Cooperative assembly of IFI16 filaments on dsDNA provides insights into host defense strategy

IFI16 Cooperatively Binds dsDNA in a Length-Dependent Manner.

We hypothesized that the DNA length-dependent responses of IFI16 in vivo arise directly from differences in binding affinity for different-length DNA ligands but not from an as yet unidentified regulatory mechanism mediated by downstream effectors. To test this idea, we developed fluorescence-anisotropy (FA) binding assays using various fluorescein-amidite (FAM)-labeled DNA fragments. Monitoring FA of FAM-dsVACV72 (a 72-bp fragment derived from VACV; Fig. S1D) (3) with increasing concentrations of IFI16 resulted in a sigmoidal binding isotherm best fit with a Hill equation [an apparent binding constant (K_Dapp) of 65 ± 19 nM and a Hill constant of 1.9 ± 0.2; Fig. 1A]. This result indicated that full-length IFI16, unlike its isolated HIN200 domains (3, 9, 15), binds dsDNA in a cooperative manner. We then tested the binding of shorter dsDNA variants of VACV and HSV fragments that were used in the previous in vivo study (3). In contrast to predictions from the previously proposed model (9), we found that the binding affinity decreased with decreasing length of the dsDNA fragments independent of sequence (Fig. 1B, Fig. S1D, and Table S1). Importantly, the calculated Hill constants also decreased with shorter dsDNA (e.g., Fig. 1 B and C), suggesting that an absence of cooperativity results in weaker affinity (see also Table S1).

A possible underlying mechanism for the cooperative length-dependent binding is that IFI16 assembles into a distinct oligomer on dsDNA, because such a protein cluster would result in a more stable complex than the previously proposed noninteracting beads on a string (9). To test this idea, we first determined the stoichiometry between IFI16 and dsDNA ligands by performing FA experiments using FAM-dsDNA concentrations at least six times higher than their respective K_Dapp values (e.g., Fig. 1D). The plot of dsDNA length vs. the number of bound molecules revealed that about 15 bp are required to accommodate one full-length IFI16 (Fig. 1E), which agrees with a previous report suggesting that one HIN200 domain takes up about eight to nine bases. Notably, because no significant binding was observed from FAM-dsHSV15, these results also suggest that oligomeric IFI16 is required for tight binding (see also below). To confirm the length-dependent binding, we normalized the K_Dapp of each fragment to the number of available binding sites on each dsDNA fragment (e.g., four for dsHSV60). Plotting normalized K_Dapp vs. dsDNA-length revealed that the binding affinity of IFI16 still increases with increasing DNA length (Fig. 1F), supporting the idea that IFI16 cooperatively oligomerizes on dsDNA in a length-dependent manner. Importantly, these results correlate with the previously reported dsDNA length-dependent responses in vivo (3) and thus suggest that cooperatively assembling stable IFI16⋅dsDNA complexes via oligomerization is critical for regulating its cellular activity.

