MAVS polymers smaller than 80 nm induce mitochondrial membrane remodeling and interferon signaling

Double‐stranded RNA (dsRNA) is a potent proinflammatory signature of viral infection and is sensed primarily by RIG‐I‐like receptors (RLRs). Oligomerization of RLRs following binding to cytosolic dsRNA activates and nucleates self‐assembly of the mitochondrial antiviral‐signaling protein (MAVS). In the current signaling model, the caspase recruitment domains of MAVS form helical fibrils that self‐propagate like prions to promote signaling complex assembly. However, there is no conclusive evidence that MAVS forms fibrils in cells or with the transmembrane anchor present. We show here with super‐resolution light microscopy that MAVS activation by dsRNA induces mitochondrial membrane remodeling. Quantitative image analysis at imaging resolutions as high as 32 nm shows that in the cellular context, MAVS signaling complexes and the fibrils within them are smaller than 80 nm. The transmembrane domain of MAVS is required for its membrane remodeling, interferon signaling, and proapoptotic activities. We conclude that membrane tethering of MAVS restrains its polymerization and contributes to mitochondrial remodeling and apoptosis upon dsRNA sensing.


Introduction
Recognition of viral nucleic acids by innate immune receptors is one of the most conserved and important mechanisms for sensing viral infection. Many viruses deliver or generate double-stranded RNA (dsRNA) in the cytosol of the host cell. Cytosolic dsRNA is a potent proinflammatory signal in vertebrates. Endogenous dsRNAs are modified or masked through various mechanisms to prevent autoimmune signaling, and genetic deficiencies in these dsRNA modification pathways can cause autoimmune disorders [1][2][3]. Cytosolic dsRNA is primarily sensed by the RIG-I-like receptors (RLRs) RIG-I (DDX58), MDA5 (IFIH1), and LGP2 (DHX58) [4], which activate the mitochondrial antiviral-signaling protein (MAVS) [5][6][7][8]. RIG-I recognizes dsRNA blunt ends with unmethylated 5 0 -di-or triphosphate caps [9][10][11][12]. MDA5 recognizes uninterrupted RNA duplexes longer than a few hundred base pairs [11,13]. LGP2 functions as a cofactor for MDA5 by promoting the nucleation of MDA5 signaling complexes near dsRNA blunt ends [14,15]. Binding to dsRNA causes RIG-I to form tetramers and MDA5 to cooperatively assemble into helical filaments around the dsRNA [16][17][18][19]. RIG-I and MDA5 each contain two N-terminal caspase recruitment domains (CARDs). The increased proximity of the CARDs upon RLR oligomerization induces the CARDs from four to eight adjacent RLR molecules to form a helical lock-washerlike assembly [19,20]. These helical RLR CARD oligomers bind to MAVS, which has a single N-terminal CARD, via CARD-CARD interactions [5]. Binding of MAVS CARDs to RLR CARD oligomers nucleates the polymerization of MAVS CARD fibrils with amyloid-like (or prion-like) properties including resistance to detergents and proteases [19][20][21]. MAVS polymerization is required for signaling, and the spontaneous elongation of MAVS fibrils following nucleation is thought to provide a signal amplification mechanism [21]. MAVS fibrils then recruit proteins from the TRAF and TRIM families to form multimeric signaling platforms, or signalosomes [21]. MAVS is localized primarily on the outer mitochondrial membrane [5] but can also migrate via the mitochondria-associated membrane (MAM) to peroxisomes [22], which function as an alternative signaling platform to mitochondria [23]. MAVS signalosomes activate both type I interferon (through IRF3) and NF-jB-dependent inflammatory responses [11,13,21]. Overexpression of MAVS induces apoptotic cell death, and this proapoptotic activity is dependent on its transmembrane anchor (TM) and mitochondrial localization, but independent of the CARD [24,25]. A loss-of-function MAVS variant is associated with a subset of systemic lupus patients [26].
In the current model of MAVS signaling, RLR CARD oligomers trigger a change of state in the CARD of MAVS, from monomer to polymeric helical fibril. MAVS fibrils grow like amyloid fibrils by drawing in any proximal monomeric MAVS CARDs [20]. This model is based partly on the observation that purified monomeric MAVS CARD spontaneously assembles into fibrils of 0.2-1 lm in length [19,21]. The fibrils, but not the monomers, activate IRF3 in signaling assays [21] with cell-free cytosolic extracts. Moreover, a purified MAVS fragment lacking the TM can, in its polymeric fibril form, activate IRF3 in crude mitochondrial cell extracts that contain endogenous wild-type MAVS [17]. However, this signaling model is based primarily on signaling assays and structural studies performed in a cell-free environment with soluble fragments of MAVS lacking the TM. MAVS has been reported to form rod-shaped puncta on the outer mitochondria membrane upon activation with Sendai virus [27], but evidence that MAVS forms polymeric fibrils in cells remains inconclusive, and furthermore MAVS fibrils are not sufficient for signaling. Indeed, the MAVS TM is required for interferon induction and cell death activation [5,21,25], and several viruses including hepatitis C virus suppress type I interferon production by cleaving it off [8,[28][29][30][31]. The sequence between the CARD and TM of MAVS, which represents 80% of the MAVS sequence, is also required for downstream signaling [32]. How this sequence and the TM function together with the CARD in cell signaling remains unclear.
In the cellular context, MAVS CARD fibrils are subject to multiple physical constraints, including tethering to RLR-dsRNA complexes (at one end of the fibril) and to the mitochondrial membrane (via the TM of each MAVS molecule in the fibril). Here, we address the question of how the current model of MAVS signaling can be reconciled with these physical constraints, and the requirement of the TM, for cell signaling. Imaging of MAVS signaling complexes by super-resolution light microscopy with effective optical resolutions of up to 32 nm reveal that in the cellular context MAVS signaling complexes are significantly smaller than expected-no more than 80 nm. Moreover, MAVS signaling is associated with remodeling of mitochondrial compartments and apoptosis, and both of these activities are dependent on the TM of MAVS. Our data indicate that MAVS forms smaller signaling complexes than previously thought [21,27].

