Cryo-EM structure of an active bacterial TIR–STING filament complex

Stimulator of interferon genes (STING) is an antiviral signalling protein that is broadly conserved in both innate immunity in animals and phage defence in prokaryotes1–4. Activation of STING requires its assembly into an oligomeric filament structure through binding of a cyclic dinucleotide4–13, but the molecular basis of STING filament assembly and extension remains unknown. Here we use cryogenic electron microscopy to determine the structure of the active Toll/interleukin-1 receptor (TIR)–STING filament complex from a Sphingobacterium faecium cyclic-oligonucleotide-based antiphage signalling system (CBASS) defence operon. Bacterial TIR–STING filament formation is driven by STING interfaces that become exposed on high-affinity recognition of the cognate cyclic dinucleotide signal c-di-GMP. Repeating dimeric STING units stack laterally head-to-head through surface interfaces, which are also essential for human STING tetramer formation and downstream immune signalling in mammals5. The active bacterial TIR–STING structure reveals further cross-filament contacts that brace the assembly and coordinate packing of the associated TIR NADase effector domains at the base of the filament to drive NAD+ hydrolysis. STING interface and cross-filament contacts are essential for cell growth arrest in vivo and reveal a stepwise mechanism of activation whereby STING filament assembly is required for subsequent effector activation. Our results define the structural basis of STING filament formation in prokaryotic antiviral signalling.

Stimulator of interferon genes (STING) is an antiviral signalling protein that is broadly conserved in both innate immunity in animals and phage defence in prokaryotes [1][2][3][4] . Activation of STING requires its assembly into an oligomeric filament structure through binding of a cyclic dinucleotide [4][5][6][7][8][9][10][11][12][13] , but the molecular basis of STING filament assembly and extension remains unknown. Here we use cryogenic electron microscopy to determine the structure of the active Toll/interleukin-1 receptor (TIR)-STING filament complex from a Sphingobacterium faecium cyclic-oligonucleotide-based antiphage signalling system (CBASS) defence operon. Bacterial TIR-STING filament formation is driven by STING interfaces that become exposed on high-affinity recognition of the cognate cyclic dinucleotide signal c-di-GMP. Repeating dimeric STING units stack laterally head-to-head through surface interfaces, which are also essential for human STING tetramer formation and downstream immune signalling in mammals 5 . The active bacterial TIR-STING structure reveals further cross-filament contacts that brace the assembly and coordinate packing of the associated TIR NADase effector domains at the base of the filament to drive NAD + hydrolysis. STING interface and cross-filament contacts are essential for cell growth arrest in vivo and reveal a stepwise mechanism of activation whereby STING filament assembly is required for subsequent effector activation. Our results define the structural basis of STING filament formation in prokaryotic antiviral signalling.
Activation of STING signalling results in assembly of oligomeric filament structures that have been observed in both human innate immunity and bacterial antiphage defence [4][5][6][7][8][9][10][11][12][13] . Key early findings supporting STING oligomerization as a required step of activation include observation of STING puncta formation in cells 9 , electrophoresis analysis of STING multimeric complexes 10,11 and artificial activation of STING on fusion to multimerization domains 12 . More recently, insight into the structural basis of mammalian STING oligomerization has been obtained through analysis of STING crystal packing 7 and cryogenic electron microscopy (cryo-EM) structures of tetrameric STING complexes 5,13 . Strict conservation of filament formation in prokaryotic STING signalling suggests that prokaryotic and metazoan STING signalling domains share an ancient mechanism of signal induction 4 . To define the molecular basis of STING filament formation, we reconstituted signalling of the S. faecium TIR-STING (Sf STING) antiphage effector in vitro and used single-particle cryo-EM to determine the structure of the activated complex ( Fig. 1a and Extended Data Figs. 1, 2, 3 and 8). In response to the nucleotide second messenger c-di-GMP produced during cyclic oligonucleotide-based antiphage signalling system (CBASS) immunity, Sf STING rapidly assembles into oligomers that form single filaments and antiparallel double-filament structures that make supra-molecular contacts between STING and TIR domains of opposing filaments (Extended Data Figs. 1 and 2). The TIR domains are not as well resolved in the main double-filament class, probably owing to conformational heterogeneity, and we therefore focused structural analysis on the single-fibre filaments. A 3.3-Å-resolution cryo-EM reconstruction of the dominant class of single-fibre filaments reveals that Sf STING oligomerizes through formation of a repeating laterally translated array of parallel stacked protein dimers that buries more than 3,000 Å 2 of surface area between two pairs of dimers and locks the STING cyclic dinucleotide (CDN)-binding domain and associated TIR effector domains into filamentous assemblies capable of reaching greater than 300 nm in length (Fig. 1a).

