Tau accumulation in astrocytes of the dentate gyrus induces neuronal dysfunction and memory deficits in Alzheimer’s disease

Alzheimer’s disease (AD) is characterized by the accumulation of the tau protein in neurons, neurodegeneration and memory loss. However, the role of non-neuronal cells in this chain of events remains unclear. In the present study, we found accumulation of tau in hilar astrocytes of the dentate gyrus of individuals with AD. In mice, the overexpression of 3R tau specifically in hilar astrocytes of the dentate gyrus altered mitochondrial dynamics and function. In turn, these changes led to a reduction of adult neurogenesis, parvalbumin-expressing neurons, inhibitory synapses and hilar gamma oscillations, which were accompanied by impaired spatial memory performances. Together, these results indicate that the loss of tau homeostasis in hilar astrocytes of the dentate gyrus is sufficient to induce AD-like symptoms, through the impairment of the neuronal network. These results are important for our understanding of disease mechanisms and underline the crucial role of astrocytes in hippocampal function. Alzheimer’s disease is often considered a disease of neurons. This study reveals that astrocytes are also impaired by the disease and that these cells contribute more to memory deterioration than previously thought.


Results
Accumulation of 3R tau in hilar hippocampal astrocytes of individuals with AD. We examined the density of cells expressing a pathological form of phospho-tau in different regions of the hippocampus in individuals with AD and in healthy age-and sex-matched donors (Supplementary Table 1 and Extended Data Fig. 1a-c). We used immunohistochemistry with an AD2 antibody, which recognizes the phosphorylated Ser396 and Ser404 epitopes 7 (Fig. 1a). We found strong variability in the density of AD2 + cells between hippocampal structures but also between individuals with AD, whereas healthy donors showed no AD2 immunoreactivity (Fig. 1b). For each individual, we found that the granule cell layer (GCL) of the dentate gyrus and CA1 and CA3 regions exhibited a higher density of AD2 + cells than the hilus and molecular layer (ML) of the dentate gyrus (Fig. 1b). However, when individuals were graded according to the Braak staging, we found a strong correlation between stage and the density of AD2 + cells in the dentate gyrus, including the ML, GCL and hilus (Fig. 1c,d). A high variability in the number of hippocampal amyloid plaques was also observed between individuals and stages. However, we found no correlation between Braak stage and hippocampal plaque number (Extended Data Fig. 1d-f). Thus, in the hippocampus, the dentate gyrus and in particular the hilus, are increasingly affected by the progression of tau pathology in AD.
We next examined the presence of 3R and 4R tau isoforms in the hilus of healthy donors and individuals with AD using isoform-specific antibodies (Extended Data Fig. 2). The density of 3R tau inclusions in the hilus was higher in individuals with AD who showed the presence of hyperphosphorylated tau, and this increase was exacerbated in individuals who also exhibited amyloid Tau accumulation in astrocytes of the dentate gyrus induces neuronal dysfunction and memory deficits in Alzheimer's disease Articles NATuRE NEuRoScIENcE plaques in the hilus. In contrast, the density of hilar 4R inclusions was only increased in individuals devoid of hyperphosphorylated tau or amyloid plaques, suggesting a transient increase along the course of the disease (Fig. 1e-h).
AD is considered to be primarily a neuronal disease. However, tau has also been found in astrocytes of individuals with AD, with less well-known consequences 8 . We therefore examined the presence of 3R or 4R tau in astrocytes in the hilus of individuals with AD. We observed more 3R tau inclusions per astrocyte and more astrocytes with 3R tau inclusions in individuals with AD than in controls. Furthermore, this accumulation was greater in individuals with the presence of hyperphosphorylated tau and was exacerbated when the hilus exhibited amyloid plaques (Fig. 2a-d). In contrast, no change in astrocytic accumulation of 4R tau was found with disease state (Fig. 2e-g). In the non-astrocytic (S100β − ) compartment, we found a high variability of 3R and 4R tau accumulation, which was not associated with disease state (Extended Data Fig. 3a-d).
The increased accumulation of 3R tau in astrocytes in the disease state was not due to modifications of S100β expression, since the density of S100β + cells was similar between individuals with AD and control donors (Extended Data Fig. 3e-h). Thus, disease state is associated with 3R tau accumulation in hilar astrocytes.
Synaptic failure is a major hallmark of AD resulting in a decrease 9 or an increase in the density of synaptic proteins 10 , depending on disease state and reactive mechanisms. We therefore assessed the expression of the presynaptic protein synaptophysin and the postsynaptic protein PSD95 using immunohistochemistry. The density of PSD95 (Fig. 2h,i)

