|Home | About | Journals | Submit | Contact Us | Français|
Apoptosis in response to TRAIL or TNF requires the activation of initiator caspases, which then activate the effector caspases that dismantle cells and cause death. However, little is known about the dynamics and regulatory logic linking initiators and effectors. Using a combination of live-cell reporters, flow cytometry, and immunoblotting, we find that initiator caspases are active during the long and variable delay that precedes mitochondrial outer membrane permeabilization (MOMP) and effector caspase activation. When combined with a mathematical model of core apoptosis pathways, experimental perturbation of regulatory links between initiator and effector caspases reveals that XIAP and proteasome-dependent degradation of effector caspases are important in restraining activity during the pre-MOMP delay. We identify conditions in which restraint is impaired, creating a physiologically indeterminate state of partial cell death with the potential to generate genomic instability. Together, these findings provide a quantitative picture of caspase regulatory networks and their failure modes.
Activation of initiator and effector caspases is a defining characteristic of apoptosis. Effector caspases cleave essential cellular substrates and directly dismantle cells, while initiator caspases have a more limited range of substrates and act primarily to regulate effector caspases (Fuentes-Prior and Salvesen, 2004). The precise, all-or-none control of caspase activation is critically important for control of cell fate: effector caspases must not turn on prematurely, but once active they must fully cleave their substrates and provoke cell death, lest cells survive with partially digested cellular contents and damaged genomes (Vaughan et al., 2002). Single-cell studies show that effector caspases are activated minutes before death, but many hours after exposure to a death ligand (Goldstein et al., 2000; Rehm et al., 2002; Tyas et al., 2000). Little is known about the state of caspase regulatory networks during the long delay between receptor engagement and the induction of cell death, a significant gap in our understanding of the regulatory logic of apoptosis.
The chain of events that initiates extrinsic cell death begins with ligand-induced assembly of death-inducing signaling complexes (DISCs) on TNF, Fas, or TRAIL receptors (Kischkel et al., 1995; Martin et al., 1998). The initiator caspases-8 and -10 are activated by enforced dimerization at DISCs (Boatright et al., 2003; Donepudi et al., 2003), after which they cleave effector pro-caspases, causing a >100-fold increase in effector activity (Bose et al., 2003; Han et al., 1997). In the case of procaspase-3, activating proteolysis occurs at Asp175, separating the enzyme into large and small subunits that remain physically associated (Fuentes-Prior and Salvesen, 2004). Similar biochemistry is observed for intrinsic apoptosis following intracellular injury, except that caspase-9 is typically the initiator (Srinivasula et al., 1998). Effector caspases were once thought to auto-activate via homotypic procaspase cleavage, but recent work shows that this is not the case: the activation of procaspase-3 by proteolysis is mediated exclusively by initiator caspases (Liu et al., 2005). Subsequent to cleavage by initiators, effector caspases are regulated by trans-acting factors such as X-linked inhibitor of apoptosis protein (XIAP), which blocks the proteolytic activity of caspase-3 by binding tightly to its active site (Huang et al., 2001). XIAP may also encode an E3 ubiquitin ligase that promotes caspase-3 ubiquitination and its subsequent proteasome-mediated degradation (Chen et al., 2003; Suzuki et al., 2001).
In most cells (those with type-II regulation (Scaffidi et al., 1998)), XIAP-mediated inhibition of caspase-3 is relieved by a pathway that involves mitochondrial outer membrane permeabilization (MOMP; (Deng et al., 2002; Li et al., 2002; Sun et al., 2002; Zhang et al., 2001). MOMP is regulated by the Bcl-2 family of proteins, which function either as pro-apoptotic (e.g. Bid or Bax) or anti-apoptotic (e.g. Bcl-2) factors (Kim et al., 2006). Initiator caspases directly cleave Bid, which then activates Bax by altering its conformation (Eskes et al., 2000; Luo et al., 1998). Activated Bax translocates to the mitochondria, forming pores that allow Smac/Diablo and cytochrome c to translocate from their normal locations in the mitochondrial inter-membrane space to the cytosol (Munoz-Pinedo et al., 2006; Oltvai et al., 1993). Smac binds tightly to XIAP, blocking XIAP's caspase-3-inhibitory activity (Du et al., 2000; Verhagen et al., 2000), while cytochrome c binds Apaf-1 and caspase-9 to form the apoptosome. Finally, in at least some cells, active caspase-3 cleaves procaspase-6, which then generates additional caspase-8 in a feedback loop (Cowling and Downward, 2002; Murphy et al., 2004). Thus, the dynamics of extrinsic apoptosis are shaped to a lesser or greater degree by three cooperating but asynchronous processes: direct cleavage of effector caspases by initiators, cytosolic translocation of pro-apoptotic mitochondrial proteins, and feedback from effector to initiator caspases.