MAVS activation by cytosolic RNA induces mitochondrial membrane remodeling
The polymerization of purified soluble fragments of MAVS into helical fibrils is well documented [19,21]. In the cellular context, however, MAVS is tethered to the outer mitochondrial membrane via its TM, and binds via its CARD to oligomeric or polymeric RLR-dsRNA complexes [19,21,33]. To examine how the physical constraints imposed by membrane tethering and association with RIG-I or MDA5 may affect MAVS CARD fibril formation, we imaged cells containing active MAVS signaling complexes by superresolution fluorescence microscopy. Mouse embryonic fibroblasts (MEFs) and 3T3 cells were imaged by structured illumination microscopy (SIM) and stimulated emission depletion microscopy (STED). Because fluorescent proteins fused to the N terminus, C terminus, or juxtamembrane region of MAVS were not suitable for STORM, cells were labeled with a monoclonal antibody against a linear epitope within residues 1-300 of MAVS and a fluorescently labeled secondary antibody. An antibody with an overlapping epitope was shown previously to recognize MAVS in the fibril form in nonreducing semidenaturing electrophoresis [21]. Immunofluorescence of MAVS and TOM20, an outer mitochondrial membrane marker, showed that the two proteins localized to the same mitochondrial compartments (Fig. 1). We observed changes in the distribution of MAVS and in overall mitochondrial morphology (using the TOM20 marker) upon infection of 3T3 cells with a West Nile virus (WNV) replicon (Fig. 1A). Infection with the WNV replicon caused mitochondrial compartments to form more fragmented and less filamentous structures closer to the nucleus. Introducing the dsRNA mimic poly(I: C) into MEFs by electroporation (0.6-1 pg per cell) recapitulated the changes in MAVS distribution and mitochondrial morphology observed upon infection with the WNV replicon (Fig. 1B). These changes were also recapitulated by cotransfecting MEFs derived from MAVS knockout mice (MAVS KO MEFs) [34] with poly(I:C) RNA and a plasmid encoding MAVS (0.6-1 pg of each per cell, Fig. 1C). Mitochondrial remodeling associated with poly(I:C) treatment was quantified as statistically significant reductions in: (a) the distance from the nucleus (from an average of 9.0 to 4.5 lm, P = 2.0 9 10 À4 ) (Fig. 1E); (b) the fraction of the cytosolic area occupied by mitochondria (from an average of 24.2% to 16.7%, P = 0.0015) (Fig. 1F); and (3) the length of mitochondrial compartments measured as unbranched segments of skeletonized TOM20 fluorescence (from an average of 2.42 to 1.98 lm, P = 0.0478) (Fig. 1G). The amount of MAVS plasmid and the electroporation method used in transfections were selected to yield MAVS expression levels that were in the physiological range (see below). Notably, no MAVS filaments longer than the resolution limit were observed. The resolution of SIM and STED was approximately 110 and 80 nm in the imaging (xy) plane, respectively (and 350 and 600 nm in z, respectively). These resolutions were also sufficient to resolve differences in the positions of individual MAVS and TOM20 protein complexes so that the immunofluorescence signals from the two proteins formed an alternating pattern within mitochondrial compartments rather than strictly colocalizing (Figs 1A,C and 2A). We confirmed that transfection with poly(I:C) RNA induced translocation of IRF3 to the nucleus, which is a hallmark of interferon-b (IFNb) signaling (Fig. 2B). Importantly, we showed that the level of MAVS expression in transfected MAVS KO MEFs was comparable to the physiological level of endogenous MAVS expression in wild-type MEFs (Fig. 3). A live-cell dual-luciferase reporter assay was used to confirm that IFN-b signaling is activated in MAVS KO MEFs transfected with the MAVS expression plasmid and poly(I:C) RNA (see below). Costaining for MAVS and MDA5 (Fig. 2C) showed an increased interaction between the two proteins, defined as the average distance between MAVS and MDA5 fluorescence (Fig. 2D) [35]. However, MDA5 staining remained predominantly cytosolic and no significant increase in colocalization was detected with poly(I:C) treatment based on the Pearson correlation. Similarly, RIG-I was recently shown to partition into a MAVSassociated mitochondrial fraction and a cytosolic stress granule fraction [36]. In the absence of MAVS, mitochondria retained their filamentous morphology and failed to move toward the nucleus upon transfection with poly(I:C) (Fig. 1D). We note that the immunofluorescence and cell signaling data show relatively high levels of mitochondrial remodeling ( Fig. 1), IRF3 nuclear translocation (Fig. 2B), and background signaling (see below) in cells transfected with a control plasmid instead of poly(I:C), which is likely attributable to IFN-b signal transduction by cytosolic DNA sensors. We conclude that activation of MAVS by cytosolic RNA sensing is associated with remodeling of mitochondria into more globular and perinuclear compartments. Mitochondrial remodeling from the healthy filamentous morphology to perinuclear globular compartments through mitochondrial fission events is a hallmark of apoptosis [37]. Indeed, MAVS was shown previously to promote apoptosis independently of its function in initiating interferon and NF-jB signaling [24].

