Tetramerisation of the CRISPR ring nuclease Csx3 facilitates cyclic oligoadenylate cleavage

Type III CRISPR systems detect foreign RNA and activate the cyclase domain of the Cas10 subunit, generating cyclic oligoadenylate (cOA) molecules that act as a second messenger to signal infection, activating nucleases that degrade the nucleic acid of both invader and host. This can lead to dormancy or cell death; to avoid this, cells need a way to remove cOA from the cell once a viral infection has been defeated. Enzymes specialised for this task are known as ring nucleases, but are limited in their distribution. Here, we demonstrate that the widespread CRISPR associated protein Csx3, previously described as an RNA deadenylase, is a ring nuclease that rapidly degrades cyclic tetra-adenylate (cA4). The enzyme has an unusual cooperative reaction mechanism involving an active site that spans the interface between two dimers, sandwiching the cA4 substrate. We propose the name Crn3 (CRISPR associated ring nuclease 3) for the Csx3 family.

domain proteins associated with type III CRISPR loci have been identified but not yet described 46 (Shmakov et al., 2018, Shah et al., 2018, and recent work has characterized the cA 4 -activated 47 DNA nickase Can1, which is present in Thermus thermophilus (McMahon et al., 2020).

49
Thus, type III CRISPR systems are capable of directing a multi-faceted antiviral defence on 50 detection of foreign RNA in the cell. However, activation of this anti-viral state is known to generate 51 collateral damage to host nucleic acids (Rostol and Marraffini, 2019) that could lead to dormancy 52 or cell death. While this could be an acceptable outcome for an infected cell, if a viral infection can 53 be cleared then the cOA-signalling pathway needs a mechanism to remove the cOA molecules 54 and return the cell to a basal state. Some archaea encode a dedicated ring nuclease 55 , recently named as the Crn1 family (CRISPR-associated ring nuclease 56 1) (Athukoralage et al., 2020b). Crn1 is a specialised CARF domain protein that splits the cA 4 ring 57 into two linear A 2 products to switch off the antiviral response . In other 58 systems, the CARF domains of Csx1/Csm6 family nucleases slowly degrade cOA, thus acting as 59 bi-functional, self-limiting nucleases ( an anti-CRISPR (Acr). This enzyme, known as AcrIII-1, binds cA 4 by means of a distinct protein 62 fold (DUF1874) unrelated to the CARF domain, and degrades cA 4 rapidly using conserved active 63 site residues (Athukoralage et al., 2020b). These viral ring nucleases can abrogate type III 64 CRISPR immunity by rapidly destroying the cA 4 infection signal. This enzyme has been co-opted 65 into some bacterial type III CRISPR systems, where it likely acts to degrade cA 4 following 66 clearance of phage infection. In this context, it has been named as CRISPR-associated ring 67 nuclease 2 (Crn2) (Athukoralage et al., 2020b, Samolygo et al., 2020.

69
Ring nucleases thus appear to be an important constituent of virus:host conflict in the expanding 70 arena of cyclic nucleotide signalling. Here, we focus on the Csx3 family of proteins, which is found 71 associated with many type III CRISPR systems (Shah et al., 2018, Shmakov et al., 2018. Csx3 72 from Archaeoglobus fulgidus has been described as an RNA deadenylation enzyme and 73 crystallised in complex with a pseudo-symmetric RNA tetraloop (Yan et al., 2015). Subsequent 74 structural analysis revealed that the Csx3 protein is a distantly related member of the CARF family 75 of proteins (Topuzlu and Lawrence, 2016). Spurred by these observations, we undertook a 76 detailed study of the Csx3 protein. We demonstrate that Csx3 is, in fact, a cA 4 -specific ring 77 nuclease, for which we propose the name Crn3 (CRISPR-associated ring nuclease 3). The 78 enzyme has an unusual cooperative reaction mechanism involving the association of two dimers, 79 bridged by the cA 4 substrate, to complete the active site. This mechanism may allow the cell to 80 respond appropriately to changing cA 4 and Csx3 levels during viral infection and preserve type III 81 CRISPR immunity. 82

