Neuronal tau species transfer to astrocytes and induce their loss according to tau aggregation state

Abstract

Deposits of different abnormal forms of tau in neurons and astrocytes represent key anatomo-pathological features of tauopathies. Although tau protein is highly enriched in neurons and poorly expressed by astrocytes, the origin of astrocytic tau is still elusive. Here, we used innovative gene transfer tools to model tauopathies in adult mouse brains and to investigate the origin of astrocytic tau. We showed in our adeno-associated virus (AAV)-based models and in Thy-Tau22 transgenic mice that astrocytic tau pathology can emerge secondarily to neuronal pathology. By designing an in vivo reporter system, we further demonstrated bidirectional exchanges of tau species between neurons and astrocytes. We then determined the consequences of tau accumulation in astrocytes on their survival in models displaying various status of tau aggregation. Using stereological counting of astrocytes, we report that, as for neurons, soluble tau species are highly toxic to some subpopulations of astrocytes in the hippocampus, whereas the accumulation of tau aggregates does not affect their survival. Thus, astrocytes are not mere bystanders of neuronal pathology. Our results strongly suggest that tau pathology in astrocytes may significantly contribute to clinical symptoms.

tau, glia, gene-transfer, design-based stereology, Thy-Tau22

Introduction

For decades, the study of neurodegenerative diseases has focused on mechanisms of toxicity and cell death occurring in neurons.¹ The general assumption was that most of the deleterious events leading to brain dysfunction were essentially sustained by neuronal cell-autonomous molecular cascades. However, several other cell types, such as astrocytes, surround neurons and interact with them, playing major roles in maintaining their normal function and their survival. Among many key functions underlying brain homeostasis, astrocytes build neuronal circuits by regulating both the formation and the elimination of synapses during development,² while in pathological conditions, they are involved in the removal of cellular debris, apoptotic cell bodies and misfolded proteins such as amyloid-β and α-synuclein.^3,-6 Their implication in tauopathies has been much less investigated although the accumulation of tau aggregates in astrocytes has long been described in this family of diseases. In both progressive supranuclear palsy (PSP) and corticobasal degeneration (CBD), neuronal and astrocytic tauopathy coexist yet their distribution does not necessarily match.^7–9 In Alzheimer’s disease, astrocytes were first described to remain free of tau aggregates despite their close proximity to neurofibrillary tangles. However recent studies have now reported frequent co-occurrence of Alzheimer’s disease lesions with astrocytic tau deposition known as ageing-related tau astrogliopathy (ARTAG).^8,10 Under normal physiological conditions, tau is mainly expressed by neurons and very little by astrocytes¹¹; changes in astrocytic tau expression in pathological situations has not been investigated yet. Further, whether the occurrence of astrocytic tau deposition is cell autonomous or derives from the transfer and uptake from neurons remains unclear. Lastly, to what extent accumulation of tau in astrocytes leads to their dysfunction or their death has been under-investigated. To address such questions, we used our recently developed rodent models of tauopathies overexpressing tau variants with a distinct propensity to aggregate.¹² We show that different tau species differentially impacted the pathology development in neurons and astrocytes. Using new gene transfer-based tools, we also demonstrate a transfer of tau between neurons and astrocytes. Finally, we report severe toxicity of soluble hyperphosphorylated forms of tau for specific populations of astrocytes.

Material and methods

Animal experiments

We used different lines of mice for the analysis of hippocampal tau pathology: (i) adult male C57BL/6J mice (2 months old, 22 g; Janvier, Charles River, n = 56); (ii) adult male heterozygous Aldh1L1-GFP mice (5–6 months old, 30 g; n = 42); these mice were back-crossed on a pure C57bl6 background, and were a kind gift from Dr E. Audinat; (iii) adult male homozygous tau knock-out mice and their wild-type littermates (5–6 months old, 30 g; n = 10), these mice were purchased from Jackson lab [Stock: Mapttm1(EGFP)Kit/J] and bred on a C57BL/6 genetic background for more than 10 generations (Supplementary Table 1); (iv) aged transgenic heterozygous Thy-Tau22 mice (6, 12 and 18 months old, 30 g; a kind gift from L. Buée’s lab, Lille, France; n = 12 total). All animal studies were conducted according to the French regulation (EU Directive 86/609 French Act Rural Code R 214-87 to 131). The animal facility was approved by veterinarian inspectors (authorization n°B 92-032-02) and complies with Standards for Humane Care and Use of Laboratory Animals of the Office of Laboratory Animal Welfare (OLAW: n°#A5826-01). All procedures received approval from the local ethical committee (Comité d’Ethique en Expérimentation Animale CEA) and the French Ministry of Research (APAFIS#4794-2016040515053618 v2). Animals were grouped by five and housed in a temperature-controlled room maintained on a 12-h light/dark cycle. Food and water were available ad libitum and nesting material was added to the cage.

AAV vector construction and production

Three human 1N4R tau constructs were designed including wild-type tau (hTAU^WT), pro-aggregating tau (hTAU^ProAggr) as well as the pro-aggregation peptide alone (TauRD.ΔK280). Control groups overexpressed either the green fluorescent protein (GFP) or infra-red fluorescent protein (iRFP). See Figs 1A and 4A for more details on construct design. For AAV production, transgenes were inserted into a shuttle plasmid using Gateway^® recombination cloning technology (Invitrogen). The different tau constructs and the control GFP and iRFP genes were expressed either under control of the chimeric cytomegalovirus/chicken β-actin (CBA) promoter, the neuron-specific synapsin II (Syn2) promoter or the astrocyte-specific GFAabc1d promoter.¹³ The constructs were cloned into a single stranded, rAAV2-based shuttle vector containing the woodchuck hepatitis virus post-transcriptional regulatory element (WPRE) sequence and the bovine growth hormone polyadenylation (bGHpA) signal as termination. All constructs were packaged into AAV9 capsids by the MIRCen viral production platform as described¹⁴; for further details see the Supplementary material. Vector concentrations were assessed by quantitative PCR and expressed as viral genomes per millilitre (Vg/ml).