To investigate the DNA length-dependent binding mechanism further, we then performed competition binding assays in which increasing concentrations of various unlabeled DNA fragments were added to the FAM-dsVACV72⋅IFI16 complex (Fig. 2A). dsDNA fragments shorter than ∼30 bp did not compete up to the highest mass concentrations used in these assays (Fig. 2A). For example, at the same mass concentration (140 µg/mL), unlabeled dsVACV72 completely competed off the labeled DNA while dsHSV33 did not show any competition (Fig. 2A). All lengths of ssDNA failed to show significant inhibition (15–72 bases), indicating that full-length IFI16 preferentially binds dsDNA despite the reported ssDNA-binding activity of IFI16^HinA (15) [ssAG60 shown as an example (30 repeats of adenosine and guanosine); Fig. 2A]. Additionally, dsAG60 and dsHSV60 showed essentially the same IC₅₀ values (the concentrations of competitor DNA at the half-maximal inhibition), confirming that IFI16 binds dsDNA independent of its sequence (Fig. 2A). dsDNA fragments longer than 120 bp competed well at similar mass concentrations (Fig. 2B); however, the FA of bound FAM-dsVACV72 increased with low concentrations of these longer dsDNA fragments, suggesting that the size of IFI16⋅FAM-dsDNA increased (Fig. 2B). Closer inspection of these results revealed that signals increased up to the binding-site normalized concentration of each competitor equivalent to the amount of IFI16 used in these assays (200 nM; indicated by the arrow in Fig. 2B). Because we have a large excess of IFI16 to FAM-dsVACV72 (5 nM; 0.2 µg/mL), we interpreted the initial increase to be caused by free IFI16 molecules preferentially binding to subsaturating concentrations of the longer competitor dsDNA and the resulting complexes interacting with IFI16⋅FAM-dsVACV72 (see also below). The apparent increase of bound fraction would result in significantly overestimated IC₅₀ values for the longer fragments (thus true IC₅₀ << 10µg/mL). Nonetheless, these results consistently suggest that IFI16 prefers to bind dsDNA fragments significantly exceeding its footprint with a switch-like selectivity.

IFI16 Oligomers Are Distinct Protein Clusters.

In contrast to the previous model in which IFI16⋅dsDNA complexes resemble noninteracting beads on a string (9), our binding experiments suggested that IFI16 assembles into distinct oligomeric clusters on dsDNA. These two models can be distinguished by monitoring formation of FAM-dsDNA⋅IFI16 complexes in the presence of increasing amounts of unlabeled DNA using native gel electrophoretic mobility-shift assays (EMSAs): the clustering mechanism would show a concerted transition without resulting in significant intermediates, whereas the noninteracting mechanism would clearly display intermediate species (25, 26). We first performed EMSAs with various FAM-dsDNA fragments. Despite extensive efforts to optimize conditions, IFI16⋅dsVACV72/HSV60 complexes did not fully enter into the gel matrix. This was likely caused by the unusually high pI of IFI16 (pI: 9.3), as similar results were observed for AIM2 (pI: 9.8) and other DNA-binding proteins with high pI values (10, 27); thus we limit analyses of EMSA for qualitative purposes only. Nevertheless, judging by the disappearance of free FAM-dsDNA and by the formation of only a few transient intermediates, these results are consistent with our FA assays in which IFI16 binds dsDNA cooperatively and eventually form an oligomer on each dsDNA in a length-dependent manner (Fig. 3A). We then performed a competition EMSA in which increasing concentrations of unlabeled dsAG60 were added to FAM-dsVACV72⋅IFI16 (Fig. 3B). As predicted from the clustering model, we found that IFI16 bound dsVACV72 in a concerted manner without forming distinct intermediates. Importantly, the lack of intermediates in this assay indicates that IFI16 preferentially clusters even if there are nonadjacent excess binding sites (i.e., excess dsDNA). Next, to further confirm these results, we cross-linked IFI16 using a cysteine-specific cross-linker [1,4 bismaleimidyl-2,3-dihydroxybutane (BMDB)] in the presence or absence of dsDNA and visualized resulting complexes using SDS/PAGE; IFI16 has eight cysteines spatially distributed on the opposite side of the DNA-binding patches of its two HIN200 domains. As expected, we found that IFI16 forms oligomers almost exclusively in the presence of dsDNA (Fig. 3C). Further supporting the clustering model, IFI16 was still efficiently cross-linked in the presence of excess DNA (i.e., excess nonadjacent binding sites) (Fig. 3D).

IFI16 Oligomerizes on dsDNA in a Switch-Like Manner.

