Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2021 May 1.
Published in final edited form as: Biotechnol Bioeng. 2020 Feb 3;117(5):1470–1482. doi: 10.1002/bit.27280

Apoptosis detection via automated algorithms to analyze biomarker translocation in reporter cells

A H Rezwanuddin Ahmed 1, Zeynep Dereli-Korkut 1, Joanne Haeun Lee 1, Sidra Piracha 1, M Lane Gilchrist 2, Xuejun Jiang 3, Sihong Wang 1
PMCID: PMC7703573  NIHMSID: NIHMS1648191  PMID: 31956989

Abstract

Rapid, efficient, and robust quantitative analyses of dynamic apoptotic events are essential in a high-throughput screening workflow. Currently used methods have several bottlenecks, specifically, limitations in available fluorophores for downstream assays and misinterpretation of statistical image data analysis. In this study, we developed cytochrome-C (Cyt-C) and caspase-3/−8 reporter cell lines using lung (PC9) and breast (T47D) cancer cells, and characterized them from the response to apoptotic stimuli. In these two reporter cell lines, the spatial fluorescent signal translocation patterns served as reporters of activations of apoptotic events, such as Cyt-C release and caspase-3/−8 activation. We also developed a vision-based, tunable, automated algorithm in MATLAB to implement the robust and accurate analysis of signal translocation in single or multiple cells. Construction of the reporter cell lines allows live monitoring of apoptotic events without the need for any other dyes or fixatives. Our algorithmic implementation forgoes the use of simple image statistics for more robust analytics. Our optimized algorithm can achieve a precision greater than 90% and a sensitivity higher than 85%. Combining our automated algorithm with reporter cells bearing a single-color dye/fluorophore, we expect our approach to become an integral component in the high-throughput drug screening workflow.

Keywords: biomarker translocation, caspase-3, caspase-8, cytochrome-C, detection algorithm, reporter cells

1 |. INTRODUCTION

Apoptosis is a programmed cell death mechanism essential for multiple biological processes including tissue homeostasis, development, and regulation of the immune system. Pathological conditions, such as developmental defects, cancer, and autoimmune diseases, can arise from the dysregulation of apoptotic controls. Apoptosis is characterized by its various hallmarks, including the loss of cell volume, chromatin condensation, and DNA fragmentation. Apoptosis is a well-regulated, noninflammatory, caspase-guided destruction of the cell. Apoptosis can be broadly classified into two main pathways–the intrinsic pathway and the extrinsic pathway (Dereli-Korkut et al., 2013; Jiang & Wang, 2004).

The intrinsic pathway is invoked by a stimulus internal to the cell, such as radiation and chemotherapeutic drugs. It involves the release of cytochrome-C (Cyt-C) from mitochondria, which in turn triggers the formation of an apoptosomal complex along with Apaf1 and procaspase-9. This complex is further autoprocessed to generate catalytically active caspase-9. Activated caspase-9 cleaves and activates downstream executioner caspases-3 and −7, and a further cascade of cellular events that lead to cell death (Dereli-Korkut et al., 2013; Goldstein, Muñoz-Pinedo, et al., 2005; Jiang & Wang, 2004; Leibowitz & Yu, 2010). The mitochondrial release of Cyt-C is accompanied by the release of various other proteins, including Smac/DIABLO and Omi/HtrA2, which bind to the inhibitor of apoptosis proteins, releasing their inhibitory effect on caspases to amplify the apoptotic signal (Leibowitz & Yu, 2010; Luetjens et al., 2001; Srinivasula et al., 2000). Various members in the Bcl-2 family of proteins confer apoptotic or antiapoptotic effects to a cell, closely correlating to their capability to confine or release mitochondrial Cyt-C and control mitochondrial membrane potential. As Cyt-C is integral in the electron transport chain, its release from the mitochondria, if approaching to completeness, results in the inevitable death of the cell even in the absence of downstream caspase activation (Goldstein, Waterhouse, Juin, Evan, & Green, 2000; Heiskanen, Bhat, Wang, Ma, & Nieminen, 1999; Jiang & Wang, 2004; Luetjens et al., 2001; Martinou, Desagher, & Antonsson, 2000; Waterhouse et al., 2001).

The extrinsic pathway involves death receptors on the membrane of the cell. The extrinsic pathway can transition into the intrinsic pathway to amplify the external stimulus or directly act on activating the executioner caspase cascade, pushing the cell to apoptosis. For example, tumor necrosis factor (TNF)-related apoptosis-inducing ligand (TRAIL) can activate the extrinsic pathway of apoptosis. TRAIL binds to and leads to the trimerization of death receptors on the cell membrane (e.g., DR4 and DR5), whose death domain (DD) recruits an adaptor protein, Fas-associated death domain (FADD), activating the latter. The death effector domain of the FADD further recruits and activates caspase-8, forming an active death-inducing signaling complex (Ashkenazi & Dixit, 1998; Dereli-Korkut et al., 2013; MacFarlane, 2003; Nagata, 1997; Nagata, 1999; Scaffidi et al., 1998). Depending on the cell type, caspase-8 can act as an initiator caspase to further activate caspase-3 and −7 downstream, or trigger the activation of Bid, which sequesters Bax (a proapoptotic member of the Bcl-2 family) in mitochondria forming supramolecular openings in the outer mitochondrial membrane, causing the release of Cyt-C and downstream activation of caspase-9 (B. Antonsson, Montessuit, Sanchez, & Martinou, 2001; Ashkenazi & Dixit, 1998; MacFarlane, 2003; Nagata, 1999; Rehm et al., 2009; Scaffidi et al., 1998; Suliman, Lam, Datta, & Srivastava, 2001).

Drug screening is an important facet in the drug development pipeline, with one of several end goals being to scope out toxic drugs toward tumorigenic cells while sparing the healthy tissues. Examples include the use of TRAIL, which is effective on cancer cells that overexpress the DR4 or DR5 receptors while sparing normal cells (e.g., normal hepatocytes) due to their decoy receptors (Dereli-Korkut et al., 2013; MacFarlane, 2003). On the other hand, doxorubicin is used to treat several categories of cancers, including, but not limited to, that of breast, lung, bladder, lymphoma, and leukemia, can pose severe cardiomyotoxicity and induce secondary mutagenic cancers (Chatterjee, Zhang, Honbo, & Karliner, 2010; Rheingold & Meadows, 2003). Because drug responses of individual patients are not the same, drug screening broadly encompasses two categories, the personalized or common track. While the former uses a patient’s specific biopsy or genetic profiling to administer a suitable drug clinically, the latter is a broader application in terms of developing a drug that works for the majority of the common populace. In both cases, a method to identify the effectiveness of a drug is vital, and the ability to streamline the screening into a robust, reliable, high-throughput process is necessary. There are several assays available commercially and/or for research purposes from simple live-dead assays and surface apoptosis markers such as Annexin-V, to cytosolic monitoring of apoptosis through caspase reporter biomarkers (Carpenter et al., 2006; Gelles & Chipuk, 2016; Helmy & Azim, 2012; Lamprecht, Sabatini, & Carpenter, 2007; Saraste & Pulkki, 2000; Saraste, 1999; Xiao, Gibbons, Luker, & Luker, 2015; Ziraldo, Link, Abrams, & Ma, 2014, 2015). Most apoptosis assays are end-point assays lacking abilities to report dynamic drug responses temporally and intracellular spatially. On the other hand, most image analysis studies using apoptosis reporter cell lines have the following limitations. They (a) rely on proprietary software for image analysis that can be heavily manual and biased in terms of adjusting the correct “thresholds” to read a measurable effect; (b) are not robust due to lack of mathematically accuracy, wherein images visually giving different information to human eyes would pass undetected on an algorithm; or (c) use multiple fluorophores, such as in epifluorescence, fluorescence lifetime imaging, or fluorescence resonance energy transfer imaging, with one or more fluorophores marking cell organelles for image registration, while another fluorophore marks the biomolecule that characterizes a particular cell event like apoptosis, which leaves less room for using other fluorophores of similar color as part of a secondary assay. These three limitations add a bottleneck to the high-throughput analysis using reporter cell lines.

