Actin remodeling mediates ROS production and JNK activation to drive apoptosis-induced proliferation

Stress-induced cell death, mainly apoptosis, and its subsequent tissue repair is interlinked although our knowledge of this connection is still very limited. An intriguing finding is apoptosis-induced proliferation (AiP), an evolutionary conserved mechanism employed by apoptotic cells to trigger compensatory proliferation of their neighboring cells. Studies using Drosophila as a model organism have revealed that apoptotic caspases and c-Jun N-terminal kinase (JNK) signaling play critical roles to activate AiP. For example, the initiator caspase Dronc, the caspase-9 ortholog in Drosophila, promotes activation of JNK leading to release of mitogenic signals and AiP. Recent studies further revealed that Dronc relocates to the cell cortex via Myo1D, an unconventional myosin, and stimulates production of reactive oxygen species (ROS) to trigger AiP. During this process, ROS can attract hemocytes, the Drosophila macrophages, which further amplify JNK signaling cell non-autonomously. However, the intrinsic components connecting Dronc, ROS and JNK within the stressed signal-producing cells remain elusive. Here, we identified LIM domain kinase 1 (LIMK1), a kinase promoting cellular F-actin polymerization, as a novel regulator of AiP. F-actin accumulates in a Dronc-dependent manner in response to apoptotic stress. Suppression of F-actin polymerization in stressed cells by knocking down LIMK1 or expressing Cofilin, an inhibitor of F-actin elongation, blocks ROS production and JNK activation, hence AiP. Furthermore, Dronc and LIMK1 genetically interact. Co-expression of Dronc and LIMK1 drives F-actin accumulation, ROS production and JNK activation. Interestingly, these synergistic effects between Dronc and LIMK1 depend on Myo1D. Therefore, F-actin remodeling plays an important role mediating caspase-driven ROS production and JNK activation in the process of AiP.


Introduction
In multicellular organisms, damaged cells are frequently removed by apoptosis, the major form of programmed cell death. Intriguingly, these stress-induced dying cells, prior to their removal, can actively induce proliferation of neighboring cells to compensate for their loss and maintain tissue integrity [1][2][3][4][5][6][7][8]. This phenomenon has been termed as apoptosis-induced compensatory proliferation or apoptosis-induced proliferation (AiP) [9,10]. Importantly, AiP is found not only to promote tissue recovery and regeneration, but also to drive tumor development and cancer recurrence [4,[11][12][13]. Therefore, understanding the mechanisms of AiP has significant clinical and therapeutic implications.
Caspases, an evolutionary conserved family of cysteine proteases, are essential for both apoptosis and AiP. Whilst caspases mediate execution of apoptosis, they can also promote AiP via activation of mitogenic signals such as cytokines or growth signaling pathways in a contextdependent manner [1,4,5,[14][15][16]. Studies in Drosophila have revealed distinct mechanisms of AiP mediated by different groups of caspases. In differentiating tissues where cells have exited mitosis, the caspase-3-like effector caspases DrICE and Dcp-1 activate Hedgehog signaling to drive cell cycle re-entry [14]. In contrast, in proliferating tissues where cells are actively dividing, the caspase-9-like initiator caspase Dronc activates the c-Jun N-terminal kinase (JNK), an evolutionary conserved stress response molecule, which in turn induces growth signals such as Wg, Dpp and EGFR signaling resulting in AiP [1,15,17].
Notably, in the proliferating larval eye and wing epithelia, Reactive Oxygen Species (ROS), the oxygen-containing free radicals produced during cellular metabolism, accumulate in apoptotic tissues and trigger activation of JNK [18,19]. Recent studies further showed that ROS and JNK in apoptotic cells also damage the epithelial basement membrane and act as signals to recruit macrophage-like hemocytes, which in turn contribute to further activation of JNK signaling cell non-autonomously [20][21][22]. Interestingly, the initiator caspase Dronc translocates from cytosol to the plasma membrane, where it exerts its non-apoptotic function to activate Duox, a NADPH oxidase, for ROS production [20]. Myo1D, an unconventional myosin, is critical for this process through its interaction with Dronc. However, it remains unknown what mediates this non-apoptotic action of Dronc and Myo1D to drive ROS production and JNK activation within the dying cells.
To study the mechanisms of AiP in proliferating tissues, we have previously developed two genetic assays, an overgrowth assay and a regeneration assay, to identify regulators of AiP [15]. In the overgrowth assay, hid, a pro-apoptotic gene, and p35, an inhibitor of effector caspases DrICE and Dcp-1, are simultaneously expressed under control of the ey-GAL4 driver (ey>hid-p35, Fig 1A) in the developing larval eye epithelium which is composed of the anterior proliferating and the posterior differentiating tissues. While the posterior differentiating tissue develops to the adult eye, the anterior proliferating tissue develops to adult head appendages including cuticle and sensory organs such as ocelli and bristles. Expression of P35 inhibits the effector caspases DrICE and Dcp-1, therefore AiP in the differentiating tissue, but not the initiator caspase Dronc [14,23]. Hence, Dronc-dependent AiP occurs specifically in the proliferating eye tissues in ey>hid-p35. Cells in this tissue are kept 'undead' because apoptotic responses are activated but execution of cell death is blocked. These 'undead' cells continuously drive AiP leading to a head overgrowth phenotype, a convenient readout of AiP for genetic screens (Fig 1A). In the regeneration assay, hid is expressed in a temporally and spatially controlled manner to induce a pulse of apoptosis therefore tissue ablation. This tissue is then allowed to recover to observe AiP and study its regulation during tissue regeneration (Fig 2A).
