JNK signalling regulates antioxidant responses in neurons

Reactive oxygen species (ROS) are generated during physiological bouts of synaptic activity and as a consequence of pathological conditions in the central nervous system. How neurons respond to and distinguish between ROS in these different contexts is currently unknown. In Drosophila mutants with enhanced JNK activity, lower levels of ROS are observed and these animals are resistant to both changes in ROS and changes in synapse morphology induced by oxidative stress. In wild type flies, disrupting JNK-AP-1 signalling perturbs redox homeostasis suggesting JNK activity positively regulates neuronal antioxidant defense. We validated this hypothesis in mammalian neurons, finding that JNK activity regulates the expression of the antioxidant gene Srxn-1, in a c-Jun dependent manner. We describe a conserved ‘adaptive’ role for neuronal JNK in the maintenance of redox homeostasis that is relevant to several neurodegenerative diseases.


Introduction
Active neurons generate reactive oxygen species (ROS) predominantly as a by-product of mitochondrial respiration. ROS levels are neutralized by constitutive and adaptive reductive mechanisms operating in neurons and glia, including the glutathione system [1][2][3]. In this manner, the amplitude and temporal dynamics of the ROS signal are controlled, damage is limited and transient ROS signals can be interpreted in part to support the growth and plasticity of neurons [4,5]. A physiological level of ROS has been demonstrated to regulate a range of nervous system processes, including neuronal development, synaptic plasticity, and neural circuit tuning [5][6][7].
In many neurodegenerative disorders the reductive capacity of neurons is overwhelmed, contributing to disease progression [4,8]. These damaging levels of ROS, termed oxidative stress, overwhelm neuronal antioxidant defenses. A central component of the neuronal response to ROS is the activation of c-Jun N-terminal kinase, JNK. ROS have been shown to activate JNK-AP-1 signalling [9] which regulates neuronal growth and plasticity [10][11][12][13]. In Drosophila models, excessive ROS driving increases in synaptic growth are seen in both activity generated excitotoxicity and lysosomal storage disease [5,12,14]. Similar changes in synaptic structure can be induced directly by application of oxidants such as paraquat [12] or diethylmaleate (DEM) [5], or through genetic activation of JNK via manipulation of the JNKK hemipterous [15] (hep) or the JNKKK Wallenda [11] (wnd), upstream activators of JNK.
Whether JNK activation by ROS in neurons is protective or degenerative is unknown. Recent evidence indicates JNK-AP-1 activity can prevent degeneration of injured axons [16] while JNK inhibition prevents neuronal loss after injury by promoting AKT signalling [17].
We recently found that knockdown of enzymes involved in antioxidant defense, superoxide dismutase and catalase, reshapes neuronal morphology at the Drosophila larval neuromuscular junction (NMJ) [5]. Similarly, mis-expressing antioxidant enzymes to manipulate levels of individual ROS species has profound effects on synaptic and dendritic growth and arborisation [5,12]. This suggests that plasticity at the synapse is finely tuned to the level of ROS. Given that JNK is known to regulate synaptic plasticity in response to ROS and that knockdown of antioxidant genes can change synapse structure, we hypothesized that JNK-induced structural changes in neurons are mediated in part by an antioxidant response.
To understand if JNK activity regulates antioxidant responses in neurons, we monitored ROS levels in Drosophila with mutations in highwire (hiw) and puckered (puc E69 ). In hiw mutants, unrestrained wallenda (wnd) activity activates JNK and downstream AP-1 signalling to generate exuberant synaptic overgrowth [11,18]. Puc encodes a dual specificity phosphatase, transcriptionally activated by AP-1 that counteracts JNK activity [19]. In both the hiw and puc E69 / + mutant backgrounds we identified enhanced antioxidant defenses and low levels of ROS. In hiw, this enhanced antioxidant defense can be reversed in neurons by down-regulating JNK-AP-1 signalling. Conversely, both puc E69 / + and hiw flies are resistant to chemically induced oxidative stress, while inhibition of JNK-AP-1 signalling increases ROS levels in wild type neurons. We show that in mammalian neurons exposed to oxidative stress, the JNK-AP-1 pathway coordinates the transcription and translation of the antioxidant gene Sulfiredoxin-1, which was localised predominantly in synaptic compartments. Our data indicates that activation of JNK in neurons drives antioxidant responses to shape neuronal morphology.