NATuRE NEuRoScIENcE
was exacerbated by the presence of amyloid plaques. The density of PSD95 immunostaining strongly correlated with the proportion of astrocytes accumulating 3R but not 4R tau, suggesting a link between 3R tau accumulation in astrocytes and synaptic alterations (Fig. 2j,k). Thus, in AD, the hilus of the dentate gyrus is particularly responsive to disease progression and astrocytes accumulate 3R tau, which is associated with synaptic alterations and the severity of the pathology.
Viral strategy for expressing tau in hilar astrocytes of adult mice.
Our observations from human samples raise the possibility that 3R tau accumulation in hilar astrocytes may participate in hippocampal dysfunction and disease etiology. To test this possibility, we developed a new lentiviral vector (LV) to specifically target astrocytes. We used a truncated version of the glial fibrillary acidic protein (GFAP) promoter gfaABC1D, with a B3 enhancer (gfaABC1D(B3), hereafter called G1) and the previously described miR124T neuronal detargeting system (LV-G1-GFP; Fig. 3a). To assess the astrocytic specificity of this construct in the adult mouse brain, we injected the reporter LV-G1-GFP in the hilus of the dentate gyrus of adult mice. Two weeks after injection, we examined the distribution and identity of the green fluorescent protein (GFP)-expressing cells. All GFP + cells were found in the dentate gyrus, with a majority (68.4% ± 1.7%) in the hilus (Fig. 3b,c). Regardless of their position, virtually all GFP + cells expressed GFAP (Fig. 3d) and S100β (Fig. 3e), which are astrocytic markers. Adult hippocampal stem cells that reside in the subgranular zone of the dentate gyrus also express GFAP and could potentially be targeted by the LV. To assess stem cell targeting, we injected another set of mice. Four days after LV injection, 8.1% ± 2.1% of GFP + cells exhibited radial glial-like (RGL) stem cell morphology, with a soma in the subgranular zone and a radial process extending into the GCL (Extended Data Fig. 5), suggesting that few adult neural stem cells may have been targeted by the LV. These cells represented 11% ± 1.2% of all RGL cells, and this proportion decreased to 3.0% ± 1.1% at 14 and 120 days after LV injection. Furthermore, only 0.4% ± 0.4% of GFP + cells expressed the mature neuronal marker NeuN (Fig. 3f), suggesting that the neuronal detargeting system disabled the transgene expression in new neurons as they matured. Together, these results show that this approach enabled the specific targeting of hilar astrocytes in the mouse hippocampus. We next used this strategy to express the human tau isoforms in astrocytes. Since tau 0N is a fetal isoform and tau 2N is weakly expressed in the human brain 11 , we used the 1N (1N3R and 1N4R) isoforms linked to a V5 tag (LV-G1-1N3R or LV-G1-1N4R; Fig. 3g). We previously reported that the V5 tag does not interfere with tau protein hyperphosphorylation and misfolding 12 and it enables the detection of exogenous tau by western blot (Fig. 3h) and immunohistochemistry (Fig. 3i). Four months after the co-injection of LV-G1-GFP and either LV-G1-1N3R or LV-G1-1N4R in the dorsal hippocampus, approximately 500 hilar cells per hippocampus were infected (Fig. 3j). All transduced cells were found in the dentate gyrus, with a majority (65%) in the hilus. Similarly to the LV-G1-GFP reporter construct, virtually all transduced cells were identified as astrocytes (Fig. 3k), representing half of the astrocyte population in the dorsal hilus (Fig. 3l). Astrocytes that accumulated human 1N3R or 1N4R tau exhibited a pathological conformation of tau, as assessed by immunohistochemistry using the MC-1 antibody (Extended Data Fig. 6a).
Thus, this LV enables the long-term expression of the human tau isoforms in astrocytes of the dentate gyrus and in particular the hilus, with very high anatomical and cellular specificity.
Tau isoform overexpression in astrocytes differentially affects mitochondrial distribution and function. By enabling their coupling with the cytoskeleton, tau is known to play a role in organelle distribution and, in particular, mitochondria transport 13 . To assess the consequences of tau isoform accumulation on mitochondria, we used MitoTimer 14 . To examine the effect of 1N3R or 1N4R tau isoforms on mitochondria, we co-injected, into the mouse dentate gyrus, the following combinations of LV: LV-G1-MitoTimer + LV-G1-CFP + LV-G1-1N3R-V5 or LV-G1-MitoTimer + LV-G1-CFP + LV-G1-1N4R-V5, or as control, LV-G1-MitoTimer + LV-G1-CFP (or LV-G1-GFP). Four months after injection, most astrocytes infected with the control construct (LV-G1-CFP) exhibited a uniform distribution of mitochondria in the soma, proximal processes (between 1 and 20 µm from the soma) and distal processes (more than 20 µm from the soma), which was defined as a class 1-distribution pattern. A few astrocytes (17.0% ± 9% of all control astrocytes) were devoid of mitochondria in the distal processes (defined as a class 2 distribution pattern) or in distal and proximal processes (defined as a class 3 distribution pattern; 14.5% ± 2% of all control astrocytes; Fig. 4a,b). In contrast, 1N3R tau overexpression strongly reduced the number of mitochondria located in the distal processes, as evidenced by a significant increase in the proportion of class 3 and decrease in class 1 astrocytes ( Fig. 4b and Extended Data Fig. 6b-d). In addition, 1N4R induced a redistribution of mitochondria towards the soma, albeit less drastically than 1N3R (Fig. 4b). Mitochondrial relocation towards the soma may be due to a retraction of astrocytic processes. To test this possibility, we analyzed the effect of 1N3R or 1N4R tau on astrocytic morphology. We found that 1N3R and 1N4R tau, were homogeneously distributed throughout the soma and processes of astrocytes (Fig. 4c). Furthermore, using GFP to examine astrocyte morphology, we found that the projected area of individual astrocytic territories, the area of the soma, the number of branching points, the number of segments, the number of terminal points, the total length of processes and neuropil infiltration volume (NIV; Fig.  4d,k) were similar between control and 1N3R or 1N4R astrocytes. Together, these results indicate that 1N3R, and to a lesser extent tau inclusions in hilar astrocytes (S100β, green; tau 3R and 4R, red) in control donors and donors with AD. b, Confocal micrographs and orthogonal projections showing the presence of 3R tau inclusions in hilar astrocytes. c, Histogram of the density of 3R tau inclusions in hilar astrocytes of controls or individuals with AD, who were distributed between 3 categories, depending on the presence of phosphorylated tau, amyloid plaques or both. d, Histogram of the percentage of astrocytes that contained 3R tau inclusions in controls or individuals with AD. e, Confocal micrographs and orthogonal projections showing the presence of 4R tau inclusions in hilar astrocytes. f, Histogram of the density of 4R tau inclusions in the hilar astrocytes of controls or individuals with AD. g, Histogram of the percentage of hilar astrocytes that contained 4R tau inclusions in controls or individuals with AD. h, Photomicrographs showing PSD95 immunostaining in the hilus of controls and the three categories of individuals with AD. i, Histogram of the intensity of PSD95 staining in the hilus of controls or individuals with AD. j, Correlation between the intensity of PSD95 staining and the number of hilar astrocytes expressing 3R tau. k, Correlation between the intensity of PSD95 staining and the number of hilar astrocytes expressing 4R tau.

NATuRE NEuRoScIENcE
1N4R, induce a redistribution of mitochondria from the processes towards the soma without modifying the morphology of astrocytes.
Next, we used the 555 nm/488 nm fluorescence ratio to examine the turnover and redox state of individual mitochondria. Both 1N3R and 1N4R tau increased the 555 nm/488 nm fluorescence ratio of mitochondria (Fig. 4l,m), indicating a reduced turnover and increased oxidized state of mitochondria.
To further investigate the consequences of 1N3R and 1N4R tau isoform overexpression in astrocytes on mitochondrial dynamics and function, we used rat hip-pocampal neuron-glial co-cultures infected with the following combination of LV: LV-G1-MitoTimer + LV-G1-CFP as control, or LV-G1-MitoTimer + LV-G1-CFP + LV-G1-1N3R or LV-G1-MitoTimer + LV-G1-CFP + LV-G1-1N4R. Similarly to our in vivo observations, the LV targeted almost exclusively astrocytes, and most astrocytes were co-infected (Extended Data Fig. 7a-d). Likewise, we found that mitochondria in control conditions were uniformly distributed between proximal (between 1 and 20 µm from the soma) and distal processes (more than 20 µm from the soma) of astrocytes (Fig. 5a-c). In contrast, 1N3R tau induced a drastic

