Bacteriophages suppress CRISPR–Cas immunity using RNA-based anti-CRISPRs

Bacterial strains and growth conditions

The bacterial strains used in this study are listed in Supplementary Table 2. Unless otherwise noted, the P. atrosepticum, P. aeruginosa (PA14, PAscm and PAO1) and E. coli strains were routinely grown at 25 °C, 30 °C and 37 °C, respectively, in lysogeny broth (LB) shaken at 180 rpm or on LB–agar (LBA) plates containing 1.5% (w/v) agar. When applicable, antibiotics and supplements were added at the following concentrations: ampicillin, 100 µg ml⁻¹; chloramphenicol (Cm), 25 µg  ml⁻¹; kanamycin, 50 µg  ml⁻¹; gentamicin, 50 µg  ml⁻¹ for P. aeruginosa or 15 µg  ml⁻¹ for E. coli; tetracycline (Tc), 5 µg  ml⁻¹; 5-aminolevulinic acid (ALA), 50 µg  ml⁻¹; isopropyl β-D-1-thiogalactopyranoside (IPTG), 100 µM for P. atrosepticum or 1 mM for PAO1; l-arabinose (Ara), 0.3% (w/v). Bacterial growth was measured as the optical density at 600 nm (OD₆₀₀) using a Jenway 6300 Spectrophotometer.

Phage purification and titration

The phages used in this study are listed in Supplementary Table 2. In brief, 2 ml of overnight host culture was inoculated into 50 ml LB in a 250 ml flask and incubated for 30 min. Then 100 µl of phage lysate was added to the culture and incubated overnight. A centrifugation step was done (3,220g for 20 min at 4 °C) to separate the virions from the cell debris. The supernatant was placed in a sterile universal container for storage and a few drops of NaCO₃-saturated chloroform were added before thoroughly vortexing the mixture to lyse any remaining cells. Finally, the phage titre was determined by pipetting 20-μl drops of serial dilutions of the phage stock in phage buffer (10 mM Tris-HCl, pH 7.4, 10 mM MgSO₄, 0.01% (w/v) gelatin) onto an LBA overlay (0.35% w/v) seeded with 100 μl host overnight culture. Plaques were counted after incubation overnight, with the phage titre represented as PFU per ml. Pseudomonas phages DMS3m and JBD30 were propagated on PA14 ΔCRISPR, wild-type PAO1 or PAsmc Δcas3. Pectobacterium phage ΦTE was propagated on wild-type P. atrosepticum. Pseudomonas phages were stored at 4 °C in SM buffer (50 mM Tris-HCl, pH 7.5, 100 mM NaCl, 8 mM Mg₂SO₄) over chloroform. Pectobacterium phage ΦTE was stored at 4 °C in phage buffer over chloroform.

DNA isolation and manipulation

The oligonucleotides used in this study are listed in Supplementary Table 3. The polymerases, restriction enzymes, Gibson Assembly mix, USER enzyme and T4 ligase were obtained from New England Biolabs or Thermo Fisher Scientific. DNA from PCRs and agarose gels was purified using the Illustra GFX PCR DNA and Gel Band Purification Kit (GE Healthcare) or QIAEX II Gel Extraction Kit (Qiagen). Restriction digests, ligations and E. coli transformations were done using standard techniques. Plasmid DNA was extracted from overnight cultures using the Zyppy Plasmid Miniprep Kit (Zymo Research) or QIAprep Spin Miniprep Kit (Qiagen) and confirmed by DNA sequencing. Plasmids and their construction details are listed in Supplementary Table 4. Plasmids were introduced into P. atrosepticum and P. aeruginosa strains by electroporation using standard techniques.

