|Home | About | Journals | Submit | Contact Us | Français|
Activation of procaspase-9 on the apoptosome is a pivotal step in the intrinsic cell death pathway. We now provide further evidence that caspase recruitment domains of pc-9 and Apaf-1 form a CARD-CARD disk that is flexibly-tethered to the apoptosome. In addition, a 3D reconstruction of the pc-9 apoptosome was calculated without symmetry restraints. In this structure, p20 and p10 catalytic domains of a single pc-9 interact with nucleotide binding domains of adjacent Apaf-1 subunits. Together, disk assembly and pc-9 binding create an asymmetric proteolysis machine. We also show that CARD-p20 and p20-p10 linkers play important roles in pc-9 activation. Based on the data, we propose a proximity-induced association model for pc-9 activation on the apoptosome. We also show that pc-9 and caspase-3 have overlapping binding sites on the central hub. These binding sites may play a role in pc-3 activation and could allow the formation of hybrid apoptosomes with pc-9 and caspase-3 proteolytic activities.
Programmed cell death terminates unwanted or dangerous cells in higher organisms. This allows cellular constituents to be recycled while avoiding inflammation (Danial and Korsmeyer, 2004; Green and Evan, 2002). In the intrinsic pathway, developmental cues, DNA damage, growth factor withdrawal or oncogene activation may act as signals to release cytochrome c from mitochondria, through a pore complex in the outer membrane (Moldoveanu et al., 2006; reviewed in Riedl and Shi, 2004). Cytochrome c binds to monomers of apoptotic protease activating factor-1 (Apaf-1) to trigger a conformational change that leads to nucleotide exchange and assembly of the heptameric apoptosome (Zou et al., 1997, 1999, Liu et al., 1996; reviewed in Bratton and Salvesen, 2010).
Apaf-1 is a member of the AAA+ super-family and contains an N-terminal caspase recognition domain (CARD), a central nucleotide binding and oligomerization domain and a C-terminal regulatory region (Inohara and Nunez, 2003; Danot et al., 2009). During apoptosome assembly, nucleotide binding domains (NBD) of seven Apaf-1 momomers associate to form an inner ring within the central hub, while their helical (HD1) and the winged helix domains (WHD) form a second ring that encircles the NBDs (Yuan et al., 2010). Helical domain 2 interacts with the WHD in each subunit to form an extended arm and this feature supports tandem β-propellers in the V-shaped regulatory region. Cytochrome c is bound between the two β-propellers (Acehan et al., 2002; Yu et al., 2005, Yuan et al., 2010) and serves as the initial trigger for Apaf-1 assembly (Li et al., 1997; Hu et al., 1999). Active site CARDs on Apaf-1 subunits in the apoptosome appear to be flexibly-linked to their respective NBDs in the ground state (Yuan et al., 2010).
The holo-apoptosome directs apoptosis and is formed by Apaf-1, cytochrome c and procaspase-9 (pc-9) (Rodriguez and Lazebnik, 1999; Zou et al., 1999; Hu et al., 1999; Srinivasula et al., 1998). In this complex, Apaf-1 and pc-9 CARDs associate to form a disk-like feature that sits above the central hub (Yuan et al., 2010). The pc-9 apoptosome then activates executioner procaspases, such as pc-3 and pc-7, by carrying out a limited proteolysis that rearranges critical loops to form the active sites (Shi, 2004; Riedl et al., 2001a, b). Initiator procaspases such as pc-9 are monomers in solution (Boatright et al., 2003, Chao et al., 2005; Li and Yuan, 2008; Renatus et al., 2001), while executioner caspases are constitutive dimers. Precise details of pc-9 activation are not known but recent data supports a proximity-induced dimerization model (Yin et al., 2006; Pop et al., 2006; Boatright et al., 2003; Renatus et al., 2001). However, pc-9 molecules that were re-engineered to form constitutive dimers are much less active than pc-9 apoptosomes (Chao et al., 2005). Hence, some interactions between pc-9 and the apoptosome may be required to activate the zymogen (Chao et al., 2005; Yin et al., 2006).
In this paper, we provide additional evidence that pc-9 interacts with Apaf-1 CARDs to form an activation disk. We then systematically shortened the CARD-p20 linker and found that the largest deletions significantly reduced pc-9 activation. We surmise that the CARD-p20 linker may facilitate the binding of pc-9 catalytic domains to the apoptosome. We localized this binding site in a 3D map that was calculated without imposing rotational symmetry. In the active apoptosome, the p20 and p10 catalytic domains of a single pc-9 bind to neighboring NBDs in the central hub, at a site adjacent to the CARD-CARD disk. Thus, disk assembly and pc-9 binding work together to create an asymmetric proteolysis machine. Based on the data, we propose a model for pc-9 activation that requires proximity induced association of catalytic domains with the apoptosome.