83
Csx3 is a ring nuclease specific for binding and cleavage of cA 4 84 The Csx3 protein from A. fulgidus was originally crystallised in the absence and presence of a 85 pseudo-symmetric RNA tetranucleotide, and shown to have RNA deadenylase activity in vitro (Yan 86 et al., 2015). Given what is now known about cA 4 ring nucleases, we re-assessed the structure and 87 potential function of the Csx3 protein. It has previously been suggested that the binding site for the 88 RNA tetranucleotide could be compatible with binding of a symmetric cyclic oligonucleotide 89 (Topuzlu and Lawrence, 2016), similar to those observed in other CARF family proteins. 90 Examination of the genomic context of Csx3 in selected archaeal and bacterial species confirmed 91 its close association with type III-B CRISPR systems and with other CARF family proteins such as 92 the ribonuclease Csx1 ( Figure 1A), implying a role in cyclic oligoadenylate signalling and providing 93 further impetus to re-examine the function of Csx3. Csx3 is most commonly observed in the 94 euryarchaea and cyanobacteria (Figure 1-figure supplement 1). We therefore cloned, expressed 95 and purified A. fulgidus Csx3, allowing biochemical analysis. Initial studies showed that Csx3 binds 96 cyclic tetra-adenylate with an apparent dissociation constant < 0.1 µM. In contrast, an RNA 97 oligonucleotide (49-9A) with a 9A poly-adenylate 3' tail was bound 100-fold less tightly, with an 98 apparent dissociation constant around 10 µM ( Figure 1B)

114
The enzymatic specificity of Csx3 was tested against an RNA substrate oligonucleotide 49-9A. 115 Csx3 cleaved this substrate in the presence of Mn 2+ ions, removing nucleotides from the 3' end 116 (Figure 2), in keeping with previous observations (Yan et al., 2015). The rate constant for cleavage 117 was calculated as 0.0063 ± 0.0013 min -1 at 50 °C. By comparison, Csx3 cleaved cA 4 with a single-118 turnover rate constant of 3.5 ± 0.16 min -1 at 50 °C ( Figure 2C). The ~600-fold faster cleavage rate 119 for cA 4 compared to RNA strongly suggests that cA 4 is the physiological substrate, and that RNA 120 with a 3' polyA tail represents a substrate analogue of the cyclic nucleotide, as observed previously 121 for a Csx1/Csm6 family enzyme (Han et al., 2017b). Although the physiological growth 122 temperature of A. fulgidus is around 70 °C, we conducted these assays at 50 °C to enable rate 123 determination. 124 The products of the cA 4 cleavage reaction were identified by liquid-chromatography high-resolution 125 mass spectrometry (LC-HRMS) as predominantly linear di-adenylate (A 2 P), plus a small amount of 126 A 2 P with a cyclic phosphate (A 2 >P) ( Figure 2D). This suggests there are two active sites in the 127 dimeric Csx3 structure, as seen for the ring nuclease Crn1 ( consistent with a partial deactivation of the Csx1 ribonuclease due to degradation of cA 4 by the 160 Csx3 enzyme. The effect is not as striking as when using a bona fide Acr protein, which is 161 consistent with the prediction that Csx3 functions as part of the type III CRISPR defence in cells 162 that express itrather than an Acr. It should be noted that we expressed the Csx3 enzymes using 163 a strong inducible promoter in these assays and we do not know what the relevant Csx3 164 concentrations are in virally-infected cells. 165  the presence of a conserved aspartate residue, D69, which is positioned adjacent (~4 Å) to the 194 bound RNA in the crystal structure ( Figure 4A). The importance of D69 was not explored in 195 previous studies, so we mutated D69 to an alanine to test for a role in catalysis. The D69A variant 196 was completely catalytically inactive as a ring nuclease, confirming the importance of D69 in the 197 catalytic mechanism and suggesting that catalysis requires residues on both faces of the dimer 198 (henceforth denoted as the "D69 face" and "H60 face"). A possible explanation for this was that 199 Csx3 has a shared active site that is formed when two dimers come together, forming a tetramer. 200 A diagnostic test for this, as first proposed for the shared active site of aspartate 201 transcarbamoylase (Wente and Schachman, 1987), is to mix inactive single variants and look for 202 recovery of activity, as a fraction of intact active sites that can be formed in the quaternary 203 structure. Accordingly, we took the two inactive variants of Csx3, H60A and D69A, and tested their 204 ring nuclease activity when mixed together ( Figure 4B). Ring nuclease activity was recovered 205 when the two inactive variants were combined. This result was strongly supportive of the 206 hypothesis that the reaction mechanism involves two half-sites that combine to form a single active 207 site that bridges two dimers of the enzyme. The structure of Csx3 reveals a head-to-tail filament stabilised by cA 4 226 In parallel with the biochemical analysis, we crystallised the H60A variant of Csx3 and soaked the 227 cA 4 substrate into the crystals. The structure was solved using molecular replacement with data to 228 1.8 Å resolution (Supplementary Table 1). The electron density clearly showed a molecule of cA 4 229 bound between adjacent dimers of Csx3 ( Figure 5A). Notably, the Csx3 structure presented here 230 crystallised in a different space group to those structures published previously (Yan et al., 2015), 231 which has allowed the arrangement of Csx3 dimers into a pseudo filament arrangement ( around 1020 Å 2 . The interface area between the D69 face, which forms the majority of the 237 interactions with cA 4 , is around 940 Å 2 (470 Å 2 per monomer), and for the H60 face is around 340 238 Å 2 (170 Å 2 per monomer). Interestingly, although not annotated as a tetramer, there is a buried 239 surface area of around 700 Å 2 (350 Å 2 per monomer) between two adjacent dimers. 240 There are surprisingly few residues that interact with cA 4 in the active site of Csx3 given its size.

