Crystal Structure of the F27G AIM2 PYD Mutant and Similarities of its Self-association to DED/DED Interactions

Abstract

Absent in melanoma 2 (AIM2) is a cytoplasmic dsDNA sensor involved in innate immunity. It uses its C-terminal HIN domain for recognizing dsDNA and its N-terminal Pyrin domain (PYD) for eliciting downstream effects through recruitment and activation of Apoptosis-associated Speck-like protein containing CARD (ASC). ASC in turn recruits caspase-1 and/or caspase-11 to form the AIM2 inflammasome. The activated caspases process proinflammatory cytokines IL-1β and IL-18, and induce the inflammatory form of cell death pyroptosis. Here we show that AIM PYD (AIM2^PYD) self-oligomerizes. We notice significant sequence homology of AIM2^PYD with the hydrophobic patches of death effector domain (DED)-containing proteins, and confirm that mutations on these residues disrupt AIM2^PYD self-association. The crystal structure at 1.82 Å resolution of such a mutant, F27G of AIM2^PYD, shows the canonical six-helix (H1-H6) bundle fold in the death domain (DD) superfamily. In contrast to the wild-type (WT) AIM2^PYD structure crystallized in fusion with the large maltose-binding protein tag, the H2-H3 region of the AIM2^PYD F27G is well defined with low B-factors. Structural analysis shows that the conserved hydrophobic patches engage in a type I interaction that has been observed in DED/DED and other DD superfamily interactions. While previous mutagenesis studies of PYDs point to the involvement of charged interactions, our results reveal the importance of hydrophobic interactions in the same interfaces. These centrally localized hydrophobic residues within fairly charged patches may form the hot spots in AIM2^PYD self-association, and may represent a common mode of PYD/PYD interactions in general.

Inflammasomes are large supramolecular complexes responsible for sensing cytosolic danger signals associated with microbial infection or endogenous perturbations in the cell. The assembly of the inflammasomes is important for eliciting innate immune responses through the maturation and secretion of two inflammatory cytokines, IL-1β and IL-18, and may lead to an inflammatory form of cell death known as pyroptosis ^{1; 2; 3}. Inflammasomes are typically composed of an upstream sensor molecule, an adaptor protein known as Apoptosis-associated Speck-like protein containing CARD (ASC), and an effector molecule caspase-1 (and/or caspase-11). Upon recruitment into a full inflammasome, caspase-1 activates through dimerization and auto-proteolysis in order to process pro-inflammatory cytokines.

The sensor molecules, after which the inflammasomes are named, can be grouped into two families – NOD-like receptors (NLRs) and AIM2-like receptors (ALRs), based on their distinct domain architecture. There are a total of 23 NLRs in the human genome ². Most NLRs contain an N-terminal interaction domain (PYD, CARD, or BIR), followed by a nucleotide binding and oligomeriziation domain (NACHT or NBD), along with a C-terminal leucine-rich repeat (LRR) responsible for auto-inhibition ⁴. Only a handful of these NLRs are found to assemble into functional inflammasomes, including NLRP1, 6, 7, 12, and NLRC4, while the functions of the other NLRs remain to be elucidated. ALRs, which include absent in melanoma 2 (AIM2) and interferon inducible protein 16, also assemble functional inflammasomes ^{5; 6; 7; 8; 9}. They lack the NACHT domain found in NLRs and directly bind to dsDNA, a trigger associated with microbial invasion, though a C-terminal HIN domain ¹⁰.

The NLRP3, 6, 7, and 12 inflammasomes and the AIM2 and IFI16 inflammasomes require the bipartite adaptor molecule, ASC, which contains an N-terminal pyrin domain (PYD) followed by a C-terminal caspase recruitment and activation domain (CARD). ASC is recruited by homotypic PYD/PYD interaction with the activated upstream sensor. Oligomerization of the sensor molecules, either through oligomerization of the NACHT domain in NLRs ^{2; 11} or the release of an auto-inhibition mechanism in ALRs ^{10; 12}, initiates inflammasome assembly. To date, several PYD monomeric structures have been solved using crystallography and NMR. These include the solution structure of ASC^{PYD 13}, full-length (ASC^FL) ¹⁴, NALP1^{PYD 15}, ASC2 ¹⁶, NLRP10^{PYD 17}, NLRP7^{PYD 18}, NLRP12^{PYD 19}, NALP3^{PYD 20}, and MNDA^PYD. While our paper was in preparation, the crystal structure of WT AIM2^PYD as a fusion protein to maltose binding protein (MBP) was also reported ¹².

