Targeted de-repression of neuronal Nrf2 inhibits α-synuclein accumulation

Many neurodegenerative diseases are associated with neuronal misfolded protein accumulation, indicating a need for proteostasis-promoting strategies. Here we show that de-repressing the transcription factor Nrf2, epigenetically shut-off in early neuronal development, can prevent protein aggregate accumulation. Using a paradigm of α-synuclein accumulation and clearance, we find that the classical electrophilic Nrf2 activator tBHQ promotes endogenous Nrf2-dependent α-synuclein clearance in astrocytes, but not cortical neurons, which mount no Nrf2-dependent transcriptional response. Moreover, due to neuronal Nrf2 shut-off and consequent weak antioxidant defences, electrophilic tBHQ actually induces oxidative neurotoxicity, via Nrf2-independent Jun induction. However, we find that epigenetic de-repression of neuronal Nrf2 enables them to respond to Nrf2 activators to drive α-synuclein clearance. Moreover, activation of neuronal Nrf2 expression using gRNA-targeted dCas9-based transcriptional activation complexes is sufficient to trigger Nrf2-dependent α-synuclein clearance. Thus, targeting reversal of the developmental shut-off of Nrf2 in forebrain neurons may alter neurodegenerative disease trajectory by boosting proteostasis.


Introduction
The transcription factor Nrf2 is a widely expressed stress-responsive regulator of several aspects of homoeostatic physiology. Under basal conditions it is held in the cytoplasm and targeted for ubiquitin-mediated degradation by its inhibitor Keap1. However, in response to signals, including oxidative stress and heavy metal toxicity, its interaction with Keap1 is altered, preventing degradation, and allowing it to translocate to the nucleus 1 . Here, it activates expression of a battery of genes containing antioxidant response elements in their promoter, including antioxidant pathway genes, and xenobiotic detoxification genes [1][2][3][4] .
Importantly, Nrf2 has also recently been found to be capable of regulating proteostasis, both directly and indirectly, through a number of gain-and loss-of-function studies. Nrf2 activity can be manipulated by conventional overexpression or knock-down, and endogenous Nrf2 dependent gene expression can also be activated by disrupting the ability of Keap1 to promote Nrf2 degradation, using a number of small molecules, most commonly acting via the electrophilic modification of redox sensitive Keap1 cysteine residues 1,5,6 . This pharmacological activation approach has been shown in pancreatic ß-cells, mouse embryonic fibroblasts and breast cancer cells to repress cytotoxicity and unfolded protein response (UPR) over-activation in response to ER stress inducers, at least in part through maintenance of ER redox balance and disulphide chemistry [7][8][9] . In addition to modulating the UPR, Nrf2 controls other key arms of cellular proteostasis machinery. In liver cells, fibroblasts and human ES cells, Nrf2 has been shown to control capacity of the ubiquitin proteasome system (UPS). In addition, Nrf2 boosts the macroautophagy pathway, as evidenced from studies in HeLa cells and MEFs 10,11 .
In the brain, many neurodegenerative diseases are associated with neuronal protein aggregate accumulation, such as α-synuclein (Parkinson's disease, Lewy Body dementia), TDP-43 (motor neuron disease, frontotemporal dementia (FTD)) and Tau (Alzheimer's disease, FTD), and both genetics, molecular pathology and animal studies point to defects in proteostasis in pathoprogression [12][13][14] . At first glance, this suggests that small molecule activators of endogenous Nrf2 could be employed to limit aggregate accumulation by controlling proteostasis machinery via mechanisms such as those described above in non-neuronal cells. However, forebrain neurons are highly unusual in that they express very low levels of Nrf2, due to epigenetic repression of the Nrf2 promoter early in development, and the little Nrf2 that remains is highly unstable [15][16][17] , suggesting that endogenous neuronal Nrf2 may be insufficient to support a proteostatic response, and that novel strategies may be required to overcome the transcriptional repression of this key cytoprotective gene. Here we have devised novel strategies to reverse neuronal Nrf2 repression, and consequently drive endogenous Nrf2-dependent α-synuclein clearance.

