A fluorescent toolkit for spatiotemporal tracking of apoptotic cells in living Drosophila tissues

The main function of epithelia is to form and maintain a physical barrier between different compartments of the body. Within epithelia, cells are highly polarized and organized in monolayers. They are connected by adhesive complexes such as apical tight junctions and adherens junctions, as well as basal focal adhesions. This organization ensures tissue integrity, a characteristic that needs to be maintained but is constantly challenged by cell division, cell death or cell intercalation. An epithelium has to deal constantly with the plasticity required to allow tissue growth and remodeling on the one hand and the maintenance of tissue cohesiveness on the other. In recent years, live imaging has become a crucial tool with which to identify cellular mechanisms involved in tissue plasticity.

Recent work has revealed that cells dying through programmed cell death (or apoptosis) can influence their environment either through biochemical communication or through the generation of mechanical forces (for reviews see Ambrosini et al., 2016; Pérez-Garijo and Steller, 2015; Suzanne and Steller, 2013). For example, apoptotic cells play an important role in the remodeling of epithelia, such as during Drosophila dorsal closure, histoblast expansion or mouse neural closure (Teng et al., 2017; Toyama et al., 2008; Yamaguchi et al., 2011). More recently, we have provided evidence that apoptotic cells of a columnar epithelium produce an apicobasal force that leads to tissue folding in the fly developing leg (Monier et al., 2015). Apoptotic cells also play a key role in vertebrate myoblast fusion and promote the growth of some tumors (Ford et al., 2015; Hochreiter-Hufford et al., 2013; Huang et al., 2011). Altogether, it is becoming clear that apoptotic cells communicate with their surroundings, which highlights the importance of studying apoptosis in an integrated model system. Understanding the role of apoptotic cells during developmental events makes it necessary to identify them easily and unambiguously, as well as to finely characterize their dynamics. Such a task has been particularly challenging so far because epithelial cells within living tissues are tightly packed. Furthermore, cells undergoing apoptosis progressively disappear, which narrows the window of observation.

Apoptosis is orchestrated by the hierarchical activation of cysteine proteases known as caspases that promote the sequential dismantling of doomed cells (Crawford and Wells, 2011). In Drosophila, death signals, such as expression of the proapoptotic genes reaper (rpr), hid or grim, trigger the degradation of the anti-apoptotic molecule Diap1, therefore promoting activation of upstream caspases (such as Dronc), which in turn activates effector caspases such as Dcp1 or Drice that usually cleave DEVD sequences (Fuchs and Steller, 2015). Drosophila is a model of choice for studying both the role and dynamics of apoptotic cells, yet available tools to track apoptosis in living tissues remain scarce. Doomed cells can be spotted using TUNEL staining or antibodies specific to activated caspases or to caspase targets, but this approach restricts the analysis to fixed tissues (Sarkissian et al., 2014; Williams et al., 2006). Additional methods have been developed to follow caspase activation (Arama and Steller, 2006; Bardet et al., 2008; Ding et al., 2016; Koto et al., 2009; Takemoto et al., 2003; Tang et al., 2015; To et al., 2015), yet all of them have limitations. So far, they provide only limited access to apoptotic cell dynamics either because they require post-acquisition image processing, such as FRET analysis, because they only detect late apoptotic cells or because they are activated in all cells that experience caspase activation whether they die or not (see Fig. S1). This prompted us to investigate whether alternative methods of detection of caspase activity could be implemented in Drosophila to easily follow the dynamics of dying cells in living tissues.

