Single-Cell and Population-Level Analyses Using Real-Time Kinetic Labeling Couples Proliferation and Cell Death Mechanisms

SUMMARY

Quantifying cytostatic and cytotoxic outcomes are integral components of characterizing perturbagens used as research tools and in drug discovery pipelines. Furthermore, data-rich acquisition, coupled with robust methods for analysis, is required to properly assess the function and impact of these perturbagens. Here, we present a detailed and versatile method for single-cell and population-level analyses using real-time kinetic labeling (SPARKL). SPARKL integrates high-content live-cell imaging with automated detection and analysis of fluorescent reporters of cell death. We outline several examples of zero-handling, non-disruptive protocols for detailing cell death mechanisms and proliferation profiles. Additionally, we suggest several methods for mathematically analyzing these data to best utilize the collected kinetic data. Compared to traditional methods of detection and analysis, SPARKL is more sensitive, accurate, and high throughput while substantially eliminating sample processing and providing richer data.

Graphical Abstract

In Brief

To quantify cell death in high-throughput studies, Gelles et al. develop a robust method for single-cell and population-level analyses using real-time kinetic labeling (SPARKL). Example protocols and mathematical analyses detail the characterization of cell death kinetics and mechanisms, with coupled changes to proliferation, for use within high-volume comparative approaches.

INTRODUCTION

Programmed cell death pathways are conserved signaling mechanisms, which developed early in the evolution of metazoa (Oberst et al., 2008). One aspect shared between many programmed cell death pathways is a variable lag phase between exposure to a perturbagen and the commitment to a cell death program. This lag phase is the consequence of intersecting intracellular pro-death and pro-survival signal transduction and provides a cell with an opportunity to resolve the stress signal and repair accumulated damage (Biton and Ashkenazi, 2011). If these damages are not resolved, the pro-death signaling contributions will overwhelm the pro-survival reserve and trigger biological events committing the cell to death. Importantly, in apoptosis, this lag phase also contains an orchestrated and systematic dissolution of organelles and cellular components conducive to efficient clearance with minimal perturbation to neighboring cells. This process is exemplified in apoptosis by the BCL-2 family of proteins, consisting of pro-apoptotic effector proteins (e.g., BCL-2-associated X protein [BAX] and BCL-2 homologous antagonist killer [BAK]) and anti-apoptotic proteins (e.g., BCL-2 and B cell lymphoma-extra large [BCL-xL]), which ultimately serve to regulate the permeabilization of the outer mitochondrial membrane and subsequent activation of the caspase cascade (Wei et al., 2001; Chipuk et al., 2010). However, the kinetics and perpetuation of cell death signaling is highly variable between perturbagens, cell types, death pathways, and between sister cells within a population (Spencer et al., 2009; Gaudet et al., 2012). Elucidating the underlying biology that causes this variability remains a principle focus within the fields of cell death, cell biology, disease etiology, and drug discovery (Kepp et al., 2011). To this end, development of technologies to properly observe and analyze cell death is crucial to progress these fields.

Current standard methods to observe and quantify cell death remain outdated, suffer from limited throughput, and generate minimal datasets for interpretation. The detection and quantification of dead or dying cells is most commonly accomplished by flow cytometry, which requires non-trivial cell numbers, extensive sample handling, sample exposure to significant mechanical and chemical stress, and considerable delays between sample harvesting and analyses (Koopman et al., 1994). For example, experiments must be terminated in order to be analyzed and therefore only provide static endpoint data, requiring considerable effort to optimize the experimental design.

Commonly used reagents involve cell-impermeable “viability” dyes (such as propidium iodide [PI], DRAQ7, SYTOX, and YOYO3 [Y3]), which label cells following loss of plasma membrane integrity or permeabilization. Reliance on this feature for quantification does not distinguish between pathways and labels cells at the tail end of the dismantling process, thereby failing to capture the time in which cells undergo key biological processes (Vanden Berghe et al., 2010; Dillon et al., 2014). Additionally, labeling with viability dyes is not stoichiometric and often results in pseudo-binary labeling profiles following the first instance of membrane instability. Enzymatically cleaved fluorescently conjugated probes (e.g., DEVD-containing caspase-target peptides) are another common strategy despite their cost, difficulty of use, and non-specific activation (Yu et al., 2001; McStay et al., 2008; Onufriev et al., 2009). Alternative methodologies use metabolic activity or biochemical measures as surrogate readouts for cell viability, but interpretations from this data are obfuscated by the underlying biology of the perturbagens and ultimately do not directly rely on cell death machinery (Chan et al., 2013).

Design

Here, we integrate and advance multiple previously described methods for observing and quantifying cell proliferation and cell death using live-cell high-content imagers and refer to this workflow as single-cell and population-level analyses using real-time kinetic labeling (SPARKL). We utilize fluorescently labeled annexin V (AV) due to stability, signal strength and longevity, stoichiometric binding, and relevance to cell death machinery. Additionally, we test and verify several fluorescent labeling reagents compatible with long-term incubation and time-course studies as well as providing examples of how they can be used to investigate aspects of cell death kinetics, which were not feasible with analogous workflows. Our non-toxic label-and-go methods outperform previous methodologies while maintaining high sensitivity, high accuracy, and a high-throughput zero-handling rapid protocol. Furthermore, we expand on the versatility of our method by providing several example adaptations and mathematical analyses to explore the depth of the collected kinetic data. Collectively, we will demonstrate how data gathered and analyzed using SPARKL can be used to investigate fundamental aspects of cell death biology by applying these workflows to models of death receptor (DR)-mediated apoptosis, the mitochondrial pathway of apoptosis, ferroptosis, and necroptosis.

RESULTS AND DISCUSSION

Automated Live-Cell Imaging and Detection of Fluorescent Labeling Provides Rich Datasets for Analyzing, Characterizing, and Identifying Cell Death in Response to Perturbagens

We developed SPARKL chiefly to capture kinetic cell death data since traditional methods utilizing flow cytometry are limited to endpoint data. Additionally, we aimed to design a workflow compatible with high-throughput studies while requiring minimal handling and retaining high sensitivity and specificity. Here, we capitalize on the advent of high-content in-incubator live-cell imagers to capture cellular phenotypes, measure fluorescent signals, and analyze data in real time through regular scanning schedules. This technology eliminates the requirement for disruptive handling (e.g., trypsinization, pipet-induced shear forces, centrifugation, and incubation in labeling buffers) and is non-invasive to cells in culture. The repeated observation of samples and coupled automated real-time analysis also collects richer datasets in a workflow that is less labor and time intensive than analogous workflows using flow cytometry (Figure 1A).

Figure 1. SPARKL Provides Data-Rich Kinetic Analyses of Cell Death in Real Time Using Live-Cell Imaging.

(A) Comparison of workflows for cell death detection assays using traditional flow cytometry methods or SPARKL live-cell imaging.

(B) MEFs were co-treated with TNFα + CHX to induce cell death, incubated with titrations of AV-FITC (67, 100, 133, 200, 400, 1,000, 2,000, and 4,000 ng/mL) and Y3 (200, 250, 333, 400, 500, 667, 1,000, and 2,000 nM), and subjected to the SPARKL protocol.

(C) MEFs cultured in DMEM lacking or containing phenol red were treated as in B. Mean integrated fluorescence intensity for positive events are depicted as a heatmap.

(D) WT and DKO MEFs were treated as indicated and incubated with AV-FITC and Y3. Scale bar denotes 100 μm. Images were white-point corrected as detailed in STAR Methods.

(E) Kinetic data analyzed from images stacks corresponding to cells from (D). Dissimilar axis range reflects difference in cell number and is visualized based on internal controls.

(F) Comparative endpoint data extracted from (E) at the indicated time points.

Treatments: CHX (50 μg/mL), erastin (10 μM), TNFα (10 ng/mL), and VP16 (25 μM). Error bars denote SD.

