Ask1 and Akt act synergistically to promote ROS-dependent regeneration in Drosophila

How cells communicate to initiate a regenerative response after damage has captivated scientists during the last few decades. It is known that one of the main signals emanating from injured cells is the Reactive Oxygen Species (ROS), which propagate to the surrounding tissue to trigger the replacement of the missing cells. However, the link between ROS production and the activation of regenerative signaling pathways is not yet fully understood. We describe here the non-autonomous ROS sensing mechanism by which living cells launch their regenerative program. To this aim, we used Drosophila imaginal discs as a model system due to its well-characterized regenerative ability after injury or cell death. We genetically-induced cell death and found that the Apoptosis signal-regulating kinase 1 (Ask1) is essential for regenerative growth. Ask1 senses ROS both in dying and living cells, but its activation is selectively attenuated in living cells by Akt1, the core kinase component of the insulin/insulin-like growth factor pathway. Akt1 phosphorylates Ask1 in a secondary site outside the kinase domain, which attenuates its activity. This modulation of Ask1 activity results in moderate levels of JNK signaling in the living tissue, as well as in activation of p38 signaling, both pathways required to turn on the regenerative response. Our findings demonstrate a non-autonomous activation of a ROS sensing mechanism by Ask1 and Akt1 to replace the missing tissue after damage. Collectively, these results provide the basis for understanding the molecular mechanism of communication between dying and living cells that triggers regeneration.

How cells communicate to initiate a regenerative response after damage has captivated scientists during the last few decades. It is known that one of the main signals emanating from injured cells is the Reactive Oxygen Species (ROS), which propagate to the surrounding tissue to trigger the replacement of the missing cells. However, the link between ROS production and the activation of regenerative signaling pathways is not yet fully understood. We describe here the non-autonomous ROS sensing mechanism by which living cells launch their regenerative program. To this aim, we used Drosophila imaginal discs as a model system due to its well-characterized regenerative ability after injury or cell death. We genetically-induced cell death and found that the Apoptosis signal-regulating kinase 1 (Ask1) is essential for regenerative growth. Ask1 senses ROS both in dying and living cells, but its activation is selectively attenuated in living cells by Akt1, the core kinase component of the insulin/insulin-like growth factor pathway. Akt1 phosphorylates Ask1 in a secondary site outside the kinase domain, which attenuates its activity. This modulation of Ask1 activity results in moderate levels of JNK signaling in the living tissue, as well as in activation of p38 signaling, both pathways required to turn on the regenerative response. Our findings demonstrate a non-autonomous activation of a ROS sensing mechanism by Ask1 and Akt1 to replace the missing tissue after damage. Collectively, these results provide the basis for understanding the molecular mechanism of communication between dying and living cells that triggers regeneration.

Author summary
One of the early events that occur after tissue damage consists on the production of Reactive Oxygen Species (ROS), that signal to the surrounding tissue to initiate wound healing and regeneration. Many signaling pathways, such as JNK and p38, respond to oxidative stress and are necessary for regenerative growth. As the link between ROS and

Introduction
Organisms are continuously exposed to a wide variety of environmental stressors that can cause deterioration and cell death. Tissues overcome the effect of those stressors by replacing damaged cells to restore homeostasis. Therefore, understanding the early signals that initiate the response to damage is an essential issue in regenerative medicine. Regeneration can be monitored in Drosophila imaginal discs, which are well-characterized epithelial sacs capable to regenerate after genetically-induced apoptosis or when parts are physically removed (reviewed in [1]). Compiling evidence supports that reactive oxygen species (ROS) fuel wound healing and oxygen-dependent redox-sensitive signaling processes involved in damage response [2][3][4]. Actually, genetically-induced apoptosis in the imaginal discs, using the Gal4/UAS system, leads to the production of ROS which propagate to the surrounding neighbors [5][6][7]. Although oxidative stress has been associated with several pathologies, it has been described that low levels of ROS can be beneficial for signal transduction [8].
After genetic ablation in a specific zone of the wing imaginal disc, dying cells produce high levels of ROS, and show high JNK activity [5,7]. High levels of JNK have been associated to dying cells and are involved in a feedback amplification loop to enhance the apoptotic response to stress [31]. However, the living cells nearby the dying zone show low levels of ROS, which are beneficial for the cell as they turn on the activation of p38 and low levels of JNK [5]. Both MAP kinases in the living cells are required for a cytokine-dependent regenerative growth. A key question is how the balance between the beneficial or detrimental effects of ROS is controlled and, in particular, how ROS control JNK and p38 activity. A candidate molecule to perform this function is the MAPKKK Apoptosis signal-regulating kinase 1 (Ask1), which responds to various stresses by phosphorylation of the kinases upstream JNK and p38 [32][33][34]. Hence, in a reduced environment, thioredoxin (Trx) inhibits Ask1 kinase activity by directly binding to the N-terminal region of Ask1. Upon oxidative stress, the redox-sensitive cysteines of Trx become oxidized, resulting in the dissociation of Trx from Ask1.
Consequently, Ask1 is oligomerized and its threonine-rich kinase domain is phosphorylated, inducing Ask1 activation [35,36]. Mammalian Ask1 is highly sensitive to oxidative stress and contributes substantially to JNK-dependent apoptosis. Nevertheless, recent studies have also revealed other functions of this kinase, including cell differentiation and survival [37].
Ask1-interacting proteins promote conformational changes that lead to the modulation of Ask1 activity and result in various cellular responses. For example, Ask1 is a substrate for phosphorylation by Akt1, a serine/threonine kinase activated by lipid kinase phosphatidylinositide 3-kinase (Pi3K) pathway in response to insulin receptor activation. This phosphorylation is associated to a decrease of Ask1 activity in vitro [38]. The Pi3K/Akt pathway, which is conserved between mammals and Drosophila, is one of the main effector signals for the regulation of tissue growth [39][40][41]. In Drosophila, loss-of-function mutants of various components of the pathway result in reduced body size or lethality [42]. Conversely, mutants of the phosphatase PTEN, an antagonist of Pi3K, result in high activity of Akt and tissue overgrowth [43][44][45]. Thus, it is conceivable that Pi3K/Akt signaling is involved in regenerative growth, but whether Pi3K/Akt and Ask1 interact for controlling regeneration is unknown and deserves attention.
In this work, we genetically induced cell death in wing imaginal discs to explore the link between ROS production and regeneration. We found that Ask1 acts as a sensor of ROS upstream the JNK and p38 pathways. Moreover, we describe that Akt1 is necessary for modulating Ask1 activity in living cells to trigger regeneration. In addition, our results indicate that oxidative stress generated in the damaged cells signals the neighboring living cells to promote tissue repair.