Results
Activation of endogenous Nrf2 can drive α-synuclein clearance in astrocytes but not neurons To establish whether activators of endogenous neuronal Nrf2 could promote protein aggregate clearance in neurons, we employed a model of cortical neuronal α-synuclein accumulation, a core pathological signature in Lewy Body Dementia. We chose this model as it had been recently shown that ectopic over-expression of Nrf2 in cortical neurons could modulate proteostasis and promote α-synuclein clearance 18 . Thus, while this approach drives Nrf2 to unphysiological levels, the study did reveal that neurons possessed the downstream machinery to clear α-synuclein in response to elevated Nrf2.
Mixed neuron/astrocyte cortical cultures (~10% astrocytes) were transfected with an α-synuclein-encoding vector. We observed that the quantity of α-synuclein plasmid, and the level of α-synuclein expression detected (by immunofluorescence) was approximately linearly related, confirmation that the detection approach was quantitative and able to reveal increases or decreases in expression (Fig. S1a,b). The expression of α-synuclein was cell-wide, as is known to be the case when α-synuclein is pathologically over-expressed, such as in the Thy1-αSYN mouse model, or in human Parkinson's disease patients with SNCA gene triplication [19][20][21][22] . In these α-synuclein expressing cells, instead of over-expressing Nrf2 18 , cells were treated with the electrophilic Nrf2 activator tBHQ for 24 h, after which accumulated α-synuclein was quantified (blind) in both neurons and astrocytes.
We analysed both cell types since astrocytes express higher levels of Nrf2 than neurons 15,17 and we therefore hypothesized that astrocytes would respond more robustly to tBHQ than neurons. We found that tBHQ treatment (10 µM) significantly reduced α-synuclein accumulation in astrocytes, but failed to reduce α-synuclein in cortical neurons (Fig. 1a-d). We confirmed that this effect of tBHQ in astrocytes was mediated by Nrf2: tBHQ was found to have no effect on α-synuclein in astrocytes derived from Nrf2 -/mice (Fig. 1c,d). Thus, αsynuclein clearance induced by endogenous Nrf2 activation can be achieved by tBHQ treatment of astrocytes but not neurons. It is important to note that these experiments are simply a proof-of-concept that Nrf2 activation has the potential to clear α-synuclein in cells where Nrf2 is activatable.

tBHQ induces Nrf2-independent induction of Jun and neurotoxicity
To investigate the transcriptional responses to tBHQ associated with these marked differences in α-synuclein clearance capacity in neurons vs astrocytes, we first performed RNA-seq on the mixed astrocyte-neuron cultures (described above) treated ± tBHQ (10 µM) and observed transcriptional induction of 25 genes (> 2 FPKM, > 1.5fold, P_adj < 0.05, Fig. 1e, Supplemental Table S1). While this list contains many known Nrf2 target genes, we wanted to determine whether any non Nrf2-mediated responses were apparent. We cross-referenced these genes to RNA-seq data we have obtained from astrocytes over-expressing Nrf2 (FAC-sorted from the GFAP-Nrf2 mouse 23 ). Of the 22 genes that were expressed in this GFAP-Nrf2 RNA-seq data set >2 FPKM, 20 are significantly up-regulated in GFAP-Nrf2 astrocytes, relative to wild-type (p_adj<0.01, n = 5, Fig. 1f, Supplementary  Table S1). Moreover, there exists independent published evidence that all of the 20 genes are direct Nrf2 targets [24][25][26][27][28] . We also confirmed that the tBHQ-induced expression of a selection of genes-Hmox1, Srxn1, and Slc7a11, was strongly abrogated in Nrf2-deficient cultures (Fig. 1g). Of note, the tBHQ-induced gene Jun was not significantly up-regulated in GFAP-Nrf2 astrocytes, indicative of Nrf2independent induction, and in agreement with the absence of published evidence that Jun is an Nrf2 target gene.
Jun is induced in neurons by multiple insults, including oxidative/electrophilic stress, and induction can contribute to neuronal apoptosis 29 , so we considered the possibility that Jun induction in neurons by tBHQ treatment represents the early signs of a neurotoxic stress response to the electrophilic tBHQ. We observed that concentrations of tBHQ ≥ 10 µM killed neurons in a dosedependent manner (Fig. 2a,b) which was inhibited by the pan-caspase inhibitor qVD-Oph, suggestive of apoptoticlike cell death (Fig. 2c). Moreover, tBHQ-induced neuronal death was inhibited by TAM67 30 , a dominant negative form of Jun (Fig. 2d,e). The electrophilic/prooxidative nature of tBHQ may be a cause of neuronal apoptosis because tBHQ treatment promoted glutathione depletion (Fig. 2f), and both depletion, and neuronal death could be rescued by of cells with a cell permeable form of the (nucleophilic) antioxidant glutathione ( Fig. 2d-f), suggesting that oxidative stress is a key factor (though ROS levels were not measured). Thus, not only do cortical neurons fail to mount any Nrf2-dependent transcriptional or proteostasis response to tBHQ, they are vulnerable to neurotoxicity when exposed to tBHQ, triggered via an Nrf2-independent, Jun-dependent mechanism.
