parametric data are presented in box-and-whisker graphs using the Tukey method. The sample size (n) for each experiment represents the number of animals tested. Each data point represents the average value of all technical replicates per animal. The final sample size is represented in all figures. Significance was set at P < 0.05 for all statistical tests; ns, P > 0.05; *P < 0.05; **P ≤ 0.01; ***P ≤ 0.001.

Results

Amyloid–tau coexpression exacerbates tau tangles and axonal dystrophy in the APP/PS1xTau22 mouse model

Studies in animal models have shown that the presence of pathological tau can increase, decrease or have no effect on Aβ plaque deposition, a phenomenon that seems to be model and age dependent.^6,18,22-24 We initially sought to understand how human mutant tau influences Aβ neuropathology in the APP/PS1xTau22 transgenic model. Nine-month-old APP/PS1 and APP/PS1xTau22 mice had widespread diffuse and dense-core plaques throughout the cortex and hippocampus, but plaque coverage was comparable between groups (Fig. 1A). Levels of soluble Aβ₄₂, as measured by ELISA, were also no different between APP/PS1 and APP/PS1xTau22 animals (Fig. 1B). Consistent with previous reports, APP/PS1-derived female mice had significantly higher Aβ plaque percentage area compared with male mice (Supplementary Fig. 1D).

Investigations into amyloid–tau interactions, in both experimental animal models and human brain, have consistently shown that amyloid potentiates the progression of tau pathology.²⁵ Introducing the human mutant APP/PS1 transgenes into the Tau22 transgenic line likewise boosted the development of hippocampal AT8⁺ tau tangles and neurites, beginning as early as 6 months of age (Supplementary Fig. 1A). By 9 months of age, the number of AT8⁺ neurons was also elevated in the cortex of APP/PS1xTau22 mice compared with Tau22 alone (Fig. 1C). We postulate that the expansion of AT8⁺ tangles is likely to be caused by accelerated tau phosphorylation, rather than a shift in the onset of tangle formation, because the difference in hippocampal AT8 percentage area between Tau22 and APP/PS1xTau22 mice increased from a factor of 1.55 to a factor of 2.00 between 6 and 9 months of age (Fig. 1C and Supplementary Fig. 1A). Phospho-tau epitopes associated with more mature tau lesions, as revealed by antibodies AT100 and PHF1, were also more abundant in the hippocampus and cortex of 9-month-old animals harbouring both amyloid and tau (Fig. 1D and E). Changes in tau deposition were not attributable to differences in human tau protein expression (Supplementary Fig. 1B), and no significant sex effect was detected at 9 months of age (Supplementary Fig. 1C).

We wondered whether amyloid alters tau biochemical properties that promote its assembly into pathological structures. Initially, we used a well-established in vitro biosensor assay to measure tau bioactivity in PBS-soluble cortical extracts. When misfolded tau is exogenously applied to HEK293 cells expressing the mutant P301S tau repeat domain fused with CFP or YFP, recruitment and aggregation of endogenous tau fragments generates a FRET signal, which can be quantified by flow cytometry.^20,26 Tau isolated from 9-month-old Tau22 and APP/PS1xTau22 mice showed comparable levels of seeding activity (Supplementary Fig. 1E). It has also been demonstrated that the occurrence of AD plaque pathology in human brain renders tau more resistant to enzymatic degradation by proteinase K (ProK), probably through conformational alterations. We therefore incubated PBS-soluble cortical lysates with 0.25 μg/mL of ProK and assessed the resulting proportion of tau fragments smaller than 28 kDa, as shown by western blotting (Supplementary Fig. 1F). Tau extracted from Tau22 and APP/PS1xTau22 mice exhibited similar susceptibility to enzymatic digestion (Supplementary Fig. 1G). Lastly, we examined the oligomeric state of PBS-soluble aggregates by semi-denaturing electrophoresis. Nine-month-old Tau22 and APP/PS1xTau22 animals had almost identical proportions of high- and low-molecular weight tau species (Supplementary Fig. 1H). These data suggest that enhanced spread of neurofibrillary tangles in APP/PS1xTau22 mice is not attributable to differences in tau seeding or to the clearance or relative abundance of high-molecular weight oligomeric assemblies.

