Direct Keap1-Nrf2 disruption as a potential therapeutic target for Alzheimer’s disease

1The transcription factor Nrf2 (nuclear factor E2-related factor 2) targets cellular defence genes containing antioxidant response elements (ARE), which include antioxidant enzymes (glutamate cysteine ligase; GCL), drug metabolising enzymes (cytochrome P450s, glutathione S-transferases; GSTs), molecular chaperones, DNA repair enzymes and proteasome subunits. Activation of these protective genes in response to Nrf2 enables the cell to maintain redox balance and to remove damaged proteins under conditions of oxidative and xenobiotic stress.

2Such cellular stress is a key feature of several neurodegenerative diseases. Markers of oxidative damage are increased in the brains of Alzheimer’s disease (AD), Parkinson’s disease (PD), Huntington’s disease (HD) and, in CSF, of Amyotrophic Lateral Sclerosis (ALS) patients. Some evidence also suggests that AD, PD and ALS patients have reduced xenobiotic metabolism and that the AD-causing Aβ42 peptide may act as a xenobiotic. As Nrf2 is inhibited in several neurodegenerative diseases, including AD and ALS, as well as in APP/PS1 mutant mouse models of AD, deficits in this important cellular protection pathway may, in part, explain the cellular damage associated with these conditions. Conversely, Nrf2 over-expression protects against toxicity induced by the Aβ42 peptide in mammalian cells, and prevents neuronal pathology in mouse models of ALS, PD and AD. Activation of Nrf2 is, therefore, increasingly implicated as an attractive target for the prevention of several neurodegenerative conditions.

3Potent activators of Nrf2 have been developed, and confer protection against chemically-induced neurotoxic insults and improve memory deficits in mouse models of AD. Many of these Nrf2 activators, however, have been reported to exert toxicity due to off-target effects. Additionally, unregulated activation of Nrf2 can have detrimental consequences, with prolonged, ubiquitous activation shortening lifespan in Drosophila, and mutations in the Nrf2 inhibitor Keap1 (kelch-like ECH-associating protein 1) causing cancer in humans. Hence, a better understanding of the mechanisms by which Nrf2 function is inhibited in neurodegenerative disease may enable the design of drugs targeted for the prevention of these conditions with minimal side-effects.

4Nrf2 activity is tightly regulated by two main inhibitors, Keap1 and GSK-3 (glycogen synthase kinase-3) (S1 Fig). GSK-3 plays a well-established role in the pathogenesis of AD, and GSK-3 inhibitors, including lithium, prevent pathology in animal models of AD, ALS and PD. Emerging evidence also suggests that inhibiting Keap1 can ameliorate neuronal degeneration, with Keap1 RNA interference (RNAi) protecting against Aβ42 and MPTP toxicity in cells, and heterozygous loss of Keap1 protecting against neuronal pathology in Drosophila models of PD. Both GSK-3 and Keap1 may, therefore, serve as valid candidates for mediating the inhibition of Nrf2 in neurodegenerative diseases, and their inhibition may enable the prevention of Nrf2 deficits, neuronal stress and degeneration specifically in these conditions. A comparative analysis of the efficiency of their inhibition in rescuing Nrf2 deficits in neurodegenerative diseases specifically, however, is required.

5We aimed therefore to investigate the role of Nrf2 in mediating the protective effects of GSK-3 and Keap1 inhibition against Aβ42 toxicity. Using an inducible Drosophila model of AD, we confirmed that Aβ42 inhibits activity of the fly homolog of Nrf2 (cap-n-collar isoform C, cncC) in neurons, consistent with previous findings in mice. Both inhibition of GSK-3, using lithium, and loss-of-function mutations in Keap1 protected against Aβ42 toxicity in this model. We found, however, that neuronal protection in response to Keap1 inhibition correlates with the rescue of Aβ42-induced Nrf2 defects, whereas lithium treatment appears to exert neuro-protection independently of Nrf2.

6Consistent with Nrf2 activation, Keap1 inhibition prevented the enhanced sensitivity of Aβ42-expressing flies to xenobiotic stress, but exerted minimal protection against oxidative damage in comparison to lithium treatment. Combined modulation of Keap1 and lithium additively protected against Aβ42 toxicity, in comparison with either treatment alone, but did not improve their respective effects on xenobiotic and oxidative damage. This further supports the divergent beneficial effects of these manipulations. Down-regulation of Keap1 alone additionally protected against Aβ42 toxicity by mechanisms correlating with enhanced degradation of Aβ42 peptide.

7Overall our data highlight Keap1 as an efficient target for the amelioration of Nrf2 deficits and protection against neuronal damage in AD. Finally, we show for the first time that a newly-described direct inhibitor of the Keap1-Nrf2 protein-protein interaction can indeed protect against the synapto-toxicity of naturally-derived Aβ oligomers in primary mouse neuronal cultures. As current Nrf2 activators may exert toxicity due to off-target effects, our data suggest that blocking the specific interaction of Nrf2 with Keap1 may provide an exciting new avenue for the discovery of disease-modifying treatments for AD, and potentially other neurodegenerative conditions.

8Over-expression of APP in mice inhibits expression of Nrf2 target genes. Comparing expression, at equivalent levels, of various Aβ species in adult fly neurons, we have shown that the presence of aggregating Aβ42 peptides may specifically mediate this effect. Using Nrf2/cncC reporter flies (gstD1-GFP;), RU486 induction of a single, site-directed, copy of Arctic mutant Aβ42 (ArcAβ42), but not WT Aβ40 or Aβ42 peptides, significantly reduced Nrf2/cncC activity compared to un-induced gstD1-GFP-expressing controls (Fig 1A and 1B). Only ArcAβ42 flies develop pathological phenotypes under these expression conditions. Moreover, inducing high levels of WT Aβ42 using an independent random-insertion line, which does cause neuronal decline, also significantly reduced Nrf2/cncC activity (Fig 1C). This suggests that the inhibition of Nrf2 may be related to the concentration-dependent ability of Aβ42 to aggregate and exert toxicity.

