Evolutionarily conserved regulators of tau identify targets for new therapies

Tauopathies are neurodegenerative diseases that involve the pathological accumulation of tau proteins; in this family are Alzheimer disease, corticobasal degeneration, and chronic traumatic encephalopathy, among others. Hypothesizing that reducing this accumulation could mitigate pathogenesis, we performed a cross-species genetic screen targeting 6,600 potentially druggable genes in human cells and Drosophila. We found and validated 83 hits in cells and further validated 11 hits in the mouse brain. Three of these hits (USP7, RNF130, and RNF149) converge on the C terminus of Hsc70-interacting protein (CHIP) to regulate tau levels, highlighting the role of CHIP in maintaining tau proteostasis in the brain. Knockdown of each of these three genes in adult tauopathy mice reduced tau levels and rescued the disease phenotypes. This study thus identifies several points of intervention to reduce tau levels and demonstrates that reduction of tau levels via regulation of this pathway is a viable therapeutic strategy for Alzheimer disease and other tauopathies.


In brief
Using cross-species screens, Kim et al. identified 11 genes that when knocked down decrease tau in the mouse brain. Furthermore, they showed that adult knockdown of USP7, RNF130, and RNF149 in a tauopathy mouse model rescued the disease phenotypes. Mechanistic studies revealed that these three genes regulate tau by counteracting the C terminus of Hsc70-interacting protein (CHIP)-mediated tau ubiquitination.

INTRODUCTION
Misfolded and aggregated proteins are common pathological features in many neurodegenerative diseases, from idiopathic conditions such as Alzheimer disease (AD) and Parkinson disease (PD) to rarer inherited conditions such as Huntington disease. 1,2 In these diseases, collectively known as proteopathies, despite their different phenotypes and pathologies, the misfolded and/ or abnormally high quantities of the disease-related proteins create toxic gains of function across the neural system. 3,4 A subgroup of proteopathies termed tauopathies shares a further connection as they involve the accumulation of tau proteins. There are six isoforms of tau, produced by alternative splicing of microtubule-associated protein tau (MAPT) but all of them stabilize microtubules and therefore play roles in intraneuronal transport. 5,6 Hyperphosphorylation of tau promotes its accumulation notably in AD, where it forms the hallmark neurofibrillary tangles (NFTs), 6 and in a number of other conditions including Pick disease, frontotemporal lobar degeneration with tau pathology (FTLD-tau), progressive supranuclear palsy (PSP), corticobasal degeneration (CBD), and chronic traumatic encephalopathy (CTE). [6][7][8][9][10] There is currently no treatment for any of these disorders.
Although each of these diseases has its own course, in all tauopathies, the cascade of aberrant posttranslational modification, misfolding, aggregation, trans-neuronal propagation, and deposition such as NFTs correlates with worsening clinical symptoms in both human patients and mouse models. The notion that protein levels can be pathogenic is supported by rare genetic forms of these diseases, such as frontotemporal dementia caused by an extra copy of the MAPT gene. 11 In this case, the tau protein itself is not mutated; it is simply overabundant. Similarly, in mice, overexpression of wild-type (WT) tau leads to age-dependent cognitive impairment and synaptic dysfunction. 12 Conversely, reducing endogenous tau levels prevents amyloid-b-induced deficits in an AD mouse model and protects mice against excitotoxicity. 13 Such findings strongly suggest that elevated tau protein levels are causally linked to disease pathogenesis-and led us to hypothesize that there must be genes and molecular pathways that regulate tau levels-and manipulation of these genes can lower tau levels, providing a therapeutic strategy for a whole family of currently untreatable diseases.
Previous work from our lab has supported this hypothesis. Using a very limited genetic screen of 636 genes, we identified NUAK1 and TRIM28 as potent tau regulators. 14, 15 Notably, (B and C) In vitro validation of putative hits from the primary screen using ELISA and immunoblot (IB) in 293T cells. (B) Quantification of the average tau protein level measured by ELISA. Each bar above a gene represents an individual shRNA. The effect of candidate-targeting shRNAs was normalized to the effect of nontarget (NT) shRNAs. Genes were selected as tau modifiers if two independent shRNAs (p < 0.1) reduced or increased tau levels by >15% (indicated by blue and red dotted lines). Genes with white bars represent candidates with non-significant effect. (C) IB image representing the effects of USP7, RNF130, and RNF149 knockdown quantified in the bar graph in (B). (D) Illustration depicting intracerebroventricular (ICV) injection of shRNA-expressing AAV8 at postnatal day 0 (P0) to suppress target gene expression in the mouse brain. CBA, chicken b-actin promoter. (E) Western blot and its quantification showing the effects of USP7, RNF130, or RNF149 knockdown on the level of endogenous tau in the mouse brain. Brain tissues were harvested 21 days after infection at P0.
(legend continued on next page) NUAK1 haploinsufficiency reduced tau levels by 15%-20% in P301S tau transgenic (PS19) mice, a model of neurodegenerative tauopathy and AD, and even this modest reduction in tau protein was sufficient to reverse multiple disease features. Other known tau regulators include the E3 ubiquitin ligase C terminus of Hsc70-interacting protein (CHIP) and p300/CBP. CHIP polyubiquitinates and degrades tau. 16,17 p300/CBP acetylates and stabilizes tau, leading to its accumulation and toxicity. [18][19][20] To date, there is no obvious way to increase CHIP activity and thus reduce tau levels. However, reduction of tau acetylation has been shown to decrease total tau and acetylated tau, taurelated pathology, and memory deficits in a tauopathy mouse model. 19,20 Thus, we rationalized that reducing tau levels would result in reduction of acetylated tau as well as all other modified tau species. To this end, we decided to undertake further unbiased screens to identify tau regulators that can reveal possible avenues for therapy and lead to a better understanding of tau biology.
In this study, we used a cross-species screening strategy to examine 6,600 potentially druggable genes in both human cells and fruit flies. We identified and validated 82 regulators of tau protein levels and pursued mechanistic studies of three of them. Specifically, we found that genetic knockdown (KD) of USP7, RNF130, or RNF149 in the brain reduced pathological tau species (p-tau and tau oligomers) and mitigated memory and learning deficits and other pathological features in a tauopathy mouse model.