Interestingly, Li and co-workers (Zhang et al., 2013) recently generated four cyclized chimera proteins containing a caspase 3 cleavage site that behave as switch-on fluorescent-based indicators of apoptosis in cultured mammalian cells. When cleaved by a caspase3-like protease, the non-fluorescent indicator becomes rapidly fluorescent and thus detects the activation of such caspases (DEVDases) in real-time (Fig. 1A; Fig. S2). We have adapted these constructs to the Drosophila model. Here, we show that three of these fluorescent-based constructs allow efficient detection of apoptotic cells within a whole tissue (fixed or live), both in normal developmental conditions or after apoptosis induction. We therefore introduce new tools that allow easy visualization of the whole apoptotic process within living Drosophila tissues, from the very beginning of the apoptotic process to cell fragmentation and formation of apoptotic bodies. Moreover, the possibility to combine apoptosis detection with any other fluorescent fusion protein makes these sensors particularly useful.

In order to determine whether the GFP-based variant of caspase 3-like proteases activity indicator (GC3Ai, Fig. 1A) is an efficient reporter of the activation of the apoptotic pathway in Drosophila, as it is in mammalian cells in culture, we generated stable transgenic UAS::GC3Ai fly lines. First, we asked whether the expression of this construct per se leads to potential side effects. We could show that its ubiquitous expression has little impact on viability (see Fig. S3A). Furthermore, comparison of the level of apoptosis in presence or in absence of UAS::GC3Ai demonstrates that this molecule has little effect, if any, on the rate of apoptosis (see Fig. S3B,C). We next characterized the pattern of GC3Ai fluorescence (hereafter called the activation pattern) by expressing GC3Ai in the eye using the GMR::Gal4 driver (Fig. 1B). GC3Ai is detected in a small number of cells within the eye disc (Fig. 1C, control), a pattern reminiscent of the apoptotic pattern in this tissue (White et al., 1996). As expected, GC3Ai activation is abolished in cells that are homozygous mutant for the H99 deficiency (which removes the three pro-apoptotic genes hid, rpr and grim) or for the activator caspase Dronc (Fig. 1C,D). GC3Ai activation also disappears if apoptosis is blocked either by the expression of the anti-apoptotic factor Diap1 [which suppresses apoptosis by blocking both activator and effector caspases (Hay et al., 1994)] or p35 [which blocks only effector caspases (Hay et al., 1995)] (Fig. 1C,D). Conversely, expression of the pro-apoptotic genes rpr or hid greatly increases the activation domain of GC3Ai, reaching around 50 to 60% of the whole GMR domain (Fig. 1C,E). Finally, if p35 is co-expressed with Rpr, the activation of GCA3i diminishes drastically (Fig. 1C,E). Altogether, these results indicate that GC3Ai activity responds to the apoptotic machinery in the fly, as it does in mammalian cells (Zhang et al., 2013). Furthermore, the fact that GC3Ai activity is fully suppressed by p35 (Fig. 1C,D,E), which does not inhibit activator caspases (Hay et al., 1995), suggests that the GC3Ai biosensor reports essentially, if not only, the activity of effector caspases.

As the experiments presented above suggest that the GC3Ai apoptosensor is activated by effector caspases, we hypothesized that GC3Ai staining should colocalize with staining of activated effector caspases in fixed tissues. We therefore expressed GC3Ai in the leg disc at the onset of metamorphosis using the ap::Gal4 driver (Fig. 1F), a tissue in which the pattern and morphology of dying cells have previously been described (Monier et al., 2015). We counterstained with an antibody against the active form of the effector caspase Dcp1 (Sarkissian et al., 2014) and found that 99% (±1%) of Dcp1-positive cells are also GC3Ai positive (Fig. 1G, quantifications detailed in the Materials and Methods). Furthermore, we show that GC3Ai and the active form of Dcp1 are both detected in the same population of cells in a variety of tissues at distinct developmental stages, ranging from the embryonic ectoderm to larval tissues (imaginal discs and brain) and the pupal notum (Fig. S4). Altogether, our results indicate that GC3Ai efficiently detects apoptotic cells, thus validating the construct as a GFP apoptosensor in Drosophila.