The actin cytoskeleton, a structural network found in all eukaryotic cells, consists of actin and actin-binding proteins [24,25]. Actin exists in two forms: globular monomers (G-actin) and filamentous polymers (F-actin). Actin filaments are highly dynamic and constantly undergo assembly and disassembly. The spatial and temporal regulation of F-actin mediates multiple cellular signaling responses including cell division, motility and phagocytosis. The evolutionary conserved LIM domain kinase 1 (LIMK1) and Cofilin are essential for the formation of actin filaments [26,27]. Cofilin, also called actin-depolymerizing factor (ADF), promotes rapid F-actin turnover by severing actin filaments [28]. LIMK1 phosphorylates Cofilin and inhibits its activity, therefore promotes F-actin polymerization. Although AiP relies on cell-cell interactions and signal transduction, it is not yet clear whether and how the F-actin network is involved in this process.
Using both overgrowth and regeneration assays, in this study, we showed that knockdown of LIMK1 or overexpression of Cofilin suppresses both AiP-dependent tissue overgrowth and regeneration. Consistent with this, F-actin polymerization, regulated by LIMK1 and Cofilin, was observed in AiP. It depends on the initiator caspase Dronc and is required for activation of JNK and its upstream ROS production. Furthermore, Dronc and LIMK1 genetically interact and drive F-actin remodeling in AiP via Myo1D.

LIMK1, a key factor promoting F-actin polymerization, is required for AiP
We have recently shown that Myo1D, an unconventional actin-based motor protein, is critical for AiP [20]. This suggests a potential role of actin dynamics and actin-based motility in AiP. To address this, we examined roles of LIMK1 and Cofilin, two key factors regulating F-actin polymerization, using the ey>hid-p35 assay (Fig 1A). AiP-dependent overgrowth phenotypes observed in ey>hid-p35 animals vary and can be categorized into three groups: severe, moderate and weak (Fig 1B-1E, quantified in Fig 1J) [15]. Using this approach, we identified three independent LIMK1 RNAi lines as suppressors of AiP (Fig 1F-1I, quantified in Fig 1J). Compared to the control ey>hid-p35 flies, in which 91% show either severe or moderate overgrowth phenotypes, the LIMK1 RNAi line 57012 suppresses it to 38%. The other two LIMK1 RNAi lines, 42576 and 26294, suppress it to a lesser degree, with the percentage of severe or moderate overgrowth flies reduced to 51% and 58%, respectively. qPCR measurements suggest that this difference correlates with the strength of the RNAi (Fig 1K). The line 57012 was used in the follow-up experiments because it is the strongest LIMK1 RNAi line. To further confirm that loss-of-LIMK1 suppresses AiP, LIMK1 2 , a hypomorphic but viable mutant of LIMK1 [29], was used. The ey>hid-p35-induced overgrowth phenotype is strongly suppressed in the hemizygous LIMK1 2 mutants with only 26% of flies showing the overgrowth phenotype ( Fig 1J). As LIMK1 phosphorylates and inhibits Cofilin, we examined the roles of Cofilin for regulation of AiP. Similar to loss of LIMK1, overexpression of Cofilin also strongly suppresses the ey>hid-p35-induced overgrowth phenotype ( Fig 1J).
To further confirm that the suppression of the ey>hid-p35 overgrowth phenotype by LIMK1 was due to a reduction of cell proliferation, we examined late 3 rd instar larval eye imaginal discs. Compared to the ey>p35 control, the number of mitotic cells indicated by the PH3 antibodies increases in ey>hid-p35 discs (Fig 1L and 1M, quantified in Fig 1P). This increase is suppressed by RNAi knockdown of LIMK1 or overexpression of Cofilin (Fig 1N, 1O and 1P). Notably, this suppression is not due to potential roles of LIMK1 and Cofilin in regulation of apoptosis as knockdown of LIMK1 or overexpression of Cofilin does not suppress hid-induced apoptosis (S1 Fig). Therefore, LIMK1 and Cofilin, two key regulators of F-actin dynamics, play critical and antagonistic roles in regulating AiP. Furthermore, we measured expression of LIMK1 and Cofilin using RT-qPCR. Interestingly, LIMK1, but not Cofilin, is transcriptionally increased in ey>hid-p35 discs compared to the control (Fig 1Q and 1R). Altogether, these data suggest that LIMK1 is induced and required for AiP-dependent tissue overgrowth.
