JAK/STAT signalling mediates cell survival in response to tissue stress

Tissue homeostasis relies on the ability of tissues to respond to stress caused by damaging environmental insults. Physical wounding, toxins, reactive oxygen species and UV irradiation all induce cellular damage and thus disrupt tissue integrity. Mounting an appropriate response to these insults is essential for tissue repair and to prevent chronic cellular stress, which can lead to disease (Fulda et al., 2010). While much progress has been made to elucidate signalling pathways that seal wounds, remove damaged cells, promote regenerative proliferation or mediate patterning of regenerated tissue, little is known about how crosstalk between these pathways coordinates repair processes to restore homeostasis.

Studies on Drosophila progenitor organs called imaginal discs have provided deep insights into cellular adaptations to tissue stress. Surgical excision (Bryant, 1975; Haynie and Bryant, 1976; Katsuyama et al., 2015) or cell ablation induced by pro-apoptotic transgenes (Grusche et al., 2011; Herrera et al., 2013; Smith-Bolton et al., 2009) has revealed cellular responses that promote wound healing (Sun and Irvine, 2014; Kashio et al., 2014; Razzell et al., 2011). Disc size is restored by accelerated proliferation of cells proximal (Sustar et al., 2011; Bosch et al., 2008) and distal (Herrera et al., 2013) to the wound site. Lacking tissue-resident stem cells, fate plasticity displayed by parenchymal disc cells facilitates repatterning of replaced tissues (Herrera et al., 2013; Herrera and Morata, 2014; Repiso et al., 2013; Schuster and Smith-Bolton, 2015). In contrast, regeneration of the Drosophila adult midgut is driven by tissue-resident stem cells (Ohlstein and Spradling, 2006; Jiang et al., 2009; Osman et al., 2013; Staley and Irvine, 2010). Strikingly, while midgut and imaginal discs utilize stem cell-dependent and -independent repair processes, both tissues activate similar signalling pathways that underlie potentially highly conserved stress responses.

The c-Jun amino-terminal kinase (JNK)/mitogen-activated protein kinase (MAPK) cascade is one of the earliest pathways activated in damaged tissues and it is triggered by loss of epithelial polarity (Brumby and Richardson, 2003; Uhlirova et al., 2005; Igaki, 2009; Wu et al., 2010; Zhu et al., 2010), apoptosis (Ryoo et al., 2004; Shlevkov and Morata, 2012) or physical wounding (Bosch et al., 2005; Rämet et al., 2002; Lee et al., 2005). JNK regulates stress responses via activation of the transcription factor AP-1 (Eferl and Wagner, 2003), which comprises Jun and Fos homo- and heterodimers. JNK activation is required for cytoskeletal rearrangement during wound closure (Ríos-Barrera and Riesgo-Escovar, 2013; Bosch et al., 2005) and promotes elimination of damaged cells by injury-induced apoptosis (Bogoyevitch et al., 2010; Chen, 2012; Shlevkov and Morata, 2012; Luo et al., 2007; Moreno et al., 2002). Importantly, JNK drives compensatory proliferation to replace damaged tissues, through cell-autonomous (Bosch et al., 2008; Sun and Irvine, 2011; Grusche et al., 2011) and non-autonomous (Bergantinos et al., 2010; Mattila et al., 2005; Ryoo et al., 2004) mechanisms. A central question remaining unanswered is how these two different responses – proliferation and apoptosis – are brought into equilibrium downstream of JNK to restore tissue homeostasis.

Apoptotic JNK targets that facilitate clearing of damaged cells include the Diap1 inhibitors reaper (rpr), head involution defective (hid) and p53 (Shlevkov and Morata, 2012; Luo et al., 2007; Moreno et al., 2002). Activation of rpr, hid and p53 drives further activation of JNK via the initiator caspase Dronc (Shlevkov and Morata, 2012). JNK also activates the TNFα homologue Eiger (Pérez-Garijo et al., 2013), leading to activation of JNK in nearby cells (Pérez-Garijo et al., 2013; Bergantinos et al., 2010; Wu et al., 2010). This positive feedback could drive excessive cell elimination and therefore, needs to be restrained to prevent unlimited spatial propagation of pro-apoptotic JNK activity. It is not known how spatial constraints on JNK propagation are established.

Importantly, apoptosis of damaged cells stimulates compensatory proliferation by JNK-dependent activation of growth pathways including Wnt/β-catenin, TFGβ/SMAD, Janus kinase/Signal transducer and activator of transcription (JAK/STAT), Hippo/Yorkie and EGF (Sun and Irvine, 2011, 2014; Wells et al., 2006; Kondo et al., 2006; Huh et al., 2004; Morata et al., 2011; Pastor-Pareja and Xu, 2013). While the requirement for these pathways differs depending on context (Smith-Bolton et al., 2009; Herrera et al., 2013; Fan and Bergmann, 2008), many have been implicated in metazoan responses to stress (Sun and Irvine, 2014; Pahlavan et al., 2006; Chen, 2012). Importantly, prevention of apoptosis in damaged cells causes sustained proliferative signalling and eventual tumour formation in Drosophila (Kondo et al., 2006; Perez-Garijo et al., 2004, 2009; Ryoo et al., 2004; Huh et al., 2004; Martin et al., 2009), emphasizing how strongly apoptotic, survival and proliferative signals need to be balanced to re-establish normal tissue size.