Article
contacts with the guanosine Hoogsteen edge, mirroring interactions required for high-affinity complex formation between human STING and the cGAS product 2′,3′-cGAMP (refs. 14 , 15 ; Fig. 1c). The partially closed 3′,3′-cGAMP-bound Sf STING conformation allows incomplete oligomerization and is incompatible with stable filamentous packing beyond about 4-6 units (or about 20 nm). Compared to the fully active Sf STING-c-di-GMP filament structure, 3′,3′-cGAMP recognition induces partial closure of the lid domain and an overall conformation of the STING CDN-binding domain (CBD) that probably weakens contact sites observed for c-di-GMP-induced filaments. Additionally, a lack of well-defined density for the TIR domains in the 3′,3′-cGAMP filaments suggests conformational flexibility that may impact stability of the filaments. Sf STING binds to c-di-GMP with about 300 nM apparent affinity (K d ) and a low nanomolar concentration of c-di-GMP is sufficient to initiate robust NADase activity 4 . Our previous results demonstrate that 3′,3′-cGAMP binds with slightly weaker affinity (about 700 nM K d ) and is unable to induce robust TIR NADase activity, findings that are now explained by our cryo-EM analysis. Thus, complete closure of the lid domain and structural compression around the high-affinity ligand c-di-GMP is essential to translate CDN recognition into a conformation sufficient to seed STING protein-filament formation and downstream signal induction.
In the Sf STING-c-di-GMP complex, individual dimeric units adopt a two-fold symmetric conformation and form the basic repeating building block of the filament structure (Fig. 1a, right). In each Sf STING dimer unit, the canonical V-shaped STING CBD is positioned above two TIR enzymatic NADase domains that dock against the base of the receptor ( Fig. 1a and Extended Data Figs. 4 and 5). The Sf STING CBD contains a unique β-hairpin insertion immediately following the stem dimerization helix, but otherwise adopts the same minimized fold and highly conserved CDN-binding pocket previously observed in crystal structures of Flavobacteriaceae sp. and Capnocytophaga granulosa bacterial STING (ref. 4 ). In the active-state Sf STING filament structure, a short linker sequence connects the α-helix stem of each STING domain to the TIR effector domain (Extended Data Fig. 4e). Previously, structures of a TIR-STING homologue from the oyster Crassostrea gigas and the human transmembrane domain-containing STING in inactive states revealed a twisted linker sequence that connects the effector domain to the STING domain located across the dimeric interface 4,5,13 (Extended Data Fig. 4e). The active-state conformation of the chicken Gallus gallus STING-2′,3′-cGAMP complex exhibits a parallel linker orientation similar to the Sf STING filament, suggesting that parallel linker orientation is a defining feature of both prokaryotic and metazoan STING activation 5,6 .

