Tau protein aggregation is associated with cellular senescence in the brain

Abstract Tau protein accumulation is the most common pathology among degenerative brain diseases, including Alzheimer's disease (AD), progressive supranuclear palsy (PSP), traumatic brain injury (TBI), and over twenty others. Tau‐containing neurofibrillary tangle (NFT) accumulation is the closest correlate with cognitive decline and cell loss (Arriagada, Growdon, Hedley‐Whyte, & Hyman, 1992), yet mechanisms mediating tau toxicity are poorly understood. NFT formation does not induce apoptosis (de Calignon, Spires‐Jones, Pitstick, Carlson, & Hyman, 2009), which suggests that secondary mechanisms are driving toxicity. Transcriptomic analyses of NFT‐containing neurons microdissected from postmortem AD brain revealed an expression profile consistent with cellular senescence. This complex stress response induces aberrant cell cycle activity, adaptations to maintain survival, cellular remodeling, and metabolic dysfunction. Using four AD transgenic mouse models, we found that NFTs, but not Aβ plaques, display a senescence‐like phenotype. Cdkn2a transcript level, a hallmark measure of senescence, directly correlated with brain atrophy and NFT burden in mice. This relationship extended to postmortem brain tissue from humans with PSP to indicate a phenomenon common to tau toxicity. Tau transgenic mice with late‐stage pathology were treated with senolytics to remove senescent cells. Despite the advanced age and disease progression, MRI brain imaging and histopathological analyses indicated a reduction in total NFT density, neuron loss, and ventricular enlargement. Collectively, these findings indicate a strong association between the presence of NFTs and cellular senescence in the brain, which contributes to neurodegeneration. Given the prevalence of tau protein deposition among neurodegenerative diseases, these findings have broad implications for understanding, and potentially treating, dozens of brain diseases.


| INTRODUCTION
The underlying processes driving chronic neurodegeneration in Alzheimer's disease (AD) and related neurodegenerative disorders are largely unknown, and disease-modifying treatments remain elusive.
The experimental data from various studies indicate that tau pathology may be associated with cellular senescence. This complex stress response induces a near permanent cell cycle arrest, adaptations to maintain survival, cellular remodeling, metabolic dysfunction, and disruption of surrounding tissue due to the secretion of toxic molecules (Childs et al., 2016). While many of these features have been described in AD brains and transgenic animal models throughout the literature (e.g., aberrant cell cycle activity, p16 INK4A co-localization with NFTs (Arendt, Rodel, Gartner, & Holzer, 1996), decreased lamin B1, and heterochromatin relaxation (Frost, Bardai, & Feany, 2016); a role for cellular senescence in AD-associated neurodegeneration has not been investigated. We hypothesized that tau accumulation may activate this stress response and thereby initiate a chronic degenerative process culminating in neuron loss and brain dysfunction. To test this hypothesis, we examined human brain tissue with NFT pathology and utilized AD transgenic mouse models that develop tau-associated pathologies. Also, we employed methods to genetically reduce NFTs and pharmacologically clear senescent cells. Our results indicate that NFTs induce cellular senescence in transgenic mice and postmortem human brain tissue. We also found that senolytics decreased cortical NFT burden, brain atrophy, and neuron loss in an advanced age (20 months old) transgenic mouse model of tau-associated neurodegeneration.

| NFT-bearing neurons from postmortem AD brain tissue displayed a senescence-like transcriptomic profile
We queried the publicly available GEO Profiles database (Barrett et al., 2013) for gene sets specific to NFTs. We evaluated laser capture-microdissected cortical neurons containing NFTs from AD brains (GEO accession GDS2795) and compared them to adjacent histopathologically normal neurons for a within-subject study design (Dunckley et al., 2006). NFT-containing neurons upregulated genes involved in cell survival and viability, inflammation, cell cycle progression and molecular transport and downregulated apoptosis, necrosis, and cell death pathways (Figure 1a). NFκB, a pro-survival master transcriptional regulator of inflammation, was the highest predicted upstream regulator of the NFT gene expression profile. In agreement with inflammatory activation, other predicted upstream regulators included IFNG, TNF, TLR4, IL1B, and CXCL1 ( Figure 1b).
Collectively, the molecular pathways identified in the NFT analyses resembled cellular senescence.