RESULTS
Parallel genetic screens in Drosophila and human cell lines identify new modulators of tau levels To identify genes that regulate tau levels, we performed parallel genetic screens in human cells and Drosophila ( Figure 1A). We used a stable human cell line based on Daoy cells expressing a bicistronic construct containing EGFP-fused WT tau441 (2N4R) and DsRed separated by an internal ribosomal entry site (IRES). In this screen, we targeted 6,600 druggable genes using a pooled shRNA library (10 shRNAs per gene) and rescreened a subset of ubiquitin-related genes (858 genes) using a pooled gRNA/CAS9 library (10 gRNAs per gene) (Table S1). After transduction with the pooled libraries, we sorted cells by the lowest and highest 5% EGFP/DsRed ratios, which represent the cell populations with decreased and increased tau levels, respectively. Next-generation sequencing of these subpopulations as well as the bulk unsorted cell population identified enriched/depleted shRNAs and gRNAs that target 1,029 genes ( Figure 1A; Table S2). Hits were ranked in order of the number of enriched/depleted shRNAs and gRNAs and their degree of enrichment/depletion compared with bulk data. 21 In parallel, we conducted a genetic loss-of-function screen in a transgenic Drosophila model that expresses human WT tau441 (2N4R) and develops an abnormal ommatidia phenotype in the presence of tau toxicity. We crossed this tau transgenic Drosophila model with 3,898 inducible RNAi alleles targeting 2,645 Drosophila genes orthologous to 4,385 human genes (Table S3A) used for the cell-based screen and looked for modifiers of tau-induced ommatidial degeneration. This Drosophila screen identified 613 genes that modulated tau-induced degeneration, which are orthologous to 1,674 human genes (Figure 1Apanel 2; Tables S3B and 3C).
We prioritized targets based on the overlap between these two screening strategies ( Figure 1A-panel 3), reasoning that crossspecies conservation would likely point to biologically conserved pathways. We selected 213 genes that, upon inhibition, either decreased or increased tau protein levels in human cells and either mitigated or enhanced toxicity in Drosophila, respectively. We also selected 133 genes that regulated tau levels in human cells but lacked Drosophila orthologues or displayed a very strong effect in cells ( Figure 1A-panel 3). These selected hits were further examined for their effects on endogenous tau levels in human 293T cells using ELISA and immunoblot (IB) analysis (Table  S4A). From this validation, 83 hits showed significantly altered tau levels upon genetic KD ( Figures 1B, 1C, S1A, and S1B; Table S1 and S4B). Notably, the data revealed that a number of hits converge on functional pathways, as illustrated for proteins in the Sumo conjugation pathway (Tables S1A and S4B).
Next, we selected 11 genes from our newly identified tau regulators, based on their more robust effect size and/or tractable molecular function, to test their effect on tau levels in the mouse brain. For this, we delivered adeno-associated virus serotype 8 (AAV8)-shRNAs for each target gene via intracerebroventricular (ICV) injection in neonatal mice (postnatal day zero or P0) and used a previously identified tau modulator (TRIM28) as a positive control ( Figures 1D and S2A). By bilateral injection of AAV8-shRNA at 6-10 3 10 10 particles per hemisphere, we achieved suppression of the target gene by %50% in the mouse brain without toxicity, showing no notable abnormality until 3 weeks before harvest ( Figure S2B). The IB analysis of brain homogenates from the injected mice revealed that KD of the 11 candidate tau regulators significantly reduced endogenous tau levels by at least 15% (Figures 1E and 1F; Table S5). Among these in vivo validated tau regulators, we selected ubiquitin-specific protease 7 (USP7), RING-type E3 ubiquitin transferase RNF130 (RNF130), and RING-type E3 ubiquitin transferase RNF149 (RNF149) for more in-depth studies, rationalizing that genes in the ubiquitin pathway will likely reveal some mechanistic insights into regulators of tau degradation.
USP7 stabilizes tau by suppressing CHIP-mediated tau ubiquitination USP7 is a deubiquitinase that cleaves ubiquitin, thereby protecting its substrates from proteasomal or autophagic degradation. Given its function, we tested whether USP7 deubiquitinates (legend continued on next page) ll OPEN ACCESS Article tau and protects it from degradation by performing in vitro ubiquitination assays. We immunoprecipitated (IP) tau in human cells expressing HA-tagged ubiquitin (HA-Ub). IB analysis of tau IP samples revealed that the selective USP7 inhibitor (P5091) significantly increased tau polyubiquitination in a time-and dose-dependent manner in both 293T cells ( Figure 2A) and human embryonic stem cell (hESC)-derived neurons ( Figure 2B). In addition, P5091 negatively correlated with tau levels in human neuroblastoma cells and hESC-derived neurons in a dosedependent manner ( Figure 2C). Conversely, USP7 expression abrogated tau ubiquitination; a catalytically inactive USP7 mutant (H456A; mtUSP7) had no such effect ( Figure 2D). USP7 overexpression significantly increased tau levels, whereas mtUSP7 overexpression had a minimal effect on the tau level ( Figure 2E). These data demonstrate that USP7 stabilizes tau by deubiquitinating it ( Figure 2F).
Our reciprocal co-immunoprecipitation (coIP) results indicate the interaction between USP7 and tau in the mouse brain (Figure 2G). To investigate whether USP7 physically interacts with tau and determine which protein domains mediate this interaction, we conducted coIP assays in 293T cells expressing USP7-FLAG and individually expressing different tau fragments ( Figure 2H). IB analysis of tau IP samples showed that tau's microtubule-binding repeat region (MTBR) is necessary ( Figure 2I) and sufficient ( Figure 2J) to bind USP7. A previous study reported that the CHIP interacts with tau's MTBR and ubiquitinates tau for proteasomal degradation. 17 Because both USP7 and CHIP bind tau's MTBR, 17 we examined whether USP7 competes with CHIP for tau binding. In coIP experiments, the CHIP-tau interaction was reduced by the presence of USP7 protein, while the USP7-tau interaction was less affected by CHIP. This suggests that USP7 negatively affects the CHIP-tau interaction. Interestingly, we found that CHIP-IP or USP7-IP could pull down USP7 or CHIP, respectively, as well as tau ( Figures 3A and 3B), and the USP7 inhibitor (P5091) did not interfere with CHIP-tau interaction (data not shown), showing that CHIP and USP7 interact with each other independently of tau ( Figure 3C). Furthermore, USP7 deubiquitinated CHIP in human cells ( Figure S3A), but the polyubiquitination of USP7 was not altered by CHIP ( Figure S3B). These data indicate that USP7 can regulate both tau and CHIP.
Given our observation that USP7 can deubiquitinate tau and CHIP, we tested whether USP7 stabilizes tau against CHIPmediated tau degradation by examining whether the effect of USP7 on tau ubiquitination and stability is dependent on CHIP. We found that depletion of CHIP completely abolished the accumulation of polyubiquitinated tau ( Figures 3D and 3E) and reduction of tau levels in 293T cells ( Figure 3F) and in mouse primary cortical neurons ( Figures 3G and S3C) typically seen after USP7 inhibition. Supporting this result, expression of USP7 completely abolished the CHIP-mediated tau ubiquitination but not tau ubiquitination by CUL4A (also identified in our screen); de-ubiquitination of tau occurred only after CHIP polyubiquitinated tau ( Figure 3H). Likewise, USP7 suppressed the CHIPdependent lysine 27-and lysine 48-specific linkage polyubiquitination of tau in 293T cells ( Figure S3D). Notably, the depletion of CHIP nullified the effect of USP7 expression on tau levels in 293T cells ( Figure 3I).
These data demonstrate that USP7 stabilizes tau by protecting it from CHIP-mediated ubiquitination ( Figure 3J).