Live-cell reporters of caspase activity and MOMP reveal considerable heterogeneity in the dynamics of apoptosis from cell to cell within a population that is nominally genetically homogeneous (Goldstein et al., 2000; Munoz-Pinedo et al., 2006; Rehm et al., 2002; Rehm et al., 2003; Takemoto et al., 2003; Tyas et al., 2000). Much of the heterogeneity is due to the variable delay, many hours in duration, between exposure to a pro-apoptotic stimulus and the sudden initiation of MOMP. This delay cannot be attributed solely to slow transcriptional processes, since it is observed even in the presence of drugs that block ongoing transcription or translation (Goldstein et al., 2000; Rehm et al., 2002). In this study we examine the relative dynamics of initiator and effector caspase activation and MOMP using four experimental tools: (i) live-cell reporters specific to initiator and effector caspases; (ii) a biologically inert reporter of MOMP; (iii) flow cytometry and immunoblot analysis of endogenous substrates; and (iv) perturbation of specific proteins using RNAi, protein over-expression and small molecule drugs. Analysis of initiator and effector caspase dynamics with a newly developed mathematical model of extrinsic apoptosis reveals a long-lived cellular state in which cleaved effector caspases are held safely in check while initiator caspases are active; failure to sustain this state makes the commitment to apoptosis ambiguous, and yields damaged but undead cells with partly proteolyzed cellular contents.
To monitor caspase regulation in single living cells, we constructed three fluorescent protein (FP) fusions (Fig. 1A, B). The first, effector caspase reporter protein (EC-RP), monitors caspase-3 activity (and, to a lesser extent, caspase-7 activity) and is composed of a Förster resonance energy transfer (FRET) donor-acceptor pair (CFP and YFP) connected via a flexible linker that contains the caspase cleavage sequence DEVDR (Fig. 1B). When the linker is cleaved, energy transfer is lost and the CFP signal increases, an event that can be monitored by live-cell microscopy. The DEVDR linker in EC-RP is expected to have 20-fold greater selectivity for caspase-3 relative to caspase-8 than the DEVDG linker used in previously published caspase reporters (Rehm et al., 2002; Stennicke et al., 2000; Tyas et al., 2000), based on in vitro data showing DEVDR to reduce kcat/Km for caspase-8 ~300-fold, but kcat/Km for caspase-3 only ~14-fold relative to DEVDG (Stennicke et al., 2000).
Initiator caspase reporter protein (IC-RP) carries tandem copies of IETD in its linker, a sequence that is efficiently cleaved by caspase-8 (Luo et al., 2003), but poorly by caspases-3,7 (Thornberry et al., 1997). IETD constitutes the site in procaspase-3 for initiator caspase cleavage, and IC-RP cleavage is therefore a good readout of procaspase-3 activation. Finally, a reporter for MOMP that localizes to the inter-membrane space (IMS-RP) was created by fusing RFP to the mitochondrial import sequence of Smac (residues 1- 59) (Du et al., 2000). FP fusions to full-length cytochrome c and Smac have been described previously (Goldstein et al., 2000; Munoz-Pinedo et al., 2006; Rehm et al., 2003), but IMS-RP differs from these fusions in lacking an IAP-binding motif, and it is therefore biochemically inactive.
To validate the properties of EC-RP, IC-RP, and IMS-RP in vivo, HeLa cells stably expressing the reporter proteins were treated with TRAIL and cycloheximide (CHX) and fluorescence signals monitored every 3 min over an 8-12 hr period. IMS-RP distribution was monitored using an image-processing algorithm that detects shifts from punctuate mitochondrial to diffuse cytosolic fluorescence. When cells were treated with varying doses of TRAIL, IMS-RP relocalized at the same time as a co-expressed Smac-CFP fusion (Karbowski et al., 2004) and ~6-9 min prior to the appearance of apoptotic cellular morphology (Fig. 1C and Supplemental Movie 1). IMS-RP translocation was blocked in TRAIL-treated cells by RNAi-mediated depletion of caspase-8 and Bid, upstream components of the extrinsic cell death pathway, but not by RNAi of downstream components such as Smac (Fig. 1D). Thus, IMS-RP appears to be a faithful reporter of protein translocation at MOMP, a process whose dynamics differs from that of falling mitochondrial membrane potential as measured using dyes (Munoz-Pinedo et al., 2006).
