Diversity and abundance of ring nucleases in type III CRISPR-Cas loci

Abstract

Most type III CRISPR-Cas systems facilitate immune responses against invading mobile genetic elements such as phages by generating cyclic oligoadenylates (cOAs). Downstream effectors activated by cOAs are typically non-specific proteins that induce damage to essential cellular components, thereby preventing phage epidemics. Owing to these toxic effects, it is crucial that the production and concentration of cOAs remain under tight regulatory control during infection-free periods or when deactivating the immune response after clearing an infection. Type III CRISPR loci often encode enzymes known as ring nucleases (RNs) that bind and degrade specific cOAs, while some effectors are auto-deactivating. Despite the discovery of several classes of RNs, a comprehensive bioinformatic analysis of type III CRISPR-Cas loci in this context is lacking. Here, we examined 38 742 prokaryotic genomes to provide a global overview of type III CRISPR loci, focusing on the known and predicted RNs. The candidate RNs Csx16 and Csx20 are confirmed as active enzymes, joining Crn1–3. Distributions and patterns of co-occurrence of RNs and associated effectors are explored, allowing the conclusion that a sizeable majority of type III CRISPR systems regulate cOA levels by degrading the signalling molecules, which has implications for cell fate following viral infection.

This article is part of the discussion meeting issue ‘The ecology and evolution of bacterial immune systems’.

1. Introduction

CRISPR-Cas is an adaptive prokaryotic immune system that incorporates genomic fragments of invading mobile genetic elements (MGEs) into the host’s chromosomal CRISPR array [1,2]. These fragments, called spacers, are expressed during subsequent infections as CRISPR RNA (crRNA) and help interference complexes to target invading nucleic acids via sequence complementarity [3]. Type III CRISPR-Cas systems use multi-subunit interference complexes, hallmarked by Cas10—a large protein that can harbour two main functions [3]. While a minority of Cas10 proteins use an HD (histidine-aspartate) nuclease domain to cleave single stranded DNA non-specifically [4,5], over 90% of Cas10s have a cyclase domain that generates second messenger signalling molecules [6–9]. The signal molecules produced by Cas10, either cyclic oligoadenylates (cOA) or S-adenosyl methionine-AMP [10], accumulate in the cell and activate downstream effector proteins that are typically encoded by genes in the same operon. The effector proteins contain a sensory domain that captures the signal, leading to the allosteric activation of the effector domain that may be a nuclease, protease or a membrane-disrupting domain (see reviews [11,12]). Activation of effectors thus tends to have toxic consequences for cells and their actions can lead to growth arrest. This outcome is sometimes called abortive infection; an umbrella term under which diverse immune mechanisms are often grouped, all sharing the ecological rationale of sacrificing the host for the sake of the population (see review [13]). Whether phage defence mechanisms actually lead to cell death in vivo is debated [14], and the presence of ring nucleases (RNs) in CRISPR type III systems provides a mechanism to avoid this outcome.

RNs are small proteins that degrade cOA signal molecules, thus thwarting the activation of downstream effectors. They are generally encoded in CRISPR-Cas loci and are also used by viruses as anti-CRISPR (Acr) proteins [15]. It is not currently known whether their primary function in cells is to deactivate defence in an ongoing infection state or to curtail signal molecule levels in a non-infected state. To date, three families of RNs have been experimentally verified. The first RN to be discovered, CRISPR-associated ring nuclease 1 (Crn1), was identified biochemically from a lysate of Saccharolobus solfataricus [16]. Crn1 is a metal-independent phosphodiesterase that binds cA₄ in its dimeric CRISPR-associated Rossmann fold (CARF) domains and degrades it into linear A₂ > P (cyclic 2′,3′ phosphate) products. Crn1 is most commonly found in crenarchaeal type III CRISPR loci and has a relatively low cA₄ cleavage rate, which may be tuned to remove cA₄ from the cell without disrupting the immune response [16].

Crn2 and its virally encoded counterpart AcrIII-1 are also cA₄-cleaving RNs [15]. This class of RN is characterized by the DUF1874 domain, which forms a dimeric cA₄ recognition domain. AcrIII-1 degrades cA₄ much more quickly than Crn1, consistent with its role in subverting type III CRISPR signalling during viral infection. This family of RNs has also been found fused to the bacterial effector Csx1, constituting a self-limiting cA₄-activated ribonuclease [17].