9Consistent with a specific inhibition of neuronal Nrf2/cncC activity in response to Aβ42 expression, ArcAβ42 reduced GFP levels in the heads, but not bodies (S2A Fig), of RU486-induced flies. Microarray analyses further revealed several pathways and processes that might mediate toxicity in response to ArcAβ42 expression in fly neurons (Fig 1D, Arc Aβ42). Comparing this data-set to a previously published microarray analysis using flies ubiquitously over-expressing Nrf2/cncC (Fig 1D, cncC), we found that Aβ42 expression induced reciprocal effects on gene ontology (GO) categories normally regulated by the Nrf2 pathway. Aβ42 activated processes that are down-regulated, and inhibited processes that are up-regulated, by cncC (Fig 1D & S2B Fig), further confirming its suppressive effect on Nrf2 signalling.

10Overall these findings suggest that the inhibition of Nrf-2 in AD is robustly conserved in Drosophila, and that this deficiency is caused by the Aβ42 peptide directly. Moreover, as we also observed a suppression of Nrf2/cncC in flies over-expressing human 0N3R tau or the ALS-related C9orf72 (GR) 100 di-peptide repeat protein (DPR) in adult neurons (S3A and S3B Fig), our findings suggest that the accumulation of toxic proteins may lead to the generalised defect in Nrf2 signalling observed in neurodegenerative diseases.

11Drosophila thus provide an excellent context for further analysis of the mechanisms by which Aβ42 regulates Nrf2 in vivo. As cncC, MafS and Keap1 mRNA transcripts were unaltered by Aβ42 in our flies (S3C Fig), we hypothesised that Aβ42 modulates Nrf2 activity by altering its biochemical interaction with upstream regulators.

12Keap1 is a well-known negative regulator of Nrf2 activity in response to oxidative and xenobiotic stressors, but its role in mediating damage in response to neurotoxic proteins has not been widely studied in vivo. Genetically reducing Keap1, alone, extended lifespan (Fig 2A) and rescued neuronal-specific motor defects in both WT (S4 Fig) and ArcAβ42-expressing flies (Fig 2B) using two independent alleles, Keap1del and Keap1EY5. GSK-3 has more recently been shown to inhibit Nrf-2, independently of Keap1, but has long been implicated in the pathogenesis of several neurodegenerative diseases including AD. We have previously shown that lithium, a well-described GSK-3 inhibitor, prevents Aβ42 toxicity in our fly model. Thus both GSK-3 and Keap1 may serve as effective targets for the prevention of neurodegeneration in AD.

13We next compared the effects of manipulating both Keap1 and GSK-3 regulatory pathways on Aβ42 toxicity and Nrf2/cncC activity in a parallel study. Using a dose of lithium (25 mM) that prevented Aβ42-induced lifespan-shortening to a similar extent as heterozygous loss-of-function mutations in Keap1 (Fig 2C, P = 0.739 comparing +RU, + LiCl 25 mM to +RU, Keap1 del), this comparative analysis revealed that both manipulations rescued lifespan-shortening in Aβ42-expressing flies in an additive manner (Fig 2C, P<0.001 comparing +RU + LiCl 25 mM to +RU + LiCl 25 mM, Keap1 del), consistent with an independent mechanism of action.

14At these therapeutic levels, genetically reducing Keap1, alone, significantly rescued the decline in Nrf2/cncC reporter expression observed in Aβ42-expressing flies (Fig 2D, p<0.05 comparing +RU vs +RU, Keap1 del). By contrast, although 25 mM lithium was sufficient to rescue Aβ42 toxicity, a dose of 50–100 mM lithium, which may exert toxic effects, was required to rescue Nrf2/cncC activity as measured using both microarray (S5A Fig) and gstD2 expression (S5B Fig) analyses. A heterozygous loss-of-function mutation in cncC (cncCK6; S5C Fig) did not alter the ability of lithium to protect against Aβ42-mediated lifespan-shortening, further suggesting that Nrf2 is not required for lithium to exert its protective effects in vivo. Consistent with our previous finding that lithium reduces Aβ42 levels, and prevents toxicity, by blocking translation, early intervention concurrent with RU486 induction was required for lithium to protect against Aβ42-induced climbing defects (S5D Fig) and was correlated with reduced Aβ42 levels from the point of induction (S5E Fig). This suggests that, at therapeutically active concentrations, the protective effect of lithium against Aβ42 toxicity is not predominantly mediated by activating Nrf2, but rather by blocking Aβ42 peptide accumulation through inhibition of translation.

15Our data provide the first evidence that specific Keap1 inhibition can protect against Aβ42 toxicity in vivo, and they highlight Keap1 as a more efficient target for the prevention of Aβ42 peptide-induced Nrf2 inhibition in comparison with lithium treatment.

16As Nrf2 plays an important role in protecting against oxidative and xenobiotic stress, we investigated the nature of the Aβ42-induced molecular damage that was ameliorated by Keap1 inhibition (Fig 3). UAS-ArcAβ42>elavGS control and heterozygous Keap1 mutant flies were induced with RU486 for 14 days before measuring sensitivity to xenobiotic (DDT; dichlorodiphenyltrichloroethane) or oxidative (hyperoxia or paraquat) stressors (see methods).

17Reducing Keap1 alone protected against Aβ42-induced sensitivity to DDT (Fig 3A) and paraquat (Fig 3C), but not hyperoxia (Fig 3B; +RU vs +RU, Keap1 del). Lithium treatment, on the other hand, protected against sensitivity of Aβ42 flies to DDT to a similar extent as Keap1 inhibition (Fig 3A), but more significantly protected against sensitivity to both hyperoxia (Fig 3B) and paraquat (Fig 3C)-induced oxidative damage. Moreover, when combined, lithium did not add significantly to the ability of reduced Keap1 to protect against the sensitivity of Aβ42 flies to DDT (Fig 3A, +RU, +LiCl 25 mM Keap1 del vs +RU, Keap1 del) and, conversely, reduced Keap1 did not add to the ability of lithium to protect against paraquat (Fig 3B) or hyperoxia (Fig 3C, +RU, +LiCl 25 mM Keap1 del vs +RU, +LiCl 25 mM). Consistent with the apparently different mechanisms of neuronal protection by Keap1 and lithium, this finding suggests that Keap1 inhibition acts mainly through protection against xenobiotic stress and lithium predominantly by limiting oxidative damage.

