Temporal Heterogeneity Metrics in Apoptosis Induced by Anticancer Drugs

Abstract

The apoptotic process is highly heterogeneous and asynchronous. A long-standing question is how many parameters define the time and reversibility of the apoptotic response at a single-cell level. We characterized at the single-cell and population levels the time sequence of apoptotic events in response to anti-cancer drugs using extrinsic and intrinsic apoptotic stimuli. We show that the temporal sequence of major apoptotic events is the same in response to all anti-cancer drugs studied: the apoptotic volume decrease and Na⁺ influx occur rapidly and are tightly coordinated with mitochondrial outer membrane depolarization (MOMP), mitochondrial inner membrane depolarization and a decrease in the production of reactive oxygen species (ROS). Phosphatidylserine externalization usually starts after MOMP and precedes caspase 3/7 activation. Activation of caspases 3/7 is a slow process that always starts after MOMP, with significant delay. Cell-to-cell variability of the MOMP onset is described by Gaussian distribution, whereas the γ-distribution model describes cellular variability in the duration of MOMP-to-caspase activation stages. Cells from the pre-MOMP stage to the after-caspase 3/7 activation stage coexist for many hours. We demonstrated by FACS that cells in the pre-MOMP stage can recover after apoptotic stimuli, rarely recover after MOMP but before caspase 3/7 activation, and are unable to recover after caspase 3/7 activation. We propose a double-stroke model for apoptosis execution.

Introduction

Apoptosis is a form of programmed cell death that occurs in a largely asynchronous manner in cell populations (Kroemer et al. 1995; Messam and Pittman 1998; Rehm et al. 1992; Green 2005; Albeck et al. 2008; Hellwig et al. 2008; Bhola and Simon 2009; Rehm et al. 2009; Spencer et al. 2009). Indeed, heterogeneity is an intrinsic feature of cell populations and is the basis of cell fate regulation in a variety of processes (Spiller et al. 2010). The current understanding of the mechanisms involved in apoptotic processes draws mainly from integrated measurements of overall cell populations (Galluzzi et al. 2009) and, more recently, from single-cell analyses using time-lapse fluorescence microscopy (Rhem et al. 2002; Hellwig et al. 2008; Rhem et al. 2009; Spencer et al. 2009). Although valuable information has been gained from these studies, the relationship between changes in cell morphology (cell rounding and cell volume), mitochondria outer membrane permeabilization (MOMP), and executive caspase activation, along with their impact on the heterogeneity of cell response, is poorly understood.

The key events in the progression and execution of apoptosis include apoptotic volume decrease (AVD), cell shrinkage, phosphatidylserine (PS) externalization on the plasma membrane, nuclear condensation, blebbing, and fragmentation. These features are orchestrated by the activation of cysteine proteases, namely caspases (Green 2005). In some cases, morphological changes are preceded by MOMP and are thought to occur in parallel with the executive caspase activation (Kroemer et al. 1995; Albeck et al. 2008; Spencer et al. 2009; Tait and Green 2010, 2013). However, the relationship between AVD, MOMP, caspase activation, PS externalization and morphologically distinguishable events is poorly investigated.

In the present study, we compared apoptosis execution at the single-cell level using extrinsic and intrinsic apoptotic stimuli and found a striking similarity in the sequence of apoptotic events at the population and single-cell levels. Each cell under any treatment undergoes the following sequential events: cell rounding; rapid MOMP, accompanied by loss of mitochondrial membrane potential and a decrease in the production of reactive oxygen species (ROS) by mitochondria; volume decrease; and PS externalization on the cell surface in parallel with the gradual activation of caspases 3/7 accumulation, presumably in the nucleus. Cell-to-cell variability at the MOMP stage is described by Gaussian distribution, whereas the γ-distribution model describes cell-to-cell variability in the pre-MOMP and MOMP-to-caspase activation stages. Sorting different subpopulations after induction of apoptosis, we found that cells before MOMP can effectively recover after apoptotic stimuli, whereas passing the MOMP stage makes cell recovery unlikely, and the activation of executive caspases makes cell recovery completely impossible. Finally, a double-stroke model for apoptosis execution/cell recovery is proposed.

Materials & Methods

Cell Culture and Transfections

HeLa cells were cultured in DMEM supplemented with l-glutamine, HEPES and penicillin/streptomycin (all from Gibco-Life Technologies; Grand Island, NJ) and 10% newborn bovine serum (NBS; Sigma-Aldrich, St. Louis, MO). Jurkat, H9, THP-1 and K-562 cells were cultured in RPMI-1640 supplemented with l-glutamine, HEPES and penicillin/streptomycin and 10% NBS. HeLa cells stably expressing Smac-DIABLO-GFP were kindly provided by Dr. D. R. Green and cultured in complete DMEM.