In the next step of the death pathway, pc-9 apoptosomes cleave pc-3 to create hemi-active caspase-3/pc-3 dimers and fully active caspase-3 dimers. We confirmed that active caspase-3 binds to the pc-9 apoptosome (Bratton et al., 2001a; Hill et al., 2004; Yin et al., 2006), while unprocessed pc-3 does not bind efficiently to the complex (Bratton et al., 2001a). Strikingly, caspase-3 binding caused a dramatic loss of pc-9 activity because these proteases have overlapping binding sites. We suggest that hemi-active caspase-3/pc-3 dimers may bind transiently to caspase-3 binding sites on the apoptosome, to increase the probability of an activating cleavage in the remaining pc-3 subunit. In this scenario, hybrid apoptosomes may be formed through feedback binding of caspase-3 dimers to the central hub.
We recently reported the structure of a human apoptosome with bound pc-9 CARDs (Yuan et al., 2010). To prepare this complex, we inserted a thrombin site in the linker between the CARD and p20 domain of pc-9. We then engineered single and triple mutants that removed known cleavage sites in the p20-p10 linker to create pc-9t (D315A) and pc-9tm (E306A/D315A/D330A). After assembling apoptosomes with pc-9tm, a single clip by thrombin released the pc-9 catalytic domains (p20-p10) in low salt buffer (LSB; Yuan et al., 2010). This created a pc-9 CARD apoptosome that was less prone to aggregation than the pc-9 complex. Procaspase-9 and Apaf-1 CARDs formed a disk-like feature that sits above the heptameric platform in this complex (Yuan et al., 2010; Figures S1A, S1B). However, the disk-like feature may be blurred because it is linked to the platform by potentially flexible, CARD-NBD linkers.
To further characterize the CARD-CARD disk and its role in pc-9 activation, we determined the stoichiometry of pc-9 to Apaf-1 in active apoptosomes with quantitative mass spectrometry. In one experiment, apoptosomes were assembled with an excess of pc-9 (D315A) molecules that contained a TEV site within the CARD-p20 linker (pc-9TEV). Complexes were purified on a 10-40% glycerol gradient (Figure S2A, lane 8) and a peak fraction was used to generate tryptic peptides, which were identified by tandem mass spectrometry (LC/MS-MS) (Experimental Procedures). The peptide MS/MS total ion current (TIC) signal intensities from each protein were averaged together to estimate the amount of each component in the complex (Asara et al., 2008; Jiang et al., 2010). With this approach, the ratio of pc-9TEV to Apaf-1 was found to be ~0.8 : 1 (Figure S2C). As a control, the observed ratio of Apaf-1 to cytochrome c was ~1 : 1.15 (Acehan et al., 2001). In a second experiment, we used glycerol gradients to determine the amounts of pc-9tm and cytochrome c that were just sufficient to saturate binding to the apoptosome. Peak fractions of this complex are shown in Figure S2B. We then assembled pc-9tm apoptosomes and analyzed the complexes by quantitative mass spectrometry (Figure S2C). The measured ratio for pc-9, Apaf-1 and cytochrome c in the complexes was ~0.8 : 1 : 1.6. Based on the data, we surmise that the apoptosome may bind ~5-7 pc-9 monomers to form the CARD-CARD disk in the active complex.
We then analyzed ~1000 side views of double-ring apoptosomes, which occurred at a frequency of ~2% in images of frozen-hydrated, pc-9 CARD apoptosomes (Yuan et al., 2010). In total, 20 side-view classes were obtained with EMAN (Ludtke et al., 1999) and 8 representative class averages are shown in Figure 1A. Two striking observations can be made from the classes. First, double-ring particles are formed by interactions between CARD-CARD disks, which are sandwiched between opposing platforms. From the images, a single disk is estimated to be ~35-50Å thick (depending upon the degree of inter-digitation). Second, the position of opposing platforms is quite variable, as large lateral and vertical displacements are present. This gives further support to our hypothesis that CARD-NBD linkers of Apaf-1 molecules may act as flexible tethers for the disk (Yuan et al., 2010).
To test the idea of a flexible tether, we created a mutant Apaf-1 with a thrombin cleavage site inserted in the linker between the CARD and helix α8 (Apaf-1tb; Figure 2A). Apoptosomes were assembled with Apaf-1tb and their mobility on a glycerol gradient was similar to wild type complexes (Figure 2B, lanes 6-8). When these complexes were treated with thrombin, CARDs were released en masse and remained at the top of the gradient (Figure 2C, lanes 1-3), while heptameric platforms started to aggregate and ran in fractions 9-12 (Figure 2C). These results are consistent with Apaf-1 CARDs being flexibly-arrayed on the apoptosome (Figure 2E), where they can interact with pc-9 CARDs. In addition, the CARD-NBD linker is disordered in the crystal structure of Apaf-1 (residues 1-591; Riedl et al., 2005). Hence, it was not surprising that the thrombin site in the linker is accessible in Apaf-1 monomers (Figure 2D).
