Mechanistic insights into caspase-9 activation by the structure of the apoptosome holoenzyme

An intact holoenzyme was assembled by incubating the full-length human Apaf-1 (residues 1–1,248) with excess amounts of equine cytochrome c (CytC) and human caspase-9 in the presence of 1 mM dATP. The assembled holoenzyme exhibited excellent solution behavior, as judged by gel filtration (Fig. S1A). The assembled holoenzyme was imaged under cryo-conditions on an FEI Tecnai Polara microscope, and 2,579 micrographs were collected (Fig. S1B); 401,784 particles were autopicked for reference-free 2D classification (Fig. S1C). After 3D classification, a subset of 240,130 particles was used for image construction, with different regions of the holoenzyme exhibiting vastly different resolution limits (Fig. S2).

A lobe of compact globular density, present in nearly all particles and similar to that reported previously (14, 15), is located right above the center of the apoptosome (Figs. S2 and S3). An additional but smaller lobe of sickle-shaped density, present in about 10% of the particles, is found at the edge of the central hub, close to the WD40 repeats. The local resolution in these regions was truncated to ∼18 Å in the published structure of the apoptosome holoenzyme (14, 15). In our structure, application of local masking strategy for the heptameric apoptosome platform, the globular density above the central hub, and the sickle-shaped density at the edge of the central hub led to improved overall resolutions of 4.4, 5.1, and 6.9 Å, respectively (Fig. S3). The central hub of the apoptosome holoenzyme displays an identical structure as that reported previously (16).

Apaf-1 sequentially contains a CARD at the N terminus, a nucleotide-binding oligomerization domain, and two β-propellers of seven and eight WD40 repeats at the C terminus. The nucleotide-binding oligomerization domain comprises a nucleotide-binding domain (NBD), a small helical domain, a winged-helix domain, and a large helical domain. Caspase-9 consists of an N-terminal CARD and a protease domain. As previously observed (16), the central hub and the spokes of the apoptosome platform are formed by the nucleotide-binding oligomerization domains and the WD40 repeats, respectively (Fig. 1A). Eight CARDs are identified in the globular density above the central hub, including a layer of four Apaf-1 CARDs (ApCARDs) docking onto the central hub and a second layer of four caspase-9 CARDs (C9CARDs) on top of the ApCARDs (Fig. 1A).

The assignment of the central CARD complex was based on the observation that two ApCARDs and one C9CARD form a stable heterotrimeric complex through both type I and type II interfaces (17) (Fig. 1B). Helices H2 and H3 of ApCARD interact with helices H1 and H4 of C9CARD in the type I interface (18), whereas the intervening turns between H1 and H2 and between H5 and H6 of ApCARD stack against the turns between H2 and H3 and between H4 and H5 of C9CARD in the type II interface. The centrally located hetero-octameric CARD complex, measuring 75 Å in diameter and 60 Å in height, can be viewed as four overlapping heterotrimeric complexes (Fig. 1C). For three of the four C9CARDs in the central CARD complex, each interacts with two neighboring ApCARDs through type I and type II interfaces (Fig. 1C). The fourth C9CARD associates with ApCARD through a type II interface only. Because both types of interface are known to be stable (17), assembly of the central CARD complex in this fashion maximizes the number of type I and type II interfaces, and hence is thermodynamically most favorable.

At 6.9 Å resolution, the sickle-shaped density at the edge of the central hub allowed identification and docking of two ApCARDs and one C9CARD, which are in perfect registry with the reported heterotrimeric complex (17) (Fig. 1D). Notably, this density was described as a single protease domain of caspase-9 in the previous holoenzyme structures (14, 15), as well as in a recent report (19).

In the central CARD complex, four C9CARDs are located on the upper layer, and only three of the four ApCARDs directly contact the apoptosome platform (Fig. 2A). The 5.1-Å resolution is insufficient for accurate assignment of specific interactions involving amino acid side chains in the CARD complex. Despite this caveat, the close proximity of the secondary structural elements between ApCARDs and the central hub allows modeling of potential interactions, which may guide experimental scrutiny of their functional importance.