STORM shows MAVS signaling complexes are smaller than expected
Purified monomeric MAVS CARD spontaneously forms fibrils 0.2-1 lm in length [19,21]. The fibrils, but not the monomers, activate IRF3 in cell-free assays [21]. However, SIM and STED imaging of cells with actively signaling MAVS did not resolve any clearly apparent fibrils (Fig. 1). To determine whether MAVS forms fibrils too small to resolve by SIM or STED, we employed a higher resolution imaging modality, stochastic optical reconstruction microscopy (STORM), to image cells containing active MAVS signaling complexes. STORM can yield effective resolutions of 20 nm in the imaging plane [38]. MEFs were immunolabeled with an antibody against MAVS and a secondary antibody conjugated to Alexa Fluor 647, which was selected for its favorable blinking characteristics and high photon output. MAVS KO MEFs were cotransfected with MAVS and poly(I: C) RNA as described for SIM and STED imaging. No MAVS filaments longer than the resolution limit were observed in the STORM images (Fig. 4). Instead, most of the MAVS immunofluorescence was present in fluorescent foci with irregular shapes, varying in diameter from 30 to 80 nm. These dimensions coincide with the effective resolution of the STORM imaging (see below). Unexpectedly, despite the global change in mitochondrial morphology associated with poly(I:C) treatment, there was no significant difference in the shapes and size range of MAVS foci in cells transfected with poly(I:C) or with a control plasmid.
Quantitative image analysis indicates MAVS fibrils are shorter than 80 nm Since our ability to visualize submicrometer MAVS fibrils is critically dependent on the effective resolution of the imaging experiment, an accurate measurement of the imaging resolution is necessary to determine the minimum fibril length that can be resolved. We measured the resolution of our STORM images with the Fourier ring correlation (FRC) method, using an FRC of 0.143 as the threshold to measure resolution [39]. This criterion is the widely accepted standard for resolution assessment in cryo-electron microscopy (cryoEM) [40][41][42]. FRC curves calculated from our STORM images indicate that the resolution in the imaging plane ranged from 32 to 66 nm (Fig. 5A,B). Moreover, there was a correlation between the measured resolution of the images and the visual appearance of the MAVS foci. More specifically, images with the lowest resolutions (Fig. 4G,A) had more diffuse density, whereas images with the highest resolutions ( Fig. 4E,I,C) had more pronounced foci, regardless of poly(I:C) treatment. This suggests that the appearance of clearly visible foci is determined by the imaging resolution rather than by the size of the imaged object.
The calculated FRC resolution of the lowest resolution STORM image was 66 nm. This is consistent with the SIM and STED data ( Fig. 1A,C), which showed no correlation between poly(I:C) treatment and the appearance of clearly distinguishable MAVS fluorescence foci.
To quantify any subtle effects that poly(I:C) treatment might have on the clustering of MAVS foci, we performed cluster analysis on the STORM images using the pair correlation function. A hierarchical cluster model of two nested clusters was necessary to obtain a good fit to the pair correlation curve (Fig. 5C,D) [43,44]. The clustering analysis showed that the two-cluster size parameters of the nested cluster model were not significantly different in poly(I:C)treated and control samples (Fig. 5E), confirming the lack of correlation between poly(I:C) treatment and the size of MAVS fluorescent foci.
Reconstruction of fine structural features by superresolution microscopy depends on the precision with which the position of each fluorophore can be localized, and on the density and spatial distribution of active fluorophores along the labeled sample [39]. In the case of immunolabeled samples, the primary and secondary antibodies increase the spacing between the molecule of interest and the fluorophore significantly (by 20-35 nm) [45]. To determine whether 200-nm MAVS fibrils would in principle be visible in STORM images with the localization precision, labeling strategy and fluorophore properties inherent to our study, we generated an atomic model of 200-nm MAVS CARD fibrils bound to primary and secondary antibodies, and simulated the appearance of the fibrils with the same localization precision as our STORM images at different labeling efficiencies (Fig. 4K,L). Immunolabeling increases the overall diameter of the fibrils from and TOM20 in NIH 3T3 cells infected with a GFP-labeled West Nile reporter virus (WN replicon), with immunofluorescently labeled MAVS in green and TOM20 immunofluorescence in red, or with the MAVS staining, TOM20 staining or the GFP infection marker shown separately in gray. Yellow results from green-red overlap and is indicative of MAVS-TOM20 colocalization. Scale bars: 10 lm in the overview panel, 1 lm in the inset panels. (B) Confocal images of MAVS and TOM20 in wild-type MEFs, with MAVS (green) and TOM20 (red) immunofluorescence, or with TOM20 staining shown separately in gray. Cells were transfected with either poly(I:C) RNA (+poly(I:C)) or an empty plasmid (Àpoly(I: C)) to control for the effects of transfection. Scale bar: 10 lm. (C) Structured illumination microscopy (SIM) of MAVS immunofluorescence (green) in MAVS KO MEFs cotransfected with MAVS and poly(I:C) RNA (+poly(I:C)), or MAVS and a control plasmid (Àpoly(I:C)). Scale bars: 10 lm in the overview panel, 2 lm in the inset panel. (D) Confocal images of TOM20 in MAVS KO MEFs transfected with either poly(I:C) RNA (+poly(I:C)) or an empty plasmid (Àpoly(I:C)) but no plasmid encoding MAVS, with immunolabeling of MAVS (green) and TOM20 (red), or with TOM20 staining shown separately in gray. DAPI nuclear staining is shown in blue (panels B-D). Scale bar: 10 lm. (E) Average distance of TOM20 fluorescence from the nucleus, defined as DAPI fluorescence. (F) Quantification of the fraction of the cytosolic area occupied by TOM20 fluorescence, from bright field immunofluorescence imaging (1009 magnification) of 39 cells [28 Àpoly(I:C), 11 + poly (I:C)]. (G) Average length of mitochondrial compartments measured as the length of unbranched segments of skeletonized TOM20 fluorescence, from the same images used in (F). Statistical significance in panels (E-G) was calculated using a one-sided t-test. Statistical significance was assigned as follows: *P < 0.05; **P < 0.01; ***P < 0.001, n = 32. 8.5 to 79 nm, but fibrils were nevertheless clearly visible with labeling efficiencies > 40%. Notably, the diameter of immunolabeled MAVS CARD fibrils is similar to the lowest resolution measured for the STORM images (79 vs. 66 nm, respectively). Therefore, immunolabeled MAVS fibrils shorter than 80 nm in axial length would be expected to appear as globular foci in STORM. Hence, the absence of visible fibrils in our images remains consistent with the presence of helical MAVS fibrils up to 70-80 nm. A MAVS CARD helical assembly of this length would contain 136-156 MAVS molecules based on the 5. 13 A axial rise per protomer [19]. By comparison, purified MAVS fragments form filaments 200-1000 nm in length in solution [19,21].