Ask1 is necessary for regenerative growth
The Gal4/UAS/Gal80 TS transactivation system has been extensively used to temporarily and spatially activate pro-apoptotic genes such as reaper (rpr) [10,11,46]. To study the capacity to regenerate, we induced cell death in the wing disc using the wing-specific sal E/Pv -Gal4 strain to activate UAS-rpr, in a temporally controlled manner thanks to the tub-Gal80 TS thermo sensitive Gal4-repressor (henceforth sal E/Pv >rpr). We first kept the embryos at 17˚C until the 8th day (192 hours after egg laying) to activate the tub-Gal80 TS and therefore to prevent rpr expression. Subsequently, we moved those larvae to 29˚C for 11 hours, to inhibit tub-Gal80 TS , and activate apoptosis in the sal E/Pv -Gal4 zone. Then, larvae were returned to 17˚C to avoid further apoptosis and allow tissue to regenerate. After adults emerged, wings were dissected and regeneration was analyzed. With these experimental conditions, 100% of the wings of sal E/Pv >rpr flies contained the full set of veins and interveins, which demonstrates that the missing parts were regenerated ( Fig 1A).
The same experiment was carried out in Ask1 mutant backgrounds. The Ask1 MB06487 mutant flies are viable in homozygosis but lethal over the deficiency Df(3R)BSC636, which suggests that is a hypomorphic allele. The Ask1 MI02915 mutant is lethal in homozygosis as well as over the Df(3R)BSC636, which suggests that is an amorphic allele. We found that after genetically-induced cell death in Ask1 MB06487 or Ask1 MI02915 heterozygous background, full regeneration is only achieved in 15% and 23% of the wings, respectively (Fig 1A). The phenotypes of the abnormally regenerating wings vary and consisted in lack of some veins or parts of a vein or anomalous vein segments (S1A Fig).
To further address the role of Ask1 in regeneration, we used a double transactivation system to simultaneously induce cell death in the sal E/Pv domain and inhibit Ask1 with UAS-RNAi (UAS-Ask1 RNAi ) in an adjacent compartment (Fig 1B). RNAi knockdown of Ask1 (UAS-Ask1 RNAi ) reduces Ask1 mRNA total levels to approximately 25-30 percent of that observed in controls (S1B Fig). The UAS-Ask1 RNAi transgene was activated in the anterior compartment using the Gal4/UAS system (ci-Gal4>UAS-Ask1 RNAi ) and cell death was induced in the sal E/Pv domain using the Gal80-repressible transactivator system LHG (LexA-Hinge-Gal4 activation domain), a modified form of the lexA lexO system (sal E/Pv -LHG>lexO-rpr) ( Fig 1B) [5,47]. Both transgenes were activated at the same time, following the same protocol as in the previous experiment. The resulting adult wings (ci>Ask1 RNAi sal E/Pv >rpr) lacked some veins or interveins and their size was reduced in all cases observed. However, the inhibition of Ask1 in the anterior compartment for the same time but without cell death, did not affect vein pattern nor wing size (ci>Ask1 RNAi ) ( Fig  1B). Similar results were obtained after driving UAS-Ask1 RNAi to a different compartment (dorsal) and inducing cell death in sal E/Pv (ap>Ask1 RNAi sal E/Pv >rpr); in this case, we found that 82% of wings could not regenerate properly. As for the anterior compartment, Ask1 RNAi in the dorsal compartment without cell death did not affect vein pattern nor wing size (ap>Ask1 RNAi ) (Fig 1B). To discard any toxicity due to the insertion of the transgene, all experiments were carried out in parallel but constantly at 17˚C to maintain tub-Gal80 TS activity and block transgene (UAS-or lexO-) expression. In these conditions no defects in the wings were detected (sal E/Pv > OFF in Fig  1A and 1B).
To evaluate whether these anomalies were due to impairment of proliferation, we analyzed the mitotic index, calculated as the number of cells positive for the phosphorylated form of Ask1 sensor for regeneration Histone 3 (P-H3) in the anterior compartment (ci>), where the Ask1 RNAi transgene was expressed. It is well known that apoptosis in discs induces compensatory proliferation of nearby cells (reviewed in [1]). Accordingly, genetic ablation (sal E/Pv >rpr ci>GFP) resulted in an increase of mitosis compared to the neutral (sal E/Pv >GFP ci>RFP) or to the Ask1 knockdown (sal E/Pv >GFP ci>Ask1 RNAi ) conditions. However, combining genetic ablation in the sal E/Pv zone with Ask1 knockdown in the nearby anterior compartment, the number of mitosis associated to damage did not increase (sal E/Pv >rpr ci>Ask1 RNAi ) (Fig 1C). This observation confirms that Ask1 reduction impairs regenerative growth.