Mechanism of Sf STING oligomerization
TIRs are widespread NADase effector domains encoded in CBASS, Pycsar and Thoeris antiphage defence systems 4,16,17 , but no previous TIR active-state structures exist to explain the mechanism of NAD + hydrolysis. The Sf STING TIR domain is most closely related to plant immune proteins and secreted bacterial effectors that catalyse glycosidic bond hydrolysis during immune defence and interspecies conflict [17][18][19][20][21] . The Sf STING residues F83, F85, L87 and L89 within the highly conserved TIR helix αC interface form extensive hydrophobic packing interactions that bridge the dimeric unit, and the Sf STING TIR domain also contains a unique βD′ and βE′ strand insertion in the TIR 'CC loop' that further extends the dimer interface ( Fig. 2a-c and Extended Data Fig. 5). Structural comparison with the human SARM1 TIR-ribose structure demonstrates that the Sf STING NAD + -binding pocket is formed by two regions: a set of hydrophobic residues, F6, W33, F37 and L47, positioned to stack the substrate nicotinamide; and a set of hydrophilic residues, S10, R78 and N80, positioned to coordinate the phosphodiester linkage and ribose of the adenosine base 19 (Fig. 2d). In addition to the highly conserved Sf STING catalytic residue E84, the NADase active site is completed by residue D110 from the opposite TIR dimer mate ( Fig. 2d and Extended Data Fig. 5).
The active Sf STING-c-di-GMP structure reveals a series of proteinprotein interfaces that explain a shared mechanism of STING filament formation. The primary STING filament interface occurs along two surfaces that pack between adjacent Sf STING dimeric units and drive lateral head-to-head oligomerization ( Fig. 3a and Extended Data Fig. 6). These surfaces centred around the hydrophobic residues V280and A309 are exposed in the closed Sf STING domain conformation, explaining a key mechanism that couples c-di-GMP recognition and filament nucleation (Extended Data Fig. 6). Individual STING-domain protomers (STING a and STING b ) within the Sf STING filament are also bridged by an electrostatic interaction between STING a R307 and STING b E290 (Fig. 3b). Notably, the previous cryo-EM structure of a chicken STING-2′,3′-cGAMP tetramer contains residues Q278 and D280 involved in a similar interaction and hydrophobic surfaces packed along the same STING-STING protein interface, revealing remarkable conservation of an ancient mechanism of STING oligomerization 5 (Fig. 3a,b). In the full Sf STING filament structure, the STING domains are more tightly packed compared to those in the minimal chicken/human STING tetramer model. Additionally, a modest approximately 2° shift between packed Sf STING dimeric units is observed in both the single-fibre and wrapped double-fibre cryo-EM reconstructions, resulting in the active Sf STING filament structure adopting a slight curve (Extended Data Fig. 1). Assembly of the complete Sf STING filament allows formation of a second cross-filament interface between the STING domain residues N278 and Q279 and residue E95 in the TIR domain associated with the adjacent protomer ( Fig. 3c and Extended Data Fig. 6). Cross-filament TIR a -TIR b interactions are also formed between two flexible Sf STING TIR loops (BB loop: P32 a -G43 a ; and DD loop: A101 b -K118 b ) that stack on top of one another (Fig. 3d). Comparison of the active Sf STING-c-di-GMP complex with the inactive apo C. granulosa bacterial STING structure reveals substantial rearrangements in the STING domain necessary to enable reorganization and TIR-domain packing 4 (Extended Data Fig. 4). Close packing is required to allow the TIR D110 loop to reach across and complete the dimer-mate active site, providing an explanation for how Sf STING filament formation triggers NADase domain activation (Figs. 2d and 3d and Extended Data Fig. 5).

Filamentation controls NADase activity
We next combined the bacterial STING filament structure with biochemical and cellular analysis of Sf STING function to establish a molecular model of STING activation. Measuring degradation of a fluorescent NAD + analogue, we observed that Sf STING alterations predicted to disrupt STING-STING, STING-TIR and TIR-TIR cross-filament interaction surfaces each strongly inhibit NADase enzymatic activity in vitro ( Fig. 4a and Extended Data Fig. 7). The Sf STING substitutions V280D, E290K and R307E within the STING oligomerization interface disrupted all detectable NAD + hydrolysis. Likewise, Sf STING variants with substitutions in the STING-STING interface (N208D and A309R) and STING-TIR interface (R52E, K142D, N278E, Q279E and D285K) exhibit weak NADase activity only at 10-100× protein concentration, suggesting defects in the ability to oligomerize and catalyse NAD + cleavage ( Fig. 4a and Extended Data Fig. 7). Each Sf STING filament interface mutant retains the ability to form a stable, high-affinity complex with c-di-GMP, demonstrating that inhibition of NADase function is not due to impaired protein stability or ligand interaction (Extended Data Fig. 7 and Supplementary Fig. 1). Negative-stain electron microscopy analysis confirmed that the absence of NADase activity is a direct result of Sf STING interface mutants specifically losing the ability to form an active filament complex ( Fig. 4b and Extended Data Fig. 7). Replacement of the TIR BB loop within the principal TIR a -TIR b interaction site with a glycine linker sequence (ΔA36-K41) resulted in complete disruption of NADase function ( Fig. 4a and Extended Data Fig. 7). However, this Sf STING TIR mutant retains the ability to oligomerize into a filament in the presence of c-di-GMP, demonstrating that STING-STING interactions are the main driver of filamentation and that secondary TIR-TIR cross-filament interactions are required only for induction