MAVS TM is required for mitochondrial remodeling and IFN-b signaling
The MAVS CARD fibrils are sufficient to activate IRF3 in cytosolic extracts, but the transmembrane domain (TM) of MAVS is absolutely required for MAVS to activate IRF3 and induce interferon [5,21]. We have shown that MAVS signaling activation causes changes in overall mitochondrial morphology similar to those associated with apoptosis, consistent with the documented proapoptotic activity of MAVS, which is dependent on the TM but not the CARD of MAVS [24]. To determine whether the MAVS TM is required for MAVS-dependent mitochondrial remodeling, we used STED microscopy to image MAVS KO MEFs transfected with a plasmid-encoding MAVS with the TM deleted (MAVS-DTM). As reported previously [5], we found that MAVS-DTM had diffuse cytosolic staining (Fig. 6A). Notably, cells expressing MAVS-ΔTM showed no visible aggregation, and little or no mitochondrial remodeling upon induction with poly(I: C), consistent with a recent report that the TM is required for the formation of high molecular weight MAVS aggregates [46]. Previous studies have measured MAVS signaling activity from cytosolic or mitochondrial cell extracts. We confirmed that MAVS KO MEFs transfected with wild-type MAVS and poly(I:C) following the same protocol used for super-resolution imaging induced IFN-b signaling in the dual-luciferase reporter assay (Fig. 6B). In contrast, cells expressing MAVS-ΔTM failed to activate IFN-b signaling. The signal-to-noise ratio was low in the assay, however, due at least in part to induction of IFN-b signaling by cytosolic DNA-sensing pathways such as cGAS-STING [47] in response to the transfected plasmid DNA. We therefore performed the luciferase reporter assay in STING KO MEFs (Fig. 6B), which are defective for cGASdependent DNA sensing [48]. The signal-to-noise was higher with STING KO MEFs than with MAVS KO MEFs despite the presence of endogenous MAVS in the STING KO MEFs. A slight but statistically insignificant increase in signaling was observed in STING KO MEFs transfected with MAVS-ΔTM. This is consistent with previous work showing that purified recombinant MAVS-ΔTM can, in its aggregated form, induce aggregation of endogenous wild-type MAVS and IRF3 activation in cell extracts enriched for mitochondria [21].

