Structure and Interdomain Dynamics of Apoptosis-associated Speck-like Protein Containing a CARD (ASC)^*

Protein Cloning, Expression, and Purification

Cloning, expression, and purification of human ASC has been reported elsewhere (29).

NMR Spectroscopy for Structural Studies

NMR samples were prepared at 0.2 mm ASC, 5 mm d₁₅-Tris(2-carboxyethyl)phosphine, 0.1 mm NaN₃, pH 3.8, 5% D₂O/H₂O, and 100% D₂O. NMR experiments were acquired at 298 K in a Bruker Avance 600 MHz spectrometer equipped with a triple-resonance cryogenic probe. Sequence backbone assignments were obtained from the following experiments: [¹H,¹⁵N]-HSQC, 3D HNCO, 3D HNCACB, and 3D CBCA(CO)NH. Side-chain assignments were obtained from 3D HBHA(CO)NH, 3D (H)CC(CO)NH-TOCSY, and 3D H(CCCO)NH-TOCSY. NOE data were obtained from 3D ¹⁵N-[¹H,¹H]-NOESY (90-ms mixing time) and four-dimensional [¹H-¹³C,¹H-¹³C]-NOESY (90-ms mixing time). Information on NMR experiments for protein structure determination can be found elsewhere (30, 31). All experiments were processed with NMRPipe (32) and analyzed with PIPP (33). Spectra resulting from some of these experiments are shown in Fig. 1 and supplemental Fig. S1.

Several attempts were made to obtain residual dipolar couplings, including the use of bicelle (34) and bacteriophage (35, 36) alignment systems. ASC appears to disrupt bicelle formation resulting in the absence of protein alignment. In the presence of bacteriophages, the NMR spectrum mostly shows signals corresponding to the flexible linker, therefore, precluding the measurement of sufficient data to include in the structure calculation protocol.

Structure Calculation

Peak intensities from NOESY experiments were translated into a continuous distribution of interproton distances. Distances involving methyl groups, aromatic ring protons, and non-stereospecifically assigned methylene protons were represented as a summation averaging, (Σr⁻⁶)^−1/6 (37). Errors of 40 and 30% of the distances were applied to obtain lower and upper distance limits. 78 hydrogen bond distance restraints (r_NH-O = 1.9–2.5 Å, r_N-O = 2.8–3.4 Å) were defined according to the experimentally determined secondary structure of the protein. The TALOS program (38) was used to obtain 311 ϕ and ψ restraints for those residues with statistically significant predictions. Structures were calculated with the program X-PLOR-NIH 2.16.0 (39). The starting structure was heated to 3000 K and cooled in 30,000 steps of 0.002 ps during simulated annealing. The final ensemble of 20 NMR structures was selected based on lowest energy and no restraint-violation criteria. The 20 lowest energy conformers have no distance restraint violations and no dihedral angle violations greater than 0.35 Å and 4.5°, respectively. Structure quality was assessed with PROCHECK-NMR (40) and MolProbity (41). Structures were analyzed with MOLMOL (42). Coordinates were deposited in the Protein Data Bank with accession code 2KN6.

Backbone ¹⁵N Relaxation Measurements

Relaxation experiments were performed at 298 K in a Bruker Avance spectrometer operating at 600 MHz. The ¹⁵N T₁, T_1ρ, and {¹H}-¹⁵N NOE data were obtained with specific NMR pulse sequences (43, 44). The recycle delay to measure ¹⁵N T₁ and T_1ρ was 1 s, whereas {¹H}-¹⁵N NOE experiments used 3.2s. All experiments were acquired in an interleaved manner to minimize the effects caused by spectrometer drift. The relaxation delays of T₁ experiments were the following: 12, 36, 100, 244, 484, 964, 1284, and 1604 ms. T_1ρ experiments used a ¹⁵N continuous spin-lock field of 2.5 kHz. T_1ρ, instead of T₂ relaxation times were acquired because resonance offset effects are significant in T₂ experiments, whereas they can be corrected in a straightforward manner for T_1ρ data using the equation (45),

where θ = tan⁻¹(Ω_N/γ_NB₁), Ω_N is the resonance offset, and γ_NB₁ is the strength of the spin-lock field. T₂ values can be obtained from Equation 1, as T₁, T_1ρ, and θ are known.

Relaxation times were calculated by fitting peak-intensity dependence with the experiment relaxation times to an exponential function given by I(t) = I₀e[(−1/T)t] (T = T₁, T_1ρ). T₁ and T_1ρ values are averages of two separate measurements. The {¹H}-¹⁵N NOE values were calculated from the ratio of peak intensities obtained from experiments performed with and without ¹H presaturation. The ¹H frequency was shifted off-resonance in the unsaturated experiments. The pulse train used for ¹H presaturation utilized 162° pulses separated by 50-ms delays and was applied for a total of 2.2s. The recycle time is reasonably long; however, NOE values were corrected for incomplete ¹H magnetization recovery as previously described (44).