JNK activity drives antioxidant responses in Drosophila
To quantify levels of ROS in Drosophila, we used the Amplex red assay for detection of hydrogen peroxide [20]. We have previously found that food containing 5 mM and 10 mM DEM which depletes glutathione levels, is sufficient to induce an overgrowth at the larval NMJ [5], with 10 mM concentrations being detrimental to survival (Fig. S1, 6% of flies pupate). To understand whether DEM influences ROS levels in flies, we raised flies on food containing ethanol (vehicle) or 5 mM DEM. Using Amplex red we found that wild type flies have an increased ROS burden when raised on food containing 5 mM DEM (Fig. 1A). To investigate how JNK activity influences ROS levels under these conditions, we reared puc E69 /+ and hiw mutants on the same food and found that both genotypes have significantly lower levels of ROS than wild type flies in vehicle-containing food. Moreover, both genotypes failed to show an increase in ROS when reared on DEM-containing food (Fig. 1A) suggesting that they are resistant to the oxidative stress generated by DEM.
Given that both puc E69 /+ and hiw mutants have lower levels of ROS than wild type flies and are resistant to chemically-induced oxidative stress, we hypothesized that if JNK activity could be reduced in neurons of these flies, then ROS levels would increase back to wild type levels. To test this, we used hiw mutants and expressed dominant negative transgenes of jnk, wnd, ask1, jun and fos pan-neuronally using nSyb-Gal4 (Fig. 1B). We found that pan-neuronal expression of these dominant negative transgenes increased ROS levels of hiw mutants back to wild type controls, except for ask1 DN . This data indicates that manipulation of JNK, its upstream regulators and downstream targets, regulates ROS levels in neurons.
To further evaluate the relationship between JNK activity and ROS levels, we investigated whether expression of dominant negative jnk, ask1 and wnd affects ROS levels in wild type flies. We found that expression of all dominant negative transgenes, pan-neuronally, significantly increases the ROS levels of wild type flies (Fig. 1C). Indeed, the levels observed were similar to those observed when wild type flies are reared on food containing DEM.
The elaborate synaptic morphology of hiw mutant flies has been extensively described [11] and is a consequence of activation of the JNK-AP-1 pathway. We therefore chose to evaluate the synaptic morphology of puc E69 /+ neurons by analyzing the NMJ at Muscle 6/7, hemi-segment A3. We first monitored the effect of DEM in wild type larvae, finding that wild type larvae fed with DEM develop an overgrown NMJ, compared to larvae reared on normal food (Fig. 1D). Similar to hiw mutants, we found that puc E69 /+ larvae have overgrown NMJ synapses (Fig. 1D). Feeding of DEM has no effect on the NMJ of puc E69 /+ larvae, indicating that JNK activity alone is sufficient to change synaptic morphology, which correlates with an antioxidant response (decrease in ROS levels from Fig. 1A). Furthermore, the DEM-induced changes in synaptic morphology in wild type flies, can be rescued when dominant negative transgenes of jun, fos and ask1 are expressed pre-and post-synaptically using spinGAL4 (Fig. 1D, E). Our observations in Drosophila show that JNK activity regulates synaptic morphology but also antioxidant responses. In puc E69 /+ and hiw flies, JNK-AP-1 signalling generates an antioxidant response which lowers the total levels of ROS in flies and protects them from induced oxidative stress while also causing an increase in NMJ synapse size. In wild type flies, inducing oxidative stress changes synaptic morphology, generates a stress response and these responses can be regulated by JNK (Fig. 1F).