USP7 promotes tau aggregation in cells
Knowing that CHIP plays a critical role in the degradation of hyperphosphorylated tau (p-tau), 16,22 we examined whether USP7 affects the ubiquitination of phospho-tau or tau aggregation. Consistent with previous studies, heavy phosphorylation of tau by the GSK3b kinase significantly increased CHIP-mediated ubiquitination of tau. We found that USP7 deubiquitinated both tau and p-tau ( Figure S4A). However, the degree of USP7-mediated deubiquitination was stronger for p-tau, as reflected in the ratio of Ub-tau/total tau ( Figure S4B).
Next, we assessed the effect of USP7 on tau aggregation in a cellular model of tau seeding. In this assay, 293T cells stably expressing YFP-tagged mutant tau carrying the aggregationdriving P301S mutation were treated with tau species isolated from aged tauopathy mice and assessed for intracellular tau aggregation. We used the detergent extraction method to measure tau aggregation by YFP signal in the absence of soluble tau (Figure S4C). Tau aggregation was increased upon the expression of USP7 but not a catalytically inactive USP7 (mtUSP7), which was reversed by the USP7 inhibitor ( Figures S4D and S5). These data demonstrate that USP7 stabilizes tau, including phosphorylated tau species, thereby promoting tau aggregation in cells.
RNF130 and RNF149 stabilize tau by promoting the ubiquitination and degradation of CHIP RNF130 and RNF149 are E3 ubiquitin ligases that promote the degradation of substrate proteins. 22,23 The fact that shRNA-mediated lowering of either protein reduced tau levels in cells ( Figure 1B) and mouse brains ( Figures 1E and 1F) argued against them being ubiquitin ligases for tau. In fact, we detected neither direct binding nor significant ubiquitination of tau by either RNF130 or RNF149 (data not shown). We reasoned, therefore, that RNF130 and RNF149 modulate tau levels via other mediators. To identify the mediator, we collected affinity-purified RNF130-and RNF140-interacting proteins by immunoprecipitation in 293T cells and submitted the samples for mass spectrometry analysis (IP/MS (D) tau and polyubiquitinated tau from 293T cells expressing either wild-type USP7 or an enzymatically inactivated form (USP7 H456A; mtUSP7). (E) IB images and quantification of endogenous tau levels in 293T cells overexpressing WT USP7 or mtUSP7 at various doses. Three biological replicates were performed for each condition. (F) Illustration proposing that tau deubiquitination is reversed by USP7 inhibition, leading to tau degradation.   Table S6). Among the shared interactors, we chose CHIP and FYN kinase for further studies because both are direct regulators of tau stability that contribute to tau pathology. FYN tyrosine kinases phosphorylate tau at tyrosine 18 (Y18) and enhance tau pathology in cell and mouse models. 24,25 CHIP ubiquitinates tau (especially p-tau) and thereby regulates its degradation and aggregation. 16,22 Using the coIP assay in 293T cells, we confirmed the RNF130-RNF149 interaction, along with the RNF149-FYN, RNF130-CHIP, and RNF149-CHIP interactions ( Figure 4B).
To determine the functional consequences of interactions between CHIP and RNF130 or RNF149, we tested whether RNF130 and RNF149 influence CHIP stability. To do this, we depleted RNF130 or RNF149 with shRNA and used a cycloheximide (CHX) pulse chase assay in 293T cells to assess real-time CHIP degradation. We found that depleting either RNF130 or RNF149 extended CHIP's lifespan, suggesting that both act to destabilize CHIP ( Figure 4C). Conversely, expression of either RNF130 or RNF149 in 293T cells markedly increased CHIP ubiquitination ( Figure 4D). RNF130 and RNF149 overexpression increased endogenous tau levels, but this increase was abolished by depleting CHIP (Figures 4E and 4F). In addition, the effect of RNF130 or RNF149 shRNA on tau levels was also abolished by depleting CHIP in mouse primary neurons ( Figures 4G  and 4H). We therefore propose that RNF130 and RNF149 also regulate tau in a CHIP-dependent manner ( Figure 4I).

