Metazoan development requires cells to go through several identity changes as they transition from progenitor to a fully differentiated state. Given that the genome remains unchanged, developmental changes in cellular identity must occur through interconnected changes in the transcriptome and in the proteome. As proteases, apoptotic caspases can alter the proteome as seen in numerous examples. Transcription factor Paired Box 7 (Pax7) maintains muscle stem cell identify but its cleavage by apoptotic caspases allows differentiation into myoblasts, a step inhibited by a caspase resistant form of Pax7 (Dick et al., 2015). Caspase 3 null mutant (caspase 3^−/−) mice that survive to near-birth stages show reduced skeletal muscle mass. Isolated myoblasts from such mice failed to differentiate into myotubes in culture, indicating a requirement for caspase activity also in a late step in muscle differentiation (Fernando et al., 2002). This requirement may be through Mammalian Sterile Twenty-like kinase (MST1) because it includes caspase consensus sites and a shortened MST1 that mimics the cleaved product rescued differentiation in caspase 3 null mutant cells (Fernando et al., 2002; Murray et al., 2008). Additional instances of cellular differentiation, in either normal development or regeneration, show a requirement for apoptotic caspase activity without accompanying evidence for apoptosis [for example, (Fernando et al., 2002; Fernando et al., 2005; Koto et al., 2009),]. Caspase substrates that must be cleaved to allow differentiation include pluripotent factors such as Nanog Homeobox (Nanog) in mouse embryonic stem cells (Fujita et al., 2008) and Abnormal Cell Lineage 29 (Lin-29) during C. elegans larval development (Rougvie and Moss, 2013; Weaver et al., 2014). Likewise, apoptotic caspases contribute to two instances of cellular identity change in the Drosophila nervous system as seen in the following examples.

In the first example, apical caspase Dronc binds Numb to prevent the hyperproliferation of neuroblasts (Ouyang et al., 2011) (Figure 2A). As described above, partitioning of Numb to just one of two cells produced during an asymmetric neuroblast division allows the said recipient to differentiate. Without Numb, both daughter cells retain their progenitor status, leading to ectopic neuroblast formation (ENF) and hyperplasia. A yeast 2-hyrid screen identified Dronc as an interactor of Numb, and Dronc overexpression rescued the ENF phenotype caused by the expression of a dominant-negative numb mutant. This rescue required the presence of wild type Numb, suggesting that the rescue occurs through sequestration of mutant Numb to allow the wildtype Numb to work. The rescue did not require effector caspase activity or the catalytic activity of Dronc, suggesting that Dronc acts in a non-apoptotic manner. Full length Dronc binds Numb as well as cleaved active Dronc, suggesting that any activity provided by Dronc to limit neuroblast number can occur without fear of killing the cell. dronc loss-of-function mutants do not have extra neuroblasts but exacerbate the milder ENF phenotype of weak dominant negative numb mutants. Therefore, Dronc appears to provide an important but redundant function to limit neuroblast number.

Apoptotic caspase activity also limits the number of SOPs, progenitor cells in the PNS (Figure 2A). A subset of SOPs produce external sensory organs called macrochaetes that are readily visible on the back of the adult fly, specifically, in a triangular region called the scutellum. Mutations in dark or dronc as well as inhibition of caspase activity with a dominant-negative Dronc (Dronc ^DN ) or p35 in the scutellum produced extra SOPs and extra macrochaetes. One possible explanation for this phenotype is that caspases normally reduce SOP cell numbers through apoptosis. But this possibility was rendered unlikely by the absence of TUNEL-positive cells in the pro-neural clusters that produce the SOPs. Instead, caspases may suppress Wingless (Wg, Drosophila Wnt1) signaling to limit the SOP number; a mutant allele of wg suppressed the extra macrochaetes phenotype that results from Dronc ^DN while null mutants in Wg inhibitor shaggy (sgg, encoding Drosophila GSK3 kinase) enhanced this phenotype. Sgg contains two caspase 3/7 consensus sites. Expression of Sgg with caspase consensus sites mutated produced extra SOPs and macrochaetes, leading to the model that caspases cleave to activate Sgg, which then opposes Wg to limit the SOP number (Kanuka et al., 2005). As the non-cleavable Sgg produced a phenotype in the presence of endogenous wild type Sgg, the former was proposed to act as a dominant negative against the latter. How might caspase activity finetune SOP number without killing the cells? Additional studies provide evidence that Drosophila I-kappaB kinase (IKK)-like kinase regulates the turnover of Diap1 to temper caspase activity and enables it to provide a developmental role without inducing cell death (Kuranaga et al., 2006; Koto et al., 2009).