Next, we examined whether LIMK1 is required for AiP-dependent regenerative growth by using the regeneration assay (Fig 2A). In this assay, to induce spatially and temporally controlled apoptosis and tissue ablation, the pro-apoptotic gene hid was expressed in the dorsal portion of the larval eye disc for 12 hours. This was under the control of the driver Dorsal Eye -Fig 1. LIMK1 and Cofilin, regulators of F-actin polymerization, are required for AiP. (A) A schematic representation of the AiP-dependent overgrowth assay. ey-GAL4 is expressed in the developing larval eye disc which is composed of the anterior proliferating tissue and the posterior differentiating tissue, separated by a morphogenic furrow (MF, the grey bar). Compared to the control ey>p35 (ey-GAL4 UAS-p35), simultaneously expression of hid and p35 in ey>hid-p35 (ey-GAL4 UAS-hid UAS-p35) results in AiP-dependent overgrowth of the anterior proliferating tissue, which leads to an adult head overgrowth phenotype characterized by expanded head cuticle with ectopic sensory organs such as ocelli and bristles. Consequently, sizes of the differentiating eye tissue and adult eye are reduced in ey>hid-p35 animals. These larval and adult phenotypes were used as readouts of AiP in our analyses. (B-I) Representative adult fly head images of the indicated genotypes. Compared to the control ey>p35 (B), which is similar to wildtype, ey>hid-p35 fly head capsules display overgrowth phenotypes which can be grouped into three categories (C-E): severe (S), moderate (M) and weak (W, including wildtype-like), as previously described [15]. A majority of ey>hid-p35 flies show either severe (57%) or moderate (34%) overgrowth phenotype, characterized by overgrown head capsules with duplications of sensory organs including bristles and ocelli (C and D, arrows). (F) Knockdown of LIMK1 by RNAi does not cause any defects. (G-I) LIMK1 RNAi-57012 strongly reduces the percentage of ey>hid-p35 flies displaying severe (4%) and moderate overgrowth (34%) phenotypes, with a large increase of flies (62%) showing a weak phenotype or wildtype-like appearance. (J) Summary of the suppression of the ey>hid-p35 overgrowth phenotype by a LIMK1 hypomorphic mutant (LIMK1 2 ), expressing three independent LIMK1 RNAi lines (26294, 42576 and 57012) or cofilin. Black indicates severe, dark grey indicates moderate, and light grey indicates weak or wildtypelike phenotypes. (K) Specificity and efficiency of LIMK1 RNAi lines were determined by measuring LIMK1 transcript levels with RT-qPCR. Compared to the control without expression of LIMK1 RNAi , two independent LIMK1 RNAi lines, 26294 and 57012, suppress LIMK1 transcript levels to less than 50% and 30%, respectively. These reductions are statistically significant ( ��� P<0.001). (L-O) Late 3 rd instar eye discs labelled with the mitotic marker PH3 (green) and the photoreceptor neuron marker ELAV (magenta), anterior is to the left. ELAV is used to mark the posterior differentiating portion of the eye discs. White dotted lines indicate the anterior proliferating portion of the eye discs. Compared to the control ey>p35 (L), the size of the anterior proliferating portion and the number of mitotic cells increase in the ey>hid-p35 eye discs (M). These increases are suppressed by expressing LIMK1 RNAi-57012 (N) or cofilin (O). (P) Quantification of the number of PH3 + cells in the anterior portion of the eye discs of the indicated genotypes. Compared to ey>p35, the number of PH3 + cells are significantly ( �� p < 0.01) increased in the ey>hid-p35 discs. This increase is significantly ( �� p < 0.01) reduced in response to expression of dronc RNAi , LIMK1 RNAi or cofilin. (Q and R) Expression levels of LIMK1 (Q) and cofilin (R) were measured by RT-qPCR. Compared to the control ey>p35-mCherry (ey-GAL4 UAS-p35 UAS-mCherry) eye discs, expression of LIMK1 is increased at about 1.6-fold in ey>hid-p35 ( ��� P<0.001). In contrast, expression of cofilin is not significantly changed.  driver, which is expressed in the dorsal half of the eye disc, together with tub-GAL80 ts , a ubiquitously expressed and temperature sensitive (ts) inhibitor of GAL4. tub-GAL80 ts inhibits DE-GAL4 (DE ts ) at 18˚C, but not at 30˚C. In this regeneration assay, a temperature shift to 30˚C for 12 hours during second instar larval stage induces expression of GFP together with or without hid. This is then followed by a recovery period at 18˚C. Compared to the control with GFP expression only, a 12-hour expression of hid results in tissue ablation in the dorsal eye half, which is regenerated completely after a recovery period of 72 hours (R72h). (B-C') Early 3 rd instar eye discs, anterior is to the left, were labelled with GFP (green in B, C) and an apoptosis marker cDcp1 (magenta in B, C and grey in B', C'). Dashed lines highlight the discs. Compared to the control (DE ts >GFP, B, B'), conditional expression of hid (DE ts >hid, C, C') for 12 hours under the control of DE ts results in a strong induction of apoptosis (arrows) and loss of bilateral symmetry of the disc which are visible after a recovery period of 24 hours (R24h). (D-F') Late 3 rd instar eye discs, anterior is to the left. ELAV (magenta in D, E, F and grey in D', E', F') labels photoreceptor neurons and is used to outline the shape of the discs. Conditional expression of LIMK1 RNAi (D, D'), hid (E, E') or hid and LIMK1 RNAi (F, F') was under control of DE ts and indicated by GFP (green in D, E, F). A temperature shift to 30˚C for 12 hours was followed by a recovery period of 72 hours at 18˚C (R72h). (D, D') Following this protocol, expression of LIMK1 RNAi alone (DE ts >LIMK1 RNAi ) does not affect the eye disc morphology indicated by the GAL4 (DE-GAL4) together with the tub-GAL80 ts , a ubiquitously expressed and temperature sensitive (ts) inhibitor of GAL4 [30]. tub-GAL80 ts inhibits activity of DE-GAL4 at 18˚C. A temperature shift to 30˚C releases this inhibition and allows expression of hid, i.e. induction of apoptosis, in the dorsal half of the eye disc under the control of DE-GAL4. After a 12-hour period of hid expression, larvae were transferred back to 18˚C for tissue recovery. Compared to the control where only GFP is expressed (Fig 2B and 2B'), a clear tissue ablation was observed at 24 hours of recovery (R24h) (Fig 2C and 2C'). This apoptosis-induced tissue ablation can fully regenerate via AiP after 72 hours of recovery ( Fig 2E) [15]. However, knockdown of LIMK1 or overexpression of Cofilin hinders such tissue regeneration (Fig 2F, quantified in Fig 2G). As a control, knockdown of LIMK1 or overexpression of Cofilin does not affect the development of larval discs (Fig 2D and 2G). Taken together, LIMK1 and Cofilin were identified as regulators of AiP in both overgrowth and regeneration assays.