One significant benefit of this method is that cells are incubated in normal growth media containing fluorescent probes thereby eliminating stepwise processing. Therefore, detection reagents must be non-disruptive, highly specific, culture-stable, non-labile, and readily detectable when incubated with cells for prolonged periods of time. Most fluorescent probes used in flow cytometry are not compatible in kinetic assays due to photobleaching or cytotoxic effects with prolonged exposure. Therefore, we sought to find reagents that would have no negative effect when incubated with cells over time. AV labels apoptotic cells by detecting phosphatidylserine (PS) on the outer leaflet of the plasma membrane, which is exposed during caspase-dependent cell death (Devaux, 1991). We began these studies by testing fluorescein isothiocyanate (FITC)-labeled recombinant AV and cell-impermeable viability dye Y3 for compatibility within our method (Logue et al., 2009; Gelles and Chipuk, 2016). Mouse embryonic fibroblasts (MEFs) were cultured in growth media containing AV and Y3, co-treated with tumor necrosis factor α (TNFα) and cycloheximide (CHX) to promote apoptosis, and subjected to SPARKL using a high-content in-incubator live-cell imager. The kinetic data captured the lag phase followed by the robust labeling of MEFs undergoing death (Figure 1B). Here, cells were cultured in media containing sufficient calcium for robust AV labeling, but alternative media formulations may require calcium supplementation (Meers and Mealy, 1993). Importantly, cells not treated with TNFα + CHX exhibited no toxicity as a consequence of either fluorescent probe at the highest concentrations (Figure 1B, “UT” line). As expected, the stoichiometric method of AV labeling was observed as the concentration-dependent decrease in number of positive events. Said another way, low concentrations of AV will bind to cells exposing PS but may not generate sufficient signal (as defined by the intensity, size, and segmentation of the fluorescent event) according to the user-defined processing definition (see STAR Methods). However, both AV and Y3 were capable of strong and sustained labeling in live-cell imagers at concentrations approximately 10-fold below what is used in flow cytometric assays (Gelles and Chipuk, 2016). Detected fluorescent intensity was greater for MEFs treated in DMEM lacking phenol red than similarly treated MEFs in phenol red containing media (Figure 1C). Therefore, all future applications of SPARKL utilized phenol red free culture media.

Collection of high-content images of cells in culture is the foundation of SPARKL, which capture cellular morphology and fluorescent signal. To demonstrate the diversity and specificity of data collected with SPARKL, Bax^+/+Bak^+/+ and Bax^−/−Bak^−/− (wild type [WT] and double knockout [DKO], respectively), MEFs were cultured in media containing AV-FITC and Y3, treated with perturbagens, and subjected to the SPARKL workflow. Images from select time points were manually reviewed to observe cellular morphology and detection of fluorescent labeling (Figure 1D). Green and red pixels from each collected image were quantified in real time using a trained algorithm and reported as the number of AV⁺ and Y3⁺ events observed at each time point (Figure 1E). To assess the selectivity of our fluorescent probes, MEFs were instigated to die through the extrinsic pathway of apoptosis (via co-treatment of TNFα + CHX), the intrinsic pathway of apoptosis (via topoisomerase inhibitor, VP16), or non-apoptotic pathway of ferroptosis (via cysteine-glutamate antiport inhibitor, erastin) (Peltzer et al., 2016; Dixon et al., 2012). Both AV and Y3 demonstrated accuracy and selectivity by labeling cells in a time-dependent manner. Interestingly, VP16-treated MEFs exhibited the longest lag phase before onset of death and demonstrated AV positivity more rapidly than Y3 positivity, reflecting the biology of PS exposure prior to loss of plasma membrane integrity during apoptosis. To qualitatively compare AV positivity between WT and DKO MEFs, data collected at specific time points were extracted and visualized side by side (Figure 1F). It is worth noting that WT and DKO MEFs proliferate at different rates and therefore result in vastly different populations over the course of a short experiment (exampled in Figure 1D). As such, the y axes for the two cell lines are not congruent but are graphed appropriately based on internal control conditions (see STAR Methods); we explore methods for appropriate data comparisons and normalization in later sections. An important aspect of this methodology is that data are collected via cell imaging and can be backtracked to the original visualization of the cells in culture. As an example, collection and visualization of these data from DKO MEFs treated with VP16 reveals that the morphological phenotypes observed in these cells does not coincide with probe labeling and therefore is not indicative of cell death (Figures 1D–1F). In this way, we have demonstrated the depth of data collected using SPARKL and how it can be visualized for thorough interpretation of experimental results.

Investigations of apoptotic cell death commonly employ AV-binding assays due to the exposure of PS on the outer leaflet of the plasma membrane following caspase-mediated cleavage of scramblases and flippases (Suzuki et al., 2013; Segawa et al., 2014). Exposure of PS is indicative of the underlying apoptotic signaling and precedes loss of plasma membrane integrity caused by breakdown of cellular homeostatic processes. Therefore, apoptotic studies occasionally utilize a dual-labeling methodology of AV with a cell-impermeable viability dye (such as PI or DRAQ7) in order to characterize cells in early or late apoptosis by dichotomizing single-positive and dual-positive events (Jiang et al., 2016). However, cells undergoing non-apoptotic cell death often do not exhibit this sequential labeling because loss of plasma membrane integrity occurs without the preceding exposure of PS. As a consequence, cells label with the viability dye and AV simultaneously, which is indistinguishable from cells in late apoptosis when analyzed by flow cytometry (Figure 2A). Therefore, we investigated whether the kinetic data of SPARKL could distinguish between apoptotic and non-apoptotic cell death using a similar dual-labeling approach.

Figure 2. SPARKL Is Versatile and Capable at Quantifying and Interrogating Different Pathways of Cell Death.

(A) Cells dying by apoptotic or non-apoptotic mechanisms exhibit differential labeling by AV or Y3, which is not clearly differentiated by flow cytometric methodologies.

(B) WT MEFs were incubated with AV and Y3, treated as indicated, and subjected to SPARKL.

(C) DKO MEFs were treated and incubated as in B.

(D) Cyld^−/− MEFs reconstituted with either glutathione S-transferase (GST) or CYLD were incubated with AV and Y3, treated as indicated, and subjected to SPARKL.

(E) Mlkl^+/+ AEFs were incubated with AV and Y3, treated as indicated, and subjected to SPARKL.

(F) Representatives images collected during SPARKL from Mlkl^+/+ AEFs incubated with AV and Y3 and treated as indicated. Scale bar denotes 100 μm and insets are a conserved magnification.

(G) Mlkl^+/+ or Mlkl^−/− AEFs were incubated with AV and Y3, treated as indicated, and subjected to SPARKL.

(H) WT and Mlkl KO AEFs were cocultured in a range of ratios and treated as in (G).

Treatments: ABT-737, 1 μM; Brp, 5 μM; CHX, 50 μg/mL; erastin, 10 μM; Nec1, 10 μM; TNFα, 20 or 50 ng/mL; and VP16, 25 μM; zVAD, 40 μM.

To assess the capability of SPARKL to dichotomize kinetic labeling patterns, we cultured WT MEFs in growth media containing AV-FITC, Y3, and perturbagens engaging either the extrinsic pathway of apoptosis (TNFα + CHX), intrinsic pathway of apoptosis (CHX or VP16), or non-apoptotic ferroptosis (erastin) (Figure 2B). Cell populations instigated to die by TNFα + CHX or VP16 exhibited AV positivity prior to Y3 positivity, indicative of apoptotic cell death. By contrast, cells instigated to die by high concentrations of CHX alone or erastin demonstrated similar kinetics of labeling with AV and Y3. Additionally, we co-treated cells with ABT-737 (an inhibitor to multiple anti-apoptotic BCL-2 proteins) and/or zVAD-fmk (a pan-caspase inhibitor) to further characterize mechanisms of cell death in response to each perturbagen using the SPARKL assay (Oltersdorf et al., 2005; Thornberry et al., 1992). Cells dying via apoptosis demonstrated accelerated or attenuated rates of AV labeling in response to co-treatment with ABT or zVAD, respectively; cells dying via ferroptosis demonstrated no change in kinetics of AV labeling. By comparison, kinetics of Y3 labeling exhibited less resolution and only modest changes in cells co-treated with ABT or zVAD.

DR-mediated apoptosis can occur either through a mitochondrial-independent or mitochondrial-dependent pathway, termed type I and type II respectively, and the latter signals through members of the BCL-2 family of proteins (Jost et al., 2009). Kinetic data collected by the SPARKL workflow clearly revealed that WT MEFs treated with TNFα + CHX signal primarily via the type II pathway, as evidenced by the synergic labeling of AV in cells co-treated with ABT. To further interrogate the mechanism of cell death engaged by each perturbagen, we replicated this treatment strategy in DKO MEFs (Figure 2C). DKO MEFs treated with VP16 demonstrated minimal labeling due to the requirement of BAX and BAK for mitochondrial outer membrane permeabilization in the intrinsic pathway of apoptosis. By contrast, DKO MEFs treated with erastin exhibited similar kinetics of death compared to WT MEFs. DKO MEFs treated with TNFα + CHX died with similar kinetics as their WT counterparts, consistent with DR signaling, which is independent of BAX and BAK. Furthermore, DKO MEFs exhibited similar kinetics for both AV and Y3. Interestingly, DKO MEFs treated with CHX alone exhibited death similar to analogously treated WT MEFs but did not synergize with ABT, suggesting that CHX treatment engaged multiple pathways of cell death that were selectively amplified with particular co-treatments or in different model systems.