For caspase 3/7 staining, CellEvent^TM Caspase 3/7 Green reagent (Life Technologies), a fluorogenic substrate for activated caspase-3/7, was used, according to the manufacturer’s protocol. The cells were incubated with 5 μM substrate for not less than 30 min at 37°C and then analyzed under the microscope or with a FACSAria (BD Biosciences, San Jose, CA). For mitochondrial potential staining, tetramethylrhodamine ethyl esther perchlorate (TMRE; Life Technologies) was used, as described elsewhere (Barteneva et al. 2014). For nuclear staining, Hoechst 33342 (Life Technologies) was used at a concentration of 100 nM.

Live-Cell Microscopy

HeLa cells stained with different fluorescent probes (TMRE, caspase 3/7 substrate, Hoechst 33342, Annexin V-Alexa Fluor 647, Cell ROX Deep Red and Sytox Blue (SB) from Life Technologies; DRAQ5 and DRAQ7 from Biostatus; Shepshed, UK; Asante Green AM (Asante NaTRIUM Green-2 AM) from Abcam, Cambridge, MA) or expressing Smac-DIABLO-GFP were imaged in a 37°C chamber on a Zeiss Axiovert 200 fluorescence microscope (Carl Zeiss Microscopy; Thornwood, NY) equipped with a CoolSNAP HQ2 camera (Photometrics; Tucson, AZ) and using SlideBook 5.0 software (Intelligent Imaging Innovations Inc., Denver, CO). For observation of apoptosis, HeLa cells were plated onto glass-bottomed dishes (MatTek Inc.; Ashland, MA) and imaged at 5-min intervals for 20–30 hr in CO₂-independent medium (Gibco-Life Technologies) with l-glutamine, penicillin/streptomycin and 10% NBS. Staurosporine (STS) (1 μM; Sigma-Aldrich), TNF-related apoptosis-inducing ligand (TRAIL) (100 ng/ml; PeproTech, Rocky Hill, NJ), cycloheximide (CHI) (2.5 μM; Sigma-Aldrich), Actinomycin D (Act D) (5 μM; Sigma-Aldrich) and etoposide (50 μM; Sigma-Aldrich) were each added before imaging. Imaging was commenced after addition of STS and TRAIL immediately, and after a delay of 2–15 hr after the addition of Act D and etoposide. For continuous observation, cells were imaged using a Planfluor 10×/0.3 and PlanApo 40×/1.3 or PlanApo 63×/1.4 oil-immersion objectives. To build cell “life histories”, we used time-lapse sequences of 4 or 5 channels (DIC and 3 or 4 fluorescent channels) for up to 15 hr and consisting of no more than 200 frames. To minimize the photo activation effect, we put intervals between consecutive frames every 5 min for prolonged recording and used a low aperture objective ×10/0.3. More frequent illumination led to a loss of TMRE staining after 30–40 frames in control cells (data not shown), whereas less frequent recordings made it impossible to build life histories with sufficient accuracy. Several events were analyzed in different modes: the overall sequence of apoptotic events was monitored at 5-min time intervals using low magnification, and specific aspects of the apoptotic process were monitored using a high-power objective (×40/1.3 oil) and shorter time intervals (30 sec to 3 min).

Asante Green AM was added to the culture medium and cells preliminarily stained with TMRE and Hoechst 33342 were imaged 30–40 min after staining. Asante Green AM cannot be used in time-lapse observations because it is redistributed between cell compartments within 60–90 min after its addition into culture medium and makes the cells extremely light sensitive (data not shown). DRAQ5 was added to the culture medium and cells preliminarily stained with TMRE and caspase substrate were imaged at 5–10 min after staining.

Image Analysis

Multichannel images were spectrally compensated before analysis. For the compensation, single-staining controls were prepared, signals were measured in each fluorescence channel, according to flow cytometry standards (Shapiro 2003), and the percentage of signal from one channel was determined and subtracted from the signal of neighboring channels using Channel operation option in the SlideBook software.

Fluorescence intensity was measured in single-channel 16-bit images after compensation using ImageJ software (NIH, Bethesda, MD). TMRE, Cell ROX Deep Red, and caspase 3/7 fluorescence were quantified by measuring the average fluorescence intensity from a minimal ellipse where a cell was inscribed. When using large magnification (×40), fluorescence intensity was summarized over 2–3 optical sections spaced by 2 µm from each other. Since Asante Green AM is known to have wide emission spectrum, images in the red channel were obtained after 50% compensation with FITC channel. Multichannel 16-bit images were deconvolved using No-neighbor or Nearest-neighbor algorithms in the SlideBook software, rescaled, and 16-bit or 24-bit images were processed for presentation with Adobe Photoshop (Adobe Systems; Mountain View, CA).