The triterpenoid CDDO TFEA can drive α-synuclein clearance in neurons after epigenetic derepression of Nrf2 In order to drive endogenous Nrf2-dependent α-synuclein in cortical neurons, it is clear that there is first a need to overcome the lack of Nrf2 mRNA expression. We recently showed that neurons can be rendered more responsive to Nrf2 activators by first de-repressing the Nrf2 promoter by treatment with histone deacetylase (HDAC) inhibitors 17 . Treatment of neurons with the HDAC inhibitor trichostatin A (TSA) rescues histone H3 hypo-acetylation at the Nrf2 promoter and induces Nrf2 mRNA expression 17 (confirmed in Fig. 3a), which enables them to respond to tBHQ by up-regulating Nrf2 target genes 17 . We followed this protocol in α-synucleinexpressing cortical neurons: pre-treating them with TSA, followed by treatment with tBHQ. However, we observed that even pre-treatment of cortical neurons with TSA failed to enable neurons to mount an α-synuclein clearance in response to tBHQ treatment (Fig. 3b). We hypothesized that the pro-oxidant, electrophilic nature of tBHQ ( Fig. 2) makes it sub-optimal as a neuron-focussed Nrf2 activator. This is relevant because oxidative stress can exacerbate α-synuclein aggregation 31,32 and so the off-target effects of tBHQ may counter-act any on-target effects. We therefore wanted to determine whether alternative Nrf2-activating compounds had a more favourable toxicity profile, or whether all Nrf2-activating compounds, regardless of potency, triggered Nrf2 activation and neurotoxicity at similar concentrations.
We compared tBHQ with another type of Nrf2 activator, the triterpenoid 1[2-Cyano-3,12-dioxool-eana-1,9 (11)-dien-28-oyl] trifluoroethylamide (CDDO TFEA ), also referred to as RTA-404 33 . We chose CDDO TFEA as it is more potent than tBHQ 33 , potentially reducing the scope for off-target effects. We first confirmed that, like tBHQ, CDDO TFEA could promote α-synuclein clearance (by both immunofluorescence and western blot analysis, Fig. 3c,d) and this was not observed in Nrf2-deficient astrocytes (Fig. 3c). We then assessed the dose dependency of both tBHQ and CDDO TFEA -induced activation of Nrf2dependent responses, in comparison to their induction of an Nrf2-independent Jun stress response. We performed the comparison on both astrocytes, able to mount Nrf2-dependent and -independent responses, as well as neurons, which only mount an Nrf2-independent stress response. Using induction of the Nrf2 target gene Srxn1 as a readout, we observed its induction in astrocytes by CDDO TFEA treatment at concentrations around 200 fold lower than tBHQ (Fig. 3e), and as expected, no induction in neurons (Fig. 3f). Importantly however, Jun was not comparably induced: while 250 nM CDDO TFEA was equally effective at inducing Srxn1 in astrocytes as 50 µM tBHQ, it neither induced Jun in neurons or astrocytes (Fig. 3g,h), nor did it trigger neurotoxicity (Fig. 3i).
We concluded that CDDO TFEA has a more favourable toxicity profile than tBHQ, so we investigated whether it could induce clearance of over-expressed α-synuclein after derepression of neuronal Nrf2 by TSA pretreatment. We observed that, while TSA or CDDO TFEA alone failed to promote α-synuclein clearance, a combination of TSA followed by CDDO TFEA successfully achieved this in Nrf2 +/+ , but not Nrf2 -/neurons (Fig. 3j, k), and also induced classical Nrf2 target genes Hmox1 and Nqo1 (Fig. S2a,b). Thus, epigenetic de-repression of Nrf2 in cortical neurons enables Nrf2-dependent proteostasis processes to be pharmacologically induced by CDDO TFEA , a non-stress inducing Nrf2 activator. We next wanted to determine the influence of this intervention on endogenous α-synuclein, which is primarily presynaptic in localisation. We therefore repeated the conditions used in Fig. 3j, but studied endogenous α-synuclein, and its pre-synaptic localisation, using synapsin as a presynaptic marker. Detection of endogenous α-synuclein required longer exposure time than detection of overexpressed α-synuclein, but its synaptic co-localisation with synapsin was apparent (Fig. 3l), its co-localisation with synapsin was unaffected by TSA or CDDO TFEA alone or in combination (Fig. 3m, upper), and overall fluorescence intensity unaffected (Fig. 3m, lower), supporting the concept that this intervention preferentially clears inappropriately accumulated α-synuclein, rather than endogenous synaptic α-synuclein.