In this study, we address these issues in a two-fold manner. The first addresses the generation of cell lines that serve as single-color fluorophores that can be tracked in cells on a conventional epifluorescence microscope. Three apoptosis reporter cell lines with Cyt-C conjugated with green fluorescence protein (Cyt-C-GFP), caspase-3 reporter, and caspase-8 reporter were constructed using PC9 non-small-cell lung cancer cells and T47D ductal carcinoma cells in this study. Prior studies have established that the tagging of GFP onto Cyt-C did not affect the biological kinetics of Cyt-C, and the localization of the conjugated Cyt-C-GFP into the mitochondrial membrane can be verified with mitochondria-specific dyes (Goldstein, Muñoz-Pinedo, et al., 2005; Goldstein, Waterhouse, et al., 2000; Luetjens et al., 2001). Caspase-3 reporter is commercially available in plasmid form, while the caspase-8-reporter plasmid was generated as a modification to the caspase-3 plasmid. Both caspase reporter molecules have caspase-specific cleavage sites (DEVD for caspase-3 and IETD for caspase 8) bridging a nuclear export signal (NES) to a nuclear localization sequence (NLS) tagged to EYFP. This permits the EYFP to be present in the cytosol until the activation of the respective caspase cleaves the EYFP-NLS sequence and allows its transport to the nucleus.

The second aspect of our study concerns the development of a robust algorithm that avoids the pitfalls of using faulty statistical variables or potentially biased manual procedures currently used in mid throughput (10s to a few 100s of samples) to high-throughput (several 100s to a few 1000s of samples) image analysis systems for identifying fluorescent signal translocation in apoptotic events. We identify extractable features and criteria, which provide valuable, robust information that strongly coincide with the human perspective of identifying biomarker translocation, and use statistical measures from the images to make the process more automated and unbiased. The method can be scaled to single-cell analysis, to high-throughput analysis of single cells, or to high-throughput batch (average) analysis.

2 |. METHODS

All reagents without specific manufactures were purchased from Sigma Aldrich.

2.1 |. Construction of cytochrome-C-GFP reporter cell line and its characterization

Human lung cancer cells (PC9) were cultured in Roswell Park Memorial Institute (RPMI) 1640 media (ATCC, Manassas, VA) with 10% fetal bovine serum (FBS; Atlanta Biologicals) and 1% penicillin and streptomycin (Invitrogen) at 37°C in an incubator with 5% CO2. A stable PC9 reporter cell line expressing the Cyt-C-GFP fusion protein was constructed using pCyt-C-GFP, kindly provided by Dr. Douglas Green at St. Jude Children’s Research Hospital (Goldstein, Waterhouse et al., 2000) via electroporation with Gene PulserXcell (Bio-Rad) at 960 uF capacitance and 250 V. Stable cell lines were generated over 2 weeks by selecting PC9 cells in the presence of 650 ug/ml geneticin. After 2 weeks, cloning rings were used to select clones expressing GFP signals, which were further sorted by fluorescence activated cell sorting (FACS). Cells expressing high GFP levels were selected. The sorted cells were further cultured and used in this study.

Western blot analysis was used to confirm the expression of the fusion protein, Cyt-C-GFP. Protein gel blotting analyses were performed as previously described (Dereli-Korkut et al., 2013). Briefly, cells were lysed in lysis buffer with 1% sodium dodecyl sulfate (SDS), 50 mM Tris-HCl, 5 mM ethylenediaminetetraacetic acid, and protease inhibitors containing 0.1 mM phenylmethylsulfonyl fluoride, 1 ug/ml leupeptin, 1 ug/ml aprotinin, and 1 ug/ml pepstatin. Proteins were separated with 15% SDS-polyacrylamide gel electrophoresis, transferred to polyvinylidene fluoride membranes (Bio-Rad, Hercules, CA), and immunoblotted with the primary antibody of GFP, followed by incubation with the horseradish peroxidase-conjugated secondary antibody. Tetra-methylbenzidine substrate kit (Vector Laboratories, Burlingame, CA) was used to visualize the protein bands. Membranes were dried and scanned into digital images.

2.2 |. Colocalization of Cyt-C-GFP with mitochondria through confocal microscopy

The Cyt-C-GFP reporter cells were cultured in Nunc Lab Tek 2-well chamber slides (Thermo Fisher Scientific), and stained with 2 mM Mitotracker Red CMXRos (Invitrogen). During imaging, serum-free medium was used. Confocal images (Leica) of Cyt-C-GFP and Mitotracker Red were collected in Z-stacks to enable 3D reconstruction in LEICA software. A Leica TCS SP2 AOBS confocal microscope system equipped with argon ion and HeNe lasers and a 40 × /1.3 NA oil-immersion objective was used for all of the CLSM images. Cyt-C-GFP was excited using the 488 nm line of the argon laser, and images were taken with the detection window set between 500 and 536 nm. Mitotracker Red was excited with a 543 nm line and images were taken with the detection window set between 579 and 599 nm. The pinhole aperture was set at an Airy value of 1.0. Colocalization analysis was done in ImageJ, and further 3D analysis was done using Amira 3D (FEI Software). Another set of samples were similarly prepared in a chamber slide. Before the addition of Mitotracker Red, TRAIL (400 ng/ml) was added to the chambers and incubated for 2 hr. After incubation, cells were washed one time with phosphate buffered saline and 2 mM Mitotracker was loaded onto them. The same procedure as above for confocal imaging was followed.

ImageJ and AMIRA 3D were used for visual-friendly presentation of 3D rendered confocal results. In ImageJ, confocal Z-stacks were processed with Gaussian Blur set to 1.1 followed by gamma correction set to 0.9. In AMIRA 3D, Cyt-C-GFP fluorescence was displayed as Volren and color-coded green using the volrenGreen color map. To delineate the surface boundaries of the mitochondria, the Mitotracker Red channel 3D data were displayed using the isosurface command using a transparent red mesh.

2.3 |. Time lapse imaging of Cyt-C release after stimulation with anticancer drugs

PC9 Cyt-C-GFP reporter cells were cultured as described above and incubated with serum-free media for 2 hr followed by drug stimulation of 400 ng/ml TRAIL (TNF-related apoptosis inducing ligand). Time lapse images were recorded with a 20× objective for 12 hr at 30-min intervals (Axio-Observer Z.1 microscope; Carl Zeiss).