Although amyloid pathology was unaffected by the presence of neurofibrillary tangles in 9-month-old APP/PS1xTau22 mice, we evaluated whether plaques are differentially toxic to surrounding neurites challenged with pathological tau build-up. We analysed dense-core plaques in the cortex of 9-month-old mice and normalized all measurements by plaque size (Fig. 1F). Not surprisingly, animals with both amyloid and tau aggregates displayed almost twice as many AT8⁺ puncta within the plaque core and corona than those accumulating amyloid alone (Fig. 1G). We measured the extent of neuritic damage directly by quantifying the amount of plaque-associated SMI312⁺ axonal swellings. In accordance with previous findings,²² SMI312⁺ dystrophies were more numerous in APP/PS1xTau22 mice compared with APP/PS1 littermates (Fig. 1H). Although the availability of aggregate-prone tau might produce more AT8⁺ puncta within dense-core plaques, the co-occurrence of tau pathology also appears to render axons more vulnerable to plaque-induced dystrophy.

Treatment with aducanumab inhibits Aβ plaque progression in the APP/PS1xTau22 mouse model

To determine whether interrupting Aβ plaque deposition could halt downstream disease events in a model of mixed AD neuropathologies, we treated APP/PS1xTau22 mice with chimeric aducanumab targeting soluble and insoluble human Aβ species (Fig. 2A). We confirmed target engagement by quantifying mouse IgG colocalization with Aβ plaques (Supplementary Fig. 2A). Indeed, Aβ deposits were decorated with mouse IgG in all mice administered aducanumab, whereas control animals did not display any IgG-labelled plaques (signal intensity in the control group represents background fluorescence). We verified that there were no sex differences in IgG-Aβ binding among aducanumab-treated mice (Supplementary Fig. 2B).

Following 3 months of immunotherapy, APP/PS1xTau22 mice had significantly lower Aβ plaque coverage in both the hippocampus and the cortex in comparison to animals treated with control IgG (Fig. 2B). Treatment was equally effective in female and male mice (Supplementary Fig. 2C). Interestingly, Aβ plaque percentage area was comparable between 6-month-old mice at baseline and 9-month-old aducanumab-treated animals, suggesting that, in this transgenic model, Aβ immunotherapy is likely to slow further plaque deposition (Fig. 2B). Although aducanumab-treated mice had ∼6-fold fewer plaques than control animals (Fig. 2C), their plaque size distribution was notably different (Fig. 2D–F): control mice continued to generate smaller, more diffuse Aβ deposits over the course of the study, whereas aducanumab-treated mice had a smaller proportion of plaques between 10 and 140 μm². We also found that the average size of all plaques >180 μm² was identical between control IgG- and aducanumab-treated mice, supporting the idea that larger Aβ deposits might be more resistant to clearance by immunotherapy (Fig. 2G).

Disrupting Aβ deposition does not affect tau phosphorylation but ameliorates plaque-related dystrophy

We next evaluated whether aducanumab has secondary beneficial effects on tau pathology in APP/PS1xTau22 mice. Although our data indicate that amyloid facilitates tau phosphorylation as early as 6 months of age, blocking further plaque deposition using immunotherapy had no effect on the quantity of AT8- or AT100-immunopositive lesions in any brain region analysed (Fig. 3A and B). Tau seeding was also no different between treatment groups (Fig. 3C). Although Aβ catalyses tau phosphorylation, interrupting plaque progression does not change established tau phosphorylation patterns or slow the proliferation of tau pathology in the current transgenic model.

We also wondered whether Aβ immunotherapy could abate local plaque toxicity. We measured the number of AT8⁺ puncta and SMI312⁺ axonal swellings within and surrounding dense-core plaques of control IgG- and aducanumab-treated APP/PS1xTau22 mice (Fig. 3D). Although our randomized selection of plaques resulted in a small, but statistically significant group difference in mean plaque area of ∼100 μm² (Fig. 3D), we judged that such a small effect size should have minimal impact on the biological measure being considered. Moreover, all measurements were normalized by individual plaque size.

Even with regional tangle pathology being similar between treatment groups, strikingly, we found a decrease in the number of AT8⁺ puncta within plaques following treatment with aducanumab (Fig. 3E). We also observed a marked reduction in plaque-associated SMI312⁺ dystrophies (Fig. 3F). These results suggest that aducanumab protects neurites from plaque-induced dystrophy by modifying the plaque microenvironment, even though amyloid and tau continue to interact over the drug treatment course.