Caspase-mediated proteolysis of EC-RP and IC-RP was monitored by calculating the ratio of CFP and YFP emission, with suitable correction for background (see Materials and Methods). Due to spectral overlap, it was not possible to monitor EC-RP, IC-RP, and IMS-RP fluorescence simultaneously in single cells, and we therefore expressed the reporters in pairs. When cells co-expressing EC-RP and IMS-RP were treated with TRAIL, increases in the CFP/YFP ratio resulting from reporter cleavage were sudden and took place only after IMS-RP translocation (Fig. 1E; note that EC-RP signals were typically lost when cells detached from the slide subsequent to the appearance of apoptotic morphology). EC-RP cleavage was reduced 20-fold by RNAi of caspase-8 and 5-fold by RNAi of Bid, consistent with a requirement for MOMP in caspase-3 activation. Control experiments also established that changes in EC-RP fluorescence required TRAIL (CHX alone had no effect) and were sequence-specific (being absent with a non-cleavable DEVG-carrying reporter; Fig. S1). However, changes in cell in morphology did alter apparent FRET signals and were therefore taken into account during image analysis (see supplementary materials).
Cells co-expressing IC-RP and IMS-RP exhibited gradual increases in IC-RP signal subsequent to TRAIL treatment, and rapid increases subsequent to MOMP (IMS-RP release; Fig. 1F). The early, gradual phase of IC-RP cleavage was insensitive to Bid depletion but the fast post-MOMP phase was eliminated by it; both were blocked by depletion of caspase-8. Rapid IC-RP cleavage post-MOMP probably reflects either elevated caspase-8 activity resulting from feedback via caspase-6 or cleavage by caspase-9 (RNAi experiments were ambiguous with respect to these possibilities but the IETD sequence in IC-RP is known to be a good caspase-9 substrate). We therefore conclude that IC-RP is a selective readout of caspase-8 activity prior to MOMP but is unlikely to retain this selectivity later on; however, it is complemented by EC-RP, which is the key readout of the post-MOMP state. Direct comparison with a traditional reporter carrying a DEVDG caspase-3 recognition sequence further demonstrates the utility of the new EC-RP reporter: unlike EC-RP, or endogenous caspase substrates (see below), cleavage of the traditional reporter occurred prior to MOMP (as monitored by IMS-RP release) and was only partially blocked by Bid depletion (Fig. 1G).
To derive a quantitative picture of caspase activation dynamics, the timing of cleaved IC-RP and EC-RP accumulation and of IMS-RP release from mitochondria were measured in 50-100 cells relative to the time of TRAIL addition (defined as t = 0; Fig. 2A,B). Considerable cell-to-cell variability was observed in the time of IMS-RP release and the onset of EC-RP cleavage (from t ~1.5 to 4.5 hr) consistent with previous reports (Goldstein et al., 2000; Rehm et al., 2002). However, when live-cell movies were aligned based on the time of IMS-RP release, a clear picture of the progression in caspase activities emerged (Fig. 2C,D, E). In all cells, IC-RP cleavage rose slowly soon after TRAIL addition to ~ 50% of its maximum level at the time of IMS-RP release, after which the rate of cleavage increased substantially. At 50 ng/ml TRAIL, the pre-MOMP delay averaged 3.2 hr, but could be as long as 12-24 hr at lower doses of TRAIL (Fig. 2A,B and unpublished data). In contrast, prior to IMS-RP release, the rate of EC-RP cleavage was very low, remaining at <5% of its maximum value, and then rising rapidly over a 10-20 min period coincident with membrane blebbing and cell shrinkage, morphological manifestations of cell death (Fig. 2A,C). Thus, subsequent to TRAIL addition but prior to MOMP (IMS-RP release), cells enter a “delay” state lasting several hours, during which initiator but not effector caspases are active (Fig. 2E). The delay was even more pronounced in cells treated with TRAIL in the absence of CHX: in this case, IC-RP cleavage continued for longer and rose to substantially higher levels prior to IMS-RP release (~75% of maximum signal; Fig. 2F). Moreover, delay was not restricted to signaling downstream of TRAIL receptors: even longer delays were observed in cells treated with TNF (data not shown).
