Type III CRISPR–Cas systems preferentially acquire spacers from early-expressed phage genes in a transcription-dependent manner

Abstract

CRISPR–Cas immunity depends on acquiring spacers that generate functional CRISPR RNA (crRNA) guides. For most RNA-targeting type III systems, this poses a paradox: spacers are captured from DNA but defense requires RNA interaction, thus the system must infer transcriptional relevance of DNA fragments. We compared spacer acquisition profiles for co-occurring type III (RNA-recognizing) and type II (DNA-recognizing) systems in Streptococcus thermophilus during lytic phage infection. Type III spacer acquisition was highly enriched for early phage genes whereas type II spacers were distributed genome wide. High-throughput spacer sequencing showed that the type III early gene bias arose during adaptation, was independent of interference, and peaked immediately downstream of a conserved early phage promoter. This bias for early phage promoters was also observed in natural type III spacers identified in publicly available S. thermophilus genomes. Cloning this early phage promoter and downstream gene into a plasmid recreated the type III-acquisition hotspot. Mutation to disrupt the promoter eliminated the spacer hotspot, while mutation to disrupt translation enhanced it. Thus, type III-A adaptation preferentially samples DNA associated with the earliest phage transcriptional program, enriching for spacers that generate crRNAs targeting early phage RNA.

Graphical Abstract

Graphical Abstract.

Introduction

CRISPR–Cas systems provide the only known adaptive immunity in bacterial and archaeal cells [1, 2]. This immunity is achieved by capturing short fragments of nucleic acids from invading mobile genetic elements (MGEs), such as plasmids and viruses, and integrating them as new immune memories, or “spacers” in CRISPR arrays (a process termed adaptation) [1, 3]. These MGE-derived spacers are transcribed and processed to produce CRISPR RNAs (crRNAs) [4] that assemble with CRISPR-associated (Cas) proteins into ribonucleoprotein effector complexes (crRNPs) [5, 6]. The crRNPs surveil the cell for complementarity between the crRNA and target nucleic acids, and upon recognition either the target is degraded, cellular dormancy or death is induced, or both (a process termed interference) [6, 7].

To confer immunity, CRISPR adaptation must generate spacers, and thus crRNAs, that are functional for their cognate effector complexes. For many CRISPR–Cas subtypes, the crRNA binds and initiates interference against DNA targets, but for a minority of systems, like the type III-A, the target is RNA [6, 7]. Despite this difference during interference, both DNA-targeting and RNA-targeting systems appear to primarily use DNA fragments when generating new spacers, with a handful of exceptions [8, 9]. For RNA-targeting systems, this creates an apparent disconnect between spacer acquisition and interference, as not all DNA fragments necessarily correspond to transcribed RNAs. DNA from non-transcribed regions would not generate functional RNA-targeting spacers, and because transcription often occurs from one DNA strand, only spacers in one orientation will produce crRNAs complementary to mRNA.

RNA-targeting systems have a second challenge compared to DNA-targeting systems: the crRNA cannot initiate interference until a target has been transcribed, and infection often progresses quickly. Lytic phages exert tight temporal regulation of gene expression to maximize replication [10–12]. Early genes, expressed immediately upon injection [10, 13], are often involved in reprogramming of the host cell to promote phage propagation [14–16] and can inhibit host anti-phage defenses (e.g. anti-CRISPR proteins [17–19]). Expression of middle and late genes follows within minutes [10, 13], and the replicative cycle of lytic phages infecting lactic acid bacteria is often complete within ∼30–40 min [10, 20–22].

In DNA targeting systems, dedicated adaptation cofactors ensure that newly acquired spacers are compatible with interference [23, 24]. Cas4 in many type I systems, and Csn2 together with Cas9 in type II systems, enforces the acquisition and integration of spacers with appropriate orientation [25, 26] and flanking sequences (protospacer adjacent motifs, or PAMs) [23, 26–29]. In contrast, no such mechanisms are known for type III systems, and it is unclear whether and how RNA-targeting CRISPR–Cas adaptation selects functional spacers.