We have generalized programmed cell death pathways as either apoptotic or non-apoptotic, the latter of which has been modeled by ferroptosis. However, DR signaling can initiate non-apoptotic cell death programs, such as necroptosis, when caspases are inhibited (Vercammen et al., 1998; Kawahara et al., 1998). WT MEFs treated with TNFα + CHX demonstrated attenuated cell death when co-treated with zVAD, suggestive of apoptosis, but DKO MEFs exhibited a modest increase, suggestive of necroptosis (Figures 2B and 2C). When caspases are inhibited, the switch from DR-mediated apoptosis to necroptosis is initiated by an interaction between the receptor-interacting serine-threonine-protein kinase-1 (RIPK1) and the receptor-interacting serine-threonine-protein kinase-3 (RIPK3), which is regulated by the ubiquitylation profile of RIPK1, resulting in RIPK3-mediated phosphorylation of MLKL, which consequentially oligomerizes and permeabilizes the plasma membrane (Sun et al., 2012; Weinlich et al., 2017). It has been reported that simian vacuolating virus 40 (SV40)-transformed MEFs do not express RIPK3 sufficiently to induce a strong necroptotic program and so we expanded our investigations to different established cellular models (Moujalled et al., 2013). First, we utilized MEFs deficient in cylindromatosis (CYLD), a RIPK1 deubiquitinase, which were reconstituted with either CYLD or a vehicle control (O’Donnell et al., 2011). Cyld^−/− MEFs treated with TNFα + CHX exhibited cell death but only the MEFs reconstituted with CYLD demonstrated increased or decreased death following co-treatment with zVAD or the RIPK1 inhibitor, Necrostatin-1 (Nec1), respectively (Degterev et al., 2005) (Figure 2D). Additionally, MEFs dying by necroptosis exhibited similar labeling kinetics of AV and Y3, which parallels the labeling phenotype we observed in MEFs dying by ferroptosis.

CHX co-treatment reveals pro-death pathways by inhibiting translation of pro-survival genes downstream of TNFα signaling. To better study labeling kinetics of cells dying by necroptosis, we utilized the second mitochondria-derived activator of caspases (SMAC) mimetic, birinapant (Brp), to inhibit RIPK1 ubiquitylation by cIAP1/2 and subsequent pro-survival signal cascade (Holler et al., 2000). As a consequence of inhibiting RIPK1 ubiquitylation, the requirement of CYLD to initiate necroptosis was greatly reduced, and therefore, we moved to an MLKL-dependent model (data not shown). Primary Mlkl^+/+ adult ear fibroblasts (WT AEFs) were cultured from mice and treated with TNFα + Brp with or without co-treatment of zVAD or Nec1 (TB, TBZ, and TBN, respectively). While AV labeling kinetics appeared unchanged in TBZ-treated WT AEFs, Y3 demonstrated a dramatic and rapid labeling profile specific to a necroptotic outcome (Figure 2E). Necroptosis is defined by the regulated permeabilization of the plasma membrane by oligomeric MLKL and therefore the rapid labeling by a viability dye is consistent with this phenotype (Wang et al., 2014). However, the unchanged labeling kinetics of AV was unexpected, particularly for cells that became necroptotic hours prior. Images collected during SPARKL revealed a distinct cellular morphology in TBZ-treated WT AEFs, which did not contract, retained a topologically distinct and normal nucleus, and slowly exposed PS resulting in a weak AV signal over the large area of the cell (Figure 2F, upper panel). Curiously, a subpopulation of the cells exhibited the same weak AV signal but demonstrated no Y3 labeling (Figure 2F, upper inset panel). It has been reported that necroptotic cells are capable of exposing PS following MLKL activation, but prior to complete loss of plasma membrane integrity, which could explain these observations (Gong et al., 2017a). TBZN-treated WT AEFs resembled an apoptotic phenotype characterized by cellular contraction and formation of apoptotic bodies (Figure 2F, bottom panel). We repeated the experiment using Mlkl^−/− primary adult ear fibroblasts (KO AEFs) and observed that these cells were both less sensitive to TNFα + Brp and also completely protected by addition of zVAD, suggestive of apoptotic cell death (Figure 2G). We investigated if SPARKL could detect population heterogeneity by coculturing WT and KO AEFs. As the ratio of WT to KO AEFs decreased, we observed a concomitant decrease or absence of TNFα + Brp sensitivity and necroptotic labeling signatures (Figure 2H).

It is worth noting that minimal cell death was observed in TBN-treated WT AEFs (Figure 2E). Paradoxically, TBZN-treated WT AEFs resembled the TNFα + Brp-only condition and exhibited apoptosis-like morphology (Figures 2F–2H). Cells unable to undergo necroptosis may eventually resolve the zVAD-induced caspase inhibition and revert to an apoptotic program. Alternatively, they may be undergoing a necroptotic program that either does not require, or is revealed by the inhibition of, RIPK1 (Lin et al., 2016; Newton et al., 2016). The KO AEFs do not parallel these observations and while this could suggest that death in response to TBZN is MLKL dependent, data collected with SPARKL revealed that TBZN-induced cell death does not resemble necroptosis by either labeling kinetics or morphology. Nec1 binds the kinase domain of RIPK1, which is responsible for propagating the pro-death signal cascade, and has been shown to selectively inhibit necroptosis in cells induced via a DR and co-treatment with CHX ± zVAD (Degterev et al., 2008). However, both apoptosis and necroptosis are inhibited by Nec1 in cells which are induced via a co-treatment with a SMAC mimetic, due to requirement of RIPK1 for the formation of the caspase-8 activating complex, and we observed this differential effect of Nec1 in our studies as well (Wang et al., 2008) (Figures 2D and 2E).

We validated that the SPARKL workflow is not only high throughput and less labor intensive but also versatile and capable of observing and characterizing cell death following exposure to perturbagens. Specifically, we demonstrated that kinetic data gathered by the SPARKL workflow can reveal deeper insights regarding the mechanism of cell death when utilizing a dual-labeling approach. Cells dying by apoptosis sequentially label with AV and then with a viability dye while cells dying by ferroptosis exhibit similar kinetics for both labels. In stark contrast, necroptotic cell death exhibited the reverse phenotype in which cells robustly label with the viability dye and then slowly accumulated AV over time. It was recently reported that cells can exhibit AV labeling prior to viability dye labeling due to the formation of PS-containing “bubbles” on the plasma membrane following MLKL activation (Gong et al., 2017b). This phenotype was demonstrated to be rapid and transient and required highly sensitive technologies to detect. While we saw no evidence of this phenotype, we believe this is a consequence of the difference in signal detection sensitivity and time resolution between methodologies and therefore do not believe the results to be inherently contradictory. Furthermore, these differences exemplify the importance of utilizing multiple specialized techniques to further understand the molecular mechanisms of cell death.

Kinetic Data Gathered by SPARKL Provide Several Quantitative Metrics for Investigative and Comparative Analyses of Cell Death at Both Single-Cell and Population Level

Cell death assays using flow cytometry provide only endpoint data and are typically visualized as comparative bar graphs, signal quadrant plots, or histograms of fluorescent intensity. SPARKL is more data rich due to both the collection of kinetic data and underlying cell imaging methodology. However, while the utilization of real-time live-cell imaging is becoming more commonplace, most users do not fully explore the depth of the technology or the data it generates when analyzing their experiments. Furthermore, the advantages of these technologies are further minimized by use of endpoint data, simplified comparisons, and labeling methods repurposed from flow cytometry (Goodall et al., 2016). Therefore, we aimed to broaden the landscape of experimental designs performed with high-content imagers while interrogating how kinetic data could best be utilized to examine phenomena currently overlooked by analogous workflows.