NATuRE NEuRoScIENcE
relocation of mitochondria in proximal processes, whereas 1N4R did not change the distribution of mitochondria in astrocytes (Fig.  5c). Here too, tau isoforms did not induce morphological changes of astrocytes (Extended Data Fig. 7e-k).
The effect of 1N3R tau on mitochondrial distribution may be due to an effect on motility and dynamics. To assess this possibil-ity, we used confocal live imaging to track mitochondrial movement and found that 1N3R but not 1N4R increased the proportion of stationary mitochondria ( Fig. 5d and Supplementary Videos 1 and 2). Furthermore, by observing the movement of mitochondria relative to the soma, we found that 1N3R but not 1N4R reduced the anterograde movement and increased the retrograde movement of

NATuRE NEuRoScIENcE
Total processes length (µm)

Articles
NATuRE NEuRoScIENcE mitochondria (Fig. 5e). Thus, 1N3R tau reduced the total motility of mitochondria and induced their transfer to the soma, resulting in a decrease in mitochondria in the distal segments of astrocytic processes. Movement enables the recycling of damaged mitochondria, their fusion and fission, as well as biogenesis, all of which sustain mitochondrial function 15 . We therefore expected the scarce, stationary, distal mitochondria to exhibit morphological and functional impairment. We first examined the morphology of individual mitochondria. As compared to control conditions, 1N3R but not 1N4R tau overexpression in astrocytes reduced mitochondrial projected area, especially in the distal processes ( Fig. 5f), suggesting that the mitochondria that remained in the distal processes may be impaired. Next, we examined the redox state/turnover of mitochondria using the using the MitoTimer reporter gene. Neither 1N3R nor 1N4R altered the turnover/redox state of mitochondria in the soma. However, both isoforms increased the redox state in the mitochondria that had remained in the proximal and distal processes (Fig. 5g,h). Thus, the in vitro experiments confirmed and extended the in vivo observations showing that 1N3R tau overexpression in astrocytes induced mitochondrial relocation in the soma, concomitant with alterations in mitochondrial morphology and function in the distal processes, whereas 1N4R had a modest impact, restricted to mitochondrial motility and redox state.
In astrocytes, mitochondrial dynamics and redox state/turnover have an impact on intracellular ATP production and the regulation of intracellular calcium concentration 16 . To investigate the consequences of 1N3R and 1N4R tau overexpression in astrocytes on mitochondrial ATP production, we used a Förster resonance energy transfer (FRET)-based sensor for ATP production. To this aim, we co-infected astrocyte-neuron co-cultures with an LV encoding either the cyan fluorescent protein (LV-G1-CFP as control) or LV-G1-1N3R or LV-G1-1N4R, together with a LV encoding a FRET-based fluorescent mitochondrial ATP probe 17 (LV-G1-MitoGoAteam2). In astrocytes overexpressing 1N3R tau but not in astrocytes overexpressing 1N4R tau, we found that ATP production by individual mitochondria was significantly reduced in distal processes as compared to control astrocytes (Fig. 5i,j). In contrast, we found no difference in ATP production in mitochondria of the soma or proximal processes (Fig. 5j). Finally, we investigated the calcium concentration in the soma and proximal and distal processes using the Fluo-4 AM calcium sensor. Compared to control, 1N3R tau overexpression had no impact on calcium concentration in the soma and proximal processes but significantly decreased calcium concentrations in distal processes, whereas 1N4R did not influence calcium concentrations (Fig. 5k,l). Overall, these results show that in vitro, 1N3R tau overexpression in astrocytes induced the relocation of mitochondria from the distal processes to the soma and proximal processes, which resulted in few and impaired distal mitochondria. In contrast, 1N4R showed mild effects on mitochondrial localization and did not impair mitochondrial function. We, therefore, next focused on the functional implications of 1N3R overexpression.

1N3R tau overexpression in hilar astrocytes impairs the hippocampal neuronal network.
To assess the long-term consequences of 1N3R tau accumulation in hilar astrocytes on neurons, we injected LV-G1-CFP (control) or LV-G1-1N3R + LV-G1-CFP into the hilus of adult mice. Four months later, we evaluated the density of different neuronal populations in the hilus. In both conditions, we observed a similar density of cells (DAPI + ; Fig. 6a) and of neurons (NeuN + ; Fig. 6b) in the hilus, suggesting that 1N3R tau overexpression in astrocytes did not induce cell death. In contrast, 1N3R tau overexpression in astrocytes significantly decreased the number of neurons expressing the activity-dependent protein parvalbumin (PV, Fig. 6c), whereas the density of mossy cells was not changed (GluR2/3 + ; Fig. 6d). This suggests that 1N3R tau overexpression in astrocytes affected PV expression in interneurons and, consequently, reduced the inhibitory transmission. The dentate gyrus is one of the two major sites for adult neurogenesis to occur, and a dysregulation of astrocytes or neuronal activity may interfere with this process 18 . We quantified cell proliferation by injecting animals with the thymidine analog 5-bromo-2′-deoxyuridine (BrdU) and analyzing animals 1 d after injection. We found that the number of cells that incorporated BrdU in the subgranular zone was unchanged in 1N3R tau-overexpressing animals ( Fig. 6e), suggesting that the few RGL cells that were targeted by the LV (Extended Data Fig. 5) were not sufficient to influence cell proliferation in the dentate gyrus. To assess the later stages of adult neurogenesis, we used immunostaining against the cytoskeletal marker of immature neurons, doublecortin (DCX). The number of DCX-expressing cells was significantly reduced in 1N3R tau-expressing animals (Fig.  6f), suggesting an impaired maturation of newborn neurons.
Next, we examined inhibitory synapses in the hilus. Inhibitory synapses are composed of nanoscale subsynaptic domains where the scaffolding protein gephyrin and the GABA vesicle transporter VGAT are closely associated (within 300 nm of each other 19 ). Using three-dimensional (3D) confocal reconstructions, we evaluated the density of VGAT and gephyrin clusters in the territories of 1N3R-overexpressing astrocytes (Fig. 6g,h). As compared to control mice, we observed a decrease in gephyrin dots, resulting in a reduction of paired VGAT-gephyrin punctae, which indicates a reduction of inhibitory synapses ( Fig. 6i-l). Thus, the overexpression of 1N3R tau in astrocytes impaired inhibitory neurons and adult neurogenesis.
Together, these results suggest that 1N3R tau overexpression in astrocytes may impair the function of the neuronal network in the hilus of the dentate gyrus. To assess the basal activity of the hilus, we

NATuRE NEuRoScIENcE
examined the expression of the immediate-early gene c-Fos. LV-G1-1N3R-injected animals showed fewer c-Fos + cells in the hilus than control mice, suggesting reduced neuronal activity in the hilus of these mice (Fig. 6m). PV-expressing interneurons are crucial for the generation of gamma oscillations, which enable coincidence detection and regulate circuit performance 20 . We therefore examined evoked gamma oscillations in the dentate gyrus of acute hippocampal slices using extracellular electrophysiological recordings.