18In addition to increasing Nrf2/cncC activity, the protective effect of reducing Keap1 on Aβ42 toxicity correlated with a reduction in Aβ42 peptide levels (Fig 4A). Although Aβ42 mRNA was also slightly reduced by down-regulation of Keap1 (Fig 4B), the peptide levels were not altered until 14 days of age (Fig 4C), suggesting that reducing Keap1 may clear Aβ42 peptide by activating protein degradation mechanisms.

19We have shown previously that Aβ42 peptide is stable for several weeks following induction in our inducible fly model. To assess whether the peptide is indeed degraded as a consequence of reducing Keap1, we induced Aβ42 expression in Keap1del and control flies for one week with RU486 then measured Aβ42 levels by ELISA at several time-points following transfer to non-RU486-containing medium (Fig 4D). Aβ42 levels were equivalent between heterozygous Keap1del and control flies at the end of the induction period (0d since switch-off; age 7d), but then declined only in flies with reduced Keap1, starting from 3 days following switch-off (age 10d). This finding confirms that inhibition of Keap1 does indeed lead to degradation of the Aβ42 peptide. Moreover, analysis of insoluble proteins one week following RU486 induction for 10 days (Fig 4E), revealed that inhibition of Keap1 reduced the level of aggregated Aβ42.

20Keap1-Nrf-2 signalling has been implicated in both autophagy and proteasomal degradation. p62, a selective autophagy substrate, competes with Nrf-2 for binding to Keap1 and Keap1 may activate autophagy directly by binding to p62. However, we did not observe any alteration in autophagy activity upon reduction of Keap1 in the Aβ42-expressing flies, as measured by western blotting using an antibody specific for Drosophila ATG8 (Fig 5A). Proteasome subunits are transcriptional targets of Nrf2/cncC and over-expression of cncC in flies, either directly or by reducing Keap1 levels, increases proteasome expression and activity. Accordingly, we detected an increase in proteasome activity in the heads of Aβ42-expressing flies upon reduction of Keap1 (Fig 5B), at a time-point coinciding with reduced Aβ42 levels (age 14d). Pharmacological inhibition of this enhanced proteasome activity (S6A Fig), however, did not prevent the degradation of Aβ42 in response to Keap1 inhibition (S6B Fig), suggesting that this may not be the mechanism by which loss of Keap1 induces Aβ42 degradation. Further studies are therefore required to elucidate the precise clearance mechanisms mediating Aβ42 degradation following Keap1 inhibition.

21Our data highlight Keap1 as a valid in vivo therapeutic target for AD. Current activators of Nrf2 are electrophilic compounds, which commonly act by modifying cysteine residues on Keap1 and thus disrupt its interaction with, and inhibition of, Nrf2. Of these, the synthetic triterpenoids, such as Bardoxolone methyl (CDDO-Me), are potent Nrf2 activators and have exhibited therapeutic potential against several diseases, including neurodegeneration, in animal models and clinical trials. One problem with the mechanism of action of these electrophilic Nrf2 activators, however, is the modification of cysteine residues on other targets, which may lead to side-effects in humans. To improve specificity, compounds have more recently been designed to directly disrupt Keap1-Nrf2 binding, and these enhance Nrf2 activity in vitro and in cells. Activity of these compounds against disease models, however, remains to be examined.

22We tested the ability of a recently published direct Keap1-Nrf2 inhibitor, 22h (Fig 6A), to protect against exogenous amyloid toxicity in cultured cells. At their EC50 concentrations for Nrf2 activation (S7 Fig), we first performed a comparative analysis of 22h, CDDO-Me and the GSK-3 inhibitor, TDZD-8, for protection against natural Aβ oligomer-induced toxicity in SH-SY5Y cells (Fig 6B). Interestingly, Nrf2 and Keap1 protein levels were increased following 24 h Aβ oligomer treatment (S8 Fig), suggesting that the Keap1-Nrf2 pathway is indeed dysregulated by acute Aβ exposure. TDZD-8 was a poor activator of Nrf2, inducing its target gene NQO1 only at a single concentration of 1 μM (S7C Fig), and exerted toxicity in control SH-SY5Y cells (Fig 6B). Although CDDO-Me was a more potent activator of Nrf2 (S7D Fig), 22h significantly protected against Aβ toxicity, in comparison with both CDDO and TDZD-8, in this experimental paradigm (Fig 6B). This comparative analysis supports our suggestion that Keap1 may serve as a more efficient target, than inhibition of GSK-3, for the rescue of Nrf2-dependent effects of Aβ42 toxicity. Moreover, as TDZD-8 has previously been shown to protect against Aβ toxicity in primary neuronal cultures, our data suggest that the threshold for this protective effect of GSK-3 inhibition lies below that for its effects on Nrf2 activity. These findings indicate that direct Keap1-Nrf2 disruptors may protect against Aβ42 toxicity more effectively than established Nrf2 activators.

23To further test the effects of Keap1-Nrf2 disruption on neuronal function in response to Aβ oligomers, we measured the effects of 22h in primary mouse neurons (Fig 6C). Conditioned medium obtained from Tg2576 mouse neurons (Tg2576CM) has previously been shown to contain oligomeric Aβ species at concentrations similar to those in human CSF, and to reduce spine density of GFP-transfected WT mouse cortical neurons. We pre-treated GFP-transfected WT mouse neurons of 12 d in vitro (DIV), with 1 or 10 μM 22h for 16 h prior to administration of either wt or Tg2576 conditioned media (wtCM or Tg2576CM), with continued 22h treatment, for a further 24 h before examining spine density (see Materials & Methods). As previously reported, Tg2576CM reduced total spine density of cortical neurons compared to wtCM (Fig 6C and 6D). Strikingly, spine density was rescued by treatment with compound 22h (Tg2576CM, 0.01% DMSO vs Tg2576CM, 10 μM 22h) at non-toxic doses (wtCM, 0.01% DMSO vs wtCM, 10 μM 22h). Aβ42 levels in conditioned media were unchanged following treatment with 10 μM 22h (Fig 6E) which further correlated with increased expression of Nrf2 target genes (Fig 6F).