241
The orientation of each monomer in the dimer means the interactions with cA 4 are symmetrical. 242 The complex with an RNA fragment had two adenine, one uracil and one guanine bases in the same 252 position as the four adenine bases described here (Yan et al., 2015), there is obviously some 253 plasticity around these recognition sites. 254 We anticipated that the comparison of the structure of Csx3 with cA 4 with the apo structure 255 published by Yan et al. (Yan et al., 2015), in conjunction with activity assays on Csx3 variants, 256 might provide clues as to the key residues in catalysis. Interestingly, R71 moves around 3.3 Å in 257 order to form bidentate hydrogen bonds with an oxygen atom of a phosphate group, which we 258 predict is adjacent to the phosphodiester bond that is cleaved ( Figure 5C and text below). This 259 movement of R71 brings it within hydrogen bonding distance of D69, which has been shown to be 260 vital for activity. The position of H80 also differs between the two structures; the residue moves 261 around 3.3 Å in order to hydrogen bond with the cA 4 . However, there is the caveat that the packing 262 arrangement differs in the apo Csx3 structure, meaning H80 does not interact with a ligand or the 263 face of an adjacent dimer and thus has nothing to 'anchor' it in place. 264 There are a number of histidine residues in or near the active site which could be involved in 265 coordinating one or more Mn 2+ ions. H60 and H57 are the most likely candidates; they are both 266 absolutely conserved, and the position of the alanine residue in the H60A variant structure 267 suggests this residue does not move significantly on tetramer formation. The position of H57 is 268 identical in the apo and cA 4 -bound structures. H60 and H57 are at the symmetry plane for the 269 monomers constituting a dimer, meaning there are four histidine residues in close proximity. It is 270 therefore feasible that one or both H60 and/or H57 residues coordinate one or more metal ions, 271 and there is the possibility that just one metal ion bridges the two monomers.   that are predicted to work by catalysing in-line nucleophilic attack by a 2'-hydroxyl group of the cA 4 298 substrate on an adjacent phosphodiester bond. In this regard, the best candidate for cleavage is 299 the P-O bond of the phosphate (labelled *), as the angle formed between the 2'-OH, P and O 300 moieties is 167° ( Figure 5C), identical to that seen for the ring nuclease AcrIII-1 (Athukoralage et  301 al., 2020b). The absolutely conserved residue R71 interacts with this phosphate group via a 302 bidentate hydrogen bond and may participate in stabilisation of the transition state or oxyanion 303 leaving group. The conserved catalytic residue D69 in turn makes polar contacts with R71. These 304 two residues may serve to position each other correctly (and/or perturb their respective pK a 's) to 305 enhance catalysis. The conserved H80 residue forms a hydrogen bond with the cA 4 . As it is one of 306 only two residues from the H60 face that interacts with cA 4 , H80 may play a 'pinning' role to 307 ensure engagement of the H60 face in catalysis. Csx3 variants R71A and H80A both displayed 308 highly reduced ring nuclease activity ( Figure 5figure supplement 4), consistent with important 309 roles in cA 4 binding and/or catalysis. 310 The observations that Mn 2+ is required for catalysis, and that the primary product of cA 4 cleavage is 311 A 2 P rather than A 2 >P ( suggestive (but not diagnostic) of a mechanism whereby a metal-activated hydroxyl ion initiates 313 nucleophilic attack on the phosphodiester bond (Yang, 2011). Uncertainty arises from the 314 observation that the metal independent ring nuclease AcrIII-1 also largely generates A 2 P, 315 presumably due to the rapid hydrolysis of the cyclic phosphate following phosphodiester bond 316 cleavage (Athukoralage et al., 2020b). The conserved histidine residues H57 and H60 are in 317 appropriate positions to contribute to catalysis by coordination of the essential catalytic Mn 2+ ion(s).