To quantitatively determine the oligomerization efficiency of IFI16 in the presence of excess DNA, we developed a fluorescence resonance energy transfer (FRET) assay by labeling one batch of IFI16 with a FRET donor and another with an acceptor. Importantly, because we add increasing amounts of excess DNA to a fixed concentration of IFI16, FRET signals are expected to arise only if IFI16 preferentially binds next to one another even in the presence of excess nonadjacent binding sites. Indeed, adding increasing amounts of unlabeled dsHSV60 or dsVACV72 to an equimolar (20 nM each) mixture of FRET-labeled IFI16 produced changes of the emission ratios between the donor (decrease) and acceptor (increase) that could fit to a Hill equation (we denote the midpoints of these curves as K_Fapp values, apparent oligomerization constants; Fig. 4 A and C and Table S2). No significant FRET signals were observed from ssDNA (e.g., ssVACV72) or dsDNA shorter than 60 bp (e.g., Fig. 4B). The amplitudes of the FRET ratios from dsDNA fragments longer than 72 bp were greater but otherwise did not differ significantly from one another, indicating that all labeled IFI16 molecules used in the assay were bound to these fragments to form similarly sized oligomers (Fig. 4B). After determining the K_Fapp value for each dsDNA fragment, we plotted relative binding efficiency (normalized to the K_Fapp of dsDNA2000) vs. dsDNA length to analyze how the oligomerization (binding) efficiency of IFI16 changes with respect to the length of dsDNA. This plot showed a cooperative relationship best fit with a Hill constant of 5.2 ± 0.6 and the optimal oligomerization efficiency (infliction point) around 150 bp (indicated by the arrow in Fig. 4C). Consistent with the results from our FA binding experiments, the Hill constant of about 5 in this plot indicates that a small fixed amount of IFI16 can dramatically amplify its oligomerization efficiency with increasing length of dsDNA in a switch-like manner. Moreover, the infliction point at 150 bp suggests that about ten IFI16 molecules comprise an optimal binding cluster (15 bp per one IFI16).

IFI16 Assembles into Filamentous Oligomers on dsDNA.

To define the structure of IFI16 oligomers on dsDNA, we examined IFI16 bound to various dsDNA fragments using negative-stain electron microscopy (EM). Electron micrographs of DNA-free IFI16 produced negatively stained particles of about 22 nm (Fig. 5A). This observation is consistent with the extended conformation (211 Å) seen from a previous small-angle X-ray scattering experiment (20) and confirms that apo-IFI16 does not assume an autoinhibited conformation (9) in which IFI16^PYD docks on one of the HIN200 domains. With various dsDNA fragments (IFI16:binding sites, 1:1), we found that IFI16 forms filaments corresponding to the length of dsDNA used in each experiment (Fig. 5 B–D). For example, dsDNA600 is predicted to be 193 nm long, and the protein filaments we observed using this fragment ranged from 190 to 210 nm (Fig. 5D). The width of filaments, however, was consistently between 20–25 nm regardless of dsDNA length, indicating that IFI16 oligomerizes laterally along the length of dsDNA. Interestingly, when we generated IFI16 filaments with higher concentrations of protein and dsDNA (3 vs. 1 µM; still keeping IFI16:binding sites at 1:1), we also found a number of extended filaments that appeared to form by conjoining either end of two or more individual filaments (Fig. 5C, Left vs. Right). Notably, formation of such higher-order oligomers is consistent with our competition binding experiments in which IFI16⋅FAM-dsVACV72 was thought to interact with IFI16⋅competitor-dsDNA complexes. Finally, as expected from our solution assays, IFI16 still assembled into filaments in the presence of excess dsDNA (Fig. 5E). The incomplete coverage of dsDNA in this experiment is likely attributable to multiple nucleation events, as seen from other filament forming nucleic acid sensors (i.e., “optimal” decamer formation, presumably 50–60 nm) (26). Taken together with our biochemical data, we concluded that IFI16 cooperatively clusters into filamentous oligomers on dsDNA.

Noncooperative dsDNA Binding by the HIN200 Domains of IFI16 Results in Much Weaker Affinity.