Crn3, previously known as Csx3, constitutes the third family of cA₄-cleaving RNs. The structure of Crn3 is only distantly related to the CARF superfamily, harbouring closer resemblance to sulfate transporter and anti-sigma factor antagonist domains [18–20]. Crn3 binds cA₄ by sandwiching the molecule between two adjacent dimers that tetramerize in a head-to-tail orientation [18]. The phosphodiesterase reaction is manganese dependent and generates linear A₂-P products. Crn3 is sometimes fused to an AAA-ATPase domain of unknown function [20,21].

In addition to the dedicated RNs, several type III CRISPR effectors have been shown to degrade their cOA activators within the binding site of the sensory domain, effectively functioning as self-limiting enzymes. All cA₆-dependent Csm6 family ribonucleases studied to date display cA₆ RN activity in the CARF recognition domain [22–24]. Likewise cA₄-activated Csm6/Csx1 effectors are also capable of degrading their activator within the CARF domain [25–27]. Recently, the CalpL effector, which uses a SAVED (SMODS-associated and fused to various effector domains) domain to bind cA₄, has also been confirmed as an RN [28,29].

The CRISPR-associated Csx15, Csx16 and Csx20 proteins have been predicted to be RNs, based on sequence and genomic neighbourhood analyses [20], but this has not yet been experimentally confirmed. RNs are thus common in type III CRISPR systems but have not been studied systematically. Here, we undertook an extensive analysis of type III CRISPR loci, mapping the known and predicted RNs and exploring their association with the diverse range of effector proteins. The cA₄ RN activities of Csx15, Csx16 and Csx20 are investigated biochemically, and structural modelling is used to predict their mechanisms of cA₄ recognition.

2. Material and methods

(a). Data preparation

A total of 38 742 complete bacterial and archaeal genomes were downloaded from NCBI on 25 March 2024. For phage genomes, the May 2024 set of 28 114 curated genomes from the Millard database [30] was downloaded.

(b). Bioinformatic methods

A previously described, the Snakemake [31] pipeline was used as the basis for type III CRISPR locus characterization [6]. In short, the pipeline uses custom-built Cas10 Hidden Markov Model (HMM) profiles to find type III CRISPR loci, which are further analysed using CCTyper [32] and a panel of HMM profiles to discover effector proteins. The pipeline was modified to enable ring nuclease detection and phage genome analysis. Custom-built HMM-libraries were constructed using published sequence data in NCBI for ring nucleases Crn1, Crn2 and Crn3. For Csx15, Csx16 and Csx20, libraries were built based on blastp homology searches against the NCBI protein database using previously published sequences [20] as queries. The predicted RN Csx14 [20] appears to be a member of the Crn1 family and was included in the HMM profiles for Crn1. Each protein coding sequence within 6 kbp of a type III CRISPR locus was analysed for RNs using hmmscan from the Hmmer 3.3.2 package [33]. To prevent cross-annotation with effectors that have similar domains (e.g. CARF), there was a maximum length cut-off of 250 amino acids for RNs during annotation. Candidate RNs were screened by multiple sequence alignment and manual inspection to check for the presence of absolutely conserved residues. A small number of false positives, lacking key highly conserved residues, were identified and removed. Crn1 was noted previously as a member of the CARF7 and CARF_m13 families [20]. Attempts to assign the other RNs into CARF families using phylogenetic and clustering approaches was unsuccessful owing to high sequence/structural divergence. A table of all cellular and phage-encoded RNs including annotation details and sequences is provided in the electronic supplementary material, S1.

The steps for constructing the Cas10 phylogenetic tree were outlined in [6]. In short, each Cas10 was annotated for the presence of cyclase or nuclease domains. The Cas10 sequences were then aligned with Muscle [34]. A phylogenetic tree file was built using FastTree2 [35] with arguments -wag and -gamma, and visualized in RStudio 2024.4.0.735 [36] using ggtree [37]. The tree was annotated with CRISPR subtype data from CCTyper (in a few cases corrected manually) and with the presence of known and candidate ring nuclease families. The signal molecule associated with each locus was inferred from the type(s) of effector present and annotated for each locus in the tree.