| NFTs were associated with a senescenceassociated transcriptomic profile in tau transgenic mice
We used the rTg(tauP301L)4510 transgenic mouse line, hereon referred to as "tau NFT " to investigate a link between NFT formation and a senescence-like phenomenon in neurodegeneration. These mice develop well-characterized, aggressive, tau pathology in forebrain regions concomitant with neurodegeneration and cognitive deficits (Santacruz et al., 2005; pathology illustrated in Supporting Information Figure S1). Mice that overexpress wild-type human tau, "tau WT ," express the same level of transgenic human tau protein as tau NFT , but acquire age-dependent tau pathogenesis at a much slower rate and are used to identify effects of elevated pre-pathogenic tau (Hoover et al., 2010;Supporting Information Figures S1 and S2); age-matched tau NFT littermate mice without human tau overexpression serve as wild-type controls, "CTL". To determine whether NFT-containing neurons in mice induced a gene expression profile resembling cellular senescence, we assessed hippocampal gene expression patterns in tau NFT mice before (~2 months old) and after (~6 months old) NFT formation (GSE56772). Consistent with NFTs from human AD, mouse NFTs also caused significant activation scores for IFNG, TNF, and IL-1B, as well as enrichment in other senescence-associated JAK, STAT, CDKN2A, and BCL2 predicted upstream regulators ( Figure 1c) indicating translational relevance for using tau NFT mice to explore our hypothesis.

| Evidence of DNA damage, SASP, and NFκB activation were associated with NFTs
Senescence-inducing stressors often inflict DNA damage that drives production of the SASP (Rodier et al., 2009). Tau NFT mouse brains displayed significantly elevated histone γ-H2ax, a sensitive marker of both double-stranded DNA breaks and cellular senescence (Sedelnikova et al., 2004; p = 0.0056; Figure 1d-e). The cell cycle protein p21, encoded by Cdkn1a, is upregulated in many senescent cell types and has been associated with DNA damage during neuronal aging (Jurk et al., 2012). Similarly, elevated expression of the cyclin-dependent kinase inhibitor 2a, Cdkn2a, is one of the most robust markers of cellular senescence, and its protein product, p16 INK4A , colocalizes with NFTs in human AD (Arendt et al., 1996). Because anti-p21 and anti-p16 INK4A antibodies are notoriously poor in mouse tissue, we exclusively measured Cdkn1a and Cdkn2a gene expression. Tau NFT brains expressed three-fold higher Cdkn1a than control mice (p = 0.0178, Figure 1f), which was replicated in a separate mouse cohort (p = 0.0086, Supporting Information Figure S2f). Moreover, Cdkn2a was expressed at levels 2.7-and 2.6-fold higher in tau NFT than CTL and tau WT , respectively (p = 0.0303 and p = 0.0352, respectively; Figure 1g); this effect was replicated in an independent mouse cohort (p = 0.0016, Supporting Information Figure S2g). Senescent cells exert chronic tissue degeneration through secretion of toxic SASP (Coppe et al., 2010). Consistent with the transcriptomic profile in human NFT-bearing neurons and mouse brain tissue (Figure 1a-c), SASP genes were found to be upregulated in tau NFT brains, that is, Il1b was four-and twofold higher than CTL and tau WT , respectively; and Cxcl1 was fourfold higher than both control genotypes; Tnfa was 13-and eightfold higher than CTL and tau WT , respectively; Tlr4 was threefold higher than both control genotypes (Figure 2a-d). Further gene expression analyses allowed us to define an array specific to tau pathology in tau NFT brains (Supporting Information Figure S2e). NFκB regulates the pro-survival, pro-inflammatory SASP gene expression profile characteristic of cellular senescence (Salminen & Kaarniranta, 2011). Consistent with NFκB pathway activation and the SASP profile, nuclear-localized NFκB p65 was significantly increased in tau NFT brains (Figure 2e-f).
In all measures, tau WT mice were not significantly different from CTL. These results suggest that insoluble tau and/or post-translational modifications associated with insoluble tau, but not general tau overexpression, were responsible for the senescence-associated profile (i.e., DNA damage, NFκB activation, and upregulated SASP; Figures 1, 2 and Supporting Information Figure S2).