The interactions between ApCARDs and the central hub of the apoptosome are asymmetric. For two ApCARDs, each uses its N-terminal helix H1 and the C-terminal portion of helix H4 to interact with surface regions of the NBD (Fig. 2B). Despite variations in the detailed interactions, the surface regions of the NBD contacted by these two ApCARDs are similar. In one case, the N-terminal residues Asp2 and Arg6 may hydrogen bond (H-bond) to Thr112 and Glu116, whereas Lys58 and Lys62 are in close proximity of Gln208 and Asp209 (Fig. 2C, Middle and Right). In the other case, Lys4 may donate H-bonds to Tyr109 and Glu175 (Fig. 2C, Left). The third ApCARD is placed further away from the central hub and may only make limited contacts to the central hub.

The 6.9-Å resolution makes it challenging even to speculate on potential interactions at the edge of the central hub. In the heterotrimeric CARD complex, the N-terminal H1 helix of one ApCARD is in close proximity to an N-terminal helix of the NBD (Fig. S4A). Polar and/or charged residues on H1, such as Arg6 and Asn7, and those on NBD, such as Gln137, Gln138, and Ser141, may interact with each other through H-bonds (Fig. S4B). Similarly, polar and/or charged amino acids from the N termini of helices H1 and H4 in the other ApCARD, including Glu14, Lys18, and Gln50, may contact surface residues in WD1 (Fig. S4C).

To examine whether the interactions between the CARD complex and the central hub are functionally important, we introduced specific missense mutations into Apaf-1 that were designed to weaken these interactions but have no effect on formation of the apoptosome. We generated six such Apaf-1 variants, designated M3 through M8, and two control variants, M1 (S141W, K142W, and K144W) and M2 (L173W), that are predicted to have no effect on these interactions (Fig. 3). These Apaf-1 variants were individually incubated with caspase-9 in the presence of CytC and dATP; the proteolytic activity of caspase-9 was examined using its physiological substrate caspase-3 (C163A) (Fig. 3A). Free caspase-9 exhibited a low level of the proteolytic activity, which was drastically increased by the WT Apaf-1 apoptosome (lanes 5 and 6). As anticipated, the control variants M1 and M2 stimulated caspase-9 activity to the same level as WT Apaf-1 (lanes 7 and 8). The M3 variant, in which the mutation D209A targets an amino acid at the interface between the CARD complex and the central hub, also functioned similarly as WT Apaf-1 (lane 9). This observation suggests that a single-point mutation may be insufficient for destabilization of the interface. Supporting this analysis, three variants, M4 (D209K and S211K), M5 (Q208K, D209K, and S211K), and M6 (K58E and K62E), each involving two or more mutations at the interface, led to markedly decreased caspase-9 activity compared with WT Apaf-1 (Fig. 3A, lanes 10–12).

The Apaf-1 variants M7 (D2R, K4E) and M8 (Y109W, T112W, and E116R) contain two and three mutations, respectively, at the interface, yet both variants appeared to stimulate caspase-9 activity to a similar level as WT Apaf-1 (Fig. 3A, lanes 13 and 14). To better characterize these variants, we used a more sensitive cleavage assay using the fluorogenic peptide substrate Ac-LEHD-AFC (Fig. 3B). In contrast to the controls, the variants M7 and M8 only stimulated caspase-9 activity to ∼45–50% of the level by WT Apaf-1. Consistent with results of the other assay (Fig. 3A), the variants M4, M5, and M6 exhibited severely compromised abilities to stimulate caspase-9 activity (Fig. 3B). These results demonstrate that interactions between the central CARD complex and the central hub are essential for proper activation of caspase-9.

Mutations in these Apaf-1 variants were designed to selectively affect the interface between the central CARD complex and the central hub, but not the assembly of the apoptosome. To confirm the design, we individually incubated five Apaf-1 variants with CytC in the presence of dATP and applied the mixture to gel filtration (Fig. 3C). These variants include two controls (M1 and M3) and three loss-of-function variants (M4, M5, and M6). All five Apaf-1 variants retained the same ability as WT Apaf-1 to assemble into the apoptosomes (Fig. 3C).

Why is the interface between the central CARD complex and the central hub required for caspase-9 activation? One possibility is that compromised interactions may affect the integrity of the apoptosome–caspase-9 holoenzyme. To examine this scenario, we assessed formation of a stable holoenzyme on gel filtration by incubating the assembled apoptosome variants with caspase-9. As anticipated, caspase-9 formed a stable holoenzyme complex with WT Apaf-1 and the two control variants M1 and M3, as judged by comigration of the caspase-9 large subunit with the apoptosome (Fig. 3D). In contrast, caspase-9 failed to comigrate with the apoptosome assembled by each of the three Apaf-1 variants M4, M5, and M6. These results suggest that the reduced ability by these Apaf-1 variants to activate caspase-9 is likely attributable to a compromised ability to form the apoptosome holoenzyme.