While these studies provide compelling evidence for a non-apoptotic role of apoptotic caspases in Drosophila PNS, the data also suggest complexities. For instance, non-cleavable Sgg was less effective than inhibition of Dark or Dronc in producing extra macrochaetes, leading the authors to suggest additional substrates besides Sgg at play (Kanuka et al., 2005). A more recent paper addressed this possibility directly by using CRISPR to replace endogenous Sgg with non-cleavable Sgg (Wang and Baker, 2019). The resulting flies, surprisingly, show normal SOP and macrochaete numbers. These results can be reconciled if non-cleavable Sgg acts not by opposing endogenous Sgg but another protein that is the bona fide caspase substrate in controlling SOP number. In other words, while the data support the conclusion that apoptotic caspases play a non-apoptotic role in controlling SOP number, we may not yet know their relevant substrates in this process. The newer study identified an additional upstream regulator of non-apoptotic caspase function in limiting SOP number. Inhibition with mutations or RNAi of expanded (ex), which encodes for a signal transducer in the Hippo (Hpo) pathway, produced extra macrochaetes as did over-expression of transcription factor Yorki (Yki, Drosophila YAP), an inhibitory target of Hpo signaling. The macrochaetes phenotype of ex mutants was rescued by reducing diap1, a known transcriptional target of Yki (Wang and Baker, 2019). These results suggest that Hpo/Yki tumor suppressor pathway regulates Diap1 transcriptionally in SOP number determination. This mechanism could operate in parallel with the IKK-mediated mechanism that controls Diap1 protein turnover (Kanuka et al., 2005).

During apoptosis execution, caspase activity helps destroy organelles such as the nucleus and the Golgi (Chiu et al., 2002; Al-Ghorbani et al., 2016). Likewise, caspase activity helps eliminate organelles in some instances of cellular remodeling that accompany differentiation. The removal of the nucleus from the lens cells for light transmission and from red blood cells for efficient oxygen transport requires caspases (Wride et al., 1999; Zermati et al., 2001). In Drosophila, Drice is needed to reduce the cytoplasm and associated organelles during the final step of sperm differentiation (Arama et al., 2003). Inhibiting effector caspase activity with p35 prevented cytoplasm elimination and produced sterile males, illustrating the importance of caspase activity in sperm development. The following examples from two sets of neurons illustrate that subcellular sculpting by apoptotic caspases occurs also in the Drosophila PNS (Figure 2B).

Larval PNS neurons prune their dendrites as the larva metamorphoses into an adult. A genetic screen identified ubcD1, an E2 ubiquitin-conjugating enzyme, as a requirement for pruning in C4da neurons (Kuo et al., 2006). A relevant target of this degradation system appears to be Diap1; a diap1 mutant that was an inefficient substrate for degradation acted as a dominant negative to inhibit pruning. dronc mutants show pruning defects, suggesting that degradation of Diap1 allows Dronc activation and pruning. How is cell death avoided in such a scenario? An antibody against cleaved human caspase 3 shows signal only in dendrites and not in axons and only during the period of dendrite severing. Thus, Diap1 may be degraded locally to produce localized caspase activity that affects localized subcellular remodeling without endangering the cell. Interestingly, expression of p35, which inhibits effector caspases, did not affect pruning, suggesting that apical caspase activity but not effector caspase activity promotes pruning in C4da neurons. In ddaC neurons, a genetically encoded caspase activity reporter shows spatially-restricted caspase activation in dendrites undergoing pruning (Williams et al., 2006). Overexpression of Diap1, Dronc^DN or p35 inhibited pruning in ddaC neurons suggesting that both apical and effector caspase activities are needed to prune these neurons. Global caspase activation by overexpression of hid resulted in ddaC cell death (Williams et al., 2006), providing a contrast to localized caspase activation that allows pruning without cell death.

A more recent study shows that apoptotic caspases play a non-apoptotic role in a phenomenon called neuroprotection that occurs during axonal regeneration (Chen et al., 2016). Briefly, severing of an axon (axotomy) in Drosophila sensory or motor neurons results in neuroprotection (NP) that acts through Jun N-terminal kinase (JNK) and Mitogen-Activated Protein kinase (MAPK). NP can be visualized as preservation of severed dendrites following axotomy and is proposed to buy neurons time to maintain structure and function through the period of axon regeneration. RNAi against dronc or mutations in drice increased the fraction of dendrites protected in ddaE neurons, suggesting that these caspases normally contribute to the removal of severed dendrites, much like in pruning, and must be overcome for NP. The authors then tested the hypothesis that uncontrolled NP might interfere with subsequent axon regeneration. Indeed, neurons with dronc RNAi showed reduced regeneration, leading to the conclusion that Dronc activity promotes axon regeneration by limiting NP.

Neuronal pruning by localized apoptotic caspase activity in Drosophila has parallels in mice where proteasome activity and localized expression of an IAP protein keep the activity of caspases 3 and 9 away from the cell body and restricted to axons to be pruned (Cusack et al., 2013).