DEM induces oxidative stress in primary mammalian neurons
JNK activity in mammalian neurons is often associated with apoptosis and activated JNK is commonly used as a hallmark of cell death in neurodegenerative disease. Given that JNK-AP-1 signalling in the fly generates an antioxidant response and decreasing JNK activity in fly neurons increases ROS levels, we hypothesized that antioxidant responses in mammalian neurons would also be regulated by JNK-AP-1. Having previously validated the effects of DEM in the fly, we established a mammalian model of oxidative stress, using DEM to deplete cellular glutathione. We used this model to assay antioxidant responses after DEM treatment. We first characterized the effect of DEM on cellular GSH levels and ROS production in mature primary rat neurons, using a glutathione assay measuring total glutathione (oxidized and reduced forms). Fig. S2A shows that application of DEM to primary cortical neuronal cultures caused a concentration and time-dependent decrease in cellular glutathione. A 1 h and 4 h treatment of DEM at concentrations of 10 μM and 100 μM caused a rapid decrease in GSH (Fig. S2A). Lower concentrations of 1 μM DEM reduced cellular GSH content with less severity. Given the extent of GSH depletion by DEM, we investigated whether DEM induces mitochondrial toxicity by monitoring WST-1 absorbance (Water Soluble Tetrazolium Salts) as a measure of mitochondrial stress. Compared to vehicle treated neurons, DEM induced a small but significant decrease in mitochondrial capacity, however this was much less than treatment with equivalent concentrations of paraquat (PQ) and rotenone (RT) (Fig. S2B). To determine if DEM induces a ROS burden as it does in our fly model we measured H 2 O 2 levels in neurons treated with DEM using Amplex red reagent. The Amplex red assay revealed that DEM (100 μM) significantly increased in H 2 O 2 levels ( Fig. S2C), after 24 h compared to ethanol treated controls (Fig. S2C). DEM-induced oxidative stress is significantly attenuated when neurons A DEM induces oxidative stress in wild type flies but not puckered (puc E69 /+) or highwire (hiw) mutants. Quantification of hydrogen peroxide levels (Amplex red fluorescence) in wild type, puc E69 /+ or hiw mutant flies administered vehicle (0.16% ethanol) or DEM (5 mM). B Quantification of hydrogen peroxide levels in hiw mutants and hiw mutants pan-neuronally expressing dominant negative jnk (hiw; nSyb > bsk R53K ), wnd (hiw; nSyb > wnd DN ), ask1 (hiw; nSyb > ask1 DN ), jun (hiw; nSyb > jun DN ) and fos (hiw; nSyb > fos DN ). C Quantification of hydrogen peroxide levels in wild type flies, pan-neuronally expressing dominant negative jnk (nSyb > bsk R53K ), wnd (nSyb > wnd DN ) and ask1 (nSyb > ask1 DN ). A-C Graphs show Amplex red fluorescence values normalized to the average signal of wild type flies. Data are plotted as mean ± SEM (**p < 0.01; ***p < 0.001; one-way ANOVA; minimum 15 flies per genotype). D Representative micrographs showing synaptic overgrowth at the Drosophila third instar larval NMJ (Muscle 6/7, hemi-segment A3) in wild type, puc E69 /+ mutants and wild type flies expressing dominant negative fos (fos DN ), jun (jun DN ) and ask1 (ask1 DN ), reared on food containing ethanol or DEM. Scale bar = 30 μm. E Quantification of mean normalized bouton number from (D) genotypes. Data are plotted as mean ± SEM (***p < 0.001; one-way ANOVA; minimum 15 NMJ's analysed per genotype). F Schematic showing the contribution of JNK to neuronal antioxidant stress responses in Drosophila. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.) are pre-incubated with catalase. The duration and concentrations of DEM that we employed deplete glutathione and induce an increase in ROS with minimal toxicity to mitochondrial function.
To model chronic oxidative stress, we treated neurons for longer time points (48 h) with DEM (100 μM) and found that DEM causes a retraction of the dendritic arbor. This effect is rescued when neurons are preincubated with Catalase (100 nM, Fig. S2D) and when the components that produce glutathione, the catalytic and modifying subunits of glutamate cysteine ligase (GCL), are transduced into neurons (Fig. S3).