Despite these challenges for RNA-targeting systems, spacers found naturally in type III arrays are protective against phage lysis [30, 31], indicating that functional RNA-targeting spacers are favored. This enrichment could occur during adaptation through preferential capture and integration of functional spacers, and/or through differential survival of infected cells, whereby only cells encoding spacers that confer immunity persist in the population. Available studies point to a combination of both intrinsic adaptation biases and selection effects. For the Thermus thermophilus type III-A system challenged with lytic phage phiFa, newly acquired spacers were detected across the phage genome from both strands, but were non-uniformly distributed and survival enriched for spacers in early-expressed genes [32]. Similarly, in Staphylococcus aureus harboring a Staphylococcus epidermidis type III-A system, adaptation targeting early-expressed phage genes was strongly enriched in cells surviving infection [33]. In these infection studies it was difficult to determine whether biases for early phage genes arose during spacer acquisition or from the selective pressure imposed by phage infection.

In studies looking at new spacers derived from plasmids or the chromosome, there was evidence for intrinsic adaptation biases independent of selection for survival. In Streptococcus thermophilus, type III adaptation hotspots were observed at highly transcribed chromosomal loci (including rRNA and tRNA operons), at sites associated with transcriptional pausing (e.g. riboswitches in 5′ UTRs), downstream of genomic promoters, and within the first ∼10–50 bp of ORFs [34]. Additionally, some plasmid features formed hotspots, e.g. single-stranded origins of rolling-circle replication [34], which are primed by RNA polymerase or primases to initiate lagging strand replication [35–38]. The co-occurring type II system did not favor spacers from these different features [34]. Similarly, in Staphylococcus epidermidis, type III adaptation was also enriched near rRNA/tRNA loci, chromosomal termini, and engineered double-stranded DNA breaks [39]. While the co-occurring type II system also captured spacers near chromosomal termini and DNA breaks, the spacer enrichment over rRNA/tRNA loci was unique to type III adaptation [39]. Importantly, these enrichment patterns were observed irrespective of interference, indicating that they do not arise from survivorship-derived selection [34, 39]. Together, these observations supported a true innate type III selectivity for transcribed DNA during adaptation, but these observations were correlative and the exact requirements for type III adaptation hotspots remained unclear.

Here, we provide an in-depth comparison of spacers and their targets across the four CRISPR loci commonly found in S. thermophilus. We found that naturally occurring type III CRISPR arrays were uniquely enriched for spacers targeting early-expressed phage genes. Likewise, early expressed phage genes were enriched in type III CRISPR arrays of laboratory-isolated bacteriophage insensitive mutants (BIMs) and in high-throughput type III adaptation assays. This bias was not apparent for the two co-occurring type II systems. Further experiments revealed that the preference for early-gene targeting spacers was independent of interference, occurring at the level of adaptation, and required a conserved promoter found upstream of early genes in all examined S. thermophilus phage genomes. Taken together, these results show that type III adaptation preferentially samples DNA underlying key transcriptional events.

Materials and methods

Generating a database of S. thermophilus CRISPR–Cas systems

Streptococcus thermophilus genomes (all available complete and partial assemblies, as of October, 2025) were downloaded from NCBI. CRISPRCasTyper [40] was used to identify CRISPR–Cas systems and their arrays (along with corresponding spacers in each array). In cases where a complete set of cas genes was identified, but no array was detected, we used a custom approach to search for noncanonical or short arrays (for example, arrays with a single spacer between two repeats) within 3 kb of the identified cas genes. Briefly, a region extending 3 kb upstream and downstream from the cas genes was scanned for sequences with at least 70% identity to a reference repeat derived from S. thermophilus DGCC7710. If found, this repeat was noted as a potential CRISPR repeat sequence for that genome and nearby regions searched to annotate potential arrays. Spacers were then extracted from these predicted arrays and aligned to the S. thermophilus phage genome database described below.

Generating a database and meta-genome of S. thermophilus phages

All sequences that included the terms “streptococcus phage” and “complete genome” were downloaded from NCBI (14 Oct 2025), yielding 752 unique phage genomes. For each genome the lysin gene was located either by finding the existing annotation (“lysin,” “endolysin,” “peptidoglycan hydrolase,” “amidase,” or “cell wall hydrolase”), by homology search with known lysins using an 80% protein identity cutoff, or by manual comparison to ORFs within closely related phages. Once the lysin was located, genomes were re-oriented such that the stop codon of the lysin was at the end of the revised reference sequence. Genes downstream of the lysin, which are presumed to be early expression genes based on conserved genome organization in lambdoid phages generally [16, 41, 42] and limited transcriptional studies on S. thermophilus phages specifically [10, 13, 22], would therefore be positioned at the start of the revised sequence. Spacers were mapped to all revised phage sequences at 90% nucleotide identity using Geneious Prime 2025, and their positions noted relative to the start points. The number of spacers overlapping each nucleotide were added across all phages and plotted to show overall spacer occupancy relative to the lysin-adjacent site for a phage “meta-genome.”