Flow Cytometry and Cell Sorting

A FACSAria II cell sorter (BD Biosciences; San Jose, CA), equipped with 488 nm, 640 nm, 407 nm and 561 nm lasers, was used. TMRE dye was excited by 561-nm laser and its fluorescence was captured using a bandpass filter at 582/15 nm. SB and caspase 3/7 substrate were excited by 407-nm and 488-nm lasers, respectively, and measured with 450/50 and 530/30 bandpasses, respectively. Flow cytometry of cultured cells stained with different dyes was performed as described previously using FACSAria cell sorter (Barteneva et al. 2014). At least 10,000–30,000 events were acquired. The data on Annexin-V, caspase 3/7 and Δψ staining were analyzed by Diva 6.1 (BD Biosciences) and FlowJo software v. 10 (Treestar; Ashland, OR).

For cell sorting, cells were stained with TMRE and caspase 3/7 substrate, treated with STS or TRAIL+CHI for 1–4 hr and sorted into 4 fractions (as described later in Figure 4). Each fraction contained at least 150,000 cells. Sorted cells were transferred to complete culture medium and cultured as described above for at least 16 hr, then stained with TMRE, caspase substrate and SB, and analyzed.

Figure 4.

Rounding of HeLa cells during apoptosis. Left columns: Rapid rounding of two cells under TRAIL (100 ng) treatment. Right columns: Synchronous shrinkage of cells under staurosporine (STS; 1 μg/ml) treatment. Time (min) is given in the upper left corner. Scale, 20 μm. Images were taken with Planapo ×40/1.3 objective.

Imaging Flow Cytometry

Imaging cytometry analysis was performed on Imagestream 100 (Amnis-Millipore; Seattle, OR) equipped with 488-nm, 658-nm and 405-nm laser sources and brightfield light source, using 40× objective, as described elsewhere (Ponomarev et al. 2014). For subsequent analysis with IDEAS software (Amnis-Millipore), we collected at least 5,000 cell events. For quantitative analysis, data were exported into Microsoft Excel (MS Office; Redmond, WA) and re-analyzed.

Statistical Analysis

Statistical analysis of flow data was performed using MS Excel and Statistica 8.0 (StatSoft; Tulsa, OK) software. P-values were calculated by Student’s t-test in MS Excel. Linear regression and curve fitting was performed using Statistica 8.0 software.

Results

We analyzed the sequence of apoptotic events at population and single-cell levels by treating Jurkat, H9, THP-1, K-562 cells and HeLa cells with extrinsic (TRAIL) or intrinsic (STS, Act D, etoposide) apoptotic stimuli. Using multicolor and multiprobe fluorescence assay (mitochondrial potential dye TMRE, caspase 3/7 substrate, SB or DRAQ5 viability dye and Annexin V-Alexa 647) we showed that, after induction of apoptosis, cell populations always become highly heterogeneous (Fig. 1A, 1B, Supplemental Fig. S1) and remain heterogeneous until the last cells enter the apoptotic pathway (Fig. 2, Supplemental Figs. S2, S3). At the early stages (Fig. 2A), the first visible event during the execution of apoptosis was a loss of mitochondrial membrane potential, which is known to be a manifestation of MOMP (Green and Kroemer 2004) and rounding of adhesive cells. The late stages of apoptosis were defined by the presence of cells with active caspases 3/7 and a permeable plasma membrane (red nuclei in Fig. 2B). The most potent inducer, STS, brought >90% of Jurkat, HeLa and THP-1 cells into late apoptotic stage within 8 hr of induction, whereas Act D and etoposide did so only after >24 hr. H9 and K-562 cells were more resistant to STS and TRAIL treatments but demonstrated the same sensitivity as the aforementioned cell lines to Act D treatment (data not shown).

Figure 1.

After induction of apoptosis with different inhibitors, the cell population becomes heterogeneous. (A) Dot plots showing Jurkat cells labeled with the caspase 3/7 substrate, TMRE (Δψ_m dye) and Annexin V. Upper row, left: control (spontaneous apoptosis is < 5%); middle and right: STS (1.0 μg/ml). P2, TMRE^high cells; P3, TMRE^low/caspase ^low cells; P4, caspase^high cells; P5, Annexin V⁺ cells. Lower row, left to right: heterogeneity of cell populations according to TMRE/caspase 3/7 staining. Treatments: Act D (5 μM), etoposide (50 μM) and TRAIL (100 ng/ml) + CHI (2.5 μM). TMRE staining reveals TMRE^high/AnnexinV^– and TMRE^low/AnnexinV⁺ populations. The TMRE^high cells became negligible (<1%–2%) during the various time intervals: 8 hr (STS); 16 hr (TRAIL+CHI); 16 hr (Act D); 36 hr (etoposide). The intermediate groups (TMRE^low/caspase^intermed and TMRE^low/caspase^low) were detectable for several hours. (B) Jurkat T-cells in early apoptosis (STS treatment for 2 hr, 30 min; TRAIL, Act D and etoposide treatments for 8 hr). From left to right: four fluorescent channels and merged color images (Hoechst 33342, blue; caspase 3/7, green; TMRE, red; Annexin V, yellow). TMRE^high cells have, on average, a larger diameter. Annexin V stains mainly the plasma membrane; caspase 3/7 is accumulated primarily in the nuclei. Scale, 20 μm. Images were taken with Planapo ×40/1.3 objective. A representative of three experiments.