Oxidative stress activates JNK-dependent SRXN-1 expression, in primary mammalian neurons
JNK enzymes are key mediators of cellular responses to stress in neurons [21], including oxidative stress [22,23], and alterations in JNK activity have previously been linked to changes in dendritic complexity [24]. Phosphorylation of threonine 183 and tyrosine 185 increase JNK activity [21]. To characterize the effectiveness of both DEM and the JNK inhibitor SU 3327 on regulating JNK phosphorylation and activation, we treated primary neurons with DEM (100 μM, 1 h) alone or in the presence of SU 3327 (700 nM, 1 h pre-treatment) and assayed JNK phosphorylation ( Fig. 2A). We found a significant increase in phospho-JNK/total JNK ratios when neurons were treated with DEM, which was significantly attenuated when JNK activity was blocked with SU 3327 (Fig. 2 A,B).
Bicuculline, a competitive antagonist of GABA A (γ-aminobutyric acid) receptors, and 4-aminopyridine, a voltage activated potassium channel blocker, are commonly used pharmacological tools that increase neuronal activity in cultures of cortical neurons [25]. These tools have shown that increased neuronal activity can upregulate neuronal antioxidant genes, in an ERK-AP-1 dependent manner. A putative antioxidant target gene of this signalling pathway, is Sulfiredoxin-1 (Srxn-1) [26]. To identify whether JNK-AP-1 signalling could regulate neuronal antioxidant genes, we compared our experiments with neurons treated with Bicuculline (Bic, 50 μM) and 4-Aminopyridine (4-AP, 500 μM) as a positive control (Bic/4AP).
Using DEM, we found that JNK activity regulates Srxn-1 transcription. DEM (10 μM) induced a 1.73-fold induction in Srxn-1 mRNA expression compared to the ethanol treated controls (Fig. 2C). Similar levels of Srxn-1 mRNA were observed at higher concentrations of DEM (Fig. S4C). When neurons were pre-treated with the JNK inhibitor SU 3327 for 24 h, DEM failed to induce Srxn-1 mRNA. Similar to DEM treatment, increasing neuronal activity with Bic/4AP increased Srxn-1 mRNA (1.32-fold), however this induction was unaffected by JNK inhibition. Synaptic activity is known to induce expression of c-Fos, an AP-1 component. We determined whether oxidative stress also induced c-Fos transcription. Fig. 2D shows that there was no change in c-Fos mRNA levels with DEM treatment whereas Bic/4AP treatment robustly induced c-Fos transcription. Similar to activity-induced Srxn-1 induction, c-Fos induction by synaptic activity was unaffected by inhibition of JNK ( Fig. 2D) suggesting that synaptic activity and oxidative stress recruit signalling pathways differentially to activate potentially divergent gene transcription responses. We next evaluated whether JNK activity regulates SRXN-1 protein expression. Fig. 2E shows that treatment of neurons with DEM (100 μM) rapidly increased SRXN-1 protein expression within 30 min, with high levels detected at 4 h as assessed by Western blot using an antibody that we first validated by overexpressing human SRXN-1 in HEK293T cells ( Fig S4A). DEM-induced SRXN-1 expression was attenuated when neurons were pre-treated with SU 3327 (700 nM, 1 h pre-incubation). As expected, and in agreement with previous work [26], synaptic activity also induced SRXN-1 expression after a 4 h treatment with Bic/4AP, which was attenuated when neurons were pre-treated for 1 h with the non-competitive NMDA receptor antagonist MK801 (10 μM) but was unaffected by JNK inhibition. Moreover, synaptic activity but not DEM, induced ERK activation (Fig. 2E). Taken together, these data indicate that oxidative stress activates JNK to coordinate an antioxidant response in neurons, which is distinct from antioxidant responses coordinated by synaptic activity.

JNK-c-Jun signalling regulates antioxidant responses in mammalian neurons
To identify the cellular mediators that co-ordinate SRXN-1 expression under conditions of synaptic activity or oxidative stress, we assessed the expression and interaction of putative AP-1 components under the two conditions. Cortical neurons were treated with Bic/4AP or DEM for 1 or 4 h and c-Fos or c-Jun immunofluorescence was quantified in NeuN positive neurons. In neurons treated with DEM ( Fig. 3A and B) c-Fos immunofluorescence was found to decrease. In contrast, DEM induced a significant increase in c-Jun immunofluorescence. As expected, in neurons stimulated with Bic/4AP for 1 or 4 h, both c-Fos and c-Jun immunofluorescence was found to increase (Figs. S4, B,D). These data suggest that oxidative stress relies on Jun proteins to mount an antioxidant response in the absence of c-Fos. To further investigate the differential involvement of specific AP-1 components we immunoprecipitated c-Jun from cortical neurons treated for 1 h with either DEM or Bic/4AP and assessed its phosphorylation and association with c-Fos protein. While both DEM and Bic/4AP induce c-Jun expression and phosphorylation, c-Jun co-immunoprecipitated with c-Fos following Bic/4AP-induced synaptic activity but not during oxidative stress mediated by DEM (Fig. 3C). In mammals, c-Jun can form heterodimers with other members of the Jun family to form the dimeric AP-1 transcription factor. To identify if c-Jun was associated with another Jun family member, we probed membranes for JunB and JunD and found no association between either protein with c-Jun in DEM-treated neurons (Fig. S4E). To determine whether c-Jun alone is sufficient to induce SRXN-1 expression, lentiviral particles encoding human c-Jun (Fig. 3D) were transduced into cortical neurons. Lentiviral c-Jun expression was sufficient to induce SRXN-1 expression in neurons (Fig. 3E), however activation of Nuclear Factor Erythroid 2-related factor 2 (Nrf2) using the well characterized agonist TBHQ (10 μM, 16 h) had no effect on neuronal expression of SRXN-1. Taken together, these data suggest that oxidative stress activates JNK, which selectively recruits c-Jun to induce SRXN-1 expression in neurons (Fig. 3F).