Selection and cloning of Racr candidates

Candidate Racrs were chosen on the basis of their similarity of sequence and secondary RNA structure to the relevant CRISPR repeats in the model system (MAFFT alignments, FastTree approximately maximum-likelihood phylogenetic trees; Extended Data Figs. 9a and 10a–c) and the presence of a promoter sequence within 250 bp upstream of the Racr candidate (promoter prediction using Bprom and manual curation). The Racr candidates were synthesized as gene fragments including flanking regions (Twist Biosciences) under the control of either the predicted wild-type promoter or P_BAD (Ara-inducible) from the predicted transcription start site (TSS). RacrIF1 variants were cloned through PCR with mismatched primers and overlap PCR. For variant 3, a hammerhead ribozyme was introduced for Cas6f-independent processing^50,51. RacrIF1 was then cloned downstream of BioBrick constitutive promoters of different strength (BBa_J23112, BBa_J23110 and BBa_J23100) to evaluate dose responsiveness. Detailed information on the candidate Racrs is listed in Supplementary Table 1.

Expression of Racr candidates and related constructs

For the experiments presented in Fig. 1d and Extended Data Figs. 2b,d and 4c,d,f, RacrIF1 and its variants, canonical and hybrid crRNAs and the isolated RacrIF1 repeat, were expressed in P. atrosepticum from a plasmid with a p15A origin of replication (copy number of around 10) either under the control of its wild-type promoter (NC_018012.1: 4,787,341–4,787,695 bp) or with the predicted TSS downstream of the P_BAD promoter (NC_018012.1: 4,787,535–4,787,695 bp) (Extended Data Fig. 1f). For the titration displayed in Extended Data Fig. 4c,d, RacrIF1 (NC_018012.1: 4,787,535–4,787,695 bp) was expressed from the P_BAD promoter under different Ara concentrations or the BioBrick constitutive promoters. For the experiments shown in Fig. 3f,g, Racr candidates tested in PAO1 or PA14 were expressed from the Escherichia–Pseudomonas (ColE1-pRO1600) shuttle vector pHERD30T with their predicted TSS downstream of the P_BAD promoter. RacrIF1 (experiment in Extended Data Fig. 1e) and RacrIC1 (experiment in Fig. 3e and Extended Data Fig. 9b) were cloned with the predicted wild-type promoter for the 5′ RACE assay. In Pseudomonas strains, pHERD30T replicates from the P. aeruginosa plasmid pRO1600 oriV and replication protein (copy number of around 13)^52,53.

Phage-resistance assay

Triplicate cultures of hosts carrying either a phage-targeting spacer (+CRISPR) or a non-targeting control (–CRISPR), and containing a plasmid expressing a candidate Racr or an empty-vector control (EV), were grown overnight in 5 ml LB supplemented with the appropriate antibiotics and inducers. For P. atrosepticum, a soft LBA overlay (0.35% w/v) containing 100 μl of the overnight cultures was poured onto an LBA plate supplemented with the corresponding antibiotics and inducers. For PA14, PAsmc and PAO1, a soft LBA overlay (0.5% w/v) containing 150 μl of the overnight cultures and supplemented with 10 mM MgSO₄ was poured onto an LBA plate supplemented with 10 mM MgSO₄ and the corresponding antibiotics and inducers. Phage titres were determined by pipetting 2.5 μl (or 5 μl for ΦTE) drops of serial dilutions of phage stock (approximately 10¹⁰ PFU per ml) in phage buffer onto the agar overlay and plates were incubated overnight. Plaques were counted after incubation overnight, with the phage titre represented as PFU per ml. When plaques were too small to count, one plaque was counted in the first dilution in which no plaques were visible. Type I-F Racr candidates were tested in P. atrosepticum PCF610 carrying the ΦTE targeting plasmid pPF1423 (for assays in Figs. 1d, 3g and Extended Data Figs. 2b and 4d) and P. atrosepticum PCF188 (for assays in Extended Data Figs. 2d and 4c,f) with the phage ΦTE, and P. aeruginosa PA14 with the phage DMS3m. Type I-E Racr candidates were tested in PAsmc, type V-A Racrs in PAO1::MbCpf1::crRNA24 (PAO1::V-A) and type I-C Racrs in PAO1 tagged with a I-C CRISPR–Cas system (PAO1::I-C), all of which were infected with the phage JBD30. The respective non-targeting (–CRISPR) control strains were P. atrosepticum PCF610 with the non-targeting plasmid pPF975 or wild-type P. atrosepticum, PAscm Δcas3 and wild-type PAO1.