To identify the functional domain of IFI16 that drives filament formation, we generated the individual and tandem HIN200 domains of IFI16 (IFI16^HinA, IFI16^HinB, and IFI16^HinAB; Fig. S1) and assayed their binding to FAM-dsVACV72. Surprisingly, we did not observe significant dsDNA binding from these IFI16 variants (e.g., Fig. 6A, Upper; see also Fig. S2C and Fig. 7A). It was previously reported that dsDNA-binding affinity of IFI16^HinB was significantly influenced by buffer salt concentrations (9). Indeed, we found that the near-physiological salt concentrations (160 mM KCl) in our reaction buffer interfered with binding (Fig. 6A, Lower). Unlike full-length, the disappearance of free FAM-dsVACV72 and dsHSV60 was apparently noncooperative even under 60 mM KCl, suggesting that the DNA-bound complexes formed by the tandem HIN200 domains of IFI16 must resemble noninteracting beads on a string (Fig. 6 A and B). To detect binding of the individual HIN200 domains, we decreased the KCl concentration even further to 20 mM. Here, we found that all IFI16 variants but IFI16^PYD bound dsDNA in decreasing order of affinity: full-length, IF16^HinAB, IFI16^HinA, IFI16^HinB. Importantly, as expected from the EMSA, the fits to the isolated HIN200 domains binding to dsVACV72 were noncooperative (Hill constants, ∼1, Fig. 6C and Fig. S2 D and E). These results suggest the HIN200 domains of IFI16, individually or in tandem, do not efficiently bind dsDNA because they fail to cooperatively oligomerize.

Several additional experiments performed at 60 mM KCl confirmed that the tandem HIN200 domains of IFI16 do not oligomerize. First, in FA binding assays, we found that unlike full-length, IFI16^HinAB bound all FAM-dsDNA fragments noncooperatively and without apparent length-dependency (Fig. 6 D and E vs. Fig. 1). Second, in FA competition assays, all dsDNA fragments competed almost equally well without showing any apparent increase in bound fractions (Fig. 6F vs. Fig. 2). Third, in an EMSA competition experiment similar to Fig. 3B, IFI16^HinAB clearly displayed intermediates (Fig. 6G). Fourth, the cross-linking of IFI16^HinAB was minimal with various concentrations of dsDNA300 (Fig. 6H vs. Fig. 3 C and D). Fifth, no filament was observed in EM (Fig. 6I vs. Fig. 5B). Importantly, the lack of cross-linking even at the 1:1 complex suggests that IFI16^HinAB⋅dsDNA is organized differently than the one assembled by full-length. Finally, no significant FRET signals were observed from donor- and acceptor-labeled IFI16^HinAB using increasing concentrations of dsVACV72 (Fig. 6I vs. Fig. 4A). All of these results consistently support the idea that the HIN200 domains of IFI16 do not cooperatively cluster on dsDNA but independently bind dsDNA analogous to beads on a string.

The PYD of IFI16 Is Necessary for Cooperative DNA Binding.

Conventionally, the role of PYDs in ALRs is thought to be limited to recruiting downstream effectors (1, 19, 28). However, several independent experiments we have performed thus far suggested that IFI16^PYD is important for oligomerization on dsDNA. To test this possibility, we first confirmed that IFI16^PYD is a monomer and does not directly bind dsDNA (Figs. 6A and 7A; see also Fig. S1). Next, we constructed IFI16 variants consisting of either IFI16^PYD and IFI16^HinA (IFI16^∆B; residues 1–393) or IFI16^PYD and IFI16^HinB (IFI16^∆A; residues 1–191 plus 459–729) (see also Fig. S1). At 160 mM KCl, we found that IFI16^∆B efficiently binds dsVACV72 and IFI16^∆A binds dsVACV72 more weakly than IFI16^∆B, likely reflecting the weaker binding affinity of the IFI16^HinB in our hands compared with the previous report (3) (Fig. 7 A and B and Table S3). Importantly, unlike the IFI16 constructs without the PYD, the binding profiles were positively cooperative (Fig. 7B and Table S3). Additionally, IFI16^∆B did not show any intermediates in an EMSA competition assay (Fig. S2F).