SRXN-1 is localised at synaptic terminals and prevents dendrite loss
To understand how SRXN-1 contributes to neuronal antioxidant defenses we investigated SRXN-1 localization in neurons. Having previously identified that chronic treatment of cortical neurons with DEM causes a loss of dendrites, we transfected neurons with PSD95-GFP alone or in combination with a plasmid encoding N-terminal FLAG-tagged human SRXN-1 and treated cells with DEM (48 h). We found that overexpression of SRXN-1 prevented dendrite loss caused by long-term treatment of DEM (Fig. 4A). Interestingly, we also observed overexpressed SRXN-1 immunoreactivity in distal dendrites and on closer examination found that overexpressed SRXN-1 was localised to dendrites and dendritic spines (Fig. 4B). SRXN-1 co-localised with PSD-95-GFP, as indicated by white arrowheads in Fig. 4B.
To further investigate the synaptic localization of SRXN-1, we assessed its distribution in subcellular fractions of mouse brain using the antibody validated in heterologous cells (Fig. S4A). Although characterized as a cytoplasmic enzyme [27], SRXN-1 was not observed in the synaptic supernatant fractions, but in the LP1 (synaptosomal) and LP2 (synaptic vesicle) membrane fractions. The efficacy of the fractionation was confirmed by enrichment of the transcription factor FOXO3a in supernatant fractions over the pellet fractions. The P2 fraction represents a crude synaptosomal fraction comprising microsomes and isolated pre-and post-synaptic nerve terminals. SRXN-1 was most enriched in the LP1 fraction, indicative of synaptic membranes and small organelles, such as mitochondria. Immunoblotting for the alpha subunit of the mitochondrial F1 ATP synthase (ATP5A1) confirmed the presence of mitochondrial proteins in the LP1 fraction. SRXN-1 immunoreactivity was also present in the LP2 fraction, which is enriched in synaptic vesicles (SV), as shown by an enrichment of the SV protein synapsin I (Fig. 4C). JNK immunoreactivity was observed in all fractions, but was enriched in P2, LP1 and LP2, in-line with the distribution of SRXN-1. These data suggest that SRXN-1 associates with synaptic membranes and is consistent with the co-localization with PSD95 we observed by fluorescence microscopy. Therefore, in neurons, antioxidant proteins such as SRXN-1 and sensors of oxidative stress such as JNK are located at synapses, placing them at functionally important sites to orchestrate the adaptive antioxidant response (Fig. 4D).

Discussion
Here we have uncovered a role for JNK signalling in the regulation of neuronal antioxidant capacity using Drosophila and mammalian neurons in culture. We and others have shown previously that JNK regulates changes in synaptic morphology independently of any stimuli [11] or in response to ROS [5,12,14]. We now show that JNK activation drives an antioxidant response, which shapes cellular reductive capacity and synaptic morphology.
We show that Drosophila mutants that have enhanced JNK activity have low levels of ROS and furthermore, are resistant to chemically induced oxidative stress. These reduced ROS levels in mutant flies can be reinstated back to levels observed in wild type flies by inhibiting the JNK-AP-1 pathway. In addition to changes in ROS levels, we show that activation of the JNK-AP-1 pathway leads to overgrown synapses at the Drosophila NMJ suggesting that the status of JNK signalling has implications for structural synaptic plasticity. In mammalian cortical neurons we identified an 'adaptive' JNK-dependent antioxidant response when neurons are challenged with oxidative stress. This adaptive response relies on JNK's downstream effector c-Jun, and results in increased abundance of the antioxidant protein SRXN-1. In contrast, JNKindependent SRXN-1 expression is triggered with increased synaptic activity of cortical neurons. We therefore implicate JNK signalling in the regulation of neuronal antioxidant defenses and identify distinct differences in neuronal responses to ROS generated in conditions of synaptic activity and oxidative stress.