NATuRE NEuRoScIENcE
Transient high-frequency oscillatory activity was induced by a brief focal application of glutamate in the hilus (Fig. 6n-t) 21 . We observed two distinct types of oscillatory activity: a gamma oscillatory activity (mean peak frequency ~75 Hz) which relies on functional GABA A receptor transmission (Fig. 6o-q) 21,22 and a faster oscillatory activity (mean peak frequency ~100 Hz), similar to the nonsynaptic network synchrony described by Towers et al. 22 (Fig. 6r-t). In slices obtained from mice injected with the LV-G1-1N3R vector, we found that the power of gamma oscillations (within the range of 50 to 90 Hz) was significantly reduced as compared to slices from control mice (Fig. 6q). Likewise, the peak frequency of these gamma oscillations was significantly lower in 1N3R tau mice (68.6 Hz ± 1.6 Hz; mean ± s.e.m.) than in control mice (75.1 Hz ± 1.3 Hz, P < 0.01; Fig. 6p). In contrast, we observed no significant difference in the power and the peak frequency of the faster oscillatory activity (Fig.  6s). Thus, the accumulation of human 1N3R tau in hilar astrocytes impaired synchronous activity.
Hilar astrocytic 1N3R tau accumulation affects spatial memory of adult mice. Gamma oscillations in the dentate gyrus play a role in spatial memory 23 and are impaired in mouse models of AD 24 .
We therefore compared the hippocampal-dependent spatial memory performances of mice 4 months after the bilateral injection of LV-G1-CFP as control or LV-G1-1N3R LV (Fig. 7a). We first used the new object location task and found that the preference index for the displaced object was significantly higher than chance level for the control but not for the LV-G1-1N3R group, indicating a reduction of spatial memory (Fig. 7b−d). On another set of mice, we performed the Morris water maze test. LV-G1-1N3R-injected mice showed similar performances to control mice in the learning phase and the probe test ( Fig. 7e-g). However, they showed a deficit in reversal memory (Fig. 7h,i), indicating a difficulty to suppress old spatial memory. To assess whether the decreased performances were specific to spatial memory, four independent behavioral tests, unrelated to spatial memory, were performed on the same sets of mice: dark/light box test, which assesses anxiety; Y-maze (YM) for spatial working memory; object recognition task for nonspatial long-term memory; and contextual fear conditioning, which assesses fear memory. For all these tests, LV-G1-1N3R-injected mice performed similarly to control mice (Extended Data Fig. 8).
Thus, the long-term overexpression of 1N3R tau in hilar astrocytes is sufficient to specifically alter spatial memory. PV interneurons play an important role in hippocampal function and spatial memory 25 . However, it is unclear whether the reduc-tion of PV interneurons observed after 1N3R tau overexpression in astrocytes contributes to the behavioral impairment observed in these mice. To test this possibility, we used the neuregulin 1 peptide (NRG1p) to increase PV interneuron excitability 26 . Another set of animals was injected with LV-G1-1N3R or LV-G1-GFP. Four months later, mice were injected with NRG1p or vehicle and, 1 h later, tested on the new object location test. Similarly to the cohort of mice shown in Fig. 7d, 1N3R tau-vehicle injected mice showed a lack of preference for the displaced object, as compared to control groups. The recognition of the displaced object was however restored in LV-G1-1N3R animals injected with NRG1p (Fig. 7j), suggesting that increasing PV interneuron activity restored the effect of 1N3R tau expression in astrocytes. To assess the involvement of PV interneurons in this effect, we examined PV immunoreactivity immediately after the behavioral test. LV-G1-1N3R-injected mice showed a reduced number of PV + cells and a reduced density of PV immunoreactivity in the hilus as compared to control groups, and the density of PV immunoreactivity was restored to control values upon NRG1p injection (Fig. 7k-n).
Thus, 1N3R tau expression in hilar astrocytes reduced long-term, spatial memory performances, which were restored upon stimulation of PV interneurons by NRG1p injection.

Discussion
In individuals with AD, we found that astrocytes of the hilus of the dentate gyrus accumulate 3R but not 4R tau, and this accumulation is correlated with synaptic alterations, suggesting an important role for astrocytes in disease progression. Using a new LV to specifically target astrocytes of the hilus, we found that overexpression of the human 1N3R isoform of tau in these cells strongly impaired mitochondrial motility, distribution and function, resulting in impaired neurogenesis, reduced number of neurons expressing PV, decreased density of inhibitory synapses and reduced gamma oscillatory activity. Together, these modifications led to impaired spatial memory, which was restored by stimulating PV interneuron activity.
Although tau has been found in glial cells 27 , astrocytes do not express this protein in physiological conditions 28 , and the origin of tau in astrocytes in AD is unclear. One unsubstantiated possibility is that AD progression induces tau translation from the mRNA present in astrocytes 29 . Alternatively, astrocytes may capture extracellular tau. Indeed, tau is released in the interstitial fluid by neurons 30 , spreads between cells 31 and astrocytes can uptake tau when exposed to this protein 32 . Furthermore, tau was found in extracellular vesicles from the cerebrospinal fluid of individuals with AD 33 , which  -test (a-f, i-l, q and t) and one-sided ANOVA with two-tailed Tukey's post hoc test (p and s). Scale bars: 20 µm (a-f and m), 5 µm (g) and 250 nm (h).

NATuRE NEuRoScIENcE
may also contribute to the intercellular propagation of this protein 34 . These possibilities are currently under intense scrutiny.
In tauopathies, much attention has been given to the role of tau in neurons 35 . However, in many tauopathies, tau is found in glial cells 36  progression. Disentangling the contribution of different cell types to a given phenotype is crucial for our understanding of disease etiology. However, access to this information is often hampered by the lack of specific tools that selectively target subpopulations of cells.
Similarly, the contribution of small brain regions to specific functions is difficult to assess without tools that selectively target them. In this study, we achieved the first goal by using a new LV strategy that enabled the expression of the genes of interest in astrocytes,  c, d, f, g and i). n = animals/cells per animal; LV-G1-CFP + saline: 6/144, LV-G1-1N3R + saline: 7/204, LV-G1-1N3R + saline: 4/68; (l and n). Significance was determined by mann-Whitney two-tailed t-test (c and d), Wilcoxon signed-rank test to chance level with ### P < 0.001, ## P < 0.05 and # P < 0.01 (d-j), one-sided ANOVA with two-tailed Tukey's post hoc test (j, l, n) and two-sided ANOVA with two-tailed Dunnett's post hoc test (f, g, i). Data are presented as the mean ± s.e.m. Scale bars: 25 µm (k) and 10 µm (m).