Stereotaxic injections of AAV vectors

AAV vectors were injected bilaterally into the dorsal cornu ammonis 1 (CA1) layer of the hippocampus at the following stereotaxic coordinates¹⁵: −2.4 mm rostral to bregma, 1.5 mm lateral to midline and −1.6 mm from the skull surface, tooth bar set at 0 mm. For constructs under the CBA promoter each injection site received 2 µl containing 1 × 10¹⁰ Vg of each vector while cell-specific vectors were delivered in 3 μl containing 5 × 10¹⁰ Vg of each vector. The viral solution was delivered using a microdialysis pump (Stoelting) set at 0.2 µl/min. The methods are detailed in the Supplementary material.

Immunohistology

One or 3 months after injection, mice received a lethal dose of pentobarbital (Exagon^®, Axience, 400 mg/kg) by intraperitoneal injection. Animals then underwent intracardiac perfusion of 0.9% NaCl followed by 4% paraformaldehyde in 0.01 M PBS. Brains were collected, post-fixed overnight in 4% paraformaldehyde at +4°C and transferred into 30% sucrose solution in PBS for cryoprotection before sectioning. Twelve series of 30-μm thick coronal sections spanning the entire hippocampus were collected using a freezing microtome (Leica).

A detailed protocol for immunohistochemistry is provided in the Supplementary material.

Stereology

Unbiased stereological counting of sex determining region Y-box 2 (SOX2)-positive astrocytes was performed using the optical fractionator of Mercator software (Explora Nova, La Rochelle, France. The counting procedure is detailed in Supplementary material.

Gene expression in isolated astrocytes

Aldh1L1-GFP mice expressing GFP under the astrocyte-specific promoter Aldh1L1 were injected with either CBA-hTAU^WT, CBA-hTAU^ProAggr or CBA-iRFP. Mice were euthanized 1 and 3 months after injection, the hippocampal formation isolated and processed for fluorescence-activated cell sorting (FACS) of GFP-positive astrocytes. The sorting procedure is detailed in the Supplementary material. See Supplementary Table 3 for primers.

Other protocols

Detailed protocols for fluorescent in situ hybridization (FISH) and western blot are provided in the Supplementary material.

Gallyas silver impregnation was performed as described in d’Orange et al.¹²

Statistical analysis

Results are presented as mean ± standard error of the mean (SEM). Statistical analysis was carried out using Statistica 13 software (Statsoft, Tulsa, OK, USA). Prior to analysis, the data were assessed for normality and homogeneity of variance. If it fulfilled the criteria for general linear model, it was analysed by one-way ANOVA followed by Bonferroni’s post hoc tests for pairwise comparisons. Otherwise, a non-parametric equivalent was used. Annotations used to indicate level of significance are as follows: *P < 0.05, **P < 0.01, ***P < 0.001. The arcsine transformation of percentages was used to normalize data in order to perform statistical analysis.

Data availability

Raw data were generated at MIRCen, Commissariat à l’Énergie Atomique et aux Énergies Alternatives, Fontenay-aux-Roses, France.

Results

Tau variants induce differential neuronal and astrocytic tau pathology

In our previous work, we generated two models of pure tauopathy in the hippocampus of rats using AAV-mediated gene-transfer of the 1N4R isoform of human tau.¹² These models use a ubiquitous CBA promoter to drive transgene expression. They allow the overexpression of either wild-type (hTAU^WT) human tau leading to soluble phosphorylated tau (p-tau), or the chimeric protein (hTAU^ProAggr) containing hTAU^WT and a pro-aggregation peptide (TauRD.ΔK280) cleaved through a 2A peptide sequence (P2A), generating high amounts of neurofibrillary tangles.¹² Here, we replicated the same experimental design in mice. We delivered AAVs bilaterally in the dorsal hippocampus by stereotaxic injections and performed histological studies 1 and 3 months after injection to follow the progression of the disease (Fig. 1A). We first verified that human tau expression levels were similar between CBA-hTAU^WT- and hTAU^ProAggr- injected groups by quantifying hippocampal HT7-positive total tau integrated density on histological sections (Supplementary Fig. 1A). These results were further confirmed by western blot analysis on a separate cohort of animals (Supplementary Fig. 1B and C). We then evaluated tau pathology using several anti-tau antibodies targeting pathological hyperphosphorylation- (AT8, pS422) and conformation-dependent (AT100) epitopes. As early as 1 month post-injection, both models displayed AT8-positive p-tau staining in the pyramidal layer of the hippocampus associated with mislocalization of the protein in the somatodendritic compartment (Fig. 1B and F). Interestingly, p-tau was also detected in the soma and fine processes of astrocytic cells but only in the CBA-hTAU^ProAggr group as shown by double-immunofluorescence for pS422 tau and GFAP (Fig. 1C and G). Misfolded AT100-positive tau staining (Fig. 1H) and Gallyas silver impregnation to visualize mature aggregates (Fig. 1I) further revealed the presence of both neuronal (reminiscent of neurofibrillary tangles) and astrocytic tau pathology in the CBA-hTAU^ProAggr group in the pyramidal and surrounding layers as well as in the subiculum. Tau aggregates in the CBA-hTAU^ProAggr group resisted proteinase K digestion as AT8- and AT100-positive immunostaining in neurons and astrocytes remained after pretreatment of tissue sections (Supplementary Fig. 1D). In contrast, no misfolded tau pathology and no mature Gallyas-positive tau aggregates could be observed in any cell type when overexpressing hTAU^WT alone or the control GFP construct, as expected (Fig. 1D, E and Supplementary Fig. 2A and B). Further characterization of both our AAV-tau models revealed the presence of AT8-positive tau deposits in inhibitory GAD67- and parvalbumin-positive hippocampal interneurons (Supplementary Fig. 3A) in addition to those in CA1 pyramidal neurons. At 3 months post-injection, we also detected AT8-positive phosphorylated tau varicosities in white matter tracts containing the axons of CA1 transduced neurons and in rare oligodendrocytes of the corpus callosum (Supplementary Fig. 3B and C).