As an alternative to studying the capacity of TSA + CDDO TFEA to clear over-expressed α-synuclein overexpression, we used a system of exposing α-synuclein preformed fibrils (PFFs) to neurons for 9 days to promote αsynuclein aggregation. We first studied α-synuclein aggregate presence using an aggregate conformationspecific antibody by immunofluorescence and observed (see figure on previous page) Fig. 3 The triterpenoid CDDO TFEA can drive α-synuclein in neurons after epigenetic derepression of Nrf2. A Astrocyte-free neuron cultures were treated ± 1 µM TSA for 16 h and Nrf2 mRNA expression analysed. *P = 0.035 (Student's t test, n = 3). B Neuronal cultures were transfected on DIV3 with eGFP, plus vectors encoding either α-synuclein or a ß-globin control. 5 days post-transfection, cells were treated ± TSA (1 µM) for 8 h, and subsequently (where indicated) with 10 µM tBHQ for 24 h. Cells were then fixed and processed for α-synuclein immunofluorescence (n = 3-5). C Experiment performed as in Fig. 1c except that CDDO TFEA (250 nM) was employed instead of tBHQ. *P = 0.033, 2-way ANOVA + Sidak's post-hoc test (n = 4). D Experiment performed as in Fig. 3c except that western blotting was performed to study α-synuclein levels, rather than immunofluorescence, and only WT astrocytes studied. *P = 0.0003, 2-way ANOVA + Sidak's post-hoc test (n = 5). Cortical astrocyte cultures (E) or astrocyte-free neuronal cultures (F) were treated with different concentrations of tBHQ or CDDO TFEA for 8 h, RNA harvested and expression of the Srxn1 measured by qPCR, normalized to Rpl13a. that TSA + CDDO TFEA caused a reduction of fibril presence by~50% (Fig. 4a,b). This reduction is mirrored when analysing phosphorylation of alpha-synuclein on serine 129, and event that occurs preferentially on alpha-synuclein aggregates (Fig. 4c,d). We also performed western blot analysis of triton-insoluble, oligomeric alphasynuclein in PFF-exposed neurons, and also observed a reduction by TSA + CDDO TFEA treatment (Fig. 4e,f). We also analysed levels of endogenous alpha-synuclein by western blot. Endogenous alpha-synuclein is extractable by Triton detergent and runs as a monomer on a western blot. TSA + CDDO TFEA treatment had no influence on Triton-soluble alpha-synuclein monomer levels (Fig. 4g,h). Also of note, exposure of neurons to PFFs also had no effect on Triton-soluble alpha-synuclein monomer levels, consistent with their Triton-insoluble nature (Fig.  4g,h). Collectively these data do not support the hypothesis that endogenous alpha-synuclein is depleted by TSA + CDDO TFEA over the course of the experiment, but we acknowledge that studying later timepoints in the future may be desirable.
