Tauopathy in the APPswe/PS1ΔE9 mouse model of familial Alzheimer’s disease

Despite compelling evidence that the accumulation of amyloid-beta (Aβ) promotes cortical MAPT (tau) aggregation in familial and idiopathic Alzheimer’s disease (AD), murine models of cerebral amyloidosis are not considered to develop tau-associated pathology. The absence of neurofibrillary lesions in amyloidosis mice remains a challenge for the amyloidocentric paradigm of AD pathogenesis. It has resulted in the generation of transgenic mice harboring mutations in their tau gene, which may be inappropriate for studying a disease with no known TAU mutations, such as AD. Here, we have used APPswe/PS1ΔE9 mice to show that tau pathology can develop spontaneously in murine models of familial AD. Tauopathy was abundant around Aβ deposits, with Gallyas- and thioflavin-S-positive perinuclear inclusions accumulating in the APPswe/PS1ΔE9 cortex by 18 months of age. Age-dependent increases in Gallyas signal correlated positively with binding levels of the paired helical filament (PHF) ligand [18F]Flortaucipir, in all brain areas examined. Sarkosyl-insoluble PHFs were visualized by electron microscopy. Tandem mass tag proteomics identified sequences of hyperphosphorylated tau in transgenic mice, along with signs of RNA missplicing, ribosomal dysregulation and disturbed energy metabolism. Human frontal gyrus tissue was used to validate these findings, revealing primarily quantitative differences between the tauopathy observed in AD patient vs. transgenic mouse tissue. Levels of tau mRNA were not different between APPswe/PS1ΔE9 and littermate control animals. As physiological levels of endogenous, ‘wild-type’ tau aggregate secondarily to Aβ in transgenic mice, this study demonstrates that amyloidosis is both necessary and sufficient to drive tauopathy in experimental models of familial AD.


Introduction
Genetically-inherited and sporadic forms of Alzheimer's disease (AD) are characterized by a common set of hallmark brain lesions, which include the accumulation of amyloid-β (Aβ) peptides into plaques, neuroinflammation, aggregation of hyperphosphorylated MAPT (tau) into neurofibrillary tangles (NFTs), and neurodegeneration. Transgenic mouse models that reproduce aspects of the aforementioned lesions have been generated based on mutations in the amyloid precursor protein (APP) and presenilin 1 (PSEN1) and 2 (PSEN2) genes, which are known to cause familial AD (1). Despite playing important roles in evaluating APP processing, Aβ toxicity, and amyloid-targeting therapeutic strategies, transgenic mice are not being regarded as models that can replicate the full spectrum of AD histopathology (2). In particular, while the overexpression of mutant APP and APP/PSEN1 has been shown to yield amyloidosis (3), neuroinflammation (4) and neurodegeneration (5) in mice, it is generally not considered to promote the conversion of endogenous tau into neurofibrillary structures (6).
To address the in vivo role of tau hyperphosphorylation and NFT formation in AD pathogenesis, human MAPT (TAU) has been introduced into the mouse genome, either mutated or nonmutated, on a Tau-knockout background (7,8). TAU overexpressing mice demonstrate progressive neurofibrillary pathology, albeit in the marked absence of cerebral amyloidosis, which is required for a neuropathological diagnosis of AD. Moreover, mutations in TAU have been linked to non-AD tauopathies, most commonly frontotemporal lobar degeneration [FTLD; (9)], a condition with neuropathological hallmarks distinct from AD. Thus, murine models of amyloidosis and combined amyloidosis-tauopathy models have been widely criticized for their translational relevance to the human condition. It has been argued that virtually all existing murine models would be considered as 'not' AD (10) according to the ABC scoring system of neuropathology (11). The inability of amyloidosis mice to develop neurofibrillary lesions is thought to contribute to the poor translation of preclinical research into clinical benefits (12), and has raised concern about the amyloidocentric model of AD pathogenesis (13).
Two principal explanations have been put forward for the lack of tau-associated pathology in amyloidosis mice (14). First, adult mice express fewer isoforms of the tau protein than humans (three vs. six), which might render them less liable to the post-translational modifications (PTMs) that are associated with the accumulation of tau into NFTs, such as phosphorylation (15). However, murine tau has been shown to readily fibrillize in vitro upon treatment with polyanionic factors, including RNA (16), and there is ample evidence of tau hyperphosphorylation in the transgenic mouse brain [(17), Table S1], indicating that no differences exist in the propensity of murine and human tau for aggregation and PTMs. A second reason that is often cited for the absence of tauopathy in amyloidosis models is that the murine lifespan may be too short for the complete sequence of neurofibrillary pathology to unfold in transgenic mice. Although age scaling studies suggest otherwise (18), the aging factor has been neglected in the design of preclinical studies.
Transgenic Fischer rats (TgF344-AD), expressing human APP harboring the Swedish double mutations (KM670/671NL) and PSEN1 lacking exon 9 (APPswe/PS1ΔE9), both under control of the mouse prion protein promoter, develop progressive neurofibrillary pathology (19). In this study, transgenic APPswe/PS1ΔE9 mice that were constructed in an identical manner as TgF344-AD rats were used to demonstrate neurofibrillary pathology in aging amyloidosis mice. 6 exclusively in AD tissue ( Fig. 1Q-Z). Coronal brain sections of 20-month-old Tg2576 mice, harboring the Swedish double mutations, were used to examine 6E10-and Gallyas-positive pathology in a second mouse model of amyloidosis (Fig. 1AA-AD). Amorphous argyrophilic signal (AC) and perinuclear lesions (AD) were also present in the Tg2576 mouse brain, albeit at lower levels compared to 18-month-old APPswe/PS1ΔE9 mice.
The fraction of brain tissue occupied by Gallyas-   Within each brain area analyzed, there was a positive correlation between the age-dependent increase in the binding levels of [ 18 18 F]Flortaucipir binding sites in aging APPswe/PS1ΔE9 mice. Fresh-frozen brain sections from APPswe/PS1ΔE9 and age-matched wild-type (WT) animals were incubated with 38.4±9.6 MBq [ 18 F]Flortaucipir for a period of 60 min (specific activity: 145±68 GBq/µmol). Autoradiography data are presented as the mean specific binding of [ 18 F]Flortaucipir (kBq/mL) ± standard error of the mean in brain regions of 5-6 animals/group. By 24 months of age, [ 18 F]Flortaucipir binding in APPswe/PS1ΔE9 mice had increased across all brain areas examined vs. age-matched WT animals. The age-dependent increase in

