Cyclic nucleotide-induced helical structure activates a TIR immune effector

Cyclic nucleotide signalling is a key component of antiviral defence in all domains of life. Viral detection activates a nucleotide cyclase to generate a second messenger, resulting in activation of effector proteins. This is exemplified by the metazoan cGAS–STING innate immunity pathway1, which originated in bacteria2. These defence systems require a sensor domain to bind the cyclic nucleotide and are often coupled with an effector domain that, when activated, causes cell death by destroying essential biomolecules3. One example is the Toll/interleukin-1 receptor (TIR) domain, which degrades the essential cofactor NAD+ when activated in response to infection in plants and bacteria2,4,5 or during programmed nerve cell death6. Here we show that a bacterial antiviral defence system generates a cyclic tri-adenylate that binds to a TIR–SAVED effector, acting as the ‘glue’ to allow assembly of an extended superhelical solenoid structure. Adjacent TIR subunits interact to organize and complete a composite active site, allowing NAD+ degradation. Activation requires extended filament formation, both in vitro and in vivo. Our study highlights an example of large-scale molecular assembly controlled by cyclic nucleotides and reveals key details of the mechanism of TIR enzyme activation. A bacterial antiviral defence system generates a cyclic tri-adenylate that binds to a TIR–SAVED effector, inducing formation of a superhelical structure with adjacent TIR domains organizing into an active site, allowing NAD+ degradation.

Cyclic nucleotide signalling is a key component of antiviral defence in all domains of life. Viral detection activates a nucleotide cyclase to generate a second messenger, resulting in activation of effector proteins. This is exemplified by the metazoan cGAS-STING innate immunity pathway 1 , which originated in bacteria 2 . These defence systems require a sensor domain to bind the cyclic nucleotide and are often coupled with an effector domain that, when activated, causes cell death by destroying essential biomolecules 3 . One example is the Toll/interleukin-1 receptor (TIR) domain, which degrades the essential cofactor NAD + when activated in response to infection in plants and bacteria 2,4,5 or during programmed nerve cell death 6 . Here we show that a bacterial antiviral defence system generates a cyclic tri-adenylate that binds to a TIR-SAVED effector, acting as the 'glue' to allow assembly of an extended superhelical solenoid structure. Adjacent TIR subunits interact to organize and complete a composite active site, allowing NAD + degradation. Activation requires extended filament formation, both in vitro and in vivo. Our study highlights an example of large-scale molecular assembly controlled by cyclic nucleotides and reveals key details of the mechanism of TIR enzyme activation.
Cyclic nucleotide second messengers play a central role in prokaryotic antiviral defence by type III clustered regularly interspaced short palindromic repeats (CRISPR) 7,8 , cyclic nucleotide-based antiphage signalling systems (CBASS) 3 and the pyrimidine cyclase system for antiphage resistance (PYCSAR) 9 . These systems activate potent effector proteins that destroy key cellular components such as nucleic acids, cofactors or membranes to disrupt viral replication 3,10 . One example is the TIR domain, which functions as an enzyme that degrades NAD + to cause cell death in plants infected with pathogens 5,11 , antiphage immunity in the bacterial Thoeris 12,13 and TIR-STING (ref. 2 ) systems, and programmed nerve cell death in metazoa 6 . TIR domain activation requires effector subunit assembly, but the molecular mechanisms are still not fully understood.