In the previous section, we characterized the labeling profiles for cells instigated to die by several cell death pathways as observed by real-time live-cell imagers. Not only did the labeling order differ between pathways, but the kinetics and sensitivity of those labeling events were also observably different. To articulate the relative labeling kinetics for each pathway, we expressed kinetic data as the difference between AV- and Y3-positive events over time (Figure 3A). Representation of data in this manner clearly identifies the period when differential labeling is most observed and how this period is variable between treatments or pathways of cell death. Applying the same method to WT and KO AEF data revealed dissimilar sequential labeling in the TNFα + Brp treatment (Figure 3B). Thus, these data suggest that TNFα + Brp-induced cell death in KO AEFs was more apoptosis-like but signaled by a different mechanism in the WT AEFs. Inspecting the labeling data taken from a coculture of WT and KO AEFs clearly demonstrated how labeling phenotypes are more difficult to identify in heterogeneous populations. The cocultured AEFs retained similar labeling kinetics of AV but exhibited notably different Y3 labeling when compared to either homogeneous population (Figures 3C and 2G). Similarly, expressing data as the difference between relative AV- and Y3-positive events reveals a trend concomitant with the combination of the two cellular populations (Figure 3D).

Figure 3. A Dual-Reporter SPARKL Workflow for Characterizing Modes of Cell Death at Population and Single-Cell Level.

(A) Number of Y3⁺ events were subtracted from the number of AV⁺ events for cells analyzed in Figure 2B.

(B) Same as in A for cells analyzed in Figure 2G.

(C) WT and Mlkl KO AEFs were cocultured in a 1:1 ratio and treated as indicated.

(D) AV⁺ and Y3⁺ events for cells treated and analyzed in C were normalized as a percentage of total signal subtracted for each time point.

(E) Histogram data from cells treated and analyzed as in A–D shown as number of new AV (green) or Y3 (red) events occurring during each hour.

(F) WT MEFs and AEFs were treated as indicated. Panels show individual cell analysis where single positive is defined as AV⁺ and double positive is defined as AV⁺Y3⁺.

(G) Same data as in (F) reanalyzed where single positive is defined as Y3⁺ and double positive is defined as Y3⁺AV⁺.

(H) WT MEFs and AEFs treated and analyzed in (F). Population distribution of time-to-signal is shown for the following metrics: AV, Y3, double positive, Y3 following AV, and AV following Y3.

(I) WT MEFs and WT AEFs were treated and analyzed as in (F) and (G) without co-treatment of ABT or zVAD, respectively. Violin plots were compared with data from (H).

Green, red, and yellow plots are scaled to the left axis, while the transparent plots are scaled to the right axis. Solid line denotes the mean and dotted lines denote the quartiles. Significance calculated by Mann-Whitney U test and noted as follows: ****p < 0.0001, ***p < 0.001, **p < 0.01, *p < 0.05; ns, not significant.

Treatments: ABT-737, 1 μM; Brp, 5 μM; CHX, 50 μg/mL; erastin, 10 μM; Nec1, 10 μM; TNFα, 20 or 50 ng/mL; VP16, 25 μM; and zVAD, 40 μM.

Kinetic data collected by SPARKL represent populations of dying cells labeled with fluorescent probes, but each cell is detected as a binary process according to the specifics of automated detection (see STAR Methods). We utilized a coculture of AEFs as an exaggerated example to illustrate how data interpretation can be less clear in heterogeneous populations. However, heterogeneity in response to a perturbagen, and subsequent kinetics of cell death and labeling, is the culmination of many biological and stochastic factors that are not necessarily so overt in isogenic populations (Roux et al., 2015; Inde and Dixon, 2018). Therefore, we explored methods in which SPARKL could assess heterogeneity in a population dying in response to a perturbagen. Instead of viewing data as cumulative, kinetic data can be processed and expressed as the number of new events over time. Comparing histograms of events per hour revealed more rapid labeling of AV compared to Y3 in apoptotic models, overlapping histograms in ferroptosis, and stark Y3-first sequential labeling in cells undergoing necroptosis (Figure 3E). These data better illustrate labeling trends within the cell population and therefore can be analyzed for population statistics (such as reporting mean, median, and deviation of cell events within the population). Additionally, data expressed in this manner can reveal subpopulations that label differentially and therefore could reflect different biological mechanisms. For instance, we consistently observed “shelves,” which were most noticeable when the perturbagen resulted in rapid and robust cell death.

Live-cell imagers observe the same cells over time and we utilized this technological advantage for our dual-labeling workflow to quantify the time-to-signal for both fluorophores. WT MEFs and AEFs were instigated to die, analyzed by this method, and graphed as the time to single positive followed by time to double positive. Labeling kinetics were treatment specific, but the time in which AV-labeled cells sequentially labeled with Y3 was highly variable (Figure 3F). For example, 75% of MEFs treated with TNFα + CHX + ABT labeled with AV by 5 h and co-labeled on average 3.2 h later; the slowest 10% were variable. This analysis observed cells once they became AV⁺ and then assessed the time in which the cell became double positive. Cells that labeled simultaneously or in reverse order are visualized as a bar with no red phase, and this pattern appeared more often in non-apoptotic cell death. To corroborate this observation, the same data were reanalyzed to track the time to Y3⁺ and then double-positivity (Figure 3G). Few apoptotic cells exhibited AV labeling after Y3 (visualized by the lack of green phase in most bars) but ferroptotic cells exhibited approximately 50% of Y3-first sequential labeling. This analysis revealed a bifurcation within the population of treated AEFs: half of the population demonstrated a phenotype consistent with necroptosis (i.e., rapid onset of Y3 followed by variable AV labeling), while the other half exhibited a phenotype reminiscent of a mitochondrial pathway of apoptosis (i.e., lengthy and highly variable labeling time and Y3 labeling after AV). This phenotype could be a consequence of Brp-mediated inhibition within the cell or the association of oligomeric MLKL with membranes other than the plasma membrane, such as the mitochondria (Wang et al., 2014).

The labeling metrics of the population can be statistically compared to further characterize pathway- or treatment-specific patterns of cell death (Figure 3H). Instances of sequential labeling can be confirmed and quantified by comparing individual label kinetics to double-positive kinetics. For example, Y3 and double-positive kinetics of VP16 + ABT-treated MEFs are not significantly different, which indicated that double-positivity depended on the occurrence of Y3. Similarly, the reverse phenotype was observed in AEFs undergoing necroptosis. Ferroptotic MEFs exhibited significantly distinct labeling kinetics for AV, Y3, and double-positivity, which indicated no sequential labeling and likely reflects the biochemistry of the currently unknown method by which lipid peroxidation results in cell death (Feng and Stockwell, 2018). Additionally, this method provides a convenient strategy to assess synergistic treatments (Figure 3I). For example, MEFs induced to undergo apoptosis exhibited significantly different labeling kinetics in response to ABT while ferroptotic MEFs did not. Importantly, the pattern of distributions, while occurring earlier, was conserved and therefore reflected a treatment-mediated synergy of the death pathway. In stark contrast, the switch from apoptosis to necroptosis in AEFs exhibited significantly different labeling kinetics as well as different patterns of distribution. Collectively, these data demonstrate a method and rationale for use of high-content live-cell imagers to characterize and compare differential labeling kinetics at the single-cell resolution.

While there are instances to perform the deep analyses we described above, it does not represent the common usage of live-cell imagers for cell death analyses–namely high-volume comparative studies. To take full advantage of the high-throughput capacity of these technologies, complex kinetic data must be expressed as simplified key parameters for comparison. Here, we will review best practices for comparing SPARKL data as well as several mathematical methods for rapid parameterization of the collected data (Figure 4A). Instinctually, investigators often use the maximum number of events to compare cell death data and use language akin to “A had more or less death than B.” However, the amplitude of kinetic curves reflects extraneous factors such as cell number or the penetrance of a particular perturbagen. This is why graph axes are set by control treatments and do not need to be the same scale between cell types (such as data in Figure 1E). Additionally, using the maximal number of events (i.e., death) is highly reductionist and does not reflect metrics that are uniquely collected by real-time methodologies which are collectively referred to as the “kinetics” of cell death (e.g., times of onset, 50% death, plateau, and maximal rate of death). Therefore, we recommend drawing comparisons from cell death kinetics.

Figure 4. Mathematical Analysis of the SPARKL Kinetic Data Provides Several Parameters for High-Throughput Comparative Assays.

(A) Example methods for mathematically analyzing kinetic data and extracting parameters for cell death comparative studies such as AUC, derivatization, or non-linear fit analyses.