Current models propose that JAK/STAT signalling is a major mediator of compensatory, cancerous and developmental proliferation. Cytokine-like ligands in Drosophila are encoded by upd1 (unpaired 1), upd2 and upd3 and signal through conserved pathway components encoded by single genes, specifically the receptor Domeless (Dome), the Janus kinase Hopscotch (Hop) and the transcription factor Stat92E (Arbouzova and Zeidler, 2006). As in mammals, JAK/STAT plays diverse roles in fly development by regulating haemocyte activation (Myllymäki and Rämet, 2014), appendage patterning (Ayala-Camargo et al., 2013, 2007; Johnstone et al., 2013) and stem cell maintenance (Gregory et al., 2008). A proliferative function for JAK/STAT was suggested by early studies of tissue growth during disc development (Bach et al., 2003; Tsai and Sun, 2004; Mukherjee et al., 2005). Subsequently, JAK/STAT has been widely implicated in promoting proliferation of neoplastic cells (Classen et al., 2009; Wu et al., 2010; Davie et al., 2015; Bunker et al., 2015; Amoyel et al., 2014) and aberrant non-autonomous proliferation of wild-type cells in fly models of tumorigenesis (Vaccari and Bilder, 2005; Herz et al., 2006; Moberg et al., 2005). Activation of Upd gene transcription upon tissue damage has been linked to compensatory proliferation in imaginal discs and adult guts (Jiang et al., 2009; Pastor-Pareja et al., 2008; Katsuyama et al., 2015; Wu et al., 2010; Lin et al., 2009; Bunker et al., 2015; Santabárbara-Ruiz et al., 2015). Other studies suggest that JAK/STAT is involved in cell competition, where signalling either promotes ‘winner’ cell state (Schroeder et al., 2013; Rodrigues et al., 2012) or compensatory proliferation in response to ‘loser’ cell elimination (Kolahgar et al., 2015).

Despite this wealth of work, little is known about target genes mediating proliferative JAK/STAT function. In fly tumours, STAT-responsive enhancers have been mapped (Davie et al., 2015), but few target genes regulating tissue size have been described (Tsai and Sun, 2004; Betz et al., 2008; Hasan et al., 2015). However, several studies suggest that developmental functions of JAK/STAT are mediated by the transcriptional repressors Chinmo (Flaherty et al., 2010), Zfh1 (Ohayon et al., 2009; Leatherman and Dinardo, 2008) and Zfh2 (Perea et al., 2013; Guarner et al., 2014; Ayala-Camargo et al., 2013).

We wanted to better understand the precise role of JAK/STAT in cellular adaptations to stress and to gain further insight into how JAK/STAT might regulate compensatory proliferation after tissue damage. Because JAK/STAT is required for regeneration in tissues as diverse as the grasshopper leg to the mouse liver (Yamada et al., 1997; Bando et al., 2013; Cressman et al., 1996; Li et al., 2002; Wuestefeld et al., 2003; Zhao et al., 2014), dissecting the function of this highly conserved pathway is essential for our understanding of regenerative processes. To achieve this, we employed an imaginal disc model of tissue stress, which utilizes ectopic expression of the TNFα homologue Eiger (Smith-Bolton et al., 2009) (Fig. S1), triggering JNK activation (Igaki and Miura, 2014; Andersen et al., 2015) and apoptosis (Fig.S1D-D‴). This model has been previously used to study tissue regeneration in response to local cell ablation (Herrera et al., 2013; Smith-Bolton et al., 2009). Expression of Eiger induces multiple hallmarks of local and systemic stress responses, such as compensatory proliferation and dILP8-induced developmental delay (Fig. S1E,G) (Smith-Bolton et al., 2009). Because Eiger induces tissue stress in wild-type discs, it represents an ideal model to distinguish JAK/STAT functions in physiological stress from secondary effects incurred by developmental patterning or tumorigenic models. Instead of promoting compensatory proliferation, we found that JAK/STAT promotes survival of JNK-signalling cells, thereby acting as a central regulator of injury-induced apoptosis to restrain excessive tissue damage and facilitate initiation of compensatory responses.

We expressed Eiger under the control of rotund (rn)-GAL4, which drives expression in the wing disc pouch, fated to give rise to the future wing blade. We induced eiger expression on developmental day 7 and temporally limited UAS-eiger expression to 40 h by using a temperature-sensitive GAL80 repressor (Fig. S1A-C). In agreement with previous studies (Herrera et al., 2013; Smith-Bolton et al., 2009), we observed extensive cell ablation in discs (Fig. S1E-E′), which gave rise to a 50% reduction in adult wing size (Fig. S1H). G-trace lineage labelling (Evans et al., 2009) of rn-GAL4-expressing cells followed by FACS analysis revealed that about 85% of rn-GAL4 lineage cells were eliminated after 40 h of Eiger stimulation (Fig. 1A,A′, Fig. S2A-A″). This correlated with an induction of a 2 day developmental delay at the larval-pupal transition (Fig. S1G).