cA 3 activates TIR-SAVED to degrade NAD +
Type III CRISPR and CBASS systems both use the SMODS-associated and fused to various effector domains (SAVED) cyclic nucleotide-binding sensor domain 14,15 . Here we focused on a type II-C CBASS from the Gram-positive bacterium Microbacterium ketosireducens 16 that has a nucleotide cyclase (CD-NTase), and TIR-SAVED and NucC effectors, along with a ubiquitin-like modification system of unknown function (Fig. 1a). We designed synthetic genes for the expression of the CD-NTase and TIR-SAVED proteins in Escherichia coli and purified the recombinant proteins using cleavable amino-terminal His-tags and gel filtration (Supplementary Fig. 2). We investigated the activity of the cyclase by incubating the protein with a range of nucleotides, including [α-32 P]ATP for visualization, and analysis by thin-layer chromatography. A radioactive product running at the position of a cyclic tri-adenylate (cA 3 ) standard was observed when ATP was present in the reaction (Fig. 1b), and the addition of other nucleotides did not result in any observable change (Extended Data Fig. 1a). The analysis of the reaction products by liquid chromatography confirmed that the cyclase uses ATP to produce cA 3 that co-elutes with a synthetic 3′,3′,3′-cA 3 standard (Fig. 1c).
To investigate the activity of the TIR-SAVED effector, we used etheno-NAD + (ɛNAD + ), an NAD analogue that emits a fluorescent signal on cleavage by TIR proteins 2,11 (Fig. 1d). We screened a range of commercially available cyclic nucleotide molecules for the ability to activate the TIR-SAVED effector, observing that only cA 3 resulted in generation of a fluorescent signal (Extended Data Fig. 1b). We proceeded to couple the cyclic nucleotide production by the cyclase with the NADase assay to follow the activation of TIR-SAVED. In the presence of ATP, the cyclase activated the TIR-SAVED effector to degrade NAD + (Fig. 1e). Together these data demonstrate that the CD-NTase synthesizes a 3′,3′,3′-cA 3 product that can activate the NADase activity of TIR-SAVED.
The initial rate of NAD + cleavage increased linearly with cA 3 concentration up to a value of 0.5 µM cA 3 , which corresponded with the concentration of TIR-SAVED in the assay, consistent with a 1:1 molar ratio of cA 3 to TIR-SAVED in the activated form of the effector (Extended Data Fig. 1c). We next examined the Michaelis-Menten parameters of the NADase activity of TIR-SAVED, determining a Michaelis constant, K m , of 470 µM and a catalytic efficiency, k cat /K m , in which k cat is the catalytic rate constant, of 2.08 × 10 3 M −1 s −1 (Extended Data Fig. 1d). This K m value falls within the concentration range of NAD + found in mammalian cells and E. coli (200-640 µM) 17,18 . Various bacterial TIR proteins have estimated K m values between 196 and 488 µM (ref. 4 ). In the Thoeris system of Bacillus cereus, the NADase enzyme ThsA activated by cyclic ADP ribose has a K m of 270 µM for NAD and a k cat /K m of 2.1 × 10 3 M −1 s −1 (ref. 13 ), in good agreement with our observations.

Activation of TIR-SAVED kills cells
CBASS defence systems tend to be phage and host species specific, and the mechanism of activation in response to phage infection remains largely unknown 19 . As we could not analyse CBASS activity in the cognate host, we took advantage of the observation that TIR-SAVED is activated by cA 3 by coupling the effector with a type III CRISPR system from Mycobacterium tuberculosis, which generates a range of cyclic oligoadenylate species including cA 3 on detection of target RNA 20 . Here we replaced the cognate Csm6 effector with M. ketosireducens TIR-SAVED and programmed the CRISPR system with a guide RNA complementary to the tetracycline resistance gene tetR ( Fig. 1f and Extended Data Fig. 2). When the active CRISPR system was present along with wild-type (WT) TIR-SAVED, transformation of a target plasmid containing the tetR gene resulted in no cell growth on plates containing tetracycline (Fig. 1g). This phenotype required the production of cA 3 (Extended Data Fig. 2) and the NADase activity of TIR-SAVED, as the variant E84Q, which targets the active site (Fig. 1g), relieved the effect 2,11,12 . This suggests that TIR-SAVED, activated by cA 3 , is responsible for cell death by NAD + hydrolysis.