MAVS induces cell death in response to cytosolic RNA
An early hallmark of apoptosis is the depolarization of the inner mitochondrial membrane [49], which is followed at later stages of cell death by loss of nuclear DNA content due to DNA fragmentation [50]. Overexpression of MAVS in HEK293T cells was shown previously to induce apoptosis [24]. To determine whether MAVS KO MEFs expressing physiological levels of MAVS induced apoptosis in response to activation with cytosolic dsRNA, we conducted cell death assays on cells transfected with MAVS and poly(I:C) RNA following the same protocol as for super-resolution imaging. Inner mitochondrial membrane depolarization and loss of nuclear DNA content were measured with the fluorescent markers 3,3 0 -dihexyloxacarbocanine iodide (DiOC6) (Fig. 6C) and propidium iodide (PI) (Fig. 6D), respectively. We found poly(I:C) treatment induced a loss of mitochondrial membrane potential in one third of cells transfected with wildtype MAVS, and loss of nuclear DNA content indicative of cell death in 21% of cells, at 16 h posttransfection, the same time point used for super-resolution imaging (Fig. 6C,D). In contrast, poly(I:C) treatment of cells transfected with MAVS-ΔTM induced a loss of mitochondrial membrane potential and DNA content in only 13% and 8% of cells, respectively.