In agreement with Eiger triggering JNK activation (Igaki and Miura, 2014), we found that Eiger-stimulated discs broadly activate the JNK-responsive TRE reporter (Fig. 1C) (Chatterjee and Bohmann, 2012). The TRE reporter was specifically activated at the wound site, whereas reporter activity is completely absent in control discs (Fig. 1B,C). Consistent with non-autonomous activation of JNK by Eiger (Pérez-Garijo et al., 2013) and tissue damage (Bosch et al., 2005; Herrera et al., 2013; Wu et al., 2010), TRE activation extended beyond a G-trace-labelled domain, in which eiger expression was induced (Fig. 1C and Fig. 2F). FACS analysis demonstrated that while 4.5% of cells in the disc belonged to the surviving rn-GAL4 lineage, 29% of cells in the disc activated TRE-GFP (Fig. S2A-A′,B-B′).

We first wanted to quantify cell proliferation in Eiger-stimulated discs near the wound site, as well as in the whole disc between 0 h (R0) and 48 h (R48) into the recovery period after the end of cell ablation (Fig. S1A,E-E″). We used an image segmentation algorithm (Fiji) to specifically measure total disc size, the size of the surviving rn-GAL4 lineage labelled by G-trace, the number of mitotic cells marked by phospho-Histone 3 (pH3) or the number of cells in S-phase marked by BrdU incorporation (Fig. 1D-G, Fig. S1F). We found that at R0, G-trace-labelled cells in the pouch account for about 3% of the total disc volume (Fig. 1D). This is in close agreement with FACS measurements (4.5%, Fig. S2A′), which also detect an additional subset of G-trace-labelled cells in the notum. This analysis verified that volume quantifications of cell populations in Fiji accurately approximate cell counts. Our analysis revealed a marked increase in cell proliferation between R0 and R24, both near the wound site (Fig. 1F), as well as in the entire disc (Fig. 1G, Fig. S2E). Mitotic rates do not increase further between R24 and R48 (Fig. 1F,G). When we quantified the volume of the total disc and that of the G-trace-labelled population, we found that they increased in size by 36% and 126%, respectively (Fig. 1D,E).

While cell ablation can be induced more efficiently by expression of pro-apoptotic transgenes, such as UAS-reaper (Herrera et al., 2013), we did not observe broad activation of JNK in this context (Fig. S2C,D). Because Eiger-stimulated discs displayed broad activation and provided genetic access to JNK-signalling cells, we utilized expression of Eiger to investigate the role of JAK/STAT in JNK-mediated responses to tissue stress.

To understand if JAK/STAT plays a role in imaginal disc regeneration, we analysed JAK/STAT activity in response to eiger expression. JAK/STAT activation can be visualized using the 10xSTAT-dGFP reporter (Bach et al., 2007), which is almost undetectable in wild-type discs (Fig. 2A). Analysis of the 10xSTAT-dGFP reporter revealed that JAK/STAT is upregulated in response to eiger expression induced at day 5 or day 7 of development (Fig. 2B,C). Reporter activation can also be observed upon reaper expression, suggesting that stress induced by ectopic cell death is sufficient to cause JAK/STAT activation (Fig. S3E). Importantly, in Eiger-stimulated discs, JAK/STAT activation was detected beyond the pouch periphery and thus extended beyond JNK domains marked by TRE activity (Fig. 2F). JAK/STAT signalling declined within the following 24 h suggesting that it is downregulated after eiger expression ceased (compare Fig. S3A-C with Fig. S3B′-C′). We furthermore examined JAK/STAT activation when eiger was induced at later stages of development. Surprisingly, JAK/STAT activity was almost undetectable when eiger was induced at day 8 (Fig. 2D, Fig. S3D-D′).

Previous reports and our studies demonstrate that the JAK/STAT ligands upd1, upd2 and upd3 are transcriptionally upregulated in response to stress from physical injury (Katsuyama et al., 2015; Pastor-Pareja et al., 2008) or tumorous growth (Wu et al., 2010; Bunker et al., 2015), indicating that pathway upregulation may be driven by JNK-dependent Upd gene transcription. We found that transcription of Upd genes is highly elevated in Eiger-stimulated discs, whereas transcription of dome, hop and Stat92E is not (Fig. 2E). Recapitulating the decline in 10xSTAT-dGFP activity in response to eiger induction at day 8, Upd gene transcription was strongest at day 5 and day 7 but decreased by more than half when eiger was induced at day 8 (Fig. 2E).

Loss of JAK/STAT activation at day 8 was not due to a decline in Eiger-mediated JNK activation, because TRE activity was as strong on day 8 as on day 7 (Fig. S3F,G). This suggests that even though JAK/STAT activation coincides with JNK activity (Katsuyama et al., 2015; Pastor-Pareja et al., 2008), JNK alone may not be sufficient to stimulate this pathway. This is supported by broader activation of JAK/STAT compared with JNK (Fig. 2F). Strikingly, the failure to activate JAK/STAT at late developmental stages correlated with a pronounced decline in the ability of larvae to induce a developmental delay at the larval-pupal transition (Fig. S3H) and with a strong reduction in recovered adult wing size (Fig. S3I). These correlations suggest that JAK/STAT may be a crucial mediator of JNK-induced compensatory responses in imaginal discs.