Tau-related neuron loss, dendritic atrophy and synapse degeneration persist after Aβ immunotherapy

Given that the scope of tau pathology is tightly associated with the severity of neuron loss and cognitive symptoms in AD, we investigated whether amyloid accumulation affects tau-mediated neurodegeneration in APP/PS1xTau22 mice. CA1 neuron loss has been described as early as 12 months of age in the Tau22 mouse model.¹⁵ Given that the Thy1.2 transcriptional promoter drives the production of human mutant tau predominantly in the hippocampus, we focused our analyses on this brain region. In the present experiment, 9-month-old Tau22 mice had a statistically significant reduction in the number of NeuN⁺ CA1 neurons compared with WT animals (Fig. 4A and B). No statistically significant difference was observed between APP/PS1 and APP/PS1xTau22 at this age (Fig. 4B). Considering previous reports that amyloid worsens tau-mediated neurodegeneration in very aged, 24-month-old transgenic animals,²² we hypothesized that the sample and effect sizes in this study might be too small to recognize subtle changes in neuron density at this earlier time point.

We also noticed considerable differences in the size of the stratum radiatum between genotypes. Indeed, pyramidal dendrites were shorter in Tau22 and APP/PS1xTau22 mice at 9 months of age (Fig. 4C). This phenotype was specific to the presence of tau pathology but was not influenced by amyloid coexpression. The length of axon-containing stratum oriens was unaffected by AD neuropathologies (Fig. 4D). These results are in line with published work demonstrating that hyperphosphorylated tau becomes mislocalized to the neuronal somatodendritic compartment, both in human AD brain and in transgenic mouse models.²⁷

We evaluated the stratum radiatum further to ascertain whether dendritic atrophy was accompanied by synapse degeneration. We performed super-resolution imaging of pre- and postsynaptic terminals far from Aβ plaques, such that any observed changes in synapse density could not be attributed to the known local effects of diffuse Aβ oligomers within the plaque corona. The number of presynaptic Bassoon⁺ puncta was preserved across genotypes (Fig. 4E). In contrast, APP/PS1 and Tau22 mice exhibited an 11% and 15% reduction, respectively, in postsynaptic PSD95⁺ puncta (Fig. 4F). Interestingly, APP/PS1xTau22 animals suffered from a 28% loss in postsynaptic boutons, probably resulting from the additive effects of soluble Aβ and tau neurofibrillary pathology (Fig. 4F). Given that postsynaptic terminals within the stratum radiatum originate from dendrites of tau-burdened pyramidal neurons, we anticipated that synapse degeneration would be most prominent within the postsynaptic compartment.

Next, we sought to determine whether chronic treatment with aducanumab could mitigate tau-related neurodegeneration. Considering that Aβ immunotherapy was ineffective at curbing upstream tau neurofibrillary spread, it was unsurprising to find that neuron density and dendritic length were indistinguishable between control IgG- and aducanumab-treated mice (Fig. 5A). We also asked whether targeting oligomeric Aβ species beginning at 6 months of age might decelerate or reverse the effects of soluble Aβ and tau fibrils on potentiated synapse loss in 9-month-old APP/PS1xTau22 mice. Given that presynaptic terminals were unaffected by either neuropathology, we expected both treatment groups to have equal numbers of presynaptic Bassoon⁺ puncta (Fig. 5B). Interestingly, the density of postsynaptic PSD95⁺ puncta remained low following administration of aducanumab (Fig. 5C).

Aducanumab stimulates parenchymal- and plaque-associated gliosis and microhaemorrhages in mouse model

To establish how amyloid and tau together alter glial hypertrophy and/or proliferation in the APP/PS1xTau22 mouse model, we first measured changes in percentage tissue coverage of microglia- and astrocyte-specific markers, Iba1 and GFAP, respectively. At 9 months of age, hippocampal Iba1 percentage area was elevated in all transgenic groups when compared with WT littermates (Supplementary Fig. 3A). However, only Aβ-expressing mice had increased cortical Iba1 expression, probably because tau pathology is not particularly evident in isocortical regions of Tau22-derived mice (Supplementary Fig. 3A). In contrast, GFAP expression in both the hippocampus and the cortex was notably amplified in APP/PS1xTau22 mice compared with transgenic animals with amyloid or tau alone and with WT controls (Supplementary Fig. 3B). In line with previous findings in the APP/PS1xTg4510 mouse line,^6,18 the effects of amyloid and tau co-deposition appear to modify astrocytic responses to a greater degree than those of microglia.