(B) WT and DKO MEFs were incubated with AV and Y3, treated as indicated, and subjected to SPARKL. AUC values were calculated from labeling kinetics.

(C) First derivative of data collected from cells treated as in (B) depicting the rates of population cell death at each time point.

(D) Comparison of maximum rates of death as calculated from first derivatives in (C).

(E) Fit curves using a lag one-phase growth function (solid line) applied to data from cells treated in (B) (dotted line). Curve for Y3 labeling kinetics in cells treated with VP16 and zVAD was considered an ambiguous fit and therefore excluded from parameter extraction and comparisons.

(F) Comparison of lag phase time (t’) from AV and Y3 data fits for cells treated as indicated from (E).

(G) Comparison of initial rate of death (R_D) from AV and Y3 data fits for cells treated as indicated from (E).

(H) 2D graphs of parameters calculated in (E) and graphed in (F) and (G) provide a convenient method for comparing treatment trends across labels and cell types.

Error bars denote SD and are not shown in instances where they are smaller than the data symbol.

Treatments: ABT-737, 1 μM; CHX, 50 μg/mL; erastin, 10 μM; TNFα, 10 ng/mL; VP16, 25 μM; and zVAD, 100 μM.

The simplest method is to calculate the area under the curve (AUC), which compresses kinetic data into a single parameter for convenient comparisons. WT and DKO MEFs were instigated to die using a variety of perturbagens, co-treated with ABT or zVAD, and detected by AV and Y3 labeling (Figure 4B). AUC values were greater in cells co-treated with ABT and reduced in cells co-treated with zVAD for perturbagens activating apoptosis but unchanged in cells undergoing ferroptosis. AUC calculated from Y3 detection exhibit similar trends in response to ABT and zVAD albeit to a lesser degree (Figure 4B). These examples retained the observed trends of AV and Y3 labeling within the raw kinetic data.

Other metrics of interest are rate of death and the time-to-maximal rate of death. The first derivative was calculated for data collected from MEFs treated with perturbagens and labeled with AV (Figure 4C). The derivative reports the rate of cell death within the population, or the slope of kinetic curves, at each time point. MEFs undergoing apoptosis demonstrated greater and earlier rates of cell death in response to ABT while exhibiting lower rates in response to zVAD (Figure 4C, yellow and purple data). Additionally, data can be presented as the time-to-maximal rate of death; however, this parameter is more informative for samples exhibiting significant death and labeling over the course of data collection (Figure 4D). These mathematical analyses represent simple but robust methods to quantitatively compare experimental conditions.

To quantify data in a manner that retains the richness of kinetic data, we employed a non-linear fit analysis to the SPARKL-collected data. A lag one-phase exponential (LOPE) function provided the best fit for our data and has been utilized successfully by other groups for modeling cell death kinetics (Forcina et al., 2017) (Figure 4E). The LOPE function calculates two useful parameters during the fit analysis: the duration of the lag phase (t′) and the rate of cell death following the lag phase (R_D). This model assumes that R_D is the maximal rate of cell death, which is not always accurate, but consistently fit data more accurately than other models (for example, sigmoidal functions). We do not suggest using the plateau’s amplitude (D_M) for comparative studies as it reflects cell number; however, we discuss data normalization in a later section. AV-labeled MEFs undergoing apoptosis demonstrated ABT-mediated reduction in t′ while remaining unchanged in ferroptosis (Figure 4F). R_D trends calculated from AV data were harder to interpret since the metric only differed in MEFs co-treated with CHX and ABT (Figure 4G). This may be an indication that perturbagen concentrations were excessive or that the time of maximal death significantly differs from the R_D calculated using LOPE. Interestingly, t′ from Y3 data did not reflect changes in response to ABT and R_D values increased (Figures 4F and 4G, red data). Data viewed as two-parameter plots reveal treatment- or cell-type-specific trends and has been shown to be effective for characterizing cell death mechanisms for perturbagens of unknown biology or multi-parametric drug panels (Forcina et al., 2017; Fallahi-Sichani et al., 2013) (Figure 4H). This method for comparative study does not appear to benefit from the dual-label approach, so we recommend using whichever label is more relevant to the experimental design.

The advent of high-content live-cell imagers provides investigators the opportunity to collect and analyze cellular data in a high-volume, high-throughput manner in real time. However, these technologies are not often used to their fullest potential and instead serve as a surrogate for flow cytometric workflows. To capitalize on the advantages of these technologies, we have provided several methods and best practices for utilizing and interpreting the data in comparative studies. We described a multiplex labeling workflow to demonstrate the variety of experimental designs compatible with these machines as well as several methods to parameterize and appropriately compare collected data. Collectively, these examples showcase the versatility, range of applications, breadth and depth of collected data, and analyses investigators can apply to experimental paradigms by using the SPARKL workflow.

Simultaneous Collection of Proliferation and Cell Death Data to Accurately Characterize Perturbagen-Induced Cytostatic and Cytotoxic Effects

The high-throughput capacity of SPARKL permits investigators to conduct exhaustive optimizations and drug titrations with relative ease. However, perturbagens at sub-lethal concentrations may still affect rates of cell proliferation, either directly or as a consequence while mitigating the underlying stress. Additionally, it has been shown that growth rate inhibition metrics can correct otherwise confounding data in large-scale drug screens (Hafner et al., 2016). Therefore, investigators require a method to properly assess the cytostatic effects of perturbagens on cells in culture. This is commonly accomplished by either measuring confluency or surface area metrics, which do not accurately reflect cell numbers, or by requiring the investigator to generate cell lines expressing fluorescent proteins, which is laborious, limiting, and not conducive to high-throughput analyses (Artymovich and Appledorn, 2015). Previously, we described a method utilizing non-toxic carboxyfluorescein succinimidyl ester (CFSE) to label cells prior to treatment and data acquisition (Gelles and Chipuk, 2016). This method required additional handling steps, suffered mild photobleaching, and was not compatible with certain cell types. Therefore, we sought to develop a label-and-go protocol of SPARKL that would simultaneously collect proliferation and cell death data in real time without requiring a separate cell-labeling step (Figure 5A).

Figure 5. Non-toxic Label-and-Go Reagents Used with SPARKL Can Normalize Data to Account for Changes in Cell Proliferation.

(A) Label-and-go method to collect proliferation data during AV-binding assay for composite cytostatic and cytotoxic investigations.

(B) WT MEFs were incubated with titrations of SYTO nucleotide labeling reagents (0.03–20 μM) in the presence of AV-594 and subjected to SPARKL.

(C) WT MEFs were treated as indicated and incubated with AV-594 and SYTO21. Representative images captured during SPARKL are shown for indicated time points. Scale bar denotes 100 μm.

(D) Kinetic data analyzed from images stacks corresponding to cells from (C).

(E) WT MEFs were seeded at the indicated numbers, incubated with AV and SYTO21, treated as indicated, and analyzed by SPARKL.

(F) Cell death data from E was normalized using SYTO21 to calculate percent death.

(G) Endpoint data of AV⁺ events from E and corresponding percent death normalization from (F).

Treatments: CHX, 50 μg/mL; TNFα, 10 ng/mL. Error bars denote SD.

We investigated if cell-permeable, low-affinity cyanine nucleic acid stains were suitable for prolonged incubation with cells. To assess cytostatic and cytotoxic effects, MEFs were incubated with varying concentrations of SYTO green fluorescent dyes and AV conjugated with AlexaFluor 594 (Figure 5B). Each tested SYTO dye exhibited toxicity at high concentrations and degrees of cytostatic behavior at low- to mid-range concentrations. Low-concentration ranges of SYTO16 and SYTO21 demonstrated no adverse effects to cell proliferation or viability, and we ultimately selected SYTO21 for use within SPARKL workflows. SYTO21 primarily labeled cells in the nucleus, was readily detected and segmented, and capable of labeling daughter cells post-mitosis (Figure 5C). Additionally, SYTO21 signal was retained in cells undergoing apoptosis and resulted in SYTO21⁺AV⁺ events. MEFs incubated with SYTO21 demonstrated an increase in SYTO21⁺ events over time, indicative of proliferation. By contrast, MEFs undergoing apoptosis via treatment of TNFα + CHX exhibit no change in the number of SYTO21-positive events and robustly labeled with AV-594 (Figure 5D). Therefore, SYTO21 is suitable as a fluorescent probe to track cell proliferation and population growth.