Structure of the TIR-SAVED complex
The structure of the TIR-SAVED monomer was predicted using Alpha-fold2 (ref. 21 ) as implemented by the Colabfold server 22 (Supplementary  Fig. 3). Size-exclusion chromatography combined with small-angle X-ray scattering indicated that TIR-SAVED tends to be monomeric in solution (Extended Data Fig. 3a,b; Supplementary Fig. 4). However, addition of cA 3 to the protein resulted in a marked increase in global particle size as shown by dynamic light scattering (Fig. 2a) and a marked shift in the elution volume on size-exclusion chromatography (Extended Data Fig. 3c), suggesting the presence of high molecular weight complexes. These data strongly suggest that cA 3 binding induces multimerization of the TIR-SAVED protein, reminiscent of that observed for the SAVED-domain-containing Cap4 protein, in which head-to-tail multimers of two or three subunits were observed by electron microscopy 23 .
Analysis by cryogenic electron microscopy (cryo-EM) demonstrated that cA 3 drove the formation of ordered TIR-SAVED filaments (Fig. 2b). The assembly is characterized by a right-handed superhelical solenoid with 22 nm diameter and a 14 nm pitch. Seventeen TIR-SAVED monomers are present in each turn of the filament (Fig. 2b). Local resolution analysis performed with the CryoSPARC package highlighted a slightly anisotropic resolution, going from 2.5 to 5 Å (Extended Data Fig. 4). Three-dimensional (3D) variability analysis of a structure including two turns of the solenoid highlighted a degree of flexibility over the filament (Supplementary Figs. 5 and 6 and Supplementary Videos 1-6). The variability analysis shows a slight variation in the radius of the filaments, as well as the pitch of the structure; therefore, we chose to proceed with the cryo-EM reconstruction of one turn of the complex in isolation, and with the fitting of the four higher-resolution protomers, which provides information on protein/protein and protein/ligand interface.
To build a model of TIR-SAVED, we used four copies of the Alphafold2 output as a starting model for rigid body fitting using Chimera and the rigid-body-fitted tetramer was then refined as described in the Methods ( Fig. 2c and Extended Data Figs. 4f and 5). Four densities that correspond to the characteristic shape and size of the cA 3 ligand 23 were fitted and refined (Fig. 2d). The final atomic model confirmed the head-to-tail assembly in which the cA 3 was located in the binding pocket of the SAVED domain of subunit 1, interacting with conserved residues including W394 and K199. The binding site is completed by interactions with Article subunit 2, including with the conserved residue R388 (Fig. 2e), resulting in 'sandwiching' of cA 3 between the two SAVED domains (Supplementary Video 7). The SAVED domain is a distant cousin of the cyclic oligoadenylate-sensing CARF domain associated with type III CRISPR systems 15 , and is found in 30% of CBASS operons 23 . SAVED domains fused to nucleases bind cyclic dinucleotides and trinucleotides, activating the associated nuclease domain for DNA degradation 23,24 . Our data suggest that head-to-tail stacking of SAVED domains, potentiated by cyclic nucleotide binding, is a defining feature of effector activation.

Generation of a composite active site
TIR domains are ubiquitous, performing both protein scaffolding and enzymatic roles in different contexts across all three domains of life (reviewed in ref. 4 ). As is the case for TIR-SAVED, the catalytically active TIR domains, which degrade NAD + to cause cell death, typically rely on multimerization linked to activation 2,4,5,11,25,26 , but the molecular mechanism for this activation is not well understood. In the filament assembly, TIR domains exhibit a conserved interaction interface involving the BB loop (Extended Data Fig. 6) of subunit 1, which is held in a configuration that exposes the active site, due to interaction with the DE loop from the adjacent subunit (Fig. 2f). This BB-loop interface is also observed in TIR protofilament formation such as for the catalytically inactive TIR domain of the human MAL Toll-like receptor protein 27 and the active form of human SARM1 (ref. 28 ). The BB loop is suspected, in other NAD + -consuming TIR proteins, to regulate the access to the active site 11,29,25,26 , and for the RUN1 TIR domain, it has recently been proposed that the DE loop of the adjacent subunit contributes to NAD binding 29 . We therefore tested the hypothesis that a composite NADase active site is formed at the interface between two adjacent TIR domains in the TIR-SAVED filament (Fig. 2f). We first investigated the BB loop, showing that a variant protein with the double alteration D45A/L46A completely lacked NADase activity (Fig. 2g) without affecting cA 3 -dependent multimerization (Extended Data Fig. 7a). We proceeded to explore the role of the putative NAD + -binding site by replacing the highly conserved residue Y115 in the DE loop, which is suitably positioned to interact with NAD + in the TIR-SAVED filament (Fig. 2f). The Y115A variant had only 10% of the NADase activity of the WT enzyme, together with a threefold increase in K m for εNAD + , suggesting a direct role in the catalytic cycle rather than just substrate binding ( Fig. 2g and Extended Data Fig. 7b,c). This supports the model of a composite NADase active site that has also been proposed for the TIR-NLR RPP1 immune receptor 26 and SARM1 (ref. 28 ), and is likely to be broadly relevant for catalytic TIR proteins.