24These findings suggest that directly blocking the Keap1-Nrf2 interaction can protect neurons downstream of extracellular amyloid toxicity.

25As the prevalence of ageing-related neurodegenerative diseases, such as Alzheimer’s, is predicted to rise dramatically, and with a lack of disease-modifying therapies currently available, there is an urgent need to find new therapeutic targets to slow down neuronal degeneration in these conditions. Accumulating evidence suggests that down-regulation of the cell protective transcription factor Nrf2 may enhance neuronal susceptibility to molecular damage, and that its activation may confer neuronal protection. Nrf2 is, therefore, a promising novel target for the prevention of several neurodegenerative diseases, but its activation must be tightly regulated to have beneficial effects. Moreover, although current activators of Nrf2, including electrophilic compounds, such as sulforaphane, and the more potent bardoxolone methyl (CDDO-Me), can exert neuroprotective effects, all of these compounds are indirect activators and likely to exert multiple off-target effects, potentially leading to toxicity. A better understanding of the precise mechanisms regulating Nrf2 activity (S1 Fig) under particular pathological conditions would, therefore, enable the design of drugs targeted to the prevention of specific diseases with minimal side-effects. Our work aimed to investigate the mechanisms regulating Nrf2 inhibition specifically in response to the AD-related Aβ42 peptide, with a view to identifying novel targets for the prevention of AD and other ageing-related neurodegenerative conditions.

26Our study firstly confirms, in Drosophila, previous reports that Nrf2 target genes are down-regulated in AD brain, and in mouse models of AD. Some studies suggest that this inhibition may be mediated by Aβ42 directly, with exogenous, synthetic, Aβ42 peptide reducing Nrf2 target gene expression in primary mouse neurons, and blocking Nrf2 nuclear translocation following injection into the hippocampus of rat and mouse brain. As Aβ42 peptide is expressed in our fly model independently of APP processing, our study further confirms that this direct effect on Nrf2 activity occurs in response to naturally-derived Aβ42 peptide conformations in vivo (Fig 7B). Importantly, other disease-related proteins, including tau and C9orf72 DPRs, also suppressed Nrf2/cncC signalling, but the non-toxic Aβ40 peptide did not. This suggests that inhibition of the Nrf2 pathway is a generalised response to the accumulation of aberrant proteotoxic proteins.

27The mechanisms mediating the effect of Aβ42 on Nrf2 activity remains to be established. Although a recent report, using sweAPP-expressing cells, has suggested that Nrf2 transcripts can be replenished by altering DNA methylation, Aβ42 did not alter mRNA expression of Nrf2/cncC, or of its co-transcription factor MafS, in our Drosophila model. This suggests that in vivo Nrf2 de-regulation in AD may be post-transcriptional. We have shown that Aβ42 oligomers increased Nrf2 and Keap1 proteins in SH-SY5Y cells after 24 h treatment, suggesting that Nrf2 is initially stabilised in response to acute amyloid exposure. This may represent a protective response to the initial toxic insult, as similar effects on Nrf2 have been observed at early time-points following transient focal ischaemia and correlate with preservation of peri-infarct regions of the brain under these conditions. Keap1 is also an Nrf2 target gene, however, and may subsequently be upregulated to control Nrf2 activity following the initial exposure to Aβ42 in our study. If sustained, this increase in Keap1 levels could provide a potential mechanism for the inhibition of Nrf2 observed in AD brain and other chronic neurodegenerative conditions. This hypothesis is supported by observations that Nrf2 protein is downregulated and Keap1 upregulated in mouse brain following 15 days exposure to synthetic Aβ42. Further work is required, however, to investigate the detailed timing of these events following chronic in vivo exposure to natural Aβ oligomers.

28Since Aβ42 did not directly affect Nrf2/cncC gene expression, we investigated the role of its upstream inhibitors, GSK-3 and Keap1, on toxicity. GSK-3 has a well-documented role in Alzheimer’s, and is suggested to provide a pivotal connection between the characteristic amyloid plaque and neurofibrillary tangle pathologies of this disease. Activation of GSK-3 has been proposed to exert neuronal toxicity by many mechanisms, including increasing apoptosis and inflammation, and impairing axonal transport, synaptic function, cell cycle regulation and adult neurogenesis. Conversely, several GSK-3 inhibitors, including lithium, protect against AD-pathology in mice and some have been tested in clinical trials for AD. Although GSK-3 inhibition increases Nrf2 activity in AD models, however, only one study has shown epistatically that Nrf2 mediates the neuroprotective effect of lithium treatment against paraquat toxicity in cells. The role of Nrf2 in mediating the protective effect of GSK-3 inhibition in AD has also not been empirically investigated.

29Our current study confirms that lithium treatment increases transcription of Nrf2 target genes in a dose-dependent manner. Concentrations of lithium required to activate Nrf2 (≥50 mM-100 mM LiCl), however, have previously been shown to exert toxicity in Drosophila. Lower doses of lithium (25 mM) were sufficient to protect against Aβ42 toxicity, to a level comparable with reducing Keap1, but did not significantly activate Nrf2/cncC. Moreover, genetically reducing Nrf2/cncC function did not prevent the lifespan-extending effects of lithium, in Aβ42 flies, even at a maximizing concentration (50 mM). Our data, therefore, suggest that lithium mediates neuroprotection against Aβ42 independently of its effects on Nrf2 (Fig 7B). Consistent with our previous observations that lithium (10–100 mM) reduces Aβ42 peptide levels by inhibiting translation, early lithium administration reduced Aβ42 peptide, from the point of induction, in our current study and this was required for lithium to exert its protective effects.