2.4 |. Construction of pCaspase8-sensor and caspase-3 and -8 reporter cell lines

The pCaspase8-sensor was constructed to detect the onset of caspase 8 activity in mammalian cells. It was designed by adapting the pCaspase3-sensor (Clontech) plasmid. The pCaspase3-sensor encodes the enhanced yellow-green variant (EYFP) fused at the 3′ end to three copies of the NLS of the simian virus 40 large T Antigen. At the 5′ end, the gene contains a sequence encoding the NES of Map Kinase Kinase (MAPKK) and caspase-3 specific cutting sequence, DEVD. Since the NES of MAPKK dominates the NLS, the full length fluorescent fusion protein is localized to the cytosol. When caspase-3 is activated and cuts at DEVD, the NES is cleaved from the fusion protein and the truncated EYFP-NLS fusion translocates to the nucleus via the NLS. The translocation of the fluorescent protein from the cytosol to the nucleus indicates caspase-3 activation at a cellular level (Kamada, Kikkawa, Tsujimoto, & Hunter, 2005). To construct pCaspase8-sensor, the caspase-3-specific cleavage site in the pCaspase3-sensor was replaced with oligonucleotides coding caspase 8-specific cleavage sequences by using polymerase chain reaction (PCR)-mediated overlap extension. Primers encoding caspase 8 specific cleavage sequences used in two rounds of PCR are listed as bellow.

  • Primer a: 5′ GCCTGGCATTATGCCCAGTACATGACCTTATGGGAC TTTCCTACTTGC 3′

  • Primer bΔ: 5′ GTCGGTCTCGATGCCACCATCAGTTTCAATGCCCTT CCTCTTCTG 3′

  • Primer cΔ: 5′ ATTGAAACTGATGGTGGCATCGAGACCGACGGCGT GGACGAGGTG 3′

  • Primer d: 5′ GGCTGATTATGATCAGTTATCTAGATCCGGTGGATC CTACCTACCTTTC 3′

As a result, the amino acid sequence was modified from G-G-D-E-V-D-G-G, the caspase-3-specific cleavage site, to G-G-I-E-T-D-G-G-I-E-T-D-G-G, which corresponds to the caspase-8 cleavage site. The detailed procedures of pCaspase8-sensor construction and confirmation are included in Supporting Information Material SI.

The caspase-8 reporter cell lines were constructed using human non-small-cell lung cancer cells (PC9) and human breast cancer (T47D) via transient transfection. PC9 and T47D cells were cultured using RPMI medium with 10% FBS and 1% penicillin/streptomycin. At 85% confluence, the cells were transfected with pCaspase8-sensor using Lipofectamine 3000 (Life Technologies). Twenty four to forty eight hours after transfection, about 30%–40% of cells were carrying the fluorescence signal. The nucleus was stained with Hoechst 33342 dye for visual verification of nuclear translocation. Apoptosis in PC9 cells was induced with 400 ng/ml of TRAIL and in T47D cells with 500 ng/ml of Fas-L. Caspase activity was captured using time lapse fluorescence microscopy (Axio-Observer Z.1; Carl Zeiss).

2.5 |. Image analysis and translocation detection algorithm

Analysis of signal translocation was done in a custom algorithm developed in MATLAB (details are provided in Supporting Information Material SI). Briefly, fluorescence images for every time point are segmented into frames enclosing single cells (see Supporting Information Material SII, Section I: Image Preparation). Each frame is further cleaned up by despeckling and background removal. A region of interest (ROI) mask is created for each frame by thresholding and morphological operations. Using the radon transform values along a few primary angles, a weighted autocorrelation value is generated, normalized to the area of the ROI. The first such weighted correlation value is assigned the name “Global Correlation Value,” or Gcorr. Further normalization of Gcorr to the maximum possible weighted correlation (Maxcorr) provides the “Local Normalized Correlation Value,” or Ncorr (see Supporting Information Material SII, Section III; Steps 7–8). Traditional analytic parameters utilized in signal translocation measures are also calculated from the ROI image—the area of the cell as the sum of the pixels in the mask, and the coefficient of variation (CV) generated from the intensity values of the pixel population within the ROI. The last two related parameters generated are the “Smoothing Ratio” (SR) and “dp” values. Briefly, the SR values are generated from an in-frame ratio between an average-filtered version of the image to the original image. The difference between sequential values of SR along the time axis is assigned the value dp. To identify translocation of signal within the cells at any time point, four criteria are identified, where meeting any of the criteria qualifies a data point as being identified as a point of interest of translocating signal: (a) Ncorr-dp criteria, which is passed if there is a significant increase (>1.3–1.8 std. dev.) in dp along with a drop in Ncorr, (b) Single-drop criteria, which is a large drop (>2–2.2 std. dev.) in the Ncorr value qualifying the point as an outlier value compared to the fluctuation throughout the whole graph, (c) Double-drop criteria, which is a significant (>1.1–1.5 std. dev.) consistent drop in Ncorr across two consecutive frames, and (d) Matched double-drop criteria, which is any increase in SR (dp > 0) across two consecutive time points, accompanied by corresponding drops in Ncorr across those same consecutive time points without the change being significant as in the Ncorr-dp and double-drop criteria above. The threshold for significant change is optimized based on the statistical deviation from the training data used, and varies according to cell type being used, and the signal being measured. From our training samples, we find that the above standard deviation (std. dev.) values served as an adequate cutoff to identify points of interest, specifically (a) for criteria 1, 1.3 and 1.8 for Cyt-C and caspase-3/-8 respectively, (b) for criteria 2, 2.1 suffices for all sample types used, and (c) for criteria 3, 1.1 and 1.3 for Cyt-C and caspase-3/−8, respectively.

The confusion matrix to determine the robustness and reliability of the algorithm was constructed using the standard definitions of true positive (TP), true negative (TN), false positive (FP), false negative (FN), accuracy ([TN+TP]/[All samples]), precision (TP/[FP+TP]), sensitivity (TP/[TP+FN]), and specificity (TN/[TN+FP]). True positives (or negatives) correspond to samples where the algorithm and visual indication recognize a signal translocation (or lack thereof). False positives include cases where the algorithm recognizes translocation but there is no visual indication. False negatives are cases where there are visual cues of translocation occurring but the algorithm fails to detect as such.

3 |. RESULTS

3.1 |. Reporter cell lines: Cyt-C-GFP PC9 and caspase-8-sensor cell lines

Stable transfected PC9 cells with pCyt-C-GFP and pEGFP-N1 are shown in Figures 1a and 1c, respectively. The Cyt-C-GFP is significantly punctuated (Figure 1a) while the EGFP signal is diffused throughout the cytosol of the cell (Figure 1c). Figure 1e is the result of flow cytometry analysis of PC9 cells expressing the fusion protein of Cyt-C-GFP after cell sorting, showing ~85% of the transfected cells are GFP positive (region “R1”). The presence of fusion protein Cyt-C-GFP in the Cyt-C-GFP reporter cell line was further confirmed using western blot analysis as shown in Figure 1f. GFP has a known molecular weight of 27 kDa, while Cyt-C has a molecular weight of 12 kDa. On the western blot membrane exposed to Anti-EGFP antibody, a prominent band in between the 89 and 36-kDa reference lanes is indicative of the presence of the Cyt-C-GFP fusion protein at ~39 kDa. Another light band observable around the 31-kDa band corresponds to the cleaved GFP. The other faint bands in between these two bands (39 and 27 kDa) might be the result of partial proteolytic cleavage of the fusion protein.

