Astrocytic TDP-43 dysregulation impairs memory by modulating antiviral pathways and interferon-inducible chemokines

Transactivating response region DNA binding protein 43 (TDP-43) pathology is prevalent in dementia, but the cell type–specific effects of TDP-43 pathology are not clear, and therapeutic strategies to alleviate TDP-43–linked cognitive decline are lacking. We found that patients with Alzheimer’s disease or frontotemporal dementia have aberrant TDP-43 accumulation in hippocampal astrocytes. In mouse models, induction of widespread or hippocampus-targeted accumulation in astrocytic TDP-43 caused progressive memory loss and localized changes in antiviral gene expression. These changes were cell-autonomous and correlated with impaired astrocytic defense against infectious viruses. Among the changes, astrocytes had elevated levels of interferon-inducible chemokines, and neurons had elevated levels of the corresponding chemokine receptor CXCR3 in presynaptic terminals. CXCR3 stimulation altered presynaptic function and promoted neuronal hyperexcitability, akin to the effects of astrocytic TDP-43 dysregulation, and blockade of CXCR3 reduced this activity. Ablation of CXCR3 also prevented TDP-43–linked memory loss. Thus, astrocytic TDP-43 dysfunction contributes to cognitive impairment through aberrant chemokine-mediated astrocytic-neuronal interactions.

Here, we investigated whether astrocytic TDP-43 is altered in the human brain and how these alterations influence brain function.We found that TDP-43 accumulates in the cytoplasmic compartment of astrocytes in AD and FTD.In mice, induction of analogous TDP-43 alterations in astrocytes throughout the brain or specifically in the hippocampus caused progressive memory loss that correlated with aberrant changes in antiviral factors expressed by hippocampal astrocytes.In particular, cytoplasmic TDP-43 accumulation increased astrocytic levels of interferon (IFN)-inducible chemokines and presynaptic levels of the corresponding chemokine receptor CXCR3, which promoted neuronal hyperexcitability and memory loss.Thus, dementia-associated TDP-43 alterations in hippocampal astrocytes may contribute to cognitive decline through aberrant engagement of antiviral mechanisms and increases in astrocytic-neuronal chemokine signaling.throughout the brain, including the neocortex, hippocampus, thalamus, striatum, and spinal cord in double transgenic mice, but not in single transgenic or nontransgenic (NTG) controls (Fig. 2, B to D, and fig.S1).Expression of hTDP-43-ΔNLS was robust in astrocytes, which showed characteristic branching and intermingling with neurons, but was minimal or undetectable in neurons (NeuN-positive cells) and microglia/macrophages (Iba1-positive cells) (Fig. 2, B to D, and fig.S1).
Previous studies have reported that targeting hTDP-43-ΔNLS to neurons causes severe motor deficits and early mortality in mice (18) and that an ALS-linked mutant form of hTDP-43 (M337V) expressed in astrocytes is similarly detrimental (30).However, we found that targeting hTDP-43-ΔNLS to astrocytes did not affect lifespan or alter motor, exploratory, and anxiety-linked behaviors (fig.S2).The mice were monitored up to the age of 23 to 24 months and had normal nesting, burying, and social behaviors but had mildly increased grooming by 14 to 15 months of age and increased incidence of ulcerative dermatitis (fig.S2).These results highlight the cell type-specific and mutation-specific effects of TDP-43 on brain function and suggest that cytoplasmic TDP-43 accumulation in astrocytes, in contrast to neurons, is not sufficient to cause early mortality, motor deficits, or changes in other innate behaviors.
TDP-43 pathology frequently occurs in patients with AD or hippocampal sclerosis (5,(9)(10)(11)(12)43) and correlates with memory impairments (10,44,45), implicating TDP-43 alterations in memory loss.To determine whether astrocytic hTDP-43-ΔNLS impairs hippocampus-dependent learning and memory, we tested the mice in the Morris water maze.In this task, mice learn to locate a hidden platform using spatial cues (46).One and 3 days after the training was completed, mice underwent memory testing (probe trials) in which the platform was removed and each trained mouse was placed back into the maze for 60 s.Preferences for the target quadrant and the platform location used in training were assessed as compared to other quadrants and analogous nontarget locations in the maze.Mice with intact memory typically have significant preferences for the target quadrant and platform location, whereas mice with diminished memory have no significant preferences for the target quadrant and platform location as compared to other nontarget locations of the maze.
At 3 to 4 months of age, double transgenic ΔNLS mice had normal learning during training and normal memory in a probe test conducted 1 day after training (Fig. 2, E and F).However, the mice were moderately impaired in a probe test conducted 3 days after training (Fig. 2, G and H).By 9 to 10 months of age, double transgenic ΔNLS mice had severely impaired performance in probe tests conducted 1 day and 3 days after training (Fig. 2, J to N) but had no deficits in learning or swimming (Fig. 2I and fig.S3, A and  B).By 12 months of age, ΔNLS mice also had deficits in novel object recognition, an alternative memory-dependent test, but no changes in total exploration (Fig. 2O and fig.S3, C and D).Thus, astrocytic hTDP-43-ΔNLS causes progressive memory deficits but does not markedly impair locomotion or other behavioral functions.