As assays for caspase activity, transfected reporter proteins have the drawback that they contain artificial cleavage sites and are over-expressed. To confirm that caspase dynamics measured with FPs were representative of endogenous caspase substrates, we turned to flow cytometry. Cleavage of initiator caspase substrates (designated IC substrates) was monitored using an antibody specific to a fragment of caspase-3 cleaved at Asp175 (the final residue in the IETD caspase-8 recognition site (Urase et al., 1998), and cleavage of effector caspase substrates (EC substrates) was monitored using an antibody specific to cleavage of PARP at Asp214 by caspases-3,7 (Tewari et al., 1995). When cells harvested 0-4.5 hr after TRAIL addition were stained with both IC and EC cleavage-specific antibodies and examined by 2D flow cytometry, a broad distribution spanning ~102 fluorescence units was observed for IC substrate cleavage, whereas EC substrate cleavage was bimodal with clear negative (non-staining) and positive (antibody-staining) peaks (see frequency distributions above and to the right of the scatter plots; Fig. 3A). Discretization of the data into low, intermediate, and high levels revealed intermediate EC substrate cleavage to occur in fewer than 5% of cells whereas intermediate IC substrate cleavage occurred in up to ~35% of cells (Fig. 3B); however, the state of intermediate IC substrate cleavage was transient, eventually progressing to complete cleavage in all cells.
Comparison of flow cytometry and live-cell data was facilitated by plotting EC-RP and IC-RP fluorescence on a scatter plot (Fig. 3C; note that, because both EC-RP and IC-RP signals increase monotonically, cells progress in time from lower left to upper right quadrants). A non-linear input-output relationship between initiator and effector caspases was clearly evident regardless of the assay: imaging and flow cytometry both showed that EC substrate cleavage was low following TRAIL treatment but that IC substrate cleavage increased steadily until a threshold value was reached (red lines in Fig. 3), at which point both IC and EC substrates were rapidly and fully processed (Fig. 3C). Imaging showed that the transition to rapid cleavage coincided with IMS-RP release, and both assays agreed that the level of IC substrate cleavage at the MOMP transition was 1.5- to 2-fold higher in cells treated with TRAIL alone than in cells treated with both TRAIL and CHX (Supplemental Fig. 2).
As a final way to characterize the relationship between initiator and effector caspases, we used immunoblotting to examine procaspase-8, procaspase-3, and PARP cleavage following exposure to TRAIL. To make population-level measurements of pre-MOMP states possible, MOMP was blocked by over-expression of Bcl-2, a potent negative regulator of pore formation (Fig. S3). Under normal conditions, >90% PARP was cleaved within 4.5 hr after TRAIL addition, but overexpression of Bcl2 reduced PARP cleavage to <5% at 4.5 hr (Fig. 3D). Nonetheless, procaspase-3 was processed with similar efficiency in both cases (Fig. 3D), consistent with the kinetics of procaspase-8 cleavage and activation, which also proceeded to completion within 6 hr (Fig. 3E). Thus, immunoblotting confirms the existence of a pre-MOMP state in which procaspase-8 processing leads to cleavage of procaspase-3, but the resulting cleaved caspase-3 does not proteolyze substrates such as PARP or cytokeratin.
The absence of caspase-3 activity during the post-TRAIL, pre-MOMP waiting period implies the action of a potent inhibitor such as XIAP. We therefore expected that, in cells prevented from undergoing MOMP (by RNAi of Bid or overexpression of Bcl-2), depletion of XIAP would relieve caspase-3 inhibition. EC substrate cleavage was indeed restored under these conditions, but with a significant difference: cleavage of EC-RP lacked the delay period and sudden onset observed in control cells, and instead rose gradually soon after TRAIL addition (Fig. 4A). By flow cytometry, a breakdown in the normal input-output relationship between initiator and effector caspases was observed, with intermediate levels of EC substrate cleavage apparent in ~35% of cells (Fig. 4B). Moreover, the extent of EC substrate cleavage did not increase between 4.5 and 24 hr, indicating that it had reached a steady state of 10-20% processing (Fig. 4C; this was more easily observed in the case of Bcl-2 overexpression than Bid depletion, presumably due to greater cell-to-cell uniformity in the former). The absence of ongoing EC substrate cleavage under these conditions was also confirmed by quantitation of cPARP levels on immunoblots (Fig. 4D). Under these circumstances, the effects of XIAP depletion were found to be dose-dependent: a second siRNA oligo that more efficiently reduced XIAP levels (17-fold vs. 8-fold) increased the extent of EC-substrate cleavage (Fig. 4E, F). Nonetheless, a sustained state of partial substrate cleavage, corresponding to ~50% PARP degradation, was still observed (Fig. 4F).
To determine the consequences of partial EC substrate cleavage for cell fate, cells at t = 4.5 (with ~8-fold depletion of XIAP) were replated into fresh medium and the fraction of survivors found to be ~45% (Fig. 4G,H; with ~17-fold depletion, ~25% of cells survived). We would expect such cells to suffer DNA damage via DFF40/Caspase-activated DNAse (CAD) (Samejima and Earnshaw, 2005), and increased levels of the DNA damage marker phospho-histone H2AX accompanied partial PARP cleavage (data not shown). However, we were unable to monitor DNA damage at the single cell-level and could not distinguish whether surviving cells (as opposed to those that die) sustain genomic damage. Nonetheless, it is clear that simultaneous disruption of MOMP and depletion of XIAP causes a dramatic failure in normal all-or-none commitment to cell death.