Ask1 activity is higher in apoptotic cells than in regenerating cells
The Drosophila Ask1 locus encodes two protein isoforms of different lengths, Ask1-RB and Ask1-RC, with predicted molecular weights of 136.5 kDa and 155.5 kDa, respectively. Both have a protein kinase-like domain that contains a highly conserved core of threonines conferring functionality to the protein (reviewed in [32]). It has been reported that after Trx release, the Ask1 oligomer undergoes conformational change leading to trans-autophosphorylation of the threonine residue corresponding to Thr838, Thr845 and Thr747 in human, mouse and Drosophila Ask1, respectively [32]. In human cells, phosphorylation of the N-terminus Ser83 by Akt results in attenuation of Ask1 activity in vitro [38]. This is noteworthy, because attenuation of Ask1 could result in moderate levels of activity necessary for turning on JNK and p38 in living cells. Drosophila Ask1 lacks this residue and the Akt consensus motif in the N-terminus (S2 Fig, S1 Appendix, S2 Appendix). However, we found that the longer Ask1-RC isoform contained a conserved domain of unknown function DUF4071 with a highly conserved Ser (human Ser174, Drosophila Ser83), which we hypothesized that could be key in Ask1 regulation (Fig 2A, S2 Fig).
Active Ask1 can be traced with antibodies against the phosphorylated threonine residues of the highly conserved kinase domain (henceforth P-Thr) [48] and the attenuated form with specific antibodies against phosphorylated Ser83 (P-Ser83). We first tested both antibodies in wild-type wing imaginal discs. We detected low levels of P-Thr all over the disc including some transient high activity during mitosis ( Fig 2B and S3A Fig), as occurs in dividing mammalian cells [49]. In Ask1 mutant backgrounds, both the general and the mitosis-associated P-Thr levels were reduced but not abolished (S3B Fig), and Ask1 RNAi discs did not show any effect on P-Thr levels in mitosis ( Fig 2E). This discrepancy may be due to the hypomorphic condition of the mutant combinations, but also to that the antibody may recognize some unspecific epitope in mitotic cells. We also found basal levels of P-Ser83 in wild-type discs ( Fig  2C). Ectopic expression of Ask1 RNAi suppressed the low endogenous P-Ser83 Ask1 levels (S3C Fig).
Remarkably, upon apoptosis, high levels of P-Thr were localized in dying cells ( Fig 2D) and absent or, in some cases, increased weakly above the basal levels in nearby living cells. This increase of P-Thr in apoptotic cells was inhibited after Ask1 RNAi expression ( Fig 2E). In contrast, P-Ser83 accumulated in living cells adjoining the apoptotic zone and was completely absent in apoptotic cells. The increase in P-Ser83 varied from discs with strong accumulation near the dying domain to those with an extended increase in the whole wing pouch (2D and S4A Figs). Similar results were obtained after killing cells with a different pro-apoptotic gene (sal E/Pv >hid) (S4B Fig). In the presence of apoptosis and blocking Ask1, the P-Ser83 increase in living cells was inhibited (sal E/Pv >rpr ci>Ask1 RNAi , Fig 2F). Moreover, P-Ser83 was also found to be elevated at the wound edges after physical injury (S4C Fig). Together, these observations indicate that living cells respond to damage by phosphorylation of Ask1 Ser83. In contrast to the high activity of Ask1 in dying cells (high P-Thr), the presence of P-Ser83 in living cells could be indicative of tolerable levels of Ask1 achieved by attenuation of P-Thr activity. To test this hypothesis, we mutated Ask1 at serine 83 to alanine and cloned it into a UAS vector (UAS-Ask1 S83A ). In parallel, we also cloned a wild-type form of Ask1 (UAS-Ask1 WT ). We ectopically expressed Ask1 WT in the posterior compartment of the wing disc (hh-Gal4 UAS-Ask1 WT ), which resulted in an increase of P-Ser83, in addition to low levels of P-Thr ( Fig 2G). Interestingly, the ectopic expression of Ask1 S83A in the posterior compartment (hh-Gal4 UAS-Ask1 S83A ), did not show a rise in P-Ser83, but strong activation of P-Thr ( Fig 2H). The high levels of P-Thr were not homogeneously distributed, varied from disc to disc, and were always found within the posterior compartment (hh>), where the transgene was activated. This observation concurs with the P-Ser83 residue as responsible for the attenuation of P-Thr.
We next analyzed whether the Ser83 residue was key for the damage response. Using the double transactivation system, we found that the expression of UAS-Ask1 WT or UAS-Ask1 S83A transgenes in the anterior compartment without cell death did not cause any defects in wing morphology, size or vein and intervein patterning (sal E/Pv >GFP ci>Ask1 WT or Ask1 S83A in Fig  2I). After genetic ablation, expression of UAS-Ask1 WT resulted in full regeneration in 89% of wings (sal E/Pv >rpr ci>Ask1 WT ). In contrast, expression of UAS-Ask1 S83A led to a fall to only 29% of individuals being capable of regenerate and the rest showed strong effects on wing morphology, with altered pattern of veins and interveins, as well as the appearance of notches, which together are indicative of disrupted regeneration (sal E/Pv >rpr ci >Ask1 S83A in Fig 2I). The suppression of the ability to regenerate by UAS-Ask1 S83A , which may act as a dominant negative allele, is likely due to the lack of non-autonomous P-Ser83 increase in the living cells near the damaged zone.