dCas9-based transcriptional activation complexes targeted to the Nrf2 promoter induce α-synuclein clearance Finally, we investigated whether a more targeted activation of the neuronal Nrf2 promoter could induce αsynuclein clearance. We employed a synthetic transcription factor complex based on nuclease-defective Cas9 fused to the viral transactivation domain VP64, in combination with three specifically designed sgRNAs to target the complex to the Nrf2 promoter 34 . In addition, the sgRNAs contained a minimal hairpin aptamer, appended to both the sgRNA tetraloop and stem loop 2, which binds to dimerized bacteriophage MS2 coat proteins (hereafter, sgRNA(MS2)). Co-expression of a fusion protein comprised of MS2 and the p65 trans-activating subunit of NF-kB, leads to its recruitment to the dCas9-VP64/sgRNA complex, and further strengthens the transactivating power of the complex 34 . We confirmed that co-expressed dCas9-VP64 and MS2-p65 could activate an Nrf2 promoter-driven luciferase reporter, when co-expressed with Nrf2 promoter targeting sgRNA(MS2), but not with control sgRNA(MS2) (Fig. 5a). Furthermore, we found that this synthetic Nrf2-activating dCas9-transactivator complex could prevent the accumulation of α-synuclein (Fig. 5b,c). To determine whether the prevention of the accumulation of α-synuclein by the Nrf2-activating dCas9-transactivator complex was due to any off-target effects, we repeated the experiment in cortical neurons prepared from Nrf2 -/mice. We observed no effect of the synthetic Nrf2-activating dCas9-transactivator complex on α-synuclein accumulation in Nrf2 −/− neurons, confirmation of the Nrf2-dependency of the process (Fig. 5b,  c). Thus, the targeted activation of the neuronal Nrf2 promoter using a synthetic dCas9-transactivator complex is sufficient to induce an Nrf2-dependent proteostasis response.

Discussion
Research into the protective effects of Nrf2 activation in the CNS has hitherto mainly focussed on its activation in astrocytes. Activation of astrocytic Nrf2 can promote neuroprotection via a non-cell-autonomous mechanism due to enhanced production and release of glutathione, which is then utilized by nearby neurons to enhance their own antioxidant defences 4,35-38 . Nrf2 activators can also repress brain inflammatory deregulation through modulating macro-and micro-glial responses, mechanisms thought to be at the heart of Multiple Sclerosis drug Tecfidera's (dimethyl fumarate) mechanism of action 39,40 , although Nrf2-independent actions of Tecfidera have also been reported 41,42 . However, while the activation of Nrf2 in glia can promote neuroprotection indirectly via inhibition of inflammatory or redox dysregulation in the brain, a significant proportion of Nrf2-dependent genes can only exert their cytoprotective effects cell-autonomously. The influence of many Nrf2 regulated detoxification enzymes, cellular oxidoreductases, molecular chaperones, proteostasis machinery and other response genes are likely to only be cytoprotective in the cell in which they are expressed 43 . As such, the inability of cortical neurons to express sufficient Nrf2 to support a response to Nrf2 activators (Fig. 1h, Supplemental Table S2, 17) suggests that they lack an important adaptive homoeostatic pathway. The biological reason for the repression of Nrf2 expression in cortical neurons early in development appears to be to facilitate their maturation by providing a more flexible redox environment to potentiate key signalling pathways 17 . However, the resultant lack of Nrf2 into maturity potentially renders neurons vulnerable to particular insults, and reliant on support from surrounding glia, which exhibit protective Nrf2 activation following mild trauma such as preconditioning episodes of ischaemia 44,45 , unlike neurons which have distinct transcriptional and post-transcriptional responses to ischaemia 46 . The findings that artificial Nrf2 over-expression in neurons can protect them against diverse insults, including oxidative and ethanolic stress, amyloid-induced deficits, and nerve crush-induced injury 47-50 as well as α-synuclein clearance 18 certainly suggests that Nrf2 activation in neurons may be beneficial. However, this must be viewed in the context of the challenges associated with activating an Nrf2 response in cells where Nrf2 levels are so low, due to transcriptional repression, that Nrf2 target genes are not activated by tBHQ (Fig. 1h). Of note, it is apparent that some Nrf2 target genes can be induced in neurons by synaptic activity by Nrf2-independent routes, contributing to the general protective effects of neuronal activity [51][52][53][54][55] . However, the complement of Nrf2 target genes induced in this way is far from comprehensive, and insufficient to fully compensate for Nrf2 hypofunction. Another question is whether derepression of Nrf2 in mature neurons is safe, or whether continued Nrf2 repression is needed. Of note, viral overexpression of Nrf2 in neurons of the hippocampus ameliorates deficits in an Alzheimer's disease model 48 , suggesting that Nrf2 de-repression could be well-tolerated.