A network interaction graph was made with Gephi (https://gephi.org/) using RN/effector co-occurrence data. Co-occurrences between effectors and between RNs were removed to highlight those between RNs and effectors.

To search for ring nucleases in phage genomes, the proteomes of all 28 114 phage genomes in the Millard phage database were analysed using our RN HMM libraries using Hmmer 3.3.2 [33] similar to the type III CRISPR-Cas analysis outlined above. Hits were analysed manually using multiple sequence alignment to detected conserved residues, and a few false-positive hits were removed manually. CARF families based on motifs were assigned as with the CRISPR RNs. The phage-encoded RNs with annotation details and phage data are provided in the electronic supplementary material, S1.

(c). Cloning, expression and purification of Csx15, 16 and 20

Synthetic genes (electronic supplementary material, table S1) encoding Csx15, Csx16 and Csx20, codon optimized for expression in Escherichia coli, were purchased from Integrated DNA Technologies (IDT), Coralville, USA, cloned into the pEHisV5TEV vector [38] between the NcoI and BamHI sites and transformed into DH5α cells. Construct integrity was confirmed by sequencing (Eurofins Genomics, DE).

The constructs were transformed into E. coli C43 (DE3) and proteins were expressed according to the standard protocols previously described [38]. In brief, 2 l of culture were induced with 0.4 mM isopropyl-β-D-1-thiogalactoside at an optical density of approximately 0.8 and grown for 4 h or overnight at 25°C. Cells were harvested (4000 rpm; Beckman Coulter JLA-8.1 rotor) and resuspended in lysis buffer and lysed by sonication. Proteins were purified with an immobilized metal affinity chromatography (IMAC) column (HisTrapFF, Cytiva, Marlborough, USA), washed with five column volumes of loading buffer and eluted with a linear gradient of loading buffer plus 0.5 M imidazole. Following his-tag removal by tobacco etch virus (TEV) protease, proteins were subjected to a second IMAC step and the unbound fraction collected. Size exclusion chromatography was used to further purify the proteins, which were eluted isocratically as described previously [38]. Pure proteins were concentrated, aliquoted and stored frozen at −70°C.

(d). Ring nuclease activity of Csx15, Csx16 and Csx20

Ring nuclease activity was assayed by incubating 1 µM of each protein with 100 µM synthetic cA₃, cA₄ or cA₆ (Biolog) in reaction buffer (20 mM Tris-HCl pH7.5, 250 mM NaCl and 5 mM ethylenediaminetetraacetic acid (EDTA) at 30°C for 60 min. The reaction was quenched by adding methanol and vortexing. The mixture was then dried, before resuspension in water for high pressure liquid chromatography (HPLC) analysis on an UltiMate3000 HPLC system (Thermo Fisher scientific) with a C18 column (Kinetex EVO 2.1 × 50 mm, particle size 2.6 μm). The column temperature was set at 40°C and absorbance was monitored at 260 nm. Samples were analysed by gradient elution with solvent A (20 mM ammonium acetate, pH 8.5) and solvent B (methanol) as a flow rate of 0.3 ml min⁻¹ as follows: 0−0.5 min, 1% B; 0.5−6 min, 1–15% B; 6−7 min, 100% B.

3. Results

(a). Structural models and ring nuclease activity of Csx15, Csx16 and Csx20

We showed previously that 92% of type III CRISPR loci probably function via nucleotide signalling, with active Cas10 polymerase domains [6]. Of these, the predominant signalling molecule is cA₄, so it is perhaps unsurprising that the three stand-alone RNs identified to date, Crn1−3, are all specific for cA₄ [15,16,18]. Csx15, 16 and 20 were previously identified as candidate RNs based on genome context, sequence conservation and structural predictions [20]. However, these predictions have not been confirmed biochemically, so we first sought to test this specific hypothesis.