Implications for signal transduction by MAVS
Taken together, our super-resolution light microscopy data suggest that in live cells, MAVS signaling complexes contain significantly smaller fibrils than those formed in vitro by soluble MAVS fragments. The absence of visible fibrils in SIM, STED, and STORM images of cells expressing physiological levels of MAVS activated with various stimuli (synthetic or viral dsRNA), and a quantitative assessment of the resolution of the STORM images are all consistent with an upper limit in the range of 70-80 nm for the longest dimension of any MAVS fibrils within the MAVS signalosome. Unexpectedly, despite the global change in mitochondrial morphology associated with poly(I:C) treatment, there were no significant differences in the shape and distribution of MAVS foci in cells transfected with poly(I:C) or with a control plasmid. Cluster analysis of STORM images revealed no significant correlation between poly(I:C) treatment and the size of the clusters. Our confocal, SIM, STED, and STORM data are consistent with each other and broadly consistent with previous fluorescence microscopy studies [21,51], although a 3D SIM reconstruction in one previous study showed MAVS forming rod-shaped puncta ranging from 100 to 650 nm in length, with a median length of 350 nm (n = 74) [27]. However, 350 nm is similar to the axial resolution of SIM and hence could be an overestimate of the MAVS cluster length.
An 80-nm-long MAVS CARD helical assembly would still contain 156 MAVS molecules, a sufficiently large number to retain potential for polymerizationdependent signal amplification and to function as a mitochondrial signaling platform capable of recruiting the necessary number and diversity of downstream signaling proteins to elicit a robust IFN-b response [19,20]. Given the smaller than expected size of MAVS signaling assemblies, a maximal MAVS-dependent signaling response may require a larger number of fibril nucleation events than previously thought. However, these nucleation events need not be independent, and MAVS signaling complexes may nevertheless assemble cooperatively, with assembled complexes promoting the nucleation of additional complexes without MAVS CARD filament extension beyond approximately 80 nm, but potentially forming two-or threedimensional networks of microfibrils. Such a trade-off of reduced filament length in favor of increased nucleation was recently shown to occur elsewhere in the dsRNA-sensing pathway. Indeed, LGP2 increases the initial rate of MDA5-dsRNA binding and limits MDA5 filament assembly, resulting in the formation of more numerous, shorter MDA5 filaments to generate a greater signaling activity [14]. Imaging at higher resolution, for example, by electron microscopy, is required to elucidate the structural organization of MAVS signaling complexes.
We have shown that the MAVS TM is required for MAVS-associated mitochondrial remodeling and confirmed that the TM is required for MAVS-dependent IFN-b signaling and apoptosis in live cells and under the same experimental conditions used for the superresolution imaging experiments in this study. The diffuse cytosolic localization of MAVS-ΔTM suggests that MAVS oligomerization is regulated by elements between the CARD and TM, and that the TM is required to overcome this regulation. It is tempting to speculate that the physical constraints imposed by MAVS aggregation and association with RLR-RNA complexes exert a physical force on the outer mitochondrial membrane capable of inducing distortions that could conceivably contribute to mitochondrial remodeling or leakage associated with apoptotic cell death.

West Nile Virus infection
West Nile reporter virus particles were generated in HEK 293T cells by cotransfecting the cells with plasmids encoding the structural proteins (C, prM, E) and a subgenomic replicon containing GFP (pWNVII-Rep-G-Z), as described previously [52,53]. The plasmids were kind gifts from T. Pierson (NIH). Supernatant collected 48 h after transfection was filtered through 0.2-lm membranes and used to infect 3T3 cells. Twenty-four hours postinfection, the cells were fixed with 4% formaldehyde solution and labeled for immunofluorescence.   Fig. 4 (see Materials and methods). An FRC value of 0.143, indicated by a horizontal purple line, was used as the threshold to measure resolution. (B) Effective resolution of each STORM image calculated from the FRC curves (FRC = 0.143). Cluster analysis of MAVS immunofluorescence in the STORM images. (C) Each image was divided into tiled 3 9 3 lm windows covering the field of view. Windows with a mean fluorescence intensity greater than half the mean intensity of the whole image were selected for analysis (shaded in blue). (D) A nested two-cluster model (red curve) fitted to the pair correlation function (PCF, blue curve), shown for representative individual cluster analysis windows from images of poly(I:C)-treated and control samples. (E) Box plots of the four parameters fitted in the cluster analysis, l 1 , r 1 , l 2 , r 2 , as defined in the Materials and Methods. The cluster density parameters, l 1 and l 2 , were normalized against the mean intensity of the cluster analysis window, I win . Statistical significance in panels was calculated using a one-sided t-test.

1553
The

Mitochondrial remodeling analysis
Widefield images were acquired at 1009 magnification with an inverted Nikon TE2000 microscope equipped with a Niji LED light source (Bluebox Optics, Huntingdon, UK), a 1009/1.49NA oil objective (Nikon, Tokyo, Japan), and NEO scientific CMOS camera (Andor, Belfast, UK). A custom IMAGEJ macro script (Mitochondria_remodelling_analysis.ijm, Data S1) was written to analyze mitochondrial distance, area, and length parameters with IMAGEJ [54]. More specifically, distance from the nucleus was measured as the shortest distance from each TOM20 fluorescence pixel to a DAPI fluorescence pixel. Second, the fraction of the cytosolic area occupied by mitochondria was measured in each cell. Third, the length of mitochondrial compartments was measured by skeletonizing the mitochondrial fluorescence signal, segmenting the skeleton at branchpoints and measuring the length of each resulting segment. These measurements were performed on 28 cells without poly(I:C) and 11 cells with poly(I:C) treatment.

MAVS-MDA5 interaction analysis
The averaged distances of the collective points in the MDA5 and MAVS fluorescence signals after treatment with poly(I:C) or a control plasmid was calculated with the Interaction Analysis plugin for IMAGEJ from MosaicSuite (MOSAIC Group, Dresden, Germany) [35]. Colocalization was measured by calculating the Pearson correlation coefficient with the Coloc 2 plugin in IMAGEJ.