The observation that EC substrates can be cleaved to a steady sub-maximal level is surprising given the efficiency of caspase-3 as a protease. To understand how partial cleavage might arise and then persist, we turned to a mathematical model we had recently developed to describe reactions between TRAIL-receptor binding and cleavage of effector substrates (Albeck et al., unpublished findings); the model includes (i) the direct action of caspase-8 on caspase-3, (ii) induction of MOMP via competition among pro- and anti-apoptotic Bcl-2 family proteins, and (iii) caspase-6-mediated feedback between caspase-3 and caspase-8. The biochemistry of these processes is represented in the model by a set of elementary reactions that are cast as ordinary differential equations with parameter values derived from the literature or estimated by fitting to experimental data. Overall, the model accurately recapitulates the dynamics of extrinsic apoptosis in unperturbed and RNAi-treated single cells (Albeck et al., unpublished findings; see Supplementary Information for details of model construction and validation). Model-based simulation serves as an effective tool for investigating quantitative relationships among cooperating and competing regulatory processes under various experimental conditions. By simulation we found that even a tiny fraction of the procaspase-3 present in cells (1%, or ~102 - 103 molecules; (Stennicke et al., 1998) is sufficient to fully process EC substrates in several hours (Fig. S4). However, when XIAP is present at normal levels in the model (corresponding to the optimum fitted value), caspase-3 catalytic activity is effectively restrained (Fig. 5A). Critically, the model shows that partial EC substrate cleavage is absolutely dependent on the inactivation of caspase-3 soon after its production by initiator caspases; we had modeled this inactivation as proteasome- and ubiquitin-dependent based on previous reports (Chen et al., 2003; Suzuki et al., 2001) but without knowing its physiological significance. The onset of MOMP normally causes Smac to be released from mitochondria, sequestering XIAP and generating a pulse of caspase-3 far in excess of that required to cleave all EC substrates in the cell, regardless of whether or not caspase-3 is degraded (Fig. 5A,B). When Bcl-2 is over-expressed or Bid depleted, thereby blocking MOMP, simulation shows that XIAP is present in excess over active caspase-3, and very little EC substrate is cleaved. However, when XIAP levels are reduced so as to simulate RNAi-mediated protein depletion, the direct action of caspase-8 on procaspase-3 serves as a limited source of caspase-3, which is then ubiquitinated and degraded. As a consequence, a pulse of active caspase-3 is generated whose limited duration results in cumulative caspase activity sufficient to cleave only a subset of the EC substrates present in a cell (~25% in the simulation in Fig. 5A, B). The involvement, in simulation, of caspase-3 degradation in the pre-MOMP delay and partial EC substrate cleavage (Fig. 5A,B) leads to two experimentally testable predictions: (i) when MOMP is blocked in TRAIL-treated cells, active caspase-3 should not accumulate due to ongoing proteasome-mediated degradation, but should be stabilized in the presence of a proteasome inhibitor, and (ii) increasing proteasome inhibition in pre-MOMP cells should result in increasing EC substrate cleavage.
Confirming the first prediction, immunoblotting with three different anti-caspase-3 monoclonal antibodies revealed that TRAIL-treated Bcl-2-overexpressing cells contained low levels of caspase-3 p20, an active form arising from cleavage of pro-caspase-3 at D175 by initiator caspases, and lacked detectable p17, which arises from autocatalytic cleavage of the p20 form (Fig. 5C; epitope masking or destruction have previously been ruled out as a possible explanation for the phenomenon) (Tawa et al., 2004). Both p20 and p17 are active, but since generation of p17 requires caspase-3 catalytic activity, accumulation of p20 in the absence of p17 is characteristic of XIAP-mediated caspase-3 inhibition (Bratton and Cohen, 2003). In our experiments, the p17/p20 forms of caspase-3 were present at significant levels in extracts only when MG-132 was added at saturating concentrations (with p17 predominating, Fig. 5C), consistent with the idea that p17/p20 destruction is proteasome-dependent. When p32, p20 and p17 levels were quantified from Western-blot data, we estimated a p17/p20 half life of ~20 min in the absence of MG132 and ~60 min in its presence (Fig. 5D). These calculations are not straightforward because they must account for non-linear production of p20 and p17 as pro-caspase becomes depleted (see Supplementary Materials) but we nonetheless conclude that caspase-3 is indeed degraded at a fairly rapid rate in pre-MOMP cells and that a significant fraction of this degradation is proteasome-dependent. Substantially slower, proteasome-independent p17/p20 degradation is also evident, as suggested previously (Tawa et al., 2004).