JAK/STAT has been reported to play a role in promoting cell proliferation in wild-type (Bach et al., 2003; Tsai and Sun, 2004; Mukherjee et al., 2005), tumorous (Classen et al., 2009; Wu et al., 2010; Davie et al., 2015; Bunker et al., 2015; Amoyel et al., 2014) and surgically injured discs (Katsuyama et al., 2015). To test whether JAK/STAT is generally required to drive compensatory proliferation, we quantified mitotic events after Eiger stimulation in discs with impaired JAK/STAT signalling. We genetically reduced JAK/STAT activity by two approaches: (1) we reduced gene dosage of JAK/STAT components in the entire animal by heterozygosity for dome^G0441, hop³⁴ and Stat92E^85C9 alleles or (2) we interfered with JAK/STAT signalling exclusively in eiger-expressing cells through expression of RNAi constructs targeting JAK/STAT components (Fig. S5A), a dominant-negative dome (dome^Δcyt) or the inhibitor Socs36E under the control of rn-GAL4-tubGAL80^ts.

Strikingly, even though 50% of the tissue in Eiger-stimulated discs activated JAK/STAT (Fig. 2B,C), discs heterozygous for dome^G0441, hop³⁴ and Stat92E^85C9 alleles did not show any evidence of a reduction in mitotic events or total disc size (Fig. 3A-F). Instead, we observed a mild increase in Eiger-stimulated discs heterozygous for hop³⁴. These results are the opposite to what we would expect if JAK/STAT regulated compensatory proliferation. We thus wanted to confirm these findings by interfering with JAK/STAT specifically in Eiger-stimulated cells. rn-GAL4-driven co-expression of dome-RNAi, dome^Δcyt or Socs36E did not reduce the number of mitotic events within surviving eiger-expressing cells (Fig. 3G-K). Instead, mitotic rates slightly increased, whereas they remained unchanged in the rest of the disc (Fig. 3K, Fig. S4A,C). Importantly, effects on eiger-expressing cells co-expressing transgenic constructs are strongest at R0 but decline as GAL4 activity decreased (Fig. 3K). Together, our results strongly argue against the previously assigned role of JAK/STAT in directly promoting compensatory proliferation in response to stress.

While the total disc size of Eiger-stimulated discs was similar to that of discs with genetically impaired JAK/STAT activity (Fig. 3F, Fig. S4D), we were surprised to notice that the number of G-trace-labelled cells that survived eiger expression was dramatically reduced when JAK/STAT signalling was inhibited. Specifically, we found that expression of dome-RNAi, dome^Δcyt or SOCS36E caused a 3- to 10-fold reduction in the volume of G-trace-labelled populations (Fig. 3L, Fig. S4B).

These observations suggested that more eiger-expressing cells die when JAK/STAT signalling is impaired. Indeed, we observed a 2-fold increase in the volume positive for activated Caspase-3 if JAK/STAT activity was exclusively reduced in eiger-expressing cells (Fig. 4A-E) and up to a 10-fold increase in the volume of Eiger-stimulated discs heterozygous for dome^G0441, hop³⁴ or Stat92E^85C9 (Fig. 4F-I). The more pronounced effect observed for heterozygous tissues likely arises as a result of the tissue-wide reduction of JAK/STAT activity in this background. Reduction was also achieved in domains that display JAK/STAT and JNK activation but are located outside of the rnGAL4 lineage (Fig. 1C, Fig. 2B,C,F). Importantly, genetic reduction of JAK/STAT in developing wild-type discs under the same conditions does not cause any elevation of apoptosis (Fig. S5B-E). Combined, these data highlight a specific role for JAK/STAT as an important mediator of cell survival, specifically in JNK-signalling cells.

To understand how JAK/STAT promotes cell survival, we tested if genetically reducing JAK/STAT activity causes further elevation of JNK, which could divert JNK-dependent compensatory responses towards apoptosis. To this end, we monitored JNK activity using TRE reporters in eiger-expressing discs heterozygous for dome^G0441, hop³⁴ and Stat92E^85C9 alleles. Our results suggested that after cell ablation, 14% of the wing pouch area activated the TRE reporter (Fig. 4J,N). In Eiger-stimulated discs heterozygous for dome^G0441, hop³⁴ and Stat92E^85C9, we observed a 2- to 3-fold increase in the area positive for TRE activation (Fig. 4K-N). These results indicate that reduction of JAK/STAT signalling promotes non-autonomous expansion of JNK signalling beyond eiger-expressing cells and that this may underlie the observed increase in the apoptotic index in JAK/STAT-impaired Eiger-stimulated discs (Fig. 4A-I).

Our observations prompted two important predictions. First, broad activation of JNK in JAK/STAT-impaired discs suggested that JAK/STAT acts as a suppressor of JNK signalling. Repression of JNK by JAK/STAT could either be mediated by direct transcriptional effects on JNK core components, or indirectly, by suppression of apoptosis and prevention of Dronc-driven positive feedback activation of JNK. This mechanism would restrain non-autonomous activation of JNK, excessive apoptosis and tissue damage. A second prediction implies that interfering with JAK/STAT increases the extent of tissue damage due to elevation of cell death. Thereby, the ability of discs to mount an appropriate regenerative response is reduced as cells required to drive regenerative responses are eliminated by excessive cell death.