NATuRE NEuRoScIENcE
with negligible expression in non-astrocytic cell types, both in vitro and in vivo. Upon injection into the hilus, the limited diffusion of the LV further enabled the exclusive targeting of the dentate gyrus, since all transduced cells were found in the dentate gyrus and about 70% in the hilus. This targeting, both at the anatomical and at the cellular levels, enabled us to reproduce in mice, the observations we obtained from human participants.
Using this approach, we found that 1N3R and 1N4R tau overexpression in astrocytes differentially altered their mitochondrial localization, trafficking and function, as well as calcium buffering. These effects may be mediated by several mechanisms: first, tau competes with kinesin/dynein cargoes for microtubules, which are involved in mitochondria transport in astrocytes. Interestingly, tau affinity for microtubules differs between the 3R and 4R isoforms 37 , a difference that may underlie their differential effect on astrocytic mitochondria. Furthermore, according to models of multiple-motor-driven cargo transport, the higher solubility and kinesin inhibitory activity 38 of 3R tau compared to 4R tau may induce a strong steric inhibition of the binding strength of mitochondria to microtubules, leading to their immobilization. Alternatively, the effect of tau on mitochondrial transport may be due to posttranslational modifications, such as phosphorylation or truncation, which modulate tau functions. Indeed, tau truncation produces N-terminal fragments, which modulate kinesin velocity, and overexpression of which alters the mitochondrial system 39 . Furthermore, tau overexpression in astrocytes disrupts the intermediate filament network 40 , which may impair the transport of other cargo, including peroxisomes and endosomes 41 , disrupts the blood-brain barrier 42 , reduces the expression of glutamate transporters and reduces gliotransmitter release 43 . Alone or in combination, these effects are consistent with our observations of impaired mitochondrial transport and function. In turn, since astrocytes are involved in diverse brain functions 44 , the impairment of astrocytic function is expected to impact on the neuronal network and on behavior.
It is noteworthy that the impairment of a few hundred astrocytes in the hilus of the dentate gyrus altered hippocampal function and spatial memory. Although our results do not rule out that other brain areas may display astrocytic tau accumulation in the course of AD, the dentate gyrus is crucial for hippocampal function and memory performances 45 . In particular, our results point to two major effects of disrupted astrocytes on the function of the dentate gyrus: adult neurogenesis and PV interneuron function. Adult neurogenesis occurs in several steps, from stem/progenitor cell proliferation to the differentiation and maturation of new neurons, several of which are regulated by astrocytes 46 . Here, we found that 1N3R tau expression in astrocytes did not affect proliferation, but strongly reduced the number of immature neurons, similarly to recent observations in the human AD brain 47 . These results are consistent with the role of gliotransmitters in the maturation of adult-born hippocampal neurons 18 . Since immature granule neurons play a role in hippocampal-dependent memory 48 , their reduction may contribute to the memory impairment in LV-G1-1N3R-injected mice. PV interneurons also play a key role in hippocampal function. By exerting an important control over granule neurons, they fine-tune their activity and enable pattern separation. Furthermore, PV interneurons enable the generation of gamma oscillations, which support coincidence detection and regulate circuit performance 49 . The strong reduction of PV immunostaining in LV-G1-1N3R-injected mice suggests that PV interneurons likely underlie their memory impairments, a possibility that is further supported by the NRG1p-mediated behavioral rescue.
Astrocytes are key actors in brain physiology and play a fundamental role in the regulation of neural functioning 50 and adult neurogenesis 18 . Our observations suggest that these cells may play a greater role than expected in AD. Although the extent to which astrocytes are involved in the etiology of AD remains unclear, our results show that their impairment can contribute to memory disturbances and may dramatically worsen disease symptoms.

online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41593-020-00728-x.

Human participants. qaBrains were collected with individuals' informed consent and the authorization of the Ethics Commission of the Lausanne University
Hospital and CHU of Lille. A total of 30 brains were examined: 21 from individuals with sporadic AD without known familial history and 9 from control individuals who showed no sign of neurological disorder and who were age-and sex-matched to individuals with AD (Supplementary Table 1 and Extended Data Fig. 1a-c). All individuals with AD had been hospitalized in the Service of Old Age Psychiatry of the Lausanne University hospital or CHU of Lille and diagnosed according to the revised Diagnostic and Statistical Manual of Mental Disorders, Third Edition (DSM-IIIR) criteria. Clinical diagnosis and Braak stage were confirmed by postmortem neuropathological examination in the Department of Pathology of the Lausanne University Hospital for each case, following previously described protocols [51][52][53] . Tau phosphorylation was assessed using an antibody specific for the AT8 epitope in the anterior hippocampus, in the prefrontal, parietal and temporal associative isocortex and in the primary visual cortex. The brains were removed with a postmortem delay of no more than 60 h and stored in buffered 10% formaldehyde until sampling.
Quantification of human samples. Quantification of AD2 + cells, Aβ plaques, PSD95 and synaptophysin optical density was performed on a minimum of four sections (separated by at least 160 µm) per individual, using a digital camera (3CCD Hitachi HV-F202SCL) mounted on a slide scanner microscope (×20 objective; Zeiss axioscan Z1). AD2 inclusions were analyzed only when their surface was between 150 µm 2 and 2,000 µm 2 and Aβ plaques between 350 µm 2 and 5000 µm 2 . To determine the presence of 3R and 4R tau inclusion in s100β-expressing astrocytes, we analyzed 15-20 stacks/individuals with AD, 40-60 S100β + astrocytes/individual and tau isoform, for a total of 1,840 astrocytes for the presence of 3R tau and 1,537 astrocytes for the presence of 4R tau.
The volume of S100β + , 3R and 4R tau inclusions was determined using autoregressive algorithms of the Imaris surface plugin 55 . Individual astrocytes were considered to contain 3R or 4R tau inclusions when the volume of these inclusions represented at least 5% of the volume of the soma.
Primary rat hippocampal cultures. For co-cultures, timed-pregnant rats (RjHan:WI, Janvier) were killed by CO 2 inhalation, and E17 embryos were collected in Petri dishes containing Hank's balance salt solution (HBSS; Gibco, Life Technologies) to enable hippocampal dissection. Cells were prepared following published protocols 66 . The cells were plated at a density of 3 × 10 5 cells per cm 2 in neuronal medium in multiwell dishes. For astrocyte cultures, P1 rat pups were used (Charles River), and cells were plated at a density of 20,000 cells per cm 2 in multiwell dishes. Hippocampal co-cultures and primary hippocampal astrocytes were infected at DIV4 and DIV8 respectively, with 0.6 pg p24 antigen per cell of LV-G1-CFP, LV-G1-GFP, LV-G1-1N3R, LV-G1-MitoTimer or LV-G1-MitoGoAteam2, corresponding to 1.2 pg of p24 antigen for a double infection and 1.8 ng for a triple infection.
Recombinant tau proteins were produced as described previously 67 . The membrane was incubated 1 h at room temperature (RT) in a blocking solution containing 5% milk-TNT (15 mM Tris, 140 mM NaCl and 0.05% Tween 20) and transferred to the antibody solution diluted in 5% milk-TNT (mouse monoclonal antibody against 3R tau (RD3, Millipore; 1/2,000) or mouse monoclonal antibody against 4R tau (RD4, Millipore; 1/1,000)) for incubation overnight at 4 °C. Tau proteins were revealed by enhanced chemiluminescence (ECL; GE Healthcare) using horseradish peroxidase (HRP)-conjugated anti-mouse secondary antibody (Vector, 1/50,000). A negative control without the primary antibodies was used to exclude nonspecific signal.
Protein extraction and immunoblotting. Mouse hippocampus and cortex were dissected and suspended in PBS to a final concentration of 1 µg µl −1 . Next, 10 µg of sample was loaded onto a 4-12% Bis-Tris (Criterion gel, Bio-Rad), followed by transfer onto a 0.45-µm nitrocellulose membrane. After three rinses, membranes were incubated with a blocking solution for 30 min at RT before incubation with rabbit polyclonal anti-V5 (Millipore, AB3792; 1:10,000) overnight at 4 °C. The membrane was then incubated with the secondary anti-rabbit HRP antibody for 45 min at RT (Vector; 1:5,000). Signal was visualized using ECL western blotting detection reagents (GE Healthcare) in an Amersham Imager 600.