Pi3K/Akt signaling attenuates Ask1 activity in living cells
Next, we decided to identify the upstream signal responsible for Ser83 phosphorylation. In mammalian cells, attenuating phosphorylation of Ask1 is driven by the serine-threonine Akt kinase, the core kinase of the insulin pathway [38]. We wondered whether the Drosophila Akt1 as well as its upstream Pi3K92E kinase (the Drosophila Pi3K kinase also known as dp110), were required for Ser83 phosphorylation.
We first found that the active phosphorylated form of Akt (P-Akt) was increased in the living tissue and decreased or was absent in the dying zone (Fig 3A-3D). To test whether the Akt1 phosphorylated the endogenous Ask1 Ser83 in vivo, we specifically inhibited this kinase with a UAS-Akt RNAi in a stripe of cells at the center of the disc, the ptc domain (ptc>Akt RNAi ), and found that P-Ser83 was reduced (Fig 3E and 3F). In addition, ectopic activation of Akt1 (UAS-myr-Akt1.S), a constitutively activated membrane-anchored form of Akt1 [50], resulted in an increase in the levels of P-Ser83 (Fig 3G and 3H).
We next studied the role of P-Akt on Ask1 in the context of regeneration. In the presence of cell death and blocking Pi3K92E, with the dominant negative form dp110 DN in the adjacent compartment, we observed a reduction in the accumulation of P-Akt and P-Ser83 induced after genetic ablation (sal E/Pv >rpr hh> dp110 DN in Fig 3I-3K). Moreover, in the overlapping zone of dying cells and dp110 DN , the accumulation of P-Thr was unaffected ( Fig 3L and S5  Fig). We also scored the effects on wing regeneration after blocking Akt and Pi3K92E kinases. Regeneration was impaired when the dominant negative form of Pi3K92E was expressed in the anterior compartment (sal E/Pv >rpr ci>dp110 DN ) as well as in the Akt1 1 heterozygous background (sal E/Pv >rpr Akt 1/+ ), an allele that encodes a catalytically inactive protein [51] ( Fig  3M). Moreover, regeneration was severely affected in double heterozygous flies containing Akt1 1 and Ask1 MB06487 or Ask1 MI02915 alleles ( Fig 3M). Neither the dominant negative form of Pi3K92E nor the allelic combination Akt1 1 and Ask1 MB06487 or Ask1 MI02915 affected wing development in the absence of cell death (sal E/Pv >OFF wings in Fig 3M). These results further support the notion that Pi3K92E/Akt1 and Ask1 genetically interact and that their epistatic interaction is key to drive regeneration.