Conjugation-efficiency assay

For the experiment shown in Fig. 1e, conjugation efficiency was assessed in a similar manner to that described previously⁵⁴. E. coli ST18 was the donor for the conjugation of the untargeted control (–CRISPR, pPF953) and type I-F (+CRISPR, pPF954) targeted plasmids. Plasmid pPF954 contains a protospacer targeted by spacer 1 from CRISPR1 (type I-F) and the canonical GG PAM. Recipients were wild-type P. atrosepticum that have either a plasmid expressing RacrIF1 from the P_BAD promoter (+RacrIF1, pPF2846) or an empty-vector control (–Racr1F1, pPF781). Strains were grown overnight in triplicate in 5 ml LB supplemented with Cm and Ara for recipients, or 5 ml LB supplemented with Tc and ALA for donor strains. One ml of overnight culture was pelleted and washed twice with LB supplemented with ALA to remove the antibiotics. Pellets were resuspended in 0.5 ml LB supplemented with ALA and Ara, and the OD₆₀₀ was adjusted to 1. Donors and recipients were mixed in a 1:1 ratio, and 10 μl was spotted on LBA supplemented with ALA and Ara, and incubated at 25 °C for 24 h. Next, the mating spots were scraped with a sterile loop and resuspended in 0.5 ml PBS, and dilution series were plated either onto LB supplemented with Cm and Ara for recipient counts or with the addition of Tc for selection of transconjugant counts. Conjugation efficiency was calculated as the ratio of transconjugants per recipient cells.

Co-expression and purification of Cas6f and RNA

For co-expression and purification of Cas6f and RNA variants, plasmids pPF2644 (His₆–Cas6f and type I-F crRNA repeat–spacer–repeat), pPF2868 (His₆–Cas6f and RacrIF1), pPF2869 (His₆–Cas6f and RacrIF1^GCmut) and pPF2640 (His₆-Cas6f alone) were transformed into E. coli LOBSTR cells. Overnight cultures were used to inoculate 500 ml LB plus kanamycin in a 2 l baffled flask and incubated at 37 °C and 180 rpm to an OD₆₀₀ of 0.2–0.3, followed by incubation at 18 °C and 180 rpm to an OD₆₀₀ of 0.6. Expression was induced with 1 mM IPTG, and proteins were expressed for 20 h at 18 °C and 180 rpm. Cells were collected at 10,000g for 10 min at 4 °C, and the pellet was resuspended in 10 ml g⁻¹ (wet-cell mass) lysis buffer (50 mM HEPES-NaOH, pH 7.5, 300 mM KCl, 5% (v/v) glycerol, 1 mM dithiothreitol (DTT) and 10 mM imidazole) supplemented with 0.02 mg ml⁻¹ DNase I, one tablet cOmplete EDTA-free protease inhibitor (Roche), 0.67 mg  ml⁻¹ lysozyme and 0.1 mM phenylmethylsulfonyl fluoride. Cells were lysed by ultrasonication and the lysate was clarified by centrifugation at 15,000g for 15 min at 4 °C. The cleared lysate was affinity purified using a 1 ml HisTrap™ FF (Cytiva) column equilibrated in lysis buffer and eluted using a gradient against elution buffer (lysis buffer containing 500 mM imidazole). Elution fractions were pooled and concentrated using a 10 kDa Nominal Molecular Weight Limit Amicon Ultra-4 Centrifugal Filter Unit (Amicon) and loaded onto a Superdex 75 Increase 10/300 GL (GE Healthcare) column equilibrated in SEC buffer (20 mM HEPES-NaOH, pH 7.5, 100 mM KCl, 5% (v/v) glycerol and 1 mM DTT). Protein concentrations were determined using a NanoDrop One Spectrophotometer (Thermo Fisher) and a Qubit Protein Assay Kit (Invitrogen). Aliquots of protein were stored at −80 °C. Protein samples were separated on an SDS–PAGE gel (Bolt 4 to 12%, Bis-Tris, 1,0 mm (Invitrogen)) and stained with Coomassie blue.