Article
of NADase catalysis. We also observe that a D110A mutant retains the ability to recognize c-di-GMP and form long protein filaments but loses all ability to initiate NADase activity, providing further evidence for the essential role of filament formation in the activation mechanism of Sf STING (Extended Data Fig. 7). We expressed Sf STING in Escherichia coli, a bacterium that constitutively produces the activating ligand c-di-GMP, and confirmed that each Sf STING filament interaction interface is essential for STING-induced growth arrest in vivo ( Fig. 4c and Extended Data Fig. 7).
Our results provide a complete structural model of bacterial STING filament formation and effector domain activation in CBASS antiphage defence (Fig. 4d). STING-mediated antiphage defence begins when the associated CBASS protein CdnE, a cGAS/DncV-like nucleotidyltransferase that recognizes a yet unknown phage cue, senses bacteriophage infection and initiates synthesis of the antiviral nucleotide second messenger c-di-GMP (refs. 4,22,23 ). c-di-GMP is a high-affinity ligand that binds STING in a central chamber formed at the receptor homodimeric interface. Cognate CDN signal recognition induces a conformational change in the STING β-strand lid domain that envelopes c-di-GMP in a closed receptor complex. Next, the conformational change induced on complete lid closure exposes surface contact sites that create an interface for nucleating STING filament formation. STING filament extension is driven primarily by STING-STING contacts and cross-filament contacts between STING and the associated TIR effector domain. Finally, filament assembly leads to TIR-TIR interactions that rearrange the NADase active site to stimulate NAD + degradation and an abortive infection response that prevents phage propagation. In further support of our model of TIR NADase activation in CBASS antiphage defence, another study has determined the high-resolution structure of a distinct CBASS effector named TIR-SAVED that demonstrates that cyclic oligonucleotide binding induces a curved protein filament responsible for TIR NADase activation 24 . In animal cells, protein oligomerization has emerged as a general principle controlling rapid induction of innate immune signalling 25 . Our structural analysis of bacterial STING activation defines the molecular basis of STING filament formation and demonstrates remarkable conservation of oligomerization as a unifying mechanism controlling both prokaryotic and metazoan antiviral immune defence.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-04999-1. Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

Cryo-EM sample preparation and data collection
On exposure to the activating ligand c-di-GMP, solutions of purified Sf STING immediately begin filament formation and become visibly cloudy. For the first c-di-GMP dataset, Sf STING at 1 mg ml −1 was rapidly mixed with a 3× molar concentration of c-di-GMP (84 μM), immediately applied to glow-discharged 1.2/1.3 Cu 300 mesh grids (Quantifoil), and frozen in liquid ethane within 10 s of mixing using a Vitrobot Mark IV (Thermo Fisher) set at 4 °C and 100% humidity with no wait time, 3 s blot time and +8 blot force. For the second c-di-GMP dataset, Sf STING at 1 mg ml −1 was pre-incubated with 1 mM benzamide adenine dinucleotide before rapid mixing with 84 μM c-di-GMP and frozen as above. Semi-automated data collection was performed with SerialEM v3.8.5 and v3.8.6. Grids were imaged on a Titan Krios (Thermo Fisher) operating at 300 kV equipped with a BioQuantum K3 imaging filter with a 20-eV slit width and a K3 summit direct electron detector (Gatan) in counting mode at a nominal magnification of 105,000× corresponding to a calibrated pixel size of 0.825 Å. For the first dataset, a total exposure time of 1.6 s, corresponding to a total dose of 55.5 electrons Å −2 , was fractionated over 49 frames. For the second dataset, a total exposure time of 1.29 s, corresponding to 51.7 electrons Å −2 , was fractionated over 51 frames. The defocus targets were −1.2 to −2.1 μm for the first dataset and −1.2 to −2.5 μm for the second dataset.
For the 3′,3′-cGAMP dataset, Sf STING at 1 mg ml −1 was rapidly mixed with a 3× molar concentration of 3′,3′-cGAMP (84 μM) and frozen as described above. The 3′,3′-cGAMP dataset was collected on a Talos Arctica (Thermo Fisher) operating at 200 kV equipped with a K3 direct electron detector (Gatan) in counting mode at a nominal magnification of 36,000× corresponding to a calibrated pixel size of 1.1 Å. A total exposure time of 4.494 s, corresponding to a total dose of 52.9 electrons Å −2 , was fractionated into 50 frames. The defocus targets were −1.4 to −2.6 μm.