ROS promote Ask1 activation
Next, we modulated the ROS levels in order to determine if Ask1 can sense oxidative stress after damage. We fed sal E/Pv >rpr larvae with food supplemented with N-acetyl cysteine (NAC), a potent non-enzymatic scavenger that decreases ROS production, and examined Ask1 phosphorylation (Fig 4A). After induction of cell death, we found a significant decrease in both P-Thr and P-Ser83 in discs from NAC-fed larvae (Fig 4B and 4C; S3 Appendix). In addition, we fed larvae with H 2 O 2 -supplemented food and observed a significant increase in both P-Thr and P-Ser83 levels (Fig 4D-4F). This oxidative stress-induced increase was blocked in the hypomorphic Ask1 MB06487 homozygous mutant (Fig 4E and 4F; S3 Appendix). It is known that in mammalian cells Ask1 can also be activated by endoplasmic reticulum (ER) stress [52]. Feeding larvae with tunicamycin, an inhibitor of N-glycosylation in the ER that induces ER stress, led to an increase in Ask1 activation, which was also blocked in Ask1 MB06487 mutant discs (Fig 4E and 4F; S3 Appendix).
Furthermore, we examined whether Akt1 activation in the living cells is targeted nonautonomously by ROS produced by the dying cells. To examine this hypothesis, we enzymatically blocked ROS production using ectopic expression of the ROS scavengers Superoxide dismutase 1 and Catalase (Sod1:Cat) in the rpr ablated region (sal E/Pv >rpr, Sod1: Cat) and monitored the pixel intensities of two adjacent zones in each disc, the anterior (A) and posterior (P) to the sal E/Pv (Fig 4G). The activation of the Sod1:Cat transgene concomitantly with sal E/Pv >rpr did not inhibit apoptosis and P-Thr was detected in the dying cells ( S6 Fig). However, in these conditions, we found that P-Akt leveis in neighboring cells were reduced in both A and P zones when Sod1:Cat was co-expressed with rpr. Accordingly, we found that the increase of P-Ser83 in neighboring cells was blocked when Sod1:Cat was co-expressed with rpr ( Fig 4G; S3 Appendix). Together, these results demonstrate that the oxidative stress generated from the dying cells targets Akt1 and Ask1 in the neighboring undamaged tissue.

Ask1 activates p38 and JNK in regeneration
We have shown that the Ask1 P-Ser83 attenuating signal is essential for regeneration, and that this signal depends on ROS and Akt. We next hypothesized that this P-Ser83-mediated attenuation does not silence Ask1, but instead maintains the low levels of Ask1 activity necessary for regeneration. Therefore, we analyzed the expression of its known targets, the stress-activated MAP kinases JNK and p38 [48,53], both required for ROSdependent regeneration [5,6]. We induced cell death in the sal E/Pv domain and simultaneously blocked Ask1 in the anterior compartment, and used the posterior as an internal control for the same disc (ci>Ask1 RNAi sal E/Pv >rpr). We found most phosphorylated p38 (P-p38), as an indicator of p38 activity, in the posterior compartment (Fig 5A and 5B). The effects of Ask1 on JNK activity were monitored by matrix metalloproteinase 1 (Mmp1) expression as it is one of the bona fide transcriptionally regulated read-outs of the JNK pathway [54]. After sal E/Pv >rpr cell death, Mmp1 was found in both compartments, but in ci>Ask1 RNAi sal E/Pv >rpr discs Mmp1 mostly accumulated in the posterior (Fig 5C and 5D). The effects of Ask1 interference on JNK pathway were strengthened after staining with P-JNK antibody. P-JNK was found accumulated around the dead zone in sal E/Pv >rpr discs. However, when Ask1 was knockdown in the anterior compartment (ci>Ask1 RNAi sal E/Pv >rpr discs), P-JNK was reduced in this compartment and accumulated in the posterior (S7A and S7B Fig). To confirm that Ask1 is responsible for JNK and p38 activation, we tested whether Ask1 was required for their activation after physical damage. We made two incisions in hh>GFP,Ask1 RNAi discs, one in the UAS-Ask1 RNAi compartment and another into the control compartment. We found that P-p38 localized at the wound edges in the control compartment, whereas the levels of P-p38 were reduced at the wound edges of the Ask1 RNAi compartment (Fig 5E and 5G). Likewise, cut Ask1 RNAi discs resulted in less Mmp1 than in the control compartment (Fig 5F and 5H). In addition, we tested the puckered (puc) reporter of the JNK pathway, which normally accumulates in wounds, and found to drop in heterozygous Ask1 discs after a physical injury (S7C and S7D Fig). Together, these results indicate that Ask1 is necessary for the activation of p38 and JNK during regeneration.