Given that Aβ stimulates glial expansion, either alone or in the presence of tau pathology, we hypothesized that restraining plaque development would temper microglial and astrocytic reactivity in APP/PS1xTau22 mice. To our surprise, aducanumab had no effect on Iba1 percentage area in any brain region analysed (Fig. 6A). However, animals receiving 3 months of aducanumab displayed increased microglial recruitment towards remaining plaques, which is consistent with the anticipated effects of microglial Fc receptor binding to Aβ-directed antibodies (Fig. 6A and B and Supplementary Fig. 4A). Intriguingly, we observed a statistically significant treatment × sex interaction, whereby female mice administered aducanumab exhibited an exaggerated microglial plaque response in comparison to male mice receiving the same treatment (Supplementary Fig. 4B).

Remarkably, GFAP expression in the hippocampus and cortex almost doubled in aducanumab-treated mice compared with APP/PS1xTau22 controls (Fig. 6C). Cortical GFAP⁺ astrocytes were most appreciable in superficial cortical layers I and II (Fig. 6C) and formed unusually wide plaque coronae (Fig. 6C and D). These plaque-associated astrocytes radiated well beyond the halo of plaque-responsive microglia, which were more tightly localized to the plaque border (Fig. 6A and B). We did not identify any sex differences in plaque-related astrocytosis (Supplementary Fig. 4C).

It is well documented that Aβ immunotherapy in patients can lead to oedema and to more rare but potentially serious adverse events, characterized by brain haemorrhage, together termed amyloid-related imaging abnormalities. Likewise, aducanumab administration in APP/PS1xTau22 mice resulted in the appearance of Prussian Blue-positive haemosiderin deposits that occurred much more frequently than in control IgG-treated mice (Fig. 6E). Microhaemorrhages were most common along brain meninges and large penetrating arterioles. Variability in Prussian Blue staining was unrelated to sex differences (Supplementary Fig. 4D).

Discussion

The finding that patients receiving Aβ immunotherapy experience discordant Aβ plaque clearance and symptom relief suggests that factors in addition to amyloid accumulation shape the trajectory of the AD cascade in a time- and context-dependent manner. Once a patient becomes symptomatic, a critical loss of neurons has already taken place.²⁸ Moreover, secondary disease mechanisms spurred by incipient Aβ accumulation may, at a point, proceed independently.^29,30 Given that preclinical studies have used animal models expressing amyloid alone almost exclusively, our objective was to test the broader efficacy of Aβ immunotherapy in the context of a more complex, AD-like neuropathological milieu.

Initially, we developed a new APP/PS1xTau22 mouse model that mirrors the hierarchical deposition of extracellular Aβ plaques and tau neurofibrillary inclusions. Results from our initial analyses indicate that amyloid and tau might not interact reciprocally: although it is clear that Aβ propels tau phosphorylation, tau does not appear to influence Aβ deposition in turn. This finding supports the hypothesis that Aβ acts upstream of developing tau pathology in the AD brain. We propose that a feedback effect of tau on amyloid might emerge only at later disease stages or in model systems with more aggressive phenotypes, as has been reported in the 5XFADxTau22 line.²³ Data from our earlier studies demonstrate that introducing pathological tau to APP/PS1 mice results in decreased Aβ plaque load at 12 months of age, but not earlier.^6,18 The finding that neurofibrillary pathology dampens plaque deposition in older transgenic mice highlights the complexity of these interactions. Furthermore, although these mice phenocopy many important AD-like characteristics, the constitutive overexpression of mutant human peptides in a murine setting might not fully reflect what happens in human AD in response to therapeutic intervention.

Multiple tau phospho-sites were differentially modified in APP/PS1xTau22 mice, but, unexpectedly, these changes occurred independently of tau seeding activity and oligomeric and conformational states, which were similar in tau-only and amyloid-coexpressing animals. Aβ might propel tau phosphorylation via direct action on neurons or via non-cell-autonomous mechanisms. For example, Aβ-induced tau phosphorylation could be mediated by microglial inflammasome activation.³¹ Of course, both processes might take place simultaneously.