Cryo-EM image processing and model building
Data processing was performed in cryoSPARC v3.1.018 (ref. 27 ) and RELION-3.1 (ref. 28 ). For the c-di-GMP datasets, patch-based motion correction and CTF estimation was performed in cryoSPARC. Micrographs with severe contamination or poor contrast transfer function (CTF) fits were removed. Automated particle picking was performed in cryoSPARC with the template picker, using templates generated from either the filament tracer (first dataset) or the blob-based picker (second dataset). The particles were extracted with a box size of 320 and downsampled to a box size of 160 for initial two-dimensional (2D) classification and refinement steps.
For the c-di-GMP-bound Sf STING single-filament reconstruction, particle coordinates from the filament tracer and template-based picking in the first c-di-GMP dataset were combined, and duplicate coordinates closer than 40 Å were removed. A total of 277,287 coordinates corresponding to single-filament classes after heterogeneous refinement were imported into RELION. The second dataset had more bundled filaments and did not contribute to the single-filament reconstruction. Global and local (12 × 8 patches) motion correction was repeated in RELION using MotionCor2 v1.4.0 (ref. 29 ), followed by CTF estimation with GCTF v1.06 (ref. 30 ). After 2D classification and 3D refinement, 270,695 particles were subjected to signal subtraction using a mask around the central filament, followed by 3D classification without alignment. A total of 206,965 particles were reverted and subjected to two rounds of CTF refinement and a round of Bayesian polishing. One 3D classification without alignments was performed with the polished particles using a mask around the central filament. A class containing 26,447 particles that best resolved both the TIR and STING domains was selected for a final round of 3D refinement. In our analysis, Sf STING activation is observed as individual filaments that range in size with some filaments reaching >300 nm in length (about 85 dimer copies, about 6.3 MDa). Particles selected for processing and high-resolution structural analysis include density for at least 5 Sf STING dimer copies.
For the c-di-GMP-bound Sf STING double-filament reconstruction, multiple rounds of heterogeneous refinement were performed independently on each dataset in cryoSPARC to isolate particles contributing to the best reconstructions of a double filament after 3D non-uniform refinement. The final reconstructions contained 176,549 particles in the first dataset and 178,579 particles in the second dataset. As no further density corresponding to the benzamide adenine dinucleotide analogue or other differences were observed in the maps, the double-filament particle coordinates from the two datasets were combined and subjected to local motion correction, CTF refinement and non-uniform refinement. For all datasets, attempts to apply symmetry or helical parameters resulted in inferior reconstructions because the Sf STING dimers are not exactly symmetrically related in the oligomeric complexes.
For the 3′,3′-cGAMP dataset, patch-based motion correction and CTF estimation was performed in cryoSPARC. Micrographs with severe contamination or poor CTF fits were removed. Automated particle picking was performed in cryoSPARC with the template picker using templates generated from the blob-based picker. The particles were extracted with a box size of 280 and subjected to 2D classification followed by ab initio reconstruction and 3D non-uniform refinement. The resulting map and corresponding 261,685 particle coordinates were exported to RELION. Global and local (12 × 8 patches) motion correction and CTF estimation was repeated in RELION using MotionCor2 and GCTF respectively. After a round of 3D classification, 105,567 particles in classes with clear density for all four strands were subjected to CTF refinement, Bayesian polishing and 3D refinement without and with a mask around the two most defined strands.
The FsSTING (PDB 6WT5) CBD was used as a starting model docked into the single-fibre c-di-GMP-bound Sf STING density in Coot followed by iterative manual model building 31 . The c-di-GMP-bound Sf STING dimer was used as the starting model in the c-di-GMP-bound double filament and 3′,3′-cGAMP-bound oligomer. In the c-di-GMP double filament, individual secondary structure elements of the TIR domains of the central dimer that interacts with the STING domain of the other filament were placed by rigid fitting and manually adjusted in Coot. N-terminal portions of the TIR domain where side chains were not visible were converted to polyalanine. The TIR domains of all other Sf STING dimers in the c-di-GMP double filament and 3′,3′-cGAMP oligomer were removed. The 3′,3′-cGAMP-bound Sf STING dimers probably contain a combination of the 3′,3′-cGAMP orientation modelled and an approximately 180° rotation. Multiple rounds of Phenix real-space refine 32 was applied with manual correction in Coot in between. Model validation was performed in Phenix using MolProbity (ref. 33 ). Figure panels were generated using ChimeraX (ref. 34 ) and PYMOL (v2.5.1). Software for data processing and modelling was configured in part by SBGrid (ref. 35 ).