Astrocytic TDP-43 alterations in the hippocampus are sufficient to cause memory loss
We next examined whether the memory deficits induced by astrocytic hTDP-43-ΔNLS could be caused by direct effects of astrocytic TDP-43 on the hippocampus, a brain region known to be crucial for memory and susceptible to aging and dementia-associated pathology.To selectively target hippocampal astrocytes in vivo, we stereotaxically microinjected adeno-associated viral (AAV) vectors encoding Cre-dependent hTDP-43-ΔNLS or hTDP-43-wild-type (WT) into the hippocampus of transgenic Aldh1l1-Cre mice, which express Cre recombinase predominantly in astrocytes (Fig. 3A) (47).Although the Aldh1l1 promoter might be active in other cell types in some contexts and niches of the brain (48), recent studies using single-cell transcriptomics and immunolabeling indicate that Aldh1l1 activity is very low or absent in hippocampal stem cells, including radial glia (49,50).Moreover, Aldh1l1 promoter-regulated transgene expression is not detected in neuronal progeny (50).Nonetheless, to further restrict transgene expression to astrocytes, we also placed the AAV vector-encoded hTDP-43-ΔNLS and hTDP-43-WT transgenes under the control of the astrocytic promoter hGfaABC1D (51).This approach created a two-promoter system that requires both Aldh1l1 and Gfap promoter activities for transgene induction.To ensure high efficiency of transgene expression, we used the PHP.eBAAV capsid, which has over 40-fold greater efficiency in transducing neural cells as compared to the standard AAV9 capsid (52).
After delivering the vectors intracranially, we performed immunostaining for hTDP-43 protein and markers of astrocytes and other neural cell types.We found that hippocampal transductions with PHP.eB AAV vectors encoding hTDP-43-ΔNLS or hTDP-43-WT were highly efficient, astrocyte-selective, hippocampus-specific, and stable for months after AAV microinjection (Fig. 3, B to D, and fig.S4, A to C). hTDP-43-ΔNLS was localized in astrocyte cell bodies and processes, whereas hTDP-43-WT was enriched in astrocytic nuclei.However, similar to hTDP-43-ΔNLS, hTDP-43-WT was also moderately extranuclear in vivo and in isolated astrocytes (fig.S4, B, D, and E), as described previously in other cell types (53,54), suggesting that both manipulations can cause accumulation of TDP-43 in the cytoplasm.
We next used the Morris water maze to assess learning and memory in AAV-injected transgenic Aldh1l1-Cre mice and AAVinjected NTG littermate controls.Three months after microinjection, Aldh1l1-Cre mice that received AAV vector encoding hTDP-43-ΔNLS had normal learning and probe performance (Fig. 3, E  and F).However, by 6 months after microinjection, these mice had impaired probe performance as compared to control groups (Fig. 3, G to I), similar to the results obtained in double transgenic hTDP-43-ΔNLS mice (Fig. 2).NTG mice that received the AAV vector encoding hTDP-43-ΔNLS but did not express Cre performed similarly to control Aldh1l1-Cre mice that did not receive the AAV vectors, ruling out nonspecific effects by the AAV vectors.Notably, Aldh1l1-Cre mice that received the AAV vector encoding hTDP-43-WT had normal learning but impaired probe performance by 9 months after microinjection (Fig. 3, K to O), consistent with previous studies showing that accumulation of WT TDP-43 is also detrimental (53)(54)(55).All groups had similar swim speeds (Fig. 3, J and P).
In addition, using our two-promoter system, we tested whether chronic astrocytic overexpression of other unrelated proteins would similarly cause progressive memory loss.To test this, we transduced hippocampal astrocytes with a PHP.eB AAV vector encoding hM4Di-mCherry (instead of hTDP-43) under the control of hGfaABC 1 D promoter.We did not detect memory deficits in Aldh1l1-Cre or NTG mice that received this vector (fig.S4, F to K), further ruling out potential nonspecific effects of AAV transduction and protein overexpression.Together, these results suggest that dysregulation of TDP-43 in hippocampal astrocytes is sufficient to cause progressive memory deficits and that TDP-43 plays essential roles in astrocytic modulation of hippocampal function.
Astrocytic TDP-43 affects IFN-inducible chemokines and other antiviral factors in a cell-autonomous manner Astrocytes contribute to memory loss in disease (39,56,57), but the exact mechanisms are not known.In addition, the effects of TDP-43 on astrocytes and astrocytic-neuronal interactions remain unclear.
In other cell types, TDP-43 dysfunction can alter inflammatory cascades (58,59), which might promote cognitive decline.Therefore, we next examined whether double transgenic ΔNLS mice had altered transcription of genes linked to neuroinflammation and glial reactivity in different brain regions that expressed hTDP-43-ΔNLS.Targeted transcriptional profiling was carried out in four different brain regions across four genotypes using a microfluidicbased high-throughput reverse transcription quantitative polymerase chain reaction (RT-qPCR) and a custom-designed panel of curated neuroinflammation-related genes.Tau-P301S mice were used as a technical positive control because this model has robust hippocampal gliosis and neuroimmune responses.In contrast to the broad changes in hippocampal gene expression detected in Tau-P301S mice (60, 61), we found highly selective changes in gene expression in double transgenic ΔNLS mice (Fig. 4A).IFN-inducible chemokines Cxcl9 and Cxcl10 were among the top genes affected and were highly increased in the hippocampus but showed minimal changes in other brain regions, including the neocortex, striatum, and thalamus (Fig. 4, A to C, and fig.S5, A to C). Various markers of astrocytic and microglial reactivity were minimally affected at the RNA and protein levels (Fig. 4A and fig.S5, A to C and F).Notably, the increases in CXCL9 and CXCL10 proteins were localized to hippocampal astrocytes, but not neurons, microglia/macrophages, or astrocytes in other examined brain regions (Fig. 4, D and E, and fig.S5, D and E), similar to previous findings in humans with AD (62).Increases in astrocytic CXCL9 and CXCL10 protein levels were also detected in transgenic Aldh1l1-Cre mice that received intrahippocampal injections of PHP.eB AAV vectors encoding hTDP-43-ΔNLS or hTDP-43-WT (fig.S5G), suggesting that astrocytic induction of these chemokines does not require widespread transgene expression and is not unique to the transgenic tet-off system.In addition to CXCL9 and CXCL10, the related chemokine CXCL11 also binds to CXCR3 with high affinity.However, in C57Bl/6 mice (which were used in this study), CXCL11 is not expressed at the protein level because of a premature stop codon (63,64).
Increased expression of CXCL9 to CXCL11 is typically induced by antiviral or IFN-associated signaling, which regulates glial functions and is linked to cognitive changes, dementia, and neuropsychiatric disorders (65)(66)(67)(68)(69)(70).One of the primary triggers for antiviral or IFN-associated signaling is activation of pattern recognition receptors (PRRs) that detect non-self or aberrant nucleic acids.Thus, we next tested whether sensors of nucleic acids were altered in transgenic mice and isolated astrocytes expressing hTDP-43-ΔNLS.Double transgenic mice expressing astrocytic hTDP-43-ΔNLS had increased levels of multiple sensors of aberrant double-stranded RNA (dsRNA), including RIG-I (retinoic acid-inducible gene 1; also known as Ddx58), MDA5 (melanoma differentiation-associated gene 5; also known as Ifih1), and PKR (dsRNA-dependent protein kinase; also known as Eif2ak2) (Fig. 4F).We also detected increased levels of other antiviral factors, including viperin (virus inhibitory protein, endoplasmic reticulum-associated, IFN-inducible; also known as cig5 or Rsad2) and IRF-7 (IFN regulatory factor-7), a master regulator of IFN responses (Fig. 4G).Similar to the chemokines, the increases in viperin protein levels were localized to astrocytes but not neurons (Fig. 4H and fig.S5H).In contrast to the dsRNA sensors, we did not detect changes in the genes encoding double-stranded DNA (dsDNA) sensor cyclic guanosine 5′monophosphate-adenosine 5′-monophosphate (AMP) synthase (cGAS) or stimulator of IFN genes (STING) (fig.S5I), suggesting that astrocytic TDP-43 alterations may preferentially affect dsRNA sensors and other alternative dsDNA-sensing mechanisms.
Next, we examined whether the changes in astrocytic gene expression were cell-autonomous and linked to functional changes in immune-related signaling.For these experiments, primary astrocytes were isolated from the hippocampus of double transgenic ΔNLS or NTG control mice generated in breeding cages with standard chow without DOX.Cell culture purity and transgene efficiency and specificity were evaluated as described in Materials and Methods.As expected, isolated ΔNLS but not NTG astrocytes expressed hTDP-43-ΔNLS protein (Fig. 5A).Total TDP-43 levels were less than twofold greater than endogenous TDP-43 levels (Fig. 5, A and B), and TDP-43 aggregates were not detected (see the Supplementary Materials).Nonetheless, isolated ΔNLS astrocytes had increased levels of multiple IFN-related gene transcripts, including dsRNA-related PRRs (Fig. 5C), Rsad2/viperin, Ifnb1, and Ifng (Fig. 5D), suggesting that these gene changes are at least in part cell-autonomous.Isolated ΔNLS astrocytes also had increased levels of XCL10 protein in the lysates and secreted into the medium (Fig. 5, E and F), as well as enhanced levels of IFN-γ in the lysates (Fig. 5G).In addition, isolated ΔNLS astrocytes had increased levels of phosphorylated nuclear factor κB (NF-κB) (Fig. 5, H and I), supporting the notion that TDP-43 dysfunction affects immune-related signaling pathways in astrocytes (71), similar to neurons and microglia (59,72).However, treating astrocytes with SN50, a cell-permeable peptide that inhibits NF-κB nuclear translocation (73), did not prevent the increases in Cxcl10 and Cxcl9 (Fig. 5, J and K).In contrast, treating astrocytes with Stattic, a potent inhibitor of signal transducer and activator of transcription 3 (STAT3) (74), was effective in blocking Cxcl10 and Cxcl9 gene expression changes (Fig. 5, J  and K).
To further test whether the changes in signaling and chemokine expression were induced by TDP-43-ΔNLS, some ΔNLS and NTG astrocytes isolated from the hippocampus were maintained in DOX-supplemented medium.As expected, DOX-treated ΔNLS astrocytes had minimal expression of hTDP-43, and total TDP-43 levels were reduced by approximately half as compared to ΔNLS astrocytes maintained without DOX (Fig. 5, L and M).DOX-treated ΔNLS astrocytes had reduced levels of Cxcl10 gene expression (Fig. 5N) and reduced levels of phosphorylated NF-κB and STAT3 (Fig. 5O) as compared to astrocytes maintained without DOX, suggesting that NF-κB and Janus kinase-STAT pathways are both engaged by TDP-43 alterations.These signaling pathways were not affected in NTG control astrocytes treated with DOX.Thus, TDP-43 alterations enhanced astrocytic antiviral gene expression and immune-related signaling and increased chemokine production.
Although the approximately twofold increases in total levels of TDP-43 induced moderate changes in gene expression and signaling, these changes were chronic and associated with marked functional effects on antiviral responses to pathogens.To examine these effects, we used a mimic of viral dsRNA, polyinosinic-polycytidylic acid [poly(I:C)], as a positive control for induction of antiviral responses.Transfection with poly(I:C) induces robust astrocytic antiviral responses at least in part via toll-like receptor 3 and MDA5 (75).As expected, isolated astrocytes acutely transfected with poly(I:C) had robust induction of antiviral genes and reduced levels of infection by vesicular stomatitis virus (VSV), a negativesense RNA virus (Fig. 6A and fig.S6A).In comparison to NTG controls, ΔNLS astrocytes had increased levels of VSV infection (Fig. 6A), suggesting that antiviral responses were impaired.ΔNLS astrocytes also had increased infection by adenovirus (Fig. 6B), a dsDNA virus.Cell culture density was similar between treatments and genotypes (fig.S6, B and C), indicating that the observed increases in viral infections were unlikely to result from altered cell survival or proliferation.We also tested whether astrocytes were more susceptible to herpes simplex virus-1 (HSV-1), a highly neurotropic dsDNA virus increasingly implicated in AD (76,77).ΔNLS astrocytes had increased levels of HSV-1 infection (Fig. 6, C to E).Thus, in addition to affecting cognitive function, alterations in astrocytic TDP-43 impaired antiviral defenses, which may predispose the cells to infectious pathogens.However, further studies are necessary to address how astrocytic proteinopathy affects host responses to microbial pathogens and possibly other neuroimmune challenges and whether alterations in type I and/or type II IFN pathways or other mechanisms, such as metabolic alterations (78), are contributing factors.