Filamentation is essential for activation
The composite active site of TIR-SAVED requires adjacent TIR domains to be brought together during the activation process. Nevertheless, it does not necessarily follow that the extended helical conformation of the effector observed here is relevant for function, as it could be an in vitro artefact of the assembly. To investigate this, three different site-directed variants were generated and assayed (Fig. 3a)  includes a substitution in the TIR active site, can multimerize but lacks NADase activity (Extended Data Fig. 8). Variant R (R388E), in which the charge of a conserved residue contributed by subunit 2 in the cA 3 interface is reversed (Fig. 2e), retained the ability to bind cA 3 but was unable to multimerize and was catalytically inactive ( Fig. 3b and Extended Data Fig. 8).
Finally, we generated variant KW (K199E/W394A), targeting two conserved residues in the cA 3 -binding site of subunit 1 (Fig. 2e). This variant no longer bound cA 3 and remained monomeric ( Fig. 3b and Extended Data Fig. 8b). Although each of these variants was inactive in isolation, they could be combined to generate further variations of the TIR-SAVED complex in a defined manner. First, we investigated the combination of the R and KW variants. Each has one WT and one compromised cA 3 -binding surface, and they can be combined along with cA 3 to generate a TIR-SAVED dimer with one composite active site (where the R and KW variants function as subunit 1 and 2, respectively, in Fig. 2e). The dimeric composition was confirmed by analytical gel filtration and native polyacrylamide gel electrophoresis (Extended Data Fig. 9a,b). Unexpectedly however, the dimer of TIR-SAVED showed no NADase activity on addition of cA 3 (Extended Data Fig. 9c), despite the provision of both halves of the shared active site. We therefore combined the R variant with the E variant, a combination that can result in the generation of filaments of varied size, but with only one active site as the R variant can be extended only by successive addition of inactive E variant subunits (Fig. 3a). By assaying this combination, we observed a recovery of NADase activity that increased progressively as the E variant was added. The initial activity continued to increase up to the highest ratio (1:32) of R/E studied, representing two turns of the helix on average-indicating that longer filaments of TIR-SAVED have higher levels of NADase activity even if there is only one active site in each filament ( Fig. 3c and Extended Data  Fig. 9d). These observations highlight the requirement that catalytic TIR domains assemble into multimers, not just dimers, for activation.
Although these observations confirmed the importance of filament assembly for TIR-SAVED activation in vitro, we wished to explore whether the same held true in vivo. To explore this, we created a second copy of the tir-saved gene to allow expression of combinations of variant proteins in our plasmid challenge assay. For the second copy of the gene, we used alternative codons to avoid any problems caused by recombination between two highly similar sequences in a plasmid. Using this experimental design, we tested whether the results obtained in vitro could be recapitulated in vivo (Fig. 3d). The R variant alone afforded a modest degree (1 log) of protection in vivo, far lower than observed for the WT protein. A combination of the R and KW variant genes produced a null phenotype, confirming the lack of activity observed in vitro. However, when the R variant was combined with the inactive E variant, a substantial level of effector activity was observed in the plasmid challenge assay, confirming the requirement for filament formation observed in vitro (Fig. 3c,d). Thus, we conclude that extended filament formation is relevant and essential for the function of the TIR-SAVED effector.