To test the prediction that inhibiting proteasome activity not only promotes caspase-3 p17 accumulation but also induces EC substrate cleavage, Bcl-2-overexpressing cells were exposed to MG-132 for 30 min at concentrations spanning the drug's IC50, and TRAIL then added. Simulation showed that, under these conditions, levels of EC substrate cleavage at 24 hr (which approximates an end-point assay) should vary in a graded manner from partial to complete (Fig. 5E). Flow cytometry of EC substrates strongly supported this prediction: as the concentration of MG-132 increased, the median level of substrate cleavage rose gradually from 5% to 89% (Fig. 5F). Moreover, at low MG-132 concentrations, substrate cleavage ranged from 5% to 50% at 4.5 hr and no further increase was observed at 24hr. When MG-132 exceeded 2μM, however, substrate cleavage was extensive and many cells lysed, making the 24 hr time point unreliable (dotted lines in Fig. 5F). Taken together, our data strongly support the hypothesis that ubiquitin-mediated caspase-3 degradation is essential for preventing EC substrate cleavage in pre-MOMP cells. The effects of perturbing Bcl-2, XIAP, and caspase-3 degradation can be viewed systematically in a bivariate landscape of predicted EC substrate cleavage relative to parameter values. Perturbations such Bcl-2 over-expression shift cells away from a normal state characterized by rapid switching (position 1 in Fig. 5G and H) to one in which EC substrate cleavage is strongly suppressed (position 2). This suppression by Bcl-2 over-expression is relieved by decreasing the level of XIAP (by RNAi) or the rate of caspase-3 degradation (by MG-132 treatment), resulting in shift toward cell states characterized by slow and incomplete switching (positions 3 and 5) or a restoration of effective switching (positions 4 and 6).
Switch-like activation of caspase-3 under normal conditions requires rapid all-or-none inactivation of XIAP. RNAi-mediated depletion of Smac, the inhibitor of XIAP, should therefore interfere with caspase-3 activation and EC substrate cleavage. As expected, EC-RP cleavage was gradual in Smac-depleted cells (Fig. 6A), and endogenous EC substrates were partially cleaved in >50% of cells at t = 4.5 hr (Fig. 6B). Because Smac acts downstream of MOMP, partial EC-RP cleavage is observed only in Smac-depleted cells that have experienced MOMP (as measured by IMS-RP release, Fig. 6A) and that are destined to die due to loss of normal mitochondrial function (Colell et al., 2007). Thus, “partial cell death” does not occur under these conditions. Remarkably, co-depletion of XIAP and Smac restored normal caspase-3 activation as assayed by EC-RP cleavage and flow cytometry (Fig. 6C,D, E), suggesting that it is not the absolute levels of Smac and XIAP that are important, but rather their ratio. Once again, the effects of perturbing Smac alone, or Smac/XIAP together could be visualized on a bivariate landscape of EC substrate cleavage (Fig. 6F).
The hypothesis that XIAP plays a critical role in controlling caspase-3 activity appears inconsistent with reports that apoptosis is regulated normally in XIAP knockout mice (Harlin et al., 2001). Cells in which XIAP had been depleted 8- to 17-fold by RNAi did exhibit responses to TRAIL similar to those of control cells, although with two subtle differences (Fig.6G-I): (i) fully cleaved EC substrates appeared ~1 hour sooner, and (ii) pre-MOMP cells exhibited small but reproducible increases in partial EC substrate cleavage (arrows in Fig. 6G-I). Upon subsequent onset of MOMP, this premature and partial substrate cleavage was overwhelmed by rapid activation of the bulk of caspase-3 and the phenotype therefore masked. The phenotypic consequences of XIAP depletion for initiator and effector caspase activation dynamics are therefore transient in our experiments, except when Bcl2 is over-expressed or Bid depleted, preventing timely onset of MOMP.