We did not find any evidence that negative JNK regulators are transcriptionally activated by Stat92E. We therefore wanted to understand if JAK/STAT regulates cell survival by impinging on pro-apoptotic JNK target genes rpr and hid, which act in combination with grim as inhibitors of IAP (inhibitor of apoptosis) proteins. To this end, we first performed a bioinformatic Clover analysis of the promoter regions of rpr, hid and grim using highly stringent parameter selections (Frith et al., 2004). As previously suggested (Shlevkov and Morata, 2012; Luo et al., 2007; Moreno et al., 2002), we identified multiple AP-1 binding motifs associated with these loci (Fig. S7A, data not shown). In agreement with studies on JNK-induced apoptosis (Shlevkov and Morata, 2012; Luo et al., 2007; Moreno et al., 2002), we specifically observed induction of hid expression in Eiger-stimulated discs (Fig. 5A). The mild increase in hid levels is likely to be an underestimate because eiger-expressing cells make up only 4.5% of the entire disc used for qRT-PCR analysis (Fig. S2A,A′). Eiger-expressing discs heterozygous for Df(3L)H99, a deficiency removing rpr, hid and grim loci, displayed a pronounced ‘undead cell’ phenotype (Fig. S6A,B) (Perez-Garijo et al., 2009; Martin et al., 2009; Kondo et al., 2006), suggesting that upregulation of hid contributes to Eiger-mediated cell death. While one previous study reports that Eiger-mediated induction of apoptosis is independent of hid activation (Igaki et al., 2002), we suggest that the small adult eye phenotype observed upon eiger co-expression with the strong apoptosis inhibitor p35 is a consequence of epithelial tissue architecture disruption rather than a failure to prevent Eiger-induced cell death (Fig. S6L). Together, previous reports and our results support the notion that induction of cell death in eiger-expressing discs is specifically mediated by AP-1/JNK-dependent activation of hid.

Strikingly, our Clover analysis also revealed multiple, highly clustered mammalian ZEB1-binding motifs the hid promoter (Fig. S7A) as well as in the promoter of the kay gene coding for the AP-1 component dFos (Fig. S7B). ZEB1 binding motifs localized to highly conserved DNA sequences as shown by mVISTA analysis (Frazer et al., 2004; Bray et al., 2003) and to open chromatin regions, likely representing regulatory elements, as demonstrated by overlays with previously published ATAC-seq data sets (Fig. S7) (Davie et al., 2015). ZEB proteins are homologous to Drosophila Zfh1 and Zfh2 (Zinc-finger homeobox) proteins, which act as transcriptional repressors (Postigo et al., 1999; Postigo and Dean, 1999). Both Zfh1 and Zfh2 were previously identified to be downstream effectors of JAK/STAT (Leatherman and Dinardo, 2008; Ayala-Camargo et al., 2013). This is reflected by Zfh2 expression mirroring JAK/STAT activation patterns in developing wing discs (Fig. 5E) (Ayala-Camargo et al., 2013) and the ability of JAK/STAT to induce Zfh2 expression de novo (Fig. S6C,D). We hypothesized that Drosophila ZEB proteins could directly mediate repression of kay in Eiger-stimulated discs (Fig. 5A), thereby restraining JNK activation to promote cell survival. At the same time, ZEB proteins could compete with AP-1 for transcriptional repression of hid, thereby limiting AP-1/Hid-induced apoptosis to promote cell survival. Indeed, in eiger-expressing discs heterozygous for Stat92E^85C9, the upstream regulator of Zfh proteins, we observed strong upregulation of hid and kay expression compared with discs expressing eiger alone (Fig. 5B).

To first understand if Zfh1 or Zfh2 expression was altered in tissue damage, we performed immunofluorescence and qRT-PCR analysis on Eiger-stimulated discs. We found that levels of Zfh2, but not Zfh1, are elevated in response to eiger expression (Fig. 5C-F). Zfh2 upregulation occurred specifically in regions with high levels of JAK/STAT activation (compare Fig. 2B,C with Fig. 5F). In addition, we found that transcripts of zfh2 but not of zfh1 were elevated (Fig. 5G). The observed mild increase in zfh2 transcription represents a strong underestimate, as cells expressing zfh-2 de novo make up only a small portion of entire imaginal discs used for qRT-PCR analysis. In agreement with Zfh2 being a JAK/STAT effector, heterozygosity for STAT92E^85C9 reduces Zfh2 expression in Eiger-stimulated discs (Fig. S6E-G).

To test if Zfh2 in Eiger-stimulated discs is required to promote JAK/STAT-dependent survival by repressing hid and kay transcription, we performed genetic experiments to reduce or increase Zfh2 function. We found that expression of an RNAi construct targeting zfh2 (Fig. S6H) increased apoptosis in eiger-expressing discs almost 4-fold (Fig. 5H-J). In contrast, overexpression of UAS-zfh2 strongly promoted survival of eiger-expressing cells and resulted in a 10-fold increase in the size of surviving rn-GAL4-derived cell populations (Fig. 5K-M). While we found that Zfh1 levels were not upregulated in response to stimulation with Eiger, overexpression of zfh1 also promoted survival of eiger-expressing cells, even phenocopying overexpression of the strong apoptosis inhibitor p35 (Fig. S6I-L). The survival-promoting function of either Zfh1 or Zfh2 suggests that both proteins can induce potent survival signals in stressed tissues, similar to developmental contexts (Ohayon et al., 2009; Guarner et al., 2014).