To estimate the size of the disk, we visualized pc-9 CARD apoptosomes by rotary metal shadowing after freeze-fracture and etching (Heuser, 1989; Supplemental Methods). In top views, a strongly-contrasted, disk-like feature is visible above the platform (Figure 1B, white arrows). The disk had a diameter of ~100Å (after correcting for the metal coating) and was present in most particles. We also imaged ground state apoptosomes as a control. In these images, we observed a weakly contrasted ring above the hepatmeric platform. These rings (and partial rings) may represent flexibly-bound Apaf-1 CARDs that interact with each other to some extent (Figure 1C, white arrows). Based on these studies, we conclude that pc-9 and Apaf-1 CARDs form a novel disk that is flexibly-tethered to the central hub. Clearly, disk formation is a key step in activation because it would concentrate pc-9 catalytic domains in a region above the apoptosome to facilitate zymogen activation.
Procaspase-9 has a potential linker of ~69 residues between the CARD and p20 domain, although the actual linker may be comprised of the 49 residues inclusive of Asn92-Gly140 (Figure 3A; Renatus et al., 2001). In a recent experiment, a 6 residue linker was sufficient to activate two pc-9 catalytic domains attached to a GCN4 leucine zipper (Yin et al., 2006). Hence, the function of the much longer linker in pc-9 is mysterious. To study the role of the CARD-p20 linker, we made a series of truncations that removed 7, 20 and 30 residues directly downstream of an inserted thrombin site (Figure 3A). When factoring in extra residues from the thrombin site, these mutant procaspases have effective linker lengths of −1, −14 and −24 residues relative to wild type pc-9. A ribbon diagram of the pc-9 dimer is shown in Figure S3 with CARD-p20 linkers drawn roughly to scale for wild type and Δ30L molecules. We then measured the proteolytic activity of pc-9t linker mutants with Ac-LEHD-AFC, in the presence and absence of the apoptosome. For this experiment, we titrated the relative amount of each protein to give a pc-9 to Apaf-1 ratio of ~1:1 (see glycerol gradients, Figure S4). Initial activity curves derived from the fluorescent product were obtained for complexes in a physiological buffer (PB) and in each case, we also measured baseline activities after cutting the pc-9t linker with thrombin to release the p20-p10 domains. Initial rates were normalized to the activity of the pc-9t apoptosome (100) and are plotted as a histogram in Figure 3B. These data provide three important observations.
First, apoptosomes with pc-9t (+6 linker) showed a 50% reduction in their activity relative to complexes made with pc-9t Δ7L/−1. Thus, lengthening the wild type linker can lower activity. We suggest that the thrombin site my impair function of the CARD-p20 linker, perhaps by mediating local aggregation, which lowers activation by ~ 2.5-fold. Second, proteolytic activity was maximal for pc-9t Δ7L/−1, which has a linker that is nearly the same length as in wild type pc-9. Third, when the linker was shortened further (Δ20L/−14 and Δ30L/−24), proteolytic activity of the complexes dropped dramatically (~4 and 9-fold, respectively). To show the generality of our results, we also tested the activity of apoptosomes with wild type pc-9 molecules that contained a Δ30L/−30 mutation in the CARD-p20 linker. These experiments were done to eliminate possible side effects due to the thrombin site in the CARD-p20 linker and due to the D315A mutation in the p20-p10 linker. As shown in the next sections, each of these mutations may affect the overall activity level of pc-9 on the apoptosome. Based on fluorescence-based proteolysis assays, we found a significant defect in pc-9Δ30L apoptosomes relative to pc-9 complexes (Figure 3C). This shows that truncation of the CARD-p20 linker to ~19 residues leads to a significant loss of pc-9 activity on the apoptosome (~80%).
Procaspase-9 apoptosomes cleave and activate pc-3. We investigated this reaction using a pc-3 substrate in which the active site residue (Cys163) was mutated to alanine, to prevent self cleavage during bacterial over-expression. In this experiment, we used a 4-fold excess of pc-3 (C163A) relative to pc-9t and carried out a 30 min time course at 37°C for apoptosomes with pc-9t and with each of the 3 linker mutants. The cleavage of pc-3 to give p20 and p10 subunits was reduced significantly with pc-9tΔ24L and Δ30L apoptosomes, relative to complexes with pc-9t or pc-9tΔ7L (Figure 3D). In all cases, mock reactions showed no pc-3 cleavage (not shown). In addition, the overall activity level with pc-3 as a substrate paralleled the activity of these complexes with a small polypeptide substrate. We conclude that the defect in pc-9tΔ30L apoptosomes extends to the proteolytic processing of pc-3. In addition, apoptosomes with the pc-9tΔ7L mutant had the highest activity with either substrate. Not unexpectedly, these experiments suggest that the optimum linker length for activation is present in wild type pc-9 zymogens.