PYD belongs to the death domain (DD) superfamily, which also includes two other subfamilies – CARD and death-effector domain (DED) ²¹. Many PYDs also oligomerize, in addition to interactions with other PYD-containing proteins. This property is observed in DD signalosomes such as the MyDDosome and PIDDosome complexes ^{22; 23}. In other cases, these homotypic complexes can form extensive cytosolic signaling filaments, such as the DED filaments of caspase-8 ^{22; 23} and the CBM filamentous complex in NF-κB activation ²⁴. Therefore, the oligomerization property of AIM2^PYD may represent a suitable activation mechanism of the AIM2 inflammsome and warrants more careful studies. Here we used biochemical and structural studies to deduce the mode of AIM2 oligomerization.

Sequence similarity with DED and mutagenesis of AIM2^PYD

As a first step in the elucidation of AIM2 inflammasome formation, we decided to pursue the structure determination of the PYD of AIM2 (AIM2^PYD). However, AIM2^PYD (residues 1–100) was insoluble when expressed alone. Even when fused with the solubility tag Sumo, AIM2 still formed large aggregates. Upon searching for suitable mutation sites to solubilize and de-aggregate AIM2^PYD, we noticed a short, but significant, sequence homology between AIM2^PYD and the DEDs of FADD, v-FLIP and caspase-8 centered at the known conserved hydrophobic patches of DEDs (Figure 1(a)). In particular, the sequence homology placed F27 of AIM2 as the homologous residue for the surface exposed F25 of FADD, which has been shown to be important for apoptosis induction ²⁵. While F25W and F25Y essentially maintained the activity of WT FADD, F25G and F25V are almost completely compromised in the cell death induction function. The mutations on F25 disrupted FADD self-association ²⁶ and rendered the proteins monomeric ²⁵. They also compromised the ability of FADD to interact with the FLICE-inhibitory protein c-FLIP, a caspase-8-like protein ²⁷. In the tandem DED structure of a viral caspase-8-like protein known as the v-FLIP MC159 from the poxvirus Molluscum contagiosum virus, the F25-analagous residue in the similar hydrophobic patch of the first DED (DED1, F30) interacts with the second DED (DED2) to form a rigid dumbbell shaped structure ^{28; 29}. In caspase-8, the analogous single site F122G mutant exhibited weakened interaction with FADD, while the double mutant, F122G and L123G, completely abolished its interaction with FADD ²⁸.

Figure 1.

Sequence similarity between PYDs and DEDs, and identification of monomeric AIM2 PYD mutants. (a) Sequence alignment of human AIM2 PYD with human caspase-8 DED1 and DED2, and human FADD DED. The conserved hydrophobic positions are highlighted in red for that corresponding to type Ib and in green for that corresponding to type Ia interactions. F25 of FADD is analogous to F27 of AIM2, and has been shown to be important for apoptosis induction, self-association, and c-FLIP and caspase-8 interaction ^{25; 26; 27}. Additional residues that are identical with the AIM2 sequence are highlighted in yellow. Residue numbers are shown. (b) Gel filtration profiles of His-Sumo-AIM2^PYD WT and F25 mutant proteins expressed in E. coli using the pSMT3 vector. The proteins were purified using Ni-NTA columns. WT: mostly in the aggregation fraction; F27W, F27Y: in both aggregate and monomeric fractions; F27L: mostly in the monomeric fraction; F27G: only in the monomeric fraction. (c) Gel filtration profiles of His-Sumo-AIM2^PYD WT and the L10A/L11A mutant protein expressed in E. coli using the pSMT3 vector. (d) Sequence alignment among PYDs in the region around L10 and F27 of AIM2.

To determine if F27 of AIM2^PYD also impacts its solubility and aggregation solution behavior, we generated the F27Y, F27L, F27W and F27G mutants on the His-Sumo-AIM2^PYD construct. All mutations caused the proteins to shift to the monomeric fraction, at least to some extent (Figure 1(b)). The more conserved substitutions F27W and F27Y showed a mixture of aggregated and monomeric proteins, while the F27L mutant is mostly monomeric and the F27G mutant is completely monomeric. Upon removal of the Sumo-tag, the F27G AIM2^PYD mutant showed solubility greater than 20 mg/ml. These data suggest that F27 is involved in self-association of AIM2^PYD.

Based on the definition of the three types of asymmetric interactions that have been observed in the DD fold superfamily ^{21; 30}, F27 resides on a type Ib surface. Modeling and sequence alignment revealed that the corresponding type Ia surface likely includes residues L10 and L11 on the predicted helix H1 (Figure 1(a)). In previously reported interactions in the DD superfamily, the type Ia and the type Ib surfaces form a conserved asymmetric interaction pair ²¹. Indeed, the mutant L10A/L11A of AIM2^PYD also showed almost all monomeric species (Figure 1(c)), revealing that the type Ia and Ib surfaces are both involved in AIM2 self-association. Both the type Ia and the type Ib residues are also conserved among the different PYDs (Figure 1(d)), suggesting that the type I interaction may be a common feature in PYD/PYD interactions.