We predicted that interfering with JAK/STAT signalling and therefore with Zfh2 function, increases the extent of tissue damage incurred by Eiger due to elevation of cell death. Consistent with these predictions, we found that Eiger-stimulated discs with genetically reduced JAK/STAT activity developed into significantly smaller adult wings (Fig. 6A,B). Expression of dome^Δcyt, Socs36E or RNAi constructs targeting multiple pathway components, including zfh2, as well as heterozygosity for dome^G0441, hop³⁴ and Stat92E^85C9, caused a significant drop in adult wing size index by 30-90% (Fig. 6B). Importantly, genetic downregulation of JAK/STAT in wild-type discs at day 7 does not cause a comparable reduction in wing size (Fig. S8A), emphasizing that the survival-promoting function of JAK/STAT is specifically required during tissue stress responses.

Notably, rn-GAL4-driven overexpression of upd1, upd2 or zfh2 in eiger-expressing cells did not increase adult wing size (Fig. 6B). Extra Upd may not translate into JAK/STAT hyperactivation, because of pathway saturation in eiger-expressing cells. While zfh2 overexpression promoted cell survival, it did not rescue other defects such as loss of epithelial polarity (Fig. 5L), which interferes with wing morphogenesis.

To test whether JAK/STAT was also required for stress responses induced by physical wounding, we analysed adult wings that developed from discs of surgically pinched larvae (Pastor-Pareja et al., 2008; Kashio et al., 2014). Reducing JAK/STAT function in the posterior compartment by expressing an RNAi construct targeting hop caused a pronounced reduction in adult wing sizes developing from discs, in which pinching had been targeted to the posterior compartment as visualized by co-expression of GFP (Fig. S8B). In contrast, no reduction in size of the posterior compartment was observed for adult wings derived from undamaged control discs raised under the same conditions (not shown).

These data suggest that cellular responses to genetically or surgically induced damage critically rely on JAK/STAT activation to facilitate restoration of normal tissue homeostasis. Combined, our results strongly imply that a reduction in final tissue size upon JAK/STAT inhibition reflects an excessive loss of tissue due to cell death rather than a failure of the tissue to undergo compensatory proliferation.

We wanted to investigate further if the observed reduction in adult wing sizes upon JAK/STAT inhibition is exclusively caused by a loss of tissue to cell death or if other regenerative processes may be disturbed. A process that contributes to successful tissue restoration is the induction of a dILP8-dependent developmental delay at the larval-pupal transition, which extends the time available for repair before metamorphosis (Colombani et al., 2012; Garelli et al., 2012). We found that interfering with JAK/STAT in eiger-expressing cells also caused a profound reduction in developmental delays induced by Eiger stimulation (Fig. 6C).

To understand if altered Ilp8 expression caused this observation, we quantified expression of a GFP-reporter driven from the endogenous Ilp8 locus (Garelli et al., 2012). We found that the reporter was strongly expressed in the pouch of Eiger-stimulated discs (Fig. S8C). In contrast, interfering with JAK/STAT by expression of dome-RNAi or dome^Δcyt in eiger-expressing cells significantly reduced the area of GFP expression (Fig. 6D, Fig. S8D,E). This suggests that loss of Ilp8 expression, caused by impaired JAK/STAT in JNK-signalling cells, underlies the observed failure to efficiently induce a developmental delay.

To test whether JAK/STAT signalling is sufficient to induce developmental delays, we expressed the JAK/STAT-ligands Upd1 or Upd2 in wild-type discs using MS1096-GAL4. However, no difference in pupariation timing between Upd-expressing and stage-matched wild-type larvae was observed (Fig. S8F). While a recent study links Ilp8-expression to JAK/STAT signalling (Katsuyama et al., 2015), our data implies that Ilp8 is not a direct target gene of STAT92E. Instead, we suggest that cells that normally express Ilp8 in response to JNK activation are more likely to die when JAK/STAT signalling is reduced, thereby preventing efficient expression of Ilp8 and induction of a developmental delay. The failure to induce this important systemic response reduces the time available for repair and probably contributes to the decrease in recovered adult wing sizes that we observed upon genetic reduction of JAK/STAT signalling (Fig. 6B).