Astrocytic TDP-43 alterations are linked to increased presynaptic levels of the chemokine receptor CXCR3 and CXCR3-mediated neuronal impairments
Given that memory was impaired in ΔNLS mice, we next explored how changes in astrocytic IFN-inducible factors might affect neuronal functions.Upon release, IFN-inducible chemokines CXCL9 to CXCL11 activate the shared G protein-coupled receptor CXCR3 (79), which can be expressed by neurons (62), microglia (80), and potentially other cell types.Notably, patients with FTD or AD show increased levels of CXCL10 (81) and have hippocampal CXCR3 expression predominantly in neurons (62), suggesting that neuronal CXCR3 might play a role in these disorders.We found that hippocampal Cxcr3 RNA and protein levels were increased in ΔNLS mice (Fig. 7, A and B, and fig.S7A) and the majority of CXCR3-immunoreactive puncta were localized in neurons (Fig. 7, C and D).However, ΔNLS mice did not have increased levels of CXCR3 immunoreactivity within neuronal cell bodies (Fig. 7, C and D).Coimmunolabeling for CXCR3 and different presynaptic and postsynaptic markers revealed that ΔNLS mice had twofold increases in CXCR3 selectively within synaptophysin-positive puncta, but not within Postsynaptic density protein-95 (PSD-95) or gephyrinpositive puncta (Fig. 7, E to H, and fig.S7B), suggesting that the increases in hippocampal CXCR3 levels were localized to presynaptic terminals.
CXCR3 activation triggers G i/o -coupled signaling, which can inhibit presynaptic neurotransmitter release (82,83).However, the presynaptic effects of CXCR3 have not been previously defined.Thus, we next assessed whether activation of presynaptic CXCR3 affects neuronal activity using the multielectrode array (MEA) system.For these experiments, primary mouse neurons were transduced with PHP.eB AAV Syn1-Cxcr3-2HA-neurexin-1α.In this vector, the neurexin-1α sequence targets CXCR3 to presynaptic terminals, which simulates the presynaptic enrichment of CXCR3 observed in ΔNLS mice and limits artificial effects of CXCR3 on neuronal excitability (84).We confirmed that the vector was functional in neurons and that CXCR3 stimulation with chemokines induced characteristic intracellular signaling (fig.S7, C and D).We found that acute chemokine treatment inhibited spontaneous neuronal activity within minutes of treatment (Fig. 8A), suggesting that CXCR3 rapidly suppresses neuronal firing, possibly through modulation of calcium channels and presynaptic vesicles (82,83).
In contrast to the acute effects, chronic stimulation of CXCR3 increased spontaneous neuronal activity (Fig. 8B), indicating that CXCR3 also promotes long-lasting increases in neuronal firing, likely through chronic changes in the presynaptic compartment.Chronic stimulation of CXCR3 enhanced the levels of synaptophysin-positive puncta, a marker of presynaptic vesicles, but did not affect PSD-95, a marker of postsynaptic compartments (Fig. 8, C and D, and fig.S7E).Notably, neurons transduced with AAV-CXCR3, but not treated with chemokines, and neurons treated with chemokines, but not transduced with AAV-CXCR3, performed similarly to untransduced vehicle-treated control neurons, ruling out nonspecific effects of the AAV vector and treatment on neuronal responses.
To test whether similar presynaptic changes were present in the hippocampus of ΔNLS mice, we compared the levels of different synaptic markers by quantitative immunofluorescence.Similar to neuronal cultures overexpressing presynaptic CXCR3, double transgenic ΔNLS mice had increased levels of synaptophysin-positive puncta (Fig. 8, E and F) without marked changes in the levels of PSD-95 or gephyrin, which are markers of excitatory and inhibitory postsynaptic zones, respectively (fig.S7, F and G).There were also no detectable changes in the levels of bassoon, a scaffolding protein in excitatory and inhibitory presynaptic compartments, and no changes in synaptotagmin-2, a marker of inhibitory presynaptic vesicles (fig.S7, H and I) (85).Thus, astrocytic TDP-43 modulates excitatory presynaptic compartments without markedly changing the density of synapses.
We next investigated whether astrocytic TDP-43 altered functional electrophysiological readouts of hippocampal transmission and presynaptic release.For these experiments, acute hippocampal slices were obtained from double transgenic ΔNLS mice or littermate controls at 5 to 6 months of age.The Schaffer collateral pathway was stimulated while recordings of field excitatory postsynaptic potentials (fEPSPs) were obtained in the striatum radiatum of the dorsal CA1 region.We found that ΔNLS mice had increased basal synaptic transmission as evidenced by enhanced fEPSPs (Fig. 8G).This effect was most apparent at higher stimulus intensities.The mice also had impaired paired-pulse facilitation, as reflected by lower fEPSP2/fEPSP1 ratios (Fig. 8H).Notably, ΔNLS mice had increased responses to the first but not second pulses as compared to control mice (Fig. 8, I and J), indicative of enhanced probability of neurotransmitter release, which likely contributed to the observed increases in basal synaptic transmission.No significant differences were detected in fiber volley amplitudes, which reflect presynaptic action potentials, or in linear regressions of fEPSP slopes and fiber volleys (fig.S7, J and K).
To further test whether the presence of astrocytes with TDP-43 alterations is sufficient to affect neuronal activities and to confirm that this effect is not dependent on other cell types, we generated primary astrocytic-neuronal cocultures in which neurons were derived from NTG mice and astrocytes were derived from transgenic Aldh1l1-Cre mice to enable astrocyte-selective transgene expression.Before the addition of neurons to the cultures, isolated Aldh1l1-Cre astrocytes were transduced with a PHP.eB AAV vector encoding Cre-dependent hTDP-43-ΔNLS under the control of the hGfaABC 1 D promoter.Consistent with our findings in ΔNLS mice, NTG neurons cocultured with Aldh1l1-Cre astrocytes expressing hTDP-43-ΔNLS had altered spontaneous firing patterns as compared to neurons cocultured with control Aldh1l1-Cre astrocytes (Fig. 8K).Notably, spontaneous neuronal activities were suppressed by selective blockade of CXCR3 with SCH 546738 (12 nM) to approximately 40% of baseline activities, and this suppressive effect was significantly more pronounced in neurons cultured in the presence of hTDP-43-ΔNLS-expressing astrocytes (Fig. 8L), revealing an increased involvement of CXCR3.Together with our findings in isolated neurons and hippocampal slices, these results suggest that astrocytic TDP-43 is linked to CXCR3-dependent neuronal hyperactivity.Given these findings, astrocytic TDP-43 alterations may promote memory loss in part through CXCR3-induced effects on hippocampal function.Thus, we next tested whether genetic ablation of the gene encoding CXCR3 can prevent TDP-43-related memory loss.For these experiments, we generated Aldh1l1-Cre mice that were either WT or functional null (KO) for Cxcr3 and performed bilateral injections of the PHP.eBAAV vector encoding hTDP-43-ΔNLS into the hippocampal formation at 4 to 9 months of age (Fig. 9, A and B).Given that Cxcr3 is X-linked, we used littermate male mice that were either WT or KO for Cxcr3 and expressed Cre recombinase, which enabled cell-selective hTDP-43 expression in hippocampal astrocytes.The mice were then assessed in the Morris water maze at 7 to 12 months of age.
Similar to other results (Fig. 3), Aldh1l1-Cre/Cxcr3-WT mice that received AAV vector encoding hTDP-43-ΔNLS had impaired memory as compared to control mice that lacked hTDP-43-ΔNLS (Fig. 9, C to F), without changes in swim speeds (Fig. 9G).In contrast, Aldh1l1-Cre/Cxcr3-KO mice with hTDP-43-ΔNLS performed more similarly to control mice lacking hTDP-43-ΔNLS (Fig. 9, D to F), suggesting that alterations in astrocytic TDP-43 cause memory loss in a CXCR3-dependent manner.In addition, under control conditions, the behavior of Aldh1l1-Cre/Cxcr3-KO mice was not different from Aldh1l1-Cre/Cxcr3-WT mice, suggesting that CXCR3 impairs memory upon changes in astrocytic function.Together, our study reveals that astrocytic TDP-43 pathology is linked to maladaptive antiviral changes and increased chemokine signaling that disrupts hippocampal synaptic transmission and contributes to memory loss (Fig. 10).