Wider aspects of TIR domain activation
Here we have explored a CBASS system with a TIR-SAVED effector that oligomerizes on activation by cA 3 . Oligomerization with 'open symmetry' (the ability to polymerize to infinity) seems to be a widespread property of prokaryotic innate immune effectors, encompassing the bacterial STING, SAVED and PYCSAR proteins 2,9,23 . We have demonstrated that this oligomerization is essential for TIR-SAVED function: first, as it probably allows access to the active site by formation of the BE interface, prising open the BB loop; second, as residues from the adjacent subunit participate in substrate binding and the catalytic cycle. Thus, the bacterial TIR-domain effectors seem to conform to the emerging paradigm of signalling by cooperative assembly formation proposed for the eukaryotic signalling complexes 30 . Notably, the bacterial TIR-SAVED filament exhibits the evolutionarily conserved head-to-tail arrangement in the TIR domain at the BE interface also found in eukaryotic TIR enzymes such as SARM1 (ref. 28 ) and RPP1 (ref. 29 Fig. 4), which is likely to be a prerequisite for TIR activation in all systems. However, our data demonstrate that assembly of this interface alone is not sufficient to result in an active state, as dimeric TIR-SAVED is inactive both in vitro and in vivo. Activation requires assembly of larger complexes, which may be essential to stabilize the TIR domain in an active conformation. In RPP1 and SARM1, TIR domains form tetrameric and octameric complexes with the AE interface formed by interactions between the αA and αE helices of TIR domains, perpendicular to the BE interface (Fig. 4). Here the TIR-SAVED assembly is unique as there is only a single filament of TIR domains, and it is notable that the SAVED domain occupies the space where the AE interface forms in the eukaryotic proteins (Fig. 4). Oligomerization-dependent activation of bacterial TIR effectors is likely to be the ancestral mechanism underlying the whole family of catalytic and non-catalytic TIR signalling complexes, exemplified by the accompanying study of the bacterial TIR-STING effector 31 . Our study reveals that activation is tightly controlled, requiring more than two TIR domains to be brought together. This suggests that the stabilization of the active form of TIR proteins requires extensive protein interactions, and possibly an alteration in dynamics, beyond the formation of the crucial BE interface-a property that may have evolved to avoid 'accidental' activation of TIR domains and autotoxicity.

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/s41586-022-05070-9.  29 (c), which are shown in the same orientation. The two-stranded TIR structure formed in the eukaryotic complexes involves the AE interface mediated by the αA and αE helices. In the TIR-SAVED filament, these two helices are oriented towards the SAVED domain.
Springer Nature or its licensor holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Cloning and mutagenesis
The synthetic genes encoding M. ketosireducens CD-NTase and TIR-SAVED were codon-optimized for expression in E. coli and purchased from Integrated DNA Technologies. Genes were cloned into the pEhisV5Tev vector 33 between the NcoI and BamHI restriction sites. The constructs were transformed into competent DH5α (E. coli) cells, and plasmids were extracted using GeneJET plasmid miniprep kit (Thermo Scientific) to verify sequence integrity by sequencing (Eurofins Genomics). Then plasmids were transformed into E. coli C43 (DE3) cells for protein expression. The TIR-SAVED variants E84Q, R388E, K199E/W394A, D45A/L46A and Y115A were generated on both constructs containing the tir-saved gene: the expression vector (pEhisV5spacerTev); and the vector used for the plasmid immunity assay (pRAT-Duet). Synthetic gene sequences and primers used for mutagenesis are listed in Supplementary Table 1.

Cyclic nucleotide analysis by thin-layer chromatography
Synthesis of cyclic nucleotides by the CD-NTase was analysed by using α-32 P-labelled ATP mixed with 'cold' NTPs. The reactions were carried out at 37 °C for 2 h in cyclase buffer (50 mM CAPS, pH 9.4, 50 mM KCl, 10 mM MgCl 2 , 1 mM MnCl 2 , 1 mM dithiothreitol), as a high pH activates CD-NTases in vitro 34 . A 20 µM concentration of CD-NTase was incubated with 50 µM ATP and 30 nM [α-32 P]ATP in a final volume of 20 µl. Reactions were stopped by addition of phenol/chloroform and products were isolated by chloroform extraction. Then 1 µl of the final volume was spotted on a silica gel thin-layer chromatography plate (Supelco Sigma-Aldrich). The plate was placed into a pre-warmed humidified chamber with running buffer composed of 30% H 2 O, 70% ethanol and 0.2 M ammonium bicarbonate, pH 9.2. Separated products were visualized by phosphor imaging. A control reaction run with a characterized type III CRISPR system, VmeCMR 32 was performed with 1 µM purified VmeCMR, 2 µM target RNA which had been incubated for 2 h at 37 °C in the reaction buffer (10 mM MgCl 2 , 10 mM Tris-HCl, pH 8, 50 mM NaCl).

Cyclic nucleotide analysis by liquid chromatography
To analyse the nature of the cyclic nucleotide produced by the M. ketosireducens CD-NTase, the reaction was carried out with 50 µM protein in cyclase buffer with 250 µM ATP for 2 h at 37 °C. The reaction was diluted twofold with water and ultracentrifuged using spin filters with a molecular weight cutoff of 3 kDa (Pall). Liquid chromatography analysis was performed on the Dionex UltiMate 3000 system. Sample separation was carried out on a Kinetiex EVO C18 2.6 µM (2.1 × 50 mm column, Phenomenex) with a 0-8% gradient of acetronitrile with 100 mM ammonium bicarbonate as the solvent. The flow rate was set at 300 µl min −1 and the column compartment temperature was set at 40 °C. Data were collected at a wavelength of 250 nm, and a 20 µM cA 3 commercial standard was used for comparison (Biolog).