| SA β-gal activity did not correlate with NFTs or brain atrophy
In regenerative tissues and in vitro cultures, senescent cells may exhibit SA β-gal activity, which is a measure of lysosomal galactosidase activity at pH 6.0 and indicative of altered/expanded lysosomal F I G U R E 1 Neurofibrillary tangles were associated with cellular senescence-associated gene pathways in human Alzheimer's disease neurons and tau transgenic mouse brains. (a) Pathways and predicted upstream regulators identified by ingenuity pathway analyses (IPA, QIAGEN) as significantly enriched in Alzheimer's disease patient-derived neurons with neurofibrillary tangles compared to non-tanglecontaining neurons; z-score plotted on x-axis and (p-value) indicated in bar graph. Cellular functions and (b) predicted upstream regulators employed by neurofibrillary tangle-containing neurons derived from Alzheimer's disease patient are shown. (c) Predicted upstream regulators of gene transcription in tau NFT mice after the onset of neurofibrillary tangles (~6 months old vs.~2 months old); z-score plotted on x-axis and (pvalue) indicated in bar graph. (d-e) Representative immunoblot generated by capillary electrophoresis on chromatin-bound fractions from mouse forebrain homogenate probed with anti-γ-H2ax antibody. (e) Densitometric normalization of γ-H2ax to total protein content (CTL: n = 3; tau WT n = 4; tau NFT : n = 5; ANOVA, p = 0.0056. Mice aged 16 to 18 months old). (f-g) Quantitative gene expression on RNA isolated from CTL (open bar, n = 3), tau WT (closed bar, n = 3), and tau NFT (red bar, n = 4) mouse forebrain targeting (f): Cdkn2a, p = 0.0066, and (g) Cdkn1a, p = 0.0207. Gene expression was analyzed by one-way ANOVA Tukey's multiple comparison post hoc. Data are graphically represented as mean ± SEM compartments (Severino, Allen, Balin, Balin, & Cristofalo, 2000). The examination of the gene that codes for the hydrolase enzyme, galactosidase beta (β) 1 (Glb1), revealed that tau NFT mice expressed higher Glb1 gene expression than controls (Supporting Information Figure S3). However, staining for β-gal hydrolase activity at pH 6.0 revealed fewer positive cells than controls. Furthermore, SA β-gal-reactive cells were observed even in very young mice (1 month old) and the number of SA β-gal-reactive cells was positively correlated with brain mass (R 2 = 0.4852, p = 0.0039 Supporting Information Figure S3). While our results indicate that SA β-gal reactivity did not correlate with other senescence markers or brain atrophy, the observed increase in Glb1 gene expression along with a decrease in lysosomal activity at pH 6.0, compared to controls, is suggestive of tau-associated lysosomal defects, which have been reported by others (Caballero et al., 2018;Wang, Martinez-Vicente, et al., 2009).

| NFT-containing brain tissue displayed aberrant cellular bioenergetics
Mitochondrial dysfunction is obligatory for SASP production and cellular senescence (Correia-Melo et al., 2016;Hutter et al., 2004). To examine mitochondrial bioenergetics, we performed high-resolution respirometry to yield accurate quantitative measurements of oxidative phosphorylation in response to specific substrates for complex I, complex II, fat oxidation, and electron-transfer system (ETS) capacity.
Across genotypes, we compared cortex, hippocampus, and cerebellum. This allowed for the evaluation of specific differences in oxygen consumption due to elevated transgenic tau (comparing CTL with tau wt and tau NFT ), pathogenic tau-specific effects (comparing tau wt to tau NFT ), as well as the interaction among brains regions and tau expression (e.g., cortex and hippocampus express transgenic tau and develop NFTs, but cerebellum does not). We found a significant genotype main effect for oxygen flux in both cortex and hippocampus, indicating that global respiratory capacity was impaired in NFT-containing brain regions (p < 0.0001; Figure 3), an effect primarily driven by CI + CII respiration coupled to ATP production (cortex: p = 0.0034; hippocampus: p = 0.0215; Figure 3g,h, respectively), and uncoupled or maximum respiratory capacity (cortex: p = 0.0248; hippocampus: p = 0.0261; Figure 3g,h, respectively). These changes were different between tau NFT and each of the control mouse lines, CTL, and tau WT mice. Because tau WT and tau NFT mice express comparable total tau levels, alterations to respiratory capacity cannot be attributed to tau overexpression. Citrate synthase activity is a surrogate marker of total mitochondrial content/mass and was similar across genotypes and brain regions ( Figure 3i) suggesting that the defects in cellular respiration were due to altered mitochondrial quality and not content/mass. Moreover, tau NFT cerebellum did not show deficits in cellular respiration or Cdkn2a upregulation (Figure 3j,k), indicating that F I G U R E 2 Neurofibrillary tangles were associated with upregulation of SASP gene expression and NFκB activation. (a) Quantitative gene expression on RNA isolated from CTL (open bar, n = 3), tau WT (closed bar, n = 3), and tau NFT (red bar, n = 4) mouse forebrain targeting SASPassociated genes Il1b, p = 0.0025; (b) Cxcl1, p = 0.0040; (c) Tnfa, p = 0.0114; and (d) Tlr4, p = 0.0144. (d) Immunoblot generated by capillary electrophoresis on subcellular fractionated mouse forebrain homogenate probed with anti-NFκB p65 antibody. Total cellular p65 (top blot) and nuclear-localized p65 protein levels (bottom blot) were (e) normalized to total protein content. Total p65, p = 0.0758; nuclear p65, p = 0.0223. CTL: open bar, n = 3; tau WT : closed bar, n = 4; tau NFT : red bar, n = 5. In all experiments, mice were aged 16-18 months old; both males and females were included. Significance was determined by one-way ANOVA Tukey's multiple comparison post hoc. Data are graphically represented as mean ± SEM senescence-associated mitochondrial dysfunction was present only in brain regions with persistent pathogenic tau expression.