Isolation and Transmission Electron Microscopy (TEM) of sarkosyl-insoluble tau
The general methods of Sahara et al. (20) and Greenberg and Davies (21) were evaluated for the extraction of PHFs from the 24-month-old APPswe/PS1ΔE9 TG brain (Fig. S3). Although longer filaments were isolated by the procedure of Sahara et al., the Greenberg and Davies method was chosen for the isolation of sarkosyl-insoluble tau from 3-and 24-month-old mice, based on immunoblotting experiments, solubility considerations, and to allow for comparisons with literature data (22). Soluble and insoluble tau levels were measured in mouse brain homogenates by using the mouse Total Tau Meso Scale kit (Meso Scale Diagnostics LLC). TEM was used to visualize filaments in the sarkosyl-insoluble extracts from the TG mouse and AD patient brains by negative staining.
Tau protein levels increased with age in the pellet obtained by centrifuging WT and   Overview of negatively-stained filament types in the sarkosyl-insoluble fraction from 24-month-old APPswe/PS1ΔE9 and AD brain tissue. Fibrils of ~20 nm in width, appearing as straight filaments (a) or as two intertwined fibrils (e), each with a diameter of ~10 nm. PHFs with axial periodicities of ~80 nm (b & f; arrows) were present in APPswe/PS1ΔE9 mice, and more frequently observed in AD patient material. The inserts show 'stacked' PHFs, which were denser in the AD preparation. Structures commonly identified in the detergent-insoluble fractions of the mouse and human brain included bent fibrils of ~7 nm in width (c & g), and rod-shaped particles (d & h; arrows). Scale bars: 200 nm (a, b, e, f), 100 nm (c, d, g, h).