DISCUSSION
Memory loss is common in aging-related neurological disorders, including AD, ALS-FTD, hippocampal sclerosis, and other conditions.However, the exact causes of memory loss are not clear and treatment options are limited.Increasing evidence implicates glial dysfunction and abnormal glial-neuronal interactions in various pathophysiological processes (35)(36)(37).In particular, astrocytes have been implicated in various central nervous system (CNS) disorders (86-89) and, similar to neurons, are functionally diverse (90,91) and affect information processing (57).Aberrant changes in astrocytes can contribute to behavioral and cognitive deficits and promote memory loss associated with aging and disease (39,57,92).However, the mechanisms by which astrocytes impair memory and other cognitive processes are not fully defined.
Here, we demonstrate that human hippocampal astrocytes accumulate extranuclear TDP-43 in AD and FTD.Astrocytic TDP-43 accumulation in the hippocampus was sufficient to impair memory, but not other neurocognitive functions, and it altered hippocampal neural activity and presynaptic function.Consistent with the selective impairments in memory, we detected marked increases in IFN-inducible chemokines preferentially in hippocampal astrocytes and increases in the corresponding chemokine receptor CXCR3 in hippocampal presynaptic terminals.These findings suggest that astrocytes in the hippocampus have a distinct response to TDP-43 alterations as compared to astrocytes in other brain regions.Widespread and chronic transgene expression in astrocytes did not cause motor impairments, early mortality, or other severe deficits.In contrast, animal models with analogous TDP-43 manipulations in neurons have severe ALS-associated phenotypes, including motor impairments and early mortality (18).These distinct and selective effects suggest that responses to TDP-43 accumulation are cell type-specific and heterogeneous across astrocytes in different brain regions.
Similar to TDP-43, ALS-linked mutations in superoxide dismutase (SOD1) cause neurotoxicity, motor impairments, and early mortality.Analogous to our findings, astrocyte-targeted expression of mutant SOD1 is not sufficient to trigger onset of motor neuron disease in mice (93).However, astrocytic SOD1 is required for disease progression, and astrocytes carrying mutant SOD1 are selectively damaging to isolated motor neurons but not other neuronal subtypes (94,95).Similarly, AD-linked tau accumulation in hippocampal astrocytes promotes selective neuronal deficits (96).Together, our findings and previous work indicate that various dementialinked protein alterations in astrocytes cause highly context-dependent effects on neurons and might facilitate selective neuronal vulnerability that contributes to variable disease manifestations in neurodegenerative conditions.
Most dementia cases involve dysregulation of WT rather than mutant TDP-43.Overexpression of WT TDP-43 in model systems is sufficient to impair cell function at least in part through the effects of TDP-43 outside the nucleus (54).We found that expression of either hTDP-43-WT or hTDP-43-ΔNLS, but not Fig. 10.Summary of the main findings.In AD and FTD, hippocampal astrocytes had aberrant accumulation of cytoplasmic TDP-43.These alterations were linked to brain region-specific and cell-autonomous changes in astrocytic antiviral pathways and increased production of IFN-inducible chemokines.The corresponding chemokine receptor CXCR3 was increased selectively in hippocampal excitatory presynaptic terminals and promoted neuronal hyperactivity and memory loss.Thus, dementiaassociated TDP-43 dysregulation in astrocytes causes chemokine-mediated changes in excitatory transmission leading to cognitive deficits.
control proteins targeted to hippocampal astrocytes, induced progressive memory deficits, suggesting that even modest alterations in WT TDP-43 within hippocampal astrocytes can impair memory and contribute to dementia-associated cognitive decline.Hippocampal astrocytes were not similarly vulnerable to control vectors, thus ruling out nonspecific effects of AAV injections or chronic protein overexpression as major drivers of the observed phenotype.Notably, microglia and other neural cells in the hippocampus might indirectly modulate astrocytic functions and the effects of TDP-43 alterations.Microglial progranulin and Triggering receptor expressed on myeloid cells 2 (TREM2) insufficiency may contribute to TDP-43 pathology (55,97).
Astrocytic TDP-43 alterations were accompanied by cell-autonomous changes in antiviral gene expression, increased phosphorylation of NF-κB and STAT3, increased levels of IFN-γ, release of chemokines, and functional changes in astrocytic innate defense against viral pathogens.Together, these results point to an aberrant TDP-43-linked antiviral phenotype (aTAP) that may affect cognitive function and neuroimmune responses.Our study focused on astrocytes, but TDP-43 is present in most cell types and increasingly linked to IFN-related pathways in different cell populations (98,99), suggesting that aTAP is not specific to astrocytes.TDP-43 accumulation in neurons triggers the antiviral cGAS-STING pathway, at least partly through abnormal mitochondrial DNA release (54).We did not detect changes in astrocytic cGAS or STING genes, possibly because aTAP engages distinct mechanisms in different cell types.Antiviral signaling involves multiple dynamic and cell-specific mechanisms (100).TDP-43 can also affect retrotransposon activity (101,102), which may also contribute to aTAP.Thus, TDP-43 likely influences multiple intracellular targets that affect neuroimmune signaling and antiviral cascades.
Although we focused primarily on the neurocognitive effects, our results in isolated cells implicate TDP-43 in modulating innate responses to viral pathogens.A link between dementia and viral infections has been suggested (103)(104)(105), but the effects of TDP-43 pathology on neural responses to infections are not known.We tested three different viral pathogens, used several independent methods to measure viral infections, and assessed different time points and viral doses.Convergent results across these different conditions suggest that alterations in TDP-43 allow pathogens to exploit weaknesses in astrocytic antiviral responses, which might affect innate immunity in the brain.
IFN-related pathways in astrocytes and other neural cells modulate brain function and have been implicated in AD, ALS-FTD, and other CNS disorders (66,69,70,(106)(107)(108)(109), but the roles of TDP-43 in these pathways have not been fully elucidated.We found that alterations in TDP-43 increased astrocytic IFN-inducible chemokines, among other genes, and promoted presynaptic increases in neuronal CXCR3, the shared receptor that is likely overactivated by the increased levels of chemokines.Although previous studies have reported CXCR3 expression in microglia (110) and infiltrating immune cells (111), we did not detect increases in CXCR3 in non-neuronal areas within the hippocampus.Similar to our results, CXCR3 has been detected in human neurons and neuronal processes (62).Acute activation of presynaptic CXCR3 suppressed neuronal activity, whereas chronic activation of CXCR3 increased neuronal activity.In a similar manner, neurons maintained in the presence of astrocytes with chronic TDP-43 accumulation had increased spontaneous neuronal activity, which was reduced by pharmacological inhibition of CXCR3.These results suggest that alterations in astrocytic TDP-43 promote CXCR3-dependent neuronal hyperactivity.Moreover, we found that alterations in astrocytic TDP-43 affected presynaptic function without changing the number of synaptic zones.Our findings are consistent with recent studies showing that G i/o -coupled receptors can enhance the number of presynaptic vesicles without altering the number of synapses (112).One mechanism promoting this effect may be presynaptic cyclic AMP (cAMP)/protein kinase A-dependent phosphorylation of synapsin-1, which results in the removal of vesicles from presynaptic clusters (112).Chronic activation of presynaptic G i/o -coupled CXCR3 may reduce synapsin-1 phosphorylation and allow additional vesicle recruitment to active zones, thus facilitating presynaptic neurotransmitter release.In support, ΔNLS mice had increases in hippocampal transmission and presynaptic release probability, which may promote an abnormal shift to low-pass filtering of excitatory firing and thereby impair the encoding of spatial memory within the hippocampus.Genetic ablation of CXCR3 alleviated memory deficits but did not affect memory in mice without astrocytic TDP-43 manipulation.Thus, alterations in astrocytic TDP-43 cause neural hyperexcitability and memory deficits that are dependent on CXCR3.
These findings implicate blockers of chemokines and CXCR3 as potential therapeutic interventions for TDP-43-associated cognitive impairments.Notably, human CXCL10 levels are increased in progranulin-linked FTD (81), AD (113), and amnestic mild cognitive impairment (113), either in the cerebrospinal fluid or in astrocytes (62), and these levels correlate with cognitive performance (113).Moreover, neural hyperexcitability has been reported in dementia (114), and global KO of CXCR3 prevents memory deficits in transgenic mice with amyloid-β pathology (115).CXCR3 blockers have reached clinical trials for peripheral inflammatory conditions (116) but have not yet been tested in patients with dementia or other cognitive disorders.
In summary, our findings suggest that TDP-43 alterations in astrocytes contribute to cognitive decline in dementia.We describe a previously unknown chemokine-mediated mechanism that is likely downstream of astrocytic TDP-43-linked antiviral changes that affects hippocampal presynaptic function and neuronal activities.Together, our results implicate astrocytic TDP-43 impairments and aTAP in dementia pathogenesis and point to chemokine signaling as a potential therapeutic target.