To understand if JAK/STAT generally acts as a survival-promoting pathway in the context of tissue stress, we revisited the role of JAK/STAT in established fly tumour models. Previous studies suggest that activation of JAK/STAT drives tumorous overgrowth in discs mutant for tumour suppressor genes, such as scribbled (scrib) or Psc-Su(z)2 (Wu et al., 2010; Classen et al., 2009). scrib cells, similar to eiger-expressing cells, exhibit strong JNK activation, correlating with elevated transcription of Upd cytokines (Wu et al., 2010; Bunker et al., 2015; Leong et al., 2009; Brumby and Richardson, 2003). While scrib cells have a growth disadvantage if surrounded by wild-type cells, they efficiently cooperate with oncogenic Ras^V12 to create invasive tumours in clonal assays (Brumby and Richardson, 2003; Wu et al., 2010). Larvae carrying MARCM-induced Ras^V12; scrib² clones fail to pupariate (Fig. 7M), suggesting that Ilp8 activation correlates with tumour load (Garelli et al., 2012). We found that Ras^V12; scrib² clones covered about 47% of eye antennal discs, compared with 19% for wild-type clones (Fig. 7A,B,D). When we probed Ras^V12; scrib² clones for activated Dcp-1, we did not observe any difference in apoptotic patterns compared with wild-type tissue (Fig. 7F,G).

When we completely removed JAK/STAT function in Ras^V12; scrib² clones by homozygosity for a Stat92E^85C9 allele, Ras^V12; scrib² clone size was reduced from 47% to 28% of the eye disc area (Fig. 7B-D). Comparison of cell division rates within Ras^V12; scrib² and Ras^V12; scrib², Stat92E^85C9 clones did not reveal any significant changes upon loss of JAK/STAT function (Fig. 7B′,C′,E). However, in contrast to Ras^V12; scrib² clones, Ras^V12; scrib², Stat92E^85C9 clones displayed a 4.3-fold increase in areas of Dcp-1 activation (Fig. 7G,H,I), suggesting that increased cell death of Stat92E mutant cells underlies the smaller sizes of Ras^V12; scrib², Stat92E^85C9 clones (Fig. 7A-D). The reduction in Ras^V12; scrib², Stat92E^85C9 clone size reduced total tumour load and allowed a significant proportion of host larvae to progress to pupal stages (Fig. 7M).

Strikingly, Ras^V12; scrib² clones displayed ectopic activation of Zfh2 (Fig. 7J,K) but not of Zfh1 (not shown), indicating activation of a JNK-JAK/STAT-Zfh2 stress module by neoplastic transformation. Importantly, ectopic expression of Zfh2 was completely abolished within Ras^V12; scrib², Stat92E^85C9 clones (Fig. 7L). These results support the notion that Zfh2 expression is regulated by JAK/STAT in multiple contexts of cellular stress and that stress-dependent Zfh2 activation in imaginal discs directly correlates with cell survival.

Altogether, our results support a model in which cellular stress caused by genetic cell ablation, physical wounding or tumorous growth drives activation of JAK/STAT to promote survival of JNK-signalling cells. Activation of JAK/STAT signalling thereby facilitates the induction and execution of local and systemic compensatory responses, rather than promoting compensatory cell proliferation directly (Fig. 7N).

JNK activation during tissue stress drives elimination of damaged cells by apoptosis (Bogoyevitch et al., 2010; Chen, 2012; Shlevkov and Morata, 2012; Luo et al., 2007; Moreno et al., 2002) and compensatory proliferation necessary to replace lost tissues (Bosch et al., 2005; Bergantinos et al., 2010; Mattila et al., 2005; Ryoo et al., 2004; Sun and Irvine, 2011). Our work provides an answer to the central question of how these Janus-faced responses can be balanced by JNK. We show that in contrast to prevalent models, JAK/STAT signalling downstream of JNK does not promote compensatory proliferation, but instead supports cell survival and thereby contributes to tissue growth (Fig. 7N). The decision between JAK/STAT-dependent survival and JNK-induced apoptosis may depend on a relative cellular ratio between JAK/STAT and JNK signalling, which is determined by the position of cells within respective signalling gradients (Fig. S9B,B′). As a consequence of its pro-survival function, JAK/STAT indirectly supports regenerative growth by allowing JNK to initiate dILP8-dependent developmental delays and, possibly, by facilitating Hippo/Yorkie-driven proliferation (Sun and Irvine, 2014). In fact, JNK-dependent Yorkie activation (Sun and Irvine, 2011) may occur independent of Jun/Fos-mediated transcription (Sun and Irvine, 2013), therefore JAK/STAT and Zfh2-dependent repression of kay/fos could suppress induction of JNK-dependent apoptosis without suppressing activation of Yorkie and thus of Hippo/Yorkie-driven compensatory proliferation. Hippo/Yorkie signalling, in turn, may sustain activation of unpaired and JAK/STAT signalling (Staley and Irvine, 2010; Bunker et al., 2015; Sarikaya and Extavour, 2015) even if JNK signalling is low. This signalling crosstalk would facilitate the stabilisation of a wound-proximal tissue domain supporting compensatory survival and proliferation.