Unlike the effector caspases, caspase-9 contains an N-terminal CARD that is required for its activation. The recruitment of caspase-9 by the Apaf-1 apoptosome strictly depends on their respective CARD domains (18, 20). Two CARD complexes, one above the central hub and the other at the periphery, are prominently featured in the assembled apoptosome holoenzyme. We suspected a regulatory role by C9CARD in addition to its known function in holoenzyme assembly. Indeed, the isolated protease domain of caspase-9 was previously found to exhibit a higher level of proteolytic activity compared with the full-length caspase-9 (12). To investigate this scenario, we generated two caspase-9 variants: residues 98–416, which no longer contains the C9CARD, and residues 140–416, which only contains the protease domain (Fig. 4A). Compared with full-length caspase-9, the CARD-deleted variant (residues 98–416) exhibited a markedly increased proteolytic activity toward the physiological substrate caspase-3 (C163A) (Fig. 4A, lanes 6 and 7). Removal of the linker sequences in caspase-9 led to an even higher level of activity (lane 8). Similar results were obtained using the fluorogenic peptide substrate Ac-LEHD-AFC, where removal of the CARD increased caspase-9 activity by at least 10-fold (Fig. 4B). These results demonstrate that the CARD domain of caspase-9 directly inhibits its proteolytic activity. This finding strongly argues that assembly of the CARD complex in the apoptosome holoenzyme may serve to relieve the inhibitory effect of C9CARD.

How does the CARD inhibit the proteolytic activity of the caspase-9 protease domain? One possibility is that, similar to effector caspases, free caspase-9 only has robust proteolytic activity in its homodimeric form, and C9CARD inhibits homodimerization. This hypothesis is particularly attractive because caspase-9 inhibition by the BIR3 domain of XIAP is mediated through heterodimerization that involves the homodimerization interface of caspase-9 (21). To investigate this scenario, we engineered three constitutively homodimeric caspase-9 variants using an established strategy, in which a five-residue β-strand at the homodimeric interface of caspase-3 was grafted onto caspase-9 (22). The resulting caspase-9 variants, hereafter referred to as DC9, remain mainly as stable homodimers on gel filtration (Fig. 5A). Compared with WT caspase-9, the full-length DC9 exhibited a much higher level of proteolytic activity toward the physiological substrate caspase-3 (C163A) (Fig. 5B, lanes 7 and 8) or the fluorogenic peptide Ac-LEHD-AFC (Fig. 5C). In particular, the proteolytic activity of the full-length DC9 is ∼12-fold higher than that of WT caspase-9 in the peptide-based assay (Fig. 5C), consistent with an inhibitory role of C9CARD in caspase-9 homodimerization.

Intriguingly, compared with the full-length DC9, removal of the CARD in DC9 (residues 98–416) led to a 2.5-fold increase of the proteolytic activity, and further truncation of the linker sequences preceding the protease domain resulted in an additional threefold increase (Fig. 5C). We speculate that DC9 exists in an equilibrium between a large population of homodimers and a small population of monomers, and the presence of C9CARD and the ensuing linker sequences inhibit homodimerization. This speculation appears to be supported by the appearances of the peaks for the various DC9 variants on gel filtration (Fig. 5A). Instead of the proposed inhibition of caspase-9 homodimerization, C9CARD may directly inhibit the proteolytic activity through binding to the active site or another region of caspase-9. The above experimental results fail to rule out this possibility.

Incubation of the full-length DC9 with the Apaf-1 apoptosome yielded a sixfold increase of the proteolytic activity. In fact, DC9 (residues 140–416) displayed a higher proteolytic activity toward the substrate Ac-LEHD-AFC than the apoptosome–caspase-9 holoenzyme (Fig. 5C). Intriguingly, the proteolytic activity of DC9 (residues 140–416) is slightly higher than that of the full-length DC9 in the presence of the Apaf-1 apoptosome (Fig. 5 B and C). This finding demonstrates that the homodimerized protease domain of caspase-9 has a proteolytic activity that is at least comparable to or higher than that of the apoptosome caspase-9 holoenzyme.