Figure 2.

Early (A) and late (B) apoptosis in HeLa cells. Left column: DIC image. Right column: Overlay of four fluorescence channels in the same field of view. Hoechst 33342, blue; caspase 3/7 substrate, green, TMRE, red and DRAQ7, magenta. Early in apoptosis, caspase activity could be determined only in rounded cells. DRAQ7-positive (i.e., dead) cells could be seen only in late apoptosis. Notice the large number of inflated cells (arrowheads) with a high level of caspase 3/7 expression. Time (in hr) and treatment is given in the upper left corner. Scale, 20 μm. Images were taken with Planapo ×40/1.3 objective.

MOMP, Δψ_m and ROS Production Timeline

To precisely determine the timeline of MOMP processes and the loss of Δψ_m, we analyzed HeLa cells stably transfected with GFP-Smac/DIABLO several hours after induction of apoptosis using time-lapse microscopy for 1–2 hr. MOMP is able to be determined by the exit of cytochrome c or Smac-DIABLO; upon the exit of these proteins, transfected cell staining pattern shifts from the mitochondria into a diffuse staining pattern (Goldstein et al. 2000). Stably transfected cells allowed us to follow, simultaneously, TMRE staining and MOMP under fluorescence microscopy. Loss of Δψ_m occurred simultaneously with permeabilization of the outer mitochondrial membrane; however, some mitochondria retained TMRE staining several minutes after exit of Smac/DIABLO into the cytosol (Fig. 3A). Since the two processes are tightly coordinated at the single-cell level, we further used the loss of TMRE staining as a surrogate marker for MOMP in all treatments. Next, we compared the loss of Δψ_m with the production of ROS. Upon treatment with Act D, etoposide or TRAIL, the loss of Δψ was usually a single-step event over 5–10 min, which occurred in parallel with a loss of ROS production (Fig. 3). However, in response to STS treatment, residual Δψ_m often remained for >1 hr (Fig. 3B) and, in this case, ROS production continued (Supplemental Fig. S4). Using TMRE as a surrogate marker of MOMP, we determined the timing of MOMP in the whole cell population. Prolonged time-lapse observations revealed a great deal of heterogeneity in MOMP timing in the HeLa cell population (Fig. 3C). This heterogeneity demonstrated nearly a Gaussian distribution in all cases (Fig. 3C), with a similar coefficient of variation for all drugs tested (Supplemental Table S1).

Figure 3.

Mitochondrial outer membrane potential (MOMP), Δψ_m loss (TMRE) and Na⁺-influx occur simultaneously in a cell but asynchronously in the cell population. (A) A time-lapse of HeLa cells undergoing apoptosis (staurosporine, STS). MOMP in two cells: Smac-DIABLO-GFP release from mitochondria in transfected HeLa cells along with loss of Δψ_m (TMRE). Smac-DIABLO-GFP, green; TMRE, red. Loss of Δψ_m occurs simultaneously with permeabilization of the outer mitochondria membrane, although some mitochondria retain TMRE staining several minutes after complete exit of Smac-DIABLO-GFP into the cytosol (indicated by arrows). Time (min) is given in the lower left corner. (B) Loss of Δψ_m is accompanied by a loss of ROS: Dynamics of the signal in HeLa cells stained for Δψ_m (TMRE; red curve) and ROS (Cell ROX Deep Red; green curve) under Actinomycin D (Act D; left) and STS (right). For each plot, 10 cells were synchronized at the time of MOMP (indicated by arrows). 90% loss of TMRE staining always occurs within 10 min. (C) Histograms showing the timing distribution from the beginning of treatment until MOMP in HeLa cells (treatments indicated in each plot; also in Supplemental Table S3). Red line indicates an approximation with gamma-function. All histograms are well described by Gaussian distribution, since the shape parameter in gamma distribution is large. (D, E) Asante Green AM staining of HeLa cells after STS treatment (D) and TRAIL+CHI treatments (E). Distance between optical sections is given in µm. Arrowheads show cells with low Δψ_m (red) and high Asante Green AM signal (green). Hoechst 33342 is stained blue. Scale (A) 10 μm; (D, E) 20 µm. A representative of four experiments.

Morphological Changes in Early Apoptotic Cells (Cell Rounding and AVD)

Next, we analyzed the timing of changes in cell morphology of HeLa cells in relation to the timing of MOMP. Cell behavior under the action of TRAIL, Act D and etoposide was similar (Fig. 4): flat cells rapidly rounded up and remained attached to the substrate via thin protrusions (filopodia). Rounded cells continued to move on the substrate for ~1–2 hr, then become immotile and remained poorly attached to the substrate for many hours until final swelling. Rounding of HeLa cells proceeded asynchronously in the population under the action of all drugs with the exception of STS, where all cells started shrinking within several minutes upon addition of the drug (Fig. 4); however, MOMP was asynchronous under STS treatment as well.