| Cdkn2a upregulation occurred with NFT onset and correlated with NFT density
We pursued multiple genetic approaches to determine whether senescence was mechanistically linked to NFT density, NFT onset, or merely protein accumulation. Reducing NFT load in age-matched animals is not feasible; once NFTs form, they cannot be therapeutically eliminated. However, genetically ablating endogenous mouse tau (microtubule-associated protein tau, Mapt) reduces NFT pathology and neurodegeneration in tau NFT mice (tau NFT -Mapt 0/0 ; Wegmann et al., 2015). The reduced tau pathology corresponded with 60% lower

| Cdkn2a upregulation was specific to NFT tau pathology and correlated with brain atrophy
To determine whether Cdkn2a expression was driven specifically by NFTs, or whether AD-associated Aβ protein deposition also increased Cdkn2a, we utilized 3xTgAD mice that acquire both AD-associated pathologies with Aβ deposition and NFT onset at 6 and 18 months of age, respectively (Oddo et al., 2003). In 15-month-old mice with heavy Aβ deposition and phosphorylated tau, but lacking NFT pathology (Orr, Salinas, Buffenstein, & Oddo, 2014), Cdkn2a expression was not elevated (Figure 4e). These data indicate that Cdkn2a expression was neither a response to general protein accumulation, nor to pre-NFT tau pathology, but instead required the presence of NFTs. Further, when plotted against brain weight, Cdkn2a expression was a strong predictor of brain atrophy across mouse lines (p < 0.0001, R 2 = 0.5615; Figure 4f).

| CDKN2A was upregulated in NFT-containing brains from patients with progressive supranuclear palsy
Tau pathology is common among >20 brain diseases. To investigate whether the findings in human AD neurons and transgenic mice translated to human brains with pure tauopathy (i.e., NFT pathology without other protein aggregates such as Aβ), we acquired human brain tissue with histopathologically confirmed progressive supranuclear palsy (PSP; Table 1 for patient characteristics). PSP is an ageassociated tauopathy that clinically manifests as parkinsonism with additional motor abnormalities and cognitive dysfunction (Orr et al., 2017) and is neuropathologically defined by the accumulation of F I G U R E 3 Brain regions with neurofibrillary tangles displayed altered cellular respiration. (a-c) Representative respirometric traces from cortical and (d-f) hippocampal tissues using the SUIT protocol to measure oxygen consumption (top gray traces: CTL; black middle traces: tau WT ; bottom red traces: tau NFT ). (g) Tissue mass-specific respiration analyses in cortical and (h) hippocampal tissue. Two-way ANOVA Tukey's multiple comparison post hoc: **p < 0.005. (i) Biochemical analyses of citrate synthase (CS) activity to assess total mitochondrial content in the cortex and hippocampus (n = 5/group). Experimental mice were aged 16-18 months old with n = 6/group; both males and females were included. (j) Total oxygen consumption and (k) Cdkn2a gene expression were measured in the cerebellum, a brain region devoid of NFTs. n = 3/group. Data are graphically represented as mean ± SEM. ETF_L (fat oxidation in the absence of ADP [state 2]), ETF_P (fat oxidation coupled to ATP production), CI_P (complex I activity linked to ATP production [state 3]), CI + CII_P (complex I and complex II linked respiration [state 3]), CI + CII_E (complex I and complex II linked respiration uncoupled [maximum respiration]), and CII_E (complex II activity uncoupled). Data are graphically represented as mean ± SEM MUSI ET AL.
four-repeat (4R) tau, NFTs, gliosis, and neurodegeneration (Flament, Delacourte, Verny, Hauw, & Javoy-Agid, 1991). Consistent with the results from transgenic mice, CDKN2A was upregulated in PSP brains (p = 0.0415, Figure 4g) and expression correlated with NFT deposition, specifically in the parietal lobe (ANOVA, p = 0.0008; Kendall's Tau rank correlation, p = 0.059, Figure 4h). Moreover, one individual with the worst cognitive performance, Mini-Mental State Examination (MMSE) score of 12, displayed the highest level of CDKN2A expression, and high molecular weight tau (Figure 4i). Collectively, these findings led us to conclude that NFTs were directly linked to senescence-associated Cdkn2a upregulation, which in turn was a strong predictor of neurodegeneration and cognitive decline.