Apparent rotational correlation times were obtained assuming full anisotropy, as described elsewhere (46), from relaxation data of residues that do not undergo slow conformational averaging and show {¹H}-¹⁵N NOE values larger than 0.65 (43). The parameters of the rotational diffusion tensor are shown in supplemental Table S1.

Atomic Force Microscopy Imaging

Mica surfaces were covered with 3 μl of protein solution at either pH 3.8 or 7.0 and incubated for 30 s. The surface was subsequently rinsed with the respective buffers at pH 3.8 and 7.0 and dried. Tapping mode imaging in air was conducted with a Multimode Atomic Force Microscope (Veeco Instruments, Santa Barbara, CA) using a Nanoscope IIIa controller and a J scanner. Veeco nanoprobe tips TESP7 with a resonance frequency of 320 kHz and a spring constant k = 42 newtons/m were used. Scan rates were set at 1Hz.

Molecular Modeling

The solution structure of ASC was used as the monomer template to build the model for the ASC dimer by superimposing the CARD of ASC to Apaf-1-CARD and caspase-9-CARD complex (27). The model was created with the program MOLMOL (42).

ASC Propensity to Oligomerize by NMR and AFM

ASC self-associates in vivo during apoptosis and inflammation (2) and is capable of forming functional supramolecular assemblies in vitro (22). The oligomerization of ASC poses significant challenges for NMR structural studies regarding protein solubility and particle size. It is, therefore, critical to find conditions to minimize oligomerization at the relatively high concentrations used in protein NMR (∼1 mm).

ASC solubility is very low at neutral pH. The soluble fraction of ASC at pH 7 cannot be detected with Coomassie Blue staining in polyacrylamide gel electrophoresis (detection limit 50–100 ng) and results in few observable signals with intensity slightly above the noise level in a spectrum acquired using fast NMR acquisition techniques (47) (supplemental Fig. S2). In contrast, under acidic conditions (pH ∼ 4) ASC solubility improves and can readily be detected by NMR. The dispersion of NMR signals in the [¹H,¹⁵N]-HSQC spectrum of ASC (Fig. 1A and supplemental Fig. S2) indicates that the protein is properly folded at this pH. However, NMR signal intensity decreases over time with no concomitant changes in the chemical shifts (Fig. 1B). This result suggests that ASC forms oligomers of considerable size that tumble too slowly to be observed in the NMR spectrum. The NMR signal intensity of 0.7 mm ASC decays to ∼80% after ∼3 h of sample preparation and to ∼30% within the first 5 days (Fig. 1B). In contrast, at 0.2 mm protein concentration, signal intensity decays to 98% after ∼3 h of sample preparation and to ∼80% within the first 24 h. NMR signal intensity continues decaying to ∼75% of the original signal, reaching a plateau after ∼3 days of sample preparation. Therefore, by decreasing the protein concentration to 0.2 mm, the effect of ASC oligomerization in NMR signal intensity is significantly reduced, whereas it is still possible to acquire NMR triple-resonance spectra with a signal enhancing cryogenic probe.

Because the capability of ASC to oligomerize is basic to its function, oligomerization at neutral and acidic pH was investigated by scanning atomic force microscopy. AFM images show that ASC forms oligomers of similar size and shape at both pH values (Fig. 2, A and B). The predominant species appears like disks of ∼1-nm height and ∼12-nm diameter (see the section images in Fig. 2, A and B). Taken together, the NMR and AFM data indicate that ASC is able to oligomerize in vitro into particles of similar structural features under both pH conditions. Based on these results, it is reasonable to assume that the structure of ASC is not perturbed at acidic pH. Under acidic conditions, the oligomerization reaction likely favors the monomeric form, increasing in turn the solubility of ASC and, thus, leading to a larger fraction of monomers observable by NMR.