NADase assay
The NADase activity of M. ketosireducens TIR-SAVED was analysed by using ɛNAD + as the substrate-when cleaved the ɛADP ribose product could be detected by fluorescence. Reactions were prepared in a 30 µl final volume with reaction buffer (50 mM Tris-HCl, pH 7.5, 50 mM KCl, 2.5 mM MgCl 2 ), 1 µM cyclic trinucleotide (cA 3 ; Biolog), 0.5 µM TIR-SAVED and 0.5 mM ɛNAD + (Sigma), or as stated in the figure. A master mix containing protein and activator was prepared on ice, and the substrate was added immediately before beginning analysis. Reaction samples were loaded into 96-well plates (Greiner 96 half-area) and fluorescence was measured continuously (cycle of 20 s) over 90 min using the FluoStar Omega (BMG Labtech) with an excitation filter at 300 nm and an emission filter at 410 nm. Reactions were carried out at 28 °C. A calibration curve was evaluated with the value obtained after 90 min of reaction with 10, 25, 75, 225, 500 and 675 µM ɛNAD + as the initial concentration.

Cyclase-NADase combined assays
For the cyclase-NADase combined reaction, 5 µM CD-NTase was incubated at 37 °C with 250 µM ATP in a 25 µl final volume containing 50 mM CAPS, pH 9.4, 50 mM KCl, 10 mM MgCl 2 , 1 mM MnCl 2 , 1 mM dithiothreitol. After 2 h, reactions were transferred into 96-well plates (Greiner 96 half-area) and supplemented with 0.5 µM ɛNAD + . Fluorescence measurements were performed as mentioned above, during 10 min before adding 0.5 µM TIR-SAVED. Reactions were carried out at 37 °C for a further hour. Fluorescence data were plotted over time and analysed with GraphPad Prism.

Plasmid immunity assays
To analyse the effect of TIR-SAVED NADase in vivo, we used the previously characterized 20 M. tuberculosis type III CRISPR system as an inducible producer of cA 3 when activated by the target RNA. Here we used the following plasmids to encode M. tuberculosis Csm1-5, Cas6 and a CRISPR array: pCsm1-5_ΔCsm6 (M. tuberculosis Csm1-5 under the control of the T7 and lac promoters); and pCRISPR_TetR (CRISPR array with spacers targeting the tetracycline-resistance gene targeting and M. tuberculosis Cas6 under the control of the T7 promoter). Competent E. coli C43 (DE3) cells were co-transformed with two constructs: pCsm1-5_ ΔCsm6 and pCRISPR_TetR. Plasmids were maintained by selection with 100 µg m l−1 ampicillin and 50 µg ml −1 spectinomycin. The tir-saved gene was inserted into pRAT-Duet between NcoI and SalI (multiple cloning site 1) or KpnI and BglII (multiple cloning site 2). This plasmid contains the tetracycline resistant gene targeted by M. tuberculosis Csm. The plasmid immunity assay is based on the transformation of the target plasmid containing the gene encoding TIR-SAVED into recipient cells encoding M. tuberculosis Csm. Recipient cells were prepared and transformed with the target plasmid as described previously 20 .
After the growth period in LB, cells were collected and resuspended in a LB volume adjusted to the same optical density at 600 nm (about 0.1). A total of 3 µl of a 10-fold serial dilution was applied in duplicate to selective LB agar plates supplemented with 100 µg ml −1 ampicillin, 50 µg ml −1 spectinomycin, 25 µg ml −1 tetracycline, 0.2% (w/v) d-lactose and 0.2% (w/v) l-arabinose. Plates were incubated overnight at 37 °C. This experiment was performed with two independent biological replicates using two technical replicates for each experiment. The variant Cas10(D630A) from M. tuberculosis Csm, which abolishes cyclase activity, was used as a control for no production of cyclic tri-adenylate. For colony counting, the same procedure was followed, except that 100 µl of a 300 µl growth volume was spread onto the selective LB agar plates. Two dilution factors were assayed for each condition and the experiments were performed in biological triplicates. Following incubation at 37 °C for 17-18 h, the resulting colonies were manually counted and the number of colony-forming units was reported per millilitre of culture volume. Data were statistically analysed by Prism8 (GraphPad) using non-pairing Brown-Forsythe and Welch analysis of variance tests. For multiple comparisons, the Dunnett T3 test was used.