Administration of murine chimeric aducanumab in APP/PS1xTau22 animals was extremely effective at nearly eliminating Aβ plaque deposition and minimizing the build-up of toxic, diffuse Aβ material, as has been observed in humans. However, potentiated tau phosphorylation progressed unabated when treatment was applied at 6 months of age, even in brain regions where tau tangles are limited. These data imply that, once triggered, mechanisms underlying Aβ-mediated amplification of tau phosphorylation are self-reinforcing and might not be amenable to the modifying effects of Aβ immunotherapy. This also supports the idea that there exist amyloid-dependent and -independent phases of AD neuropathology.²⁹ One limitation of the present study is that the APP/PS1 and Thy-Tau22 transgenic lines display excessive Aβ and tau production compared with humans. As a result, the APP/PS1xTau22 mouse model does not exactly replicate the sequence, pace and interactions of Aβ and tau generation, deposition and clearance as observed in AD patients; this might explain why amyloid removal appears to be more prominent in clinical trial subjects and/or why we do not find an effect of immunotherapy on tau histological phenotypes.⁹ That said, although plaque and tangle formation are driven by transgene overexpression, we would expect amyloid clearance to impact amyloid-dependent tau phenotypes, such as exaggerated synapse loss, and gliosis; these effects were not observed in our model following immunotherapy.

Although the extent of neuron loss was mild at 9 months of age, CA1 pyramidal cells in Tau22 and APP/PS1xTau22 mice had considerable dendritic atrophy without any changes in axon length. We also noted a stepwise decline in postsynaptic densities, with APP/PS1xTau22 animals presenting with the most severe impairment. In contrast, presynaptic terminals were spared. We suspect that synaptic depletion in double-transgenic mice results from the additive effects of Aβ and tau. At first, we hypothesized that, if given early enough, Aβ monotherapy could ease synapse degeneration, even when tau pathology persists. The finding that aducanumab could not curtail synaptic deficits in APP/PS1xTau22 mice should be a crucial consideration for why available immunotherapies do not alleviate symptoms and produce only modest improvement in the rate of decline of cognitive measures. Although Aβ alone is synaptotoxic, synaptic sites on dendrites with Aβ-initiated tau hyperphosphorylation might still deteriorate at an accelerated rate. It is also possible that IgG-activated microglia phagocytose PSD95⁺ terminals as collateral damage, as has been reported.^32,33 Thus, the potential protective effects of removing soluble Aβ might be negated by the unintended consequences of stimulating microglial immune responses.

Most unexpected was that removing Aβ plaques did not resolve microgliosis and even worsened parenchymal and plaque-associated astrocytosis. Although immunotherapy might prompt the beneficial clearance of Aβ plaques by microglia, it might also trigger activation of cytotoxic pathways that, in parallel, could exert secondary outcomes. For example, administration of human and murine chimeric aducanumab in mouse models of amyloidosis leads to increased expression of the cytokines Tnfα, Il1β, Ccl3 and Ccl4, in addition to the complement components C3 and C1qα.^32,34 Importantly, the effects of aducanumab on astrocytes are likely to be non-cell-autonomous, because astrocytes do not express IgG-binding Fc receptors. Treatment-dependent astrocytosis in murine models could therefore be mediated by microglia-derived signalling molecules. It has also been proposed that, following Aβ immunotherapy, reactive glia become highly associated with penetrating arterioles coated with amyloid deposits,³⁴ and that these vessels are particularly susceptible to microhaemorrhage, as we have observed in the APP/PS1xTau22 mouse model. Unravelling the link between Aβ-directed glial responses and weakening of cortical blood vessels, especially in the context of the APOE genotype, might give way to improved therapeutic strategies that limit adverse events in patients.

Data availability

The data that support the findings of this study are available from the corresponding author, upon reasonable request.

Funding

L.A.W. is the recipient of Postdoctoral Research Fellowships from the Fonds de recherche du Québec – Santé (FRQS), the Canadian Institutes of Health Research (CIHR) and the Alzheimer’s Association (24AARF-1192364). This work was funded by grants from the National Institutes of Health (NIH; 1RF1AG059789), the JPB Foundation and the Rainwater Foundation.

Competing interests

T.B. is an employee and shareholder of Biogen. B.T.H. has a family member who works at Novartis and owns stock in Novartis; he serves on the scientific advisory board of Dewpoint and owns stock; he serves on a scientific advisory board or is a consultant for AbbVie, Aprinoia, Avrobio, Biogen, BMS Cell Signaling, Genentech, Novartis, Seer, Takeda, the US Department of Justice, Vigil and Voyager; his laboratory is supported by sponsored research agreements with AbbVie, and research grants from the National Institutes of Health, Cure Alzheimer’s Fund and Tau Consortium.

Supplementary material

Supplementary material is available at Brain online.

References