Mice
All animal experiments were conducted in accordance with guidelines set by the Institutional Animal Care and Use Committee of Weill Cornell Medicine.Mice were housed in groups of two to five mice per cage and maintained on a 12-hour light/dark cycle with ad libitum access to food and water.Experiments were conducted during the light cycle and included littermate controls.Regulatable and astrocyte-specific expression of hTDP-43 was achieved using transgenic hGFAP-tTA mice with a tet-off tTA element downstream of the human GFAP (hGFAP) promoter (provided by L. Mucke, Gladstone Institutes, San Francisco, CA) (39).hGFAP-tTA mice [B6.Cg-Tg(GFAP-tTA)110Pop/J; the Jackson Laboratory, strain #005964] were crossed with tetO-hTDP-43-ΔNLS mice [B6;C3-Tg(tetO-TARDBP*)4Vle/J; the Jackson Laboratory, strain #014650], which contained a tet operator (tetO) upstream of the human TARDBP gene with a mutated nuclear localization sequence (tetO-hTDP-43-ΔNLS) leading to expression of cytoplasmic hTDP-43 (18).In double transgenic mice (referred to as ΔNLS mice), DOX can bind to tTA to prevent tetO-mediated transgene expression.DOX-supplemented chow (200 mg/kg; VWR, 89067-462) was provided to breeding pairs and offspring until weaning [postnatal day 21 (P21)] to prevent expression of transgene during embryonic and postnatal development.Thereafter, mice were maintained on standard laboratory chow (Purina 5053) to allow the induction of transgene expression.Because the parent tetO-hTDP-43-ΔNLS strain was on a B6/C3 hybrid background, we backcrossed this line onto the C57Bl/6J background (the Jackson Laboratory, strain #000664) for five generations.
Mouse CXCR3 with a C-terminal hemagglutinin tag was targeted to neuronal presynaptic terminals using the neurexin-1α targeting sequence, as described (84).A gBlock gene fragment (Integrated DNA Technologies) encoding mCherry-T2A-CXCR3-2HA-neurexin-1α (axon targeting sequence) was designed following sequences made available by S. Sternson (Addgene, plasmid #52523; RRID #Addgene_52523), synthesized, and cloned into an AAV-hSyn1 expression vector (donated by B. Roth; Addgene, plasmid #50458; RRID #Addgene_50458) using Sal I and Eco RI restriction digest to generate pAAV-hSyn1:mCherry-T2A-CXCR3-2HA-neurexin-1α.NEB 5-alpha cells (New England Biolabs) were transformed with pAAV constructs, and the integrity of inverted terminal repeats and expression-related elements in selected clones were confirmed by sequencing and restriction digests.pAAV2/PHP.eBor pAAV2/DJ particles were produced by the Stanford University Neuroscience Gene Vector and Virus Core or the University of Pennsylvania Vector Core.PHP.eB capsid vectors were provided courtesy of V. Gradinaru and B. Deverman at the California Institute of Technology.PHP capsids are a modification of AAV9 provided by the University of Pennsylvania (52).

Behavioral testing
Experimental groups were distributed randomly across home cages and consisted of age-matched littermates of both sexes.Experimenters were blinded to genotypes and treatment conditions, and mice were tested in random order.Before most behavioral testing, except for the elevated plus maze, all mice were handled for approximately 2 min/day for 7 days.Mice that were injured or in poor health, independent of genotype, were excluded from behavioral testing.Tests were performed under white light, unless otherwise noted.For all test days, mice were acclimated to the testing room for 1 hours before testing.Sample sizes were chosen on the basis of our previously published work (39) and studies by other groups (42).