To investigate whether the possible oligomerization of ASC at 0.2 mm interferes with structural and dynamics studies, NMR relaxation measurements were performed. Backbone amide (¹⁵N) magnetic relaxation experiments provide rotational correlation times (τ_c), which directly depend on the molecular size and shape (46, 48). In the presence of aggregation, measured τ_c values should be larger than theoretical values derived from the molecular size (49). Differences between the two can also originate from the low sphericity of the protein and from dynamic processes associated, for example, to interdomain motion. The former case generally results in experimental τ_c values larger than the prediction. The experimental τ_c value of ASC is small for its size (∼22 kDa) (Table 1), indicating that the fraction of ASC molecules observable by NMR tumbles as monomers. The discrepancy with the theoretical value (Table 1) could, therefore, be due to interdomain dynamics in ASC (see below). In addition, amide ¹⁵N transverse relaxation times (T₂) depend on the protein rotational correlation time but are reduced in the presence of aggregation (50). For ASC, the average T₂ value of the two domains (72.5 ± 5.5 ms) agrees with the measured correlation time. These results indicate that ASC oligomerization does not have significant effects under the conditions used for the following NMR structural and dynamics studies.

High Resolution Structure of Full-length Human ASC

The three-dimensional structure of ASC was determined with 3046 NOE-derived distances and 311 dihedral and 78 hydrogen bond restraints. The 20 lowest energy conformers of ASC do not show distance or angle restraint violations greater than 0.35 Å and 4.5°, respectively (Table 2). The ensemble of structures does not show significant deviations from covalent geometry and is well defined by the NMR data, resulting in low atomic coordinate precision for the backbone and all heavy atoms (Table 2). Structural validation data of the ASC structure obtained with MolProbity (41) in comparison to average values calculated for all NMR PYD and CARD structures deposited in the Protein Data Bank indicate that the structure of ASC is of comparable quality (Table 3). The equivalent resolution provided by PROCHECK-NMR (40) of the ASC structure compared with x-ray structures is between 1 and 1.8 Å (supplemental Fig. S3).

The NMR structure of full-length ASC shows two six-helix bundle domains (PYD and CARD) connected by a 23-residue-long linker (Fig. 3A). No interdomain NMR-derived contacts (NOEs) were observed, and neither domain shows NOEs with the linker; therefore, the PYD and CARD do not interact with each other. This result agrees with the previously observed propensity of death domains to participate in homotypic interactions within each subfamily (1) and the description of ASC as an adapter protein with two homophilic protein-protein interacting domains (51). Although long, the linker of ASC (residues 90–112) adopts some residual structure as evidenced by the presence of short-range NOEs. In addition, the ¹³C_α chemical shifts (29) deviate from random coil values. NOE data involving residues 90–94 (supplemental Table S2) and some positive ¹³C_α secondary shifts in this region (Fig. 4A) suggest that it adopts residual turn-type conformations. In contrast, ¹³C_α chemical shift deviations are almost consistently negative from residues 95 to 112 (Fig. 4A), pointing to the formation of low populated extended structures (52). These results are supported by the empirical program TALOS (38), which using a combination of ASC chemical shifts of several nuclei (¹⁵N amide, ¹³C_α, ¹³C_β, ¹³C′, ¹H_α) as input data, predicts most linker residues to populate extended structures (Fig. 4B). It is noteworthy that only five residues fall outside the extended structure region of the Ramachandran plot. Three of them (Gln-91, Gly-92, and Gly-94) belong to the fragment 90–94 that, based on NOE data, is likely adopting turn or short-helix conformations. The other two residues (Gly-99 and Gly-111) are comprised in the fragment 95–112 and fall in the left-handed helix region characteristic of Gly. The propensity of the linker to adopt extended structure is not surprising on the basis of its amino acid composition. Residues such as Ser, Ala, Gly, and Pro, present in the linker of ASC, are known to bias the polypeptide chain toward such conformations (53–54). In particular, two consecutive Pro residues (Pro-103 and Pro-104 in the linker of ASC) favor the polyproline II or collagen conformation (54), which is a common residual extended structure found in protein loops and linkers connecting domains in modular proteins (53).

In contrast to the structure of full-length ASC reported here, in the solution structure of the protein FADD, few interdomain NMR contacts have been observed between the DD and the DED (23). These interactions are not located in the consensus binding sites of each domain and are spatially close to the short linker (6 amino acids) connecting them. The linker length and flexibility could, thus, emerge as important factors in the structure and dynamics of the death domain superfamily.