In order to assess the suitability of GC3Ai for live imaging, we expressed and imaged it in the developing leg, a tissue in which apoptotic dynamics has been previously characterized using a FRET probe (Monier et al., 2015). Importantly, we found that the signal-to-noise ratio was high, even at low laser power, indicating that GC3Ai is adequate for good quality detection of apoptotic cells in vivo in optimal conditions for tissue culture. We therefore followed apoptotic cell dynamics in vivo and found that fluorescent GC3Ai is strongly and specifically detected in apoptotic cells, from a very early stage until cell fragmentation (Fig. 2A,B, Movie 1). Interestingly, the strong cytoplasmic signal obtained with GC3Ai makes it possible to follow the whole process of apoptotic cell morphological changes in detail. This includes a first step of apical constriction of the apoptotic cell (Fig. 2A, red arrowheads, 11′ to 33′), which culminates with adherens junction accumulation in the dying cell underneath yet in continuity with the adherens junctions of neighboring living cells (Fig. 2A, black and white arrowheads, 44′). At the same timepoint, apoptotic cell blebbing (Coleman et al., 2001) is visualized as round unfragmented protrusions (Fig. 2A, black arrows, 44′) while apical surface deformation of living neighboring cells (Fig. 2A, yellow arrow, superposition of dashed lines) reveals the generation of a transient apico-basal apoptotic force (Monier et al., 2015). The GC3Ai signal perdures right up to cell fragmentation (red arrows) which occurs concomitantly with detachment of the apoptotic adherens junctions (Fig. 2A, open arrowheads, 55′). Hence, apoptotic cell tracking through GC3Ai recapitulates the known morphological behavior of apoptotic cells in the leg disc, which has been initially visualized using the FRET sensor of caspase activity (Monier et al., 2015; Takemoto et al., 2003). Interestingly, both GC3Ai and the FRET sensor detect early events of the apoptotic process (detection of the apoptotic cells during apical constriction); however, GC3Ai surpasses the FRET probe in several aspects. First, it provides the ability to directly visualize an apoptotic cell under the microscope, whereas the FRET probe requires post-acquisition treatment to spot a dying cell. GC3Ai also allows 3D rendering of the cellular morphology of dying cells. Supporting this, individual apoptotic fragments can be counted and quantified using GC3Ai, which is extremely difficult or even impossible solely on the basis of the FRET probe (Fig. S5).

We next characterized the pattern of GC3Ai in fixed tissues. Interestingly, GC3Ai can be detected early during the apoptotic process, as shown by its detection in columnar cells, before the completion of apical constriction, a stage at which it is not possible to detect apoptotic cells using conventional tools such as anti-activated Dcp1 staining (Fig. 3, ‘apoptosis initiation’). We further demonstrated that more apoptotic fragments can be detected with GC3Ai than with Dcp1 staining, showing that GC3Ai allows detection of late apoptotic stages that cannot be analyzed using conventional strategies (Fig. 3, ‘apoptotic bodies’). Altogether, these results reveal that the GC3Ai pattern of activation is similar in both fixed and living tissues. GC3Ai is an efficient sensor of apoptosis that possesses an even higher sensitivity than Dcp1 in fixed tissues and allows real-time in vivo analysis, making GC3Ai a versatile tool with which to study apoptotic cell dynamics in the fly.

To further implement the apoptosensor toolkit, we generated UAS lines with three other fluorescent versions of the apoptosensor initially developed by Li and colleagues (mCherry-, Cerulean- and Venus-derived constructs, respectively RC3Ai, CC3Ai and VC3Ai) (Zhang et al., 2013). We tested these different lines by expressing them in the developing leg. Expression of the UAS-RC3Ai transgene leads to the formation of aggregates in all cell types (apoptotic or not), indicating that this C3Ai variant is inadequate to report apoptosis in the fly (data not shown). Alternative strategies, such as the one recently developed by To and colleagues (To et al., 2016) to label apoptotic cells in zebrafish could be considered in the future to engineer a red apoptosensor in Drosophila.