Proteomics of sarkosyl-insoluble tau
The sarkosyl-insoluble fractions extracted from 3-and 24-month-old mouse brain, AD and non-AD individuals, were pooled and digested with trypsin & Lys-C. The peptides were labeled with Tandem Mass Tags (TMT), fractionated, and analyzed by nanoLiquid Chromatography-

Electrospray Ionization Mass Spectrometry (LC-ESI MS/MS).
A list of tau-associated proteins quantified in the sarkosyl-insoluble proteome is shown in Table   2. Lists of between-group abundance ratios for all regulated proteins are shown in Data File S1.
There were 583 proteins identified in the sarkosyl-insoluble mouse proteome, of which 456 were also present in the human samples. Isoforms of tau with three (3R) and four (4R) microtubulebinding repeats were extracted from both human and the murine brain. In mice, all isoforms collapsed under the term microtubule associated protein (MAP; UniProt accession number: B1AQW2). Mouse MAP was regulated by age, rather than genotype. The protein was enriched 2.1-fold in 24-vs. 3-month-old TG mice, and 1.8-fold in 24-vs. 3-month-old WT mice.
Genotype-specific enrichment was observed for mouse tau isoform-B (UniProt accession number: P10637-3), a 3R isoform of tau with an extended C-terminal domain, which was identified by the sequence 205 KVQIVYKPVDLSKV 218 . Tau isoform-B increased 3.2-fold in 24month-old TG vs. WT mice, and 4.5-fold in 24-vs. 3-month-old TG animals. Human MAP (UniProt accession number: A0A0G2JMX7), containing tau isoforms P10637-2, -4, -6 & -8, was 37-fold enriched in the sarkosyl-insoluble fraction of AD compared to non-AD brain. The database for annotation, visualization and integrated discovery (DAVID, v6.8) was used for gene ontology (GO) enrichment analysis of the sarkosyl-insoluble proteome (23,24). RNA splicing, mRNA processing and translation were among the 10 most enriched biological processes associated with protein upregulation in 24-month-old APPswe/PS1ΔE9 vs. WT mice and AD vs. non-AD subjects. Ribonucleoprotein complexes, ribosomes, and exosomes were among the 10 most enriched cellular components in the insoluble extracts from the mouse and human brain (Fig. 5A). The top 10 molecular functions of the enriched proteins were associated with poly(A) RNA binding, as well as binding of molecules contributing to the structural integrity of ribosomes and the cytoskeleton (Fig. 5B). Pathway-based enrichment analysis of upregulated proteins in 24-month-old APPswe/PS1ΔE9 vs. WT mice involved GO terms such as Alzheimer's and Huntington's disease, long-term depression, cholinergic, serotonergic and glutamatergic synapse (Fig. 5C). Glycolysis/gluconeogenesis and the Krebs cycle were among the top 10 pathways for downregulated proteins (Fig. 5D). enriched KEGG pathways based on protein downregulation in 24-month-old TG vs. WT mice. Functional annotation clustering was generated by using DAVID software. Maximum enrichment probability (P value) was based on an EASE score threshold value of 0.05.