Figure 1

Tau variants induce distinct neuronal and astrocytic tau pathology. (A) Schematic representation of the vectors used for tau pathology induction and the experimental design (scale bar = 200 μm). All constructs were expressed under the CBA promoter and contained the WPRE expression-enhancing sequence as well as the bGHpA tail. Tau constructs were composed of the 1N (yellow) 4R (red) tau isoform. The hTAU^ProAggr construct enables a 1:1 expression ratio of full-length wild-type TAU (hTAU^WT) and pro-aggregation peptide (TauRD.ΔK280) after intracellular cleavage of the P2A sequence. Histological analysis was performed 1 and 3 months after injection. Tau pathology was assessed by immunohistochemistry using different antibodies for pathological phosphorylation (B and F: AT8, C and G: pS422 in green, GFAP in red) and a conformation-dependent (D and H: AT100) antibody on hippocampal sections, 1 month after injection. Mature aggregates were revealed by Gallyas silver staining (E and I) on hippocampal sections, 1 month after injection. Arrows indicate neuronal tau pathology, arrowheads indicate tau-positive astrocytes (scale bars = 40 μm). The different hippocampal layers are represented as follows: so = stratum oriens; sp = stratum pyramidale; sr = stratum radiatum. (J) Three months after injection, co-localization of AT100-positive staining with GFP-positive astrocytes (arrowhead) on hippocampal sections of Aldh1L1-GFP mice overexpressing hTAU^ProAggr further confirmed the appearance of an astrocytic tau pathology in our model (scale bars = 20 μm). (K) Confocal images of FISH in the CA1 hippocampus of CBA-hTAU^ProAggr-injected mouse reveals no co-localization between human tau transgene (blue) and AT100-positive (green) GFAP-positive astrocytes (red) at 1-month post-injection demonstrating that in astrocytes tau aggregates are independent of tau transgene expression. Arrowheads show AT100- and GFAP-double positive astrocytes (scale bars = 40 μm).

We then investigated the presence of tau inclusions in astrocytes using an alternative approach that would be independent of GFAP labelling. To that aim, we injected Aldh1L1-GFP mice with CBA-hTAU^ProAggr. Co-localization of AT100-positive staining with GFP-positive astrocytes on hippocampal sections (Fig. 1J) further confirmed the occurrence of astrocytic tau pathology in this model. Within a same animal we observed various abnormal tau distributions in astrocytes reminiscent of human astrocytic tau pathologies. The different types of astrocytic tau inclusions included tufted astrocyte-like cells with long and thin tau inclusions around the nucleus and in the proximal processes, astrocytic plaque-like lesions with round tau deposits in distal cell processes and granular fuzzy astrocyte-like cells with dense accumulations of tau in the perinuclear region and punctate tau inclusions in the proximal processes. These different types of lesions were similar to those found in PSP, CBD, and ARTAG human cases, respectively (Supplementary Fig. 2C). Interestingly, we found AT100-positive astrocytes not only in close vicinity to AT100-positive neuronal somas or dendrites but also in regions with no neuronal AT100-positive staining (Supplementary Fig. 2D). Hence our findings show that tau variants induced differential neuronal and astrocytic tau pathology. Tau was present in a soluble phosphorylated form solely in neurons in the CBA-hTAU^WT group, whereas phosphorylated and insoluble tau was detected in neurons as well as in astrocytes in the CBA-hTAU^ProAggr group. Since we used a supposedly ubiquitous CBA promoter, we ensured that the presence of astrocytic tau inclusions did not result from the direct transduction of astrocytes by the viral vector. First, we determined the cellular targeting of our AAVs by examining the nature of the transduced cells in the CBA-GFP control group. GFP-positive cells were mostly confined to the CA neuronal pyramidal layer. Double-immunostaining with a GFAP antibody demonstrated that astrocytes were not GFP-positive whereas pyramidal neurons were (Supplementary Fig. 2E), suggesting that transgene expression was restricted to neurons after AAV transduction. We then performed FISH experiments to confirm the identity of the cell types expressing the transgene mRNA in CBA-hTAU^ProAggr-injected animals. We used an anti-sense probe targeting the bGHpA sequence present in all our AAV backbones followed by double immunolabelling for misfolded AT100-positive tau and GFAP-positive astrocytes. The specificity of the probe for the transgenes’ mRNA was validated by the absence of staining in a control condition using the sense probe and in a non-injected animal (Supplementary Fig. 2F). Staining revealed that the tau transgene was essentially detected in the pyramidal neuronal layer of the hippocampus (Fig. 1K). Even though some probe-positive cells were observed in surrounding layers, these did not correspond to GFAP-positive astrocytes. Most importantly, AT100 and GFAP double-positive astrocytes did not co-localize with the transgene probe. These observations demonstrated that the presence of astrocytic tau was not driven by the overexpression of the transgene within these cells. Intra- and interregional astrocytic heterogeneity has recently been described in the hippocampus.¹⁶ We next investigated whether the hippocampal subfields containing astrocytes were differentially affected by our two types of tauopathy.