Another early event in apoptosis execution is the so-called apoptotic volume decrease (AVD), which occurs along with Na⁺ influx and K⁺ efflux (Bortner and Ciedlowski 2003). To determine the relationship between Na⁺ influx and K⁺ efflux on one hand and MOMP on the other hand at the single-cell level, we stained cells with the Na⁺-sensor Asante Green AM in parallel with TMRE. Snapshots taken during the early stage of apoptosis showed that only cells that had lost Δψ_m exhibited a significant increase in Asante Green staining (Fig. 3D, 3E). The majority of Asante Green-positive HeLa cells under TRAIL treatment were rounded; however, some of these cells remained flattened (Fig. 3E).

To characterize cell volume changes in greater detail, we utilized flow cytometry and microscopy of suspended Jurkat cells. Direct measurements of Jurkat cells show that the diameters of TMRE⁺ cells were similar under all four treatments (p>0.5 for each pair) and not significantly different from the control cells. However, TMRE⁺ Jurkat cells had a significantly larger diameter as compared with TMRE^- cells for three inhibitors (p<0.001 for TRAIL, n=126; p<0.001 for Act D, n=49 and p=0.027 for etoposide, n=48), but not for STS (p>0.2, n=51).

The volume decrease after MOMP was confirmed by imaging flow cytometry (Fig. 5A, Supplemental Table S2). The TMRE⁺ population always showed the highest forward scatter-width (FSC-W) and height (FSC-H) parameters and the TMRE^-/caspase^- population had the lowest width and height scores (Supplemental Table S2). Finally, FSC-W was used (Tzur et al. 2011) to analyze three cell populations that differed in levels of Δψ_m and caspase 3/7 activity (TMRE⁺; TMRE^-/caspase^low and caspase^high) (Fig. 5B). We thus concluded that, during MOMP, cells shrink and lose a significant amount of their internal content (Supplemental Table S3). Since all cells compared had an intact plasma membrane, the volume loss is likely explained by increased ion efflux (K⁺ and Cl^-) and/or leakage of small molecules (Bortner and Ciedlowski 2003; Kasim et al. 2013).

Figure 5.

Volume changes of apoptotic cells. (A) Cell area distribution of Jurkat cells (TRAIL+CHI/ 8.5 hr; Imagestream 100). From left to right: Gating for cells and cell area distribution; gating TMRE-positive cells (R4), TMRE-negative/caspase-negative cells (R5), caspase-positive/TMRE-negative cells (casp) and corresponding area cell distributions. (B) Forward scatter (FSC) and side scatter (SSC) distributions for Jurkat cells after etoposide treatment (6 hr). Upper row, left to right: selection of cells based on FSC parameters (P1 denotes single cells and excludes doublets and debris); FSC/SSC dot plot (P1 is shown in red); dot plots showing cell heterogeneity according to TMRE/caspase 3/7 and TMRE/Annexin V staining (colorized regions P2–P5 denote four different populations). Lower row, left to right: Histogram of FSC for all gated cells (P1); overlay of histograms of FSC for four populations (P2–P5); overlay of histograms of SSC for same populations (P2–P5) above. Statistics are the percentage of gated cells and their FSC and SSC mean values for each population. Intermediate population gated in P4 (TMRE^negative/caspase 3/7^negative) has the lowest FSC and SSC values.

Activation of Caspase 3/7 in Single Apoptotic Cells Is Gradual

In apoptotic cells, caspase 3/7 activity is remarkably different throughout the overall population (Fig. 6A, 6F, 6G), and the presence of an intermediate population (TMRE^-/caspase^low) is always evident (Supplemental Figs. S1–S3). However it should be stressed that under no treatment in all cell types examined intermediate population of TMRE⁺/caspase^high cells could be found (Supplemental Figs. S1–S3).

Figure 6.

Caspase 3/7 dynamics during apoptosis execution. (A) Caspase 3/7 activation in HeLa cells under staurosporine (STS; 1 μg/ml) treatment. Arrows point to two cells with a burst-like elevation in caspase 3/7. Arrowheads point to large blebs formed during the caspase 3/7 burst. TMRE, red; caspase 3/7, green. Overlay of fluorescence and DIC images. Time (min) is given in the upper left corner. Scale, 20 μm. (B) Caspase 3/7 substrate shows a wide dynamic range (dot plot of fluorescence intensity of caspase 3/7 substrate in the nucleus vs time (HeLa; STS/20 hr after caspase 3/7 activation). Sequential images below were set to the same brightness/contrast. Arrow points to the time of caspase burst (right figure; upper row). (C) Elevation of the average level of caspase 3/7 is slow and linear for approximately 2 hr. Dot plots show intensity averaged for 13–15 cells synchronized at the time of caspase 3/7 activation (set to 0 min) under each treatment. (D) Histograms show the timing distribution between mitochondrial outer membrane potential (MOMP) and activation of caspases 3/7. Red line is the approximation with gamma-function. Scale and shape parameter values are given in Supplemental Table S4. (E) Linear regression analysis of the delay between caspase 3/7 activation and MOMP. Correlation coefficients for all four treatments are insignificantly different from 0. (F) Imagestream 100 images: caspase 3/7 activity is heterogeneous in the population of Jurkat T-cells even during late apoptosis after all treatments (caspase 3/7, green channel vs Δψ_m, TMRE). (G) Image gallery of apoptotic cells with high and low caspase 3/7 activity (Ch3) after TRAIL (100 ng/ml; 34 hr) taken with Imagestream 100. Ch2, Bright Field; Ch6, DRAQ 5 staining of cell nuclei (nuclear fragmentation is evident in caspase^high and caspase^low subpopulations). A representative of two experiments.