Our studies suggest that, while Nrf2 activators alone cannot induce Nrf2-target gene expression or α-synuclein clearance in cortical neurons, epigenetic de-repression of Nrf2 prior to Nrf2 activator treatment can achieve this (though it would be desirable in the future to study other neuronal types affected by α-synuclein aggregates such as midbrain dopaminergic neurons). Since α-synuclein aggregates can be degraded by both autophagic clearance and proteosomal degradation, and their formation can be prevented in the first place by molecular chaperones and the UPR 56 , several mechanisms may contribute to αsynuclein clearance following Nrf2. For example, Nrf2 antagonizes UPR-associated cytotoxicity and associated CHOP10 expression 7 . Analogous results have been obtained in mouse embryonic fibroblasts, where targeted deletion of Nrf2 sensitizes them to tunicamycin-induced cytotoxicity 8 . Nrf2-driven gene expression can repress ER stress and UPR through the control of glutathione biosynthesis and recycling enzyme genes, essential for maintaining disulphide chemistry in the ER 9 . In another study, Nrf2 target gene Gpx8, a KDEL-motif containing ER-localised glutathione peroxidase, was seen to repress UPR activation. Keeping an appropriate redox balance is of direct relevance to α-synuclein aggregation, since oxidizing conditions promote α-synuclein aggregation 57,58 , potentially via oxidative or nitrative modification of αsynuclein 59,60 . In addition to modulating the UPR, Nrf2 can control other key arms of cellular proteostasis machinery, the UPS and also autophagic protein clearance. Nrf2 activators promote UPS activity through the regulation of multiple proteasome subunit genes in liver cells, fibroblasts and human ES cells, including Psma and Psmb genes 61,62 , Similarly, Nrf2 can modulate macroautophagy in part by regulating expression of cargo recognition gene p62/SQSTM1, as well as genes involved in autophagy initiation, autophagosome formation, elongation and clearance 10,11 . Of course, Nrf2 derepression via HDAC inhibition may not be feasible as a therapeutic strategy given the non-specific nature of this intervention. On the other hand that Valproate, which inhibits Class I/ II HDACs (among other things) like TSA, is a tolerated drug prescribed for certain CNS disorders (epilepsy, bipolar disorder, migraine) and so may achieve Nrf2 derepression without damaging side effects. This warrants further investigation.
Our study has also highlighted the importance of the Nrf2 activating strategy employed. The Nrf2 promoter must be first de-repressed, such as with a histone deacetylase inhibitor, prior to an Nrf2 activator being employed (Fig. 3a). The classical Nrf2 activator tBHQ, while effective in driving Nrf2-dependent gene expression, caused off-target effects in neurons, promoting glutathione depletion and oxidative neurotoxicity associated with Nrf2-independent Jun induction. The electrophilic nature of tBHQ, important for modifying redox-sensitive cysteine residues of Keap1 in order to activate Nrf2, is also likely responsible for the neurotoxicity, since it was repressed by nucleophilic antioxidant glutathione (Fig. 2d,  e). Jun transcription is known to be regulated by JNK, whose upstream activators include redox sensitive signalling molecules such as ASK1 63 .
Neuronal vulnerability to tBHQ contrasts with other cell types, and may be due to their relatively weak intrinsic antioxidant defences, itself a consequence of Nrf2 hypofunction 17 . However, the relationship between on-and off-target effects of Nrf2 activators appears to be compound-specific, since we found that CDDO TFEA could activate Nrf2 at concentrations that did not induce a stress response in neurons. Their distinct molecular structure and chemistry of these two molecules must underlie their differences. Electrophilic compounds of different structures may have different relative affinities for electron donor groups. It is possible that CDDO TFEA preferentially reacts with key Keap1 cysteine residues compared to other "off-target" nucleophiles whose oxidation leads to non-specific oxidative stress and Jun activation. However in the absence of any guiding principles to explain these differences, Nrf2 activators should be assessed for their on-vs. off-target effects, and no electrophilic Nrf2 activator is likely to be completely devoid of off-target effects. CDDO TFEA , also referred to as RTA-404, is structurally highly related to another synthetic triterpenoid RTA-408 (trade name omaveloxolone), both of which are licensed to Reata Pharmaceuticals. RTA-408 (see Fig. 3) has extensive human and primate safety and pharmacokinetic data, both when taken orally and topically applied [64][65][66][67] and is in clinical trials for Friedreich ataxia and mitochondrial myopathy 68 , with positive Phase 2 data recently reported for Friedreich ataxia 69 . In contrast, tBHQ is a food additive used as a preservative (E319) though levels are strictly limited and several studies point to potential carcinogenic or cytotoxic effects, despite its efficacy as an Nrf2 activator [70][71][72][73] .