Streptococcus thermophilus strains and phages

Streptococcus thermophilus strain JIM8232 (WT) was kindly provided by Pierre Renault (AgroParisTech, France). Streptococcus thermophilus DGCC7710 and phages 2972, 858, D3288, D5913, and D6037 were kindly provided by Sylvain Moineau (Laval University, Canada). Streptococcus thermophilus strains were grown in M17 (HiMedia), and the cultures were incubated at 37°C overnight, or at 42°C during the day.

Phage amplification and titer calculation

For phage amplification, 5 ml overnight cultures were started from a glycerol stock of S. thermophilus DGCC7710 in M17 (HiMedia) and incubated at 37°C. The next day 100 μl of overnight culture was added to 10 ml of M17 containing 10 mM CaCl₂. A glycerol scraping of phage lysate was resuspended in 200 μl of phage buffer [50 mM Tris–HCl (pH 7.5), 100 mM NaCl, 8 mM MgSO₄], and 100 μl of this phage resuspension was added to the overnight culture. This was incubated at 42°C for 5 h, then centrifuged at 3000 rpm for 15 min, and the supernatant was filtered through a 0.22-μm filter [43, 44]. A second round of phage amplification was performed, again starting with 5 ml of an overnight S. thermophilus DGCC7710 culture. The following day 10 ml of culture was grown in M17 to an OD₆₀₀ of 0.2–0.25. CaCl₂ was added (10 mM) followed by 100 μl of the first round of phage amplification. Cultures were incubated at 42°C for 5 h before filtering through a 0.22-μm filter. Phage lysates were stored at 4°C. To enumerate plaque forming units, 10-fold dilutions of phage lysate (diluted in phage buffer) were mixed with 500 μl of saturated S. thermophilus DGCC7710 culture and 3 ml top agar (M17 + 10 mM CaCl₂ + 0.7% agar) and spread evenly over M17 plates. After overnight incubation at 42°C, plaques were counted and averaged between three replicates.

Phage infection and host cell genomic DNA extraction

Streptococcus thermophilus strains were used to inoculate 10 ml M17 and grown overnight at 37°C. Cultures were diluted 1:100 in fresh M17 and incubated at 42°C to an OD₆₀₀ of 0.2. CaCl₂ was added to 10 mM and the culture divided into 10 ml aliquots. Phage lysate was added to the aliquots to a multiplicity of infection (MOI) of 1, 10, or 100. After phage infection, cultures were either incubated for 25 min, until lysis (as indicated by full clearing of liquid media—~1 h), or overnight (16 h). At each time point, cells were pelleted immediately via centrifugation at 3500 × g for 10 min at 4°C, supernatant was discarded, and the pellet was frozen at −80°C. For genomic DNA preparation, pellets were thawed at room temperature, and DNA was extracted with Quick-DNA Fungal/Bacterial Miniprep kit (Zymo Research).

BIM isolation and analyses

BIMs were isolated through an adapted protocol from [13, 45]. In short, S. thermophilus strains were used to inoculate 10 ml M17 and grown overnight at 37°C. The following day 10 ml M17 was inoculated with 100 μl of overnight culture and incubated at 42°C to an OD₆₀₀ value of 0.6. Phage dilutions were added to 300 μl cells at an MOI of 100 or greater in 3 ml of top agar and spread evenly over M17 plates before incubation at 42°C overnight. Colonies were screened for CRISPR array expansion to identify CRISPR BIMs using primers binding within the leader and first known spacer (primer sequences in Supplementary Table S1). PCR (polymerase chain reaction) products were Sanger sequenced (Eurofins genomics) and new spacers were mapped to the phage 2972 genome.