Time-lapse microscopy confirmed that activation of caspase 3/7 in HeLa cells occurred only after cell rounding and MOMP and occurred asynchronously in total cell populations in response to all of the apoptosis triggers. Caspases 3/7 activity increased in individual cells slowly and usually linearly for several hours (Fig. 6B, 6C). After the initial accumulation, the highest activity of caspases 3/7 was always in the nucleus (Supplemental Fig. S3) where it exceeded cytoplasmic caspase 3/7 activity by several-fold (Fig. 7). In some cells, caspase 3/7 activity changed in a non-linear fashion: it rapidly increased throughout an individual cell along with formation of large blebs of up to 10–20 µm in diameter and then decreased (arrows point to the bursting HeLa cells in Fig. 6A and THP-1 cells in Supplemental Fig. S3A). The burst of caspase 3/7 activity was observed under all treatments yet with different frequency (data not shown).

Figure 7.

Dynamics of loss of TMRE (red) and caspase 3/7 (green) accumulation in HeLa cells under etoposide treatment. Arrows point to two cells where an accumulation of caspase occurs over at least 6 hr. Time (hr, min.) from the beginning of treatment is given in the upper left corner. Scale, 50 μm. Images were taken with Planfluor ×10/0.3 objective.

Apoptotic Temporal Heterogeneity Is Described by γ-Distribution

Activation of caspase 3/7 always occurred after MOMP, and with significant delay (Fig. 3D). The variability in the time gap between MOMP and caspase 3/7 activation (Fig. 5D) is described by the γ-distribution for all inhibitors studied:

$$\mathrm{t}=\mathrm{G}(\mathrm{\alpha },\mathrm{\theta });\mathrm{f}\left(\mathrm{t}\right)=({{\mathrm{t}}^{\mathrm{\alpha }-1}}_{*}{\mathrm{e}}^{-\mathrm{t}/\mathrm{\theta }})/({\mathrm{Г}\left(\mathrm{\alpha }\right)}_{\mathrm{*}}{\mathrm{\theta }}^{\mathrm{\alpha }})$$

Here, t is the experimentally observed time gap between MOMP and caspase 3/7 activation in an individual cell; f(t) denotes the probability distribution function of t; α, θ are the two parameters characterizing the experimentally observed distribution; and Γ(α) is a complete γ-function. In the case of STS and TRAIL treatments, shape parameter α ≈2; for Act D treatments, α ≈1.5 and etoposide treatment results in nearly an exponential distribution (i.e., gamma-distribution with α close to 1) (Supplemental Table S2). It means that, at least in the cases of STS, TRAIL and Act D treatments, more than one independent event is responsible for this transition.

PS Externalization Does Not Correlate with Caspase Activity

Probing for PS externalization is a widely used marker of apoptosis (van Engeland et al. 1998). To put PS externalization in line with other apoptotic events, we combined an Annexin V probe with TMRE staining of mitochondria and caspase 3/7 staining. After induction of apoptosis by any treatment, the majority—but not all—of TMRE-negative Jurkat, H9, THP-1 and K-562 cells were Annexin V-positive and nearly all (>95%) showed homogeneous staining of the plasma membrane with different intensity, along with staining of the nucleus by the caspase 3/7 substrate (Fig. 1B). At the same time, no TMRE-positive cell was positive for Annexin V in all of the cultures tested (Fig. 1A; Supplemental Figs. S2, S3).

Annexin V staining of HeLa cells treated with each apoptotic stimulus shows three patterns (Fig. 8): partial staining of cell protrusions (usually balbed at the tip); bright staining of the whole cell surface (rounded cell with small blebs, Supplemental Figs. S2, S3); faint staining of the large bubbles formed by inflating cells, with a high activity of caspases. The first pattern was observed mainly after STS treatment, whereas the second and third patterns were observed after all treatments. Only the third group of Annexin V-positive cells matched with caspase positivity, whereas the two first did not. The intensity of Annexin V staining did not correlate with caspase 3/7 activity (Fig.8, left column) and, even in late apoptosis, we did not find a correlation in the intensity between Annexin V and caspase 3/7 (Supplemental Fig. S4). We conclude that PS externalization in the whole cell always happens soon after MOMP and precedes caspase 3/7 activation.