In this paper we use live-cell imaging, flow cytometry and mathematical modeling to examine the states of the core apoptotic network, comprising caspases and their immediate regulators, in HeLa cells exposed to TRAIL. Single-cell data show that cells treated with TRAIL enter a state of delay lasting many hours, during which initiator caspases are active (Figure 7A, state 2). Because procaspase-3 is a direct substrate of initiator caspases, the enzymatically active form of caspase-3 is generated steadily throughout the delay state, but its activity is held fully in check. Previous single-cell experiments examining the dynamics of initiator and effector caspase activation by death ligands have been contradictory, with the most recent reports concluding that the two are simultaneous (Kawai et al., 2004; Kawai et al., 2005; Luo et al., 2003). Our data, based on improved reporters and quantitative analysis of a large number of cells, demonstrate that this is not the case. If MOMP is blocked by Bcl-2 overexpression or Bid depletion, or if TRAIL is added at very low doses, the pre-MOMP delay state can persist with initiator caspases on and effector caspases off for many hours (Fig. 7A, state 2*).
Modeling and experiments reveal that, in the absence of new protein synthesis, maintaining a long-lasting pre-MOMP delay and then switching suddenly to rapid and decisive cell death places stringent demands on the pathways that regulate effector caspase activity. Caspase-3 is a very potent enzyme and simulation shows that even ~400 active molecules are sufficient to cleave ~1 mg/ml cellular substrate (i.e. 106-107 molecules per cell) within several hours (Fig. S4). Despite this catalytic efficiency, we observe that >90% of the 104-105 caspase-3 molecules present in a typical HeLa cell (Fig. 3F) (Stennicke et al., 1998) are processed but held in check prior to MOMP without evidence of caspase-3-mediated EC substrate proteolysis. Experiments with proteasome inhibitors strongly suggest that ubiquitin-dependent degradation of caspase-3 is essential for effective inhibition of its enzymatic activity. Modeling reveals why this is true: were caspase-3 inhibition dependent solely on XIAP's activity as a nanomolar competitive inhibitor (Kd ~1 nM) (Huang et al., 2001), a >100-fold molar excess of XIAP over caspase-3 would be required to ensure the highly efficient inhibition of proteolytic activity observed experimentally in pre-MOMP cells (Fig. S4). This arises because a competitive inhibitor of an irreversible enzymatic reaction must be present in large excess to block access of abundant substrates to the enzyme active site (Huang et al., 2001). However, XIAP is not present in large molar excess over caspase-3, and available data suggest that the proteins are present in HeLa cells at roughly similar levels (Rehm et al., 2006). Conversely, if caspase-3 is modeled as being unstable with a t1/2 of ~20 min, active caspase never accumulates to a high level, and the need for a vast excess of XIAP levels is reduced. XIAP has an E3 activity, and it is therefore reasonable to model it as having two activities, one as a competitive inhibitor of caspase-3 catalytic activity and the other as a mediator of XIAP ubiquitination (although other possibilities can be imagined, and direct tests with mutant XIAPs remain necessary). It should be noted that even when both the competitive inhibitory and E3 ligase activities of XIAP are included in our model, the degree of caspase-3 inhibition observed in pre-MOMP cells is achieved only with XIAP levels higher than those we estimate by semiquantitative Western blotting (data not shown). Thus, additional mechanisms of caspase-3 inhibition are required, among which ubiquitin-independent caspase-3 degradation with a t1/2 ~60 min is one (Fig. 5D) (Tawa et al., 2004). Our working model is that three (and possibly more) distinct processes are involved in restraining caspase-3 catalytic activity during the pre-MOMP delay: competitive inhibition by XIAP, XIAP E3-mediated destruction by the proteasome, and ubiquitin-independent proteolysis. Several developmental processes are associated with procaspase-3 cleavage in the absence of cell death and it will be interesting to ascertain if similarly complex mechanisms of caspase-3 regulation are involved (Rosado et al., 2006).
In addition to maintaining a prolonged state of delay, a second fundamental challenge for the extrinsic cell death network is achieving a rapid and unambiguous transition to death (Fig. 7A,B, state 3). In HeLa cells this transition is dependent on MOMP. Our data support the widely held notion that a threshold level of pro-apoptotic Bcl-2 family proteins (such as tBid, the level of which is reflected by the level of IC-RP cleavage) must be reached for MOMP to occur and that the height of this threshold varies considerably depending on the physiological state of a cell. For example, we find that the MOMP threshold is 1.5- to 2-fold higher in the absence of CHX than in its presence, when assayed by IC-RP cleavage, a difference that we attribute in part to changes in the levels of the unstable Mcl-1 protein (unpublished data). Once the threshold is reached, Smac is released into the cytosol whereupon it binds to XIAP 3-fold more avidly than to XIAP binds to caspase-3 (Huang et al., 2001; Huang et al., 2003). Simulations show that the rapid 10-fold reduction in XIAP activity necessary for sudden and efficient EC substrate cleavage can be achieved assuming diffusion-limited binding and roughly equimolar XIAP and Smac. If we attempt to reconcile the demands of a stable delay (< 5% EC substrate cleavage in pre-MOMP delay state, Fig. 7C) with those of rapid and efficient transition to death (> 90% EC substrate cleavage in the post-MOMP state, Fig. 7D), the expression level of XIAP must lie within a relatively narrow range (Fig. 7E). We anticipate that the precise shape of this landscape will differ among cell types, as CHX-treated HeLa cells used in this study are quite abnormal. Indeed, we have observed a different landscape of cellular states in Hct-116 cells (unpublished data), although the model can be fitted to these data with a few adjustments to the expression levels of key proteins. Looking forward, it will be interesting to determine how the core apoptotic machinery is buffered against variation beyond the range of parameter values at which effective all-or-none switching is possible, and whether a breakdown in switching is more frequent in disease.