30These findings are further supported by our observation that the specific GSK-3 inhibitor TDZD-8 has a narrow window for Nrf2 activation in cultured cells, inducing its target gene NQO1 only at a single concentration of 1 μM (Fig 6B). At concentrations ≤ 1 μM TDZD-8 has been described to prevent Aβ-induced reductions in spine density in mouse neurons, but doses ≥ 1 μM exerted toxicity (Fig 6B and). Together these data suggest that the threshold for protection against Aβ toxicity by GSK-3 inhibitors lies below the concentration required for activation of Nrf2 in neuron. This supports the hypothesis that blocking GSK-3 exerts its protective effects independently of Nrf2.

31Contrary to our findings with lithium, we provide the first in vivo evidence that Keap1 inhibition can exert neuroprotective effects in response to Aβ42 by ameliorating deficits in Nrf2 activity (Fig 7C). As with Drosophila models of PD, heterozygous loss of Keap1 protected against Aβ42-induced lifespan-shortening and climbing defects in the fly. Moreover, we show that this protective effect of reducing Keap1 correlates significantly with a rescue of Nrf2/cncC activity in response to Aβ42.

32Only one previous study has addressed the role of Keap1 inhibition, specifically, in AD. Keap1 RNA interference increased expression of Nrf2 target genes, protected against synthetic Aβ42-induced cytotoxicity and oxidative damage to proteins and lipids, and enhanced autophagy in cultured cells. Our study adds to these in vitro findings, demonstrating that Keap1 inhibition rescues Aβ42-induced Nrf2 deficits, and protects against neuronal decline in vivo. These effects occur in correlation with prevention of xenobiotic and, to a lesser extent, oxidative damage as well as the degradation of endogenous, aggregated, Aβ42 peptide.

33Using primary cortical neurons, our findings further confirm that direct pharmacological inhibition of Keap1-Nrf2 binding can protect against neuronal damage downstream of extracellular Aβ42 oligomers. The mechanisms by which reducing Keap1 enhances degradation of Aβ in vivo, however, requires further investigation. Autophagy, as measured by cleavage of ATG8, was unaltered in response to Keap1 inhibition in our Aβ42 expressing flies. Consistent with the established effect of Nrf2 on transcription of proteasome subunits, loss of Keap1 enhanced proteasome activity in Aβ42 flies, but blocking this increase, using the proteasome inhibitor Bortezomib, did not prevent the reduction in Aβ42 levels. This suggests that the reduction of aggregated Aβ42 in response to Keap1 inhibition may not be directly mediated via enhanced autophagy or proteasomal degradation. Nrf2 has also been implicated to play a role in the unfolded protein response (UPR), serving as a target for PERK and activating transcription of several components of the ER associated degradation (ERAD) pathway, including chaperones and ubiquitin-conjugating enzymes, in addition to proteasome subunits and autophagy. Future studies will therefore be required to investigate the functional role of such protein quality control pathways, potentially upstream of the proteasome and autophagy, in the clearance of Aβ42 following Keap1 inhibition in vivo.

34Overall, our data point to Keap1, in comparison with GSK-3, as an efficient target for Nrf2 activation in response to Aβ42 toxicity in vivo. Recent advances in the development of direct inhibitors of the Keap1-Nrf2 binding domain may, therefore, enable the prevention of Nrf2 deficits in neurodegenerative diseases with minimal side-effects. Our study shows for the first time that a direct, small molecule inhibitor of Keap1-Nrf2 binding, 22h, can ameliorate synaptotoxicity in response to naturally-derived Aβ oligomers in mouse cortical neurons. As synaptic loss correlates well with cognitive decline, our work suggests that these compounds show promise as therapeutic agents for AD. Moreover, as this is the first demonstration that these compounds can prevent neuronal toxicity, our findings warrant their investigation in other neurodegenerative conditions.

35Both GSK-3 inhibitors and Nrf2 activators can exert toxicity due to off-target effects and thus it is important that modifiers of these pathways achieve therapeutic benefits at low doses. More specific inhibitors of each of these targets are currently being developed. It has also been postulated that combined inhibition of both GSK-3 and Keap1 may activate Nrf2, and confer protection, at lower doses than either intervention alone, hence limiting their detrimental effects.

36We present the first in vivo study showing that combined Keap1 and GSK-3 inhibition, by lithium, confers additive protective effects against Aβ42 toxicity. Keap1 deletion and lithium treatment extended lifespan and improved climbing ability of Aβ42-expressing flies to a greater extent than either manipulation alone. Additionally, reducing Keap1 limits the dose of lithium required to reach maximal levels of protection since heterozygous Keap1del combined with low dose (25 mM) lithium extended lifespan of Aβ42 flies to a similar extent as high dose (50 mM) lithium alone. Our data, however, do not predict that these additive protective effects are due to mechanisms converging on Nrf2. Rather, low dose lithium treatment exerts protective effects independently of Nrf2 activation, whereas the neuroprotective effects of Keap1 inhibition correlate strongly with increasing Nrf2. Hence, the additive nature of combined lithium and Keap1 inhibition appears to be mediated through divergent, complementary protective mechanisms.

37Oxidative and xenobiotic damage are key features of AD brain and may potentially be explained by the down-regulation of Nrf2, which is also observed in several neurodegenerative diseases. Consistent with this idea, Keap1 inhibition and, therefore, Nrf2/cncC activation, correlated strongly with protection against Aβ42-induced sensitivity to the xenobiotic DDT in our fly model. Conversely, Keap1 inhibition exerted minimal protection against oxidative damage in comparison with lithium treatment, which strongly protected against Aβ42-induced sensitivity to both paraquat and hyperoxia-induced damage. As therapeutic concentrations of lithium, which can prevent oxidative damage, did not strongly activate Nrf2/cncC in our flies, this suggests that rescue of Nrf2 activity is not required to protect against oxidative damage in response to Aβ42. Although pharmacological activation of Nrf2, has been shown to protect against oxidative damage, in response to Aβ42 peptide, in cells and in animal models of AD, conflicting reports have also described protective effects against such damage that are mediated independently of Nrf2. Moreover, studies showing protection against Aβ42-induced oxidative damage in response to GSK-3 inhibition, using antisense oligonucleotides and in response to lithium, have not demonstrated a causal role of increasing Nrf2 activity. Our direct comparison of Keap1 and GSK-3 pathways, however, reveals that Nrf2 activity and prevention of Aβ42-induced oxidative damage do not strongly correlate.