F-actin accumulates in the process of AiP
LIMK1 is best known to inhibit Cofilin and promote F-actin polymerization. Suppression of AiP by loss of LIMK1 or gain of Cofilin suggests that formation of actin filaments might be important for AiP. We therefore examined the F-actin pattern in larval discs of various genetic backgrounds using phalloidin (PHN), a F-actin specific marker. Compared to the control ey>p35 ( Fig 3A and 3A'), in ey>hid-p35, bright F-actin signals with an increased number of actin aggregates were observed in the anterior overgrown part of the disc (arrows and arrowheads, Fig 3B and 3B', aggregate numbers are quantified in Fig 3F). It has been reported that, in epithelial tissues, apoptotic cells can be extruded basally via mechanisms depending on actin remodeling in neighboring cells [31]. We therefore labeled the ey>hid-p35 disc with the cleaved caspase-3 antibody, a marker of the activated initiator caspase Dronc in Drosophila [9], to indicate cells with high Dronc activities although execution of cell death is blocked by P35 ( S2A Fig). Notably, the majority of the aggregates we observed in ey>hid-p35 discs are not overlapping with or adjacent to the cells labelled with the cleaved caspase-3 antibodies. Moreover, these aggregates frequently localize at the apical cell cortex in the disc suggesting an accumulation of F-actin independent of cell extrusion (arrows and arrowheads, S2B Fig). Nevertheless, this F-actin accumulation including the formation of actin aggregates depends on Dronc, which coordinates apoptosis and AiP, as knockdown of Dronc suppresses it as well as the AiP-dependent tissue overgrowth (Fig 3C, 3C' and 3F). This suggests that F-actin accumulates downstream of Dronc. As expected, LIMK1 2 mutants or overexpression of Cofilin also suppresses this actin accumulation (Fig 3D-3F). To determine whether such effects are tissue-specific, we further examined larval wing imaginal discs by co-expression of hid and p35 under the control of nub-GAL4, a wing pouch-specific driver. LIMK1-dependent tissue overgrowth and F-actin accumulation were also observed in wing discs (S3 Fig). Therefore, LIMK1 and Cofilin regulate F-actin dynamics and AiP-dependent tissue overgrowth in both eye and wing epithelial tissues. normal ELAV pattern in the dorsal half of the eye disc. (E, E') Tissue damage induced by expression of hid (DE ts >hid) for 12 hours has fully recovered after 72 hours recovery (R72h) at 18˚C as indicated by the largely normal ELAV pattern in the late third instar eye discs. (F, F') A DE ts >hid eye disc that simultaneously expresses LIMK1 RNAi (DE ts >hid-LIMK1 RNAi ). The arrow in (F') highlights the incomplete ELAV pattern on the dorsal half of the disc indicating that the regenerative response was partially impaired by reduction of LIMK1. (D) Quantification of the dorsal/ventral size ratio in representative eye discs of the indicated genotypes. Compared to DE ts >hid, expression of LIMK1 RNAi or cofilin significantly ( � P < 0.5) reduces the size of the dorsal half of the eye disc. As the controls, disc sizes of DE ts >LIMK1 RNAi and DE ts >cofilin are not significantly (n.s.) different from those of DE ts >hid. https://doi.org/10.1371/journal.pgen.1010533.g002 Considering that the hid-p35-expressing discs are overgrown, the actin accumulation we observed in these tissues can potentially be a secondary effect caused by tissue overgrowth. Therefore, in addition to the whole disc analysis, we also employed a clonal analysis to determine whether F-actin accumulates in AiP independently of tissue overgrowth. We coexpressed hid and p35 in heat-shock-induced clones (hid-p35 clones) and examined F-actin in these clones 48 hours after their induction (Fig 4). These clones do not show any signs of tissue overgrowth. We focused our analysis on the larval wing disc pouch because it is a flat epithelial tissue most suitable for the observation of potential F-actin pattern changes. Compared to the control clones expressing p35 (Fig 4A-4B'), F-actin accumulates apically and cortically in the hid-p35 clones (arrows, Fig 4C-4D', and quantified in Fig 4K). As observed in whole discs, F-actin accumulation in hid-p35 clones is suppressed by loss of either dronc (via RNAi, Fig 4E, 4E' and 4K) or LIMK1 (via RNAi or mutants, Fig 4F-4G' and 4K). Notably, loss of dronc or LIMK1 alone does not affect the F-actin organization (S4A-S4C' Fig). This further confirms that LIMK1-dependent F-actin accumulation occurs downstream of Dronc. Overexpression of Cofilin promotes depolymerization of F-actin. Indeed, it reduces the F-actin levels in both control and hid-p35 clones although the clones appear to be smaller in size (Figs 4H, 4H', 4K and S4D and S4D'). Intriguingly, expression of a dominant negative mutant of JNK (bsk DN ) does not inhibit the F-actin accumulation in hid-p35 clones (Fig 4I, 4I' and 4K). Consistently, this F-actin accumulation also persists in a null mutant of Tak1 (Fig 4J, 4J' and 4K), an upstream kinase in the JNK pathway critical for AiP [15]. As controls, expression of bsk DN or the mutant of Tak1 does not affect the F-actin organization by themselves (S4E-S4F' Fig). Taken together, these data suggest that F-actin accumulates upstream or in parallel of activation of JNK during AiP.