A third state of the caspase network, observed only in perturbed cells, involves persistent partial EC substrate cleavage (Fig. 7A,B; states 4 and 5). If the extent of cleavage is insufficient (and our data suggest an LD50 for EC substrate cleavage of ~10% of full cleavage), cells survive despite damaged proteomes and genomes. DNA damage seems most likely to be pathological than protein degradation because it has the potential to cause genomic instability. Indeed, activation of effector caspases and downstream DNAses in the absence of cell death has been proposed to induce the chromosomal translocations characteristic of leukemia (Betti et al., 2005; Vaughan et al., 2002; Vaughan et al., 2005; Villalobos et al., 2006). Our work identifies at least one way in which this state can arise through breakdown in pre-MOMP inhibition of caspase-3. Consistent with this finding, Bcl-2 is a potent oncogene in cancers such as leukemia that are characterized by chromosomal translocations (McDonnell and Korsmeyer, 1991). Perhaps the oncogenic effects of Bcl-2 reflect its ability to promote accumulation of genetic lesions via a malfunctioning apoptotic machinery as well as through its well-recognized ability to prevent apoptosis. The cooperation between Bcl-2 and MG-132, a drug similar in activity to the cancer therapeutic Bortezomib, in inducing a state of partial death further emphasizes the importance of understanding the responses of individual cells to perturbations in core apoptosis pathways. Avoiding physiologically indeterminate states of partial cell death is as likely to be important in cancer treatment as well as in the normal physiology of programmed cell death.
pECFP-Smac was obtained from Dr. Richard Youle. IMS-RP was constructed by PCR of bp 1-147 of Smac-1, ligation to the 5′ end of monomeric RFP-1, and cloning into pBabe-Puro. EC-RP, IC-RP, and DEVDG-RP were constructed by ligating Venus YFP between BamHI and EcoRI sites in pECFP-C1, and ligating linkers encoding cleavage sequences as BspEI-BamHI fragments between ECFP and Venus. Multiple serine and glycine residues flanking the cleavage sequences were added to increase linker flexibility and substrate accessibility.
HeLa cells were obtained from the ATCC and cultured in DMEM supplemented with 10% calf serum and L-glutamine; Bcl-2-overexpressing cells were obtained from Dr. Fei Hua and Dr. Michael Cardone. HeLa cells stably expressing combinations of EC-RP, IC-RP, and IMS-RP were derived by transfection with Fugene 6 (Roche) and isolation of puromycin and geneticin-resistant colonies. SuperKiller TRAIL was obtained from Alexis Biochemicals, CHX from Sigma-Aldrich, and MG-132 from Calbiochem. Flow cytometry, immunoblotting, colony forming assays, and computational modeling were performed using conventional methods; see Supplementary information for details.
Time-lapse movies were recorded using a Deltavision-modified Spectris IX71 fluorescence microscope equipped with an environmental chamber (Olympus, Applied Precision) at 10× or 60× magnification with frames every 3 min. Cells grown in 8-well chambered cover glass slides (Nunc) were shifted into phenol red-free CO2-independent medium (Invitrogen) supplemented with 1% fetal bovine serum and L-glutamine for imaging. For FRET analysis, background-subtracted CFP and YFP images were divided to create a ratiometric image using ImageJ and custom plug-ins (available on request). Signals were normalized by subtracting the minimum value across all time points from each single-cell time course. IMS-RP release was analyzed by edge detection in ImageJ or by visual inspection, which enabled an unambiguous identification of the first frame of IMS-RFP release in >95% of cells.
We thank Dr. Fei Hua and Dr. Michael Cardone for Bcl-2-overexpressing HeLa cells, and Dr. Richard Youle for kindly providing pECFP-Smac. We also thank R. Ward, S. Gaudet, S. Spencer, and D. Flusberg for helpful discussions. This work was supported by NIH grant P50-GM68762 to P.K.S.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.