The structures of Csx15, Csx16 and Csx20 were modelled as dimers using Alphafold3 (AF3) ([39]; figure 1A). Two AMP ligands were included to help predict cOA binding sites. Inclusion of these ligands improved the ordering of the mobile loops that probably become structured on cOA binding. All three models were predicted with high confidence (ptm/iptm/ranking scores 0.91/0.90/0.94, 0.79/0.75/0.77 and 0.95/0.94/0.95 for Csx15, 16 and 20, respectively). To gain more insight into the likely cOA binding sites of the proteins, we mapped conserved residues from a diverse range of homologues for each protein (electronic supplementary material, figure S1) onto the AF3 structures (electronic supplementary material, figure S2). This revealed a cluster of conserved residues on one face of Csx16 and Csx20 that correspond with the modelled AMP binding sites and probably pinpoint the binding site for cOA. However, for Csx15, conserved residues were more prevalent on the ‘bottom’ face of the dimer, away from the canonical cOA binding site. This situation is reminiscent of Crn3, which forms filaments of head-to-tail dimers sandwiching cA₄ [18]. Structural comparisons using the DALI server [40] yielded hits for Csx15 with SAVED and CARF family proteins (Z-score > 5 for PDB accessions 7RWM, 8Q3Z, 8FMF and 7QDA); Csx16 yielded a significant hit (Z-score 4.8) with PDB 2J6B—a viral homologue of Crn2 [41] while Csx20 yielded a strong match (Z-score 8.0) with the Crn1 family protein Sso2081 (PDB 7YGH) [42]. Pairwise comparison of the predicted structures of Csx16 and Csx20 yielded a DALI Z-score of 8.9, consistent with a common core fold consisting of a 5-stranded β-sheet flanked by α-helices. These structural features are consistent with a derived Rossmann fold (CARF-like) structure for Csx15, Csx16 and Csx20, an observation reinforced by the two-dimensional topology maps [43] of the three predicted proteins and comparison with Crn1 (electronic supplementary material, figure S3).

Figure 1.

Structural modelling and RN activity of Csx15, Csx16 and Csx20. (A) Dimeric protein AF3 models are shown with 2 AMP molecules (yellow sticks) modelled to mimic the cOA binding site, subunits are coloured differently for ease of interpretation. (B) RN activity of Csx15, Csx16 and Csx20 against cA₃, cA₄ and cA₆, monitored by HPLC. Csx16 and Csx20 degrade cA₄ into linear products. Standards and characterized reaction products are labelled. (>p represents a 2′,3′-cyclic phosphate).

Representative examples of Csx15, Csx16 and Csx20 were cloned, expressed in E. coli and purified to homogeneity as described in the methods (electronic supplementary material, figure S4A). The proteins were tested individually for the ability to degrade cA₃, cA₄ and cA₆ under conditions of 100-fold substrate excess for 1 h (figure 1B). Csx16 and Csx20 showed clear RN activity against cA₄, fully degrading it to small linear products. Neither enzyme degraded cA₆, while only very minor activity for Csx20 against cA₃ could be observed. Csx15, on the other hand, displayed no RN activity against any cOA species (figure 1B). To explore RN activity more fully, we repeated these experiments with a 10-fold higher concentration of protein (10 µM) (electronic supplementary material, figure S4B). Under these conditions, Csx20 fully degraded cA₃ as well as cA₄, while Csx16 retained specificity for cA₄. Very limited activity of Csx15 against cA₃ and cA₆, but not cA₄, could be observed. We conclude that Csx16 and Csx20 are cA₄-specific RNs, while the function of Csx15 remains uncertain. It is possible that we have not found the correct reaction conditions to reveal the activity of Csx15. Alternatively, it could conceivably act as a cOA ‘sponge’ rather than a RN, as phage-encoded sponge proteins are known to sequester cyclic nucleotides and inhibit cellular defences [44,45].