To characterize CC3Ai and VC3Ai constructs, we first combined them with GC3Ai and expressed them in the leg disc. Sequential detection of each one of these constructs with specific filters show that they are suitable for live imaging (good quality detection with low laser power). As expected, CC3Ai and VC3Ai are detected in apoptotic cells, as shown by their co-detection with GC3Ai (Fig. 4A,B). We noted, however, that VC3Ai appears to be detected earlier (during the completion of apical constriction) than CC3Ai, which is weakly or not detected in early apoptotic cells (Fig. 4A,B). We further checked that CC3Ai and VC3Ai are co-detected in Dcp1-positive cells in fixed tissues and quantified the proportion of VC3Ai- and CC3Ai-positive cells at different apoptotic stages (Fig. S6). In summary, the three constructs differ slightly in their temporal detection, appearing successively during the apoptotic process: first GC3Ai, then VC3Ai followed by CC3Ai. Finally, we tested their sensitivity to the apoptotic pathway by analyzing their activation pattern in the imaginal eye disc in different genetic contexts. Similarly to GC3Ai, ectopic expression of rpr leads to strong activation of CC3Ai and VC3Ai, while blocking apoptosis using Diap1 or p35 expression abolishes their activation (Fig. 4C). These results validate the use of CC3Ai or VC3Ai as alternative apoptosis reporters to GC3Ai.

To conclude, the apoptosensors described here offer the opportunity to recognize caspase activity and apoptotic cells directly within Drosophila tissues. These tools do not require post-acquisition processing and can be used either in fixed or living samples, bypassing the potential limitation of antibodies. These constructs detect a wide range of apoptotic stages and constitute excellent tools for following apoptotic cell dynamics within living tissues, particularly GC3Ai, which surpasses the detection range of Dcp1. Finally, the possibility of combining these apoptosensors with any other fluorescent marker provides a considerable stepping stone for analyzing the dynamics of tagged cellular components in apoptotic cells.

CC3Ai, GC3Ai, VC3Ai and RC3Ai cDNAs (a gift from Binghui Li's Lab, Tianjin Medical University Cancer Institute and Hospital, China) were individually cloned into the EcoRI site of pUAST Drosophila expression vector (a gift from C. Iampietro, LBCMCP, CBI, France) using the In-Fusion HD Cloning System (Clontech). The forward primer used to amplify the CC3Ai, GC3Ai, VC3Ai and RC3Ai cDNAs (including HA tag) was AGGGAATTGGGAATTCGCCACCATGTACCCCTACGACGTG. The reverse primer used to amplify CC3Ai, GC3Ai and VC3Ai (including Myc tag) was ATCTGTTAACGAATTCCGCTTACAGGTCCTCCTCGCTGATC. The reverse primer used to amplify RC3Ai was ATCTGTTAACGAATTCGCGGCCGCTTACACATTAATACTA.

pUAST-CC3Ai, pUAST-GC3Ai, pUAST-VC3Ai and pUAST-RC3Ai were generated in this study. P-element transformations were performed by the CBMSO Drosophila Transgenesis Service (Madrid, Spain) using random insertion in a white background. Insertions were recovered on each chromosome.