Analytical gel filtration
To analyse the oligomeric state of TIR-SAVED, 100 µl protein (at least 100 µM) was injected into a size-exclusion column (Superose 6 Increase 10/300 GL or Superose 12, GE Healthcare) equilibrated in 20 mM Tris-HCl, pH 8.0, 250 mM NaCl and 10% glycerol. In some experiments, 158 µM cyclic tri-adenylate was added to the TIR-SAVED sample before centrifugation at 10,000g for 10 min at 4 °C and loading onto the size-exclusion column. To analyse TIR-SAVED dimerization, 83 µM R388E variant was first incubated with 500 µM cA 3 and 127 µM K199E/ W394A variant and loaded into the Superose 12 size-exclusion column in similar conditions. The eluted fractions were then analysed by native polyacrylamide gel electrophoresis in a 4-16% Bis-Tris gel (Invitrogen). Using similar gel filtration running conditions, standard proteins (number 1511901, BioRad) were eluted to calculate a calibration curve of the column. A 100 µl volume of BSA (7.3 mg ml −1 ) was injected as an extra standard. The elution volume (V e ) for each protein was determined on the basis of the elution profile. Then the K average was calculated as the ratio K average = (V e − V 0 )/(V t − V 0 ), in which V 0 is the void volume (7.77 ml) and V t is the total volume (24 ml) of the column. The calibration curve corresponds to the plot of K average versus the molecular weight of each protein in log 10 . The trend line (logarithmic) was plotted with the following standards: y-globulin (158 kDa), BSA (66 kDa), ovalbumin (44 kDa) and myoglobin (17 kDa) to obtain the best R 2 value (0.997) corresponding to the range of the target protein.

Dynamic light scattering
Dynamic light scattering measurements were performed with the Zetasizer Nano S90 (Malvern) instrument. In the protein dilution buffer (20 mM Tris-HCl, pH 7.5, 250 mM NaCl, 10% glycerol), 21 µM TIR-SAVED was prepared with 32 µM cyclic nucleotide (cyclic tri-adenylate or others) when stated in the figure. After centrifugation at 12,000g for 10 min at 4 °C and filtration with 0.22 µm filters, 12 µl of sample was loaded into the quartz cuvette (ZMV1012). The measurements were carried out at 25 °C with 3 measures of 13 runs. The curves of WT TIR-SAVED are the mean of three technical replicates and two independent experiments.

Small-angle X-ray scattering
Small-angle X-ray scattering (SAXS) datasets were recorded at the European Synchrotron Radiation Facility (Grenoble, France) on the BioSAXS beamline BM29 (ref. 35 ) using a 2D Pilatus detector. Data were collected at room temperature (20 °C) using a standard set up (automated sample mounting to a capillary by a robot) 36 . A 100 µl volume of TIR-SAVED (10.9 mg ml −1 ) was injected into the Superose 12 column 10/300 GL (GE Healthcare) in 20 mM Tris-HCl, pH 8.0, 250 mM NaCl and 10% glycerol with a flow rate of 0.4 ml min −1 .
Sample scattering curves were obtained after subtraction of the averaged buffer signals using standard protocols with PRIMUS 37 . The values for the radius of gyration, R g , and the forward scattering intensity, I(0), were extracted using the Guinier approximation. The molecular weight range was estimated by Bayesian inference 38 implemented in the ATSAS suite 3.0.5 (ref. 39 ). The theoretical SAXS curve of the TIR-SAVED structure predicted by Alphafold2 was back-calculated and fitted with the experimental SAXS datasets with the program CRYSOL 40 . SAXS parameters for data collection and analysis are reported in Supplementary Table 2.
Electrophoretic mobility shift assays Radiolabelled cA3, [α-32 P]cA 3 , was prepared with the type III CRISPR complex, VmeCMR, as described in the section entitled Cyclic nucleotide analysis by thin-layer chromatography. The reaction product was incubated with a threefold dilution range of TIR-SAVED (0.22, 0.67 and 2.0 µM) for 15 min at 25 °C in a final 15 µl volume. The reaction buffer was the same as that for NADase activity: 50 mM Tris-HCl, pH 7.5, 50 mM KCl, 2.5 mM MgCl 2 and 25 mM NaCl from the protein dilution buffer. Ficoll was added to the samples to give a final concentration of 4%, and samples were then loaded into a native 6% acrylamide gel (acrylamide/ bis-acrylamide 29:1). TIR-SAVED/cA 3 complexes were separated by electrophoresis into 1× TBE buffer for 1 h at 200 V and visualized by phosphor imaging.