Phage reinfection assays

Streptococcus thermophilus strains and BIMs were grown in 10 ml of M17 overnight at 37°C. Cultures were diluted to an OD₆₀₀ of 0.2, CaCl₂ added to 10 mM, and phage lysate was added to reach the appropriate MOI. Cultures were aliquoted into a 96-well cell culture plate at 200 μl per well. Plates were incubated at 42°C with orbital shaking in a Biotek Epoch2 Spectrophotometer and OD₆₀₀ was collected every 2 min. Growth curves were performed in triplicate, and the average of these replicates were graphed in GraphPad PRISM 10.2.2.

To quantify and isolate phage CRISPR escapers, BIMs were reinfected with phage 2972 at an MOI of 100 in top agar as described above. After incubation at 42°C overnight, plaques were counted and isolated plaques amplified in their corresponding BIM as described above. A random subset of phages was further isolated and re-amplified for sequencing to identify the mechanism of escape. For this, the region surrounding each protospacer was amplified by PCR (primers in Supplementary Table S1) and sequenced (Eurofins genomics).

Adaptation assays and analysis

Adaptation assays were performed using CAPTURE [46], optimized for this model system as previously described [34]. Briefly, PCR was performed on gDNA with a forward primer binding the CRISPR array leader and a reverse primer binding the leader-adjacent spacer (sequences in Supplementary Table S1). Amplicons were separated by gel electrophoresis and expanded products purified using a gel recovery kit (Zymo Research). A second round of PCR with CAPTURE primers (Supplementary Table S1) selectively amplified products with new spacers and added partial Illumina adapters. A final round of PCR amplification then added sequencing and barcoding adapters. PCR products were ranked and pooled by intensity to prevent over-representation of more intense PCR products, then concentrated by ethanol precipitation, and finally quantified and diluted to a suitable concentration for Illumina platform sequencing.

Pooled libraries were sequenced on Illumina instruments (either MiSeq or NextSeq 2K), then demultiplexed by index. Read 1 sequences were used in this study. For each read, flanking repeats were located and the spacer sequence in between was extracted. Spacers were aligned by bowtie [47] to a reference sequence containing the bacterial chromosome and any phage genome or plasmid relevant to the sample. Spacer alignment output files were sorted, indexed, and formatted using samtools [48] to produce read depth tracks (bedgraphs) viewable on the UCSC Genome Browser [49]. The spacer depth files were normalized to the total number of aligned spacers in that replicate (SPM = spacers per million). For “total” spacer tracks, all reads with aligned spacers for a given replicate were included when generating the read depth tracks. In addition, we created “unique” spacer tracks for each replicate. For these tracks, spacers with identical alignments (with respect to alignment start position, stop position, and strand) were treated as PCR duplicates, collapsed, and represented as a single instance. Lastly, we generated “merged” tracks: after visually comparing all replicates from a given strain and treatment to ensure consistency, we combined their respective spacers. For unique merged tracks, identical spacers were collapsed to one representative within a replicate, but not between replicates, as they were generated in separate PCRs.

In previous work [34] we described array duplications, wherein existing spacers in the array are duplicated or re-arranged into the leader-adjacent position. These existing spacers are then captured in our PCR amplicon libraries and sequenced as “new” leader adjacent spacers. This process does not require Cas1 and appears to be independent of true adaptation [34]. Consequently, in this study we did not count duplication events as adaptation and we removed spacer duplications from our merged alignment output files prior to generating read depth tracks.

To quantify spacer hotspots over the cloned ORF31 in pTRK882, we used bedtools intersect [50] to detect and count all aligned spacers that overlapped this feature. Overlaps were quantified for both total and unique spacers sets.

To optimize the number of amplicon reads containing new spacers, we compared the numbers of new, unique spacers detected per sample for three different infection time points and three phage MOIs. The time points were 25 min (adsorption), ~1 h (lysis), and 16 h (overnight) and the MOIs were 1, 10, and 100. Cells were first grown to an OD of 0.2, then inoculated at one of the three MOIs, and finally grown until the designated time point. Cells were immediately spun at 4°C, then frozen and kept at −80°C until gDNA extraction, as described above. Adaptation in the CR1 and CR2 arrays was assessed (CR3 array was not assessed as it was expected to have a similar profile to CR1). The greatest numbers of new, unique spacers were recovered at an MOI of 100 and either lysis or overnight time points; these conditions were selected for the rest of the study.