318
We see no evidence for the ion in our crystal structure, which may be due to the deletion of the 319 H60 side chain. However, we note that the five Mn 2+ ions modelled in the Csx3 structure by Yan et 320 al. have ambiguous electron density, hampered by the lower resolution data (Yan et al., 2015). In 321 addition, the coordination distances for a Mn 2+ ion with a nitrogen atom (of the histidine residues) 322 are longer than expected (Zheng et al., 2017). It remains possible that H80 participates in binding 323 a second metal ion in addition to its observed interaction with cA 4 . 324 The crystal structure of the Csx3:cA 4 complex neatly explains the observation of two distant active 325 site regions, which come together in the complex. During the catalytic cycle, cA 4 binding and 326 dimer:dimer sandwiching would lead to rapid cA 4 degradation that presumably in turn results in 327 dissociation of the tetrameric active form. In support of this model, we observed that the inactive 328 H60A variant of Csx3 had a milky, colloidal property on addition of cA 4 , which we interpreted as 329 being due to the formation of long Csx3 fibres bridged by multiple cA 4 molecules ( Figure 5D). The 330 wild-type protein did not show this behaviour, presumably because cA 4 was rapidly cleaved. We 331 confirmed the propensity of Csx3 to oligomerise upon cA 4 addition by dynamic light scattering 332 (DLS) measurements in buffer devoid of metal ions. Under these conditions, the addition of cA 4 333 resulted in mixed oligomeric species with significantly increased particle size and molecular weight, 334 whereas the molecular weight of particles formed in the absence of cA 4 were consistent with 335 homodimers of Csx3 (Supplementary Table 2). 336 The discovery that two dimers of Csx3 must associate to sandwich the cA 4 substrate to effect Here we re-examined the specificity of the Csx3 nuclease, which is found in association with type 352 III CRISPR systems in bacteria and archaea. Csx3 was originally described as a manganese 353 dependent RNA deadenylase (Yan et al., 2015). The enzyme was co-crystallised with an RNA 354 tetranucleotide, revealing a pseudo-symmetrical binding mode, and site directed mutagenesis 355 revealed a crucial role for residues including H60 in catalysisa residue that was situated on the 356 opposite face of the protein from the RNA binding site (Yan et al., 2015). This prompted the 357 authors to make the reasonable speculation that RNA is bound on one side and wraps around the 358 protein to be cleaved on the opposite side. revealed that Csx3 binds much more tightly to cA 4 than to RNA, is considerably more active as a 365 cA 4 -specific ring nuclease than an RNA deadenylase, and has properties consistent with a function 366 as a ring nuclease in vivo. These data indicate that Csx3 functions as a cA 4 -specific ring nuclease 367 in type III CRISPR systems. The default "Csx" nomenclature was designed for proteins of unknown 368 function lacking a clear association with a specific CRISPR subtype (Haft et al., 2005). We 369 therefore propose the family name Crn3 (CRISPR-associated ring nuclease 3) for the Csx3 protein 370 family. 371 The Crn1 family of ring nucleases, based on the CARF domain fold, was originally identified in the 372 Sulfolobales and related crenarchaea , where their function is thought to 373 be to remove cA 4 from the cell once a viral infection has been cleared. The anti-CRISPR ring 374 nuclease AcrIII-1 appears in many archaeal virus and some bacteriophage genomes 375 (Athukoralage et al., 2020b). Homologues of AcrIII-1 are also found associated with some bacterial 376 type III CRISPR systems, and in this context have been named as CRISPR associated ring 377