Next, we wondered if the activation defect in pc-9 molecules with a truncated linker might arise from an inability to assemble a CARD-CARD disk. To test this idea, we assembled apoptosomes with pc-9tΔ30L and released the catalytic domains with thrombin. We then froze these pc9 CARD apoptosomes, imaged the complexes and determined a structure at ~19Å resolution with EMAN2 (Tang et al., 2007). A blurred disk was present in the final, c7 symmetrized 3D map (Figure 3E), which suggests that there is no gross defect in disk formation with this mutant.
We observed a 25-30-fold activation of pc-9tΔ7L molecules on the apoptosome relative to the complexes treated with thrombin (Figure 3B). However, this value was lower than expected (Shi, 2004). We reasoned that this may be due in part, to the D315A mutation in the p20-p10 linker. To test this idea, we measured the proteolytic activities of apoptosomes containing pc-9t with the D315A mutation and of complexes without this mutation (pc-9t wt). In order to assess the effect of the thrombin site, we also evaluated apoptosomes assembled with pc-9 (D315A) or pc-9 (Figure 4A). In panels 1 and 3, the D315A point mutation significantly reduced cleavage activity towards pc-3, relative to complexes in panels 2 and 4 that contained pc-9 molecules without this linker mutation. In addition, time courses with wild type pc9t and pc-9 showed extra density on the gels at t=0, due to the p35 and p10 subunits, which arise from bacterial cleavage of the p20-p10 linker. These bands overlapped the pc-3 and p10/cytochrome c bands, respectively (Figure 4A). The relative effects of the D315A and thrombin site mutants on pc-9 activation were also quantitated with the fluorescence-based proteolysis assay, as discussed later. We conclude that the D315A mutation in the p20-p10 linker has a significant effect on the final level of pc-9 activation (see next section for a possible explanation).
To evaluate the possible role of pc-9 binding in activation, we first compared the accessibility of the CARD-NBD linker in apoptosomes assembled with Apaf-1tb that contained either pc-9 or pc-9 (D315A). In this experiment, active complexes were treated with thrombin and purified on glycerol gradients. We found that clip sites in CARD-NBD linkers were more strongly protected when pc-9 was bound to the apoptosome (Figure 4B, left), relative to complexes with pc-9 (D315A) (Figure 4B, right). To more clearly visualize the protection, we probed the complexes with thrombin and then visualized them directly on a gel (Figure 4C). Since CARD-NBD linkers are located between the disk and the platform, the protection experiments suggested that pc-9 catalytic domains may bind to the apoptosome. If this is the case, then wild type catalytic domains of pc-9 may bind with a higher affinity than domains containing the D315A mutation.
Next, we asked if pc-9 catalytic domains could be visualized on the apoptosome. We froze apoptosomes with wild type pc-9 in LSB and determined a structure without symmetry restraints, using ~20,000 particles and EMAN2 (Tang et al., 2007). To avoid bias, we started with an apoptosome that did not contain the CARD-CARD disk. After refinement, the 3D map revealed the p20 and p10 catalytic domains of a single pc-9 on the central hub, adjacent to the disk (Figures 5A-5D). The FSC0.5 indicated a resolution of 16.9Å and edge-view classes revealed the disk and adjacent pc-9 density on the platform (Figures S5A, S5B). Interestingly, helix α9 and strand β5 of an NBD are located in the “footprint” of the pc-9 catalytic domain, along with the NBD-HD1 linker (Figure 5E). In addition, bound pc-9 catalytic domains may also interact with helix α8 in the NBD of an adjacent subunit. However, a gap is present between the catalytic domains and the NBDs (see asterisk, Figure 5E). This gap cannot be accounted for by the docked models, which suggests that part of the CARD-p20 linker may become ordered to help form the pc-9/NBD interface. This may explain why pc-9 apoptosomes with a Δ30L deletion in the zymogen are significantly less active than wild type complexes (Figure 3C).
The precision of our current docking is limited by the resolution. In this case, an unambiguous choice cannot be made between the orientation in Figure 5E and a second position, defined by a 180 degree rotation about the long axis of the density. However, two aspects of pc-9 function are more consistent with the position shown in Figure 5E. First, the docking explains why a D315A mutation in the p20-p10 linker of pc-9 may down-regulate activation. In a pc-9 monomer, the uncleaved p20-p10 linker may interfere with formation of the NBD interface, because it would pass through the density footprint between the two components (Figure S6 and Figure 5E, middle and right panels). When Asp315 is cleaved, flexible L2 and L2′ ends of the p20-p10 linker may move laterally, allowing the catalytic domains to interact more efficiently with the hub. The p20-p10 loop also backtracks from Asp315 into the active site where it anchors the catalytically active Cys285 (Figure S6, right). Thus, changes in this region induced by binding and auto-processing of the p20-p10 loop could activate the pc-9 monomer. Second, pc-9 is inhibited by the BIR3 domain of the X-inhibitor of apoptosis (XIAP), which forms a complex with the N-terminus (L2′) of the p10 subunit created by auto-processing (Shiozaki et al., 2003). In our model, the BIR3-p10 complex would block activation by preventing the catalytic domains of pc-9 from binding to the hub. A higher resolution 3D map will be required to pin down the precise orientation of pc-9 catalytic domains on the hub.