(b). Distribution and co-occurrence patterns of extrinsic ring nucleases

Having explored the RN activity of Csx15, 16 and 20, we proceeded to analyse the distribution of extrinsic RNs across the Cas10 tree (figure 2). This analysis immediately demonstrated that RNs are widespread in Cas10 loci, and when found in an effector-containing locus (96% of RN instances), they are associated only with cA₄-dependent effectors. In the 536 CRISPR-Cas loci that contained a cA₄-activated effector, 211 (39%) had an associated RN. In 31 cases, RNs were found in loci with no effector, suggesting either the presence of unknown effectors in the locus or a role for RNs in trans by another type III CRISPR locus with a known effector. The most frequently observed RN was Csx20 (78 loci), followed by Csx16 (68 loci), Crn1 (44 loci), Crn3 (37 loci), Csx15 (18 loci) and Crn2 (7 loci). There was no obvious bias in the distribution of RNs with respect to CRISPR subtypes that use cOA signalling, while loci from subtypes III-C and III-F, which lack an active Cas10 cyclase, have very few associated RNs. Csm6 and Csm6-2, which are activated by cA₆, are not found in association with known RNs, which may reflect intrinsic RN activity by these effectors.

Figure 2.

Phylogenetic tree of Cas10 with ring nuclease distribution. The signalling molecule used by each system is predicted based on effector content of the locus, as defined in [6], and coloured according to the key. Instances of candidate ring nucleases Crn1, Crn2, Crn3, Csx15, Csx16 and Csx20 are indicated in concentric circles. Cas10s lacking a clear active cyclase domain are indicated by a red dot in the subtype ring.

Only 10 of the 242 RN-positive loci (approx. 4%) contained more than one RN, which supports the hypothesis that they are all performing equivalent roles. An exception to this rule was Csx15, which co-occurs with Crn1 in seven loci and as the sole RN in 11 loci. We investigated the frequency of Csx15 across all approximately 40 000 RefSeq genomes and found that Csx15 is found in genomes without a type III CRISPR-Cas locus in 25% of cases. Along with the peculiar co-occurrence pattern with Crn1 and lack of RN activity in vitro, this suggests Csx15 may not be a canonical RN.

The co-occurrence of ring nucleases and effectors is shown in figure 3A. All characterized ring nucleases preferentially degrade cA₄, so it is unsurprising (but reassuring) that they co-occur with cA₄-specific effectors. The most abundant effector in our dataset, Csx1, defined here as the cA₄-activated family of CARF-HEPN dimeric effector proteins [6] is found in association with each of the ring nucleases analysed here and has a particularly strong association with Csx20. We also analysed loci with a single known effector to quantify the proportion that encoded a RN (figure 3B). The observation that only 34% of Csx1-containing loci encode a known RN may be at least partially explained by the fact that some Csx1 family members have intrinsic RN activity [25,27]. In support of this, the Can1−2 effector family [46–48], which lack intrinsic RN activity, are associated with RNs in 66% of loci whereas for Cami1, which possesses RN activity [49,50], the proportion is only 30%. A similar pattern of association was identified by Makarova & Koonin [20]. Following this logic, the observation that the membrane bound effector Cam1 is found alongside RNs in 81% of loci suggests that this effector also lacks intrinsic RN activity, although that has yet to be tested [51]. None of the characterized cA₄-specific RNs are found associated with effectors that use a different signalling molecule. While some of these effectors, such as Csm6, are effective RNs in their own right [22–24,27], others, such as NucC, are not [52]. This could suggest that RNs targeting molecules other than cA₄ simply await discovery, or alternatively that infection outcomes are different in these cases.

Figure 3.

Co-occurrences of ring nucleases and effectors in type III CRISPR loci. (A) Gephi plot showing co-occurrence patterns between effectors and ring nucleases. Ring nucleases are in yellow, cA₃, cA₄ and cA₆-activated effectors in purple, blue and green, respectively and SAM-AMP-activated effectors in orange. (B) The proportion of effector instances that are associated with a RN. The data only includes loci with one effector and zero or one RNs.

Phage genomes were also investigated for ring nucleases. Among the 28 114 phage genomes in the Millard database, the previously described anti-CRISPR protein AcrIII-1 (the phage-encoded homologue of Crn2) [15] was by far the most abundant RN, present in 36 genomes. Additionally, Csx20 was found in four genomes and Csx16 in two genomes. The high relative abundance of AcrIII-1/Crn2 in phage genomes compared to its presence in only seven prokaryotic genomes supports the view that it probably originated from viruses as an anti-CRISPR and was subsequently adopted by cellular hosts to regulate cOA levels.