| Senolytic treatment reduced NFT burden and neurodegeneration
Senescent cells comprise a small proportion of total cellular makeup within a tissue (~15%; Herbig, Ferreira, Condel, Carey, & Sedivy, 2006). Nonetheless, genetically (Baker et al., 2011) or pharmacologically (Zhu et al., 2015) clearing even a small percentage of these cells improves health span and delays age-associated diseases (Kirkland et al., 2017). We used some of the best characterized senolytics to date, dasatinib and quercetin (DQ), to determine the utility of targeting cellular senescence to treat tau-associated neurodegeneration in late life. Beginning at 20 months old, tau NFT -Mapt 0/0 and nontransgenic Mapt 0/0 mice were randomized to receive vehicle or DQ at biweekly intervals for 3 months. When mice were 23 months old, brain structure and cerebral blood flow were analyzed with MRI and postmortem histopathology ( Figure 5 and Supporting Information Figure S5). Consistent with senescent cell removal, intermittent DQ treatment significantly reduced the number of NFT-containing cortical neurons (p < 0.0001, 35% reduction; Figure 5a,b). Relative to the existing neuronal population at this advanced age, gene expression of the NFT-associated senescence gene array was reduced by DQ F I G U R E 4 Senescence-associated Cdkn2a was significantly upregulated in mouse and human brains with neurofibrillary tangles and tracked with total tangle deposition and brain atrophy. (a) Genetically ablating endogenous mouse tau to significantly reduce neurofibrillary tangle load resulted in a concomitant 60% reduction in Cdk2na expression (two-tailed t test: p = 0.0041; n = 3/group) and (b) significant reduction in brain atrophy (two-tailed t test, p = 0.0143; n = 3/group). (c) Tracking Cdkn2a expression in tau WT mice revealed a significant age-dependent increase (one-way ANOVA: p = 0.0043; n = 3/group for tau WT and n = 4 tau P301L ). In contrast to significantly lower expression than tau NFT mice at 16 months old (p = 0.0075), Dunnett's multiple comparison test indicated that at 22 months of age, tau WT mouse Cdkn2a expression was no longer statistically lower than tau NFT mice (p = 0.0577) and by 28-30 months they were are statistically the same (p = 0.999). (d) Immunofluorescence and Bielschowsky silver staining revealed neurofibrillary tangles in 18-month-old tau WT mouse hippocampal CA1 (NeuN, neuron, green; PHF1: phosphorylated tau, red; DAPI, blue, nuclei). (e) qPCR analyses of RNA extracted from 3xTgAD mice with Aβ plaques were compared to tau NFT set at y = 1. 3xTgAD Cdkn2a expression was no different than age-matched C57BL/6 mice (two-tailed t test, p = 0.1081; n = 3 WT, n = 6 3xTgAD; n = 4 tau NFT ). Both mouse cohorts expressed significantly less Cdkn2a than tau NFT mice (one-way ANOVA: p < 0.0001). (f) Cdkn2a expression level was significantly correlated with brain atrophy (R 2 = 0.5615, p < 0.0001; n = 43). (g) qPCR analyses of RNA extracted from brains from control older adult humans (n = 10; ave. age = 85.70 years) and age-matched progressive supranuclear palsy (n = 14; ave. age = 83.86 years) indicated a 57% upregulation of CDKN2A with progressive supranuclear palsy diagnosis (unpaired t test, p = 0.0415) that (h) positively correlated with neurofibrillary tangle deposition in the parietal lobe (ANOVA, p = 0.0008; Kendall's Tau rank correlation p = 0.059). (i) Immunoblot generated by capillary electrophoresis on cortical brain homogenate from control and progressive supranuclear palsy human brains probed with total tau antibody, HT7. The individual with the highest CDKN2A expression (panel g) displayed high molecular weight tau, lane 9*. Data are graphically represented as error bars, mean ± SEM T A B L E 1 Human postmortem brain characteristics  Figure S6a). Among these genes, those highly sensitive to NFT-dependent upregulation (Supporting Information Figure S2) were most affected (i.e., Tlr4: p = 0.0459 and Cxcl1: p = 0.0142; Figure 5c; Supporting Information Figure S6a). NFTs are highly correlated with the rate of ventricular enlargement, an indicator of brain atrophy and hallmark of AD pathology (Silbert et al., 2003). Tau NFT mice recapitulate this pathology on a wild-type (Supporting Information Figure S1) and Mapt 0/0 background (p = 0.0007; Figure 5d). The DQ-dependent reduction in cortical NFTs corresponded with decreased ventricular volume pathology (28% decrease, p = 0.05, Figure 5d,e) and a reduction in cortical brain atrophy (compared to controls: p = 0.0092 and p = 0.0274, vehicle and DQ, respectively; Supporting Information Figure S5a). The absence of a full rescue of ventricular enlargement to that of control animals was not completely unexpected considering the severity of disease and age of the animals when treatment was initiated.
Aberrant cerebral blood flow is a functional defect that occurs in AD and tau NFT mice and is closely associated with cognitive impairment (Wells et al., 2015). In brain tissue with tau pathology, cerebral blood flow was elevated in tau NFT -Mapt 0/0 vehicle-treated mice (21% whole brain, p = 0.045; cortex, 48.7%, p = 0.051, Supporting Information Figure S5b,c) and consistent with previous reports of tau NFT mice on a Mapt +/+ background (Wells et al., 2015). DQ improved aberrant cerebral blood flow in tau NFT -Mapt 0/0 mice such that cerebral blood flow was no longer statistically different from controls (Supporting Information Figure S5b,c). Overall, a composite analysis of DQ treatment in tau NFT mice revealed a significant global benefit on cerebral blood flow and neurodegeneration (p = 0.0138; Supporting Information Figure S5d). To elucidate whether the DQ-dependent reduction in NFT burden, SASP expression, and improvements to brain structure and cerebral blood flow conferred neuroprotection, we measured levels of cell type-specific protein expression in the brain. DQ-treated mice expressed significantly higher levels of neuronal proteins (NeuN: 25%, synaptophysin: 40.8%; PSD95: 38.5%; p < 0.05; Figure 5f-i).
The astrocyte protein GFAP was unchanged, while microglia Iba1 expression was elevated (Iba1: 40%, p = 0.0013; Supporting Information Figure S6b-d) suggesting that DQ-mediated neuroprotection and decreased SASP were not derived from a reduction in pro-inflammatory glia (astrocytes or microglia) but instead associated with fewer NFT-containing neurons. Moreover, DQ did not alter total human tau protein levels indicating the effects were not driven by changes in tau protein expression, but rather insoluble NFTs or associated post-translational modifications (Supporting Information Figure S7). Collectively, our data suggest that cyclic senescent cell removal of NFTs with DQ produced long-lasting global effects on brain, as evidenced by histopathology and MRI analyses.