F-actin polymerization is required upstream of JNK to induce AiP
To further determine the relationship between F-actin polymerization and JNK activation in AiP, we examined the expression level of MMP1, a downstream target of JNK using clonal analysis. Compared to the control clones ( Fig 5A and 5A'), expression of MMP1 is strongly induced in hid-p35 clones (Fig 5B and 5B', quantified in Fig 5G). As expected, such an increase of MMP1 is suppressed by knockdown of dronc, expression of bsk DN or a null mutant of Tak1 (Fig 5C, 5D' and 5G). Notably, the increased MMP1-labelling in hid-p35 clones is also strongly suppressed in the LIMK1 2 mutants or by expression of Cofilin (Fig 5E-5G). This was further confirmed in the whole eye discs using TRE-red, another reporter of JNK activity [32]. Expression of LIMK1 RNAi or Cofilin suppresses the activation of JNK in ey>hid-p35 eye discs (S5 Fig). Therefore, the F-actin polymerization, regulated by LIMK1 and Cofilin, is required for activation of JNK, the critical step triggering AiP.

F-actin polymerization is required for ROS production upstream of JNK
ROS production has recently been reported to mediate activation of JNK in AiP [19,21]. We therefore investigated whether F-actin polymerization is required for ROS production using clonal analysis. Compared to the control clones expressing P35 (Fig 6A and 6A'), ROS, indicated by Dihydroethidium (DHE) staining, increase in hid-p35 clones (arrows, Fig 6B and 6B', quantified in Fig 6G). Expression of SOD, a superoxide dismutase reducing cytosolic ROS, in hid-p35 clones suppresses the clonal increase of ROS (Fig 6C, 6C' and 6G) as well as activation of JNK (S6A and S6B Fig). Duox, a transmembrane NADPH oxidase, has been reported to produce extracellular ROS and attract hemocytes to promote AiP and tissue overgrowth of the whole eye discs [21]. However, RNAi knockdown of Duox does not significantly inhibit ROS production in hid-p35 clones (Fig 6D, 6D' and 6G). The SOD and Duox RNAi experiments suggest that, in hid-p35 clones, largely cytosolic ROS are produced. Consistently, unlike the whole eye discs where extracellular ROS recruit hemocytes for the AiP-dependent tissue overgrowth [21,22], hemocytes were rarely observed to be associated with wing discs containing hid-p35 clones (S6C-S6D' Fig). Importantly, the increase of ROS in hid-p35 clones is completely suppressed in the LIMK1 2 mutants or by expression of Cofilin (Fig 6E, 6E' and 6G). In contrast, these ROS are not affected by expression of bsk DN (Fig 6F-6G). This is consistent with the published data that ROS act upstream of JNK [18,21]. Altogether, our data suggest that LIMK1and Cofilin-dependent F-actin polymerization is required for ROS production upstream of JNK during AiP.