The PYD of full-length ASC shares some general characteristics with other known PYD structures, including the long loop between helices 2 and 3 (Figs. 3A and 4), a feature of PYDs absent in other death domains. However, it is worth noting several differences with the isolated PYDs of NALP1 (55) and NALP10 (PDB entry 2DO9). These show a disordered loop in place of helix 3 in the PYD of ASC (Fig. 3B). Helices 1 and 6 are also shorter in the PYD of NALP1 (Fig. 3B). These structural differences result in high root mean square deviation values (9.7 Å for NALP1-PYD and 7.3 Å for NALP10-PYD) and might be related to their differences in biological function. In fact, NALPs contain additional domains and leucine-rich motifs, suggesting different roles in inflammasome formation. In contrast, the PYD-only protein ASC2 (56), which is thought to modulate ASC-mediated autoimmune response through PYD/PYD interactions (57), is significantly similar to the PYD of ASC at the structural level (root mean square deviation = 1.36 Å) and also displays the helix 3. Not surprisingly, the PYDs of both proteins share a high degree of sequence identity (64%). The PYD of full-length ASC and the isolated PYD (25) are structurally very similar as well (root mean square deviation = 1.37 Å). This result indicates that the presence of the CARD does not perturb the structure of the PYD in ASC and agrees with the absence of interdomain contacts.

Full-length ASC CARD shows structural peculiarities compared with other known CARDs. In all previously reported CARD structures, helix 1 is bent or broken into two smaller helices, named H1a and H1b. The hinge connecting these two fragments is involved in protein-protein interactions according to structural studies on the complex between Apaf-1-CARD and caspase-9-CARD (26, 27). Fragment H1a is missing in the CARD of ASC. Helix 1 spans residues Gln-117 to Val-126 and is preceded by a relatively ordered turn (His-113—Asp-116) (Fig. 3A). A comparison of the CARD structure of full-length ASC with the solution structure of Apaf-1-CARD (26), which is one example with the two H1a and H1b fragments, is shown in Fig. 3C. The H1a fragment displayed by Apaf-1-CARD corresponds to residues 108–116 in ASC. The region 108–112 is significantly flexible according to NMR relaxation data (see below), thus confirming the absence of H1a in ASC. The lack of H1a in the CARD of ASC might be related to its plasticity in protein-protein interactions that could facilitate participation in apoptotic and inflammatory events. In addition, large deviations in the orientation of helices 2 and 3 are also observed (Fig. 3C), which can result from the propagation of structural changes in the binding surface involving the connection between H1a and H1b. The electrostatic surface of full-length ASC CARD is significantly different from Apaf-1 CARD (Fig. 3D). ASC shows positively and negatively charged areas evenly spread throughout the surface, whereas Apaf-1-CARD shows two extensive oppositely charged patches (Fig. 3D). Within a similar fold, CARDs show structural differences pertaining to helix length, orientation (1), and electrostatic surface (examples are illustrated in supplemental Fig. S4), which might serve as a fine-tuned mechanism to tightly control the binding specificity observed in protein-protein interactions mediated by these domains.

Interdomain Dynamics in ASC

NMR relaxation of backbone amide (¹⁵N) measured as heteronuclear Overhauser values ({¹H}-¹⁵N NOE) as well as longitudinal (T₁) and transverse (T₂) relaxation times are affected by N-H bond dynamics and the molecule's rotational diffusion (58, 59). Residues adopting regular secondary structure show heteronuclear NOE values close to the theoretical maximum (∼0.83 at a spectrometer frequency of 600 MHz), whereas values lower than ∼0.65 are symptomatic of internal dynamics (43, 60). The average heteronuclear NOE values for the PYD and CARD regions of ASC are high and similar (0.78 ± 0.07 and 0.79 ± 0.08, respectively) as expected for two rigid structures (Fig. 5A). In contrast, heteronuclear NOE values decrease from the PYD C terminus and the CARD N terminus toward the linker center (Fig. 5A). These results indicate that the linker undergoes local motions on a fast time scale compared with molecular tumbling. These motions become increasingly restricted toward the connections to the DDs. Thus, the heteronuclear NOE data indicate that ASC comprises two well ordered rigid domains connected by a flexible linker.

The next step is to analyze the dynamic behavior of each domain relative to the other. Two extreme models for ASC interdomain dynamics can be envisioned; in the first, both domains tumble as a single rigid body, and in the second model, each domain is dynamically independent. Backbone amide ¹⁵N T₁/T₂ ratios are particularly useful in this type of analysis, as they are similar among the domains when they tumble as a whole and different otherwise (43). The ¹⁵N T₁/T₂ ratios of the PYD and CARD of ASC are noticeably different, indicating that they reorient at different rates (Fig. 5B). In addition, NMR relaxation-derived τ_c values of ASC individual domains are significantly larger than the predicted values (Table 1). Theoretical and NMR τ_c values of globular proteins are generally in very good agreement. As an example, Table 1 shows the theoretical and experimental τ_c values of the PYD-only protein ASC2 (56) of similar size and structure to ASC individual domains. The NMR τ_c values of both PYD and CARD in ASC are also larger than the experimental τ_c of ASC2, indicating that the former do not tumble independently. These results suggest that the PYD and CARD in ASC are in between the two extreme models, showing some interdomain flexibility and simultaneously sensing each domain the drag of the other.