Animals and stereotaxic delivery of LV.
Three-month-old male C57BL/6 mice were purchased from Janvier. All animals were housed in a temperature-controlled room (22 °C ± 1 °C) and maintained on a normal 12-h light/dark cycle with access to food and water ad libitum. Mice were separated into cages of four mice per cage. All procedures were performed in accordance with the overly strict Swiss legislation on the care and use of laboratory animals. Mice were anesthetized by intraperitoneal injection of a mixture of 100 mg kg −1 ketamine (Ketasol, Graeub) and 10 mg kg −1 xylazine (Rompun, Bayer Health Care). The animals received 2 μl of LV bilaterally injected into the dorsal dentate gyrus at the coordinates: ±1.5 mm lateral to the midline, −2 mm posterior to bregma and −2.4 mm ventral to the dura relative to bregma. The LV were injected at 0.2 μl min −1 and the needle was left in place for 5 min. Animals received injections of the analgesic buprenorphine at a dose of 0.1 ml per 100 g after injection.
Quantification of cell populations in the mouse hippocampus. Quantification of DAPI + , GFP + , PV + , GluR2/3 + , NeuN + , V5 + , DCX + and BrdU + cells per brain was performed from 1-in-6 sections spaced 300 µm apart using a digital camera (3CCD Hitachi HV-F202SCL) on a slide scanner microscope (×20 objective, Zeiss axioscan Z1). In hippocampal section containing infected cells, immunopositive cells located in the region of interest (ML, GCL, subgranular layer, CA1, CA3, cortex and hilus) were counted using image analysis software Zen 2 (black 8.0 Articles NATuRE NEuRoScIENcE edition and blue 2012 edition). Cell density was calculated by dividing the total number of cells for each acquisition by the surface of each area of interest. For PV labeling intensities, 100-150 cells were quantified in arbitrary units as the mean of all isolated pixels of soma. Each optical density was normalized via the subtraction of a slide section in which signal was absent (black). Normalization and recalibration across different experiments was achieved by using internal control animals. Animals for the quantification of c-Fos + cells were perfused 90 min after memory test (object location task).
Quantification and determination of cell phenotype in vitro and in vivo. Cell phenotypes were determined from 350 GFP + cells per mouse or 50 per neuronglial culture. Colocalization with GFAP, S100b, NeuN or IBA1 was assessed by confocal microscopy (×40 oil immersion objective, Zeiss LSM 710 Quasar) over the entire z axis. Labeled cells were rotated in orthogonal planes (x and y) to verify double labeling. All analyses were performed in sequential scanning mode to prevent crossover between channels. The estimated fraction of GFP + cells co-labeled with NeuN or GFAP was calculated for each animal. Absolute numbers of GFP + GFAP + , GFP + S100b + and GFP + NeuN + cells were obtained by multiplying the corresponding estimated fraction of co-labeled GFP + cells by the total number of GFP + cells for each animal. GFP + cells located in dentate gyrus were classified as astrocytes, RGL and neurons based on morphology. Astrocytes were characterized by a large spheroid or pyramidal soma with ramified processes. Neurons displayed an oval-shaped soma with an apical dendritic tree extending through the GCL and reaching the ML. RGL cells displayed a prototypical morphology, including a nucleus located in the subgranular zone of the dentate gyrus, a radial process extending through the GCL and extensively branching into the outer GCL and the ML, and a few basal processes extending towards the hilus 69,70 .

Morphological analyses of GFP + astrocytes in vitro and in vivo.
Approximately 15-20 hilar GFP + astrocytes per mouse (or 50 GFP + astrocytes in vitro) were imaged with a Zeiss LSM 880 Quasar confocal system (63× + 2× numerical zoom) equipped with Airyscan. Care was taken to only image astrocytes with a soma entirely contained within the thickness of the section. Images consisted of 50-75 optical sections (z = 0.3 µm). Three-dimensional reconstructions of a series of confocal images were deconvolved (Huygens SVI) and analyzed using Imaris XT (Bitplane AG) and the 'autopath' algorithm of the 'filament' plugin. Soma volume was calculated using the Imaris surface plugin 55 and was manually corrected to exclude the main processes. The NIVs were calculated as previously described 71 . Briefly, for every astrocyte analyzed (20-25 cells per group), three randomly chosen regions of interest of 15 µm × 15 µm × 10 µm, devoid of soma and large branches were imaged. Astrocytic processes were 3D reconstructed in the hilus using Imaris software and their volume was measured.
Quantification of gephyrin and VGAT punctae. Images were acquired in the vicinity of distal processes of GFP + astrocytes that were entirely contained within the section thickness. Typically, 40-50 images per group (around 10 per animals) were acquired (series of 50-75 multiple optical sections, z = 0.2 µm) with a Zeiss LSM 880 Quasar confocal system (63× + 4× numerical zoom) equipped with Airyscan. Images were imported into Imaris XT (Bitplane AG) and corrected for background. VGAT or gephyrin dots were determined using autoregressive algorithms of Spot plugin. The density of VGAT and gephyrin dots was calculated by dividing the total number of dots for each acquisition by the volume of interest. For the VGAT/gephyrin pairing analysis, we used colocalized Spot MATLAB script with 0.3 µm for the closest distance between spots. The VGAT/gephyrin ratio was then calculated.
MitoTimer mitochondrial analyses in the mouse hippocampus. The hippocampal astrocytic mitochondrial system was assessed four months after co-infection with LV-G1-CFP + LV-G1-MitoTimer or LV-G1-CFP + LV-G1-1N3R + LV-G1-MitoTimer. Multiple optical sections (z = 0.3 µm) of confocal images were acquired throughout the section of cells located in the polymorphic and subgranular layers of the dentate gyrus, with a Zeiss LSM 780 Quasar confocal system (63× + 4× numerical zoom). Fluorescence images were captured using similar mirror/filter, excitation and detection parameters as those used for the in vitro experiments. For localizing processes within mitochondria, acquisitions were imported into Imaris XT (Bitplane AG). The green (500-540 nm) and red (580-640 nm) channels from MitoTimer were merged using the Imaris channel arithmetic MATLAB plugin to visualize whole mitochondria. Mitochondrial volume reconstructions were performed using the Imaris surface plugin. To assess the redox state of the mitochondria, the mean intensities of the red and green channels were automatically calculated using the Imaris statistics plugin and normalized against those of the control condition (LV-G1-CFP).
MitoTimer mitochondrial system analyses in vitro. The astrocytic mitochondrial system was assessed 10 d after viral infection. Cells were imaged by acquisition of multiple optical sections with a Zeiss LSM 710 Quasar confocal system.