Dronc and LIMK1 genetically interact and drive F-actin polymerization via Myo1D
To further determine how F-actin dynamics is regulated in response to apoptotic stress, we examined the relationship between LIMK1 and Dronc, the caspase activating AiP in the proliferating eye and wing tissues [3,33]. Overexpression of Dronc or LIMK1 alone induces a moderate overgrowth of ey>p35 eye discs (Fig 7A-7B', compared to Fig 3A), which is associated with a weak level of F-actin accumulation and JNK activation (Fig 7A" and 7B"). In contrast, co-expression of both Dronc and LIMK1 triggers a strong F-actin accumulation, JNK activation and tissue overgrowth (Fig 7C-7C"). These data suggest that Dronc and LIMK1 genetically interact. Expression of cofilin also suppresses the hid-p35-induced MMP1 expression (F, F'). (G) Quantification of the MMP1 signal intensity in the representative wing disc pouches carrying clones of the indicated genotypes. Compared to the discs with p35 clones, the MMP1 signals are significantly ( ���� p < 0.0001) increased in the discs with hid-p35 clones. This increase is significantly ( ���� p < 0.0001) reduced in response to expression of dronc RNAi , bsk DN , cofilin or in Tak1 hemizygous null mutants or LIMK1 2 mutants. Myo1D, an unconventional myosin, has been previously shown to physically interact with Dronc and mediate its role on activation of JNK in AiP [20]. We therefore examined whether Myo1D mediates the interaction between Dronc and LIMK1. RNAi knockdown of Myo1D completely suppresses the phenotypes of F-actin accumulation, JNK activation and tissue overgrowth induced by co-expression of Dronc and LIMK1 (Fig 7D-7D"). Consistent with this, expression of Dronc or LIMK1 alone results in adult heads with moderate overgrowth phenotypes, including duplicated ocelli and bristles (Fig 7E and 7F). In contrast, co-expression of Dronc and LIMK1 causes semi-lethality with escapers showing severe adult head overgrowth phenotype (Fig 7G). This overgrowth phenotype is completely suppressed by knockdown of . This increased ROS production is suppressed by expression of SOD, a superoxide dismutase (C, C'). In contrast, RNAi knockdown of Duox, a transmembrane NADPH oxidase, does not obviously affect the level of ROS in hid-p35 clones (D, D', arrows). Increased ROS production in hid-p35 clones is also inhibited in a LIMK1 2 mutant background (E, E'), but not in response to expression of bsk DN , (F, F', arrows). (G) Quantification of the DHE signal intensity in the representative clones of the indicated genotypes in the wing disc pouches. Compared to the p35 clones, the DHE signals are significantly ( ���� p < 0.0001) increased in the hid-p35 clones. This increase is significantly ( ���� p < 0.0001) reduced in response to expression of SOD, cofilin or in   (Fig 7H). These suggest that Myo1D mediates the synergistic roles of Dronc and LIMK1 on promoting F-actin polymerization and tissue overgrowth.
Furthermore, we examined the roles of Myo1D in F-actin remodeling during AiP by clonal analysis. RNAi knockdown of Myo1D inhibits F-actin accumulation and its downstream activation of JNK in hid-p35 clones (Fig 8A-8D). Moreover, consistent with the role of LIMK1 in promoting F-actin polymerization and AiP, overexpression of LIMK1 enhances the ey>hid-p35 overgrowth phenotype indicated by an extensive tissue outgrowth (arrow, Fig 8E, compared to Figs 3B and S2A'). This results in adult semi-lethality with escapers showing severe head overgrowth phenotype with mostly loss of eyes (Fig 8G, compared to Fig 1C-1E). This enhanced phenotype is suppressed by expression of Myo1D RNAi (Fig 8F and 8H). Taken together, these data further confirm the critical role of Myo1D in mediating LIMK1-dependent F-actin polymerization and activation of AiP (Fig 8I).

Discussion
This study revealed an accumulation of F-actin in response to apoptotic stress and its essential role in activation of AiP (Fig 8I). LIMK1, an evolutionary conserved kinase promoting F-actin polymerization via phosphorylating and inhibiting Cofilin, acts as a key regulator of AiP. Our results further suggest that such F-actin remodeling depends on the initiator caspase Dronc and the unconventional myosin Myo1D. Importantly, actin polymerization mediates production of cytosolic ROS (cROS) and activation of JNK, which trigger AiP. Therefore, the dynamic behavior of actin filaments appears to be critical for the communication between apoptotic cells and their neighbors.
As the key structural network of the cell, the actin cytoskeleton is highly dynamic to support various cell behaviors including growth, proliferation, migration, and death [24,25]. It is therefore regulated both spatially and temporally to mediate the cellular responses to internal or external signals. For example, actin cytoskeleton reorganization is essential for the characteristic morphological changes during apoptosis such as cell shrinkage and plasma membrane blebbing [34,35]. In mammalian cells, effector caspases including caspase-3 were found to cleave cytoskeletal actin directly or Gelsolin, an actin-binding protein, to promote F-actin depolymerization and apoptosis [36][37][38]. Caspase-3 can also cleave and activate ROCK1, a kinase acting on cytoskeleton, leading to membrane blebbing [39]. Furthermore, in epithelial tissues in vivo, apoptotic cells are frequently extruded leading to delamination, a process depending on actin accumulation in the neighboring cells [31]. Therefore, actin remodeling is critical for execution of apoptosis. Interestingly, actin reorganization has also been observed at the induction stage of apoptosis. A recent study using time-lapse imaging detected an early accumulation of F-actin apically in irradiation-induced apoptotic cells in the Drosophila developing wing epithelium, prior to the basal extrusion of these cells and the appearance of any noticeable apoptotic characteristics such as cell shrinkage, nuclear condensation, and fragmentation [40]. The exact role of this early actin accumulation is not yet known. In this study, we have observed a similar actin accumulation in the apoptotic cells prior to the execution of cell death. This F-actin polymerization occurs downstream of the initiator caspase Dronc. It appears to be an early cellular response to the apoptotic stress as it mediates ROS production and JNK activation, which act as signals to communicate with the indicated by a duplication of bristles and/or ocelli (E, F, arrows). In contrast, co-expression of dronc and LIMK1 in ey>p35 results in semi-lethality with escapers showing severe head phenotypes, indicated by deformed head capsules with outgrown and amorphic tissues (G, arrows). These phenotypes are completely suppressed by expression of Myo1D RNAi (H).
https://doi.org/10.1371/journal.pgen.1010533.g007 Another intriguing question raised from our study is how the apoptotic stress promotes Factin polymerization. We showed that the initiator caspase Dronc is essential for this process. It genetically interacts with LIMK1 leading to actin polymerization, JNK activation and tissue overgrowth when execution of apoptosis is blocked (Fig 7). Notably, the human LIMK1 can be cleaved by caspase-3 and become constitutively active to promote membrane blebbing in apoptotic cells in culture [41]. This is unlikely the case in our study because 1) AiP in the proliferating tissues depends on the initiator caspase Dronc, but not the effector caspases; and 2) Factin polymerization observed in AiP occurs at an early stage of the apoptotic response, prior to any noticeable cellular morphological changes. However, it is still possible that LIMK1 is cleaved and activated by Dronc because the catalytic activity of Dronc is believed to be required for AiP [3,15]. Alternatively, LIMK1 may be involved in AiP without being cleaved as its expression is increased when AiP occurs ( Fig 1Q). The mechanisms regulating expression of LIMK1 and its activity therefore deserve to be further investigated.