Apoptosensor expression was achieved using the following Gal4 lines from Bloomington: ap::Gal4[md544], Dll::Gal4[EM212], act5c::Gal4, hh::Gal4, en::Gal4, Tub::Gal4 and GMR::Gal4. Manipulation of apoptosis was performed using uas::rpr, uas::p35, uas::Diap1-H, uas::Diap1-P, uas::hid[4] and uas::rpr; uas::p35 (a gift from H. Steller, Rockefeller University, USA). UAS::Histone2Av-mKO was crossed to GMR::Gal4 and used as a control in the experiments reported in Fig. 1. UAS::alphaCatenin-TagRFP was a gift from K. Sugimura (iCeMS, Japan). Unless noted, stocks were obtained from Bloomington Drosophila Stock Center (BDSC). ubi::mCD8-DQVD-Gal4 and G-TRACE flies were obtained from M. Hardwick and D. Montell (Johns Hopkins Medicine, USA; UC Santa Barbara, USA) (Ding et al., 2016; Tang et al., 2015). Dronc^ΔA8 and Drice^Δ2C8 mutants (Kondo et al., 2006) were gifts from M. Miura (University of Tokyo, Japan). Analysis of GC3Ai expression in absence of hid, rpr and grim was achieved by crossing GMR::Gal4, uas::GC3Ai; Df(3L)H99, FRT2A flies with ey::Flp; ; tub::Gal80, FRT2A flies. In this context, GMR::Gal4 is expressed only in Df(H99)^−/− clones, which can be labeled positively by anti-GFP staining.

Leg discs dissection, immunostaining and ex vivo cultures were performed as previously described (Monier et al., 2015). Embryo staining was performed as described previously (Monier et al., 2010). Confocal images were acquired using a Zeiss LSM880 confocal using 25×/0.8 immersion and 40×/1.3 immersion objectives.

Antibodies and probes used were: rabbit anti-cleaved Dcp1 #9578 (1/200, Cell Signaling Technologies); rabbit anti-GFP #ab290 (1/500, Abcam); chicken anti-GFP #1020 (1/1000, Avès lab); phalloidin-rhodamine (1/200, Invitrogen); and phalloidin-Alexa647 (1/100, Interchim). Secondary antibodies coupled to Cy5 were obtained from Jackson Laboratories and diluted 1/50.

Endogenous fluorescence of CC3Ai, GC3Ai and VC3Ai is detected directly following excitation at 405 nm or 458 nm to visualize Cerulean, 488 nm for superfolderGFP or 514 nm for Venus. Non-fluorescent GC3Ai (i.e. GC3Ai in non apoptotic cells) can be detected in fixed tissues using rabbit or chicken polyclonal antibody against GFP. FRET and FM4-64 membrane stainings were performed and analyzed as described previously (Monier et al., 2015).

To quantify the number of apoptotic cells and fragments in pupal leg discs, 3D image stacks were reconstructed using Imaris software (Bitplane) and smoothed using a 3 voxel-large median filter. Fluorescent objects (cells and apoptotic bodies) were segmented on each channel separately (Dcp1, CC3Ai, VC3Ai and GC3Ai, respectively) using the Imaris Surface tool with a manual threshold. Only objects larger than 4 µm³ were included in the count. To quantify the number of GC3Ai-positive apoptotic cells and fragments within the Dcp1-positive population, we considered objects with at least a 50% overlap over the Dcp1 and GC3Ai channels.

To quantify the response of GC3Ai to the apoptotic machinery in the imaginal eye (Fig. 1), discs were stained with rabbit anti-GFP to reveal the domain of GMR expression. 3D image stacks were reconstructed using Imaris software (Bitplane). Channels corresponding to endogenous GC3Ai fluorescence and the GMR::Gal4 domain were segmented separately, as indicated above. The overall volume of GC3Ai endogenous fluorescence was normalized by the overall volume of the domain for each eye disc.

On all graphs, data are presented as box plots with median, lower and upper quartiles, whiskers extending to 1.5 times the interquartile range, mean and individual measurements. Statistical comparison of GC3Ai fluorescence in the GMR::Gal4 domain across conditions (Fig. 1D,E) or Dcp1 staining in en::Gal4 domain across conditions (Fig. S3C) were assessed using the Wilcoxon-Mann–Whitney rank-sum test. Asterisks indicate statistical significance: *** P<0.005.

We thank Binghui Li for sharing the sensor constructs; M. Hardwick, M. Miura, D. Montell and H. Steller for sharing stocks; E. Martin for comments on the manuscript; and the Bloomington Drosophila Stock Center (NIH P40OD018537) for numerous stocks used in this study.