FIGURE 1.

FIGURE 1

Open in a new tab

Cytochrome-C-GFP (Cyt-C-GFP) reporter cell line characterization. Stable transfected PC9 cells with pCytochrome-C-GFP in (a) green fluorescence (GFP) channel and (b) its phase contrast image. For comparison, stable transfected PC9 cells with pEGFP-N1 shown in (c) GFP channel and (d) its phase contrast image. (e) Fluorescence activated cell sorting (FACS) sorting results of Cyt-C-GFP PC9 cells expressing a strong GFP signal in “R1” compared to the background green fluorescent signals from nontransfected PC9 cells. (f) Western blot membrane of Cyt-C-GFP reporter cells with anti-GFP antibody. Scalebar: 50 μm

To reveal the exact subcellular location of Cyt-C-GFP, mitochondria were visualized by mitochondrion-specific dye. Figure 2a–f are the maximum intensity projection images of confocal z-stack slices of PC9 Cyt-C reporter cells dyed with fluorescent red mitochondria tracker without (a–c) or with apoptotic drug treatments (d–f). Localization of the fusion protein Cyt-C-GFP within mitochondria in the reporter cells was confirmed by the colocalized signal in yellow (Figure 2c) of Cyt-C-GFP (Figure 2a) and red Mitotracker signal (Figure 2b), the latter tagging cellular mitochondria. Strong colocalization of green Cyt-C-GFP signal to red mitochondrial signal quantitatively measured by a Pearson correlation coefficient (R) of 0.84 is shown in Figure 2l. In contrast, after Cyt-C-GFP diffused into cytoplasm following the apoptotic drug treatment for 2 hr with 400 ng/ml TRAIL, the strongly uncorrelated localization between Cyt-C-GFP and mitochondria gives R of 0.54 (Figure 2m).

FIGURE 2.

FIGURE 2

Open in a new tab

Confirmation of working translocation mechanisms of PC9 Cyt-C-GFP reporter cells dyed with Mitotracker Red (a–f) and T47D caspase 8-sensor cells dyed with blue Hoechst nuclear dye (g–k). (a) Maximum intensity projection (MIP) of Cyt-C-GFP, (b) MIP of Mitotracker Red of the same field of view as (a), (c) overlaid image of (a) and (b) showing colocalization of Cyt-C-GFP and mitochondria by yellow signal, (d) MIP of Mitotracker Red signal in Cyt-C-GFP reporter cells exposed to tumor necrosis factor-related apoptosis-inducing ligand (TRAIL) at 400 ng/ml for 2 hr, (e) MIP of Cyt-C-GFP in the same field of view as (d), (f) overlay image of (d) and (e) showing diffused Cyt-C-GFP out of mitochondria after TRAIL stimulation via minimum yellow areas. (l and m) Results of quantitative colocalization analysis of Cyt-C-GFP signal with Mitotracker Red signal using Pearson correlation coefficient R, in which (l) results of signals between (a) and (b), (m) TRAIL-treated results between (d) and (e). (g) blue nuclei in caspase-8-reporter T47D cells, (h) green channel of (g) showing YFP signal in cytosol, (i) overlay image of (g and h), (j) concentrated YFP signal after 2 hr of Fas-L treatment, (k) overlay of blue and green channels of (j) showing the YFP signal colocalized with nuclear location after Fas-L stimulation. (n) 3D zoom-in of cell “1” from (c) that was not treated with TRAIL. The red isosurface mesh indicates the extent of the mitotracker red and indicates the organelle boundary. The Cyt-C-GFP distribution is indicated in green. (o and p) 3D zoom-in of cells “2” and “3” from (f) treated with 400 ng/ml TRAIL for 2 hr that are undergoing apoptosis at different rates. For (n–p), the mitotracker isosurfaces are cropped (as indicated by the central bisecting planes) to expose the mitochondrial localization and distribution of Cyt-C-GFP within the cytoplasm. Scalebar: (a–f) 20 μm; (g–k) 50 μm. (n–p) 10 μm. Cyt-C, cytochrome-C

The translocation of Cyt-C-GFP from mitochondria to cytoplasm is visually more vibrant in 3D rendered and zoomed in images. Figure 2n shows one half of a 3D rendering of red Mitotracker signal and Cyt-C-GFP from cell “1” in Figure 2c, showing Cyt-C-GFP embedded inside the mitochondrion boundary indicated by red isosurface mesh. Figure 2o,p shows the semi-3D rendering of two cells (cell “2” and cell “3”) from Figure 2f, going through different apoptotic rates, with cell “3” (Figure 2p) showing significant late-stage apoptotic blebbing. Both cells show a delocalization of the Cyt-C-GFP signal (green) from the mitochondria (red), compared to the control condition without drug treatment in cell “1” (Figure 2n), wherein the latter GFP is heavily colocalized with the red mitotracker signal obviated in the 3D rendering.

Figure 2g,h includes T47D cells transiently transfected with pCaspase8-sensor with fluorescent blue nucleus staining and the expression of YFP-conjugated sensor protein respectively. The overlay image (Figure 2i) of Figure 2g,h shows a uniform expression of YFP in cytoplasm in caspase-8 reporter cells without apoptotic drug treatments. The translocation of YFP from cytoplasm to nucleus (Figure 2j,k) after 2 hr of FasL stimulation confirmed that pCaspase8-sensor was successfully constructed.

3.2 |. Algorithmic detection of signal translocation

Figure 3 shows the analytics generated by the detection algorithm. A field of view perspective is shown in Figure 3a–c and Figure 3d–f for caspase-3 PC9 and caspase-8 PC9 reporter cells, respectively, both treated with TRAIL from a set of time lapse images taken at 15-min intervals. The time points of the snapshots are labeled along the figures, with Figure 3a–c representing the earlier time point (t = 0 min, frame 1), a mid point (t = 135 min, frame 10), and an end-point (t = 270 min, frame 19) respectively, for a 19-point time lapse taken at 15-min intervals. The term “frame” here is synonymous to the time-step within the experiment. Similar time point labels are applied in Figure 3d–f. Two cells with signal translocation from each set of Figure 3a–c and Figure 3d–f have been outlined with circular annotations. The cell marked “s” in Figure 3a–c has been used as an example in Figure 3g–m for providing the data analytics generated in our MATLAB-implemented algorithm with the full set of the 19-point time lapse. The sample shown here in Figure 3g is therefore of a caspase-3 PC9 reporter cell treated with TRAIL, recorded in 15-min intervals, and the time point for each snapshot is labeled atop the respective image, from frame number 1 to 19. The six graphs (Figure 3h–m) show the different image-based data, which the algorithm uses to identify translocation of the signals within cells. The CV and area data graphs are included in Figure 3j,k, respectively, for completeness in regard to classical methods of signal translocation analysis. The SR and its differential dp are plotted in Figure 3l,m, respectively, along with the mean and one significant standard deviation line in Figure 3m (+1.3 std) to correlate the translocation detection criteria as described in Section 2. A significant increase in the dp value with a corresponding drop in the Ncorr value (Figure 3h) gives the first of several criteria (NCorr-dp) for detection of signal translocation, seen at Frame Number, f = 5. Other criteria for change seen in the figure are: Single-Drop (criteria #2) for a large decrement in NCorr (f = 13–14), and a Double-drop (criteria #3) for two consecutive drops in NCorr (f = 13–16). The captured frames identified by the algorithm as translocation are highlighted in vertical bars in the normalized correlation graph (Figure 3h). Note that the first frame identified (f = 5) by the algorithm is considered to be the starting point of signal translocation, regardless of how long the dynamic effects last (up to f = 16). The global correlation value, GCorr, which was calculated as a precursor to Ncorr, is plotted in Figure 3i.