Elevated plus maze
The plus-shaped maze consisted of two enclosed arms and two open arms elevated 60 to 70 cm above the ground.Furthermore, tape was attached to the ends of the open arms (5 cm from the end of the arm) to limit falls.After 1 hour of habituation, mice were placed at the center of the maze facing an open arm.Mice could freely explore the four arms for 5 min.Time and distance traveled in each arm and center area were video recorded and tracked using EthoVision XT video tracking software (Noldus Information Technology Inc., Leesburg, VA).The apparatus was cleaned with 70% alcohol between mice.

Open field test
Mice were placed in the center of a clear plastic chamber (41 cm by 41 cm by 30 cm) with two 16 by 16 photobeam arrays detecting horizontal and vertical movements.To measure context-dependent habituation in the open field, the chambers were surrounded by distinct cues that were maintained across test days.Mice were acclimated to the chamber in 2 × 5-min trials with a 3-hour intertrial interval and assessed in the same chambers 1 and 14 days after habituation.Light in the room was set to 75% red light to limit the anxiolytic effect of 100% white light.Total exploration, rearing, and percent time spent in the center of the arena were measured with an automated Flex-Field/Open Field Photobeam Activity System (San Diego Instruments, San Diego, CA).The apparatus was cleaned with 70% alcohol between mice.Rotarod Mice were placed on the Rotarod (Rotamex-5, 0254-2002L) that was suspended 25.5 cm from a soft surface.The speed of rotation was either held constant at 12 rpm or increased from 4 to 40 rpm gradually at an acceleration rate of 0.3 rpm/s.Mice were tested on the rod in three trials with approximately 30-min intertrial intervals.The latency to fall off the rod was recorded and reported as the average of three trials.Equipment was cleaned with 70% ethanol between trials, and the light in the room was set to red light.

Morris water maze
The maze consisted of a 122-cm-diameter pool filled with water (20°± 2°C) made opaque with nontoxic white tempera paint (Colorations powder tempera paint).Spatial cues were set up around the pool before testing.All mice underwent one session of three to four pretraining trials in which they swam in a rectangular channel (15 cm by 122 cm) with a square platform (14 cm by 14 cm) hidden 0.5 cm below the water surface in the middle of the channel.If a mouse did not reach the platform within 10 s, then it was guided onto the platform by the experimenter and remained on the platform for 10 s before it was returned to its cage.One to 3 days following pretraining, mice underwent hidden platform training in the circular water maze.
For hidden platform training, the platform was submerged 1.5 cm below the surface.All mice underwent one session of four trials for three to five consecutive training days.For each trial, the platform location remained the same, but the mice were dropped in four different locations.The maximum time allowed per trial was 60 s.If a mouse did not find or mount the platform, then it was guided to the platform by the experimenter.All mice were allowed to sit on the platform for 10 s after each training trial.
Probe trials were performed 24 and 72 hours after the last hidden platform-training day.For probe trials, the platform was removed, and mice were allowed to swim for up to 60 s per trial.The drop location for the probe trials was 180°from the platform location used during hidden platform training.After 60 s, mice were guided to the platform location before removal from the pool and returned to its cage.
If the mice had any problems in learning where the platform is located, we performed a cued platform training 24 hours after probe testing.All mice underwent one session of four trials of the cued platform training.The cued (visible) platform training was performed using a different platform location and a clearly visible cue (a colorful 15-cm pole on top of the platform).All behavior was recorded and analyzed with an EthoVision XT video tracking system (Noldus).Escape latencies, distance traveled, swim paths, swim speeds, platform crossings, and proximity to the platform were recorded automatically for subsequent analysis.

Novel object recognition test
Mice were habituated to the testing chamber (40 cm by 40 cm) for 15 min.A day after habituation, mice were exposed to two identical objects in the same chamber and allowed to explore freely for 10 min once per day for two consecutive days.The next day, mice were presented with one object used during training and one unfamiliar (novel) object of a different shape and texture in the same chamber, and the mice were allowed to explore for 15 min during a test trial.The objects used for training and testing were assigned randomly to each mouse to avoid object bias, and which of the familiar objects was replaced with an unfamiliar object was varied randomly between mice to control for location bias.Chamber and objects were cleaned with Clidox-S (Pharmacal; 1:18:1 dilution) after each mouse.Behavior was recorded and analyzed with an EthoVision XT video tracking system (Noldus) the time that the mice spent next to each object were scored.

Social interaction test
Mice were allowed to freely explore a three-chamber arena (side chambers were 22.86 cm by 42.2 cm; middle chamber was 21.59 cm by 42.2 cm) with two empty inverted wire cups (8.5 cm in diameter) in each side chamber.Chambers were divided by clear plexiglass dividers, each with a half-circular opening at the bottom to serve as a free passageway between chambers.After 10 min of exploration by a test mouse, an unfamiliar mouse of the same sex as the test mouse was placed under one inverted wire cup and the other wire cup was left empty.The test mouse was allowed to freely explore the three-chamber arena for another 10 min to explore the unfamiliar partner.Each unfamiliar mouse used for testing was assigned randomly to each side to avoid location bias.Chambers and wire cups were cleaned with 70% ethanol after each mouse.Behavior was recorded and analyzed with the EthoVision XT video tracking system (Noldus).The total time that the test mouse spent next to each inverted wire cup with a mouse or empty wire cup was scored.The light in the room was set to red light.

Marble-burying test
Mice were placed in large cages (12 cm by 12 cm by 7.25 cm) covered with mouse bedding material to a depth of 5 cm.During each trial, 20 standard glass black marbles were gently placed on the surface of the bedding in a grid pattern.A mouse was placed in the center of the cage and allowed to explore for 30 min.A marble was scored as buried when it was at least three-fourth covered with bedding.The light in the room was set to red light.Between trials, the bedding was changed, and the cages and marbles were cleaned with 70% ethanol.

Nestlet-shredding test
Mice were single-housed with one cotton fiber nestlet (5 cm by 5 cm, 5 mm in thickness, ~2.5 g each) for 60 to 90 min.Nestlets were weighed before putting them in the cages.After returning the mouse to its home cage, whole nestlet pieces (not the shredded nestlet) were removed, dried overnight, and weighed.

Grooming test
Mice were sprayed three times with a light water mist on the back and neck areas.The mice were placed in large cages (12 cm by 12 cm by 7.25 cm) with no bedding.Behavior was recorded with the EthoVision XT video tracking system (Noldus) for 10 min.Cages were cleaned with 70% ethanol after each mouse.The amount of time spent grooming the head and the rest of the body was recorded and analyzed.The light in the room was set to red light.

Pole test
A vertical metal pole (1 cm in diameter and 50 cm in height) with a heavy metal base was placed in the center of a clean cage covered with bedding.Individual mice were placed on the upper portion of the pole facing the ceiling for three trials.The amount of time spent turning at the top of the pole and climbing down is video-recorded (EthoVision XT video tracking system, Noldus) and quantified.The apparatus was cleaned with 70% ethanol after each mouse.The light in the room was set to red light.

Wire hanging test
Individual mice were placed on top of a 2-mm-thick metal cloth hanger that is securely attached 40 cm above the home cage containing extra padding.For three consecutive days, the mouse was allowed to grasp the wire with the two forepaws for a maximum of 300 s for three trials.The latency of the mouse to full was recorded and quantified.The apparatus was cleaned with 70% ethanol after each mouse.The light in the room was set to red light.
WT and double transgenic astrocyte cultures were determined to contain 98.8 ± 0.4% (SE) of astrocytes and 1.2 ± 0.4% (SE) of microglia, based on immunolabeling for GFAP, Iba1, NeuN, and DAPI.These analyses were performed using a Nikon Eclipse Ti-S microscope with a Nikon PlanFluor 10× objective and NIS-Elements BR v5.02.01 acquisition software.
In double transgenic TDP-43-ΔNLS cultures without DOX treatment, hTDP-43 expression was detected in 47.8 ± 4.7% (SE) of astrocytes, based on immunolabeling for hTDP-43 and the astrocyte marker GFAP.For these analyses, a stringent lower cutoff was determined using mean signal intensity in parallel cultures with DOX treatment, which accounted for nonspecific immunoreactivity and background fluorescence.For analyses of hTDP-43 distribution in cultures, images were captured with an LSM 880 confocal microscope (Zeiss) equipped with a 63× objective (Zeiss) and Zen Black v2.3 SP1 FP3 acquisition software (Zeiss).Images were processed in Fiji v2.1.0/1.53c by subtracting the average background fluorescence and analyzing individual cells that had DAPI-positive nuclei and GFAP-positive cell bodies for levels of nuclear and extranuclear hTDP-43 immunoreactivity, respectively.