The subiculum is severely affected by tauopathy

To evaluate whether soluble and insoluble tau pathologies differentially affected hippocampal subfields in our models, we quantified tau immunolabelling in the subiculum, the pyramidal layer and in the surrounding layers stratum oriens and stratum radiatum which reflect both neurodendritic and astrocytic pathology (Fig. 2A). Regarding tau phosphorylation, AT8 burden did not differ between CBA-hTAU^WT and CBA-hTAU^ProAggr groups at any time point in any region. However there was a significant decrease with time in both CBA-hTAU^WT and CBA-hTAU^ProAggr groups in the subiculum and in pyramidal layer [two- factor ANOVA, time effect in subiculum F(1,17) = 36.1; P = 1 × 10⁻⁵; in pyramidal layer F(1,17) = 46.22; P = 3 × 10⁻⁶] (Fig. 2B). AT8-immunopositivity was stronger in the subiculum, compared to the other subregions with a higher fraction of AT8-immunopositivity at both time points. We next focused on the conformation-dependent AT100-positive staining by quantifying the percentage of AT100-positive volume (Fig. 2C) and the number of AT100-positive somas (Fig. 2D). Pathology was intense in all regions examined in the CBA-hTAU^ProAggr group, in particular in the subiculum for which the volume fraction affected was ∼5-fold that of CA1-2-3 neuropil layers. AT100-positive volume remained stable in the CA1-2-3 subregions despite a qualitative redistribution of staining from neuronal dendrites to astrocytes (Fig. 2E). On the other hand, AT100 tau pathology decreased by 61% in the subiculum from 1 to 3 months after injection (t-test, t-value = 3.02, df = 9, P = 0.014) (Fig. 2C). Altogether, these data indicate that the subiculum is particularly affected by tau lesions in our models of tauopathy, compared to the other sublayers. In addition, while p-tau load is similar on the two models, aggregated species were only detected in the CBA-hTAU^ProAggr group.

Figure 2

The subiculum appears the most vulnerable hippocampal subfield for tau pathology. (A) Schematic representation of the different hippocampal subfields analysed including the subiculum, the pyramidal layer of the hippocampus and the surrounding layers stratum oriens and stratum radiatum. (B) Phosphorylation of tau at the AT8 epitope does not differ between tau variants regardless of the hippocampal subfield (quantification of AT8-positive volume at 1 and 3 months post-injection; two-factor ANOVA and Bonferroni’s test, *P < 0.05, **P < 0.01, ***P < 0.001). However, it appears more severe in the subiculum as indicated by the elevated percentage of affected volume. (C and D) Misfolded tau pathology present in the CBA-hTAU^ProAggr group remains stable between 1 and 3 months after injection in all subregions except the subiculum (C, quantification of AT100-positive volume in the subiculum and the stratum oriens and stratum radiatum; and D, quantification of the number of AT100-positive pyramidal somas at 1 and 3 months post-injection, paired t-test, *P < 0.05). (E) Representative images of misfolded AT100-positive tau pathology in the subiculum (scale bars = 50 μm) and the CA1-2–3 layers (scale bars = 200 μm, inset = 50 μm) of the hippocampus in the CBA-hTAU^ProAggr group at 1 and 3 months post-injection (arrows indicate neuronal tau pathology, arrowheads indicate tau-positive astrocytes). (F) Representative images of neuronal NeuN and cell nuclei (DAPI) staining in the different experimental groups 3 months after AAV injection demonstrating neuronal cell loss in the pyramidal layer of the hippocampus in the CBA-hTAU^WT-injected group as indicated by the arrow (scale bars = 200 μm). (G and H) Quantification of the volume of different subfields of the hippocampus 3 months after injection on DAPI-stained sections. All analysed hippocampal subfields (G) and, as a result, the total hippocampus (H) were atrophied in the hTAU^WT group compared to the GFP control (two- factor ANOVA and Bonferroni’s test, **P < 0.01, ***P < 0.001) while no significant atrophy was observed in the hTAU^ProAggr group.

We then wondered how tau pathology related to structural atrophy, a major hallmark of tauopathies, in both models. Hence, we measured the volumes of neuronal and neuropil layers 1 and 3 months post-injection. In the CBA-hTAU^WT group, we detected a significant reduction of the volume of CA1-2-3 neuronal pyramidal layer [−44%, two-factor ANOVA, group effect F(2,27) = 8.2; P = 0.002, Bonferroni P = 0.02] and the surrounding layers stratum oriens and radiatum [−43%, two-factor ANOVA, group effect F(2,27) = 12; P = 0.0002, Bonferroni P = 0.02] at 3 months, compared to the GFP control group (Fig. 2G, 1 month data in Supplementary Fig. 4A). Extensive neuronal loss was obvious on NeuN immunostained sections in the CBA-hTAU^WT group (Fig. 2F). The volume of the subiculum was reduced by 75% from 1 month [two-factor ANOVA, group effect F(2,27) = 25.8; P = 1 × 10⁻⁶, Bonferroni P = 0.002] and remained atrophied by 76% at 3 months (Bonferroni P = 0.0002) compared to the GFP control group. Consistent with these drastic changes in the hTAU^WT group, the total volume of the hippocampus was atrophied as early as 1 month post-injection and further progressed by −46% [two-factor ANOVA, group effect F(2,27) = 15.8; P = 0.00003, Bonferroni P = 0.004] 3 months after injection compared to controls, while it remained unchanged in the other groups, including in the CBA-hTAU^ProAggr group (Fig. 2H, 1 month data in Supplementary Fig. 4B). These data show that in mice, as previously described in rats, soluble p-tau pathology leads to severe tissue loss whereas aggregated forms of tau do not. Furthermore, the subiculum is more severely affected than the CA1-2-3 layers, and astrocytes display severe tau pathology in the hTAU^ProAggr model. We therefore investigated the origin of the astrocytic pathology in this model.

Endogenous murine tau contributes to astrocytic tau inclusions but is dispensable

Although the level of tau expression in neurons and astrocytes has not been thoroughly described in clinical cases, it is known that MAPT gene duplication increases tau levels in parallel with the risk of developing tauopathies.^17,18 We wondered whether under the stressful pathological conditions associated with our models of tauopathy, astrocytes could respond by upregulating the expression of endogenous tau, favouring its aggregation in situ. To test this possibility, we injected Aldh1L1-GFP mice in the hippocampus with either CBA-hTAU^WT inducing soluble p-tau, CBA-hTAU^ProaAggr generating tau aggregates or a CBA-iRFP control vector. One or 3 months later, we dissected their hippocampi and isolated GFP-positive astrocytes by FACS (Supplementary Fig. 5A and B), and performed real time quantitative PCR (RT-qPCR) using primers specific for murine Mapt. For both time points, we observed no difference in astrocytic murine tau expression between the control iRFP- and tau-injected animals (Supplementary Fig. 5C), even in the hTAU^ProAggr group where both neuronal and astrocytic tau-positive aggregates were detected (Supplementary Fig. 5D and E). Thus, astrocytic endogenous tau expression does not increase in pathological conditions of neuronal tauopathy.