Live-cell imagers detect the number of fluorescent events and are therefore influenced by the number of cells. Data interpretation may be obfuscated when cell numbers vary between samples due to treatment-induced or cell-inherent differences in proliferation rates. Imperfect strategies to circumvent this issue undermine the advantages of kinetic monitoring (e.g., tandem analysis via flow cytometry, normalizing to endpoint data collected using DNA-intercalating dyes or inducers of cell death, using confluency as a surrogate for cell number) (Lopez et al., 2016; Giampazolias et al., 2017). To verify that SYTO21 could normalize data in real time, MEFs were seeded at several densities, incubated with SYTO21 and AV-594, and treated with either vehicle or TNFα + CHX. SYTO21 data accurately measured differences in cell number and cell proliferation in untreated MEFs. Concomitant AV⁺ data exemplified how cell number can complicate data interpretation (Figure 5E). For each time point, the AV⁺ events were normalized by the corresponding SYTO21⁺ events and expressed as percent death (Figure 5F). For comparison, AV⁺ endpoint data are shown pre- and post-normalization (Figure 5G). Furthermore, the kinetic curves exhibited similar slopes once normalized (Figure 5F, bottom panel). These data demonstrate that this method can normalize cell death data to account for differences in cell number.

Cell death data normalization is particularly relevant when perturbagens do not illicit a rapid death phenotype, which will result in more proliferation variability between experimental conditions and therefore different maximal labeling. To demonstrate this, we applied the normalization methodology to WT and DKO MEFs treated with a panel of perturbagens. AV data collected through SPARKL was normalized by the SYTO21 data and provided values that can be directly compared between the two cell lines (Figure 6A). Perturbagens engaging BAX- and BAK-mediated apoptosis demonstrated an attenuated lethal response in DKO MEFs (such as VP16 and calcium ionophore, A23187) and perturbagens capable of engaging the intrinsic pathway of apoptosis were synergized by co-treatment with ABT. When we replicated this experiment using Y3 as the cell death probe and normalized for SYTO21 labeled cells, we observed a significant loss of signal during longer experiments (Figure 6B). When analyzed well beyond induction of cell death, MEFs treated with TNFα + CHX exhibited sustained SYTO21 and AV signal but lost Y3 signal overtime (Figure 6C). The average fluorescent intensity of Y3⁺ events decreased and eventually became sub-threshold, which we believe to be indicative of changes to DNA content in apoptosis (Figure 6D). Signal loss did not appear to be due to cell detachment as this would affect detection of all probes, and the data do not demonstrate this concomitant loss of signal in either SYTO21 or AV over time.

Figure 6. Parameters Extracted from Non-linear Fit Analyses of Normalized Death Kinetics Generate Comparative Profiles for Studying Mechanisms of Cell Death.

(A) WT and DKO MEFs were incubated in DMEM containing SYTO21 and AV, treated as indicated, and subjected to SPARKL to calculate percentage of death at each time point.

(B) Cells, treatments, and analysis as in (A) for cells cultured in DMEM containing SYTO21 and Y3.

(C) WT MEFs treated as in (A) were subjected to long-term, repetitive scanning to assess signal permanence and longevity.

(D) Average integrated fluorescence intensity of events detected in C over the first 48 h.

(E) Fit curves using a LOPE function applied to data from WT cells in (A). Curve for DMSO-treated cells was considered an ambiguous fit and therefore its parameterization was non-informative.

(F) Comparison of lag phase time (t’) from (E).

(G) Comparison of initial rate of death (R_D) from (E).

(H) 2D graph of parameters calculated in (E) and graphed in (F) and (G) provides a convenient method for comparing trends across various cytotoxic treatments.

Error bars denote SD, are not shown for kinetic data to aid in data visualization, and not visible in instances where they are smaller than the data symbol.

Treatments: A23187, 5 μM; ABT-737, 1 μM; CHX, 50 μg/mL; erastin, 10 μM; TNFα, 10 ng/mL; TRAIL, 25 ng/mL; VP16, 25 μM; and zVAD, 100 μM.

Finally, we culminated our validation of SPARKL methodology by applying mathematical analyses to our normalized data. Percentages of cell death calculated by normalizing AV data to SYTO21 were fit using the LOPE function described in the previous section (Figure 6E). Samples co-treated with ABT reveal a reduction in t′, indicative of BCL-2 family contributions within the cell death pathway; by contrast, t′ is unchanged in response to ABT in MEFs undergoing ferroptosis via erastin (Figure 6F). Consistent with our previous methods of analysis, receptor-mediated apoptosis by either TNFα or TNF-related apoptosis-inducing ligand (TRAIL) demonstrated substantially different kinetics when co-treated with ABT and indicates a predominance for type II signaling in MEFs. When analyzing R_D, which became more informative in normalized data, we observed increased rates of death following co-treatment of ABT in apoptotic MEFs (Figure 6G); R_D was unchanged in MEFs treated with erastin or A23187. Cell death data from A23187-treated MEFs collectively indicates a mechanism requiring BAX or BAK but only partially sensitized by inhibition of anti-apoptotic BCL-2 proteins, which is a profile not shared among the other perturbagens exampled within this work (Figure 6H). Control-treated MEFs demonstrated ambiguous fit curves, and therefore parameter extraction was highly variable and not informative (Figures 6E–6G, red panel). Similarly, DKO MEFs not exhibiting death in response to particular perturbagens (such as VP16 and A23187) produced ambiguous fit curves and were not analyzed further (data not shown). These data exemplify the utility of mathematical parameterization on kinetic data that were normalized for differential cell numbers due to cell-inherent or treatment-induced changes to proliferation.

Conclusions

Here, we described the SPARKL workflow that is capable of simultaneously gathering proliferation and cell death data at single-cell and population-level resolutions. Using multiplex fluorescent detection of cell death, investigators can dichotomize cells engaging apoptotic or non-apoptotic signaling pathways in response to perturbagens. Additionally, we described a method to quantify cell numbers and proliferation using a non-disruptive fluorescent probe, which is compatible with “label- and-go” workflows for high-throughput studies. Collection of these rich datasets provides comprehensive insights into the cytostatic effects of perturbagens as well as their cytotoxic potential. Furthermore, we described and advanced mathematical analyses that parameterize data for convenient comparative analyses and interpretations of underlying cell death biology. The presented data demonstrate the versatility of the technology using single concentration treatments, but these workflows and analyses are well suited for experiments using perturbagen titrations and determining optimal conditions for subsequent studies. Collectively, SPARKL is a versatile, data-rich, non-intensive, and zero-handling method to visualize, detect, and quantify the kinetics of cell death, which, when coupled with mathematical analyses, enable investigators to conduct thorough high-throughput comparative studies.

Limitations

The data presented herein have utilized either immortalized MEFs or primary AEFs to demonstrate the workflows compatible with SPARKL. These adherent cells demonstrate minimal PS exposure under normal growth conditions. In our method, cells are incubated with labeled AV that may bind to stochastically exposed PS on the outer leaflet of the plasma membrane and result in fluorescent puncta. We have previously described that excessive concentrations of AV in the media also increase this effect (Gelles and Chipuk, 2016). These puncta are easily excluded while developing appropriate processing definitions (Table 1) but may pose a difficulty for cell types demonstrating significant basal PS localization on the outer leaflet. Similarly, mouse fibroblasts exhibit robust PS exposure following caspase activation; the utility of an AV-based detection is diminished in cell types that do not scramble plasma membrane lipids as rapidly (Suzuki et al., 2013). Here, MEFs tolerated the use of SYTO21 as a non-perturbing cell marker; the tolerance of other cell types to concentrations of SYTO21 may vary its efficacy. Accurate quantification of fluorescent events requires competent segmentation of fluorescent signals to avoid under- or over-counting events. Therefore, cell lines that grow in colonies or exhibit a “cobble-stone” monolayer may require more stringent optimization to adequately establish segmentation between neighboring cells. In this work, apoptotic bodies demonstrated minimal detachment from the plate following cell death and labeled sufficiently with our detection reagents. However, adherence following cell death is likely to be variable with certain cell types or perturbagens. Similarly, treatments resulting in complex cellular morphology could complicate detection and quantification of fluorescent events using this method. While many of exampled applications can be modified for suspension cells adhered to the plate by a binding agent, additional trouble-shooting is necessary and event segmentation is generally less representative due to cellular aggregates. Finally, high-content live-cell imagers require more robust signal for detection when compared to flow cytometry, and therefore, certain methods, such as transient expression of fluorescent proteins, may not be detectable in these high-throughput workflows.