Morphology and localization.
Confocal images were imported into Imaris XT (Bitplane AG). Green (500-540 nm) and red (580-640 nm) channels from MitoTimer were merged using the Imaris channel arithmetic MATLAB plugin to visualize whole mitochondria in CFP + cells and mitochondrial volume reconstruction was performed using the Imaris surface plugin 55 . Using the CFP channel to visualize the entire cell, a 20-µm diameter circle was drawn around the center of the soma of astrocytes. Mitochondria found within this circle were considered as proximal and mitochondria found further away were considered as distal. Mitochondrial distribution and morphology (length, width and surface) were automatically counted for each compartment and the frequency distribution calculated in 'proximal' and 'distal' processes for each cell.
Motility. For mitochondrial motility, single mitochondria were manually followed for each time point from live imaging acquisition based on merged green and red channels. The mean velocity and total track length were then automatically calculated by Imaris. The total track length traveled (Δ) was used to classify mitochondria as highly mobile (Δ ≥ 20 µm), mobile (6 µm ≤ Δ ≤ 20 µm) or stationary (6 µm ≤ Δ). The direction of each mitochondria was determined by visualizing the displacement vector and was defined as anterograde when the mitochondria moved towards the periphery of the astrocyte and retrograde when it moved towards the soma.
MitoTimer redox state ratio. The red and green mean intensities were automatically calculated in the soma and proximal and distal processes based on green and red merged channel volume reconstruction of MitoTimer using the Imaris statistical plugin. The fluorescence ratio (R 555/488 ) was normalized against of the fluorescence the control condition (LV-G1-CFP) for each culture.
FRET mitochondrial ATP imaging. The ATP mitochondrial system was assessed 10 d after viral infection. Astrocytes were recorded every 5 s for 5 min using a Zeiss LSM 710 Quasar confocal microscope with a ×100 objective and digital zoom set to 4. The excitation wavelength was 350 nm for CFP and 488 nm for GFP, with detection of blue (410-480 nm), GFP (493-545 nm) and OFP (580-640 nm). The OFP/GFP ratio was calculated by dividing the OFP mean intensity by the GFP intensity for ten regions of interest in the soma and proximal and distal CFP + processes of astrocytes.
Calcium imaging. The intracellular calcium concentration was assessed 11 d after viral infection. Cultures were loaded with 5 µM Fluo-4 AM (Invitrogen) for 15 min in the presence of 0.02% Pluronic F-127 (Invitrogen) at 37 °C and 5% CO 2 in the dark in a HEPES-KRH buffer at pH 7.4 and de-esterified for 10-15 min before imaging. The mean fluorescence for Fluo-4 AM was normalized over CFP fluorescence and calculated for ten regions of interest in the soma and proximal and distal CFP + processes.
Recordings. Field potentials were recorded with ACSF-filled glass electrodes (~5 MOhm). Signals were amplified (differential amplification mode, gain 1000×, NPI Ext-2 amplifiers), bandpass-filtered at 1-1,000 Hz and digitized at 2 KHz and acquired with Spike2 software (CED). In each hippocampal slice (six per animal), electrodes were positioned at four different sites in the GCL-ML interface of the dentate gyrus (Fig. 6n). Transient neuronal activity and high-frequency oscillations were evoked in the dentate gyrus by a short pressure injection (200 ms, ~7-12 bars) of glutamate (10 mM in ACF) via a glass pipette (tip diameter: ~8-12 µm) placed into the hilus 21 . These experiments were performed blindly without knowledge of the type of LV injected. After recording, slices were fixed with 4% paraformaldehyde for later verification by a second experimenter of the presence of GFP-labeled astrocytes in the dentate gyrus of the recorded slices. Only slices displaying proper LV injection and expression in the dentate gyrus were used for further analyses (for LV-G1-1N3R: 5 mice and 22 hippocampal slices; for LV-G1-GFP: 5 mice and 19 hippocampal slices).
Analysis. Time-frequency decompositions were performed using the MATLAB toolbox EEGLAB (version 14.1.1) on recording segments comprising the 2 s before and the first 12 s following glutamate stimulation. Time-frequency decompositions (from 10 Hz to 160 Hz, 1 Hz steps) were computed using Morlet waveform transforms (f0/σf of (3 0.5), 3-cycle wavelet with a slow linear increase (coefficient 0.5)). Data were assessed for the normality using the Kolmogorov-Smirnov and Shapiro-Wilk tests, and for the homogeneity of variance using the Levene test. When the data did not meet the criteria of normality, they were first log transformed before the statistical analyses. For each response type, the total power and the frequency at which the power was maximal were compared between groups. Comparisons between groups were performed using univariate ANOVAs. The Welch t-test was used, however, when the homogeneity of variance was not met. The initial voltage deflection following local glutamate application and reflecting the overall induced activity was used as covariate. Statistical analysis was performed with SPSS, and significance was set to P = 0.05.
Behavioral procedures. Four months after viral injection, tests were performed in the following order: group of animals were tested for light-dark box (day 121), Y-maze (day 123), open field (day 127) and object recognition or location (day 128) or for Morris water maze (days 120-128) and contextual fear conditioning (days 130-131). For NRG1p injections, another group of mice received one intraperitoneal injection of NRG1p (0.1 μg kg −1 in 0.9% saline; Prospec) or saline solution 1 h before the object location test and were euthanized 90-100 min after the test.
Dark/light box. The dark/light box consisted of two compartments made with acrylic transparent glass, placed in the open-field arena, a black/dark compartment (40 cm × 20 cm × 15 cm; 2 lux) and a white/illuminated compartment (same dimensions, 350 lux). Both compartments were connected by an aperture. Each mouse was released in the same corner of the illuminated compartment, and the number of exits and total time in the lit compartment were recorded for 6 min.
Y-maze test. This test was performed as previously described 72 . The symmetrical Y-maze, made of acrylic glass, consisted of three arms, each 40-cm long, 15-cm high and 5-cm wide. Each mouse was placed in the center of the Y-maze and was free to explore the arena for 6 min. After each session, the maze was thoroughly cleaned using ethanol and water and dried. The number of entries was recorded for each mouse while observing the mouse via a camera; one entry was defined as both hind paws of the animal being completely inside the arm. The measure for working memory was the percentage of alternations, that is, the number of triads divided by the maximum possible alternations (the total number of entries minus 2) × 100 (ref 73 ).
Object location test. This task is based on the spontaneous tendency of rodents previously exposed to two identical objects to preferentially explore the object that has been placed in a new location, rather than the non-displaced object 74 . The day before the exploration phase, each mouse was placed in an open-field arena (35 cm × 34 cm × 40-cm high wall with a spatial pattern inside) for habituation and allowed to explore the arena for 10 min. The total distance traveled in the open field was measured by video tracking (Noldus EthoVision), to assess general motricity and activity. The next day, two identical objects were placed in the middle of the open-field arena, and mice were allowed to explore them for 10 min. The time exploring the two objects was scored. Spatial memory was tested 24 h later when one of the objects (left or right counterbalanced) was moved to a new position. Mice were allowed to explore for 10 min. The time exploring the displaced object was calculated as the percentage of the total time exploring both objects.