38Finally, combining lithium treatment with manipulation of Keap1 did not confer additional protection against oxidative and xenobiotic damage respectively. Overall this suggests that Keap1 and lithium treatment combine to maximise protection against AD-phenotypes, through divergent effects on Nrf2 and by additively protecting against Aβ42-induced oxidative and xenobiotic damage (Fig 7D).

39Our findings provide compelling support for the use of direct Keap1-Nrf2 inhibitors for the treatment of neurodegenerative diseases, particularly AD. Future work is warranted to develop these compounds further for in vivo use, and to investigate their effects in combination with other established therapeutic targets for AD, such as specific GSK-3 inhibitors.

40Stocks were maintained at 25°C on a 12:12-h light:dark cycle at constant humidity on a standard sugar-yeast (SY) medium (15gl-1 agar, 50 gl-1 sugar, 100 gl-1 autolysed yeast, 100gl-1 nipagin and 3ml l-1 propionic acid). Adult-onset neuronal-specific expression of Arctic mutant Aβ42 peptide was achieved by using the elav GeneSwitch (elavGS)-UAS system (GAL4-dependant upstream activator sequence) and treatment with 200 μM mifepristone (RU486), as previously described. ElavGS was derived from the original elavGS 301.2 line and obtained as a generous gift from Dr H. Tricoire (CNRS, France). Aβ lines used in Fig 1 were: UAS-attP Aβ lines, as previously published, and UAS-WT Aβ42x2, obtained from Prof. P. Fernandez-Funez (University of Florida, USA). Random insertion UAS-ArcAβ42 was obtained from Dr D. Crowther (University of Cambridge, UK) and was used in all other experiments. UAS-0N3R tau line was obtained from Guy Tear (Kings College London, UK). UAS-C9orf72 (GR)100 flies are published. Keap1del was generated by P-element mediated male recombination using the P-element insertion line, Keap1[EY02632], Keap1EY5 and gstD1-GFP lines were obtained from Dr D. Bohmann (University of Rochester, USA), and cncCK6 mutant flies, originally generated in the laboratory of William McGinnis (University of California, San Diego, USA), from Dr A. Whitworth (University of Cambridge, UK). All fly lines were backcrossed six times into the w1118 genetic background.

41LiCl (Sigma) was dissolved in ddH20 at a concentration of 5 M before diluting to the indicated final concentrations in SYA medium.

42Bortezomib (New England Biolabs) was dissolved in ethanol at a stock concentration of 10 mM, and stored frozen at– 20°C, before diluting to the indicated final concentrations in SYA medium.

43Flies were raised at a standard density on SY medium in 200 mL bottles. Two days after eclosion once-mated females were transferred to experimental vials containing SY medium with or without RU486 at a density of 10 flies per vial. Deaths were scored and flies were transferred to fresh food 3 times per week. Data are presented as cumulative survival curves, and survival rates were compared using log-rank tests.

44Climbing assays were performed using methods adapted from Sofola O et al., 2010. Briefly, 15 adult flies were placed in a vertical glass column (SciLabware), tapped to the bottom, and their ability to climb subsequently analysed. Flies reaching the top (above 10 cm) and flies remaining at the bottom (below 3 cm) of the column after a 30 sec period were counted. Scores recorded, from three trials per biological repeat, were the mean number of flies at the top (ntop), the mean number of flies at the bottom (nbottom) and the total number of flies assessed (ntot). A performance index (PI) defined as ½ (ntot + ntop—nbottom)/ ntot) was calculated. Data are presented as the mean PI ± SEM obtained in three independent repeats for each group.

45cncC activity was measured by crossing UAS-ArcAβ42;elavGS flies to gstD1-GFP reporter flies, expressing green fluorescent protein (GFP) under the control of a 2.7kb genomic sequence upstream of the cncC target gene gstD1, and analyzing GFP levels by western blotting.

46Fly heads were homogenized in 2× laemmli sample buffer containing 100mM DTT. Proteins were separated by SDS-PAGE at 150V for 1h using 10% Bis-Tris gels and Mes-SDS running buffer (Invitrogen). Gels were then transferred to nitrocellulose membrane, incubated in a blocking solution containing 5% milk proteins in TBST for 1h at room temperature, then probed with GFP (Cell Signaling, 2955S, 1:1000), ATG8 (custom made anti-rabbit polyclonal against peptide EP113385 (Eurogentec), 1:1000,) or actin (Santa Cruz, sc-47778, 1:5,000) primary antibodies overnight at 4°C. Anti-horseradish peroxidase (HRP)-conjugated secondary antibodies (Abcam, 1:12,000) were used and blots were developed using the enhanced chemiluminescence method (ECL) according to the manufacturers’ instructions (Luminata Crescendo; Millipore). Proteins were visualized using a luminescent image analyzer (ImageQuant LAS 4000; GE Healthcare) and relative intensities measured using ImageQuant software. Proteins of interest were expressed as a ratio relative to actin levels in each sample.

47RNA extraction, cDNA synthesis and quantitative PCR (qPCR) reactions were performed as previously published. For gene expression in Drosophila, primers, 5’ to 3’, were as follow: Aβ forward GATCCTTCTCCTGCTAACC and reverse CACCATCAAGCCAATAATCG; cncC forward GAGGTGGAAATCGGAGATGA and reverse CTGCTTGTAGAGCACCTCAGC; MafS forward AGATCGTTCGGATGAAGCAG and reverse GTCTCCAGCTCGTCCTTCTG; gstD2 forward CATCGCCGTCTATCTGGTGGA and reverse GGCATTGTCGTACCACCTGG; and eIF-1A reference gene forward ATCAGCTCCGAGGATGACGC and reverse GCCGAGACAGACGTTCCAGA. Genes of interest were expressed as a ratio relative to eIF-1A.