PYDs are homotypic protein-protein interaction domains in which multiple binding surfaces are used in their interactions (29). For example, helices one and four dock on helices two and three to create a high-affinity oligomeric interaction (30) (see also Fig. 7D). To identify surface residues in IFI16^PYD important for the cooperative filament formation, we generated a set of mutations on the conserved surfaces residues using a homology model of IFI16^PYD based on the PYD of myeloid cell nuclear differentiation antigen (MNDA) (Protein Data Bank ID code 2DBG; 88% sequence similarity; Fig. 7 C and D). We found that mutating these surface side-chains on IFI16^PYD resulted in up to 25-fold increase in K_Dapp values for FAM-dsVACV72 compared with wild-type (Fig. 7E and Table S3). In addition, the binding profiles of IFI16^mut2 and IFI16^mut3 were essentially noncooperative (Hill constants: ∼1 for both; Fig. 7E), indicating that the cooperative dsDNA-binding mechanism requires intact IFI16^PYD. Also supporting our FA results, IFI16^mut3 bound FAM-dsVACV72 significantly weakly compared with wild-type with an altered migration pattern in an EMSA (Fig. 7F, Upper). Moreover, unlike wild-type, IFI16^mut3 showed intermediates in a competition ESMA (Fig. 7F, Lower). Finally, no FRET signal was observed between an equal mixture of donor-labeled IFI16^HinAB and acceptor-labeled full-length IFI16 with increasing dsVACV72 at 60 mM KCl (Fig. 7G). The lack of FRET signal in this experiment suggests that IFI16^PYD interacts with one another instead of interacting with adjacent HIN200 domains on the dsDNA scaffold. Taken together, we concluded that the conditional proximity induced upon encountering dsDNA by the HIN200 domains triggers IFI16^PYD-driven cooperative filament formation of IFI16.

What Could Be the Molecular Basis for DNA Length-Dependent Responses?

Conventionally, it was thought that IFI16 and its related ALRs prefer long dsDNA as it can accommodate a greater number of ALRs, which would be required to promote multimeric protein⋅protein interactions with their downstream partners (3, 9, 18). However, an important point to consider here is that if IFI16 binds dsDNA resembling beads on a string as previously thought, longer DNA fragments can be as counterproductive for multimeric interactions as short fragments. For example, even one femtogram of transfected 100-bp DNA fragments (≥9,000 copy number) provide more than 100,000 equally plausible nonadjacent binding sites for the HIN200 domains of IFI16. Furthermore, in principle, one HSV-1 genome contains more than 30,000 binding sites. Thus, because IFI16 lacks any DNA sequence specificity, without an intrinsic clustering mechanism mediated by protein⋅protein interactions, IFI16 would be randomly scattered on any self- or nonself-DNA (Fig. 8A) instead of selectively forming signaling foci on foreign DNA as observed in vivo (11–14). Our results suggest that the PYD-driven filament formation by IFI16 can provide an effective mechanism that allows multiple IFI16 (or its related ALRs) to bind adjacent to one another to form signaling foci even in the presence of excess DNA (Fig. 8B).

What Is a Potential Regulatory Mechanism of IFI16 in the Nucleus?

A striking behavior of IFI16 is that even though it is localized in the nucleus, it is randomly diffused without forming distinct foci with self-DNA (11–14). In fact, one of the longest-standing questions regarding IFI16 is how it does not engage self-DNA in the nucleus (11–14, 31, 32). We show here that the HIN200 domains of IFI16 possess negligible DNA-binding capacities under physiologically relevant salt concentrations [i.e., K_D >> 10 µM for the HIN domains in the ∼200 mM effective salt concentrations of the host cell nucleus (33)]. Importantly, the interaction between the HIN200 domains and dsDNA are exclusively mediated by the phosphate backbone [i.e., nonspecific electrostatics (9)]. Although useful in engaging a wide variety of foreign DNA, this intrinsically weak intermolecular force does not generate enough binding energy beyond formation of an encounter complex (34). Thus, despite exceeding the footprint of the HIN200 domains, the length of the exposed linker-dsDNA between nucleosomes [10–20 bp (21)] or even that of the transcription bubble [∼17 bases (22)] is too short to promote robust filament assembly of IFI16. Additionally, we envision that requiring oligomerization to achieve tight binding also plays a negative role in competing against replication/transcription machinery. Thus, our data suggest that the weak binding capacity of its HIN200 domains coupled with filament formation could sufficiently suppress its interaction with self-DNA.