Analysis of TIR NAD + cleavage activity with fluorescent nicotinamide 1,N 6 -ethenoadenine dinucleotide
Plate reader reactions to assess NADase function were prepared as described previously 4 . Reactions were built in 50 μl final volume with reaction buffer (20 mM HEPES-KOH pH 7.5, 100 mM KCl), 500 μM nicotinamide 1,N 6 -ethenoadenine dinucleotide; (ε-NAD, Sigma), 0.1-10 μM enzyme and 20 μM c-di-GMP. Reactions were prepared as master mixes in PCR-tube strips and initiated by adding nicotinamide 1,N 6 -ethenoadenine dinucleotide immediately before placing into the plate reader. Fluorescence emission at 410 nm was read continuously over 40 min using a Synergy H1 Hybrid Multi-Mode Reader (BioTek) after excitation at 300 nm. Plots were generated with GraphPad Prism 9.3.0.

Electrophoretic mobility shift assay
Sf STING interactions with radiolabelled c-di-GMP were monitored by electrophoretic mobility shift assay as previously described 4 . In brief, 10-μl reactions contained 1× buffer (5 mM Mg(OAc) 2 , 50 mM Tris-HCl pH 7.5, 50 mM KCl) with a final protein concentration of 20 μM and about 1 μM α 32 -P-labelled c-di-GMP generated by overnight reaction of purified Vibrio cholerae DncV with GTP (about 0.1 μCi). Reactions were incubated for 5 min at 25 °C and separated on a 6% nondenaturing polyacrylamide gel held at 100 V for 45 min in 0.5× TBE buffer. Gels were fixed (40% ethanol and 10% glacial acetic acid) before drying at 80 °C for 1 h. Dried gels were then exposed to a phosphor storage screen and imaged on a Typhoon Trio Variable Mode Imager (GE Healthcare).

Negative-stain EM sample preparation, data collection and image analysis
Wild-type or mutant Sf STING (1 μM) was incubated with 10 μM c-di-GMP in buffer (20 mM HEPES-KOH pH 7.5, 250 mM KCl, 1 mM TCEP) for 15 min on ice. The mixture was then directly applied to a glow-discharged (30 s, 30 mA) 400-mesh Cu grid (Electron Microscopy Sciences, EMS-400Cu) coated with an approximately 10-nm layer of continuous carbon (Safematic CCU-010) for 30 s. After side blotting, the grid was immediately stained with 1.5% uranyl formate and then blotted again from the side. Staining was repeated twice with a 30-s incubation with uranyl formate in the final staining step. EM images were collected on a FEI Tecnai T12 microscope operating at 120 keV and equipped with a Gatan 4K × 4K CCD camera at a nominal magnification of 52,000× corresponding to a pixel size of 2.13 Å and at a defocus of about 1 μm.