Discussion
The present study describes tauopathy in murine models of familial AD. Neurofibrillary alterations in APPswe/PS1ΔE9 and Tg2576 mice were demonstrated by a set of tools that are currently used for the evaluation of pathological tau clinically, such as the Gallyas silver stain and [ 18 F]Flortaucipir. The presence of PHF tau was confirmed by TEM of sarkosyl-insoluble preparations from the APPswe/PS1ΔE9 mouse brain. As murine tau possesses a remarkably high number of 76 potential serine/threonine and 4 tyrosine phosphorylation sites, an antibody-free proteomics approach was used for the detection of tauopathy-related epitopes. Of the five hyperphosphorylated sites identified, S404 has been associated with the intraneuronal and extracellular deposition of NFTs in AD (25). The pathology observed in the present study occurred at physiological levels of endogenous tau, as there was no difference in total tau mRNA and protein between APPswe/PS1ΔE9 and WT mice. Hence, in addition to progressive amyloidosis (3), neuroinflammation (4) and neurodegeneration (5), APPswe/PS1ΔE9 mice develop progressive neurofibrillary pathology of the AD type, mimicking a range of AD pathologies, in a translationally-relevant manner. The observation that endogenous tau accumulates secondarily to Aβ in models of cerebral amyloidosis is entirely consistent with post-mortem (26) and in vivo imaging data (27), showing that the development of cortical tau pathology in AD patients is associated with, and may depend on, pre-existing amyloid pathology.
Current approaches to induce tauopathy in mice have been criticized for generating models that poorly recapitulate the situation in the AD brain, as TAU in AD is neither overexpressed, nor mutated (28). FLTD-linked mutations, in particular, induce tauopathy that is not only morphologically different than that of AD (e.g. Pick bodies), but further characterized by distinct neurodegenerative processes. For example, cholinergic neurons are extensively lost in AD, but not in FTLD (29). Acetylcholinesterase inhibitors, which are prescribed for the symptomatic relief of cognitive impairment in AD, are largely ineffective in FTLD and may even worsen its symptoms (30). Thus, the pathophysiology that differentiates AD from primary tauopathies is unlikely to be modeled in mutant TAU models. Moreover, neurofibrillary alterations in TAU overexpressing mice occur in the absence of Aβ deposition, which is a defining feature of AD histopathology. The present results indicate that amyloidosis models may overcome these limitations, by reproducing both the neurofibrillary pathology of familial AD and the molecular heterogeneity that is associated with it. In addition to the spontaneous aggregation of tau in APPswe/PS1ΔE9 and Tg2576 mice, analysis of the sarkosyl-insoluble APPswe/PS1ΔE9 proteome identified proteins that have been strongly linked to AD pathogenesis, in general, and tau pathology in particular. Among them, APOE and BIN1 are encoded by genes whose variants are known to increase the risk of late-onset AD, through pathways involving interactions with both APP (31,32) and tau (33)(34)(35). Core components of the spliceosome, on the other hand, particularly Sm-D1 and Sm-D2, are closely related to the deposition of NFTs, but not plaques in familial AD (36). This literature implicates multiple mechanisms in AD tauopathy, which occur downstream of Aβ processing in cases of autosomal dominant AD (ADAD) and, as shown here, Although the sporadic and familial forms of AD share common clinical and histopathological features, it is becoming increasingly recognized that they are not precisely equivalent (37).
Positron emission tomography (PET) with [ 11 C]PIB demonstrates accumulation of Aβ in the cerebellum of familial AD cases, which is not typical of sporadic AD (38). Cerebellar deposition of hyperphosphorylated tau has been observed in ADAD cases harboring the PSEN1 E280A mutation, but not in sporadic AD (39). Thus, the pronounced cerebellar involvement in was not assessed in this study, these findings suggest that the neurofibrillary alterations observed in APPswe/PS1ΔE9 mice are not mediated by an imbalance between the production and clearance of tau. It is important to note that, unlike in TG mice, sarkosyl-insoluble tau was increased in AD vs. non-AD tissue, a finding that is consistent with literature data on the regulation of human tau in AD (43). It might be that the increased concentration of brain tau in late-stage AD is associated with heavily impaired clearance pathways or pronounced neuronal damage, processes that may not be modeled in 24-month-old APPswe/PS1ΔE9 mice. Alternatively, the present data may highlight the involvement of transcriptional and translational mechanisms, rather than production and clearance pathways, in the assembly of PHF tau.
A prevalence of 3R isoforms in the composition of NFTs has been observed in the AD hippocampus by immunohistochemical and biochemical methods (44). Moreover, a shift from 4R to 3R isoforms has been associated with the morphological evolution of tau-positive neurons from a pre-tangle to the NFT state (45). Although the literature on the regulation of tau isoforms in AD remains scarce, the present results support the notion that an imbalance in tau isoform ratio is involved in the neurofibrillary alterations of AD, with 3R isoforms being preferentially sequestered into the insoluble tau fraction. The identification of tau isoform-B, a 3R isoform that is predominantly expressed in the fetal mouse brain, supports the suggestion that immature tau isoforms participate in AD tauopathy (46), and implicates aberrant transcription and translation mechanisms in the disease process. A re-induction of fetal tau may be attributed to the deregulation of core splicing machinery, which was marked in this study and considered to occur early and selectively in AD (47). Moreover, as the selection of splice sites is determined by canonical sequences encoded into the genome, the re-expression of fetal isoforms might be a consequence of aberrant DNA replication during cell cycle re-entry (48). Cell cycle proteins that were deregulated in an age-and genotype-specific manner in this study include Sub1, cdc42, CEND1, Histone H3 and nucleolin (Data File S1). Clearly, the exact mechanisms underlying tauopathy in AD cannot be resolved by the present set of experiments. The data demonstrate, however, that the formation of PHF tau is associated with loss of regulatory control over tau splicing in vivo, which may have important implications for the origins and management of tauopathy in AD. It is tempting to speculate that tau hyperphosphorylation may partly be due to the re-emergence of fetal isoforms, which are known to be over-phosphorylated compared to adult tau (49). Moreover, it is plausible that an imbalance in tau isoform ratio mediates protein mislocalization from the axonal to the somatodendritic compartment, as distinct tau isoforms are differentially sorted across the cell (50). Of note, cofilin-dependent, 'classical' pathways of tau missorting (51) may also be involved in the pathology observed in this study, as cofilin was reduced in the sarkosyl-insoluble proteome of 24-month-old APPswe/PS1ΔE9 mice. Collectively, these data highlight the relevance of amyloidosis models for studying the diverse macroscopic and molecular aspects of AD tauopathy.
The limitations associated with models overexpressing APP and PSEN mutations have been discussed previously (2). To exclude the possibility that tauopathy is an artefact of APP or PSEN with the spread of tau pathology, and the number of brain areas containing at least one NFT has been shown to be the best explanatory variable of intellectual status in AD (55). In this context, it is worth evaluating whether measurements of tauopathy in aging APPswe/PS1ΔE9 mice correlate with the progressive cognitive impairment that these animals exhibit in the Barnes maze assay (56). Fresh-frozen coronal brain sections of male and female, 20-month-old Tg2576 and WT mice were provided by the Centre for Biological Sciences, University of Southampton, U.K.