| DISCUSSION
The inability to effectively treat tau-associated diseases arises, in part, from a limited understanding of processes driving neurodegeneration during the prodromal period. We have identified cellular senescence, the quintessence of latent tissue degeneration, as a cellular mechanism upregulated in tau-associated neurodegeneration.
Findings in NFT-developing transgenic mice, postmortem human AD, and PSP brain tissue support this concept.
Cellular senescence is an elaborate stress response that varies across tissues and even among cell types within tissues. Our experimental data provide an initial report of features consistent with cellular senescence in the brain (i.e., transcriptomics, upregulated However, we cannot exclude the possibility that other tau species and/or specific post-translational modifications may also contribute to cellular senescence. The upstream mediators driving tau-associated cellular senescence in AD and PSP also remain unknown; however, it is tempting to speculate that tau-induced cell cycle reentry may be involved (Arendt, 2012). Aberrant cell cycle re-entry causes neuronal apoptosis and AD-associated pathology (Park, Hallows, Chakrabarty, Davies, & Vincent, 2007) and requires soluble tau (Seward et al., 2013). The observed increase in NFT-associated Cdkn1a and Cdkn2a gene expression may allow stressed neurons to abort cell cycle re-entry and enter a cellular state similar to cellular senescence. In this way, NFTs formed in early pathogenic stages due to acute stress may initially protect neurons from cell death, but then contribute to neurodegeneration later in life through senescence-like mechanisms by altering the bioenergetic state of the brain and upregulating the toxic SASP.
Pathogenic tau induces a traditional neuroinflammatory response by activating microglia and astrocytes (for recent review Laurent, Buee, & Blum, 2018). Our data suggest that NFT-containing neurons may be active participants in perpetuating the inflammatory response as well. Future studies are required to better understand the contribution of SASP to the overall neuroinflammation phenotype common among many brain diseases. Nonetheless, our findings suggest that therapeutically targeting cellular senescence effectively interrupted a chronic neurodegenerative cascade to decrease NFTassociated pathology and improve brain structure and aberrant cerebral blood flow even in the presence of established tau pathology in a late-life advanced disease state. Overall, our data provide evidence that cellular senescence may be an underlying pathogenic process common among tauopathies, which opens a new field of investigative research and offers a potential druggable target to treat the >20 tau-associated neurodegenerative diseases.