STING toxicity analysis in E. coli
Sf STING and mutant constructs as well as an sfGFP negative-control construct were cloned into pET vectors for IPTG-inducible expression. E. coli BL21 (DE3) (NEB) were transformed with these plasmids and then plated on LB medium plates supplemented with 100 μg ml −1 ampicillin. After overnight incubation, three colonies from these plates were used to inoculate 5-ml MDG liquid cultures (0.5% glucose, 25 mM Na 2 HPO 4 , 25 mM KH 2 PO 4 , 50 mM NH 4 Cl, 5 mM Na 2 SO 4 , 2 mM MgSO 4 , 0.25% aspartic acid and trace metals) supplemented with 100 μg ml −1 ampicillin and grown overnight at 37 °C with 230 r.p.m. shaking. Cultures were diluted 1:50 into fresh M9ZB medium (supplemented with 100 μg ml −1 ampicillin) and grown for 3 h at 37 °C with 230 r.p.m. shaking. Cultures were then diluted to a uniform OD 600nm in M9ZB medium and further diluted 1:5 into fresh M9ZB medium supplemented with 5 μM IPTG to induce protein expression. A 200 μl volume of induced culture was added to a 96-well plate in technical triplicate and OD 600nm was recorded over 300 min in a Synergy H1 Hybrid Multi-Mode Plate Reader (BioTek) shaking at 37 °C. Plots were generated with GraphPad Prism 9.3.0.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Data availability
Coordinates and density maps have been deposited with the PDB and the Protein Data Bank in Europe under the following accession codes: Sf STING single fibres with c-di-GMP-7UN8 and EMD-26616; Sf STING double fibres with c-di-GMP-7UN9 and EMD-26617; Sf STING short fibres with 3′,3′-cGAMP (masked)-7UNA and EMD-26618; and Sf STING short fibres with 3′,3′-cGAMP-EMD-26619. Data that support the findings of this study are available within the article and its Extended Data and Supplementary Information. Source data are provided with this paper.   Fig. 3 | Cryo-EM data processing for SfSTING with 3′,3′-cGAMP. a, Section of a representative electron micrograph (n = 1,148 from 1 dataset) of SfSTING oligomers formed in the presence of 3′,3′-cGAMP. b, Data processing scheme used to generate c. c, Reconstructions of SfSTING complexes bound to 3′,3′-cGAMP, coloured by local resolution, masked and unmasked. d, Model of full-length SfSTING bound to 3′,3′-cGAMP which appears to have additional TIR to TIR suprafilament contacts (linking one filament to another). Insufficient resolved density for the TIR domains in this processed dataset limits further analysis of this interaction. The final uploaded model does not contain the amino acids in the TIR domains for this reason. The STING to STING cross-filament 'tetramer' interface is similar to the c-di-GMP single filament model but filaments do not extend beyond an average of 4-6 units.

Article
Extended Data Fig. 7 | Mutational analysis of the SfSTING cross-filament contacts. a, Raw plate reader NADase assay measuring fluorescence increase over time as a proxy for NAD + degradation. At 1 μM enzyme, WT* and E95Q* mutant SfSTING consume the ε-NAD substrate within the deadtime of the experiment (flat maximal signal). Error bars indicate standard deviation for average of three technical replicates. ΔA36-K41, Q279E, V280D, E290K, R307E, and D110A* (grey) show no activity at this enzyme concentration, and lines are obscured by other inactive construct curves. D110A* indicates that this is not a filament contact mutant but an intradimeric contact mutant. b, Bar-graph representation of NADase activity for mutant D110A similar to presentation in Fig. 4a. NADase activity is measured as fluorescence intensity at 5 min. Each bar within a set corresponds to 0.1, 1, and 10 μM enzyme. Baseline threshold indicates background fluorescent signal. Error bars indicate standard deviation for average of three technical replicates. Data representative of three independent biological replicates. One-way ANOVA comparison of mutant to WT at the same protein concentration with statistically significant mean values (P < 0.0001) marked by * or otherwise are greater than P = 0.05 and considered not significantly different. c, EMSA showing radiolabelled c-di-GMP binding to WT, D110A, and V280D mutant SfSTING. Note: lack of probable filament band for the V280D mutant. Image representative of n = 2 experiments. d, EMSA showing radiolabelled c-di-GMP binding to all mutants. Note: lack of probable filament band for majority of mutants, consistent with lack of observed filaments by electron microscopy. e, Extended negative-stain EM micrographs for all mutants in the presence of c-di-GMP. Scale bar represents 100 nm. Images representative of n ≥ 3 experiments. f, Extended plate reader growth curve assay data for most tested mutants. WT* and GFP* curves are identical to Fig. 4c. All curves represent the mean of 3 technical replicates. Data representative of >2 independent biological replicates. g, Plate reader growth curve assay for additional mutants V280D and D110A. All curves represent the mean of 3 technical replicates. Data representative of >2 independent biological replicates.