4. Discussion

By confirming the RN activity of Csx16 and 20, we have expanded the catalogue of extrinsic RNs to five families. These are widely distributed across the type III CRISPR systems that signal via cA₄—the predominant signalling molecule used by 84% of all known effector instances [6]. When combined with the observation that intrinsic RN activity is frequently observed in CARF or SAVED sensor domains of effectors, we can conclude with some confidence that an ‘off-switch’ is an important component of most type III CRISPR systems. Exceptions, such as the prophage-encoded systems described in Vibrio species that use NucC or Csx23 effectors [52,53], exist but probably represent a small minority.

The prokaryotic immune system has two major components that potentiate an antiviral response by generation of cOA species: type III CRISPR-Cas and CBASS (cyclic oligonucleotide-based antiphage signalling system) [54]. Each system responds to infection by activating a specialized cyclase that generates a cyclic nucleotide second messenger which in turn activates one or more effector proteins to provide an immune response. CBASS and type III CRISPR-Cas have much in common, with shared signalling molecules such as cA₃ and effectors such as NucC [52,55] and TIR-SAVED [6,56]. Each uses the signal amplification possible with a second messenger to activate a toxic response to infection that slows down viral replication at the cost of cell fitness. Their main point of divergence appears to be the fact that type III CRISPR systems often encode an ‘off-switch’ while, by contrast, CBASS activation appears to be a ‘one way street’ that may often result in cell death or growth arrest [57].

This fundamental difference between type III CRISPR and CBASS defence, despite all their similarities, may reflect the fact that the former can be activated early in infection (by viral mRNA) whilst the latter is viewed as a ‘last ditch’ defence, typically activated late in the infection cycle (for example, by intracellular phage capsid proteins [44]) when other defences have failed [58,59]. CBASS defence, functioning at the level of herd immunity, is clearly worth having, as the system is fairly common in bacteria [60]. Nevertheless, cells appear to avoid altruistic suicide when possible. One significant caveat is that RNs may primarily function to prevent aberrant activation of type III CRISPR-Cas defence (i.e. activation of Cas10 by non-viral RNA), rather than to clear up after a bona fide infection. The growth-arrested state facilitated by type III CRISPR-Cas may also enable other defence systems to clear the infecting virus, which could then be followed by reversal of the arrested state by RNs. Such cooperation between defence systems has been found between the RNA-targeting type VI CRISPR-Cas systems and restriction modification, where restriction enzymes cleave phage genome while the cell is under Cas13-induced dormancy [61]. Looking ahead, there is a pressing need for studies that tackle these questions using an evolutionary, population-based approach.

Although we have analysed six candidate RN families that account for a large proportion of the cA₄-signalling CRISPR systems, it is possible and, indeed, likely that further RN families remain to be discovered. The candidate RN Unk_01 [20] appears unrelated to the CARF superfamily and was not examined here, but is worthy of further study. The function of Csx15 also remains an open question at this point. By looking closely at the loci which lack a known RN and which have an effector with intrinsic RN activity, we hope to unveil new RN families in the future.

Ethics

This work did not require ethical approval from a human subject or animal welfare committee.

Data accessibility

The Snakemake pipeline and accompanying scripts are available for review at [62]. This repository also contains the HMM profiles for Cas10s, effectors and all six ring nuclease families in this study. An interactive website for browsing the bioinformatic results is available at https://vihoikka.github.io/rn_browser.

Supplementary material is available online [63].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors’ contributions

V.H.: conceptualization, data curation, formal analysis, investigation, methodology, project administration, software, visualization, writing—original draft, writing—review and editing; H.C.: formal analysis, investigation, methodology, writing—original draft; S.Gru.: formal analysis, investigation, methodology, writing—original draft; S.Gra.: investigation, methodology, writing—original draft; M.W.: conceptualization, formal analysis, project administration, resources, supervision, writing—original draft, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This work was supported by a European Research Council Advanced Grant (Grant REF 101018608 to MFW). H.C. acknowledges the support of the China Scholarship Council (code 202008420207).