Constitutive JNK activation drives an antioxidant response in Drosophila
Genetic and pharmacological activation of JNK changes neuronal morphology and this structural change is dependent on AP-1 [15,28]. JNK-AP-1 signalling has long been known to positively regulate growth and strength at the larval NMJ [15]. JNK has also been shown to mediate the effect of ROS generated during oxidative stress associated with excitotoxicity [14] and lysosomal storage disease [12]. Combined with reports that activation of the JNK pathway after neuronal injury coordinates a regenerative response [16] and that AP-1 signalling regulates dendrite growth during both development and during conditions of synaptic activity; JNK-AP-1 signalling has been demonstrated as a crucial regulator of neuronal homeostasis. Both puc E69 /+ heterozygotes and hiw mutants, which have increased JNK activity [11,19], showed reduced total ROS levels and rendered flies resistant to chemically induced oxidative stress. The reduced ROS levels in puc E69 /+ heterozygotes and hiw mutants flies could be reversed by inhibition of neuronal JNK-AP-1, suggesting that the altered ROS originates from neurons. The reduced total ROS levels, arising from protective JNK signalling, in hiw mutants might explain the recently reported resistance of neurons in highwire null flies to the damaging effects of physical blows, which mimic traumatic brain injury [29].
A key question raised by our work is whether the JNK-dependent changes in synaptic morphology are driven by the increased antioxidant defenses. We have previously identified DEM as a regulator of neuronal morphology at the NMJ [5] and here we show that this DEM-induced change in synaptic morphology is regulated by JNK.
Furthermore, we show that puc E69 /+ larvae have overgrown NMJ's which are not influenced by oxidative stress induced by DEM. Given that puc E69 /+ and hiw larvae have elevated JNK activity, reduced ROS, elaborate NMJ's and are resistant to chemically induced oxidative stress, our data suggests that changes in neuronal plasticity in these models is at least, in part, driven by an antioxidant response.

JNK coordinates antioxidant responses in mammalian neurons
It is known that synaptic activity can regulate the expression of neuronal antioxidant genes [26]. Furthermore, we have recently shown that ROS generated from synaptic activity acts as a signalling molecule to regulate neuronal plasticity, in a PI3K/DJ-1β dependent manner [5] implicating synaptic activity and ROS in the regulation of synaptic plasticity. What is currently unknown is how neurons recognize and respond to differences in physiological and pathological ROS. Our data shows that neurons recruit different signalling pathways to mediate ROS generated by synaptic activity or oxidative stress, but converge to regulate the same antioxidant gene, Srxn-1.
Increasing synaptic activity activates ERK1/2 [30,31] that can in turn phosphorylate AP-1 components [32,33]. Consistent with this, we found that increasing neuronal activity using Bicuculline and 4-AP regulates Srxn-1 mRNA and protein expression [26]. Conversely, inducing oxidative stress in neurons with DEM did not affect ERK phosphorylation of AP-1 but did activate JNK and SRXN-1 expression. JNK inhibition abolished increased SRXN-1 in conditions of oxidative stress but not during synaptic activity. This suggests that neurons use distinct kinases to upregulate SRXN-1 expression in response to synaptic activity or oxidative stress.
Synaptic activity has been demonstrated to boost glutathione synthesis, providing a positive feedback loop, where accumulating levels of ROS from increased synaptic activity are neutralized by increases in cellular glutathione [26]. JNK activity is also regulated by cellular glutathione levels. In non-stressed cells, JNK is tethered to monomeric glutathione S-transferase Pi (GST-Pi) and increased H 2 O 2 can trigger the detachment and oligomerization of GST-Pi, conjugating ROS to glutathione [34]. This suggests that JNK activity is tuned to the redox thresholds of the cell [35]. This is important given that glutathione homeostasis is altered in many neurodegenerative diseases [60,61] and decreased glutathione levels have been observed in mouse models of Alzheimer's Disease [62] and in AD patients [63].
Both ERK1/2 and JNK kinases regulate AP-1 [21,32], raising the possibility that the mechanisms converge at the level of this dimeric transcription factor. The AP-1 heterodimer of c-Fos and c-Jun [36] is well characterized but the conditions driving differential heterodimer composition are not well understood. In our work, a notable difference between synaptic activity and oxidative stress conditions was the lack of c-Fos transcriptional/translational induction by DEM and reduced dimerization with c-Jun. This indicates that DEM-induced AP-1 activity and Srxn1 transcription is c-Fos independent (Fig. 4C and D). The mammalian AP-1 complex can contain Fos, Jun, Fra, and ATF components and multiple hetero-and homo-dimeric partners are capable of binding to AP-1 DNA binding sites [37,38]. Our immunoprecipitation of c-Jun showed a clear association with c-Fos during conditions of increased neuronal activity, but this association was absent during conditions of oxidative stress. Moreover, total c-Jun increased during oxidative stress and we did not detect JunB or JunD in immunoprecipitates suggesting that c-Jun may homodimerise to generate an antioxidant response in neurons. Our data identify a clear divergence between physiological and stress-dependent signalling in neurons, where activation of JNK and lack of c-Fos defines the stress response.