Discussion
We have shown here that Ask1 acts as a sensor of ROS after damage, and that synergizes with Pi3K/Akt1 to phosphorylate the Ask1 Ser83, a key phosphorylation event to initiate regeneration. In addition, we have demonstrated that the activation of the Pi3K/Akt1 and Ask1 in undamaged cells is originated non-autonomously by ROS produced by the damaged tissue.
An essential question to understand regeneration is how damaged cells communicate to the nearby living tissue to initiate regenerative growth. One of the most classical and determining studies on epithelial regeneration was the discovery that massive cell death caused by ionizing radiation in Drosophila larvae resulted in proliferation of the unaffected neighbors to compensate the lost [55]. Since then, compensatory proliferation, a cellular response linked to regeneration, has been considered as a result of signals released from apoptotic cells (reviewed in [56]). We propose here that the oxidative stress generated by damaged or apoptotic cells signals undamaged tissue to initiate the Ask1/Akt1 machinery that will culminate with repair and compensatory proliferation.
ROS acting in signaling after wounding has been subject of extensive research in various organisms and tissues [5,6,[64][65][66][67]26,[57][58][59][60][61][62][63]. Here, we have uncovered two scenarios of Ask1 activity: high activity in apoptotic cells, with high levels of P-Thr; and low activity in undamaged cells, where P-Ser83 prevails over P-Thr. This fits with the high levels of ROS and JNK in dying cells and the low levels of ROS, low JNK and P-p38 in living cells reported previously [5]. Together, our observations address the question of how JNK signaling selectively fosters apoptosis or proliferation, and how this signal is related to high or low ROS levels. The same applies to p38, which is only activated in unharmed cells when neighbors enter in apoptosis or are damaged [5,53].
Our study demonstrates that Ask1 operates in an Akt1-dependent manner in living cells and in an Akt1-independent manner in dying cells. It is conceivable that Akt1-mediated attenuation results in either low or transient levels of Ask1 activity necessary to stimulate P-p38 and low JNK levels for regeneration. In contrast, apoptotic cells induce high or sustained levels of JNK because high levels of ROS result in high Ask1 P-Thr in the absence of Akt1 attenuation.
Based on the results presented here and previous observations, we propose a model in which oxidative stress and the absence or presence of Pi3K/Akt1 survival signals are integrated in Ask1 to control the selective activity of JNK and p38, and in turn regenerative growth ( Fig  5I). In this model, only dying cells produce high ROS levels and high JNK activity [5,7]. In addition, other mechanisms can exacerbate apoptosis through the lethal levels of JNK, and therefore increase ROS production. For example JNK, as well as p53, functions in a positive feedback after apoptosis as it transcriptionally activates the pro-apoptotic genes hid and rpr, thus amplifying apoptosis [31,68]. It is therefore likely that the high ROS that result in high Ask1 activation, inabilities the dying cells to respond to Pi3K/Akt1, preventing the Ask1 attenuation and resulting in deleterious JNK activity. We also have demonstrated that P-Akt1 is absent, or low, in dying cells, which supports a lack of Pi3K/Akt1-dependent Ask1 attenuated activity in dying cells. These cells will enter into a committed-to-die status with no return to survival. In contrast, the living neighbor cells, albeit adjacent to dying cells, show tolerable levels of ROS [5,7], and activate moderate levels of Ask1 through the Akt1-dependent phosphorylation of Ask1 P-Ser83. In these cells, Ask1 will turn on the activation of tolerable levels of JNK and p38, leading to a committed-to-regenerate status. Eventually, the JNK and p38 pathways will promote the transcription of the unpaired cytokines necessary for survival and regeneration [5,15,[69][70][71][72]. In addition to cytokines, Wnt/Wg signaling pathway acts as a mitogenic signal involved in wing disc regeneration that responds to JNK [11,18,69,73] as well as the inhibition of the Hippo pathway, which results in increased Yki activity [74][75][76].
Several phosphatases are sensitive to ROS, and could be also integrated into the model presented here. This is the case of the JNK phosphatases, which are critical molecular targets of ROS. Oxidative inhibition of JNK phosphatases results in sustained JNK activation in order to promote caspase 3 cleavage and eventually cell death [77]. Therefore, the early burst of ROS in dying cells could activate lethal levels of JNK activity through inhibition of the Drosophila JNK phosphatase puckered concomitantly with the high levels of Ask1, and together boost the committed-to-die status. Additionally, the PTEN phosphatase, which regulates many cellular processes through direct antagonism of Pi3K signaling, is also ROS sensitive. Increasing ROS can enhance insulin signaling attributable to the oxidative inhibition of PTEN [78]. PTEN inactivation enhances the activation of AKT signaling, which in turn promotes the expression of cell survival genes [78]. Thus, PTEN could be involved in the living cells to release the Akt1 activity that is necessary to reach the tolerable levels of Ask1/JNK/p38. Therefore, the low oxidative stress that reaches the living cells can trigger not only Trx2 dissociation from Ask1, but also the PTEN-oxidation dependent Pi3K/Akt1 activity required for attenuated Ask1 necessary for survival and proliferation.
Our model concerns to an inherent response in the imaginal disc epithelium. However, other signals could enhance or stabilize the ROS-dependent regenerative response. For example, ROS produced after wounding are necessary for recruitment and activation of the Drosophila macrophage-like immune cells, which will secrete the TNF ortholog Eiger. Eiger activates JNK signaling in the epithelial cells to contribute to the regenerative response [6].
Because the moderate activity of Ask1 for regeneration is achieved in an Akt1-dependent manner, and because Akt1 functions downstream of the insulin signaling, it seems conceivable an implication of nutrition in imaginal disc regeneration. The production and secretion of insulin and AKH, the fly analog of mammalian glucagon, vary depending of the food content. Thus, a challenge for the future will be to analyze how these hormones can modulate the regenerative response to damage. For example, the Drosophila TNF Eiger is produced by the fat body when larvae are exposed to low protein diet [79]. As Eiger activates JNK non-autonomously in many tissues, it could have a central role in imaginal disc regeneration, depending on food availability. The relation between nutrition and metabolism, as well as their involvement in regeneration deserves further analysis in order to achieve an integrative view of regeneration.