48For gene expression in primary mouse neuronal cultures, primers 5’ to 3’ were: Hmox1 forward AGCACAGGGTGACAGAAGAG and reverse GGAGCGGTGTCTGGGATG, Srnx1 forward GACGTCCTCTGGATCAAAG and reverse GCAGGAATGGTCTCTCTCTG, and xCT forward ATACTCCAGAACACGGGCAG and reverse AGTTCCACCCAGACTCGAAC, as previously published. Reference gene primers were to mouse actin forward AACCGTGAAAAGATGACCCAGA and reverse CACAGCCTGGATGGCTACGTA. Genes of interest were expressed as a ratio relative to actin.

49Total Aβ42 peptide, from fly heads, was extracted into guanidinium HCl (GnHCl) buffer based as previously described. Alternatively, soluble and insoluble Aβ pools were isolated by differential centrifugation followed by formic acid extraction, as previously described. Aβ42 levels were then measured using the High Sensitivity Human Amyloid Aβ42 ELISA kit (Millipore). Samples were diluted 1:100, for total Aβ42, or 1:10, for insoluble Aβ42, in dilution buffer and ELISA performed according to the manufacturers’ instructions. Protein extracts were quantified using the Bradford protein assay (Bio-Rad laboratories Ltd, UK) and the amount of Aβ42 in each sample expressed as a ratio of the total protein content (pmoles/g total protein).

50For cell culture, conditioned media was removed from cells, following compound treatment, and diluted 1:2 in dilution buffer before measurement of Aβ42 levels by ELISA as described above.

51Fly heads were homogenized in 25 mM Tris, pH 7.5 and protein content determined by Bradford assay. Chymotrypsin-like peptidase activity of the proteasome was assayed using the fluorogenic peptide substrate Succinyl-Leu-Leu-Val-Tyr-amidomethylcoumarin (LLVY-AMC), as previously described. Proteasome activity was determined as the slope of AMC accumulation over time per mg of total protein (pmoles/min/mg).

52Flies were prepared as for lifespan experiments then aged to the indicated time-points before measuring stress resistance. For xenobiotic stress, flies were exposed to DDT vapour, 1 mg/mL diluted in acetone, in glass vials in the absence of food. For oxidative stress, flies were subjected to hyperoxia (95% oxygen; PROOX model 110, Biospherix, USA) or injected with paraquat (75 nLs of 1 mg/ml in Ringers buffer) and maintained on SYA media containing RU486 ± LiCl as indicated. As survival times were short in these experiments flies were not transferred to fresh food. Survival under each stress condition was then monitored by recording the number of deaths every 2 hours from the start of decline. Data are presented as cumulative survival curves, and survival rates were compared using log-rank tests.

53The raw microarray data generated in this study are deposited in ArrayExpress (http://www.ebi.ac.uk/arrayexpress) with identifier E-MTAB-4611. UAS-ArcAβ42/+;elavGS/+ flies were treated, for 17 days, with standard SY medium alone (-RU) or medium containing RU486 in the absence (+RU) or presence of 100 mM Lithium Chloride (+RU, +LiCl). Flies used for microarray analyses were snap-frozen in liquid nitrogen and, for each array, RNA extracted from 200 heads using RLT buffer + 0.01% β-mercaptoethanol and purified with RNeasy columns (Qiagen, Valencia, CA, USA) following the manufacturer's instructions. The quality and concentration of RNA was confirmed using an Agilent Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA, USA), and further procedures followed the standard Affymetrix protocol. All samples were hybridized to the Drosophila Genome 2.0 Genechip. In total, 4–5 biological replicates of each condition (-RU, +RU and +RU, +LiCl) were performed.

54Raw data (cel files) were processed to correct for probe-sequence biases, by using bioconductor's package gcrma (http://www.bioconductor.org) in R (http://www.r-project.org). Affymetrix's MicroArray Suite 5.0 (bioconductor's package affy) was used to determine present target transcripts. Transcripts were deemed present if the p-value was <0.111 and absent otherwise. The raw data were summarized and normalized by using Robust Multichip Average (rma function, part of bioconductor's package affy. In order to identify differentially expressed genes a linear model was fitted and differential expression was assessed using the empirical Bayes moderated t-statistic as implemented in R's limma package. P-values were adjusted for multiple hypothesis testing by applying the Benjamini and Hochberg correction for false discovery rate. Summarized probe-sets were mapped to transcripts using R's package "drosophila2.db". Transcripts not mapping to any known or predicted genes were excluded from further analysis. The following freely available gene expression microarray datasets were used: control vs. hsp70-CncC (E-GEOD-30087).

55The Wilcoxon rank sum test, as implemented in Catmap, was used to perform functional analysis, that is significant enrichment of Gene Ontology categories. FlyBase (http://flybase.org) gene identifiers were mapped to Gene Ontology identifiers (FlyBase version FB2014_01). Ranks of genes were based on the p-value derived from the Bayes t-statistic for differential expression. To account for multiple hypothesis testing, an enrichment of GO terms was deemed statistically significant if the p-value derived from the wilcoxon rank sum test was ≤1.0x10-05. Gene lists were sorted by log-fold change and p-value. For all microarray experiments two sets of lists were derived; a gene list comprising most differentially up-regulated (log-fold change > 0) genes at the top of the list and most differentially down-regulated genes (log-fold change < 0) at the bottom of the list (termed up-to-down) and vice versa (termed down-to-up). If a GO category was found to be statistically significant in the up-to-down list, this GO was referred to as up-regulated, meaning that a large enough proportion of the genes in this GO category were found to be up-regulated or at the top of the list. If a GO category was found to be statistically significant in the down-to-up list, this GO was referred to as down-regulated, meaning that a large enough proportion of the genes in this GO category were found to be down-regulated or at the top of the list.