Primary astrocytic-neuronal cocultures
Cortical and/or hippocampal astrocytes at P1 to P3 were cultured as described above.Before seeding neurons, near-confluent monolayers (typically 5 to 8 days after plating) were briefly rinsed of serumcontaining medium.Neuronal suspensions were obtained from cortical and hippocampal tissue of P0 mouse pups as described above and seeded at 50,000 live cells per MEA on top of existing rinsed astrocyte monolayers in neuronal medium, as described above.

Slice preparation
Mice were deeply anesthetized with 5% isofluorane before being cardially perfused with ice-cold and oxygenated (95% O 2 /5% CO 2 ) sucrose cutting solution.Sucrose cutting solution contained 87 mM NaCl, 75 mM sucrose, 2.5 mM KCl, 1.25 mM NaH 2 PO 4 , 0.5 mM CaCl 2 , 25 mM NaHCO 3 , 1.3 mM ascorbic acid, and 10 mM D-glucose.The mice were quickly decapitated, and the brain was extracted in ice-cold sucrose cutting solution.Coronal slices (350 μm in thickness) were made on a vibrating blade microtome (Leica, VT1200s) while submerged in ice-cold and oxygenated sucrose cutting buffer.Slices were transferred to a heated (~35°C) incubation chamber containing artificial cerebral spinal fluid (ACSF), which consisted of 124 mM NaCl, 2.5 mM KCl, 1.5 mM MgSO 4 , 1.25 mM NaH 2 PO 4 , 2.5 mM CaCl 2 , and 26 mM NaHCO 3 .After approximately 30 min, the incubation chamber was allowed to equilibrate to room temperature for at least an additional 30 min.

Field potential recordings
For recordings, slices were transferred to a stage-mounted holding chamber on an upright BX51W1 microscope (Olympus).The chamber was superfused (2 to 3 ml/min) with oxygenated (95% O 2 /5% CO 2 ) and heated (~35°C) ACSF.Recordings were obtained using a Multiclamp 200B amplifier (Molecular Devices) and filtered at 2 kHz, digitized at 10 kHz, and acquired with Clampex 10.7 (Molecular Devices).Micropipettes were made with borosilicate glass pulled to a resistance of 3.5 to 5.5 megohm on a Flaming/Brown P-1000 micropipette puller (Sutter Instruments) and filled with the same ACSF.These were placed in the stratum radiatum of CA1, and a concentric bipolar stimulating electrode was placed within the same layer, upstream of the Schaffer collaterals.Care was taken to keep the recording pipette and stimulating electrode as far apart as possible (at least 200 μm) to help isolate the stimulus artifact, the fiber volley, and the field potential.For the stimulus intensity/field potential slope relationship, once a minimal current to reliably evoke a fiber volley and field potential was established, the stimulus current was systematically increased, and the subsequent fiber volley and field potential were recorded.For paired-pulse ratio recordings, stimulus intensity was set to approximately halfmaximal intensity.Recordings were analyzed using custom code in MATLAB (MathWorks), Excel (Microsoft), or Prism (GraphPad).

VSV and adenovirus infections
Primary astrocytes from NTG and double transgenic ΔNLS mice were transfected with 0.66 μg of low-molecular weight poly(I:C) (InvivoGen, #tlrl-picw) per ml of culture medium at DIV 8 using Lipofectamine 3000 5 hours before infection with VSV at 100 multiplicity of infection (MOI) or with adenovirus-eGFP (Ad5CMV-eGFP, lot #ad3586, Viral Vector Core Facility, Carver College of Medicine, University of Iowa) at indicated MOIs (see figures).A change of medium was made 2 hours after viral infections, and the cells were collected 24 hours after infection.VSV levels were measured by RT-qPCR.eGFP fluorescence was analyzed after cultures were fixed with 4% PFA in PBS and stained with DAPI.Fluorescence intensity (integrated density), area, and percent of total area were extracted using Fiji.Data are represented as eGFP normalized to the total DAPI-positive area and percentage of DAPI-positive area per condition.

HSV-1 infections
HSV-1 H129-eGFP strain was generously provided by L. Enquist (Princeton University, Princeton, NJ).Viral stocks were grown on human Vero E6 cells (ATCC, CRL-1586) maintained in DMEM with 10% FBS.Standard plaque assays were performed to titer HSV-1 H129-eGFP stocks and quantify viral load in conditioned medium from infected primary astrocytes.Briefly, stocks or medium samples were serially diluted in DMEM supplemented with 2% FBS and used to infect Vero cells.After 3 hours, cells were washed three times with PBS and further incubated in DMEM with 2% FBS, antibiotics, and 1.5% methylcellulose (37°C).After 48 hours, cultures were fixed with 4% PFA in PBS and counterstained with Hoechst 33342 (Thermo Fisher Scientific).Fluorescent signal was used to detect plaques of infected cells and quantify viral titers.Cell culture plates were imaged on a BX-X710 microscope (Keyence) with a 20× objective (Nikon).Images were stitched with BZ-X Analyzer software (Keyence) and used to count plaques.

Immunohistochemistry
Postmortem human brain tissue blocks or sections from nondemented controls, AD, or behavioral variant frontotemporal dementia (bvFTD) cases were obtained from the Neurodegenerative Disease Brain Bank at University of California, San Francisco (San Francisco, CA) and the Banner Sun Health Research Institute Brain and Body Donation Program of Sun City, Arizona.Case details are listed in table S1.Formalin-fixed tissue blocks were rinsed in PBS and incubated in 30% sucrose for 3 to 5 days at 4°C before sectioning.
Double or triple immunolabeling of free-floating mouse sections was performed with minor modifications depending on the antibodies used.All steps were performed at ambient temperature unless specified.Cryopreserved sections were rinsed in PBS, permeabilized for 30 min or longer in PBS-T, blocked with 10% normal donkey or goat serum (Jackson ImmunoResearch) in PBS-T for 1 to 2 hours, incubated in primary antibodies in 3% serum in PBS-T for up to 48 hours at 4°C, and rinsed with PBS-T.Sections were protected from light in all subsequent steps.Tissue was incubated with fluorescent secondary antibodies in 3% serum in PBS-T for 2 hours, rinsed with PBS-T, mounted, and dried on Superfrost glass slides (VWR, #75799-266) before sealing #1.5 coverglass (VWR, #89239-734) with VECTASHIELD antifade medium containing DAPI (VWR, #101098-050).When necessary, Prolong Diamond Antifade Mounting Media (Thermo Fisher Scientific #P36970) replaced VECTASHIELD mounting medium to minimize quenching of Alexa Fluor 647-conjugated antibodies.Slides were allowed to set overnight before acquiring images.