Other strategies for reducing systemic side effects of electrophilic Nrf2 activators are also being developed, such as the use of pro-drugs such as carnosic acid, whose electrophilic nature is only exposed upon oxidative modification, and thus are only active in cells experiencing oxidative stress, and have been found under certain circumstances to induce Nrf2 activity in neurons 74 As an alternative to the use of small molecules to first derepress, then activate Nrf2, we show that targeting of a dCas9-based transcriptional activation complex to the Nrf2 promoter is sufficient to drive Nrf2-dependent α-synuclein clearance (Fig. 5). Two very recent studies have shown the potential for dCas9-based targeted gene activation to alter neuronal phenotype. Fragile X Syndrome neurons have been rescued through targeted demethylation of the FMR1 gene 75 , and neuronal differentiation has been promoted in stem cells through targeted, inducible activation of NEUROG2 76 . Ours represents a third example of what is likely to represent a useful and flexible approach to manipulate neuronal properties via gene activation. A theoretical advantage of the dCas9-based systems is their selectivity and lack of off-target effects 77 , and our demonstration that Nrf2-promoter targeted complexes drive α-synuclein clearance in Nrf2 +/+ , but not Nrf2 −/− neurons represents strong evidence that they are acting specifically via Nrf2 induction. As gene therapy technologies develop, these or similar approaches may become a viable therapeutic option for controlling Nrf2 activity in neurons. As a master regulator of antioxidant, detoxification, and proteostasis genes, the effective regulation of Nrf2-driven gene expression has the potential to antagonize multiple pathways driving neurodegeneration.

Immunohistochemistry and imaging
All imaging was performed using a Leica AF6000 LX system and DFC350 FX digital camera. For α-synuclein assays, transfected cells were stimulated as indicated, then immunofluorescence performed essentially as described 81 . Briefly, cells were fixed with 4% paraformaldehyde for 20 min at room temperature (21°C), washed with PBS, then cell membranes were permeabilised by 0.5% NP40 (Life Technologies) treatment for 5 min. Neurons were subsequently incubated with mouse anti α-synuclein (BD Biosciences) at 1:1000 dilution for 3 h at room temperature, followed by Cy3-conjugated donkey anti mouse secondary antibody (1:250 dilution, 2 h at room temperature; Jackson ImmunoResearch) and FITCconjugated goat anti GFP antibody (1:500 dilution, 2 h at room temperature; Abcam). A glass coverslip was then mounted using DAPI-containing Vectashield (Vector Labs). For any single experiment, exposure time is constant for all pictures and conditions, and is set low enough such that no pixel in any cell is saturated. We performed an experiment whereby we altered the amount of αsynuclein plasmid in our transfection mixture, keeping the amount of eGFP co-transfection plasmid constant, and measured the level of α-synuclein expression by immunofluorescence. This revealed an approximately linear relationship between the quantity of α-synuclein plasmid, and the level of α-synuclein expression detected (Fig. S1a, b). Images were chosen for analysis based on GFP-signal not α-synuclein, and were analysed with the experimenter blind to condition using ImageJ software, measuring integrated density of α-synuclein immunofluorescence of cell soma. For these experiments we use morphology to distinguish astrocytes from neurons, but used neuronal and astrocytic markers to validate this as an approach. Briefly, mixed neuron-astrocyte cultures were transfected with eGFP and co-stained with either GFAP (astrocytes) or NeuN (neurons). Then, blind to the GFAP or NeuN staining, pictures of eGFP-expressing cells were classified as neurons or astrocytes by 1 st author Paul Baxter, based on morphology, after which an independent person determined which of these cells were GFAP or NeuN-positive. Of 80 cells classified as astrocytes based on morphology, 98.7% were GFAP-positive and 1.3% were NeuN-positive. Of 134 cells classified as neurons based on morphology, none were GFAP-positive and 100% were NeuN-positive. Thus, morphology is an accurate way of distinguishing neurons from astrocytes, as long as the experimenter is experienced, and the approach is validated. The following approximate number of cells were analysed: Nrf2 WT/KO astrocytes ( For neuronal death assays, GFP-transfected neurons were imaged and their locations mapped using the Leica "Mark and Find" software application. Neurons were allowed 3 h to re-equilibrate and then stimulated as indicated. Images were taken of saved locations at 24 and 48 h post stimulation, with at least two wells per condition and two pictures per well. Cell death was assessed by counting the number of surviving GFP-positive neurons pre and post-stimulation, with the user blind to the image analysis. Neuronal death was easily identifiable by the replacement of a healthy GFP-expressing cell with the presence of fragmented neurites and fluorescent cell debris. In addition, cell viability was measured using the Cell Titre-Glo assay kit (Promega).