56Statistical significance of overlaps of GOs in two microarray experiments was determined using fisher's exact test. To account for multiple hypothesis testing a p-value cut-off of ≤1.0x10-05 was used.

57SH-SY5Y cells were incubated in a humidified atmosphere at 37°C, 5% CO2 in Dulbecco’s modified medium (DMEM) supplemented with 4.5g/L glucose, 10% FBS, 1% penicillin- streptomycin, and differentiated with 10 μM retinoic acid. Cells were seeded at an appropriate density in 96-well plates prior to subsequent analyses.

58SH-SY5Y cells were seeded at a density of 2 x 104 cells per well in a 96-well plate and cultured for 2 days before treatment for 24 hours with compound or vehicle (0.1% DMSO final concentration). Cells were then lysed in 50 μl/well lysis buffer (0.1% Tween20 in 2 mM EDTA [pH 7.5]) for 15 mins at room temperature. 200 μl enzyme reaction mixture (25 mM Tris buffer [pH 7.5] containing BSA [0.067%], Tween20 (0.01%), FAD (5 μM), Glucose 6 Phosphate (G6P) (1 mM), NADP (30 μM), G6P dehydrogenase (40 units), MTT (0.03%), and menadione (50 μM)) was added to each well. After 5 min at room temperature, 40 μl/well stop solution (10% SDS) was added and the absorbance at 595 nm measured. The background optical density was measured using wells containing lysis buffer, enzyme and stop solutions without SH-SY5Y cells. The optical density values at 595 nm were averaged and the background corrected ratio of optical densities (compound treated/control) was calculated.

59SH-SY5Y cells were seeded at a density of 2 x 104 cells per well in a 96-well plate. Cells were pre-treated with compound or vehicle overnight before either Aβ-conditioned Chinese Hamster Ovary (CHO) cell media (7PA2CM) or WT-conditioned CHO media (wtCM) was added to wells at a dilution of 50% for 24 hours. Resazurin (final concentration 20 μM) was added to wells and incubated for 4 hours at 37°C, 5% CO2. The resulting fluorescence intensity was measured at 590 nm. The relative fluorescence values were averaged and normalised to wtCM, DMSO-treated control intensities.

60SH-SY5Y cells were seeded at a density of 6,000 cells per well, in a 96-well plate then, after 24 hours, treated with 50% 7PA2CM or wtCM for a further 24 hours. Plates were fixed with 4% PFA (v/v) for 15 minutes, washed three times with PBS, permeabilized with 0.5% Triton-X before staining with primary antibodies against Keap1 (1:50; ab150654, Abcam) and Nrf2 (1:100; ab62352, Abcam) in blocking solution containing 10% goat serum and 3% BSA overnight at 4 C. After 3 washes in PBS secondary antibodies (α rabbit Alexa 594, ab150080, α mouse FITC, ab6785, Abcam) were applied. Nuclei were stained in a PBS solution of 10 μM Hoechst 33342 before high-resolution digital imaging using a Zeiss LSM700 confocal microscope.

61For quantification of Nrf2 and Keap1 staining, and cell morphology, 250 cells per well were imaged using the IN Cell Analyzer 6000 automated laser-based imaging platform with confocal modality (GE Healthcare) at 40X magnification. Automated image analysis was conducted with the IN Cell Developer software using custom-developed analysis protocols. Cell nuclei were identified by acquiring images in the DAPI channel (405 nm excitation, 455/50 emission). Whole cells were identified using images of Keap1 from the FITC channel (488 nm excitation, 524/48 emission). Images were also acquired of Nrf2 in the dsRed channel (561 nm excitation, 605/52 nm emission). The whole-cell intensity of Keap1 FITC and Nrf2 Alexa 594 was measured and expressed as average fluorescence units per cell for each well. N = 5 wells per condition.

62To obtain transgenic conditioned medium (TgCM) enriched in Aβ, neurons from Tg2576 mice were maintained for 14 days in vitro (DIV) without changing the medium. Medium from wild type neurons from littermates, wild type conditioned medium (wtCM), was used as a control. The genotype of the animals was determined by polymerase chain reaction on DNA obtained from fibroblasts.

63Wild type primary neurons were obtained from cerebral cortex of CD1 mouse embryos, at embryonic day 15, as previously described. Neurons were plated to a density 6×105 viable cells/35-mm2 on glass-bottomed dishes (MatTek Corporation, Ashland, MA, USA) previously coated with poly-D-lysine (10 μg/ml) for at least 1 h at 37°C. Cultures were maintained at 37°C with 5% CO2, supplemented with neurobasal medium with 2% B27 nutrient, 2 mM L-glutamine, penicillin (100 units/ml) and streptomycin (100 μg/ml). At 12 days in vitro (DIV) neurons, transfected on day 7 with the plasmid peGFP-N1 (Clontech, Mountain View, CA) using lipofectamine 2000 (Invitrogen), were pre-treated with the Keap1 inhibitor, 22h (10 μM), or 0.1% DMSO, for 16 h. Conditioned medium (CM) from Tg2576 transgenic, Aβ-enriched, or wild type mouse neurons, with and without 22h, was then added for a further 24h before analysis.

64Neuronal morphology was assessed at 14 DIV by high-resolution digital imaging using a Zeiss LSM700 confocal microscope and analysis using NeuronStudio software (CNIC, Mount Sinai School of Medicine). Spine density was defined as the number of spines per micrometer of dendrite length according to previously published protocols. Dendritic spine densities were calculated from 6–17 neurons per condition.

65Data are presented as means ± SEM obtained in at least three independent biological samples. Log-rank, analysis of variances (ANOVA) and Tukey’s HSD (honestly significant difference) post-hoc analyses were performed using JMP (version 11.0) software (SAS Institute, Cary, NC, USA).

66Animals were maintained and treated in accordance with the Animals (Scientific Procedures) Act, 1986, following approval by the Animal Welfare and Ethical Review Body (AWERB), KCL, and the Home Office Inspectorate, and performed in accordance with the European Communities Council Directive of 24 November 1986 (86/609/EEC).