Genetic ablation and dual Gal4/LexA transactivation system
Cell death was genetically induced as previously described [5]. We used the sal E/Pv -Gal4 as a driver, which consists of the spalt wing enhancer with expression confined to the wing to score adult wing parameters. The UAS lines used to promote cell death were UAS-rpr or UAS-hid, two pro-apoptotic genes, controlled by the thermo-sensitive Gal4 repressor tubGal80 TS . We also used the sal E/Pv -LHG and LexO-rpr strains [5] for genetic ablation, utilizing the same design as for Gal4/UAS.
Embryos were kept at 17˚C until the 8th day/192 hours after egg laying to prevent rpr expression. They were subsequently moved to 29˚C for 11 hours and then back to 17˚C to allow tissue to regenerate. Two types of controls were always treated in parallel; individuals without rpr expression (UAS-GFP, moved to 29˚C for 11 hours) and individuals kept continuously at 17˚C to avoid any transgene activation.
In most of the experiments shown here, we induced cell death for 11h. However the induction of cell death for testing the effects of signals emerging from the dying zone was longer. This is the case of the P-Akt experiment in Fig 3B and 3C and for the genetic scavenging of ROS (Sod1:Cat) (Fig 4G and S6 Fig), in which sal>rpr was activated for 24h.
In dual transactivation experiments, we used sal E/Pv -LHG LexO-rpr to ablate the sal E/Pv domain (abridged as sal E/Pv >rpr), whereas Gal4 was used to express different transgenes under the control of Gal4 drivers (ci-Gal4 for anterior compartment; ap-Gal4 for dorsal compartment and hh-Gal4 for posterior compartment).

Test for regenerated adult wings and statistics
To test the capacity to regenerate in different genetic backgrounds, we used adult wings emerged from sal E/Pv >rpr individuals in which patterning and size defects can be scored easily. Flies were fixed in glycerol:ethanol (1:2) for 24 hours. Wings were dissected in water and then washed with ethanol. Subsequently, they were mounted in lactic acid:ethanol (6:5) and analyzed and imaged under a microscope.
The percentage of regenerated wings refers to fully regenerated (for genetic ablation genotypes) or normally developed wings (for testing transgenes) and was calculated according to the number of wings with a complete set of veins and interveins, as markers of normal patterning. For each sample, we scored the percentage of individuals belonging to the "regenerated wings" class. We calculated the standard error of the sample proportion based on a binomial distribution (regenerated complete wing or not) SE = p p (1-p)/n, where p is the proportion of successes in the population. Strikingly the human Ser83 site was not found on Drosophila melanogaster; which was confirmed by using GeneWise (version wise2.4.1) [80] to map the M3K5_HUMAN protein sequence (Q99683), downloaded from UniProt [81], over a window of 10kbp centered at the Ask1 transcription start site; this sequence segment was retrieved from FlyBase FB2017_06 genome version ( [82], chr3R:19,875,000-19,885,000[+]). The GeneWise output is provided as S1 Appendix, the software tool was not able to find a match on the fly genome for the initial 126 residues from the human amino term. In order to assess the relevance of the fly Ser83 two homology-based approaches were implemented. First, a search on PFAM database [83] returned the domain of unknown function DUF4071 (PF13281). The first amino acid positions for this domain are highly conserved across a set of diverse homologs, including fly Ask1-PC and human MAP3K5 but also paralogs sharing the domain. The consensus sequence for DUF4071 has been included on the right alignment block from S2 Fig for illustrative purpose. The second approach was considering a precomputed set of homolog sequences downloaded from EggNOG database (version 4.5) [84]. A cluster of homologous sequences already included fly Ask1 (KOG4279), providing a total of 338 protein sequences for 164 species; a total of 49 sequences were selected from that set, fly Ask1-PC isoform among them, that were also reliably annotated as orthologs of human MAP3K5. Fly Ask1-PB isoform protein sequence was retrieved from FlyBase and appended to the selected MAP3K5 orthologs. Then, the 50 protein sequences were aligned by MAFFT (version 7.271) [85] using the following parameters: maxiterate = 1000, localpair, op = 10. The complete alignment is available from S2 Appendix, from which two conserved blocks centered on the human and fly Ser83 domains were trimmed and processed with TeXshade [86] to produce the S2 Fig.  Note that on S2 Fig the domain highlighted in cyan, which includes the fly equivalent column highlighted in red, is highly conserved in a wide range of taxa, from sponges to humans. Other domains are conserved only within chordata or even tetrapoda species. Such conservation signature makes Drosophila Ser83 a good candidate for functional and mutational studies.
pUASt-attb_Ask1 S83A . Two serines in the 75 and 83 residues in the DUF4071 domain showed putative non-canonical AKT phosphorylation sites (ILTQQRPLSYHYGVRESF). Both residues are spatially exposed similarly to human Ser83 of ASK1, which makes them accessible to kinases. pUASt-attb_Ask1 S83A was constructed by mutating serine 83 to alanine by PCR using oligos Ask1Mut-Fwd and Ask1S83A-Rev for partial PCR1 and Ask1S83A-Fwd and Ask1Mut-Rev for partial PCR2. Complete PCR was performed using the two partial PCRs as templates with oligos Ask1Mut-Fwd and Ask1Mut-Rev. The complete PCR was then cut with EcoRI /PflMI and cloned into FI02066-cut EcoRI/PflMI. The mutation introduced a new StuI site that was used to check the mutated clones. Mutated Ask1S83A cDNA was then cut from EcoRI/BamHI and cloned into pUASt-attb. Both clones were injected by standard procedures in line zh-86Fb-attP and transgenic lines were selected. Ask1Mut
The number of mitotic cells per mm 2 (Fig 1C) was calculated after counting the number of P-Histone-H3 (P-H3) positive cells in the anterior compartment of the wing pouch and hinge, excluding the notum, for all the genotypes shown. Note that two genotypes in Fig 1C show the anterior compartment labeled with the transgene ci-Gal4 UAS-GFP (ci>GFP), whereas in the two ci>Ask1 RNAi experiments the anterior compartment was labeled with anti-Ci. In these experiments shown in Fig 1C, the induction of cell death or activation of transgenes was done for 16h at 29˚C. The number of mitosis after analyzing the stacks of confocal images was calculated using Fiji software.