Expression and purification of type I-F Cascade

For expression and purification of the type I-F Cascade shown in Fig. 2b, plasmids pPF1635 (Cas8f–Cas5f–Cas7f) and pPF2644 (His₆–Cas6f and type I-F crRNA repeat–spacer–repeat) or pPF2868 (His₆–Cas6f and RacrIF1) were co-transformed into E. coli LOBSTR cells. Protein was expressed and purified as described above with the following modifications: lysis buffer contained 15 mM imidazole, elution fractions were pooled and concentrated using a 30 kDa Nominal Molecular Weight Limit Amicon Ultra-4 Centrifugal Filter Unit (Amicon) and concentrated samples were loaded onto a HiLoad 16/600 Superdex 200 pg (GE Healthcare) column equilibrated in SEC buffer.

RNA isolation from protein fractions

For the experiment shown in Fig. 2c, the different RNA variants were isolated from the purified His₆–Cas6f or type I-F complex by phenol–chloroform extraction, ethanol precipitation and resolved on a denaturing gel containing 15% (v/v) 19:1 polyacrylamide, 7 M urea and 0.5× TBE (45 mM Tris, 45 mM Boric acid, pH 8.3, 1 mM EDTA) (Novex). The gel was stained with SYBR gold (Invitrogen) and RNA was shown using the Odyssey Fc imaging system (LICOR). For samples with purified His₆–Cas6f only, the amount of protein was normalized before RNA isolation.

Small RNA extraction and sequencing

For the experiments shown in Fig. 1b,c and Extended Data Fig. 1d,g, triplicate cultures of wild-type P. atrosepticum that have either a plasmid expressing RacrIF1 from its wild-type promoter (+RacrIF1, pPF2845) or an empty-vector control (–RacrIF1, pPF781) were grown overnight in 5 ml LB supplemented with Cm. The overnight cultures were subcultured into 25 ml LB supplemented with Cm in 250-ml flasks from a starting OD₆₀₀ of 0.05 and incubated for 15 h up to stationary phase while monitoring culture growth (OD₆₀₀). Next, 1 ml (in triplicate) of each culture was centrifuged for 1 min at 13,000g. The supernatant was discarded and the pellet was resuspended in 1 ml RNAlater Stabilization Solution (Invitrogen) and stored at −20 °C. The small RNA fraction (less than 200 nt) was extracted using the mirVana miRNA Isolation Kit according to the manufacturer’s instructions. Residual genomic DNA was removed by treatment with TurboDNase (Thermo Fisher) according to the manufacturer’s instructions, and the absence of gDNA was confirmed by PCR. RNA purity, integrity and concentration were determined using a NanoDrop One Spectrophotometer (Thermo Fisher), a Qubit RNA High Sensitivity (Invitrogen) and an Agilent 2100 Bioanalyzer system with an RNA nano chip. Library preparation and sequencing of small RNA samples were carried out by Vertis Biotechnologie (Freising). In brief, the small RNA samples were first treated with T4 polynucleotide kinase. Then oligonucleotide adapters were ligated to the 5′ and 3′ ends of the RNA samples. First-strand cDNA synthesis was done using M-MLV reverse transcriptase with the 3′ adapter as primer. The resulting cDNA was amplified with PCR using a high-fidelity DNA polymerase. The cDNA was purified using an Agencourt AMPure XP kit (Beckman Coulter Genomics) and was analysed by capillary electrophoresis. For Illumina NextSeq sequencing, the cDNAs were pooled in approximately equimolar amounts. The cDNA pool was purified using the Agencourt AMPure XP kit (Beckman Coulter Genomics) and was analysed by capillary electrophoresis. The primers used for PCR amplification were designed for TruSeq sequencing according to the instructions of Illumina. The NGS libraries (six samples) were single-read sequenced on an Illumina NextSeq 500 system using a read length of 75 bp at a depth of 10.2–11.5 million reads and were returned as sequences in FASTQ format.