Under basal conditions, the MAPT gene is expressed 5 and 10 times less in astrocytes than in neurons in humans and mice, respectively.^11,19 We wondered whether the minimal amount of tau normally expressed in astrocytes was necessary to the occurrence of tau aggregates in these glial cells. Therefore, we injected tau-deficient mice (tau knockout) and their wild-type littermates in the hippocampus with CBA-hTAU^ProaAggr or a control vector CBA-iRFP, and evaluated tau pathology 3 months later. We hypothesized that if murine astrocytic tau was required, no astrocytic tau aggregates would be detected in tau knockout mice injected with CBA-hTAU^ProaAggr. We found that misfolded and aggregated tau was reliably detected both in neurons and astrocytes with AT100 immunohistochemistry (Fig. 3A and Supplementary Fig. 6A) and Gallyas staining (Fig. 3B) in both wild-type and tau knockout mice overexpressing hTAU^ProaAggr. To evaluate the contribution of murine tau more precisely, we quantified the number of AT100-positive neurons in the CA1-2-3 neuronal layer and in the subiculum, as well as the percentage of AT100-positive volume in the surrounding layers encompassing neuronal dendrites and astrocytes. Astrocytic and dendritic pathology was markedly lower in tau knockout mice compared to wild-type littermates (−75% in stratum oriens/radiatum, t-value = 8.43; df = 8; P = 3.10-5; and −80% in subiculum t-value = 2.44; df = 8; P = 0.04; Fig. 3D) and this effect seemed independent of the number of pathological neurons as the latter was identical in both groups (t-value = −0.32; df = 8; P = 0.75; Fig. 3E). Hence endogenous murine astrocytic tau is not required for the appearance of tau aggregates, but it contributes to and potentiates their formation. Notably, co-localization of pSer422 human tau immunostaining and murine tau detected with T49 antibody evidenced an entanglement of human and endogenous murine tau in both neurons and astrocytes in wild-type littermates only (Fig. 3C and Supplementary Fig. 6B). We performed FISH experiments followed by double immunolabelling for AT100-positive tau aggregates and GFAP-positive astrocytes to determine whether tau-positive astrocytes were transduced by the AAV in the tau-injected tau knockout mice. Qualitative analysis showed that the hTAU transgene was mainly expressed in the pyramidal neurons of the hippocampus (Supplementary Fig. 6C). Most importantly, no co-localization between the transgene probe and the AT100-positive GFAP-positive astrocytes was observed, showing that AT100-positive astrocytes were not transduced by the viral vector. Hence in our hTAU^ProAggr model, when pathology initiates in neurons, tau aggregates can be detected in astrocytes even in the absence of endogenous astrocytic tau. Therefore, we next assessed whether tau species could be transferred from neurons to astrocytes.

Figure 3

Endogenous murine tau contributes to astrocytic tau inclusions formation but is dispensable. To investigate the contribution of normally expressed endogenous tau on the formation of glial tau aggregates, tau-deficient mice [tau knockout (KO)] and their wild-type (WT) littermates were injected with AAV-CBA-hTAU^ProAggr. Three months after injection, histological analysis of the hippocampus revealed misfolded AT100-positive tau lesions (A, scale bars = 200 μm, inset = 50 μm) and Gallyas-positive mature aggregates (B, scale bars = 40 μm) in both neurons (arrows) and astrocytes (arrowheads) regardless of the genotype. The different hippocampal layers are represented as follows: so = stratum oriens; sp = stratum pyramidale; sr = stratum radiatum; sub = subiculum. (C) Three months after injection, confocal images of co-localization of hyperphosphorylated pS422-positive human tau (represented in green) and T49-positive murine tau (represented in red) in the hippocampus showed mixed human and murine tau tangles in neurons (arrow) and astrocytes (arrowheads) in wild-type littermates only (scale bars = 40 μm). (D) Quantification of the AT100-positive volume showed a smaller affected fraction volume in the subiculum and the stratum oriens and stratum radiatum layers of tau knockout mice compared to wild-type littermates (paired t-test, *P < 0.05, ***P < 0.001). (E) The number of AT100-positive somas in the pyramidal layer of the hippocampus did not differ between genotypes (paired t-test, not significant).

Neurons and astrocytes can exchange tau species

To evaluate the possibility that neurons and astrocytes can exchange tau species, we engineered a reporter system based on the design of the hTAU^ProAggr vector, where the co-expression of the pro-aggregating peptide (TauRD.ΔK280 tagged with a P2A sequence) along with human wild-type tau (hTAU^WT) leads to the aggregation of the full-length tau, detectable by the human-specific and conformation-dependent AT100 antibody. Importantly, the epitope recognized by AT100 is not present in the sequence of TauRD.ΔK280 (Fig. 4B). Here, we decided to overexpress TauRD.ΔK280 and hTAU^WT separately, each under a cell-specific promoter, either neuronal (Syn2) or astrocytic (GFAabc1d) (Fig. 4A). We reasoned that AT100-immunopositive tau staining would only be detected if the two proteins came into contact to form tau aggregates, and not if they remained in their respective cell type. We therefore injected mice in the hippocampus with a mixture of Syn2-TauRD.ΔK280 and GFAabc1d-TAU^WT and used AAVs overexpressing GFP under both promoters as controls (Fig. 4C). Cell-specificity of each vector was first ascertained by co-localization of total tau staining (with HT7 antibody) and TauRD.ΔK280 (with P2A antibody) with either NeuN or GFAP staining in neurons and astrocytes respectively (Supplementary Fig. 7A and B). As expected, no AT100 staining was detected in any of the two control groups while Gallyas staining was only visible in the Syn2-TauRD.ΔK280-injected group, consistent with the capacity of TauRD.ΔK280 to aggregate endogenous murine tau into mature fibrils²⁰ (Fig. 4C). In groups co-injected with the combination of Syn2-TauRD.ΔK280 and GFAabc1d-TAU^WT, both AT100-positive misfolded and Gallyas-positive aggregated tau were detected in neuronal cells of the pyramidal layer as well as in astrocytic cells in the surrounding layers. Both stainings also revealed sparse granular material in the stratum oriens, stratum radiatum and the subiculum reminiscent of cellular debris and astrocytic plaques. Pathology increased over time (3 months post-injection in Fig. 4C, 1 month post-injection time point in Supplementary Fig. 7C). Confocal imaging confirmed that tau aggregates were present in the cell soma and processes of neurons as well as in the fine processes and cell body of astrocytes (Fig. 4D and Supplementary Fig. 7D). Taken together, these data show that tau species can transfer between neurons and astrocytes, leading to the presence of aggregated tau in both cell types.