AIM2^PYD form ordered filaments in vitro and in cells

To assess whether AIM2^PYD forms ordered aggregates, we first used negative stain EM to examine purified His-Sumo-AIM2^PYD, which elutes in the void position of a Supderdex 200 column (Figure 1(b)). The protein showed filamentous morphology (Figure 2(a)). To confirm that AIM2^PYD also forms ordered aggregates in cells, we transiently transfected in 293T cells full-length, PYD domain, and HIN domain of AIM2 in fusion to eGFP at the C-terminus (AIM2^FL-eGFP, AIM2^PYD-eGFP and AIM2^HIN-eGFP). Confocal microscopy showed that while both AIM2^FL-eGFP and AIM2^PYD-eGFP formed filamentous structures, AIM2^HIN-eGFP did not, suggesting that PYD is responsible for the ordered aggregation (Figure 2(b)). We further transiently transfected in HeLa cells AIM2^PYD WT and Type I interface mutants identified above, as fusions to C-terminal mCherry. Confocal microscopy images showed filamentous aggregates for WT, F27Y, and F27W AIM2^PYD-mCherry, while F27G, F27L, and L10A/L11A AIM2^PYD-mCherry distributed throughout the cells (Figure 2(c)). These cellular data correlate well with migration positions in gel filtration chromatography (Figure 1(b)). Together, they suggest that bulky hydrophobic residues at the predicted Type I interface are essential for filament formation.

Figure 2.

AIM2 PYD forms filamentous aggregates. (a) EM images of purified His-Sumo-AIM2^PYD, which showed ordered filamentous structures. (b, c) Confocal microscopy images of C-terminal eGFP tagged AIM2 constructs (b) and of mCherry tagged AIM^PYD WT and Type I interface mutants (c). 0.8 μg of recombinant DNA was transfected into cells using Lipofectamine 2000 (Invitroggen) accordining to manufacturer’s instructions. Twenty-four hours post transfection, cells were fixed in 0.5% paraformaldehyde followed by nuclei staining using Hoechst 33342 (Molecular probes). The images were collected on an Olympus Fluoview FV1000 confocal microscope.

Crystal structure of F27G of AIM2^PYD

We set up the F27G mutant of AIM2^PYD (residues 1–100) and obtained initial crystals that diffracted to ~4Å resolution. Using in situ proteolysis with trypsin, needle clusters grew within 3 days at 4 °C in the same condition with improved resolution to ~1.8 Å (Figure 3(a)). The structure was determined using molecular replacement in the program MOLREP ³¹. A composite of several PYD structures was used as the search model because of the low sequence identity of AIM2^PYD to any known PYD structures. Model building and refinement at 1.82 Å was carried out using Coot ³² and PHENIX ³³ (Supplementary Table 1). The final atomic coordinates comprise one residue from the vector, and AIM2^PYD residues M1 to K93. It is likely that trypsin cleaved the C-terminal tail off at residue K93. This C-terminal region is involved in crystal packing and the cleavage must have facilitated crystal growth.

Figure 3.

Crystal structure of the AIM2^PYD F27G mutant. (a) Photograph of a crystal under a transmission light microscope. (b) A ribbon diagram of the structure in rainbow colors. Helices 1–6 are labeled and the location of F27G is shown in red. (c) A stick model of helix H2 showing the kink at G27. (d) Superposition of the crystal structures of F27G (yellow) with the WT in fusion with the MBP tag (cyan). (e) Zoom-up of the superposition in the H2-H3 region, showing the longer H2 in the F27G structure and the different conformations in this region. (f) B-factors along the protein sequence for both the F27G (yellow) and the WT (Cyan) structures, showing the low values in F27G even in the H2-H3 region. (g) Superposition of the WT AIM2-MBP structure (cyan) with the F27G structure (yellow). L10/L11 and F27 residues are shown as red sticks, showing that L10/L11 are packed against MBP to avoid aggregation.