RNF149 and FYN stabilize RNF130
Given that RNF130 and RNF149 share 185 interactors (Figure 4A), we hypothesized that they must associate in the same functional complex. A cellular ubiquitination assay revealed that RNF130 and RNF149 are heavily poly-ubiquitinated and that RNF130 ubiquitination was decreased by RNF149 expression, whereas RNF149 ubiquitination was unaffected by RNF130. CHIP had no effect on the ubiquitination of either protein ( Figures 5A and 5B). Notably, we observed that RNF149 delayed RNF130 degradation but not vice versa ( Figure 5C). These data suggest that RNF149 stabilizes RNF130 by suppressing RNF130 ubiquitination.
We explored whether RNF130 and RNF149 regulate tau levels through FYN but found that neither affected FYN ubiquitination or stability in human cells. Instead, CHX chase analysis of 293T cells co-expressing RNF130 or RNF149 with FYN revealed that FYN limits the degradation of both RNF130 and RNF149, extending their respective half-lives ( Figure 5D). Next, we performed IB analysis using phosphor-tyrosine antibodies, and we found that RNF130 and RNF149 are phosphorylated on their tyrosine residues in the presence of FYN ( Figure 5E). These findings demonstrate that FYN can phosphorylate and stabilize both RNF130 and RNF149, thereby indirectly stabilizing tau.
Adult knockdown of either USP7, RNF130, or RNF149 rescues memory deficits in a tauopathy mouse model We next tested our hypothesis that reversing tau accumulation could mitigate deficits caused by overexpression of P301S mutant human tau in a mouse model of tauopathy. We introduced a doxycycline (Dox)-off tetracycline-inducible shRNA expression system into the mouse brain to allow us to reduce target gene expression at a desired time point by switching from a diet containing Dox to a regular diet without Dox ( Figure 6A). We knocked down USP7, RNF130, or RNF149 in adult PS19 mice, which express human mutant tau (1N4R P301S) at five times the level of the endogenous mouse protein and recapitulate many features of human tauopathies. 26 We withdrew the Dox diet in PS19 mice and their WT littermates at 2 months of age, the time point at which the brains are mature. To test the gene KD efficiency, we co-injected P0 mice with AAV containing inducible shRNA and AAV containing tetracycline-transactivator (TTA) and then fed them the Dox diet until 3 weeks of age; shRNA expression was efficiently switched on upon the change in diet ( Figure 6B). PS19 mice displayed the typical tau phosphorylation (pSer202/pThr205) in the brain at 2 months of age, but USP7 and RNF130 suppression reduced the levels of human tau and p-tau, respectively ( Figure 6C).
After 7 months of aging, the injected PS19 and WT littermates were subjected to cognitive behavioral tasks to examine the effect of USP7, RNF130, and RNF149 KD on learning and memory. In the novel object task, PS19 mice were unable to discriminate between novel and familiar objects; KD of USP7, RNF130, or RNF149 restored object discrimination ( Figure 6D). At 8 months of age, PS19 mice exhibited less freezing than the age-matched WT mice in the contextual fear-conditioning task, a test of hippocampus-dependent memory, but again, KD of USP7, RNF130, and RNF149 rescued this behavior ( Figure 6E). Our data show that downregulation of our target genes during adulthood can successfully ameliorate memory deficits in PS19 mice.
Adult knockdown of USP7, RNF130, or RNF149 mitigates tau pathologies After behavioral testing, we collected the mouse brains for biochemical and immunohistochemical analyses. All PS19 mice with target gene KD had less microgliosis illustrated by the reduced Iba1 staining (Figures 6F and S6A) in the cortex, hippocampus CA1, and dentate gyrus (DG) compared with PS19 control mice. KD of target genes also mitigated abnormal astrocytic activation, as demonstrated by the reduced Glial

Article
Fibrillary acidic protein (GFAP) signal in the cortex and hippocampus compared with controls ( Figures 6G and S6B).
Previous reports found that PS19 mouse brains display tau seeding activity, suggesting that misfolded tau present in the brain homogenate can enhance tau aggregation via a prion-like mechanism. 27 To measure tau seeding activity, we utilized the tau-biosensor cells (Tau RD P301S FRET Biosensor cell), in which YFP and CFP form a FRET pair upon the induction of tau aggregation. 27 After transducing the biosensor cell line with brain lysates, we measured tau aggregation based on the abundance of the FRET-positive cells. The tau-biosensor cell line formed numerous aggregates after 24 h of transduction with aged PS19 brain lysates, but no aggregates were detected after transduction with age-matched WT brain lysates ( Figure 6H, top left panels). Brain homogenates from PS19 mice with target gene KD (PS19/USP7 KD, PS19/RNF130 KD, and PS19/RNF149 KD mice) showed markedly reduced proteopathic tau seeding activities as compared with those from PS19 control mice (PS19/Ctrl) ( Figure 6H, top right and lower panel).
Lastly, we examined the levels of different tau species in brain lysates from these mice. At 9 months of age, PS19/KD mice had markedly lower levels of total tau and pathological forms of tau (p T205, p S396/S404 residues, and oligomeric tau) (Figures 7A and  7B). Interestingly, p-tau and oligomeric tau were reduced to a greater extent than total tau. Notably, CHIP levels were significantly higher in the aged RNF130 KD and RNF149 KD mice than those in control PS19 mice ( Figure 7C). We also performed IB analysis of PS19 mice at 5.5 months of age to examine the effects of target gene KD on tau pathology at an early stage before the appearance of memory deficits. We found that the KD of USP7, RNF130, or RNF149 reduced total and p-tau (pS396/S404). CHIP levels were significantly greater in RNF130 KD mice, but RNF149 KD had not affected CHIP levels yet at this age ( Figures S7A-S7C).
Collectively, our data demonstrate that USP7, RNF130, and RNF149 are in vivo regulators of tau levels, play an important role in tauopathy pathogenesis, and represent potential therapeutic targets for decreasing tau levels and mitigating tau pathology ( Figure 7D).