PLOS GENETICS
In this study, we also observed that the interaction between Dronc and LIMK1 is mediated by Myo1D. Loss-of-Myo1D inhibits actin polymerization, JNK activation, and, consequently, induction of AiP (Fig 8). We have previously shown that, during AiP, Dronc relocates from cytoplasm to plasma membrane via a physical interaction with Myo1D [20]. It is possible that LIMK1-dependent F-actin polymerization assists the subcellular localization of Dronc to promote AiP. Notably, Myo1D also mediates a role of Dronc in activation of Duox, a membranebound NADPH oxidase, to produce extracellular ROS and attract hemocytes which in turn secretes Eiger, the ligand to activate JNK non-cell autonomously [20,21]. This process is important for AiP-dependent tissue overgrowth as loss-of-Duox or expression of scavengers specifically targeting the extracellular ROS suppresses overgrowth of the ey>hid-p35 discs [21]. Interestingly, in our clonal analysis of AiP which does not induce tissue overgrowth, it is the cytosolic ROS, but not the extracellular ROS produced by Duox, triggering action of JNK (Fig 6). In this case, although the mechanism controlling the cytosolic ROS production is indicated genotypes in the wing disc pouches. Compared to the hid-p35 clones, expression of Myo1D RNAi significantly ( ���� p < 0.0001) reduces F-actin levels in hid-p35 clones. (C, C') Images of Z-stack projections of a late 3 rd instar wing discs with 48-hour old mosaic clones, positively marked by GFP and labelled with MMP1 (magenta in C and grey in C'). Expression of Myo1D RNAi reduces the MMP1 level in hid-p35 clones (compared to Fig 5B and 5B'). (D) Quantification of the MMP1 signal intensity in the representative wing disc pouches with clones of the indicated genotypes. Expression of Myo1D RNAi significantly ( ���� p < 0.0001) reduces the MMP1 levels in the wing disc pouches carrying hid-p35 clones. (E, F) Late 3 rd instar eye discs of the indicated genotypes labelled with PHN. Expression of LIMK1 enhances the overgrowth phenotype in ey>hid-p35 by inducing further tissue outgrowth (E, arrow). This overgrowth phenotype is suppressed by expression of Myo1D RNAi (F). (G, H) Representative adult fly head images of the indicated genotypes. Expression of LIMK1 in ey>hid-p35 results in semi-lethality with escapers all showing severe overgrowth of head capsules and loss of eye tissues (G). These phenotypes are completely suppressed by expression of Myo1D RNAi (H). (I) Model summarizing roles of F-actin polymerization and its regulation in AiP. The data presented in this paper suggested that F-actin polymerization occurs in apoptotic signal-producing cells and is required for production of the cytosolic ROS (cROS) and its subsequent activation of JNK leading to AiP via secreting growth signals such as Dpp/TGF-β, Wg/Wnt and Spi/EGF. In response to apoptotic stress, Dronc may interact with LIMK1 via Myo1D to trigger F-actin polymerization and the initial activation of JNK. Moreover, Myo1D also mediates activation of Duox and production of the extracellular ROS (eROS) [21]. eROS work together with JNK to disrupt the basement membrane and attract hemocytes, the Drosophila macrophages, to further promote JNK activation non-cell autonomously via secretion of the ligand Eiger (Egr)/TNF [20,22]. We propose that these mechanisms, indicated by the dashed arrows, amplifies LIMK1-mediated JNK activation leading to tissue overgrowth.
https://doi.org/10.1371/journal.pgen.1010533.g008 unknown, JNK is not required because blocking JNK has no effects. However, in overgrown ey>hid-p35 discs, JNK, together with the extracellular ROS, damages the epithelial basement membrane and recruit hemocytes to promote tissue overgrowth [22]. Moreover, it has been reported that JNK can activate a feedback amplification loop via regulation of pro-apoptotic genes such as hid and rpr in apoptotic cells [42]. Therefore, it is possible that the cytosolic ROS are required for the initial activation of JNK in apoptotic cells to trigger AiP. When these apoptotic cells are kept 'undead', activated JNK promotes and works together with production of ROS including the extracellular ROS to recruit hemocytes and further amplify the JNK activity leading to tissue overgrowth (Fig 8I). Further investigation of the underlying mechanisms is required to test this hypothesis.