Figure 4

Neurons and astrocytes can exchange tau species in vivo. (A) Schematic representation of the four vectors used for tau trans-cellular transfer experiments. Vectors contained either the full-length wild-type tau alone (hTAU^WT) or the pro-aggregating peptide alone (TauRD.ΔK280). GFP was used as control. Constructs were expressed in a cell-specific manner under either the neuron-specific Synapsin (Syn2) promoter or the astrocyte-specific GFAabc1d promoter. All constructs contained the WPRE expression-enhancing sequence as well as the bGHpA tail. (B) Schematic representation of human tau proteins showing the localization of anti-tau antibody epitopes used in this study. HT7 recognizes full-length tau while AT100 antibody recognizes only misfolded human tau and cannot bind to the pro-aggregation peptide TauRD.ΔK280. The latter was designed with a P2A site enabling its specific detection using an anti-P2A antibody. (C) Representative images of AT100- (scale bars = 200 μm, inset = 50 μm in CA1) and Gallyas- (CA1, scale bars = 50 μm) positive tau staining in the different conditions 3 months after injection. Both AT100- and Gallyas-immunopositivity revealed that the neuronal pro-aggregating peptide (TauRD.ΔK280) and the astrocytic full-length tau (hTAU^WT) came into contact leading to both neuronal (arrow) and astrocyte-like (arrowhead) tau aggregates. Tau pathology was absent from the control combinations except for Gallyas-positivity in the Syn2-TauRD.ΔK280 condition. (D) Confocal images of triple immunostaining for tau (AT100, green), neurons (NeuN, blue) and astrocytes (GFAP, red) confirmed the presence of both neuronal (top row) and astrocytic (bottom row) tau aggregates 3 months after injection in the Syn2-TauRD.ΔK280/GFAabc1d-hTAU^WT combination (CA1, scale bars = 50 μm).

To corroborate this finding, we hypothesized that tau-positive astrocytes should also be detected in transgenic mouse models where pathology is induced in neurons. Hence, we looked for astrocytic tau pathology in 6-, 12- and 18-month-old Thy-Tau22 transgenic mice which express human TAU with a double mutation G272V/P301S under a neuron-specific Thy1.2 promoter.²¹ In addition to the previously reported neuronal AT100-positive tau inclusions, confocal images of double-immunolabelling for AT100 and GFAP revealed sparse AT100-positive astrocytes in the stratum oriens and stratum radiatum of the CA1 region in close vicinity to AT100-positive dendrites and near AT100-positive granular cells of the dentate gyrus at advanced stages of pathology, i.e. 12 and 18 months of age (Fig. 5A–C, Supplementary Fig. 8A and B). These observations in a neuron-targeted tauopathy model strengthen the idea that astrocytic tau inclusions are secondary to neuronal pathology.

Figure 5

Astrocytic tau inclusions are secondary to neuronal pathology in the neuron-targeted Thy-Tau22 mouse model of tauopathy. (A and B) Confocal images in the stratum radiatum of (A) 12-and (B) 18-month-oldtransgenic Thy-Tau22 mice showed AT100 (green) and GFAP (red) double-immunopositive astrocytes (arrowhead) in close vicinity to AT100-positive dendrites (arrow). (C) At lower magnification, confocal images in the dentate gyrus of an 18-month-old Thy-Tau22 mouse revealed several AT100 (green) and GFAP (red) double-immunopositive astrocytes (arrowheads) in proximity to AT100-positive granular cells (arrow). Scale bars = 40 μm.

We demonstrated that soluble phosphorylated forms of tau (induced by CBA-TAU^WT) triggered severe hippocampal volume reduction while mature aggregates (induced by CBA-TAU^ProAggr) did not. We next examined whether astrocyte loss contributed to this atrophy.

Soluble hyperphosphorylated tau is toxic to astrocytes in the subiculum

Using an unbiased stereological counting method, we estimated the number of astrocytes labelled with a nuclear marker to facilitate their detection (Fig. 6A). We chose the nuclear transcription factor SOX2 expressed in over 97% of mature post-mitotic astrocytes in the adult hippocampus.²² The number of SOX2-positive astrocytes was counted in the hippocampus of mice injected with CBA-hTAU^WT and CBA-hTAU^ProAggr 1 and 3 months after injection. In the CBA-hTAU^ProAggr and CBA-GFP groups, this number was stable over time and in all subfields of the hippocampus. In contrast, in the group injected with CBA-hTAU^WT, the number of SOX2-positive astrocytes in the subiculum was significantly smaller by 30% compared to controls [two-factor ANOVA, group effect in subiculum F(2,27) = 18.6; P = 8 × 10⁻⁶, Bonferroni P = 0.04; Supplementary Fig. 9] as early as 1 month after injection. The astrocytic loss further aggravated by −48% 3 months post-injection (Bonferroni P = 0.0002; Fig. 6B and C). In CA1-2-3, the number of SOX2-positive astrocytes reduced with time by 25% in the hTAU^WT group [two-factor ANOVA, time effect F(2,27) = 7.9; P = 0.009, Bonferroni P = 0.004, 1 month versus 3 months post-injection], while not being significantly different from the other groups [two-factor ANOVA, group effect F(2,27) = 2.1; P = 0.14]. Altogether, our data show that soluble tau species are highly toxic to subicular astrocytes, whereas aggregated tau forms are not.