Our new interpretation of JAK/STAT in promoting cell survival in response to tissue damage and in tumours can be reconciled with previous studies: JAK/STAT mutations frequently reduced tissue size. This was interpreted as a reduction in growth (Classen et al., 2009; Wu et al., 2010; Mukherjee et al., 2005; Amoyel et al., 2014), but we suggest that it is a consequence of excessive cell death. Similarly, Upd overexpression has been previously reported to drive cell proliferation (Bach et al., 2003; Classen et al., 2009; Tsai and Sun, 2004). However, continuous overexpression of Upd can also induce apoptosis (not shown) and, thus potentially also sustained JNK-dependent compensatory proliferation, driving tissue growth. So far, only isolated studies have implicated JAK/STAT in cell survival (Betz et al., 2008; Hasan et al., 2015; Guarner et al., 2014; Ohayon et al., 2009). For example, the apoptosis inhibitor IAP has been suggested to be a positively regulated target of Stat92E, protecting cells from apoptosis (Betz et al., 2008; Hasan et al., 2015). We identify the JAK/STAT effector Zfh2 as a potential repressor of kay and hid activity – a molecular pathway expected to restrain excessive JNK-activity and induction of apoptosis (Fig. S9A). Similarly, Dpp/TGFβ-dependent repression of rpr by Schnurri has been reported to prevent JNK-mediated apoptosis. Curiously, this occurs in contexts where JNK function is required to mediate cell shape remodelling during development (Beira et al., 2014) and therefore bears conceptual similarities to a model where JAK/STAT-dependent repression of kay and hid prevents JNK-mediated apoptosis but not Hippo/Yorkie-dependent compensatory proliferation. Our work does not address whether JAK/STAT only promotes survival during stress or if this also occurs during development. Cells carrying JAK/STAT mutations are eliminated from developing wing discs (Rodrigues et al., 2012) by cell competition. However, in the light of recent studies, which implicate JNK signalling in competitive cell elimination (Kolahgar et al., 2015), more studies are clearly needed to dissect the functional contribution of stress signals in this context. Combined, these studies and our work suggest that JNK-dependent apoptosis can be counteracted by multiple molecular pathways impinging on anti- and pro-apoptotic genes.

While the role of mammalian IL-6/STAT3 in regeneration needs to be further investigated, much evidence points to JNK and JAK/STAT pathways as crucial mediators of compensatory responses and tumorigenesis in mammalian tissues (Chen, 2012). A previous report suggests that JAK/STAT activation during mouse liver regeneration potentially confers a cell-protective function, similarly facilitating initiation of compensatory responses, rather then directly promoting cell proliferation (Wuestefeld et al., 2003). Therefore, the dominant pro-survival function of JAK/STAT in response to tissue stress, which we find to be essential for successful restoration of tissue homeostasis, appears highly relevant to human contexts of cellular stress in physiological or pathological conditions.

For a detailed list of genotypes used in all experiments and shown in figures, refer to Table S1.

Fly crosses were carried out as described in supplementary Materials and Methods and eiger expression was induced as described in Smith-Bolton et al. (2009) and Fig. S1.

Larval cuticles were fixed in 4% PFA/PBS for 15 min at room temperature. Washing steps were performed in 0.1% Triton X-100/PBS (PBT), blocking in 5% NGS/PBT. Tissue samples were incubated in the following primary antibodies overnight at 4°C: mouse anti-Nub (1:100, S. Cohen, University of Copenhagen), rabbit anti-GFP (1:1000, Immunokontakt, cat. no. 210-PS-1GFP), mouse anti-H3S10p (1:2000, Abcam, cat. no. ab14955), rabbit anti-cCasp-3 (1:500, Cell Signaling, cat. no. 9661), rabbit anti-Dcp1 (1:500, Cell Signaling, cat. no. 9578), rabbit anti-β-Gal (1:500, Cappel, cat. no. 559762 ), rat anti-Zfh1 (1:500, R. Lehmann, NYU School of Medicine), rat anti-Zfh2 (1:300, C. Doe, HHMI/Institute of Molecular Biology). Secondary antibodies (Molecular Probes), DAPI and phalloidin-TRITC (Sigma) were applied and samples were incubated for 2 h at room temperature.

Flow cytometry of wing imaginal disc cells and labelling of fixed cuticle tissue with BrdU was carried out using standard techniques as described in supplementary Materials and Methods.

RNA was extracted from ∼80 wing discs using Qiagen RNAlater and RNeasy protocols. cDNA libraries were prepared using standard protocols, including Ambion TurboDNase and Invitrogen Superscript III kits. qPCR was performed using Fast SYBR Green (Applied Biosystems) on a CFX-96 Real-Time machine (Bio-Rad). Data were analysed using the ΔΔC_t method and normalized to at least two housekeeping genes (Table S2).

Bioinformatic analysis for potential transcription factor binding sites was performed using the program Clover (cis-element over-representation) (Frith et al., 2004) in combination with position-weighted matrices (PWM) obtained from the JASPAR collection (Mathelier et al., 2014) as detailed in supplementary Materials and Methods.

Wing size index (Ws) was calculated as the mean of five different wing phenotypes (w) (Fig. 5A) weighted with the frequencies at which they occurred (f) (Ws=Σw_i×f_i/Σf_i). Samples were compared in a paired manner to control wings from the same experimental replicate (Wilcoxon signed rank test, α=0.05, n≥3 sample populations).

We thank R. Smith-Bolton, D. Bohmann, E. Bach, G. Halder, S. Cohen, R. Lehmann, C. Doe, E. Sanchez-Herrero, F. Diaz-Benjumea, N. Gompel and H. Leonhardt for sharing reagents and resources. We thank Bloomington and VDRC for fly stocks, and DSHB for antibodies. We thank the LSM, IMPRS-LS and IRTG-SFB1064 graduate schools for supporting our students.