Identification of phage termini

To identify the defined termini and cos site of D5913, Oxford Nanopore full-length sequencing was used (Plasmidsaurus, San Francisco, CA). The terminal 25 bp from the 5′ and 3′ ends of each read were extracted and aligned to the D5913 genome. We generated UCSC Genome Browser read pileup tracks to visualize peaks consistent with genome termini.

Genome deletions and transformation of plasmids in S. thermophilus

To generate the ∆csm1-6 strain, established protocols for homology dependent recombination were performed [51]. Briefly, S. thermophilus cells were grown overnight in M17 at 37°C. A 1-ml aliquot of overnight culture was centrifuged at 2600 × g for 2 min and the pellet was washed with chemically defined media (CDM) twice. Cells were then resuspended in 1 ml of CDM, diluted to an OD₆₀₀ of 0.075 and further incubated at 37°C for 1 h. Cells were then frozen at −20°C. For transformation, cells were thawed over ice, aliquoted into 100 μl, and 200 ng of cleaned PCR product (Zymo Research) was added along with 1 mM ComS peptide (LPYFAGCL). PCR product encodes homology arms sequence-identical to regions immediately upstream and downstream of the Csm1-6 operon. Cells were then incubated for 4 h at 37°C before being spread on 1% M17 agar plates. Colonies were verified for correct insertions by PCR and amplicons sequenced. Plasmids used in this study were generated through Gibson Assembly with primers listed in Supplementary Table S1. Plasmid transformation was carried out as above, with cells plated on media containing erythromycin (10 μg/ml).

Results

Natural type III CRISPR spacers are uniquely enriched for early phage genes

The abundance of sequenced genomes for S. thermophilus and associated phages presented an opportunity to characterize and compare spacers across four endogenous CRISPR–Cas subtypes commonly found in this species (Supplementary Fig. S1). We analyzed 715 S. thermophilus genomes (partial and complete) for the presence of CRISPR–Cas systems and arrays. A total of 1788 systems were found that could be assigned to subtype: type II-A (CR1; n = 677), type III-A (CR2; n = 640), type II-A (CR3; n = 401), and type I-E (CR4; n = 70). These systems were associated with 2743 CRISPR arrays, although this number is likely inflated by fragmented arrays in contig-level assemblies. Arrays were oriented and spacer sequences were extracted, resulting in a total of 27,459 spacers (6402 unique). Type III-A (CR2) arrays were notably shorter, averaging only four spacers, with many arrays containing a single spacer. In contrast, type II-A arrays averaged 20 (CR1) and 14 (CR3) spacers, respectively, while type I-E (CR4) arrays averaged 12 (Supplementary Fig. S2A). The spacer sizes differed between CRISPR–Cas subtypes, with type III-A spacers generally being longer and more variable (Supplementary Fig. S2B).

To identify protospacers (targets), these spacers were aligned (requiring at least 90% identity) to a custom database of 752 complete genomes of phages infecting Streptococcus (see the “Materials and methods” section). Genes in S. thermophilus phages can be generally assigned to three temporal classes referred to as early, middle, and late genes, with holin and lysins being the last expressed genes, as represented by phage 2972 (Fig. 1A). Alignments were compiled and displayed along an axis representing a composite genome (a phage “meta-genome”, see the “Materials and methods” section) to visualize genome-wide phage targeting (Fig. 1B). There were clear differences in spacer alignments of the four CRISPR–Cas subtypes (Fig. 1B). Type III-A spacers showed a strong enrichment over early-expressed phage genes and a strong bias for protospacers mapped to the “minus” strand of the phage meta-genome. In a typical phage, for example phage 2972 (Fig. 1A), all ORFs are encoded on the same strand, in this case the “plus” strand. The phage meta-genome is also oriented with the coding strand on top. The abundant minus-strand spacers in type III-A systems (Fig. 1B) would therefore be predicted to bind mRNA and be competent for interference. In contrast, spacers from the DNA targeting type II-A and I-E systems had more balanced spacer coverage on the plus and minus strands and generally followed the abundance of system-specific PAM elements (Fig. 1B).

Figure 1.