Figure 6. Sigmoidal response of ring nuclease activity as a function of Csx3 concentration.
Plot visualising initial reaction rates of cA 4  picture is becoming clearer for the specific ring nucleases as they are identified in different 382 archaeal and bacterial genomes (Figure 7). It appears to be an emerging paradigm that cells with a 383 type III CRISPR defence require a mechanism to remove the cOA signal, either once viral infection 384 is cleared, or when the system "fires" inappropriately due to self-targeting (Athukoralage et al., 385 2020a). 386 387 Crn3 has a highly unusual cooperative catalytic mechanism 388 A. fulgidus Crn3/Csx3 is a much faster enzyme than Crn1. The single turnover reaction rate for cA 4 389 cleavage of 3.6 min -1 is closer to that of the AcrIII-1 family, which utilise a distinct active site 390 architecture, employing a conserved histidine residue as a general acid to stabilise the oxyanion 391 leaving group (Athukoralage et al., 2020b). This raises an important question: why does Crn3/Csx3 392 degrade cA 4 so quickly when it plays a role in cellular defence rather than viral offense? Clearly, 393 removing a crucial signal of viral infection that mobilises cellular defences is not something that 394 should be undertaken precipitously. A key observation is that Crn3/Csx3 functions via a highly 395 unusual cooperative mechanism where two enzyme dimers associate to sandwich a cA 4 substrate 396 molecule. The active site is thus composed of two half-sites that are present on opposite faces of 397 each dimeric moiety, with enzyme tetramers formed transiently to complete the catalytic cycle. where interdomain interactions with ATP can influence quaternary structure (Zhao et al., 2016). 403 This results in cooperative kinetics where low concentrations of Crn3/Csx3 provide very low levels 404 of ring nuclease activity that rapidly increases with increasing enzyme concentrations. 405 Furthermore, the rate of cA 4 turnover is limited by the formation of protein:cA 4 complexes or 406 dissociation of the products, rather than the catalytic step, which is significantly faster. Together, 407 these factors may provide a means to control ring nuclease activity in an appropriate manner thus 408 ensuring that cA 4 activation of ancillary ribonucleases is allowed to proceed and provide immunity. 409 Thus, alterations in the gene expression levels of Csx3 would have a large effect on the overall 410 rate of catalysis. 411 Finally, the observed specificity of Crn3/Csx3 for cA 4 implicates the corresponding type III CRISPR 412 systems as functioning via a cA 4 second messenger. This conforms to the paradigm established 413 from studies of the Crn1 and Crn2/AcrIII-1 family as well as analysis of the specificity of the 414 Csx1/Csm6 ribonucleases, which are most often activated by cA 4 (Grüschow et al., 2019).