Figure 6

Soluble hyperphosphorylated tau is toxic to astrocytes in the subiculum. (A) Confocal images of double immunolabelling for SOX2 (green) and GFAP (red) (scale bars = 40 μm). SOX2 staining reflects most of the astrocytic population of the mouse hippocampus. (B) Representative images of astrocytic SOX2 staining in the subiculum 3 months after AAV injection (scale bars = 200 μm). (C) Stereological estimation of the number of SOX2-positive cells in different subfields of the hippocampus at 3 months post-injection. In the CBA-hTAU^WT group, there was a significant reduction in the number of SOX2-positive cells in the subiculum only (two-factor ANOVA and Bonferroni’s test, *P < 0.05, ***P < 0.001).

Discussion

Accumulation of various abnormal forms of tau in neurons and astrocytes are common hallmarks of tauopathies. The expression of tau protein in astrocytes in physiological conditions being low, the origin of astrocytic tau in these diseases is still unknown. To this day, recapitulation of tau pathology in astrocytes as seen in human patients has been mainly obtained by specific tau overexpression in astrocytes^23,24 or by injecting brain homogenates from distinct tauopathy patients into tau transgenic or wild-type mouse brains.^25,26 Here, we used innovative gene transfer tools to model tauopathies in adult mouse brains and to investigate the origin of astrocytic tau. We next determined the consequences of tau accumulation in astrocytes on their survival in models displaying various status of tau aggregation. We show that astrocytic tau pathology can emerge secondarily to neuronal pathology, mediated by bidirectional exchanges of tau species between neurons and astrocytes. We further report that as for neurons, soluble tau species are highly toxic to some subpopulations of astrocytes in the hippocampus, whereas the accumulation of tau aggregates does not affect their survival.

Using AAV-based gene transfer to overexpress different variants of human tau in the hippocampus of adult mice, we observed a neuronal pathology similar to that we initially reported in our different rat models.¹² From 1 month after injection, overexpression of hTAU^WT in neurons resulted in robust tau hyperphosphorylation detected with antibodies recognizing a number of pathological phospho-epitopes (AT8, pSer422), in the absence of any mature fibrillar aggregates up to 3 months, and in the absence of any detectable tau species into astrocytes. On the other hand, overexpression of the hTAU^ProAggr construct led to the additional formation of mature neurofibrillary tangles and Gallyas-positive tau aggregates within astrocytes in the CA and subicular regions. We then investigated the origin of astrocytic tau aggregates. We first ascertained by in situ hybridization that in our AAV models, pathology was induced in neurons and that tau accumulation in astrocytes was not the result of direct transduction by AAVs. We further established that hyperphosphorylated and misfolded astrocytic tau could be detected at advanced stages of pathology (from 12 months of age) in a commonly used transgenic model (Thy-Tau22²¹) overexpressing mutant tau specifically in neurons. This is in line with observations made in 24-month-old rTgTauEC mice expressing P301L-tau under the control of neuropsin neuronal promoter²⁷ and in mice expressing N279K human tau under the human tau promoter.²⁸ These data and ours suggest that astrocytic tau pathology is secondary to neuronal pathology. We then wondered whether under stressful pathological conditions, astrocytes could respond by upregulating the expression of endogenous tau, favouring its aggregation in situ. Yet when we measured the changes in endogenous tau expression on FACS-sorted hippocampal astrocytes under tau pathological conditions, endogenous astrocytic tau expression remained unchanged compared to the control condition. Our sorting method only allowed us to measure the global expression of tau in the whole population of hippocampal astrocytes rather than in isolated tau-positive astrocytes. Clearly, further studies are required in animal models and patient samples in order to address this particular point. To gain insight into the contribution of endogenous tau to the formation of astrocytic lesions, we injected our CBA-hTAU^ProAggr vector into tau knockout mice and their littermates. Both neuronal and astrocytic pathologies were detected in tau knockout mice, confirming that endogenous tau was not required for the presence of tau lesions. The number of AT100-positive neurons was the same in both genotypes, suggesting that the same number of neurons was targeted by our vector. Nonetheless, we cannot rule out that tau pathology within each neuron may be reduced in tau knockout mice and may explain why in our study, astrocytic pathology was overtly diminished in tau-deficient mice. Hence, our results demonstrate that even though endogenous murine tau is not required, it contributes to the severity of astrocytic tau pathology. This was further confirmed by the detection of a mixture of murine and human tau in astrocytic lesions in wild-type littermates. In conclusion, we demonstrated that when full-blown tau pathology is triggered in CA1 neurons, it emerges in astrocytes, even in the absence of endogenous tau. These data led us to hypothesize that there may be a transfer mechanism of tau species from neurons to astrocytes.

To substantiate the occurrence of this transfer mechanism, we designed a novel in vivo reporter system with cell-specific promoters using AT100-positive staining as a read-out of tau species encounter. The overexpression of a pro-aggregation peptide TauRD.ΔK280 specifically in neurons and that of hTAU^WT in astrocytes led to considerable amount of AT100 staining both in neurons and astrocytes. While both release and recapture of tau species in neurons have been thoroughly explored in recent years,^29–31 less is known about these mechanisms in astrocytes. Recent in vitro studies have demonstrated the capability for astrocytes to uptake extracellular monomeric, oligomeric and fibrillar tau^32–34 but astrocytic tau release has not yet been thoroughly assessed. Injection of brain homogenates from PSP, CBD or ARTAG patients in wild-type and tau transgenic mice also gives rise to astrocytic lesions,^25,35,36 even in the absence of neuronal tau.³⁷ In both our AAV-based mouse model and the transgenic Thy-Tau22 mouse line displaying neuronal and astrocytic tau pathology, AT100-positive astrocytes were localized in close vicinity to AT100-positive neuronal dendrites. Given their capacity to eliminate synapses during development³⁸ and their close proximity to synapses bearing tau species,^39–41 astrocytes may internalize tau species while phagocytosing pathological synapses.⁴² In favour of this hypothesis, recent studies described an enrichment of complement proteins at the synapses of tau transgenic mice and tauopathy patients^43,44 mediating synapse loss, and one of the complement proteins involved, C1q, is a ligand of the phagocytic receptor MEGF10 expressed by astrocytes.^45,46 In an experimental model of induced apoptosis in vivo, astrocytes have also been shown to effectively remove small and diffuse apoptotic neuritic fragments from dying neurons.⁴⁷