Figure 8.

Phosphatidylserine (PS) externalization in apoptotic HeLa cells. Left column: Dynamics of the loss of TMRE (red), the accumulation of caspase 3/7 (green), and the externalization of PS (blue) in HeLa cells under staurosporine (STS) treatment. Time of treatment is shown in hours. Partial staining of cell protrusions with Annexin V is evident. Middle and right columns: Optical sections showing inflated cells with Annexin V-positive membrane (blue) after prolonged STS and TRAIL treatments (10 hr). Plasma membrane is homogeneously stained. Scale, 20 μm.

Pre-MOMP and post-MOMP Apoptotic Stages

To further characterize temporal apoptotic heterogeneity, we addressed the question of timing between early apoptotic morphological changes in cells (cell volume decrease and cell rounding), MOMP (using TMRE staining as a marker) and caspase 3/7 activation. The duration of pre-MOMP and MOMP-to-caspase stages did not correlate with each other for any of the drugs tested (Fig. 6E), suggesting that, at the single-cell level, these events are largely independent of one another.

Summarizing the time-lapse microscopy data, we conclude that the TMRE^-/caspase^low cellular population consists of cells that have recently undergone MOMP, whereas the caspase^high population consists of cells at a late stage of apoptosis. Interestingly, caspase-positive cells had a larger cell area than TMRE^-/caspase^low cells (Supplemental Table S3), which is in accord with AVD kinetics described for Ehrlich ascites tumor cells (Poulsen et al. 2010) and our observations on cell inflation at the late stages of apoptosis (Fig. 2B).

Thus, true cell shrinkage during apoptosis is a rapid process that happens in parallel with MOMP, and subsequent volume changes in shrunken cells reflect processes associated with the late stages of apoptosis, such as DNA fragmentation and caspase activity (Bortner and Ciedlowski 2011).

Apoptosis Reversibility

Finally, we addressed the question of apoptosis reversibility. We previously showed that TMRE staining allows discrimination of live and apoptotic/dead cells (Barteneva et al. 2014).To dissect whether cells undergoing apoptosis remain able to recover, we treated Jurkat and HeLa cells with apoptotic stimuli, sorted different subpopulations, and followed them afterwards. The TMRE^high/ caspase^negative subpopulation recovered at a relatively high rate, whereas the two other subpopulations (TMRE^-/caspase^low and TMRE^intermed/caspase^intermed) exhibited reduced recovery capacity (Fig. 9). The caspase^high subpopulation was not able to recover at all. Thus, these data indicate that a major point of irreversibility in the apoptotic pathway is marked by the loss of Δψ_m; however, complete irreversibility is achieved only when the activity of executive caspases reaches a certain threshold.

Figure 9.

Cell changes and recovery capabilities of apoptotic cells. (A–C) Results of cell sorting (after treatment with staurosporine (STS) 1.0 μg/ml). (A) Sorting gates. (B) Results of Jurkat cells sorting (left column, caspase 3/7 vs TMRE; right column, Sytox Blue (SB) vs TMRE). (C) Results of HeLa cells sorting (microscopy, TMRE^positive cells). Scale, 50 μm. Results are representative of two experiments. (D) Model summarizing the dynamics of apoptotic progression and cell recovery.

Discussion

In summary, we have monitored cellular events at the single-cell and population levels over the course of apoptosis. We determined caspase 3/7 activity by measuring the accumulation of the fluorescence product; this is a significant improvement over previous fluorescence FRET-based sensors (Rhem et al. 2002; Albeck et al. 2008; Wolf et al. 1999) because of its wider dynamic range [>1:100 vs 1:4 for FRET-sensors (Luo et al. 2003)] and its utility for quantifying the gradual accumulation of caspase 3/7 activity, as confirmed by fading at the end of the caspase burst. Our findings with respect to this gradual caspase 3/7 activation are in accordance with caspase 3/7 dynamics and the dynamics of its substrates (cleaved PARP, for example) measured in the overall population by immunoblotting or similar integrative approaches (Luo et al. 2003; Fazal et al. 2005) that are not applicable to single-cell analysis.

We have, for the first time, directly compared Na⁺ influx, cell volume decrease, and mitochondrial depolarization, and have shown that these events are tightly coordinated with MOMP. Opening of Na⁺ channels happens at least partly in parallel with MOMP and the loss of cell volume. Our data are in agreement with Dezaki et al. (2012), who reported that AVD occurs before caspase-3 activation.