This behavior is structurally illustrated by superimposing each individual domain of the NMR conformational ensemble (Fig. 6). The residual structural preferences of the linker result in a defined spatial interdomain organization that determines the orientation for binding of one domain relative to the other. Within this topological arrangement the flexibility of the linker increases the accessible space sampled by each domain, improving therefore, the chances to find interacting partners relative to proteins with two structurally fixed domains.

Interdomain motions caused by linker flexibility have been related to protein function. A classical example is the NMR study on the dynamic behavior of the two-domain protein calmodulin (43). This study shows that the long interdomain linker is highly mobile, supporting its role in calmodulin versatility to bind to multiple partners. Moreover, the linker flexibility is proposed to allow both protein halves to simultaneously interact with the target and to adopt the different domain orientations required in the formation of each complex. The binding requirements of ASC and calmodulin are analogous in that both proteins need to interact with different partners. Therefore, the flexibility of the linker could play a similar role in the interdomain dynamics of both proteins. Nevertheless, an important difference in the behavior of ASC could be that each domain binds at least one different target to form a particular complex. Like ASC, calmodulin domains show NMR relaxation-derived τ_c values that are significantly larger than the prediction (Table 1) albeit smaller than the correlation time expected for a globular protein of similar size (43). Interdomain motions in calmodulin have been further studied, resulting in the determination of the motion time scale (61). Examples of proteins with domain dynamics fitting the first extreme scenario explained above have also been reported (62). In this case the domains orient together because short or rigid linkers connect them or because they participate in interdomain contacts. In the death domain superfamily, FADD serves as an example of two domains orienting as a whole (23).

A Model for ASC Oligomerization

ASC forms homo- and hetero-oligomeric assemblies that are tightly bound to its function. NMR data and AFM images reported here indicate that ASC is also capable of self-associating in vitro (Figs. 1 and 2). Based on this information and the propensity of death domains to form homotypic interactions within subfamilies, it is possible to build a model to illustrate how ASC could oligomerize. The model (Fig. 7A) uses the structure of full-length ASC as the monomer template together with the binding interface of Apaf-1-CARD-caspase-9-CARD complex structure (27), which is currently the single three-dimensional CARD/CARD complex structure known. The structure of a PYD/PYD complex has not been determined up to date.

The CARD/CARD interaction is asymmetric, involving helices 1 and 4 of one CARD and helices 2 and 3 of the other (Fig. 7A). The asymmetry leaves two free binding sites in the dimer: helices 1 and 4 of one monomer and helices 2 and 3 of the other. The free binding sites allow additional CARD/CARD interactions, naturally leading to oligomerization. PYD/PYD interactions are suggested to also involve helices 1 and 4 of one PYD and helices 2 and 3 of the other (25). The PYD/PYD interface would be asymmetric and, therefore, would leave free binding sites upon dimer formation. Thus, the PYDs could as well participate in the self-association of ASC through homophilic interactions. The formation of the CARD/CARD and PYD/PYD interaction in the dimer model pre-establishes the relative binding orientation of the each domain to other partners (Fig. 7A). Moreover, in this model the pyrin domains are confined to a restrained space on top of the CARDs and do not obstruct the CARD/CARD interface (Fig. 7A). Interestingly, in the ASC structure the helices suggested to participate in the PYD/PYD interface are positioned as far as possible from the CARD and are, therefore, accessible for interactions with other partners (Fig. 7B). Thus, the CARD does not interfere with the PYD interface suggested to be involved in PYD/PYD interactions.

The CARDs central axes in the dimer model are positioned at an angle (Fig. 7A) that could result in the formation of a ring upon further association through the free interacting surfaces. The value of this angle (53.6°) is consistent with a ring composed of ∼6–7 monomers. Strikingly, electron microscopy data of the supramolecular apoptosome formed by Apaf-1 oligomerization show a 7-member CARD ring (63). In addition, the protein NALP1, which bears an N-terminal PYD and a C-terminal CARD flanking other domains, also forms a 7-fold symmetric ring as observed by electron microscopy (18). Both are rings of ∼12–13-nm outer diameter. Interestingly, the disk-like ASC oligomers (Fig. 2) of ∼12-nm diameter suggest that ASC could oligomerize into rings analogous to those formed by Apaf-1 and NALP1. However, ASC has been shown by confocal microscopy to also form rather large specks of ∼2 μm in diameter (8) and filaments (2). These results together with the AFM data reported here suggest that the oligomerization of ASC could be a complex process. It is likely that the disk-like oligomers observed by AFM further aggregate into larger assemblies.