Structured illumination microscopy
Cells were plated as a monolayer on high-performance Zeiss cover glasses (18 9 18 mm, 0.17 mm thickness, type 1.5 H, ISO 8255-1 with restricted thickness-related tolerance of AE0.005 mm, refractive index = 1.5255 AE 0.0015) in six-well CytoOne TC-Treated tissue culture plates (STAR-LAB, Milton Keynes, UK). Cells were stained and mounted as described above. Images were acquired on a Zeiss ELYRA S.1 system with a pco.edge 5.5 scientific CMOS camera using a 639/1.4 NA oil immersion objective lens. For optimal lateral spatial sampling a 1.69 intermediate magnification lens was used, resulting in a pixel size of 64 nm. Excitation was achieved with a 488-nm laser and emission was filtered with a 495-550-nm band pass filter. Modulation of the illumination light was achieved using a physical grating with 28-lm spacing. This patterned illumination was shifted through five phases at each of three rotational angles per image. Raw data were processed using ZEN software (Zeiss, Jena, Germany).

Stimulated emission depletion microscopy
Cells were plated as a monolayer on high-performance Zeiss cover glasses (as above) in six-well CytoOne TC-Treated tissue culture plates (STARLAB). Cells were stained and mounted as described above. Images were acquired on a Leica TCS SP8 STED confocal system with time-gated HyD GaAsP detectors using a 1009/1.4 NA oil immersion objective lens. Excitation was achieved with an NKT Super K pulsed white light (470-670 nm) laser tuned to 488 nm and STED was induced with a 592-nm laser (~40 MWÁcm À2 ). A time gate window of 2-6 ns was used to maximize STED resolution. About 20-nm pixels were used to ensure adequate spatial sampling to support maximal expected resolution. Images were deconvolved with Huygens Professional.

Stochastic optical reconstruction microscopy
In STORM, fluorophores are induced to switch, or 'blink' between fluorescent and dark states. With a sufficiently small fraction of fluorophores in the fluorescent state at any given time, spatial overlap between adjacent fluorophores is avoided and the positions of individual fluorophores can be determined with high precision using the point-spread function of each fluorophore [38]. STORM images are generated by plotting the positions of each fluorophore blinking event as points in the image plane and applying to each point a blurring factor corresponding to the localization precision for that point. The appearance of the images also depends on the efficiency of fluorescent labeling and the extent of fluorophore bleaching during the experiment. Cells were plated as a monolayer on high-performance CellPath (Newtown, UK) HiQA Coverslips (No. 1.5 H, 24 mm Ø) in six-well CytoOne TC-Treated tissue culture plates (STARLAB). Cells were stained as described above. Alexa Fluor 647-labeled samples were imaged in the following buffer condition: 10% w/v glucose, 0.5 mgÁmL À1 glucose oxidase (Sigma-Aldrich, St. Louis, MO, USA), 40 lgÁmL À1 catalase (Sigma Aldrich), and 0.1 M b-mercaptoethylamine (MEA) (Sigma-Aldrich) in PBS, pH adjusted to 7.4 with concentrated HCl. To minimize air oxidation of the fluorophore, the imaging well was filled to full capacity with the imaging buffer and sealed with a cover glass. Imaging was performed with an N-STORM microscope (Nikon) equipped with an Apochromat TIRF 1009/1.49 NA oil immersion objective lens, a single-photon detection iXon Ultra DU897 EMCCD camera (Andor), and 405 nm (30 mW) and 647 nm (170 mW) laser lines. To acquire single-molecule localizations, samples were constantly illuminated with the 647 nm laser at 2-3 kWÁcm À2 . To maintain adequate localization density, the 405-nm laser was used to reactivate the fluorophore. During imaging, Perfect Focus System was used to maintain the axial focal plane. 7-10 9 10 4 frames were collected for each field of view at a frame rate of 50-70 Hz. STORM images were reconstructed with NIS Elements Advanced Research (Nikon): lateral drift correction was performed using automated cross-correlation between frame sets and each localization point in the reconstructed image was shown as a normalized Gaussian, the width of which corresponded to the localization uncertainty calculated using the Thompson-Larson-Webb equation [55]. Each STORM image contained 1-2 million molecules within a single cell.

Fourier ring correlation
The effective optical resolution of the STORM images was assessed using the FRC method proposed by Nieuwenhuizen et al. [39] using the MATLAB code provided (frc_analysis.m, Data S2). An FRC threshold of 1/7 (0.143) was used to determine the resolution of each rendered STORM image.