| Mice
All animal experiments were carried out following National Institutes of Health and University of Texas Health Science Center at San Antonio (UTHSCSA) Institutional Animal Care and Use Committee guidelines. We used 16-to 32-month-old male and female rTg4510 and rTg21221 mice that reversibly express P301L mutant human tau or wild-type human tau 4R02, respectively, on either a wild-type or Mapt knockout Bl6/FVB genetic background (Hoover et al., 2010;Santacruz et al., 2005;Wegmann et al., 2015). Nontransgene expressing littermates from rTg4510 and rTg21221 are used as controls; since no differences were found between these control lines, only littermates from rTg4510 are used here (Supporting Information Figure S1f-i). The mice were bred by Rose Pitstick and George A.
Carlson at McLaughlin Research Institute, Great Falls, MT. Mouse euthanasia, brain dissection, and preparation were performed as previously described (Orr et al., 2012(Orr et al., , 2014.

| Ingenuity pathway analyses
The GEO accessions GDS2795 and GSE56772 were accessed from the GEO Profiles database (Barrett et al., 2013;Dunckley et al., 2006)  Findings from GDS2795 were replicated with more stringent criteria (p < 0.05 for NFT/CTL ratios with no fold change limit) allowing for 1715 genes to be uploaded into IPA with similar results. Similarly, these findings were replicated a third time using the LIMMA package, the most common method for microarray analysis. This method did not take into account within-subject design. Using a p < 0.05, 1,219 differentially regulated genes were uploaded for IPA analyses; the results were similar to the original findings. Furthermore, results from GSE56772 were replicated using gene set enrichment analysis (GSEA) with default setting and similar results were obtained.

| RNA extraction and qPCR
Frozen forebrain and cerebellum were powdered in liquid nitrogen.
RNA was extracted from~25 mg of each respective brain (or brain region) using the RNAqueous 4PCR® kit (Ambion), following the manufacturer protocol including the 15-min DNase treatment. qPCR was performed on 25 ng RNA using the TaqMan® RNA-to-CT ™ 1-step kit.
All gene expression analyses were made using TaqMan gene expres-

| Protein extraction and capillary electrophoresis
Approximately 50 mg frozen forebrain was used for subcellular fractionation and capillary electrophoresis as previously described (Orr et al., 2014;Orr, Garbarino, Salinas, & Buffenstein, 2015). Briefly, frozen tissue was powdered in liquid nitrogen and then homogenized with dounce and pestle and fractionated following manufacturer protocol (Subcellular Protein Fractionation Kit, Thermo Fisher Scientific).
Protein concentrations were determined with BCA (Bio-Rad); 2 µg protein was used for capillary electrophoresis. Antibodies were  Table S1 for complete antibody information. Protein quantification was performed by normalizing to total protein concentration (Li & Shen, 2013;Moritz, 2017; Supporting Information Figure S8).