(Immuno)histochemistry, autoradiography and proteomics
The Gallyas silver stain was performed according to Kuninaka

Statistical analysis
Parametric testing was employed following inspection of the data for normality with the Kolmogorov-Smirnov test in Prism (v6.01; GraphPad Software). Data sets were analyzed by Statistica TM v10 (TIBCO Software Inc., USA). The effects of age, genotype and brain region on the binding levels of [ 18 F]Flortaucipir were analyzed by three-way ANOVA. Gallyas-positive area fraction and tau gene/protein levels were analyzed by two-way ANOVA for the independent factors age and brain region or genotype, respectively. Where ANOVA yielded significant effects, Bonferroni post-hoc comparisons were used to detect between-group regional and agedependent differences. Levels of sarkosyl-insoluble tau between 24-month-old TG and WT mice, and PHF dimensions extracted from TG vs. AD brain were compared by two-tailed independent Student's t-tests. Significance was set at α=0.05. A 1.3-fold change cut-off value for all TMT ratios was used to rank proteins as up-or down-regulated in the proteomics study (62).

Supplementary Materials
Materials and Methods.     Data file S1. Regulated proteins in the sarkosyl-insoluble fraction.

Gallyas silver staining
The

Aβ immunohistochemistry on silver-stained sections
The biotinylated 6E10 antibody (SIG-39340, Nordic BioSite) was used to investigate the association between neurofibrillary pathology and Aβ load in APPswe/PS1ΔE9 mice. Clone 6E10 is raised against amino acids 1-16 of human Aβ, recognizing multiple amyloid peptides and precursor forms (manufacturer information). Silver-stained sections were immersed in 70%

Isolation of sarkosyl-insoluble Tau
The general procedure by Greenberg  the P1, S1, sarkosyl soluble and insoluble fractions were kept for determining Tau protein levels.
Samples were stored at -80°C until further processed.

Tau Meso Scale
Tau

RT-qPCR
For reverse transcription, quantitative polymerase chain reaction (RT-qPCR), Trizol TM -isolated RNA (2 μg) from brain sections of WT and TG mice was reverse-transcribed to cDNA, by using the Applied Biosystems TM high-capacity cDNA transcription kit (Thermo Fisher Scientific).
Samples were analyzed in triplicate on a StepOnePlus TM Real-