FIGURE 3.

FIGURE 3

Open in a new tab

Algorithm analytics for translocation identification. (a–c) one field of view in time lapse images of TRAIL-treated caspase-3 reporter PC9 cells undergoing signal translocation across the time lapse images. The cell marked “s” is used as an example to break down the algorithm analytics in g–m. (d–f) One field of view in time lapse images of TRAIL-treated caspase-8 reporter PC9 cells undergoing signal translocation. (g) Time lapse of YFP signal in one caspase-3 PC9 reporter cell during drug exposure to TRAIL. Images are labeled as frame numbers, each consecutive frame recorded 15-min apart. (h) and (i) Image-based weighted correlation values, wherein (h) is the local correlation (Ncorr) value, with algorithm-identified frames of translocation events denoted by blue bars; and (i) is the global correlation value (Gcorr), which are values before normalization to obtain Ncorr. (j) Coefficient of variation (CV) and (k) area of the cell (region of interest) in pixels after thresholding the background out. (l) Smoothing ratio (SR) generated from performing a differential smoothing filter over each image frame. The mean value line of the SR is shown for reference. (m) Difference in consecutive frames for SR (dp) from (l), where the first point is defined as zero as there is no time before t = 0. Significant changes along the graph, crossing the threshold (shown here as a+1.3 std. dev. line) are marked with “x”. TRAIL, tumor necrosis factor-related apoptosis-inducing ligand

The evaluation of the algorithm’s correspondence with visual indication is summarized in Table 1. The generated table scores the algorithm at an accuracy of 86%, a precision of 94%, a sensitivity of 88%, and a specificity of 78%. The matrix also highlights the total number of cases presented for each category, namely, (a) 85 samples (53 Cyt-C-GFP and 32 caspase-3/caspase-8) where the observed visual indication matched the algorithm’s analysis to detect translocation of signals correctly (true positives); (b) 18 samples (seven Cyt-C-GFP and 11 caspase-3/caspase-8) visually indicating no signal translocation and correctly identified by the algorithm as such (true negatives); (c) five samples (three Cyt-C-GFP, two caspase-3/caspase-8) where the algorithm identifies a signal translocation which cannot be verified with confidence visually (false positives); (d) 12 samples (seven Cyt-C-GFP, five caspase-3/caspase-8), wherein there are visual cues of signal translocation occur but the algorithm fails to detect them (false negatives). These values were obtained after optimization of adjustable parametric values from training data.

TABLE 1.

Statistical classification table of the algorithm’s performance as an error matrix

Translocation (Algorithm) No detected translocation (Algorithm)
Vision-based (Ground Truth) Translocation 53 32 85 (TP) 7 5 12 (FN)
No translocation (Control samples) 3 2 5 (FP) 7 11 18 (TN)
PC9 Cyt-C Caspase3/Caspase8 Total PC9 Cyt-C Caspase3/Caspase8 Total
Accuracy [(TN+TP)/Total]: 86% Precision [TP/(FP+TP)]: 94%
Sensitivity [TP/(TP+FN)]: 88% Specificity [TN/(TN+FP)]: 78%
Open in a new tab

Abbreviations: FN, false negative; FP, false positive; TN, true negative; TP, true positive.

3.3 |. Real-time translocation of Cyt-C-GFP, caspase-3 sensor, and caspase-8 sensor proteins

Real-time translocation detection across time lapse imaging for PC9 reporter cells are shown in Figures 4a, 4c, and 4e with control conditions (i.e., no translocation in cells without drug treatments in Figures 4b and 4d) using representative samples. The negative samples for Cyt-C-GFP (Figure 4b) and caspase-3 sensor protein (Figure 4d), which lacked any drug induced stimuli, show no detectable change in signal distribution identifiable by the algorithm, and can be verified from a visual perspective. The negative sample for Caspase-8 sensor protein is not shown due to its indistinguishable similarity in cytosolic distribution throughout the cell as the caspase-3 sensor. In a drug-treated sample for Cyt-C-GFP PC9 cells, Figure 4a shows the punctuated Cyt-C-GFP becoming more diffuse over time, being detected by the algorithm starting at the sixth frame (t = 150 min), with significant morphological (and hence signal distribution) changes in the seventh and eighth frames (t = 180, 210 min). This corresponds to the visual indication as can be seen in the time lapse images along the top row in Figure 4a. Figure 4c shows the caspase-3 sensor protein initially excluded from the nucleus for the first four frames, and starts translocating strongly toward the nucleus around the fifth frame (t = 60 min), continuing throughout the next two frames. These visual translocation time points are also consistent to the ones identified by the algorithm, as indicated by the blue bars on the graph. Figure 4e shows a similar translocation of the caspase-8 sensor protein as in the caspase-3 sensor, but over a significantly shorter time period (t = 30 min). Visual indication here may suggest the translocation initiating at t = 15 min, but the algorithm does not validate this event with certainty until the next time frame, at t = 30 min.

FIGURE 4.

FIGURE 4

Open in a new tab

Algorithm-identified cases of translocation from Cyt-C-GFP, caspase-3, and caspase-8 reporter cells. Blue bars on the graph identify time point(s) for the algorithm-based detection of signal translocation. (a and b) Cyto-C-GFP PC9 cells, where (a) treated with TRAIL and (b) without drug treatment. Signal change in (a) detected at 150,180, and 210 min (Frame # 6, 7, and 8). (c and d) Caspase-3 PC9 reporter cells, where (c) treated with TRAIL and (d) without drug treatment. Translocation identified at 60-, 75-, and 90-min frames (Frame # 5, 6, 7. (e) Caspase-8-PC9 reporter cell treated with TRAIL, showing earlier signal translocation at 15–30 min (Frame # 2–3), and later detection at 60, 75, and 90 min (Frame # 5, 6, 7). Time scale: (a and b) 30 min interval per frame. (c–e) 15 min interval per frame. Cyt-C, cytochrome-C; TRAIL, tumor necrosis factor-related apoptosis-inducing ligand

The visual and algorithm-based analyses of translocation starting time are summarized in Figure 5. Cyt-C-GFP PC9 cells show Cyt-C signal diffusion within 2 hr after TRAIL treatment with an average of 2–2.2 hr as identified by both visual cues and algorithm. Caspase-3 activation by TRAIL and FasL detected through YFP sensor translocation to nucleus is noted at an average of about 2 hr in the PC9 and T47D cells, respectively, both by visual and algorithmic analysis. Caspase-8 results have a slight discrepancy between the visual identification and algorithm detection, however, the difference is not significant (p > .05). Visual cues place most of the caspase-8 activation registering as early as 15–45 min, up to 2 hr with an average time of 45 min, while the algorithm identifies most caspase-8 activation with an average of ~1.5 hr.

FIGURE 5.