The structure revealed a six-helical bundle structure comprised of helices H1 to H6, which is conserved in the DD fold superfamily ²¹ (Figure 3(b)). Mutation to Gly at F27 did not dramatically alter the structure; it only created a minor kink in helix H2 (Figure 3(c)). In comparison with the WT AIM2^PYD structure that was crystallized as a fusion to N-terminal maltose binding protein (MBP) ¹², helix H2 in the F27G mutant is in fact longer (Figure 3(d) and 3(e)). Most conspicuously, in the published WT AIM2^PYD structure, the H2-H3 region is highly disordered with high B-factors that go up to ~120 Ų while the remainder of the residues show B-factors of ~20–40 Ų (Figure 3(f)). In contrast, in our high resolution F27G mutant AIM2^PYD structure, this region is ordered and well defined; the B-factors of residues in this region fall within the average B-factor of 18.7 Ų in the whole protein chain (Figure 3(f), Supplementary Table 1). The protein conformation in the H2-H3 region is quite different in the WT and the F27G AIM2^PYD (Figure 3(e)), suggesting that the ordered F27G mutant provides a clearer structure of this region of AIM2. In comparison with the WT structure, the F27G AIM2^PYD structure shows differences in secondary structure boundaries (Figure 4(a)), offering an additional view on the intrinsic conformational flexibility of the molecule.

Figure 4.

Sequence and structural comparisons of AIM2^PYD. (a) Sequence alignment among AIM2^PYD from different species, with secondary structures from the F27G structure (pink) and the WT structure (blue). (b) An AIM2^PYD/AIM2^PYD type I interaction model based on the Myddosome structure. The interaction residues F27, L10 and L11 are shown as red sticks. (c, d) Ribbon and electrostatic diagrams of the AIM2^PYD surfaces containing F27 (c) and L10/L11 (d), respectively. (e) Superposition between AIM2^PYD (yellow) and MC159 DED2 (pink).

Modeled AIM2 PYD/PYD interaction: hydrophobic interactions may comprise the functional epitope

The locations of F27 on a type Ib surface and L10 on a type Ia surface, and the monomeric phenotypes of mutations on these residues, suggest that AIM2^PYD oligomerizes through a type I interface. Consistent with this observation, residues L10 and L11 are positioned adjacent to the fusion partner MBP in the WT AIM2^PYD structure ¹² (Figure 3(g)). Therefore MBP would inhibit the type I interaction of AIM2^PYD in the fusion protein, explaining how MBP inhibits AIM2^PYD aggregation.

We generated a PYD/PYD interaction model by superimposing AIM2^PYD onto a type I interaction pair in the Myddosome ^{21; 30} (Figure 4(b)). It has been proposed previously that AIM2^PYD interacts with ASC^PYD mainly through charge-charge interactions ¹². Previous extensive mutagenesis on ASC^PYD also suggested the involvement of charged interactions in ASC self-association and interaction with other PYD proteins such as NLRP3 and ASC2 ^{14; 34; 35}. In our model of the type I interaction in AIM2^PYD oligomerization, both the F27 containing and the L10/L11 containing surfaces are highly charged (Figure 4(c) and 4(d)), consistent with these existing data.

In addition to charge interactions, it has also been proposed that hydrophobic surfaces may be involved in the interactions mediated by AIM2^{PYD 12}. Here we observe that the hydrophobic residues F27 and L10 appear to reside in the center of the charged patches and are likely the major contributors of the interactions. The data provide additional structural and energetic insights because the strength of the hydrophobic interaction may be further enhanced by the local charges surrounding it. In addition, previous detailed mutational studies on protein-protein interfaces have shown that the buried hydrophobic contacts at the center of an interface may be responsible for a majority of the interfacial binding energy ³⁶. Therefore, the hydrophobic residues in AIM2^PYD self-association may represent the hot spot or the functional epitope in the interaction, which has been mostly overlooked and not emphasized in the existing literature.

Structural similarities between PYDs and DEDs

In addition to the local sequence homology with DEDs (Figure 1(a)), AIM2^PYD was also unexpectedly much more similar in structure to canonical DEDs, such as DED of FADD ²⁵ and DED2 of vFLIP ²⁸, than to CARDs and DDs. A DALI structural homology search ³⁷ found that the top hits are all PYD or DED structures (Supplementary Table 2, Figure 4(e)). This observation is in keeping with a phylogenetic analysis of DD superfamily, which suggests that DED and PYD share common ancestors with PYD derived latest in the evolution of these domains ³⁸. Therefore, the structural and evolutionary features dictate the similarities between DEDs and PYDs.

Highlights.

PYD oligomerization plays a pivotal role in inflammasome activation.

• Crystal structure of an AIM2 PYD mutant reveals conformational differences from WT.

• PYD and DED evolve similar hydrophobic interactions for self-association.

• Residues F27 and L10 provide hydrophobic interactions at the Type I interface.

• Interaction through oligomerization represents a ubiquitous signaling platform.