In our 3D map, the CARD-CARD disk is located off-center relative to the central hub and appears to be tilted (Figures 5A-5D, ,6D).6D). The acentric disk has apparent dimensions of ~80 × 80 × 35Å but the actual diameter of the disk as determined from rotary metal shadowing is ~100Å (see dashed circle in Figure 5A). This is consistent with the idea that the disk is somewhat blurred in the 3D map because of flexible CARD-NBD linkers (Yuan et al., 2010). We wondered if the disk may have been forced to move off-center when the catalytic domains of pc-9 are bound to the hub or alternatively, whether the disk may assemble in an acentric position relative to the hub. To evaluate these possibilities, we reprocessed a large dataset of pc-9 CARD apoptosomes from Yuan et al. (2010) without symmetry restraints. These particles lack the pc-9 catalytic domain, yet the disk in the resulting 3D map is clearly acentric and tilted (Figures 6A-6C), as it is in pc-9 apoptosomes (Figures 5A-5D, ,6D).6D). Since the disk remains in an acentric position after catalytic domains have been released, it seems likely that co-assembly of pc-9 and Apaf-1 CARDs may create a mismatch between the disk and central hub. This also suggests that the disk itself may not have cylindrical symmetry.
In summary, the presence of a mismatch between the disk and central hub could pave the way for pc-9 catalytic domains to bind to a single site on the hub. This may occur because the acentric disk could block access to other binding sites, especially those on the far side of the hub relative to the observed binding site (Figure 5). Since no additional density is present on adjacent Apaf-1 subunits, it seems that catalytic domains of only one pc-9 may be bound to this asymmetric proteolysis machine. We also suspect that binding may be a dynamic process with catalytic domains coming on and off the platform, because a single clip with thrombin released all catalytic domains from the pc-9t apoptosome, while the disk remained intact (Yuan et al., 2010; next section).
The mechanism of pc-3 activation by the pc-9 apoptosome is not well understood. As a first step, we verified previous observations of caspase-3 binding to the apoptosome (Bratton et al., 2001a; Hill et al., 2004; Yin et al., 2006; Malladi et al., 2009). In this experiment, we added a 3-4 fold excess of unprocessed pc-3 (C163A) to pc-9t (D315A) apoptosomes in LSB and after 60 min the cleavage reaction was treated with thrombin. Proteins were then run on a glycerol gradient and visualized by SDS PAGE. Significant amounts of caspase-3 co-migrated with the pc-9 CARD apoptosome, as shown by the p20 subunit (note that the p10 subunit is overlapped with cytochrome c; Figure 7A, lanes 8-11). Released pc-9 catalytic domains, pc-3 and some caspase-3 stayed at the top of gradient. We then confirmed that pc-3 does not bind to the apoptosome. In this experiment, pc-9t apoptosomes were treated with thrombin to release catalytic domains, unprocessed pc-3 (C163A) was added for 60 min and the sample was run on a glycerol gradient. Under these conditions, no proteolysis of pc-3 occurred and most of the procaspase remained at the top of the gradient. While there was some streaking of pc-3 aggregates into the gradient (Figure 7B, lanes 1-5), these molecules were not associated with pc-9 CARD apoptosomes, which peaked in fractions 6-8. Hence, unprocessed pc-3 does not bind strongly to the apoptosome even though the zymogen is a constitutive dimer like caspase-3. These results are in line with previous data from Bratton et al. (2001a) in which cell lysates were supplemented with recombinant caspases. When taken together, the data suggest that caspase-3 binding sites on the apoptosome are able to discriminate between active and inactive conformations of the caspase. In addition, caspase-3 is able to cleave Apaf-1 at Asp271 within the NBD (Bratton et al., 2001b). In our apoptosome model, this cleavage site is buried within the interface between adjoining Apaf-1 subunits and therefore is unlikely to serve as a robust caspase-3 binding site (see Bratton et al., 2001a; Hill et al., 2004; Yin et al., 2006).
We then investigated whether caspase-3 would compete successfully with wild type pc-9 for binding sites on the apoptosome. To test this idea, we assembled apoptosomes with wild type pc-9t. These complexes were run on a glycerol gradient or pre-treated with thrombin before the gradient step. In these experiments, wild type pc-9 catalytic domains were released from the apoptosome after thrombin treatment (Figure 7C, left and center panels). Hence, the affinity of wild type pc-9 catalytic domains for the apoptosome is greatly enhanced by formation of the CARD-CARD disk. We then incubated pc-9t wt apoptosomes with pc-3 (C163A) for 60 min, treated the complexes with thrombin and ran them on a glycerol gradient. This created caspase-3/pc-9 CARD apoptosomes that migrated into the gradient (Figure 7C, right panel), while released pc-9 catalytic domains remained at the top (not shown). Since caspase-3 dimers remain bound to the apoptosome when wild-type pc-9 catalytic domains are released by thrombin, the data suggest that caspase-3 dimers are bound to the central hub, rather than being bound by flexibly-tethered catalytic domains of pc-9.