Object recognition test. This task is based on the spontaneous preference of rodents for novelty and their ability to remember previously encountered objects 75,76 . The procedure, equipment and analyses were similar to those described for the new object location test, but the pattern inside the arena was removed. One day after habituation, two identical objects were placed in the middle of the open field, and the time the animal spent exploring each object was recorded. We ensured that every mouse spent the same amount of time exploring the objects and avoided any bias due to differences in individual levels of exploration by removing the animal once it had explored the objects for a cumulative total of 30 s. Animals that did not achieve this criterion within 10 min were excluded (two animals). Recognition memory was tested 24 h after the exploration phase. Mice were reintroduced into the arena and exposed to two objects, a familiar object and a new object, for which the positions of the two objects were identical to those of session 1. The familiar object was a triplicate copy of the sample used in session 1, to avoid olfactory trails.
The mouse was allowed to explore for 10 min, and the time spent exploring each object was recorded. The nature and position (left or right) of the new object was randomized. The open field was cleaned thoroughly between the introductions of each mouse to eliminate olfactory cues. Memory for the familiar object was evaluated by calculating the preference index for the new object, expressed as the percentage of time spent exploring the new object per total time spent exploring both objects. Independent groups of mice were used for the object location and object recognition tasks, so that each animal was submitted to either one or the other task.
Morris water maze. During the training phase, mice were placed in a pool filled with opaque water set at 25 °C. Training consisted of 6 d of four trials per day. The platform was always hidden in the southeast quadrant of the pool, but the mice were released at various points around the swimming pool, with the point of release being counterbalanced every day. The intertrial interval was between 10 and 20 min. Swimming tracks were recorded using video hardware and EthoVision software (Noldus). Spatial memory was then assessed during a probe trial 24 h after the final training (day 7). After establishing robust spatial preference for the platform location, either reversal trials were performed during which the platform was placed in a different location. At day 8, mice started a new, 3-d training session (two trials per day), where the platform was located in a new position (northwest) to start reversal learning. Swim paths were recorded and analyzed by a tracking system (EthoVision, Noldus). The assessed variables were escape latencies and, for the probe trials, time spent in target quadrants and platform place proximity indices. During the probe tests, time spent in each quadrant and the numbers of crossings over the location of the platform (virtual circle) compared to the mean crossing number of the three other virtual circles in the other three quadrants was calculated.
Contextual fear conditioning. Mice were assessed for fear memory accuracy in a fear-conditioning paradigm. For training, each mouse was introduced in a conditioning chamber (FCS-NG 46000, Ugo Basile) measuring 19 × 10 × 30 cm, with a metal wire floor and transparent plastic wall, set in a white soundproof cubicle (context A). The floor of the chamber under the grid was lined with tissue paper, which was changed between mice. After 3 min, mice received a single, 2-s foot shock (0.5 mA) and were removed from the chamber 15 s later. The conditioned freezing response to context A was assessed 24 h later, upon 3-min exposure to exactly the same chamber. The freezing time, defined as the absence of all movements with the exception of those related to respiration, was recorded by overhead cameras and measured using automated scoring systems (AnyMaze).
Sample sizes, calculations and statistical analysis. Sample sizes are indicated in the legend of the corresponding figures. Human sample size was not predicted. We have used a collection of human samples composed of 9 healthy control individuals and 21 individuals with AD. For cellular and behavioral assays, the sample size was chosen to account for statistical variability of cultures (more than three cultures) and surgical and behavioral procedures (more than eight animals), based on previous studies 69,77 . Human samples were classified on the basis of neurological and neuropathological examination, in particular on the presence of tau and Aβ in the hilus. The order of culture and mouse used for infection, injection and behavioral procedures was randomized for each experiment. Investigators were blinded to group allocation when processing the tissue, performing cell counts and during confocal image acquisition and behavioral tests. The only reasons for exclusion were problems encountered during culture (such as culture contamination) or failure of the injection procedure (no fluorescence observed in the hippocampus). Values are presented as the mean ± s.e.m.; N corresponds to the number of independent experiments and n to the overall number of values. Statistical analyses were performed on raw data with GraphPad Prism software v8.0. The normality of the data was verified using a Shapiro test. Data containing two experimental groups were analyzed using Student's t-test (parametric observations), Mann-Whitney test (non-parametric observations), one-way and two-way ANOVA tests and Wilcoxon matched pairs test (non-parametric paired observations), followed by Tukey's post hoc analyses. Statistical analyses on data containing more than two experimental groups were performed using two-way ANOVA test, followed by Dunnett's post hoc analyses, to account for multiple comparisons.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support the findings of this study are available from the corresponding author upon request. The map sequence for LV construction and microscopy acquisition data have been deposited in Zenodo.org at https://doi. org/10.5281/zenodo.3953694. Source data are provided with this paper.

nature research | reporting summary
October 2018 Field-specific reporting Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.

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

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

Sample size
Sample size was not predicted. They were chosen based on previous literature. We have used a collection of human samples composed of 9 healthy control individuals and 21 Alzheimer´s disease patients. Pilot experiments provided an estimate of effect size, and indicated the appropriateness of sample sizes chosen. The other sample size was chosen to account for statistical variability of cultures (more than three cultures ) and surgical and behavioral procedures (more than eight animals), based on previous studies (