RNA-seq analysis

Generated reads in FASTQ format were initially processed by removing adaptors and low-quality reads using Trimmomatic⁵⁵. The quality of the reads was assessed using FastQC v.0.11.9 (ref. ⁵⁶) Processed reads were aligned to the P. atrosepticum (genome accession number BX950851.1) using Bowtie 2 (ref. ⁵⁷) with local parameters and the alignment was converted to BAM format using SAMtools v.1.16.1 (ref. ⁵⁸). The alignment was visualized and final images were generated using Geneious Prime 2022.1.1 (Dotmatics).

RNA structure prediction

The RNA structures in Fig. 1a,b and Extended Data Figs. 1b, 2a,c,  3d, 4e,  9a and 10 were predicted using the RNAfold web server⁵⁹ v.2.4.9 and visualized by RNA2Drawer⁶⁰ v.6.3 and Adobe Illustrator v.27.

5′ RACE

To identify the 5′ end of the mRNA encoding RacrIF1 (experiment shown in Extended Data Figs. 1e and 3a) or RacrIC1 (experiment shown in Fig. 3e and Extended Data Fig. 9b,c), 5′ RACE was used to identify the 5′ end of the RNA transcript using the template-switching enzyme from NEB. In brief, RNA was extracted from overnight cultures in triplicate (for RacrIF1, PCF610 carrying an empty-vector control (pPF781) or the RacrIF1-expressing plasmid (pPF2845); for RacrIC1, POA1::IC carrying a plasmid expressing the Acr locus under wild-type promoter expression (pSC144)) using the Zymo-Seq RiboFree Total RNA Library Kit (Zymo Research). Afterwards, a template-switching reverse-transcription reaction was used to generate cDNAs with a universal sequence of choice (introduced by a template-switching oligonucleotide) attached to the 3′ end of the cDNA (the 5′ end of the transcript) (NEB). A sequence-specific reverse-transcription primer was placed so that it binds in the respective Racr or crRNA sequence. In the second step, the 5′ end of the transcript was identified by PCR amplification with primers that bind upstream from the Racr processing site and in the template-switching oligonucleotide, respectively. Oligonucleotides used are listed in Supplementary Table 3. PCR products were visualized on gels and cleaned up. For RacrIF1 under wild-type promoter expression, the size of the 5′ RACE product was visualized on a gel and analysed on a fragment analyser (experiment shown in Extended Data Fig. 1d), while the 5′ RACE product for RacrIC1 was visualized on a gel and sent for Sanger sequencing for confirmation (experiment shown in Extended Data Fig. 3b,c). 5′ RACE was also done to confirm the identity of the RNA species isolated by phenol/chloroform extraction and ethanol precipitation from the purified type I-F complex (Fig. 2c). PCR products were A-tailed with DreamTaq polymerase (Thermo Fisher) and dATP and cloned into pGEM-T Easy Vector (Promega). Plasmids were isolated from individual colonies and Sanger sequenced (Extended Data Fig. 3b,c).

CRISPR-primed adaptation assay

The CRISPR adaptation assays shown in Fig. 2e and Extended Data Fig. 6b were performed as previously described⁶¹. A naive plasmid control (no matching protospacer, pPF953) and strong (AG PAM variant, pPF959) and medium (GT PAM variant, pPF967) priming-inducing plasmids were conjugated as described above (without Ara) into wild-type P. atrosepticum containing either a plasmid expressing RacrIF1 from the P_BAD promoter (pPF2846) or an empty-vector control (pPF781). The priming-inducing plasmids escaped targeting from the P. atrosepticum type I-F CRISPR–Cas system (Extended Data Fig. 6a). Strains with plasmids were grown in triplicate for 24 h in 5 ml LB supplemented with Cm and Tc. These ‘day 0’ cultures were then used to inoculate (1:500 dilution) 5 ml fresh LB supplemented with Cm, IPTG and Ara (without Tc selection), and incubated in the same conditions. This process was repeated for 5 days. Aliquots of culture from each day were mixed with 50% glycerol in a 1:1 ratio and frozen at −80 °C for future use. CRISPR array expansion (indicative of adaptation) was assessed by PCR using the cell glycerol stocks as a template. PCR products were loaded on a 2% agarose gel made up in 1× sodium borate buffer, run for 30 min at 180 V and stained with ethidium bromide.