To characterize these binding sites, we asked if pc-3 and caspase-3 would protect a thrombin site in CARD-NBD linkers of apoptosomes assembled with Apaf-1tb. We incubated pc-9 apoptosomes with pc-3 for either 10 min or 60 min at 37°C, before thrombin addition. Complexes were then run on glycerol gradients and visualized by SDS PAGE. As expected, we found that pc-3 did not protect the CARD-NBD linker from thrombin, since the zymogen is not strongly bound (Figure 7D, top left; fractions 9-11). However, caspase-3 was able to protect the linker (Figure 7D, top right; fractions 6-8). We then compared the ability of pc-9 and caspase-3 to protect CARD-NBD linkers in the apoptosome. For this experiment, we made appropriate complexes and probed them directly with thrombin. The degree of linker protection was much greater for caspase-3 than for pc-9 in these complexes and CARD-NBD linkers were more protected than in control apoptosomes (Figure 7D, bottom). When taken together, the data suggest that pc-9 and caspase-3 may use overlapping binding sites on the central hub, since they are both able to protect the CARD-NBD linker.
We also asked if pc-9t apoptosomes would proteolyze Ac-LEHD-AFC when caspase-3 was bound. In this experiment, we monitored pc-9 activity because caspase-3 (C163A) is catalytically inactive. The activity of pc-9t apoptosomes was normalized to 100 (Figure 7E, left). When pc-3 was added at t=0, the activity of pc-9t apoptosomes was diminished by ~30%. Note that the assay itself took ~15 min and some pc-3 was processed during this time. However, pc-9t activity in the caspase-3 complexes was reduced nearly to background levels when pc-3 was added 60 min before starting the assay (compare Figures 7E, 7F, left set). We also tested the effect of caspase-3 on apoptosomes with bound pc-9 molecules that lacked the D315A mutation, including complexes with pc-9t wt or pc-9 (Figure 7E, middle and right). In both cases, pc-9 molecules without the D315A mutation were significantly more active than pc-9t (D315A) molecules (Figure 7F), as shown previously (Figure 4). Cleavage of unprocessed pc-3 to caspase-3 resulted in a dramatic loss of pc-9 activity to near background levels in both cases (Figures 7E, 7F). Based on our data, a possible explanation for these results is that pc-9 catalytic domains may be displaced from the central hub by caspase-3. This leads to a loss of pc-9 activity while the catalytic domains remain associated with the apoptosome through their linkers to the stable CARD-CARD disk.
In this paper, we show that the activation of pc-9 on the apoptosome involves the formation of a CARD-CARD disk and may require direct binding of pc-9 catalytic domains to the central hub. We also confirmed that caspase-3 binds to the apoptosome, while unprocessed pc-3 does not bind efficiently. Unexpectedly, we found that caspase-3 and pc-9 have overlapping binding sites on the apoptosome. This property allows caspase-3 to down regulate pc-9 activity on the apoptosome by feedback binding. When taken together, these data have allowed us to propose models for the activation of pc-9 and pc-3.
Apoptosome assembly requires cytochrome c binding and nucleotide exchange (see Bratton and Salvesen, 2010; Riedl and Shi, 2004). During this process, each CARD is released from its interactions with the NBD and WHD of Apaf-1. The CARD-NBD linker is ~16 residues long and extends from the CARD C-terminus to the start of helix α8. This linker is disordered in a crystal structure of Apaf1-591 that lacks the regulatory region (Riedl et al., 2005). The CARD-NBD linker may also be flexible in solution because a cleavage site inserted within this feature is accessible to thrombin in the monomer. Apaf-1 CARDs were completely disordered in our 3D map of the ground state apoptosome (Yuan et al., 2010; Yu et al., 2005). However, we could not rule out the possibility that CARDs might have been disordered when exposed to the air water interface during freezing. To test this idea, we showed that CARDs are released en masse from apoptosomes assembled with Apaf-1tb, when the linker is cut with thrombin. We conclude that flexibly-tethered CARDs are displayed on the top surface of the ground state apoptosome in solution (Figure 8, panel 1).
In the first step of pc-9 activation, Apaf-1 CARDs interact with pc-9 CARDs to form a tilted disk-like feature that is displaced laterally on the central hub (Yuan et al., 2010 and this work). This assembly process concentrates pc-9 catalytic domains near the apoptosome and primes them for activation. We suggest that the diameter of the disk and its acentric position could limit the number of possible binding sites for pc-9 catalytic domains on the central hub (Figure 8, panel 2). This may explain why catalytic domains of a single pc-9 are bound to the hub in the active complex. In the extrinsic pathway, FAS and FADD form a distorted disk comprised of 5 FAS and 5 FADD Death Domains, arranged in roughly 2 layers (Wang et al., 2010). CARDs are members of the Death Domain super-family and may form a related structure that does not follow the cylindrical symmetry of the apoptosome. This unusual architecture might also account for the tilted appearance of the disk in our maps.