Fungi and plants encode metacaspases, which are proteases that share similar tertiary structure and an active-site cystine with caspases but cleave after Arg or Lys instead of Asp [reviewed in (Minina et al., 2017)]. Budding yeast metacaspase YCA1 is dispensable for cell death (Hauptmann et al., 2006; Kang et al., 2008; Palermo et al., 2013) but yca1 mutants show elevated stress response proteins, cytoplasmic protein aggregates and reduced replicative lifespan (Lee et al., 2010; Hill et al., 2014). These data suggest a non-lethal role for a metacaspase in removing harmful protein aggregates from the cytoplasm. Paradoxically, apoptotic caspases may provide an opposite function, to clutter rather than to clean, in Alzheimer’s disease (AD) where microtubule-binding protein Tau forms aggregates called neurofibrillary tangles (NFT) (Medeiros et al., 2011). Several lines of data suggest that cleavage of Tau by apoptotic caspases leads to NFT. Tau bears a caspase 3-consensus site and the caspase-cleaved form of Tau as well as cleaved (active) caspase 9 are detected with specific antibodies in neurons from AD patients but not non-AD controls (Rohn et al., 2002; Gamblin et al., 2003). In mouse disease models, activation of effector caspases precedes NFT formation and caspase-cleaved Tau is sufficient to promote NFT formation in wild type brains (De Calignon et al., 2010). Finally, caspase-cleaved Tau forms fibrils more rapidly than full length Tau in vitro (Gamblin et al., 2003). Collectively, these data lead to the model that apoptosis caspase activity that is not immediately followed by cell death results in Tau NFTs and disease.

A Drosophila model of AD demonstrates a non-lethal role of apical caspase Dronc in Tau cleavage and enabled an investigation of the observed correlation between circadian dysregulation and susceptibility to neurodegeneration (Means et al., 2015). Disruption of circadian rhythm by RNAi-mediated knock-down of circadian kinase encoded by double time (dbt) or its genetic interactor spaghetti in Drosophila clock neurons led to Dronc activation as detected with an antibody against the cleaved form. Dronc activation in these experiments was observed only during daytime or after light exposure during the night, could spread to nearby non-clock cells, and increased with fly age. Expression of a dominant-negative dbt in the Drosophila eye led to Dronc activation in this tissue, cleavage of co-expressed human Tau in a Dronc-dependent manner, and neurodegeneration phenotypes. These results in Drosophila parallel the relationship between apoptotic caspase activity, Tau and neurodegenerative disease observed in mammals as described in the preceding paragraph. Furthermore, Drosophila studies add to this picture by identifying Dronc activation as a possible mechanism that links circadian dysregulation with susceptibility to neurodegeneration.

Studies in mice and Drosophila show that dying cells send mitogenic signals that promote tissue homeostasis. In Drosophila larval wing imaginal discs, Dronc promotes Apoptosis-induced Proliferation (AiP) through different mechanisms. First, Dronc acts together with JNK to produce Wg that acts as a secreted mitogen (Perez-Garijo et al., 2004; Ryoo et al., 2004; Kondo et al., 2006). Second, Dronc activation leads to the elevation of extracellular Reactive Oxygen Species, which recruits macrophages that secrete Eiger/Tumor Necrosis Factor (TNF) to activate JNK and sustain mitogenic signaling (Fogarty et al., 2016; Khan et al., 2017). The role of Dronc in AiP is distinct from its role in activating effector caspases for apoptosis because AiP still occurs when the effector caspases are inhibited. In mice, effector caspases 3/7 cleave and activate Calcium-independent phospholipase A2 to result in the generation and release of prostaglandin E, a known promoter of cell proliferation (Li et al., 2010; Huang et al., 2011). Thus, apoptotic caspase activity produces non-cell autonomous mitogenic signaling in both Drosophila and mammals, albeit through different downstream targets.

A parallel story has emerged recently in Drosophila thermal nociception (sensing painful heat) where Dronc and Eiger function in a non-apoptotic manner to promote sensitization to pain (Jo et al., 2017) (Figure 2C). The exposure of Drosophila larval epidermis to UV results in both apoptosis and allodynia (extreme sensitivity to pain). The latter outcome can be observed as an adverse response to thermal stimuli that would not elicit a response in non-irradiated animals. Dronc is required for both UV-induced apoptosis and thermal allodynia while effector caspases Drice and Dcp1 are required only for apoptosis, suggesting that Dronc plays a non-apoptotic role to induce allodynia. This idea is supported by the finding that doses of UV that are too low to induce apoptosis still induced thermal allodynia in a Dronc-dependent manner. Dronc, it was found, acts in the epidermis to produce Eiger/TNF that then signals through TNF Receptor (TNFR) on sensory neurons to induce thermal allodynia. Induction of Dronc in the epidermis, even when effector caspases are inhibited to prevent Dronc-induced apoptosis, results in thermal allodynia in an Eiger-dependent manner without UV exposure, further lending evidence to a non-apoptotic role for this apoptotic caspase.