FIGURE 5

Open in a new tab

Algorithm versus human vision detection of activation time of apoptotic signals. Bar graphs depict the comparative average (±SEM) times identified of signal translocation visually against those detected by the algorithm in caspase-3 (n = 26) and caspase-8 (n = 24) reporter cells (T47D and PC9) as well as Cyt-C-GFP PC9 reporter cells (n = 70). Caspase-3 signal gives an average time detection of 1.86 ± 0.28 hr visually while 1.97 ± 0.31 hr algorithmically. Caspase-8 signal has an average time detection of 0.76 ± 0.23 hr visually and 1.47 ± 0.31 hr algorithmically. Cyt-C-GFP signal is detected on average at 2.05 ± 0.13 hr visually while 2.19 ± 0.14 hr algorithmically. Cyt-C, cytochrome-C

4 |. DISCUSSION

In this study, we developed (a) caspase-8 reporter cell lines in PC9 and T47D cells, and (b) a robust algorithm for tracking of signal translocation, punctuation, or diffusion of the biomarkers (i.e., Cyt-C GFP, caspase-8 reporter, and caspase-3 reporter molecules) in single cells or multiple cells. Incorporation of the reporter caspase-8 plasmid within the cell lines allows tracking of caspase-8 activity live without any other external agent. Commonly used methods in studying apoptosis include assaying methods where (a) multiple fluorophores are used in the assay, typically one for image registration to locate organelles or structures of interest, and another as the biomolecule of interest that translocates; or (b) several samples are employed for a temporal study since each sample serves in an end stage assay wherein they are fixed and stained with dyes. While there are strengths to both approaches, there are drawbacks in recording dynamic data through these methods. For the case with multiple fluorophores, the use of multiple colors restricts the usage of dyes with similar fluorometric signals for recording other activities within the cells. In the case for several samples being used due to the need of sacrificial samples for end-stage assays, some statistical temporal data are generated instead of truly observing dynamic data within the same sample. Having a single reporter biomolecule resolves both these difficulties by providing (a) a single fluorometric marker, such that other dyes of different fluorometric characteristics may be administered without overlapping signals, and, (b) a marker that is expressed and can be monitored live within the cells, generating dynamic data from within the same sample so as to both preserve the sample size and observe true temporal changes within the same sample instead of having a statistical estimation from other samples. We have used nuclear staining for caspase-8 reporter, and mitochondrial (Mitotracker) dye for Cyt-C reporter cells, as secondary dyes in our characterization experiments to provide a comparable metric with the classical methods of analysis. There is, however, no necessity to the use of any additional fluorescent dyes when we use our algorithm for detecting signal translocation within our inherently fluorescent reporter cell lines.

In Section 2 and Supporting Information Material, we elaborate on an image analysis protocol that avoids the general pitfalls assimilated in current image processing algorithms widely used for analyzing punctuation or diffusion of signals produced by biological trackers. We do not focus on image cleanup, as there are several available works on the methods of noise removal and image segmentation, and as such these methods can vary depending on the modality of cell culture chosen (Arslan, Ersahin, Cetin-Atalay, & Gunduz-Demir, 2013; Bardet et al., 2008; Buschhaus, Humphries, Luker, & Luker, 2018; DeCoster, 2007; DeCoster, Sowgandhi, Dutta, & Mcnamara, 2010; Dekaliuk, Pyrshev, & Demchenko, 2015; Eidet, Pasovic, Maria, Jackson, & Utheim, 2014; Filippi-Chiela et al., 2012; Forero, Pennack, Learte, & Hidalgo, 2009; Goldstein, Kluck, & Green, 2000; Helmy & Azim, 2012; Saraste & Pulkki, 2000; Saraste, 1999; Yamaguchi et al., 2011; Ziraldo et al., 2014, 2015). Our focus is strictly on the algorithms, which trace the changes in the cellular biomarker translocation with high precision, sensitivity and accuracy.

While we have discussed the specific shortcomings we chose to address in regard to assay techniques above, another aspect of the data lies in the image analysis process rather than the assay itself. Current methods rely on (a) the incorrect interpretation of statistical measures, or, (b) heavily manual parameters adjusted in proprietary software, which is not easily translated to an automated form of analysis. One such common incorrect use of analytics relies on standard deviation (STD) or CV of the pixel population to analyze the degree of spread of a biomarker signal by collectively applying a pixel-cumulative statistical analysis. This can result in a false interpretation of signal translocation when it truly occurs (or vice versa, of detecting translocation when there is none), thereby giving a significantly delayed response time or not at all. One such example is highlighted in Figure 6. Figure 6a shows a reporter caspase-3 PC9 cell without an apoptotic stimulus over time indicated by frame numbers. Note in the graphs of the analysis that our algorithm (Figure 6b) correctly classifies it as a negative sample and does not highlight any frame as a signal translocation event, while in the CV (Figure 6c) there are big fluctuations, which in the classical approach using STD/CV, may be identified as an apoptotic event. Neither statistical variable (STD nor CV) can account for a change, wherein the intensities of every pixel remains almost the same but the biomarkers simply move around, becoming more disperse or punctuated, since the pixel statistics will remain constant in the lumped STD or CV. In Figure 6d–f, we illustrate a simplified version of this viewpoint, with the larger red box as representative of a cell and the black squares as representative of fluorescent biomarker signals. All three images have the same population statistics because four black squares out of 16 squares result in the same mean, STD, and CV. However, it is clear from each image that it represents distinctly different distributions, with Figure 6d being somewhat random, Figure 6e with a more centralized signal, and Figure 6f with the signals localized to one side of the cell. Unless the biomarkers overlap and increase/decrease the intensity of individual pixels, simple translocation will be misrepresented in STD/CV values. Spatial information therefore must be included in any algorithm that analyzes translocation of signals within cells (or within an ROI).

FIGURE 6.

FIGURE 6

Open in a new tab

Examples of misinterpretation of STD/CV values. (a) Time lapse images of a capase-3 PC9 reporter cell under control conditions (no drug treatment). Numbers atop each image indicate the sequential frame number, at 15 min intervals. The local correlation graph is plotted in (b), while the CV is plotted in (c). (d–f) Example of three different distributions having the same STD/CV. The red border is representative of a region of interest, such as a cell. The black squares represent possible biomarker location, white squares indicate empty areas

By developing an algorithm that can extract features in a manner, which gives a robust analysis of translocation of biomarker signals, we provide a software tool that circumvents the deficiencies of the traditional methods of analysis. Making the feature extraction dependent on both the location and intensities of the pixel values in an image produces analytics that are sensitive to translocation of signals but is not easily misinterpreted as in the classical STD and CV parameters. The algorithm also utilizes statistical parameters of the extracted features themselves as additional tunable parameters to identify translocating signals. This bypasses the otherwise heavily manual procedure of tweaking image processing variables to generate such data by passing the criteria to statistically trained parameters. As shown in Figure 3, time points caught by the algorithm within the NCorr values may result in multiple identified points of signal translocation (blue bars). For our purposes of analysis, we use the first identified time point (within a contiguous set of time points) as the one recorded for the initiation of signal translocation. Multiple sets of contiguously identified translocation points can arise, especially when the cells go through significant morphological structural change as a natural path in the apoptosis process. In our apoptotic samples, we find that the first contiguous set of time points correspond to signal translocation, while the second set corresponds to cellular blebbing or shrinking, morphological features which are typically associated with the late stage of apoptosis. It should be noted that the order of scale (from half an hour to a few hours) of observed and algorithmically detected caspase-8, caspase-3, and Cyt-C-GFP translocation activity corresponds to that found in literature, and that on average, caspase-8 activity occurs upstream on the temporal scale before caspase-3 corresponds to the biological extrinsic pathway that is well-known (A. Antonsson & Persson, 2009; Bardet et al., 2008; Buschhaus et al., 2018; Goldstein, Munoz-Pinedo, et al., 2005; McIlwain, Berger, & Mak, 2013; Scaffidi et al., 1998; Sundquist et al., 2006). The large window of variation occurs due to the inherent heterogeneity on a cell-to-cell basis.