Table 1.

Channel Settings for Each Fluorescent Reporter Used for Processing Definitions with the IncuCyte ZOOM

	Channel	Parameters	Radius (μm)	Threshold (RFU)	Edge Sensitivity	Area (μm2)
Annexin V-AF594	Red	Top-Hat	10	5	−35	>150
Annexin V-FITC	Green	Top-Hat	25	3	−30	>100
Annexin V-FITC (low)	Green	Top-Hat	25	2	−30	>50
SYTO21	Green	Top-Hat	25	0.5	−25	>150
YOYO3	Red	Top-Hat	10	3	−46	>150

STAR★METHODS

LEAD CONTACT AND MATERIALS AVAILABILITY

Code for image processing automation has been made available at https://github.com/ChipukLab/SPARKL_pipeline.git. Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Jerry Chipuk (jerry.chipuk@mssm.edu).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell Lines

Bax^+/+Bak^+/+ and Bax^−/−Bak^−/− SV40-transformed MEFs were obtained from ATCC (Manassas, VA, USA). Cyld^−/− MEFs reconstituted with either a Flag-tagged CYLD construct or empty vector as described previously (O’Donnell et al., 2011). Primary adult ear fibroblasts (AEFs) were isolated from Mlkl^+/+ and Mlkl^−/− mice (Murphy et al., 2013). Cells were cultured in high-glucose DMEM (Cat. No. 10-017-CV, Corning, Manassas, VA, USA) containing 10% FBS (Cat. No. 10438-026, Life Technologies, Carlsbad, CA, USA), 2mM L-glutamine (Cat. No. 25030-081, Life Technologies, Carlsbad, CA, USA), and 1× Penicillin/Streptomycin (Cat. No. 10378-016, Life Technologies, Carlsbad, CA, USA), and grown in humidified incubators at 37°C with 5% CO₂. MEFs were maintained in mycoplasma-free conditions as verified by the HEK-Blue Detection Kit (Cat. No. hb-det2, Invivogen, San Diego, CA, USA). All cell genotypes were confirmed by western blot analysis.

METHOD DETAILS

Reagents and Equipment

Cell culture reagents and chemicals were from Sigma Aldrich (St. Louis, MO, USA) or Thermo Fisher Scientific (Waltham, MA, USA) unless otherwise stated. Drugs and biologics as follows: ABT-737 (Selleck Chemicals); CHX (Sigma); YOYO3 and SYTO reagents (Life Technologies/Thermo Fisher Scientific), mTNFα and mTRAIL (Peprotech, Rocky Hill, NJ, USA); zVAD-fmk (ApexBio, Houston, TX, USA), birinapant (MedChem Express). Fluorophores: FITC (Cat. No. 46425, Thermo Fisher Scientific), Alexa Fluor 488 5-SDP Ester (Cat. No. A30052, Thermo Fisher Scientific), Alexa Fluor 594 NHS Ester (Cat. No. A37572, Thermo Fisher Scientific).

Generating Fluorescent Annexin V

Annexin V Expression and Purification

Recombinant Annexin V was generated and labeled as previously described (Logue et al., 2009). His-tagged Annexin V recombinant protein was overexpressed in E. coli DH5α cultured in Terrific Broth media and induced via 0.5 mM IPTG at 37°C for 3 hours with constant shaking. Cells were harvested at 10,000 × g for 10–20 minutes and resuspended in 100 ml of lysis buffer (50 mM Tris-HCl, pH 8.5, 200 mM NaCl, 5 mM Imidazole, 0.1 mM AEBSF, 5 mM β-mercaptoethanol, and 1 tablet of protease inhibitor). Cells were lysed via sonication on ice using 30/120 second pulse/pause intervals for 5 minutes, centrifuged at 15,000 × g to pellet insoluble material, and the supernatant was loaded onto a 5 ml HisTrapFF column equilibrated with Buffer A (20 mM Tris-HCl, pH 8.5, 200 mM NaCl, 5 mM Imidazole, and 5mM β-mercaptoethanol, and 10% glycerol). Upon binding, the HisTrapFF column is washed with 20 column-volumes with buffer A to remove any non-specific binding. His-tagged Annexin V protein was eluted with a gradient over 30 column-volumes using Buffer B (20 mM Tris-HCl, pH 8.5, 200 mM NaCl, 500mM Imidazole, and 5 mM β-mercaptoethanol). Peak fractions were pooled and further purified over a Super-dex S75 16/60 column equilibrated with Gel Filtration Buffer (1× PBS, pH 7.2, 150 mM NaCl, 5 mM β-mercaptoethanol). Annexin V-containing fractions were confirmed by SDS-PAGE and concentrated with a Centricon concentrator to 2 mg/ml for storage and labeling in 1.5 ml aliquots.

Annexin V Labeling

Per labeling reaction, swell G-25 beads in PBS overnight at an approximate ratio of 0.5 g beads in 5 ml PBS. Add 100 μl of 1 M NaHCO₃ pH 8.3 to a 1.5 ml aliquot of purified recombinant Annexin V for labeling reaction. Immediately prior to use, dissolve fluorophores in high-quality unopened DMSO to a concentration of 10 mg/ml (AF488: 1 mg in 100 μl DMSO; AF594: 100 μg in 10 μl). Add dissolved fluorophore to 1.5 ml aliquot of 2 mg/ml Annexin V at a molar ratio of 12–15:1 (in this formulation, the entire solution of AF488 or AF594 is added to the Annexin V). Protect the tubes from light and mix on a rotisserie rotator for 1 hour at room temperature. Prepare a dye removal column for each 250 μl of reaction (typically, 6–8 columns will be needed). Rinse G-25 beads in PBS, resuspend in 2.5 ml (adjusted as needed for number of columns), aliquot approximately 350 ml of slurry to each column, and briefly centrifuge at 1000 × g to remove PBS. Add 250 μl of fluorophore/Annexin V reaction to each column, briefly vortex, and spin at 1000 × g to collect protein. Pool flow-through, dilute to 200 μg/ml (final volume of 10 ml), and aliquot for storage. Confirm labeling efficacy and required final concentration via flow cytometry of dying cells.

Cell Culture and Treatments

Unless stated to the contrary, experiments were performed with Bak^+/+Bax^+/+ and Bak^−/−Bax^−/− (WT and DKO, respectively) SV40-transformed MEFs (ATCC, Manassas, VA, USA). Primary adult ear fibroblasts (AEFs) were isolated from Mlkl^+/+ and Mlkl^−/− mice as follows: mouse ears were clipped, minced, and incubated in trypsin at 37°C for 1 hour with repeated vortexing before being cultured on tissue culture plates (Murphy et al., 2013). For kinetic studies, cells were seeded at approximately 2–4 × 10³ cells/well in a 96-well format and left to adhere for 18–24 hours. Prior to live-cell imaging, growth media was replaced with complete phenol-red-free DMEM containing indicated fluorescent reagents and perturbagens. Annexin V was used at 250 ng/ml; YOYO3 was used at 250 nM; SYTO21 was used at 300 nM. DMEM contains sufficient calcium concentrations for robust Annexin V labeling but other media formulations require calcium supplementation for optimal Annexin V labeling (1.5–2 mM CaCl₂). Plates were pre-warmed prior to data acquisition to avoid condensation and expansion of the plate, which hinders auto-focusing during scan intervals.

SPARKL Data Acquisition and Automated Analysis

Data Acquisition

Kinetic experiments were performed with the IncuCyte ZOOM (Model 4459, Essen Bioscience, Ann Arbor, MI, USA) residing in a tissue culture incubator maintained as above, but other high-content live-cell fluorescent imagers are capable of replicating these data (e.g., IN CELL analyzer). Experiments were conducted for 24–96 hours with data collection every 1–2 hours to avoid photobleaching of fluorescent reporters. Using the 10× objective, a single plane of view was collected per well for 96-well plate assays. Phase contrast, green channel (Ex: 440/80 nm; Em: 504/44 nm; acquisition time: 400 ms), and red channel (Ex: 565/05 nm; Em: 625/05 nm; acquisition time: 800 ms) were collected for all experiments with spectral unmixing set as 3% of red removed from green (except for studies assessing dual-labeling kinetics, in which case spectral unmixing was set at 0%). Images collected were 1392 × 1040 pixels at 1.22 μm/pixel.