| Histology
Brains were fixed in 4% PFA for 48 hr and transferred to PBS containing 0.02% sodium azide and vibratome sectioned at 30 µm. Sections were washed 3× with TBS (pH 7.4) and incubated in 50% ethanol for 5 min, followed by 70% ethanol for 5 min. The sections were then submerged in 0.7% sudan black b dissolved in 70% ethanol for 5 min to quench lipofuscin-like autofluorescence. Tissues were then rinsed three times for 1-2 min in 50% ethanol. Following this step, tissue sections were transferred from 50% ethanol to TBS and proceeded to immunofluorescence staining as described previously (Orr et al., 2015(Orr et al., , 2014. Primary antibodies used are as fol-

| Confocal image analyses
Image analyses were conducted using ImageJ's FIJI. Analyses were performed on confocal z-stacks imaged at 40x magnification. A maximum intensity image was created by compressing four z-stack planes. All analyzed DAPI fields were applied a bandpass filter under the same conditions, applied a threshold, and measured using particle analysis excluding particles smaller than 25 µm 2 . All particles measured in the analysis were checked for mislabels, and any particles that included two nuclei or exhibited abnormal/incorrect selection were excluded from analysis. Cell type was identified using NeuN (neurons) and PHF1 (NFT-bearing neurons) immunofluorescence.

| SA β-gal staining
Following euthanasia, brains were immediately removed and freshfrozen in an isopentane/liquid nitrogen slurry. The frozen brains were immediately adhered to the cryotome chuck with optimal cutting temperature compound (OCT) precooled to −18°C; 10-µm coronal sections were collected and mounted on superfrost plus microscope slides (Fisher Scientific). After sectioning, slides were fixed for 10 min in 2% paraformaldehyde/0.2% glutaraldehyde at room temperature, rinsed 3× in TBS, and stained with SA β-gal staining solution overnight (Dimri et al., 1995). Following SA β-gal staining, sections were processed for immunofluorescence as described above.

| High-resolution respirometry
HRR was conducted using two Oxygraph-2k (models D and G) machines from Oroboros Instruments (Austria). To minimize mitochondrial damage associated with mitochondrial isolation techniques, we measured oxygen consumption in fresh brain tissue homogenates (Makrecka-Kuka, Krumschnabel, & Gnaiger, 2015). Whole hippocampus, cortex, and cerebellum were homogenized with~15 strokes using a Kontes glass homogenizer in 5% w/v ice cold MiRO6. Two milligram of brain homogenate was loaded into the chamber, and experiments were carried out when oxygen concentration in each well was saturated under atmospheric conditions (~190 nM/ml O 2 ).
All reagents and SUIT protocol were described previously (Pesta et al., 2011) with small modifications. Briefly, 1.25 mM ADP was sufficient for saturation in brain homogenate, rotenone was added at a concentration of 1 µM, and FCCP was added in a single injection at a concentration of 0.5 µM.  (Ogrodnik et al., 2017). Mice were weighed and fasted for 2 hr prior to treatment. One month after the first treatment, senolytic or vehicle gavage continued on a biweekly basis for a total of six treatment sessions over 12 weeks. Within 2 weeks of the final treatment, all mice underwent MRI analyses.

| MRI
MRI experiments were performed on an 11.7 Tesla scanner (BioSpec, Bruker, Billerica, MA). A surface coil was used for brain imaging and a heart coil (Muir, Shen, & Duong, 2008) for arterial spin labeling. Coil-to-coil electromagnetic interaction was actively decoupled.
Mice were maintained on 1.5% isoflurane anesthesia for MRI duration. Anatomical MRI: Anatomical images were obtained using a fast spin-echo sequence with a matrix = 128 x 128, field of view (FOV) =  (Strupp, 1996). Anatomical MRI images were used to measure cortex, subcortex, ventricle, and whole brain volume. The desired regions of interest (ROIs) were outlined, and volumes were obtained by multiplying ROI total voxels by voxel volume (0.004 mm 3 ). Ventricular volume was obtained by thresholding anatomical image voxels to highlight regions of greater intensity, followed by ROI traces of the target regions. Whole brain volume was obtained by ROI trace after removal of the skull using a local Gaussian distribution 3D segmentation MATLAB code (Wang, Li, Sun, Xia, & Kao, 2009 comparisons were analyzed using one-way analysis of variance (ANOVA) with Tukey's post hoc or unpaired t test unless stated otherwise in the figure legends. Respirometric data, brain volume, and cerebral blood flow data were analyzed using two-way ANOVAs (genotype × respirometric parameter) and (treatment × brain region), respectively, with Tukey's post hoc comparisons.

| Human PSP brain tissue analyses
The PSP group contained 14 samples and was compared to 10 age-