Our method of analysis is automated based on the image data it is fed, but allowance for manual adjustment of the parameters used for translocation identification is included to provide flexibility if a stricter or weaker criteria is desired. Using a single fluorophore-based analysis with images taken at the wide field microscopy resolution, we developed a procedure that lends itself to high-throughput of samples, which is essential in processing large amounts of data from drug screening trials. In dynamic studies of apoptosis, having a simple intrinsic signal alleviates the need to use additional dyes or multiple sacrificial samples for every time point, thereby reducing the (a) cost of use for additional dyes, (b) imaging time for additional dyes, (c) cost of maintenance and imaging of several sacrificial samples by monitoring the live dynamics of a smaller sample size over time. The algorithm itself provides the robust, parametrically tunable approach to automate the process of identifying translocating (apoptotic) signals without much computational complexity. Our end products of this study (i.e., both the cell lines and algorithm) can streamline and handle automated accurate interpretation of apoptotic events in high-throughput cell-based drug screening platforms.

Supplementary Material

bit27280-sup-0001-supporting_information_ii
bit27280-sup-0001-supporting_information_i

ACKNOWLEDGMENTS

This study was partially supported by NIH/NCI U54 CA 132378 (CCNY MSKCC Partnership) to X. J. and S. W., NSF CBET 1055608 to S.W. and Young Investigator Prize to S.W. from Pershing Square Sohn Cancer Research Alliance.

Footnotes

CONFLICT OF INTERESTS

The authors declare that there are no conflict of interests.

SUPPORTING INFORMATION

Additional supporting information may be found online in the Supporting Information section.