DISCUSSION
The dysregulation of tau proteostasis is clearly involved in taurelated pathology, making modulation of tau levels an attractive therapeutic strategy to tackle AD and other tauopathies. Here, we found that the genetic KD of USP7, RNF130, or RNF149 in the brain reduced pathological tau species (p-tau and tau oligomers) and mitigated memory and learning deficits and other pathological features in a tauopathy mouse model (PS19 mice). Previous studies have tested the therapeutic efficacy of decreasing tau levels in a mouse model of tauopathy and non-human primates via the ICV infusion of antisense oligonucleotides (ASOs) targeting human tau mRNA. 28 However, given the invasive nature of ASO therapies and the need to treat individuals at risk for decades, our study focused on discovering potentially druggable gene products that can regulate tau levels. We reasoned that such regulators would expand our knowledge about tau biology and homeostasis and ultimately lead to the development of non-invasive therapeutic strategies for AD and other tauopathies. Of note, when we utilized a Dox-inducible system to knock down our candidate genes in PS19 mice during the adult stage, we observed robust effects on tau-related pathologies. There are three reasons we used this strategy. First, using a Dox-inducible system to knock down our candidate genes in adult PS19 mice eliminates the chance for secondary non-specific effects and potential compensatory (or pathogenic) mechanisms that might arise due to the loss of the gene's function during development. Second, the use of AAV-shRNAs obviates the multiple timeconsuming and costly breeding schemes that would otherwise be required to generate conditional gene knockouts in tauopathy mouse models, while also allowing us to test the effect of the KD of multiple candidate targets in parallel. Most importantly, the Dox-inducible system allowed us to test the effects of our manipulations after disease onset. Initiation of the KD of USP7, RNF130, or RNF149 in PS19 mice at 2 months after birth, by which time molecular signatures of tau pathology (p-tau) have already appeared ( Figure 6C), exhibited clear therapeutic benefits at later stages of the disease (9 months). Considering, however, that patients with AD normally start receiving treatment after the initial onset of clinical symptoms, it would be worthwhile to test the effects of candidate gene KD upon the first appearance of behavioral abnormalities in the PS19 mice ($6 months of age).
Although USP7, RNF130, and RNF149 have never been associated with each other before, all three converge on CHIP to influence tau proteostasis. This is particularly striking in light of our previous finding that NUAK1 phosphorylates tau at Ser356 to reduce its ubiquitination by CHIP. 15 The fact that two independent screens uncovered genes whose molecular functions converge on CHIP highlights a central role of CHIP in maintaining tau proteostasis in the brain. In fact, several studies have demonstrated a protective function for CHIP in the context of other neurological diseases such as stroke, intracerebral hemorrhage, PD, and polyglutamine diseases [29][30][31][32][33] -which at least in the latter cases involve abnormal protein accumulation. In addition, mutations in the CHIP gene have been found in several human patients with various forms of spinocerebellar ataxia (SCA; SCA16 and SCA48). 34 CHIP upregulation could possibly be neuroprotective in multiple neurological diseases, serving as a promising therapeutic strategy. A therapeutic strategy that involves inhibition of RNF130, RNF149, or FYN to enhance CHIP function could be beneficial in tauopathies, but it would be important first to better understand the molecular network and feedback mechanisms that regulate CHIP.
For example, our results suggest that USP7 can stabilize tau directly while also indirectly destabilizing it by enhancing CHIP (E and F) IB images (E) and quantification of tau and CHIP level (F) showing the endogenous tau and CHIP after RNF130 and RNF149 overexpression with or without CHIP in 293T cells. (G and H) IB image (G) and quantification graph (H) of endogenous tau and CHIP after knockdown of RNF130 and RNF149 with or without CHIP in mouse embryonic primary cultured neurons. (I) Illustration depicting RNF149-and RNF130-mediated regulation of CHIP stability. Data are represented as mean ± SEM (*p < 0.05, **p < 0.01, ***p < 0.0005).
activity. This bimodal regulation is similar to the dynamic regulation mechanism of USP7 reported for p53. In that scenario, USP7 has a greater affinity for the p53 E3 ligase MDM2 than it does for p53, and so it protects MDM2 from auto-ubiquitination, ultimately leading to the proteasomal degradation of p53. 35 Upon external stimulation, however, USP7 shifts from MDM2 to p53, causing MDM2 to be degraded and instead stabilizing p53. 36,37 Our data suggest that USP7 has a stronger preference for deubiquitinating tau than CHIP. Moreover, suppressing USP7 exerted beneficial effects in the tauopathy mice, suggesting tau may be a more favorable substrate for USP7 activity, especially under pathological conditions. An interesting future study would be to elucidate, in both healthy and pathological conditions, how USP7 switches its target preference between tau and CHIP. In AD, tau pathology follows a predictable progression, which is classified by Braak stages and correlates with the disease symptomatology. 38 Transcellular propagation of tau aggregation, in which tau pathology moves from one neuron to the next, underlies disease progression. In our cellular tau aggregation model, USP7 promoted tau aggregation upon proteopathic tau seeding. USP7 could be a key factor in propagating tau pathology.   The discovery that KD of RNF130 and RNF149 reduces tau levels through stabilizing CHIP is quite exciting and opens up the possibility of identifying small molecules that could inhibit their activity and, if safe, be used to lower tau levels. That adult KD of either of these genes in the brain was well tolerated, coupled with the fact that haploinsufficiency for either protein is tolerated in humans (https://gnomad.broadinstitute.org/) bodes well for their potential druggability.
In summary, starting from an unbiased pooled genetic screen, we identified several genes that converge onto CHIP-mediated control of tau proteostasis. Even the modest suppression of these genes in adult tauopathy mice benefited the learning and memory deficits, proving that reducing tau levels could be an attractive therapeutic strategy for tauopathies to delay or prevent disease progression. We are optimistic that future studies will expand our knowledge of the molecular pathways regulating tau proteostasis and provide a stronger foundation for pursuing novel drug targets for AD and other tauopathies.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

INCLUSION AND DIVERSITY
One or more of the authors of this paper self-identifies as an underrepresented ethnic minority in science. We support inclusive, diverse, and equitable conduct of research.

Lead contact
Further information and requests for reagents may be directed to and will be fulfilled by the corresponding author Huda Y. Zoghbi (hzoghbi@bcm.edu).

Materials availability
All materials except AAV and the aged brain tissue are available from the authors upon request. All the Drosophila strains are available from public repositories or from the authors upon request.