415
Although this conclusion may be influenced by a sampling bias, it appears that cA 4 is the default 416 cyclic nucleotide employed to signal infection and activate defences in type III CRISPR systems. It 417 is possible that the pressure applied by cA 4 -specific anti-CRISPRs has resulted in the utilization of 418 cA 3 and cA 6 second messengers in some bacterial defence systems. Given the pace of discovery 419 in this area, new cellular and viral enzymes implicated in cyclic nucleotide signalling are 420 anticipated. 421

Protein crystallisation 512
Crystallisation conditions were tested with JSCG and PACT 96 well commercial screens (Jena 513 Biosciences) with Csx3 H60A at a concentration of 13.5 mg/ml. Following optimisation, crystals 514 were obtained from 25% (v/v) Jeffamine M-600 and 100mM HEPES pH 7.5 using hanging drops in 515 a 24 well plate. 3 μl drops in a 2:1 or 1:1 protein:mother liquor ratio were added to a silanized cover 516 slip over 400 μl mother liquor and sealed with high-vacuum grease (DOW Corning, USA) and left 517 to grow at room temperature. Prior to addition of the cA 4 ligand, crystals were harvested into a 518 fresh 2 μl drop of mother liquor and 1 μl of 16mM cA 4 solution was added and left to soak for 12 519 hours. Crystals were harvested and cryoprotected with the addition of 2 μl 50% (v/v) Jeffamine M-520 600 in 0.5 μl increments, mounted on loops and vitrified in liquid nitrogen. Cowtan, 2004) were used for automated and manual refinement respectively, which included 530 addition of the ligand and water molecules. cA 4 was drawn using Chemdraw (Perkin Elmer) and 531 restraints generated in JLigand (Lebedev et al., 2012 Dynamic light scattering experiments were carried out in a 20 µl quartz cuvette using a Zetasizer 538 Nano S90 instrument (Malvern). 80 µM AfCsx3 was either measured alone or when mixed with an 539 equimolar concentration of cA 4 in phosphate buffered saline, pH 7.5 at 25 °C. Three technical 540 replicates were carried out and by default triplicate measurements were made to produce an 541 average for each technical replicate. For visual inspection of the effect of adding cA 4 to AfCsx3 542 and variants, 80 µM AfCsx3 was added to equimolar cA 4 in PBS supplemented with 2 mM MnCl 2 in 543 a 100 µl reaction volume and either heated to 25 °C or 70 °C for 10 min before photographing 544 using a 12-megapixel ƒ/1.8-aperture camera. 545 Liquid chromatography high-resolution mass spectrometry 546 Samples were generated by incubating Csx3 (10 μM dimer) with 100 μM synthetic cA 4 (BIOLOG 547 Life Science Institute, Bremen) in buffer containing 20 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM 548 DTT and 2 mM MnCl 2 for 10 min at 50 °C. Reactions were quenched by adding EDTA to a final 549 concentration of 25 mM, and then deproteinised and desalted using a C18 cartridge (Harvard 550 Apparatus). Liquid chromatography-high resolution mass spectrometry (LC-HRMS) analysis was 551 performed on a Thermo Scientific™ Velos Pro instrument equipped with HESI source and Dionex 552 UltiMate 3000 chromatography system as previously described . Data were 553 analysed using Xcalibur™ (Thermo Scientific). 554

Plasmid immunity assay comparing viral and host ring nucleases 555
These assays were performed largely as described previously (Athukoralage et al., 2020b). Cells 556 containing Mycobacterium tuberculosis (Mtb)Csm1-5, Cas6 and a CRISPR array targeting the 557 tetracycline resistance gene of pRAT-Duet were transformed with the target plasmid containing 558 genes encoding the cA 4 -activated ancillary nuclease T. sulfidiphilus (Tsu) Csx1 and a ring 559 nuclease (M. mazei Csx3 or the anti-CRISPR ring nuclease AcrIII-1 from phage THSA-485). 560 Transformants were allowed to recover on selective LB plates in the presence of 0.2 % lactose and 561 0.02 % arabinose (the former for induction of the Cas genes and the ring nuclease, the latter 562 induces the Csx1 plus ring nuclease). The experiment was run with increasing arabinose 563 concentrations (0, 0.002, 0.02, 0.2 % w/v); there was no difference between 0 and 0.002 %, and a 564 slight difference between 0.02 and 0.