Cluster analysis of MAVS immunofluorescence in STORM images
Cluster analysis was performed to assess the size of fluorescent foci from immunolabeled MAVS in the STORM images. Using a custom C++ script (nestedclusteranalysis.cpp, Data S3), each image was divided into 3 9 3 lm windows covering the field of view and windows with a mean fluorescence intensity greater than an arbitrary thresholdhalf of the mean fluorescence intensity of the whole imagewere selected for analysis (Fig. 5C-E). Cluster analysis was performed by fitted a hierarchical nested cluster model, or clusters of clusters model, to the pair correlation function (or radial distribution function), defined as the average number of points located in a ring of radius r centered around each point and normalized by the expected intensity taking into account the border of the window of analysis [44,56]. The nested cluster model was fitted to the pair correlation curve using a least squares approach. The selected model comprised two clusters (smaller 'inner' clusters that cluster into larger 'outer' clusters) defined by four parameters (l 1 , r 1 , l 2 , r 2 ) through the following equation: with r 1 and r 2 corresponding to the outer and inner cluster sizes, respectively.

Simulated rendering of immunolabeled MAVS fibrils
The atomic coordinates of an mouse antibody molecule (PDB code 1IGY) [57] were manually docked with UCSF CHIMERA [58] to an arbitrary epitope on the MAVS CARD coordinates (PDB code 3J6J) to simulate a primary antibody-MAVS complex. Two additional antibody molecules were then docked to two different arbitrary epitopes in the constant region (Fc domain) of the primary antibody. The helical symmetry of the MAVS CARD filament determined by cryo-electron microscopy (cryoEM) image reconstruction [19] was then applied consecutively to the atomic coordinate of the complex to generate a 200-nm MAVS CARD filament with bound primary and secondary antibodies. The simulated epitopes and orientations of the antibodies were selected to minimize steric clashes after the helical symmetry of the MAVS CARD filament was applied. A custom C++ code was used to render the atoms of the lysine residues, the sites of fluorophore conjugation, of each secondary antibody in the filament. Each atom was rendered on a two-dimensional image at a 5-nm scale with a given observation probability ranging from 20% to 100%, in order to emulate different antibody labeling efficiencies. The resulting image was then blurred in order to take into account the localization uncertainty of the STORM singlemolecule localizations events, approximately 10 nm.

Dual-luciferase reporter cell signaling assay
The MAVS KO MEFs or STING KO MEFs were transfected with wild-type MAVS or MAVS-ΔTM and poly(I:C) with the Neon transfection system as described above, using the same conditions as for light microscopy, except that plasmids encoding firefly luciferase under an interferon-b (IFN-b) promoter and Renilla luciferase under a constitutive promoter (Promega, Madison, WI, USA) were also included in the transfection. Cells were transfected and seeded in quadruplet sets for each condition in 96-well clear bottom black polystyrene microplates (Corning, Corning, NY, USA) at 5-6 9 10 3 cells per well. Cells were lysed and the lysates transferred to all-white, 96-well flat solid bottom plates suitable for tissue culture and luminescence measurements (Greiner Bio-One, Kremsm€ unster, Austria). Sample preparation from cell lysates and luciferase luminescence measurement were performed according to the manufacturer's protocol for the Dual-Luciferase Reporter Assay System (Promega). IFN-b-dependent induction of firefly luciferase was measured in cell lysates 16 h postinduction with a PHERAstar (Ortenberg, Germany) FSX microplate reader with dual sample injection. IFN-b signaling activity was measured as the ratio of firefly luciferase luminescence to Renilla luciferase luminescence.

Quantification of cell death
Depolarization of the inner mitochondrial membrane was quantified by staining cells with 40 nM of 3,3 0 -dihexyloxacarbocanine iodide (DiOC6; Sigma-Aldrich) or 6 mgÁmL À1 propidium iodide (PI) for 1.5 h in PBS. DiOC6 and PI fluorescence was quantified by flow cytometry. Quantification was set at 5000 for cell count and FL1 (530 nm) and FL3 (670 nm) were used to acquire DiOC6 and PI signals, respectively. Data were acquired on an LSR II flow cytometer (BD Biosciences, San Jose, CA, USA) and analyzed with FLOWJO (TreeStar, Ashland, OR, USA).

Statistical analysis
No statistical methods were used to predetermine sample size, experiments were not randomized, and the investigators were not blinded to experimental outcomes. Unless otherwise noted, errors are presented as the standard deviation of the mean of four replicates conducted in a single independent experiment. Statistical significance was calculated with PRISM 8 (GraphPad Software, San Diego, CA, USA) using an unpaired t-test without prior assumptions regarding the standard deviations of each set of quadruplet measurement. Statistical significance was assigned as follows: *P < 0.05; **P < 0.01; ***P < 0.001, n = 4.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Data S1. IMAGEJ macro script used to analyze mitochondrial distance, area, and length parameters.