How Can IFI16 Selectively Engage Foreign DNA and Assemble into Large Signaling Foci?

IFI16 is known to selectively colocalize with viral genomic DNA to form signaling foci within the nucleus (e.g., HSV-1, KSHV) (11–14). Although the total amount of viral DNA would never exceed that of self-DNA, there are two features of foreign DNA that are much more conducive to filament formation by IFI16. First, the entire naked DNA genome is exposed immediately after the invasion (35–37). Second, although the viral genome packages into chromatin with the host histones, it is less dense and much more loosely packed than the nuclear self-DNA (35–37). Importantly, we find that the relative binding affinity of IFI16 to various dsDNA fragments changes cooperatively not only with increasing IFI16 concentrations (Fig. 1) but also with increasing number of binding sites (dsDNA lengths; Figs. 3 and 4). These highly cooperative relationships suggest that IFI16 is capable of amplifying its clustering behavior in a switch-like manner. For example, the Hill constant of about 5 from Fig. 4C suggests that even the low basal amount of IFI16 (38) prefers by more than 2,000-fold to oligomerize on a 150-bp fragment rather than simply binding to a 15-bp fragment. Additionally, by these cooperative mechanisms, the binding efficiency diminishes in the same manner it is amplified; thus, these results suggest that IFI16 can clearly define an “on” or “off” state with respect to its concentration and the length of dsDNA. Collectively, the filament formation by IFI16 provides a compelling mechanism by which it could selectively engage foreign DNA while minimizing its interaction with self-DNA.

The Role of PYD.

In contrast to the autoinhibitory role of the PYD of AIM2 (9, 18), we find that IFI16^PYD plays an unexpected positive role in dsDNA binding. Several surface residues in IFI16^PYD that mediate the cooperative dsDNA binding of IFI16 are highly conserved (Fig. 7C), thus suggesting that other related ALRs use a common oligomerization mechanism. Interestingly, we find that the PYD-driven filament assembly could provide two important tactical advantages that might not be attainable by the previously proposed noninteracting model (9). First, it allows formation of an ordered array of IFI16^PYD oligomers on foreign DNA even with the low basal concentration of IFI16 in vivo (38). The IFI16^PYD cluster would then be instrumental for subsequent steps in IFI16 signaling pathways, because assembling PYD/CARD oligomers by receptor molecules is required for recruiting downstream partners (23, 24, 39). Second, the PYD-driven filament assembly ensures formation of a high-affinity IFI16⋅foreign DNA complex for signaling, which would also directly interfere with replication of pathogens. Indeed, IFI16 can directly suppress replication of human cytomegalovirus (HCMV) without downstream activation of IFN-β (40). The authors of this report (40) also showed that this intrinsic antiviral activity of IFI16 required its PYD, supporting the positive role in engaging foreign DNA in vivo. Finally, further consistent with the positive role of IFI16^PYD, very recent results published while this paper was under review showed that IFI16 oligomerizes on HCMV genome via its PYD in human fibroblasts and that a viral protein (pUL83) sequesters IFI16^PYD to evade the host immune response (41). Future studies could reveal how PYDs regulate the function of IFI16 and its related ALRs in more detail.

Potential Role of Extended Filaments.