Human Apaf-1 and WT caspase-9 were purified as described (16). The holoenzyme was assembled in the presence of 1 mM dATP by incubating Apaf-1, equine CytC, and caspase-9 at a molar ratio of 1:2:2. The dimeric caspase-9 was generated as described (22).

Cryo-EM images were recorded manually on an FEI Tecnai Polara electron microscope operating at 300 kV. The 26 movie frames of each micrograph were aligned by whole-image motion correction (23). Contrast transfer function parameters were estimated by CTFFIND4 (24), and 2D and 3D classifications were carried out using RELION (version 1.4-alpha) (25). The particle-polishing approach was applied (26). Reported resolutions are based on Fourier Shell Correlation (FSC) curves between independently refined half-maps, using the 0.143 criterion (27). The resulting maps from refinement were postprocessed by RELION and sharpened by a negative B-factor (28). Local resolution variations were estimated using ResMap (29). The structures of Apaf-1 apoptosome (PDB code 3JBT) (16) and 2:1 ApCARD-C9CARD complex (PDB code 4RHW) (17) were docked into the overall maps by COOT (30) and fitted into density by CHIMERA (31).

For apoptosome assembly, Apaf-1 mutants were individually incubated with equine CytC at a molar ratio of 1:2 in the presence of 1 mM dATP. For holoenzyme assembly, the apoptosome mutants were individually incubated with WT caspase-9 at a molar ratio of 4:1.

The effect of Apaf-1 mutations on caspase-9 activity was measured using caspase-3 (C163A) and Ac-LEHD-AFC. The molar ration between Apaf-1 and caspase-9 is 5:1. The assay was performed as described (8). The basal activities of caspase-9 truncations were also measured as described (8).

The full-length human Apaf-1 with a C-terminal 10xHis tag was purified as described previously (16). The wild-type (WT) caspase-9 was cloned into a pET29 vector with a C-terminal 6xHis tag, overexpressed in Escherichia coli BL21 (DE3), and purified by nickel affinity chromatography (Ni-NTA, Qiagen) and anion-exchange chromatography (Source-15Q; GE Healthcare). The holoenzyme was assembled in the presence of 1 mM dATP by incubating Apaf-1, equine cytochrome c, and caspase-9 at a molar ratio of 1:2:2. The holoenzyme was further purified by gel filtration [Superose-6 (10/30), GE Healthcare] in 20 mM Hepes (pH 7.5), 10 mM KCl, 1.5 mM MgCl2, 1 mM EDTA, and 1 mM DTT.

All caspase-9 variants were cloned using a PCR-based method. The caspase-9 variant proteins were purified identically as for WT caspase-9. The dimeric caspase-9 was generated as described (22).

Three-microliter aliquots of the assembled Apaf-1 apoptosome bound to caspase-9 (the holoenzyme), at a concentration of ∼5 μM, were applied to glow-discharged Quantifoil 200-mesh CuR 2/2 grids. Grids were blotted in a Vitrobot IV (FEI Company) at 4 °C for ∼2–3 s with 100% humidity and then plunge-frozen in liquid ethane. Cryo-EM images of the holoenzyme were recorded manually on an FEI Tecnai Polara electron microscope operating at 300 kV. A pixel size of 1.34 Å, defocus values between 1.4 and 3.0 μm, a dose rate of ∼20 electrons/Ų/s, and an exposure time of 1.6 s were used on an FEI Falcon-III detector.

The 26 movie frames of each micrograph were aligned by whole-image motion correction to correct beam-induced movements (23). Contrast transfer function parameters of the resulting micrographs were estimated by CTFFIND4 (24). Using RELION (version 1.4-alpha) (25), 401,784 particles were autopicked from 2,579 micrographs (Fig. S1B). After two rounds of reference-free 2D class averaging (Fig. S1C) and one round of 3D classification, 240,130 good particles were selected for 3D autorefinement (Fig. S2). A 40-Å low-pass filtered cryo-EM reconstruction of Apaf-1 apoptosome (EMD-6480) (16) was used as an initial model in the 3D autorefinement. The particle-polishing approach was subsequently applied to the autorefined particles for particle-based beam-induced movement correction (26). Nonsymmetric autorefinement of these movement-corrected particles resulted in a map at an overall resolution of ∼5.1 Å. The map exhibited a clear density of the heptameric apoptosome platform and two relatively poor densities sitting above the central hub of the platform: one disk-like density positioning in the center, and one sickle-like density locating in the periphery of the central hub (Fig. S2).