CRISPR-primed plasmid clearance assay

Plasmid clearance, visualized in Fig. 2f, was measured as previously described³⁵. Cells from the CRISPR-primed adaptation assay (glycerol stocks) were diluted in 1 mL of PBS (1:1,000) and analysed using a BD LSRFortessa Cell Analyzer (BD Biosciences). A threshold was applied for FSC and SSC to detect bacterial cells. The mCherry was excited using a yellow–green laser (561 nm) and detected with a 610/20 nm bandpass filter; 20,000 events were recorded per sample using BD FACSDiva Software v.8 (BD Biosciences). Subsequent analysis was done using FlowJo Software v.10.8.1 (BD Biosciences). Cells were gated on SSC-A/SSC-H and SSC-A/FSC-A, then bifurcated (using BifurGate) into mCherry+ and mCherry− populations (Extended Data Fig. 6c). The ratio of mCherry− cells to total cells indicates the proportion of cells that cleared the plasmid.

SRUFinder

We built a dedicated bioinformatic algorithm⁴⁰ for finding SRU candidates in DNA sequences. The algorithm is available as a python package (https://pypi.org/project/srufinder) and a conda package (https://anaconda.org/russel88/srufinder) and is available at Zenodo. The algorithm is depicted as schematics in Extended Data Fig. 7. As queries, the algorithm uses a database of 17,823 non-redundant CRISPR repeat sequences with known associated subtypes (https://github.com/Russel88/SRUFinder/blob/master/data/repeats.fa). Repeats were obtained from the CCtyper⁶² web server (v. December 2020) and de-duplicated using cd-hit-env⁶³ at 100% identity and coverage. First, open reading frames (ORFs) were predicted using prodigal⁶⁴ in meta mode, and all ORFs with confidence ≥80% were masked from the input sequence. Next, repeat sequences were aligned with BLASTn⁶⁵ against the masked input sequence with task = blastn − short and word size  =6. Matches with identity less than 90% were discarded. Furthermore, matches with coverage ≥90% were considered to be full matches, whereas matches with coverage between 50% and 90% were considered partial matches. If any alignments overlapped, only the match with the highest bit score was kept. Then all full matches within 100 bp were clustered into arrays, and these repeats were disregarded as potential SRUs. Furthermore, if a partial match was within 100 bp of a solitary full match, it was considered a mini-array if the identity between the two was ≥90% (biopython pairwise2.align.global, default match/mismatch penalties, −1 open/extend gap penalties, no end gap penalty)⁶⁶. Then the remaining potential SRUs were aligned against the flanking 100 bp (biopython pairwise2.align.local, default match/mismatch penalties, −1 open/extend gap penalties). Because BLAST was observed to miss identifying repeats with several mismatches to the query, candidate SRUs showing partial matches (identity greater than 70%) to any of the two flanking regions (100 bp) were discarded to ensure that the SRUs were truly solitary. The remaining SRUs were then filtered by a bit score threshold of 41.1. This cut-off was set by running the algorithm on both intergenic (as described above) and intragenic (as above, but with ORF masking reversed) on the IMG/VR3 database³⁹, and using recursive partitioning trees (rpart 4.1–15; ref. ⁶⁷) to determine the best cut-off for distinguishing potential SRUs in intergenic regions (true candidates) from potential SRUs in intragenic regions (probably false-positive matches). We found that 84.0% of the matches with a bit score ≥41.1 were from intergenic regions, compared with 23.9% of matches with any bit score being from intergenic regions; 84.6% of matches with bit score of less than 41.1 were from intragenic regions.