A growing body of evidence suggests that cooperative assembly of supermolecular structures plays a critical role in regulating signal transduction pathways (23). Indeed, formation of large aggregates by other PYD/CARD-containing inflammatory signaling molecules has been observed under overexpression conditions (28, 30). However, we find that extending the size of filaments by conjoining two or more individual oligomers in an end-to-end manner is unique to IFI16 (Figs. 2B and 5C). In fact, this behavior of IFI16 also provides a tangible explanation for why IFI16 and relatively short DNA (60–70 bp) also cluster into a few large foci as observed from viral DNA (3). We envision that formation of such supermolecular structures would increase the cooperative nature of IFI16 signaling pathways, thus further enhancing its fidelity. Moreover, such oligomers may represent its aberrant state that could promote autoimmunity. First, it might provide means to amplify the signaling efficiency (and stability) of IFI16⋅dsDNA complexes formed by relatively short dsDNA fragments (e.g., apoptotic self-DNA). Second, such an abnormal structure could be recognized as a danger signal by other host immune receptors, leading to activation of inflammatory responses. Future studies could reveal the mechanism of the higher-order oligomer assembly and its function in more detail.

Filament Formation as a Broad Host Defense Strategy.

Conventionally, assembling filaments on foreign nucleic acids has been considered as a defense strategy reserved for sensing intracellular RNA (28, 42). For example, a principal mechanism by which melanoma differentiation associated protein (MDA)5 distinguishes self- from nonself-RNA is via cooperatively assembling into filaments along the length of long dsRNA with its two CARDs forming clusters tracing the center of filaments in a helical trajectory (25, 26, 42–45). IFI16 and MDA5 are unrelated proteins, and, not surprisingly, mechanisms allowing these proteins to assemble into filaments are significantly different. For example, the RecA-like RNA-sensing domain of MDA5 wraps around dsRNA and promotes filament formation in a head-to-tail manner; the CARD clusters are not required for filament assembly. In contrast, the DNA-binding domains of IFI16 do not directly promote filament formation, and yet its PYD is necessary to drive this process. Interestingly, our data suggest that an optimal binding unit consists of approximately ten IFI16 protomers (Fig. 4C). It is tempting to speculate that the filament assembly is then accomplished by propagating these decameric units along the length of dsDNA. Another difference between IFI16 and MDA5 is that the helicase domain regulates the lifetime of MDA5 filaments via ATP hydrolysis, although it is not clear what triggers the disassembly of IFI16 filaments. Further structural and kinetic studies could reveal the architecture and dynamics of IFI16 filaments in more detail. Overall, regardless of these mechanistic differences, our results suggest that assembling filaments of intracellular dsDNA/dsRNA receptors is not only an effective mechanism but also a broadly conserved host defense strategy against foreign nucleic acids.

Protein Expression and Purification.

Human IFI16 full-length (residues 1–729), IFI16^PYD (residues 1–93), IFI16^HinA (residues 192–393), IFI16^HinB (residues 518–729), IFI16^HinAB (residues 192–729), IFI16^∆A (residues 1–191 + 459–729), and IFI16^∆B (residues 1–393) were cloned into a pET21b vector (Novagen) using standard PCR methods. The identity of each IFI16 construct and mutant was confirmed by DNA sequencing. IFI16 constructs were expressed using T7 express cells (NEB). The cells were grown until OD₆₀₀: 0.5 at 37 °C and protein expression was induced by 0.2 mM isopropyl β-d-1-thiogalactopyranoside overnight at 18 °C. Each protein construct was purified using Ni²⁺-nitrilotriacetic acid (NTA) affinity chromatography (Qiagen), followed by cation exchange chromatography using a Resource S column (GE Healthcare) with a NaCl gradient (50 mM to 1 M) and Superdex-200 or Superdex-75 size-exclusion chromatography (GE Healthcare). Micrococcal nuclease (NEB; 100 units per 1 L of cells) was added during cell lysis to eliminate bacterial DNA contamination. All proteins were eluted as monomers in size-exclusion chromatography (Fig. S1A), and UV-visible spectroscopy confirmed that they were free from nucleic acid contamination (Fig. S1B). Each protein was greater than 95% pure (Fig. S1C) and concentrated and stored at −80 °C in 20 mM Hepes⋅NaOH at pH 7.4, 200 mM NaCl, 1 mM EDTA, 5 mM DTT, and 10% (vol/vol) glycerol.