This new evidence of tau transfer to astrocytes encouraged us to investigate a potential toxic effect on astrocytes survival. Few anatomo-pathological studies have provided stereological counts of astrocytes in affected regions of tauopathy patients. Astrocytic loss was reported in patients brains following chronic traumatic encephalopathy,⁴⁸ and in rat models of traumatic brain injury⁴⁹ while apoptotic astrocytes occur at an early stage in the orbito-frontal cortex in Pick’s disease.⁵⁰ Recently, Bussian et al.⁵¹ described increased numbers of senescent astrocytes in tau PS19 mice, although it remains unclear whether senescence leads to cell death. Morphological changes such as simplification of astrocytic arborization or fragmentation of processes have been reported in patients with Alzheimer’s disease^52–54 and 3xTg-AD transgenic mice.⁵⁵ However, actual death of astrocytes has not been clearly evidenced in Alzheimer’s disease using GFAP, ALDH1L1, and GS as astrocytic markers or TUNEL staining, at least in restricted parts of the cortex.^56–58 While subicular astrocytes seemed particularly vulnerable in our mouse model of soluble tauopathy, the number of astroglial cells in the CA region was not significantly reduced. We hypothesize that soluble tau species are not detected in astrocytes in our hTAU^WT model because astrocytes that have recaptured soluble p-tau die quickly. Intra-regional variability of astrocytes has recently been described.¹⁶ Therefore, the proportion of vulnerable and resilient astrocytes may differ between the subiculum and the CA region. To explain the absence of significant astrocyte loss in CA, another possibility is concomitant enhanced death and proliferation. Furthermore, CA astrocytes may engulf soluble p-tau species but degrade them more efficiently than subicular astrocytes. Regional heterogeneity in astrocytic phagocytosis properties may account for the diversity of tau deposition within astrocytes but this remains to be investigated.

Another important finding in our study is that accumulation of aggregated tau in astrocytes, as opposed to soluble tau species, does not compromise the survival of the overall CA and subicular astrocytic population. This is also true for neurons in our study and is in line with our previous study in rats where we showed that soluble tau species were more neurotoxic compared to mature fibrillar tau.¹² While both components of the hTAU^ProAggr vector, i.e. hTAU^WT and TauRD-ΔK280, have been shown to be toxic on their own,^12,59–61 they counteract each other when expressed within the same cell. Our interpretation is that as TauRD-ΔK280 strongly aggregates hTau^WT into innocuous fibrillary material, this process also results in the neutralization of noxious soluble hyperphosphorylated tau generated by both tau species in the cytoplasm. Both stereological counts of the number of astrocytes and hippocampal volume measurements support the notion that soluble hyperphosphorylated tau species are harmful to brain cells. In the CBA-hTAU^ProAggr group, the reduction of AT100-immunopositive staining in the subiculum with time may therefore reflect degradation of abnormal tau rather than cell death. Nonetheless, cell survival does not preclude cell dysfunction. It was shown that tau oligomers internalized by astrocytes disrupt intracellular Ca²⁺ signalling and Ca²⁺-dependent release of gliotransmitters, affecting synaptic transmission in neighbouring neurons.³⁴ Furthermore, in a number of conditions including Alzheimer’s disease and ageing, some astrocytes change their transcriptomic profile and adopt a new phenotype as they lose some neuro-supportive functions⁶² and become senescent⁵¹ suggesting that similar alterations may occur in the context of glial tauopathies.

Our demonstration of tau transfer between neurons and astrocytes implies that astrocytes are not mere spectators of neuronal pathology during tauopathies. While soluble tau species rapidly become toxic to a subpopulation of astrocytes, the effect of long-term accumulation of tau aggregates needs further investigation. We pinpointed the subiculum as a region particularly vulnerable to the formation of tau lesions and to soluble tau toxicity. This brain region receives major inputs from CA1, the visual and the entorhinal cortices, and sends projections back to CA1 neurons and to the retrosplenial and perirhinal cortices.^63,64 As such, it plays an essential role in the formation of complex spatial representations and learning of object-place associations. Hence, it is anticipated that functional cognitive deficits should follow subicular tauopathy. The contribution of astrocytes to the pathophysiology and to clinical symptoms in tauopathies is still elusive but evidence of their involvement now opens promising therapeutic perspectives.

Acknowledgements

The authors thank Audrey Vautheny, Nathalie Déchamps and Jan Baijer for skillful technical advice and assistance at the Cytometry and Cell Sorting facility, Institut de Biologie François Jacob CEA Fontenay-aux-Roses, and Jason Martin for proofreading this manuscript. They also thank Julien Mitja, Laurent Vincent, Kristell Bastide, Corina Dussaud, Benoît Larrat and Sébastien Mériaux for their help with mouse housing and Christoph Schmitz for helpful advice on stereology.

Funding

This work was partially funded by ANR-11-INBS-0011 - NeurATRIS: A Translational Research Infrastructure for Biotherapies in Neurosciences and by ANR-18-C816-0008-03.

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.

 
• AAV

  adeno-associated virus

 
• CA

  cornu ammonis

 
• CBA

  cytomegalovirus/chicken β-actin

 
• GFP

  Green fluorescent protein

 
• hTAU^ProAggr

  human Pro-aggregant chimeric tau construct

 
• hTAU^WT

  human wild-type tau construct

 
• iRFP

  infra-red fluorescent protein