REFERENCES

  1. Antonsson A, & Persson JL (2009). Induction of apoptosis by staurosporine involves the inhibition of expression of the major cell cycle proteins at the G(2)/m checkpoint accompanied by alterations in Erk and Akt kinase activities. Anticancer Research, 29(8), 2893–2898. [PubMed] [Google Scholar]
  2. Antonsson B, Montessuit S, Sanchez B, & Martinou JC (2001). Bax is present as a high molecular weight oligomer/complex in the mitochondrial membrane of apoptotic cells. Journal of Biological Chemistry, 276(15), 11615–11623. 10.1074/jbc.M010810200 [DOI] [PubMed] [Google Scholar]
  3. Arslan S, Ersahin T, Cetin-Atalay R, & Gunduz-Demir C (2013). Attributed relational graphs for cell nucleus segmentation in fluorescence microscopy images. IEEE Transactions on Medical Imaging, 32(6), 1121–1131. 10.1109/tmi.2013.2255309 [DOI] [PubMed] [Google Scholar]
  4. Ashkenazi A, & Dixit VM (1998). Death receptors: Signaling and modulation. Science, 281(5381), 1305–1308. [DOI] [PubMed] [Google Scholar]
  5. Bardet PL, Kolahgar G, Mynett A, Miguel-Aliaga I, Briscoe J, Meier P, & Vincent JP (2008). A fluorescent reporter of caspase activity for live imaging. Proceedings of the National Academy of Sciences of the United States of America, 105(37), 13901–13905. 10.1073/pnas.0806983105 [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Bouchier-Hayes L, Muñoz-Pinedo C, Connell S, & Green DR (2008). Measuring apoptosis at the single cell level. Methods, 44(3), 222–228. 10.1016/j.ymeth.2007.11.007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Buschhaus JM, Humphries B, Luker KE, & Luker GD (2018). A caspase-3 reporter for fluorescence lifetime imaging of single-cell apoptosis. Cells, 7(6), 57 10.3390/cells7060057 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Carpenter AE, Jones TR, Lamprecht MR, Clarke C, Kang IH, Friman O, … Sabatini DM (2006). CellProfiler: Image analysis software for identifying and quantifying cell phenotypes. Genome Biology, 7(10), R100 10.1186/gb-2006-7-10-r100 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Chatterjee K, Zhang J, Honbo N, & Karliner JS (2010). Doxorubicin cardiomyopathy. Cardiology, 115(2), 155–162. 10.1159/000265166 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. DeCoster M (2007). The Nuclear Area Factor (NAF): A measure for cell apoptosis using microscopy and image analysis. Modern Research and Educational Topics in Microscopy, 1, 378–384. [Google Scholar]
  11. DeCoster MM, Maddi S, Dutta V, & Mcnamara J (2010). Microscopy and image analysis of individual and group cell shape changes during apoptosis. Microscopy: Science, Technology, Applications and Education, 2(4), 836–843. [Google Scholar]
  12. Dekaliuk M, Pyrshev K, & Demchenko A (2015). Visualization and detection of live and apoptotic cells with fluorescent carbon nanoparticles. Journal of Nanobiotechnology, 13(1), 86 10.1186/s12951-015-0148-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Dereli-Korkut Z, Gandhok H, Zeng LG, Waqas S, Jiang X, & Wang S (2013). Real-time detection of cellular death receptor-4 activation by fluorescence resonance energy transfer. Biotechnology and Bioengineering, 110(5), 1396–1404. 10.1002/bit.24804 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Eidet JR, Pasovic L, Maria R, Jackson CJ, & Utheim TP (2014). Objective assessment of changes in nuclear morphology and cell distribution following induction of apoptosis. Diagnostic Pathology, 9, 92 10.1186/1746-1596-9-92 [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Filippi-Chiela EC, Oliveira MM, Jurkovski B, Callegari-Jacques SM, da Silva VD, & Lenz G (2012). Nuclear morphometric analysis (NMA): Screening of senescence, apoptosis and nuclear irregularities. PLoS One, 7(8):e42522 10.1371/journal.pone.0042522 [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Forero MG, Pennack JA, Learte AR, & Hidalgo A (2009). DeadEasy caspase: Automatic counting of apoptotic cells in Drosophila. PLoS One, 4(5):e5441 10.1371/journal.pone.0005441 [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Gelles JD, & Chipuk JE (2016). Robust high-throughput kinetic analysis of apoptosis with real-time high-content live-cell imaging. Cell Death & Disease, 7(12):e2493 10.1038/cddis.2016.332 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Goldstein JC, Kluck RM, & Green DR (2000). A single cell analysis of apoptosis. Ordering the apoptotic phenotype. Annals of the New York Academy of Sciences, 926, 132–141. [DOI] [PubMed] [Google Scholar]
  19. Goldstein JC, Waterhouse NJ, Juin P, Evan GI, & Green DR (2000). The coordinate release of cytochrome c during apoptosis is rapid, complete and kinetically invariant. Nature Cell Biology, 2(3), 156–162. 10.1038/35004029 [DOI] [PubMed] [Google Scholar]
  20. Goldstein JC, Munoz-Pinedo C, Ricci JE, Adams SR, Kelekar A, Schuler M, … Green DR (2005). Cytochrome c is released in a single step during apoptosis. Cell Death and Differentiation, 12(5), 453–462. 10.1038/sj.cdd.4401596 [DOI] [PubMed] [Google Scholar]
  21. Goldstein JC, Muñoz-Pinedo C, Ricci JE, Adams SR, Kelekar A, Schuler M, … Green DR (2005). Cytochrome c is released in a single step during apoptosis. Cell Death and Differentiation, 12, 453–462. 10.1038/sj.cdd.4401596 [DOI] [PubMed] [Google Scholar]
  22. Heiskanen KM, Bhat MB, Wang HW, Ma J, & Nieminen AL (1999). Mitochondrial depolarization accompanies cytochrome c release during apoptosis in PC6 cells. Journal of Biological Chemistry, 274(9), 5654–5658. [DOI] [PubMed] [Google Scholar]
  23. Helmy IM, & Azim AM (2012). Efficacy of ImageJ in the assessment of apoptosis. Diagnostic Pathology, 7, 15 10.1186/1746-1596-7-15 [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Jiang X, & Wang X (2004). Cytochrome C-mediated apoptosis. Annual Review of Biochemistry, 73, 87–106. 10.1146/annurev.biochem.73.011303.073706 [DOI] [PubMed] [Google Scholar]
  25. Kamada S, Kikkawa U, Tsujimoto Y, & Hunter T (2005). Nuclear translocation of caspase-3 is dependent on its proteolytic activation and recognition of a substrate-like protein(s). Journal of Biological Chemistry, 280(2), 857–860. 10.1074/jbc.C400538200 [DOI] [PubMed] [Google Scholar]
  26. Lamprecht MR, Sabatini DM, & Carpenter AE (2007). CellProfiler: Free, versatile software for automated biological image analysis. Biotechniques, 42(1), 71–75. 10.2144/000112257 [DOI] [PubMed] [Google Scholar]
  27. Leibowitz B, & Yu J (2010). Mitochondrial signaling in cell death via the Bcl-2 family. Cancer Biology & Therapy, 9(6), 417–422. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Luetjens CM, Kogel D, Reimertz C, Dussmann H, Renz A, Schulze-Osthoff K, … Prehn JH (2001). Multiple kinetics of mitochondrial cytochrome c release in drug-induced apoptosis. Molecular Pharmacology, 60(5), 1008–1019. [DOI] [PubMed] [Google Scholar]
  29. MacFarlane M (2003). TRAIL-induced signalling and apoptosis. Toxicology Letters, 139(2–3), 89–97. [DOI] [PubMed] [Google Scholar]
  30. Martinou J-C, Desagher S, & Antonsson B (2000). Cytochrome c release from mitochondria: All or nothing. Nature Cell Biology, 2, E41–E43. 10.1038/35004069 [DOI] [PubMed] [Google Scholar]
  31. McIlwain DR, Berger T, & Mak TW (2013). Caspase functions in cell death and disease. Cold Spring Harbor Perspectives in Biology, 5(4), a008656 10.1101/cshperspect.a008656 [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Nagata S (1997). Apoptosis by death factor. Cell, 88(3), 355–365. [DOI] [PubMed] [Google Scholar]
  33. Nagata S (1999). Fas ligand-induced apoptosis. Annual Review of Genetics, 33(1), 29–55. 10.1146/annurev.genet.33.1.29 [DOI] [PubMed] [Google Scholar]
  34. Rehm M, Huber HJ, Hellwig CT, Anguissola S, Dussmann H, & Prehn JH (2009). Dynamics of outer mitochondrial membrane permeabilization during apoptosis. Cell Death and Differentiation, 16(4), 613–623. 10.1038/cdd.2008.187 [DOI] [PubMed] [Google Scholar]
  35. Rheingold SR, Neugut AI, & Meadows AT (2003). Holland-Frei Cancer Medicine In Kufe DWPR & Weichselbaum RR (Eds.), Holland-Frei Cancer Medicine (2, 6). Hamilton (ON): BC Decker; https://www.ncbi.nlm.nih.gov/books/NBK13212/ [Google Scholar]
  36. Saraste A (1999). Morphologic criteria and detection of apoptosis. Herz, 24(3), 189–195. [DOI] [PubMed] [Google Scholar]
  37. Saraste A, & Pulkki K (2000). Morphologic and biochemical hallmarks of apoptosis. Cardiovascular Research, 45(3), 528–537. [DOI] [PubMed] [Google Scholar]
  38. Scaffidi C, Fulda S, Srinivasan A, Friesen C, Li F, Tomaselli KJ, … Peter ME (1998). Two CD95 (APO-1/Fas) signaling pathways. EMBO Journal, 17(6), 1675–1687. 10.1093/emboj/17.6.1675 [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Srinivasula SM, Datta P, Fan XJ, Fernandes-Alnemri T, Huang Z, & Alnemri ES (2000). Molecular determinants of the caspase-promoting activity of Smac/DIABLO and its role in the death receptor pathway. Journal of Biological Chemistry, 275(46), 36152–36157. 10.1074/jbc.C000533200 [DOI] [PubMed] [Google Scholar]
  40. Suliman A, Lam A, Datta R, & Srivastava RK (2001). Intracellular mechanisms of TRAIL: Apoptosis through mitochondrial-dependent and -independent pathways. Oncogene, 20(17), 2122–2133. 10.1038/sj.onc.1204282 [DOI] [PubMed] [Google Scholar]
  41. Sundquist T, Moravec R, Niles A, O’Brien M, & Riss T (2006). Timing your apoptosis assays. Cell Notes, 16, 18–21. [Google Scholar]
  42. Waterhouse NJ, Goldstein JC, von Ahsen O, Schuler M, Newmeyer DD, & Green DR (2001). Cytochrome C maintains mitochondrial transmembrane potential and Atp generation after outer mitochondrial membrane permeabilization during the apoptotic process. The Journal of Cell Biology, 153(2), 319–328. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Xiao A, Gibbons AE, Luker KE, & Luker GD (2015). Fluorescence lifetime imaging of apoptosis. Tomography, 1(2), 115–124. 10.18383/j.tom.2015.00163 [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Yamaguchi Y, Shinotsuka N, Nonomura K, Takemoto K, Kuida K, Yosida H, & Miura M (2011). Live imaging of apoptosis in a novel transgenic mouse highlights its role in neural tube closure. The Journal of Cell Biology, 195(6), 1047–1060. 10.1083/jcb.201104057 [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Ziraldo R, Link N, Abrams J, & Ma L (2014). Automatic image analysis of multicellular apoptosis process. Paper presented at the 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society. [DOI] [PubMed] [Google Scholar]
  46. Ziraldo R, Link N, Abrams J, & Ma L (2015). Towards automatic image analysis and assessment of the multicellular apoptosis process. IET Image Processing/IET, 9(5), 424–433. 10.1049/iet-ipr.2014.0531 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

bit27280-sup-0001-supporting_information_ii
NIHMS1648191-supplement-bit27280-sup-0001-supporting_information_ii.pdf (167KB, pdf)
bit27280-sup-0001-supporting_information_i
NIHMS1648191-supplement-bit27280-sup-0001-supporting_information_i.pdf (290.6KB, pdf)

RESOURCES