Event Detection and Data Visualization

Automated event detection was accomplished using the ZOOM software (v2018A) and user-defined processing definitions using images collected using the specific cell lines and fluorescent reporters pertaining to the experiment. Processing definitions settings are shown in Table 1; the settings for “AV-FITC (low)” were used for necroptotic studies where the entire cell surface labeled. Kinetic data graphs are expressed as calculated events per well and exported by the IncuCyte software. The y-axis scale was determined for each experiment using internal control treatments to assess maximal death fora population in parallel to the experimental data (commonly TNFα+CHX with ABT-737 or CHX with ABT-737; the pan kinase inhibitor staurosporine is not suggested as the morphology of cells complicates event identification). Fluorescent intensity data were analyzed using the average integrated intensity (AU × μm²) metric as exported by the IncuCyte software. Images of cell culture presented in this manuscript were white-point corrected for clarity and to normalize over-compensated visualization of green and red pixels in frame containing no significant signal. White-point correction was accomplished in post by adjusting temperature and tint parameters to overlap peak contributions of red, green, and blue pixels.

Data Handling and Processing for Parameterization

Data exported from the IncuCyte ZOOM v2018A (Essen Bioscience) were organized, handled, and further analyzed using Excel v16.16.9 (Microsoft) and Prism 8.1.1 (Graphpad Software Inc.). Event count data were converted into histograms using Equation 1, where #Signal⁺ refers to the number of events using either fluorescently-labeled AV or Y3 for a given well “i” at time point “j”:

$$\text{\#Events}/{\text{hour}}_{\text{well}=\text{i},\text{time}=\text{j}}=\left({\text{\#Signal}+}_{\text{i},\text{j}}-{\text{\#Signal}+}_{\text{i},\text{j}-1}\right)$$ (Equation 1)

Area under the curve (AUC) values were calculated with the analysis tool in Prism and utilized the trapezoid rule ignoring peaks that were less than 10% of the distance from minimum to maximum Y values. The first derivative of collected data was calculated with the analysis tool in Prism and smoothed the data using 4 neighbors and 2nd order smoothing polynomial. Non-linear curve fits were calculated using the “plateau followed by one-phase association” function in Prism using the least squares fitting method given by Equation 2. Parameters extracted by Equation 2 are defined as follows: D(t”) is the amount of death for given time point t”; D₀ is the initial and minimum amount of death; D_M is the plateau of maximal death; R_D is the initial and maximal rate of death; t’ is the time at which the lag plateau ends and exponential increase in death begins. All parameters were left unconstrained for mathematical fitting. Fits listed as ambiguous by Prism only occurred in conditions that did not result in death positivity and are highlighted within the text.

$$\text{D}\left(t^{″}\right)={\text{D}}_0+\left({\text{D}}_M-{\text{D}}_0\right)\left(1-{\text{e}}^{-{\text{R}}_{\text{D}}\left(t^{″}-t^{\prime }\right)}\right)$$ (Equation 2)

Experiments using SYTO21 to normalize data with cell counts was accomplished using Equation 3, where #Dead was determined by the event counts using either fluorescently-labeled AV or Y3 for a given well “i” at time point “j”:

$$\%{\text{Death}}_{\text{well}=\text{i},\text{time}=\text{j}}=\left(\frac{{\text{\#Dead}}_{\text{i},\text{j}}}{{\text{\#SYTO}+}_{\text{i},\text{j}}}\right)$$ (Equation 3)

Single-cell Sequential Labeling Kinetics

For experiments investigating sequential labeling of individual cells, analysis utilized event masks generated by the IncuCyte ZOOM software which were exported for further processing using external software packages. Specifically, masks for green and red channels as well as the overlap masks were used for these analyses. Cell event masks were recognized and spatio-temporally defined using ImageJ and Fiji software packages (ImageJ, National Institutes of Health, Bethesda, ML, USA) (Schneider et al., 2012; Schindelin et al., 2012) as detailed below. Images exported from the ZOOM software were imported into ImageJ as stacks for each mask collection (green, red, and overlay), converted to 8-bit grayscale and binary, and subjected to translation alignment using the Stack-Reg and TurboReg plugins (Thévenaz et al., 1998). Reference regions of interest (ROIs) were autonomously defined by applying the Analyze Particles tool on the final image slice of the overlay mask. These reference ROIs were applied to all slices within the image stacks corresponding to the green, red, and overlay channels using the Multi-Measure tool. Positive pixel intensity within ROIs defined the slice in which the signal occurred. These data were then sorted and aligned in Excel by ROI reference number, thereby linking identified signal slices from each channel to the same ROI for graphing and interpretation. This workflow was used to quantify the progression of either green or red single-positivity to double-positivity for single cells within in a population. This workflow was automated using custom code which has been deposited to Github.

QUANTIFICATION AND STATISTICAL ANALYSIS

Histogram data were calculated using Excel v16.16.9 (Microsoft); AUC, derivative, and non-linear fit analysis were conducted using Prism 8.1.1 (Graphpad Software Inc.). Graphing and all statistical analyses were performed using Prism 8.1.1. Graphs of kinetic data have been smoothened for convenience of visualization and these steps have not affected the interpretation of the data or downstream analyses which were performed on the raw collected data. Unless otherwise noted, all experiments were conducted in triplicate and the presented data are the mean of replicates and representative of repeated experiments. When shown, error bars denote standard deviation of replicates; select graphs omit error bars for convenience of visualization. Violin plots display the entire population of collected data and have been smoothened to aid in visualization. Instances of flat ends are an artifact of the method used to create the frequency distribution and do not reflect missing or truncated data points. Median and quartiles are shown as a solid and dotted line, respectively. Population statistics were conducted by performing a Mann-Whitney U test for each indicated comparison denoting significance as follows: ****P < 0.0001, ***P <0.001, **P < 0.01, *P < 0.05, ns = not significant. Single-cell sequential labeling analyses were performed using exported event masks collected by the IncuCyte ZOOM v2018A and ImageJ/Fiji software packages as described in the Method Details. Data showing frames from the IncuCyte ZOOM were manually post-processed for white-point correction using GIMP-2.10 (https://www.gimp.org). Figures and diagrams were assembled using Inkscape v0.92.2 (https://inkscape.org).

DATA AND CODE AVAILABILITY

The code generated during this study to automate single-cell sequential labeling analyses are available at https://github.com/ChipukLab/SPARKL_pipeline.git (https://doi.org/10.5281/zenodo.3458574) under the GNU General Public License v3.0.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals, Peptides, and Recombinant Proteins
ABT-737	Selleck Chemicals	Cat#: S1002
mTNFα	Peprotech	Cat#: 315-01A
mTRAIL	Peprotech	Cat#: 315-19
STYO21	Thermo Fisher Scientific	Cat#: S7556
YOYO-3 Iodide	Thermo Fisher Scientific	Cat#: Y3606
zVAD-fmk	APExBIO	Cat#: A1902
Experimental Models: Cell Lines
Bax+/+Bak+/+ SV40 transformed MEF	ATCC	Cat#: CRL-2907; RRID: CVCL_U630
Bax−/−Bak−/− SV40 transformed MEF	ATCC	Cat#: CRL-2913; RRID: CVCL_U626
Cyld−/− SV40 transformed MEF	Adrian Ting (Icahn School of Medicine at Mount Sinai, NY, USA)	O’Donnell et al., 2011
Mlkl−/− adult ear fibroblasts	Warren Alexander (University of Melbourne, Melbourne, Australia)	Murphy et al., 2013
Recombinant DNA
pProEx.Htb.annexinV	Seamus Martin (Trinity College, Dublin, Ireland)	Accession#: NM_001154; Logue et al., 2009
Software and Algorithms
Excel v16.16.9	Microsoft	N/A
ImageJ/Fiji v2.0.0-rc-69/1.52n	National Institutes of Health	N/A
IncuCyte ZOOM v2018A	Essen Biosciences	N/A
Prism v8.1.1	Graphpad Software	N/A
Sequential labeling pipeline	This paper	https://doi.org/10.5281/zenodo.3458574

Highlights.

• SPARKL workflows use live-cell imagers to capture the kinetics of cell death in real time

• Multi-parametric analyses of cell death kinetics reveal mechanisms for comparative study

• Multiplex workflows accounts for cell-inherent or drug-induced proliferation changes

• Single-cell analyses quantify differential response to drugs in isogenic cell populations