Bioinformatic search for SRUs in databases

Prophages were extracted using VIBRANT 1.0.1 (ref. ⁶⁸) from the 104,858 high-quality genomes from GTDB (Version r95, 2020/10/06; ref. ³⁷), which yielded 437,636 prophages from 69,688 of the genomes. SRUfinder⁴⁰ was then run against these GTDB prophages, the PLSDB plasmid database (27,939 plasmid genomes³⁸) and the IMG/VR3 database (2,332,702 virus genomes³⁹). A virus dendrogram was created from the taxonomic information provided in the IMG/VR3 metadata and presented in Extended Data Fig. 8a. SRUFinder⁴⁰ was also run against the PHASTER database (65,668 prophage and virus genomes⁶⁹), but these SRUs were used only for finding candidates for validation. The filtered output of SRUFinder⁴⁰ can be accessed in Supplementary Data 1.

GTDB prophage analysis

A phylogenetic tree of the GTDB-derived prophages containing SRUs was made by calling genes with prodigal⁶⁴ and extracting 40 single-copy marker genes⁷⁰ using fetchMGs 1.2 (https://github.com/motu-tool/fetchMGs). Each marker gene was then aligned separately with mafft 7.310 (ref. ⁷¹), the alignments were concatenated and a tree was inferred using FastTree 2.1.10 (ref. ⁷²). Clades were collapsed with the collapse_tree_at_resolution function from R-package castor version 1.7.2 (ref. ⁷³) at resolution 0.01 with rename_collapsed_nodes = TRUE. The tree was visualized with iTOL v.5 (ref. ⁷⁴). Identification and subtyping of cas operons in the chromosomes was done with CCtyper 1.2.1 (ref. ⁷⁵). To determine whether there was a non-random association between the subtype of the SRU and the subtype of any cas operon in the chromosome, we firstly restricted the analysis to SRUs where the host had any cas operon (that is, IMG/VR3 hits were excluded given the lack of known host associations). For this subset (170 SRUs of 188 total in GTDB prophages), 82.9% of the SRUs had a cas operon of matching subtype in the host. When the subtype of the SRU was permuted, there was a mean association of 32.3% with a standard deviation of 2.7% across 1,000 permutations. The data are illustrated in Extended Data Fig. 8b and the alignments of SRU sequences from strains with CRISPR–Cas, along with their corresponding consensus CRISPR repeats, are available in Supplementary Data 2.

Association with acr genes

To establish whether racr candidates were co-located with acr genes, we used acr genes predicted by machine learning from ref. ⁷⁶. Only predicted Acrs with a score greater than 0.5 were considered. Acr protein sequences were aligned against all virus and plasmid genomes containing SRUs with tblastn v.2.11.0+(ref. ⁶⁵). Only matches with E-values ≤0.01 were kept. If matches were overlapping, only the match with the highest bit score was retained. To determine whether SRUs and acr genes are genetically co-located more often than random, the number of acr genes within 1 kb of an SRU was counted. This was then compared to the same statistic across 1,000 permutations in which the location of the SRU across the virus or plasmid genome was random. The data are depicted in Fig. 3c and a one-tailed P value was calculated as:

Statistics and reproducibility

The specific test used for assessing statistical significance is indicated in the figure legends. The exact P values of the statistical analyses are stated in Supplementary Table 5. Protein purifications, RNA isolations and the phage infection assay on PAO1::I-C were independently repeated twice. Small RNA-seq, 5′ RACE, conjugation efficiency, primed adaptation and the phage infection assay with induction of Racr expression at different ALA concentrations were performed once with three independent biological replicates. All the other phage infection assays were independently repeated at least three times.

Data visualization

Unless stated otherwise, data processing and visualization were done in Microsoft Excel v.16, Prism v.9.2.0 (GraphPad), SnapGene v.7.0.2 and Geneious Prime v.2022.1.1, and subsequently edited in Adobe Illustrator v.27. For gel source data, see Supplementary Fig. 1.

Reporting summary

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