For measurement of α-synuclein and synapsin colocalization, untransfected neurons were treated as above with Trichostatin or CDDO TFEA , then fixed, permeabilised and incubated over-night at 4oC with mouse α-synuclein antibody and rabbit synapsin antibody (1:1000 dilution; Synaptic Systems), followed by 2 h incubation with Alexa Fluor 594 conjugated anti-mouse and Alexa Fluor 488 conjugated anti-rabbit antibodies. Images were taken at 60x magnification using a Nikon A1R confocal microscope. Pearson's colocalization coefficient was calculated using the JACop plugin for ImageJ 82 , with 18 images taken per condition across 3 independent experiments.
Glutathione depletion assays were performed as described (Baxter et al., 2015), substituting tBHQ (50 µM) for H 2 O 2 . Briefly, cells were treated with GSH-ee 1 mM for 1 h, then with tBHQ 50 µM for the times indicated. thirty minutes before the end of stimulation, neurons were treated with 50 μM MCB, and allowed to incubate at 37°C. Cells were then washed once with fresh TMo, and lysed in K 2 HPO 4 buffer containing 0.5% Triton-X-100. Lysates were centrifuged at 15,700 × g (13,000 r.p.m.) at 4°C for 10 min, and supernatants were transferred to a black 96-well plate for fluorescence measurement (excitation 405 nm, emission 520 nm) with a FLUOstar OPTIMA. Lysates were then assayed for protein concentration using a BCA assay, to which fluorescence values were normalized to.

RNA Isolation, qPCR and RNA-sequencing
RNA was isolated using the High Pure RNA Isolation Kit (Roche) according to the manufacturer's instructions, including a column based 15 min DNAse I treatment. For qPCR, cDNA was synthesised from 1 to 3 µg of RNA using the Transcriptor First Strand cDNA Synthesis kit (Roche) using a mix of both random Hexamer primers and anchored Oligo(dT) 18 primers. qPCRs were run on a Stratagene Mx3000P qPCR system (Agilent Technologies) using FastStart SYBR Green Master (Rox) mix (Roche) using 6 ng of initial RNA per 15 µl qPCR reaction and 200 nM of forward and reverse primers to standard PCR and amplification conditions. Gene of interest expression was normalised to RPL13A mRNA expression. The following primers were used: Rpl13a -F: GATGAATACCAACCCCTCC, R: CGAAC AACCTTGAGAGCAG.
Slc7a11 (xCT) -F: ATACTCCAGAACACGGGCAG, R: AGTTCCACCCAGACTCGAAC Hmox1 -F: AGCACAGGGTGACAGAAGAG, R: GGA GCGGTGTCTGGGATG For RNA-sequencing, libraries were prepared from RNA samples by Edinburgh Genomics using the Illumina TruSeq stranded mRNA-seq kit, according to the manufacturer's instructions (Illumina). The libraries were pooled in equimolar proportions and sequenced on an Illumina HiSeq 2500 platform in high output mode (v4 chemistry). Sequencing was performed to a depth of~35 million paired-end reads per sample, with 3 biological replicates per condition. Sequencing reads were mapped to the Mus Musculus (mm10) reference genome using the STAR RNA-seq aligner version 2.5.3a 83 , with a summary of per-gene read counts generated from the mapped reads with featureCounts version 1.5.2 84 using annotations from Ensembl version 90 85 . Relative expression levels of genes are expressed as fragments per million reads per kilobase of message (FPKM). Differential expression analysis was performed using DESeq2 (R package version 1.16.1) 86 with a significance threshold set at a Benjamini-Hochberg-adjusted p-value <0.05. Raw data are deposited at EBI (E-MTAB-5688).

Statistical analysis
Statistical testing of the RNA-seq data is described in the RNA-seq methods section. Sample size for experiments was based on powering an experiment at 80% to detect a 30% effect size, based on the variance of data in previously published experiments from our laboratory. Other testing involved a 2-tailed paired Student's t-test, or a one-or two-way ANOVA followed by Sidak's or Dunnett's post-hoc test, as indicated in the legends. For t-tests, variance was generally found to be similar, abrogating the need for Welsh's Correction. No data were exlcuded. Throughout the manuscript, independent biological replicates are defined as independently performed experiments on material derived from different animals.