Imaginal disc culture and physical injury
Wing discs were dissected from third instar larvae in Schneider's insect medium (Sigma-Aldrich), and a small fragment was removed with tungsten needles. To visualize Mmp1 staining, the discs were cultured in Schneider's insect medium supplemented with 2% heat-activated fetal calf serum, 2.5% fly extract and 5 μg/ml insulin, for 5 hours at 25˚C. For P-p38, discs were injured in Schneider's insect medium and immediately fixed and stained. Ex vivo images were taken using a Leica SPE confocal microscope and processed with Fiji software.
Ex vivo discs were also monitored live after a cut, using puc-Gal4 UAS-GFP as a JNK reporter, using the same medium. For nuclei visualization in ex vivo culture, we used NucRed Live647 (Life Technologies; 1 drop per slide) added 20 minutes before imaging.

ROS scavenging
To prevent ROS production in sal E/Pv >rpr discs we pursued two different protocols. First, we inhibited ROS chemically (Fig 4A-4F). To do this, standard fly food was supplemented with the antioxidant N-acetyl cysteine (NAC 100 μg/ml; Sigma-Aldrich). NAC treatment was dispensed on the 7 th day of development at 17˚C. On the 8 th day, experimental larvae were moved to 29˚C for 11 hours to promote cell death, whereas controls were transferred to a vial with standard food and moved to 29˚C for the same time period. Afterward, the larvae were move back to 17˚C to allow tissue recovery. Second, we decreased ROS production genetically by the ectopic expression of the Sod1 and Catalase enzymes using a recombinant fly UAS--Sod1:UAS-Cat (Sod1:Cat) in the sal E/Pv >rpr domain (Fig 4G and S6 Fig). For those experiments, Sod1:Cat and rpr were activated for 24 hours at 29ºC.

Oxidative stress induction
Third instar larvae were transferred to vials with 5mL of special medium containing 1.3% UltraPure TM LMP agarose (Invitrogen), 5% sucrose (Fluka) and the desired concentration of 0.1% H 2 O 2 (Merck) or 1ng/μl tunicamycin (Sigma-Aldrich). To avoid loss of oxidative capacity, these substances were added to the media at a temperature below 45˚C. The larvae were fed for 2 hours prior to dissection and fixation of the discs. Controls without H 2 O 2 or tunicamycin were always handled in parallel.

Quantitative real-time PCR analysis
Control (hh>RFP) and experimental (hh>Ask1 RNAi ) conditions were induced for 16h at 29˚C (192h after egg laying) using the Gal4/tub-Gal80 TS system. Wing imaginal discs (n = 40) were dissected after induction in three biological replicates for each condition. RNA was extracted with the Zymo Research ZR RNA MicroPerp (R1060/R1061) and RNA Clean and Concentrator (R1015/R1016) kits following standard protocols. RT-PCR was performed using SYBR green Master Mix (Roche). Specific primers for Ask1 and Rps18 are detailed in S1B Fig. ΔΔCt method was used to normalize the data. Ask1 RNAi and control samples were normalized against the housekeeping gene Rps18. Average standard error of the mean (SEM) of the three biological replicates was computed for each one based on three technical replicates by ΔΔCt method.