SRXN-1 is present at the synapse
SRXN-1 is an oxidoreductase that reduces sulfinylated proteins such as 2-cys peroxiredoxins (PRDX), small redox sensitive proteins widely expressed in mammalian neurons [39]. Both cysteine residues in PRDX can undergo reversible hyper-oxidation by peroxides to form sulfinic acid (Cys-SO 2 H) [40]. Hyper-oxidation of PRDX proteins inactivates their peroxidase activity. This inactivation is known to occur in neurons where it can be seen to follow a circadian time course [41], and occur during glutamate induced excitotoxicity [42] in addition to oxidative stress conditions [43]. SRXN-1 catalyses the reduction of Cys-SO 2 H in an ATP dependent manner, restoring PRDX activity [44]. Both PRDX and SRXN-1 are regulated by synaptic activity [26]. Our data indicate that the cellular levels of SRXN-1 can also be tuned by the intracellular concentration of GSH and H 2 O 2 levels.
We additionally show that SRXN-1 can be found at cellular locations that are particularly vulnerable to ROS -synaptic terminals and dendrites [45]. Overexpressed human SRXN-1 localised to dendrites and dendritic spines and endogenous SRXN-1 protein was particularly enriched in synaptosomal membrane fractions. We also found that JNK kinases were present in synaptosomes. The localization of SRXN-1, a cytosolic protein, in a synaptic membrane fraction is intriguing. However, in non-neuronal cells, SRXN-1 has been reported to translocate from the cytosol to a 'heavy membrane' fraction enriched in mitochondria in response to oxidative stress [46]. Mitochondrial SRXN-1 has a role in re-activation of oxidized peroxiredoxins [40,47] and our data indicate that SRXN-1 might reside constitutively with neuronal mitochondria, consistent with a greater basal ROS burden in neurons.
Taken together, our data from both fly and mammalian models identify a fundamental conserved role of JNK signalling in regulating neuronal redox homeostasis through an adaptive antioxidant response. JNK signalling is activated specifically by oxidative stress and activates c-Jun containing AP-1 transcription factors independently of c-Fos. In this manner, JNK imparts mechanistic specificity to activate AP-1 under conditions that have relevance to a number of neurodegenerative diseases.

Culture of neurons and cell lines
Timed-mated female Wistar rats (Charles River UK) (RRID: RGD_737929) were maintained in accordance with the UK Animals (Scientific Procedures) Act (1986). Cortices were dissected from postnatal day 1 (P1) mixed sex rat pups. Animals were euthanised using pentobarbital injection followed by cervical dislocation, according to Home Office guidelines. Cortical cell suspensions were obtained as previously described and cytosine arabinoside (AraC, 2.4 μM final concentration) was added to the growth medium at 1 day in vitro (DIV) [48].

WST-1 assay
After treatment, primary rat neurons were washed with TM and then incubated with phenol red free TM containing WST-1 reagent (Sigma, Cat# 5015944001, 20 μl/ml) for 4 h at 37 • C. 200 μl was transferred to a 96 well plate and absorbance measured using a plate reader (BMG Fluostar λ = 440 nm).

Immunocytochemistry
Cells were washed with phosphate buffered saline (PBS) and fixed for 30 min at room temperature with 4% paraformaldehyde (containing 4% sucrose; Sigma). Cells were permeabilized in 0.5% NP40 in PBS for 5 min at room temperature. All primary and secondary antibodies, dilutions and suppliers can be found in Table 1. Primary antibodies were incubated overnight at 4 • C. Corresponding Alexafluor secondary antibodies (1:500, Thermo Scientific) were incubated for 1 h at room temperature before mounting with Fluoromount (Sigma).