Data and code availability
This study did not generate standardized datatypes for public repositories. This paper does not report original code. DOI is listed in the key resources table. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS
Mouse models CFW (MGI catalog #5911387) and CD-1 (IMSR catalog #CRL:22) mice were used for in vivo validation of selected tau modifiers. Timed pregnant CFW or FBV mice were used for embryonic primary cortical neuronal culture experiments. Tau P301S transgenic mice (PS19 line) were bred in our laboratory. USP7-, RNF130-, or RNF149 knockdown mice in PS19 background were generated by intracerebroventricular injection of the recombinant AAV8 harboring shRNA for each given targets. Transgenic offspring for these experiments were generated by mating PS19 males with FVB females. All mice were maintained in a 14 hrs:10 hrs light:dark cycle at 68-72 F and 30%-70% humidity, with standard mouse chow and water ad libitum. All procedures were reviewed and approved by the Baylor College of Medicine Institutional Animal Care and Use Committee in accordance with the guidelines of the US National Institutes of Health.

Druggable genome library
Druggable genome libraries were generated as previously described. 21 Druggable genes were grouped into seven sub-libraries (enzyme, kinase/phosphatase, transcriptional regulation/chromatin, ion channel/transporter, ubiquitin regulation, GPCR, V4-others) based on their functional categorization. The ten best shRNAs targeting each gene were designed using the shERWOOD algorithm. Each shRNA was synthesized as 97mer oligos on chip by Agilent Technologies and integrated into MSCV (murine stem cell virus) retroviral vectors.

METHOD DETAILS
Pooled shRNA screen The cell-based primary screen was performed as previously described. 21 Daoy DsRed-IRES-tau441:EGFP cells were transduced by single copy integration of pooled shRNA (MSCV infection at 0.3 multiplicity of infection [MOI]) and selected with puromycin (1mg/ml, for 72 h). Cells were infected with a representation of 1000 x of each shRNA in biological quadruplicates. At 9 days post-infection, cell populations representing the lowest 5% (decreased tau) and the highest 5% (increased tau) EGFP/RFP ratio were collected using a cell sorter (Sony, SH800) at 100-200X representation. For the control group (bulk group), all transduced cells were collected at 2000X representation. Four experimental replicates were generated by independent infection, With sorted cells, two independent round of PCR were carried out on the genomic DNA to amplify shRNA half-hairpins with adaptor/indices and generate sequencing libraries. After purifying libraries by pippen prep, its DNA concentration, and purity were measured by Qubit, qPCR and Bioanalyzer. Sequencing was performed on the Illumina HiSeq2500 using 150-bp single end reads. Subsequently, we determined the relative enrichment or depletion of each shRNA in sorted groups compared to the bulk group using the CRISPRcloud pipeline. 39 We prioritized genes using the following criteria: (1) shRNAs showing the same direction of change in both sorted groups (low and high) were excluded (conflict score), (2) unidirectional enrichment or depletion in a single sorted group (directional score), (3) the number of significantly enriched or depleted shRNAs targeting individual genes (hit ratio), (4) the degree of enrichment or depletion of a significant shRNA (fold change).

Drosophila screen
For the Drosophila druggable genome screen we identified Drosophila homologs using the Blast algorithm applied to protein sequences, applying a cutoff e-value of E-10 (2645 Drosophila genes corresponding to 4385 human homologs). We obtained a total of 3898 available inducible hairpin RNAs targeting the Drosophila genes in the Vienna Drosophila Resource Center repository (stockcenter.vdrc.at). The [w 1118 ; UAS-Tau (19y), GMR-GAL4)/CyO] was used as previously described 14 and animals were crossed and raised at 28 C. The eyes (ommatidia) were imaged either using scanning electron microscopy as previously described 40 or using a Leica MZ16 imaging system for fresh eyes.
In vitro validation HEK293T cells (30%-40% confluency in the 96-well plate format) were infected by individual lentivirus containing pGIPZ-shRNA at 3-5 MOI in biological quadruplicates. Over a period of 9 days after infection, the transduced cells were selected by puromycin treatment and split into 96-and 24-well plate format. At 6 days post infection, knockdown efficiency was determined by qPCR for a subset of the infected cells to ensure enough target gene suppression. After cell-lysate preparation in 96-well plates, tau protein levels were measured using a commercial tau ELISA kit (Thermo, KHB0042) according to the manufacturer's instructions. Total protein concentrations were measured in parallel using BCA protein assay for loading control and to ensure effects of each shRNA on cell viability; samples with <50% of average protein concentration were excluded. Genes whose knockdown significantly decreased or increased tau protein levels by >15% in R2/3 shRNAs from biological quadruplicates in a comparison to nontargeting shRNA were subjected to the additional immunoblotting assay (p<0.1) in which tau level was compared to nontargeting (NT)shRNA and selected as hits based on same criteria as ELISA assay.

Behavioral testing Novel object recognition
The novel object recognition test was modified from a previous method. 41 On the day of the test, subject mice were pre-habituated in the procedure room for at least 30 min. Individual mice transferred to an evenly illuminated sound-proofed box (light at 150 + 10 lux and background noise at 60+5 dB) that is identical to the operation box, and they were allowed to explore freely for 5 min. After habituation, individual mice were placed in the operation box and were allowed to explore two identical objects that were placed on opposite sides of the box at equal distances from the nearest corners for 5 minutes and then returned to their home cages (training phase). One hour later, trained mice were individually placed back in the same operation box, where one of the two familiar objects was switched to a novel object of the same size but different color and shape. Mice could freely explore the two objects and we recorded the amount of time they spent exploring each object during the 5-minute testing phase. ''Object exploration'' was defined as the time the mouse spent either sniffing an object, touching it with its forepaw, or both. We calculated a preference index for each object during training and testing phases using the following formula: Preference index = ðtime exploring one object = time exploring object pairsÞ 3 100% Preference index during the training phase, which should be 50%:50%, was used as an environmental control to rule out the location effect of the objects.