We now present a model of pc-9 activation on the apoptosome. Based on our data, the apoptosome binds 5-7 pc-9 zymogens to assemble a CARD-CARD disk and form an active complex. We suggest that pc-9 activation may occur by proximity induced association of catalytic domains with adjacent NBDs in the central hub. Moreover, binding may be quite dynamic with different copies of the tethered catalytic domains being transiently activated as they bind to the central hub, after bound catalytic domains have dissociated. These studies also showed that pc-9 catalytic domains provide some protection against thrombin cleavage of CARD-NBD linkers, which are located between the disk and central hub. Conversely, pc-9 (D315A) did not protect CARD-NBD linkers, presumably because catalytic domains of the mutant do not bind as strongly. To account for the observed level of protection by wild type pc-9, we suggest that the acentric disk may protect 3-4 CARD-NBD linkers of Apaf-1, while the disk also blocks access to additional binding sites on the hub. Hence, the holo-apoptosome may contain a single active pc-9 (Figure 8, panel 2).
We also found that pc-9 activation requires an optimal length for the CARD-p20 linker because truncations in this loop resulted in a significant loss of activity. We suggest that the first ~20 residues of the CARD-p20 linker may act as a spacer to facilitate the binding of catalytic domains to the central hub, while the C-terminal region of the linker may help form the interface between p20-p10 domains and adjacent NBDs. This would explain why pc-9 activation is greatly reduced in our linker truncation mutants. In addition, the CARD-p20 linker may serve as a flexible tether for inactive pc-9 catalytic domains that are bound to the disk (Figure 8, panel 2). We also found a 50% loss in pc-9 activation when a 6 residue thrombin site was inserted into the CARD-p20 linker. We speculate that the thrombin site may lead to local effects, such as transient aggregation of the linkers, which impede activation.
The p20-p10 linker of pc-9 is much longer than in other procaspases and may play a role in activation. Cleavage of this “activation loop” was initially thought to be required for activity (Thornberry and Lazebnik, 1998), but subsequent studies showed that uncleaved pc-9 is activated on the apoptosome, but to a lesser extent (reviewed in Bratton and Salvesen, 2010). We found that the activity of pc-9t (D315A) with an uncleaved p20-p10 linker is diminished by ~2 fold, relative to processed pc-9t wt. This activity loss is correlated with a weaker protection of CARD-NBD linkers in the apoptosome by pc-9 (D315A), relative to pc-9. One could argue that an intact p20-p10 linker may impede pc-9 dimerization due to its proximity, even though the loop itself is not in the interface (Renatus et al., 2001). However, our data suggest that an intact p20-p10 loop may interfere with the binding of pc-9 catalytic domains to the apoptosome. This would lower the efficiency of zymogen activation and may explain why the p20-p10 linker of pc-9 undergoes auto-processing.
A generally accepted model for pc-9 activation is based on the idea that local proximity may promote pc-9 dimerization, which in turn triggers activation (Yin et al., 2006; Pop et al., 2006; Boatright et al., 2003; Renatus et al., 2001). In particular, pc-9 catalytic domains were fused to the GCN leucine zipper with a 6 residue linker and these zymogens were activated, presumably by dimerization (Yin et al., 2006). Conceptually, it is easy to understand why a short linker may activate pc-9 by dimerization because CARD-less pc-9 molecules crystallize as a hemi-active dimer at high protein concentration (Renatus et al., 2001). Contrary to our expectations, we found that proteolytic activity on the apoptosome was significantly reduced in pc-9 zymogens with a 30 residue deletion in the CARD-p20 linker. These pc-9 molecules retained 19 residues between the CARD and Gly140. We reasoned that this greatly shortened linker should have been sufficient to promote activation of adjacent pc-9 molecules tethered to the disk, if dimerization were the preferred mechanism.
In a second elegant study, a pc-9 CARD and it’s linker (residues 1-152) were coupled to pc-8 catalytic domains and this chimeric molecule was activated on the apoptosome (Pop et al., 2006). In this experiment, activation may have been due to the similar tertiary structures of pc-8 and pc-9 catalytic domains, CARD-CARD interactions between the chimeric zymogen and Apaf-1, and the presence of the entire CARD-p20 linker of pc-9 at the appropriate position in this molecue. In our structure of the pc-9 apoptosome, the CARD-p20 linker may form part of the NBD/catalytic domain interface. Hence, the CARD-p20 linker may have helped to activate catalytic domains in chimeric pc-9/pc-8 molecules, by promoting binding to the central hub. In contrast, a second activation model has been proposed in which local binding of pc-9 catalytic domains to the apoptosome would activate the zymogen (Chao et al., 2005; Yin et al., 2006). Data presented in this paper suggest that aspects of both models are correct.