Microscopy and image analysis
Images were collected on an inverted Zeiss microscope (880) with 20x or 63x Plan Apochromat objectives using Zeiss filter sets for DAPI and Alexa 488, 546 or 633. Images were taken at a resolution of 2048 × 2048 pixels.

Synaptosome preparation
The subcellular fractionation of mouse brain was performed as previously described [53]. Briefly, seven mixed sex mouse forebrains were dissected and transferred into cold homogenization buffer (320 mM sucrose, 1 mM EDTA, 5 mM Tris, pH 7.4, 4 • C), homogenized in a glass Teflon homogenizer, and centrifuged in an SS-34 rotor (1000g, 10 min, 4 • C). The supernatant (S1) was collected and the pellet re-suspended in 15 ml of homogenization buffer and centrifuged again (1000g for 10 min). S1 supernatants were combined, and pellets (P1) collected. S1 lysates were further centrifuged at 13000g for 20 min at 4 • C. Supernatant (S2) was collected and the pellet (P2) re-suspended in 21 ml homogenization buffer. The P2 pellet was further centrifuged at 13000g for 20 min at 4 • C. P2 was re-suspended in 320 mM sucrose (300 μl/forebrain), transferred into a glass-Teflon homogenizer with 9 vol of ice-cold distilled water and immediately homogenized for 3 up/down strokes at 2000 rpm. Homogenates were incubated on ice for 30 min with 0.1 volume of 1 M HEPES-NaOH (pH 7.4) and centrifuged at 25000g for 20 min at 4 • C. Pellets (LP1, synaptosomal membranes) were collected and the supernatant (LS1) further centrifuged in a 70 Ti rotor at 165,000g for 2 h at 4 • C. The resulting supernatant (LS2) was collected and the pellet (LP2, synaptic vesicles) re-suspended in 40 mM sucrose (2 ml). Protein concentration was assayed using Bradford reagent and 20 μg of each fraction was prepared in 1x Laemmli buffer for western blotting.

Co-immunoprecipitation
Neurons in 10 cm dishes were lysed as described. Lysates were incubated with primary antibodies for c-Jun at a 1:50 dilution and rotated overnight at 4 • C. Lysates were incubated with 15 μl (30 μl bead slurry) Protein G, Sepharose (GE Healthcare, 17-0618-01) for 2 h at 4 • C rotating. Samples were spun down, transferred to spin-x columns (Sigma, CLS8163-100 EA), washed 3 times with lysis buffer and the beads finally incubated with 4 x Laemmli (containing mercaptoethanol) for 10 min, then eluted through the spin columns. Samples were further heated at 95 • C for 5 min before western blotting. Proteins were detected using conformation specific antibodies (Cell Signalling, L27A9, mAb #3678) which prevent detection of heavy and light antibody chains.

Glutathione assay
Total levels of glutathione were assayed using a colorimetric Glutathione Assay Kit (CS0260, Sigma). Briefly, at 14 DIV, primary neurons in 35 mm dishes were treated with DEM in transfection medium for 1, 4 and 24 h. Cells were then washed with PBS (4 • C), lysed with 5% sulfosalicylic acid and snap frozen in liquid nitrogen. Lysates were then defrosted at 37 • C, centrifuged at 10,000g for 10 min and 10 μl samples were used for the glutathione assay in accordance with the manufacturer's instructions.

Drosophila stocks and husbandry
Drosophila were raised on 4-24® instant Drosophila medium (Carolina Biological Supply Company, USA) supplemented with a yeast Table 1 Antibodies used in this study.

Survival analysis
Hatched 1st instar larvae were collected and raised in standard food with different DEM concentrations (0 mM, 1 mM, 5 mM and 10 mM). For each survival experiment, at least 2 vials, each containing 50 larvae, were maintained until fly eclosion. The number of eclosed flies was recorded.

Immunohistochemistry and NMJ analysis
Third instar wandering larvae were dissected, fixed, antibody stained, imaged and analysed as described previously [13]. All NMJ analysis was performed double-blind. Primary antibodies detailed in Table 1. Confocal microscopy was performed using a Zeiss LSM 880 on an Axio Observer.Z1 invert confocal microscope (Zeiss). Z-stacked projections of NMJ's and VNCs were obtained using a Plan Neofluar 40x/0.75 NA oil objective. NMJ lengths were measured from stacked NMJ images using the NeuronJ plugin for ImageJ (National Institutes of Health) as described previously [13,54].

Declaration of competing interest
The authors declare no competing interests.