Contextual fear conditioning
The contextual fear conditioning test was performed with modifications from our previous study. 15 On training day, individual mice were placed in an soundproof operation box (20 cm length x 20 cm width x 35 cm height) and allowed to explore freely for 2 minutes. Three consecutive electrical foot-shocks were delivered to the mice at 1.5 mA for 2 seconds with 1-minute intervals. Foot-shocked mice were then returned to their home cage. The next day, context testing was conducted in the same operation box for 5 minutes by measuring the freezing response of individual mice. The freezing response was scored as the total time mice spent 'frozen' during the test period. Data acquisition, control of stimuli, and data analysis were performed automatically using the Coulbourn/Actimetrics FreezeFrame3 System. Mouse brain sample preparation CFW mice were sacrificed by isoflurane inhalation at 3 weeks post injection. The hippocampus and subregion of cortex above the hippocampus was harvested and immediately frozen on dry ice. PS19 mice and age-matched WT littermate mice were sacrificed by sodium pentobarbital overdose and transcardially perfused with ice-cold PBS. Brains were isolated and split into two hemispheres for either biochemical or histological study. Cortex and hippocampus from the hemisphere were dissected and lysed together. The remaining hemisphere was fixed by 4% paraformaldehyde (PFA) for 24 hours and cryopreserved in 1X PBS containing 30% sucrose at 4 C. Frozen samples were homogenized using an electric pestle (handheld polytron, WPR, 47747-370) in 10x volumes/weight of cold PEPI buffer (1x PBS containing 5 mM EDTA, protease inhibitor cocktail and phosphatase inhibitor cocktail). Part of the homogenate was then diluted 1:1 with RIPA buffer (1x PPS containing 5 mM EDTA, protease inhibitor cocktail, phosphatase inhibitor cocktail, 1% deoxycholate, 1% Triton X-100, and 1% SDS). After mixing well, the supernatant was collected by centrifugation at 15,000 rpm for 20 min. For RNA extraction, 200 ml of homogenate was mixed in 600 ml Trizol LS (Fisher, 10296-010) followed by RNA precipitation using chloroform and ethanol according to the manufacturer's instructions. RNA was further purified using the RNeasy mini kit (QIAGEN).
FRET measurement of seeding activity Tau-seeded transduction of tau RD P301S FRET Biosensor cell line and flow cytometry analysis were conducted as previously described with minor modifications. 27 The biosensor cells were plated in 96-well plates at a density of 20,000 cells/well. After 24 h, when cells reached 60%-70% confluency, we made seed-transduction complexes by combining [15 mL Opti-MEM (Gibco, 31985070) + 0.1À0.4 mL Lipofectamine 2000 (Invitrogen, 11668-500)] with [7.5 mL Opti-MEM + 0.5-2.0 mg protein extract from brain homogenate]. The transduction complexes were then incubated for 30 min at room temperature (RT) and added to cells. After 24 h of seed transduction, cells were dissociated with trypsin (Life Technologies, # 25200072), and resuspended in 1x PBS containing 2 mM EDTA and 3% FBS. Upon excitation by a 405nm laser we measured the FRET signal with the BD LSRFortessaÔ cell analyzer. For each experiment, we recorded CFP and FRET-YFP signals of 5,000-8,000 singlet events at 485/22nm-filter and 525/50 nm-filter, respectively. Seeding activity of given brain homogenates was assessed by the percentage of FRET-positive cell in a bivariate plot of FRET vs. CFP. In this bivariate plot, to assess the number of FRET-positive cells we created a gate by using the FRET-negative signal exhibited by biosensor cell treated with lipofectamine alone and the FRET-positive signal exhibited by 293T cells expressing CFP fused with YFP.
Immunofluorescence and microscopy assay Fixed brains were frozen on dry ice and sagittally sectioned (35 mm thickness) using a sliding microtome (Leica, SM2010 R). Floating sections were stored in the antifreeze buffer [50 mM NaPO4 (pH 7.4), 25 % glycerol, and 30% ethylene glycol (v/v)] at -20 C until use. Sagittal sections located 1-1.5 mm from the midline were selected for staining and analysis based on landmarks in the hippocampus, lateral ventricle, and striatum. A total of 2-3 sections were imaged per animal. Brain sections were immunostained with GFAP (1:2,000; Novus Biologicals, 53809), and Iba1(1:2,000; Wako Chemicals, 019-19741) antibodies at 4 C overnight. Sections were then incubated in Alexa FluorÒ 555 conjugated goat anti-rabbit IgG secondary antibodies (Invitrogen, A32732) and Hoechst (1:2,000, Abcam, ab228551) for 2 hours at RT. Fluorescent images were collected from nonoverlapping fields within the somatosensory cortex (above CA1), hippocampus CA1, and the dentate gyrus. A single optical plane of 0.977 mm in depth was collected in blue (Hoechst) and red (GFAP or Iba1) channels using fluorescent microscopy (Carl Zeiss) at 100 X magnification (895.26 mm x 670.8 mm per field). For the GFAP staining, the whole hippocampus was captured at 50 X magnification (1790.52 mm x 1341.60 mm per field). For representative images, Z-stacked fluorescent images (14 optical images) were collected from the same field of the cortex and hippocampus CA1 region using confocal microscopy (Carl Zeiss) at 200 X magnification (416.80 mm x 416.80 mm).
tau aggregation assay in cell model To assess in vitro tau aggregation upon extracellular tau seeding, seeding potent tau species (''tau seeds'') were purified from 8-10month old PS19 mice through fractionation using ultracentrifugation. Cortical and hippocampal tissues from these PS19 mice were collected and homogenized in cell lysis buffer containing 1x protease and phosphatase inhibitor cocktails on shaker for 30-60 min at 4 C. Supernatant was collected via centrifugation at 5000 rpm for 10 min at 4 C which was further ultra-centrifuged at 200,000G for 40 min at 4 C. Triton X-100-insoluble pellets were then washed with ice-cold lysis buffer three times. The pellet was resuspended in 1X PBS and sonicated (probe sonicator, 30% amplitude, pulsed ten times with 2''/1'' on/off). The concentration of tau was determined by tau ELISA kit (Thermo, KHB0042).
We generated a seed-recipient cell line that stably expresses YFP fused human tau 441 (2N4R) bearing a preaggregation mutant, P301S (HEK293T_tau 2N4R-P301S-YFP), by lentiviral infection and antibiotic selection (Puromycin, 1 mg/ml, 5 days). The seed-recipient cells were plated at 50,000 cells per well in 24-well format plate and were reverse-transfected with plasmid containing cDNA of USP7, USP7 mutant, or FYN using TransITÒ-293 (Mirus, MIR2704) transfection reagent according to the manufacturer's protocol. 24 hours later, at 60%-70% confluency, transduction complexes were made by combining [15 mL Opti-MEM (Gibco) + 0.5 mL Lipofectamine 2000 (Invitrogen)] with [15 mL Opti-MEM + proteopathic seeds]. The transduction complexes were then incubated 30 min at room temperature (RT) and added to cells. To examine the effects of USP7 inhibition on tau aggregation, we added DMSO or USP7 inhibitor (P5019, 10 mM) to cells 6 hours before transfection. At 48 hours after tau seeding, cell medium was replaced with 2 % Triton X-100 in 1X PBS and cells were incubated for 1 min at RT. Then cells were added with the same volume of 2 % Hexadecyltrimethylammonium bromide (HDTA, Sigma, H5882) and further incubated for 10 min at RT to remove soluble proteins. Cells were then fixed with 4% PFA for 20 min at RT. Tau YFP aggregates in each well were captured by cell imaging microplate reader (BioTek, Cytation5 Cell imaging Reader) at 25X magnification (3402x3001 pixels, 13.2 x 11.6 mm per field) and analyzed using Image J.