We propose that pc-9 activation occurs through proximity induced association. In the first step, a CARD-CARD disk is formed by pc-9 and Apaf-1 CARDs. During assembly the disk adopts an off-center position relative to the hub. We speculate that the acentric disk may block access to potential binding sites on the hub, leaving only one available binding site for pc-9 catalytic domains (Figure 8, panel 2). In the second step, pc-9 catalytic domains bind to the hub and are activated by conformational changes of appropriate loops in the zymogen. This activation step may be facilitated by auto-processing of the p20-p10 loop.
Caspase-3 associates with apoptosomes in lysates from apoptotic cells (Bratton et al., 2001a; Hill et al., 2004) and also forms a complex with the apoptosome when using purified proteins (Yin et al., 2006). In addition, the apparent Km of pc-9 apoptosomes for their pc-3 substrate is much lower than for pc-9 catalytic domains attached to a leucine zipper. This suggested that pc-3 activation may require transient binding to uncharacterized sites on the apoptosome (Yin et al., 2006). In extending these studies, we verified that uncleaved pc-3 does not bind efficiently to the apoptosome (Bratton et al., 2001a). We also confirmed that pc-3 molecules which have been processed by the pc-9 apoptosome are able to bind robustly to this complex (this work, Yin et al., 2006 Bratton et al., 2001a, Malladi et al., 2009). Hence, this binding site is able to discriminate between inactive pc-3 and active caspase-3.
Strikingly, we found that caspase-3 inhibits pc-9 activity. When taken together, our data suggest that caspase-3 dimers may dislodge pc-9 catalytic domains from the central hub (Figure 8, panels 4-5). This may occur because pc-9 and caspase-3 have overlapping binding sites, as suggested by protection of the CARD-NBD linker of Apaf-1 from thrombin cleavage. In this model, pc-9 catalytic domains remain tethered to the disk by their CARD-p20 linker but are inactive. When combined with structural data on the pc-9 apoptosome, this argues that bound catalytic domains of pc-9 are active. We also note that an asymmetric apoptosome with bound pc-9 catalytic domains would serve no obvious purpose, if dimers of flexibly-attached pc-9 catalytic domains were the active species.
The observation of a specific and robust binding site for caspase-3 dimers further suggests that hemi-activated caspase-3/pc-3 dimers could bind to the platform by their activated subunits (Figure 8, panel 3). This would keep hemi-activated caspase-3/pc-3 dimers near the apoptosome, thereby improving the probability of a second cleavage in the remaining pc-3 subunit when it dissociates (for simplicity, we are ignoring clips that remove the N-terminal prodomains). Finally, why is there a mismatch between the disk and platform? The active apoptosome may provide multiple binding sites for caspase-3 dimers on the central hub, even though catalytic domains of a single pc-9 are bound (Figure 8, panel 2). In this scenario, the apoptosome could bind hemi-activated caspase-3/pc-3 dimers at adjacent sites that are not sterically-blocked by the disk or by bound pc-9 catalytic domains, without losing pc-9 activity. Since intra-cellular concentrations of pc-9 and pc-3 are ~10 and 20 nM respectively (Kim et al., 2008; Stoka et al., 2001), the initiator apoptosome would activate pc-3 in an efficient manner. However, in some cells the pc-3 concentration has been estimated to be as high as ~100 nM (Yin et al., 2006). In this case, feedback binding could create a wheel of death with caspase-3 dimers and pc-9 forming a hybrid complex (Figure 8, panel 4; Bratton et al., 2001a; Malladi et al., 2009). If sufficiently high concentrations of caspase-3 are present then an executioner apoptosome could be formed (Figure 8, panel 5) that is in equilibrium with free caspase-3 dimers. Additional studies are now needed to probe the structure and function of pc-9 and caspase-3 on the apoptosome.
The preparation of pc-9 and Apaf-1 mutants and other experimental methods, including glycerol gradients, pc-9 activity assays, quantitative mass spectrometry and metal shadowing of the complexes, are described in Supplemental Methods. Procaspase-9 apoptosomes were prepared for electron cryo-microscopy as described (Yuan et al., 2010) with some modifications and data were collected on a TF20 microscope at 120 kV with a 4kx4k CCD (TVIPS). Image processing was done with EMAN2 and EMAN1 (Tang et al., 2007). The resolution of asymmetric pc-9 apoptosome and pc-9 CARD apoptosome maps was determined to be 16.9 and 11Å, respectively. Both maps were filtered to ~17Å resolution to provide an optimal view of the acentric disk. All crystal structures and models were docked into density maps with Chimera (Goddard et al., 2005) and the catalytic domains of pc-9 were manually adjusted in the density.
We thank Guy Salvesen and Yigong Shi for providing pc-3(C163A) and pc-9 expression clones, respectively. We also thank Fenghe Du and X. Wang for sf21 insect cells and helpful discussions on Apaf-1 expression and purification. The Ludtke lab was supported by NIH grants R01GM080139 and P41RR02250. The Akey laboratory was supported by an NIH grant (RO1 GM63834).
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.