DNA.

Each FAM-labeled and unlabeled DNA fragment shorter than 120 bp was commercially purchased from Integrated DNA Technologies (IDT; Fig. S1D). Duplex formation was achieved by mixing the sense and the complementary strands in 1:1 molar ratio in buffer A (see below) and heating in 95 °C for 10 min. The duplex samples were subsequently cooled to room temperature (25 ± 2 °C) on the bench top. The longer dsDNA fragments were generated by PCR using a plasmid. Under our reaction conditions, all duplex DNAs are at least 15 °C below their predicted melting temperatures (SciTools; IDT).

Biochemical Assays.

Unless noted otherwise, all assays were performed at room temperature (25 ± 2 °C) with 40 mM Hepes⋅NaOH at pH 7.4, 160 mM KCl, 1 mM EDTA, 2 mM DTT, 10% glycerol, and 0.1% Triton-X100 (referred to as buffer A). The FA of FAM-labeled DNA (at least 15- to 20-fold lower concentrations than the K_Dapp for each experiment) was monitored with increasing concentrations of IFI16 variants by using a Tecan M1000 plate reader (excitation at 470 nm and emission at 528 nm) and 384-well nonbinding plates (Corning), and FA values were determined by iControl data analysis software (Tecan). The concentration of each FAM-dsDNA was kept at least 20-fold less than its respective K_Dapp in each of our binding assay to prevent any length bias (typically 2.5–5 nM per assay). For EMSA, each DNA was incubated with indicated concentrations of IF16 variants for 10 min and the bound complexes were separated from unbound dsDNA using 6% Tris⋅boric acid-EDTA (TBE) gels with 0.25× TBE as a running buffer. The results were then visualized by scanning on a Typhoon imager (GE Healthcare; excitation at 488 nm; emission at 532 nm). FRET assays were also performed in a 384-well plate format using the Tecan M1000. All experiments were performed at least three times independent of one another, and errors were calculated by using the SDs (number of independent experiments, n ≥ 3). Data were fit to a Hill, quadratic, or competition form of binding equation using Kaleidagraph (Synergy Soft).

Protein Cross-Linking and Labeling.

IFI16 constructs were incubated with or without indicated dsDNA fragments for 20 min in buffer A minus glycerol and with DTT replaced by tris(2-carboxyethyl)phosphine (TCEP). BMDB (10.2-Å spacer arm) was then added to each sample and cross-linking reactions proceeded at room temperature for 20 min. After quenching the reactions with 50 mM DTT, the final mixtures were applied to 4–15% TGX SDS/PAGE (Bio-Rad) and visualized by silver-stain.

IFI16 and IFI16^HinAB were labeled using twofold molar excess of Dylight 550 maleimide or Dylight 650 maleimide (Pierce) in buffer A with DTT replaced by TCEP. Labeling reactions were quenched by adding excess β-mercaptoethanol and the labeled-proteins were purified using Ni²⁺-NTA chromatography followed by size-exclusion chromatography. The labeling efficiency of IFI16 to each dye was approximately one to one, and there was no conceivable difference in FRET data from two different batches of labeled proteins.

Electron Microscopy.

To preserve the stability of IFI16 oligomers during EM experiments, we first cross-linked IFI16⋅dsDNA complexes using BMDB. The resulting complexes were adsorbed to glow discharged carbon grids for 2 min, then blotted and transferred through two consecutive drops of 1% uranyl formate or 1% uranyl acetate for a total of 1–2 min. The carbon film was then quickly dried by aspiration. Images were collected with a Philips BioTwin CM120 (FEI) transmission electron microscope, operating at 80 kV, and digitally captured with an XR-80 CCD. The images from control experiments show no protein contamination in our buffer or DNA samples (Fig. S2B).