Drosophila genetics and stocks
Genetic crosses for all experiments were reared at 25˚C unless otherwise noted. Two ey>hid-p35 stocks were used to analyze AiP-dependent tissue overgrowth phenotypes. Their exact genotypes are UAS-hid; ey-Gal4 UAS-p35/CyO,tub-Gal80 and UAS-hid; ey-Gal4 UAS-p35/ TM6B,tub-Gal80. The stock UAS-hid; ey-Gal4 UAS-p35/CyO,tub-Gal80 was used to cross with mutants or transgenic lines to identify potential AiP regulators by scoring the adult hyperplastic overgrowth phenotypes of the F1 offspring. These phenotypes were categorized into three groups, weak (W), moderate (M) and severe (S), as previously described [15]. The stock UAS-hid; ey-Gal4 UAS-p35/TM6B,tub-Gal80 was used for the analyses of larval eye discs because of the convenience to identify the correct larval genotypes. Accordingly, the ey>p35 controls used in each experiment are the ey-Gal4 UAS-p35 on either second or third chromosome as appropriate.  [43], UAS-bsk DN [44], UAS-dronc [45], UAS-Duox RNAi [46] and GMR-hid [47] are as described. were raised at 18˚C. Expression of the UASconstructs (GFP, hid, LIMK1 RNAi , cofilin) was induced by a temporal temperature shift to 30˚C for 12 hours. This was then followed by a 72 h recovery period at 18˚C before the late 3 rd instar eye discs of each genotype were dissected and analyzed. Full details of the DE ts >hid assay have been described previously [15].

Mosaic analysis
For mosaic analysis with clones expressing transgenes, a heat shock inducible FLP-out assay combined with the GAL4-UAS system was used [48,49]. Late second instar (64-72h posthatching) larvae of the following genotype were heat shocked at 37˚C for 7 minutes and then grew at 25˚C for 48 hours before they were dissected and analyzed at the late 3 rd instar stage.

Quantitative reverse transcription PCR (RT-qPCR)
To measure expression of LIMK1 and cofilin in ey>hid-p35 (UAS-hid;; ey-Gal4 UAS-p35/+) versus the control ey>p35-mCherry (ey-Gal4 UAS-p35/UAS-mCherry), 100 third instar larval eye discs of each genotype were used to extract total RNA with an RNeasy Plus Mini Kit (Qiagen). For the analysis of LIMK1 RNAi efficiency, T80-GAL4 was used to drive expression of LIMK1 RNAi , 26294 or 57012, ubiquitously in all 3 rd instar larval imaginal discs. Total RNA was then isolated from 100 imaginal discs collected from either the control T80-GAL4 or T80-GAL4 UAS-LIMK1 RNAi 3 rd instar larvae using the TRI Reagent Solution (Invitrogen). Following RNA extraction, cDNA was generated from 300ng of total RNA with the GoScript Reverse Transcription System (Promega) for each genotype. This was followed by the realtime PCR using the SensiFAST SYBR Hi-Rox kit (Bioline) with an ABI Prism7000 system (Life Technologies). LIMK1 or cofilin mRNA levels were normalized to the reference gene ribosomal protein L32 (RPL32) by using the ΔΔCt analysis. Three independent biological repeats were analyzed in triplicate for each experiment. The following primers suggested by the FlyPrimerBank [52] were used: LIMK1 Fw, GTGAACGGCACACCAGTTAGT; LIMK1 Rv, ACTTGCACCGGATCATGCTC; RPL32 Fw, AGCATACAGGCCCAAGATCG; RPL32 Rv, TGTTGTCGATACCCTTGGGC. The primers used for cofilin were Fw, GTGAAA-GAAGGCGGAAGGTTAA, and Rv, CACAGTTACACCAGAAGCCATT, as described [53].

Imaging, quantification and statistical analysis
Adult fly eye images were taken using a Zeiss stereomicroscope equipped with an AxioCam ICC1 camera. Fluorescent eye and wing disc images were taken with a Zeiss LSM 710 confocal microscope. For quantification of the number of PH3-positive cells (Fig 1P) and F-actin aggregates (Fig 3F), ImageJ with the cell counter plugin was used to score representative larval eye discs (n�10) from each genotype. For quantification of the fluorescence intensity of PHN labeling (Figs 4K and 8B), MMP1 labeling (Figs 5G, 8D and S6B) or DHE staining (Fig 6G) in mosaic analysis and TRE-red signals in larval eye discs (S5F Fig), ImageJ was used to measure the fluorescence intensity of the area of interest (ROI) and its corresponding background area in each representative disc to calculate their mean value difference as the normalized fluorescence intensity of ROI. The ROIs and their background areas measured were selected using drawing tools in discs as follows: (1) to measure PHN or DHE signals, ROIs are the representative clones (n�15) in wing disc pouches while the background is their adjacent non-clonal areas; (2) to measure the MMP1 signals, ROIs are the whole wing disc pouches (n�8) while the background is the non-clonal areas in the same wing pouches; and (3) to measure the TRE-red signals, ROIs are the whole eye discs (n�10) while the background is the antenna disc areas without TRE-red signals. For the DE ts >hid regeneration assay, at least 10 representative eye discs from each genotype were measured for their sizes of dorsal versus ventral half of discs using the "histogram" function in Adobe Photoshop CS6, as previously described [54]. The statistical analysis in all experiments was conducted with GraphPad Prism 7 using either an unpaired student's T test (Figs 1Q, 1R, 8B, 8D and S6B) or a one-way ANOVA with Dunnett's multiple comparison test and plotted with Mean ± SEM.