By applying a local mask around the central disk-like density and performing 3D classification without any alignment, particles were classified into seven classes. The resulting maps of the seven classes all contain the central disk-like densities, which can be well aligned by rotation around the symmetry axis of the platform (Fig. S2). For example, the map of the second class can be aligned with that of the first class by clockwise rotation of 3 × 360°/7 degrees. By adding 3 × 360°/7 to the column “_rlnAngleRot” in the RELION particle star file of the second class, the particles of this class were rotated 3 × 360°/7 degrees. The particles of the third and the fourth classes were clockwise rotated by 360°/7 and 2 × 360°/7, respectively; the particles of the fifth, the sixth, and the seventh classes were counter clockwise rotated by 2 × 360°/7, 3 × 360°/7, and 360°/7, respectively. 226,842 orientation-corrected particles were then aligned and refined locally, within 5°, resulting in a markedly improved resolution to ∼5.1 Å for the central disk-like region (Figs. S2 and S3A). The final reconstruction of the central disk-like density showed that this region consists of eight CARD domains, into which atomic coordinates from crystals structures were docked.

By applying a local mask around the peripheral sickle-like density and performing 3D classification without any alignment, particles were classified into seven classes (Fig. S2). One of the seven classes showed strong density in this region. The reconstruction of 40,562 particles from this class led to improved quality of the peripheral sickle-like density at an average resolution of ∼6.9 Å (Fig. S3A), allowing identification of three CARDs in this region. By applying a local mask around the heptameric apoptosome platform, the classification and autorefinement of the platform with a C7 symmetry resulted in a map at 4.4 Å resolution (Figs. S2 and S3A), which displays highly similar features as those in the reported 3.8 Å Apaf-1 apoptosome (EMD-6480) (16). Reported resolutions are based on FSC curves between two independently refined half-maps, using the 0.143 criterion (27). The resulting maps from refinement were postprocessed by RELION for correction of the modulation transfer function of the detector and sharpened by a negative B-factor (28). Local resolution variations were estimated using ResMap (29).

The structures of Apaf-1 apoptosome (PDB code 3JBT) (16) and 2:1 ApCARD-C9CARD complex (PDB code 4RHW) (17) were docked into the overall maps by COOT (30) and fitted as rigid-bodies into density by CHIMERA (31).

The effect of Apaf-1 mutations on the assembly of apoptosome and holoenzyme was assessed by gel filtration analysis. For apoptosome assembly, Apaf-1 mutants were individually incubated with equine cytochrome c at a molar ratio of 1:2 in the presence of 1 mM dATP before loading into a Superdex-200 column (Increase 5/150; GE Healthcare). For holoenzyme assembly, the apoptosome mutants were individually incubated with WT caspase-9 at a molar ratio of 4:1. The column was pre-equilibrated with 100 mM KCl, 20 mM Hepes (pH 7.5), and 5 mM DTT. Fractions were then examined by SDS/PAGE and Coomassie staining.

The effect of Apaf-1 mutations on Apaf-1 assembly was measured by the activity of apoptosome-activated caspase-9 activity. In the presence of 14 μM caspase-3 (C163A) as the substrate, the assembled apoptosome mutants were incubated with 45 nM caspase-9 at a molar ratio of 5:1 at 37 °C for 60 min. The reaction was stopped by adding an equal volume of 2×SDS loading buffer. The cleavage activities were examined by SDS/PAGE and Coomassie staining. In the presence of 200 μM fluorescent peptide Ac-LEHD-AFC, caspase-9 at 80 nM was mixed with various apoptosome mutants at a molar ratio of 1:3 at 22 °C. The fluorescence intensity was monitored using a fluorescence spectrophotometer (F-4600, Hitachi) with an excitation wavelength of 400 nm and an emission wavelength of 505 nm. For measurement of the basal activities of caspase-9 variants, caspase-9 at 500 nM was mixed with 200 μM Ac-LEHD-AFC in a preheated assay buffer containing 100 mM KCl, 20 mM Hepes 7.5, and 5 mM DTT.