Thermal shift assays
A 2 µM concentration of TIR-SAVED WT or variants were incubated with a range of cA 3 concentrations (0, 0.4, 2 and 10 µM) in the following buffer: 20 mM Tris-HCl, pH 7.5, 250 mM NaCl, 10% glycerol supplemented with 5× SYPRO Orange Fluorescent Dye (BioRad). A temperature gradient was applied from 25 to 95 °C with 1 °C increments, and fluorescence was measured in a Stratagene MX3005. The curves are the mean of two independent experiments with technical triplicates.

Transmission electron microscopy
Samples for negative-stain electron microscopy analysis were prepared by diluting purified TIR-SAVED protein alone, or with an equimolar ratio of cyclic trinucleotide as indicated, to a concentration of 1 mg ml −1 in buffer (20 mM Tris-HCl pH 8.0, 250 mM NaCl, 10% glycerol).
Electron microscopy images of negatively stained TIR-SAVED were collected using a JEOL 1200 transmission electron microscope operating at 120 keV and equipped with a Gatan Orius CCD (charge-coupled device) camera at a nominal magnification of ×100,000, and a pixel size of 9.6 Å. A 4 µl volume of the diluted sample was applied onto a glow-discharged 400-mesh copper grid (Agar Scientific) coated with a layer of continuous carbon, followed by a 1-min absorption step and side blotting to remove bulk solution. The grid was immediately stained with 2% uranyl acetate at pH 7 and then blotted from the side and air-dried before imaging.
Cryo-EM grids were prepared using an FEI Vitrobot Mark IV (Thermo Fisher) at 4 °C and 95% humidity. A 3 µl volume of TIR-SAVED complex was applied to holey carbon grids (Quantifoil Cu R1.2/1.3, 300 mesh), glow-discharged for 45 s at a current of 45 mA in an EMITECH K100X glow discharger. The grids were then blotted with filter paper once to remove any excess sample, and plunge-frozen in liquid ethane. All cryo-EM data presented here were collected on a JEOL CRYO ARM 300 microscope, equipped with a DE-64 direct detector at the Scottish Centre for Macromolecular Imaging, Glasgow, UK. A total of 3,907 videos were collected in accurate hole centring mode using SerialEM 3.8 (ref. 41 ). The CryoSPARC 3.3.1 software 42 was used for motion correction, CTF estimation and manual exposure correction, as well as for the selection of the 2,319 videos used in the analysis. CryoSPARC 3.3.1 was also used for the whole single-particle reconstruction workflow, from manual particle picking to classification to generate templates for autopicking and subsequent 2D classification and 3D processing, including per-particle motion correction 43 , sharpening and 3D variability analysis 44 , obtaining a structure with an overall resolution of 3.8 Å. The final reconstruction was obtained from 596,378 particles selected from classes representing both circular and elongated particles at a sampling rate of 0.997 Å per pixel and had an overall resolution of 3.8 Å, as calculated by Fourier shell correlation at 0.143 cutoff during post-processing. The Alphafold2 model was fitted to the map using the Chimera software, taking into account the handedness. The inverted handedness was the only compatible solution and was therefore chosen for further modelling and refinement using the software packages Coot 45 , Refmac-Servalcat as implemented in the CCP-EM suite 46 and PHENIX 47 . The cryo-EM data collection, refinement and validation statistics are summarized in Extended Data Table 1.

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

Data availability
The electron microscopy data deposition D_1292120121 has been assigned the following accession codes: PDB ID 7QQK and EMD-14122. Source data are provided with this paper. All other data are provided in the Supplementary Information.