Our findings indicate that the timing of pre-MOMP and post-MOMP events provides a sufficient window to explain the variability in phenotypic response to apoptotic stimuli. The MOMP-to-caspase 3/7 transition is described by γ-distribution. We found that, for the STS and TRAIL treatments, shape parameter is α≈2, suggesting that at least two independent events happen during this transition. Based on the conventional mitochondria-dependent pathway (Kroemer et al. 1995; Green 2005; Tait and Green 2013), we suggest these events are (1) the formation of the APAF1 complex after the release of cytochrome c: it requires assembly of several subunits; and (2) the activation of caspase 9: it could be as slow as the activation of caspase 3. For Act D, α ≈1.5 and for etoposide, an exponential distribution gives a better approximation (i.e., γ-distribution with α close to 1) (Supplemental Table S2). The relatively fast activation of caspase 3/7 in the post-MOMP stage after etoposide treatment could be explained by plural effects of this drug. Etoposide not only activates the mitochondrial (intrinsic) pathway, even in the absence of p53 and Bax (Karpinich et al. 2002; 2006), but also activates caspase 8 in CD95-independent pathway (Boesen-de Cock et al. 1998; Day et al. 2009). It is important to note that activation of caspase 3/7 always occurs after MOMP, even in the case of the apoptosis type I pathway (in our study, it was in H9 cells treated with TRAIL). However, the average delay between two events depends on the stimulus and on the cell type. Thus, we suggest that, even in the type I pathway, mitochondria are also involved in the apoptosis execution cascade and previous reports on the possible mitochondria-independent activation of executive caspases (reviewed in McIlwain et al. 2012) did not take into account the inhibitory effect of mitochondria on caspase 8 precluding the activation of caspases 3/7 (Spencer and Sorger 2011).

Our data confirm that, during apoptosis, PS externalization is mediated in two ways. Bright staining of TMRE^positive cells (Fig. 7) could be explained by the Ca2⁺-dependent activation of scramblases coupled to the down-regulation of flippase activity (Daleke and Lyles 2000), which is independent of MOMP (Balasubramanian et al. 2007). Staining for the active caspases results from the irreversible cleavage of flippases of PS (Suzuki et al. 2013; Segawa et al. 2014) and is ubiquitous. We suggest that several mechanisms are involved in PS externalization: first (reversible) PS externalization might be not related to MOMP and caspase activation, whereas, later on, it becomes irreversible due to caspase 3/7 activation.

AVD in our experiments was an early process that occurred in parallel to MOMP and Na⁺ influx, which can be explained by the fact that volume loss is an energy-dependent process (Okada et al. 2004) and is driven by actomyosin contraction generated by changes in osmotic pressure (Stewart et al. 2011); this is similar to the cell rounding process during mitosis. Our findings concur with those reported by Yang et al. (2011), who validated Na⁺ inhibitors as anti-apoptotic agents.

We found that only cells with intact mitochondria could recover after a washout of the apoptotic stimulus. Although MOMP was often described as a “point-of-no-return” in apoptosis execution (Goldstein et al. 2000), more recent work has demonstrated that MOMP can be incomplete, leaving some mitochondria intact (Tait and Green 2010); thus, the likelihood of cell survival after MOMP remains to be elucidated. We have found that, at the population level, MOMP is generally associated with irreversible apoptotic induction; however, in individual cells, apoptosis becomes absolutely irreversible only after the activation of executive caspases, as a small fraction of cells from intermediate populations are able to recover (Fig. 7D, 7E). Thus, apoptotic recovery is highly probable for cells prior to MOMP, it is still possible for cells after MOMP, and becomes impossible only once the executive caspase activity passes a certain threshold.

We propose a model for apoptosis execution, which encompasses the following stages (Fig. 9 D): 1) the pre-MOMP stage, which is variable and largely dependent on the stimulus (for adherent cells, this stage includes cell rounding); 2) the stage that follows apoptotic insult, which includes (i) MOMP, (ii) a rapid loss of mitochondrial membrane potential (although this might be incomplete), (iii) a loss of cell volume, and (iv) a change in the Na⁺/K⁺ balance; 3) the stage that involves post-MOMP events, which includes (i) rapid externalization of PS, (ii) gradual activation of executive caspases, and (iii) a secondary increase in cell volume (inflation).

Single-cell imaging and biochemical reaction modeling of apoptosis have provided a backbone for the analysis of stochastic variability of apoptotic cellular response (Bhola and Simon 2009; Rhem et al. 2009; Spencer et al. 2009). Apoptosis should be described as including two types of processes. (i) The first type involve fast processes, determined by the opening of specific channels—in the internal and outer mitochondrial membranes, and in the plasma membrane (efflux of potassium and influx of sodium; dynamic range: sec and min); besides, other channels are responsible for cell rounding (attached cells), for cell shrinkage (AVD), and for nuclear shrinkage (to be determined). (ii) The second is relatively slow enzymatic processes (activation of executive caspases; dynamic range: hours). The temporal heterogeneity of apoptotic processes in single cells that arises from the combination of fast and slow processes and apoptosis morphological steps (for example, cell shrinking and rounding, and nuclei fragmentation) cannot be readily explained or modeled by fluctuations in protein concentrations.

We conclude that adding quantitative morphological parameters to experimental protocols aimed at defining the chain of early events involved in apoptosis execution will provide a powerful means of dissecting and modeling apoptotic mechanisms in cancer biology and in the wider context of other diseases and developmental processes.