Viral production
Recombinant virus containing shRNA or ORF was produced in low-passage HEK293T cells via triple-transfection at 80%-90% confluency, using Lipofectamine 2000 (Invitrogen, 11668019) or TransITÒ-293 (Mirus, MIR2704) transfection reagent according to manufacturer's instructions. For pooled shRNA libraries, viral packing was done by transfection with plasmid DNA mixture of pMSCV-shRNA, Gag/Pol, and VSV-G at 7:1:1 ratio (total 45 mg) in a 150 mm culture dish format. For lentiviral packing, cells were transfected with a 4:3:1 ratio of viral shuttle vector (pGIPz, pLenti, pLOC), psPAX2, and pMD2.G. Medium was collected at 48 and 72 hours posttransfection. The viral suspension was then frozen and kept at -80 C until needed. Lentivirus was concentrated 50-fold using Lenti-X concentrator (Clontech, 631231) in case a higher titer and more purified virus was required. Recombinant AAV8 was produced using a triple transfection with pAAV-, capsid-, and helper plasmid in a 150 mm dish format and the resultant virus was purified and concentrated on an iodixanol step gradient, as previously described. 42,43 To increase the yield of virus particles, cell-associated and medium-containing secreted AAVs were collected separately at 72 hours post-transfection and combined before purification. Otherwise, the collected medium was concentrated 50-fold using AAVancedÔ Concentration Reagent (SBI, AAV100A-1) and used for in vitro applications.

Viral titering
Retroviral and lentiviral titering was performed on Daoy or HEK293T cells using a serial dilution of the virus according to Open Biosystems pGIPZ method (Thermo Fisher Scientific) as previously described. 21 The titer of pooled shRNA containing retrovirus was determined using flow cytometry. Daoy cells were plated on 6-well plates at 2 3 10 4 cells per well. 4 hours later cells were infected with three different amounts of viral suspension to achieve around 5, 10 or 30% infectivity based on the previously determined viral titer or estimated average viral titer (53 10 5 genome copy number [GC]/ml). 24 hours post-infection entire cells in each well were transferred into a 100 mm culture dish and cultured in fresh media containing 1m/ml puromycin (puro+ media) for at least 72 hours until the complete cell-lethality in uninfected group. As a control group, cells were cultured in puro media. The number of surviving cells from each culture dish were counted using cell flow cytometry, and viral titers were estimated using the following formula: GC ml = ð# Cells in puro + media = # Cells in puro À mediaÞ 3 À 2 3 10 4 Á viral stock vol: ðmlÞ On average, the viral titer of retroviral suspension or lentiviral suspension corresponded to 2-10 3 10 5 or 2-10 3 10 6 transduction units/ml. Purified AAV viruses were titered using a quantitative-PCR based titering method as previously described. 42,43 In vivo gene delivery The knockdown efficiency of each shRNA in pAAV was tested in Neuro2A cell as previously described. 21 The two most potent shRNAs targeting each gene were packed into AAV serotype 8 (AAV8). AAV8 harboring a shRNA expression cassette was bilaterally infused into mouse brain at 4-10 3 10 10 viral genomes per hemisphere using intracerebroventricular (ICV) injection at post-natal day 0 as previously described. 44,45 Transgenic offspring were generated by mating PS19 males with FVB females; pups were injected with the mixture of AAV8 containing TRE-YFP-miRE-miRE (8 3 10 10 viral particles/hemisphere) and the TA2S-T2A-tTA2S expression cassette (2.0 3 10 10 viral particles/hemisphere) using ICV injection at post-natal day zero (P0). Injected PS19 mice and WT littermates were fed with a doxycycline (DOX)-containing diet to prevent shRNA expression until 2 months post-injection. At 2 months, mice were switched to a DOX-free diet to induce shRNA expression until 5-9 months of age for behavioral and pathology assessments.
Cell culture, transfection, infection, and drug treatment All cell lines were cultured according to the manufacturer's protocol. Cells were transfected with plasmids encoding cDNA or short hairpin RNA using Lipofectamine 2000 (Invitrogen,11668019) or TransITÒ-293 (Mirus, MIR2704) transfection reagent, according to the manufacturer's protocol, and incubated for 48 hours followed by cell lysate preparation. Cells were infected with lentivirus expressing ORF at 3-10 MOI in the various cell culture size formats. After 24 hours of infection, cells were incubated with the appropriate antibiotics at least for 72 hours. Cells were further maintained until additional treatment or harvest. HEK293T and BE2C cells were