Microscopy and image analyses
Primary neurons were evaluated for synaptic content using a 40× objective on an ImageXpress Micro Confocal Au4tomated High-Content Analysis System (Molecular Devices) at the Weill Cornell Medicine Automated Optical Microscopy Core Facility.Briefly, four regions of interest (ROIs) were imaged per well, and images were processed to assess the number and total area of synaptophysin-1 or PSD-95-positive puncta using Fiji.Briefly, images of individual channels were background-subtracted and stringently thresholded to the brightest 5 to 10% of all pixels, and particles equal or larger than 1 μm in length were counted.The total area and number of thresholded puncta were measured for each image.
To measure astrocytic TDP-43 in human brain tissue, postmortem hippocampal tissue was immunostained with anti-TDP-43 and anti-GFAP antibodies and imaged on a BX-X710 microscope (Keyence) with a 40× objective (Nikon), and astrocytes were analyzed within the dentate gyrus and Cornu Ammonis regions.Tables S1 and S2 provide detailed information about the human cases and the numbers of cells analyzed per case.Image analysis was performed using Fiji.The patterns and intensities for TDP-43 and GFAP labeling were not consistently colocalized across cases and individual cells, indicating minimal bleed-through between channels.
Astrocytes were defined by drawing ROIs encompassing primary processes and cell soma with strong GFAP fluorescence together with overlapping or adjacent DAPI-positive nucleus.ROIs were extracted, and the intensity of nuclear TDP-43 immunoreactivity was measured for pixels within a DAPI-thresholded mask.The intensity of extranuclear TDP-43 immunoreactivity in GFAP-positive ROIs was measured for pixels in a GFAP-thresholded mask after eliminating the nuclear area defined by the DAPI mask.Intensities were corrected for background fluorescence in each image by subtracting the mean intensity of five circular ROIs drawn in areas that did not have GFAP, TDP-43, or DAPI labeling above background levels.
To evaluate cell-specific localization and intensity of protein expression in mouse brain sections, slides were imaged on a BX-X710 microscope (Keyence) with a 20× objective (Nikon) using the tiling function.Images were stitched with BZ-X Analyzer software (Keyence) and further processed for brightness and contrast using Fiji.To further evaluate cell localization of proteins in mouse brain sections, z-stacks of immunostained tissue were acquired on a Zeiss LSM 880 Confocal Laser Scanning Microscope with a 63× objective, and three-dimensional renderings of maximal projections were made using Imaris software (Oxford Instruments).
Localization of CXCR3 protein in neuronal cell bodies and synaptic compartments was evaluated using a Zeiss LSM 880 Confocal Laser Scanning Microscope.Images of the CA1 stratum radiatum region of the hippocampal formation were acquired using a 63× objective, 4× zoom, 16-line averaging, and a pixel dwell time of 5.3 μs.Filter and detector configurations were optimized using single-antibody controls, and each of the three channels was imaged sequentially to further minimize potential cross-bleed.Images were analyzed using Fiji to measure the number, area, and intensity of puncta that were immunoreactive for CXCR3 and NeuN, synaptophysin-1, synaptotagmin-2, PSD-95, or gephyrin.
Briefly, images of individual channels were background-subtracted, and noise was removed using the "despeckle" function.Images were stringently thresholded to the brightest 5 to 10% of pixels, and particles equal or larger than 1 μm in length were counted.The total area, mean size, and mean intensity of thresholded puncta were measured for each image.The fractional overlap of CXCR3 and each synaptic marker was measured from processed images as Mander's coefficient using the Just Another Colocalization Plugin (119).A total of 9 to 12 images from three to four mice were analyzed per genotype.
For brain tissue, approximately 30 mg of mouse hippocampal tissue was dissected from flash-frozen forebrain in ice-cold PBS under a dissecting scope (AmScope) and refrozen on dry ice in 1.5-ml Eppendorf tubes.Frozen samples were thawed on ice for 1 to 2 min before adding 150 μl of ice-cold lysis buffer to each tube (RIPA with protease and phosphatase inhibitors).Samples were immediately homogenized with a Fisherbrand Bead Mill 24 (Thermo Fisher Scientific, #15-340-163) for 40 s at a speed setting of 5 in a prechilled adaptor tube rack.Samples were centrifuged at 1000g for 2 min at 4°C before sonication in an ice-chilled EpiSonic 2000 water bath (5-s on, 2-s off, 5-min total, amplitude of 40).RIPA-soluble extracts were clarified by centrifugation at 100,000g for 30 min at 4°C in a Beckman Ultracentrifuge, and protein content was measured using a detergent-compatible Bradford assay (Thermo Fisher Scientific).
CXCL10 concentrations in lysates and medium were determined using the IP-10 (CXCL10) Enzyme-linked immunosorbent assay (ELISA) kit (Thermo Fisher Scientific, #BMS6018), according to the manufacturer's instructions.IFN-γ concentrations in cell lysates were determined using IFN-γ ELISA kits from R&D Systems (#DY485-05) and Thermo Fisher Scientific (#88-8314-22), according to the manufacturers' instructions.For each kit, 3 μl of cell lysate were diluted in 97 μl of the provided Sample Diluent, and 50 μl of medium was diluted in 50 μl of the provided Sample Diluent.Concentrations were determined on the basis of standard curves, and values were corrected for the dilution factor.

Microfluidic qPCR
RNA was extracted using the RNeasy Mini Kit with on-column deoxyribonuclease treatment following manufacturer's instructions (QIAGEN, #74106 and #79256).Cultured primary cells were rinsed once with ice-cold PBS, scraped in freshly prepared extraction buffer, and frozen at −80°C until extracted.Saline-perfused, microdissected mouse brain tissue was frozen on dry ice and stored at −80°C until RNA extraction.Tissue was homogenized in fresh extraction buffer using a bead mill for 20 s at a speed setting of 5 in a prechilled adaptor tube rack.
The number of VSV viral copies was determined by RT-qPCR using PowerUp SYBR Green Master Mix (Thermo Fisher Scientific, #A25741) according to manufacturer's instructions.RT-qPCR was performed in a CFX96 Touch Real-Time PCR Detection System (Bio-Rad).All primer sequences are detailed in table S3.
Microfluidic RT-qPCR was performed similarly to previously described protocols (120).Briefly, cDNA was synthesized with the Protoscript First Strand Synthesis Kit (New England Biolabs, #E6300L) and preamplified for 14 cycles against a pool of primers (table S3) using PreAmp Grandmaster mix (TATAA Biocenter, Sweden, #TA05) before exonuclease I treatment (New England Biolabs, #M0293L).Preamplified cDNA was diluted at least fivefold with nuclease-free water and mixed with SsoFast EvaGreen with Low ROX (Bio-Rad, #1725211) and chip-specific DNA Sample Reagents before loading into primed Flex Six or 96.96 Dynamic Array chips (Fluidigm, #100-6308, BMK-M-96.96).Individual primers were mixed with DNA assay reagent (Fluidigm) and loaded into chip inlets.Chips were primed and loaded using an IFC Controller HX (Fluidigm) before measuring and analyzing amplification and melting curves on a BioMark HD System (Fluidigm).Cycle of quantification (Cq) values were thresholded equally for all inlets across each chip run and normalized to the average of reference genes (Actb and Gapdh for tissue samples and Actb and/or Tbp for cultured cells) before determining ΔΔCq and fold change relative to experimental control groups.

Statistical analyses
Statistical specifications are reported in the figures and corresponding figure legends.Replicates for the main figures are specified in table S4.Data are presented as means ± SEM.All statistical tests were performed using GraphPad Prism 8, except Fisher's exact test, which was performed using IBM SPSS Statistics for Windows, version 24.0.The criterion for data point exclusion was established during the design of the study and was set to values above or below two SDs from the group mean.Two-sided Student's t test was used to determine statistical significance between two groups.Welch's correction was used to account for unequal variances.Differences among multiple groups were assessed by oneway or two-way analysis of variance (ANOVA) or mixed-effects model, followed by Dunnett's or Bonferroni's multiple comparisons