Natural spacers of the four CRISPR–Cas systems in S. thermophilus genomes show sub-type-specific targeting biases. (A) Schematic of phage transcriptional temporal classes using phage 2972 as an example. Early, middle, and late expressed genes are shown in green, blue, and red, respectively. The lysin gene ORF is shown in purple. This linear representation of the genome has early genes on the left-hand side, but in the conventional representation, early genes are on the right-hand side. (B) A set of 752 complete phage genomes was curated and given a common layout by linearizing each genome with the stop codon of the lysin gene as the last coordinates and the next base as position “1.” Spacers from sub-typed CRISPR arrays were mapped to the set of phage genomes. Tracks show cumulative aligned spacers. Spacers that aligned to the phage coding strand (top track) are shown in blue; spacers aligned to the noncoding strand (bottom track) are shown in pink. The proportion of all spacers aligned to the top versus bottom strand is shown for each CRISPR type. The proportion of available PAM sequences on top and bottom strands for the set of targeted phage genomes is shown as well. (C) Sequences immediately upstream and downstream of aligned spacers were extracted, and sequence logos were generated to visualize PAMs (or absence of a PAM, for type III-A).

Alignments also allowed us to determine the upstream and downstream sequence context of each protospacer. We observed candidate PAM motifs consistent with prior reports: CR1 (type II-A) showed a 3′-NNRNAA motif, CR3 (type II-A) showed a 3′-NGGNG motif, and CR4 (type I-E) showed a 5′-AA PAM (Fig. 1C) [52–54]. The type III-A protospacers had no apparent PAM, as expected (Fig. 1C) [34].

Spacers of laboratory-generated type III-A BIMs are enriched over early phage genes, and this bias is not explained by survival or growth bias

We next evaluated whether the type III bias for early-expressed phage genes is also found in laboratory-generated CRISPR bacteriophage insensitive mutants (BIMs), i.e. S. thermophilus cells that survive phage infection by acquiring new, protective spacers. Streptococcus thermophilus strain DGCC7710 has four co-existing CRISPR–Cas systems (Supplementary Fig. S1) and is susceptible to multiple isolated phages, facilitating infection studies. However, the CR2 (type III-A) system in this strain acquires new spacers at very low efficiency (unpublished data) and CR2 BIMs are not observed [55]. Therefore, we used a previously generated strain that has the JIM8232 type III-A system cloned in place of DGCC7710’s native type III-A system [13, 34]. Using this strain, we recovered BIMs resistant to phage 2972, with new spacers primarily in the CR1 array, followed by the CR3 array and the transplanted CR2 array, in proportions roughly similar to what has been previously found [34, 43, 53, 55]. No BIMs for CR4 (type I-E) were isolated, consistent with previous results showing infrequent spacer acquisition by this system [55]. For CR1 and CR3 type II BIMs, we sequenced 30 and 22 isolates, respectively, to identify spacer targets. The acquired spacers targeted sites across the entire length of the phage genome (Fig. 2A and Supplementary Tables S2 and S3), with no apparent preferred regions or ORFs. For CR2 (type III-A) we selected 130 BIMs for sequencing (Supplementary Table S4), and found that BIM spacers were strongly biased toward early genes, with only 2 spacers targeting middle genes and 3 spacers targeting late genes. Of the 125 BIMs with spacers against early expressed genes, 25 (i.e. 20%) targeted a single gene, ORF31, which encodes a cro-like repressor (Fig. 2A). Type III-A spacers also exhibited a strong strand bias, with nearly all spacers aligned to the minus strand, consistent with the requirement for RNA recognition by type III systems. Five exceptions were noted with spacers complementary to the template strand of annotated ORFs; these five spacers were clustered around the same region between ORF37 and ORF38. Interestingly, RNA-Seq data previously generated in our lab [13] revealed the presence of an antisense RNA in this region (Supplementary Fig. S3), potentially explaining how these spacers can support immunity.

Figure 2.

Isolation and characterization of CRISPR BIMs following infection with phage 2972. (A) Sequences of 30 isolated type II (CR1) BIMs, 22 isolated type II (CR3) BIMs and 130 type III (CR2) BIMs were aligned to the phage 2972 genome; distributions show enrichment of CR2 BIMs around ORF31. Middle, late, and early expressed genes are shown in blue, red, and green, respectively. The lysin gene ORF is shown in purple. A small hypothetical ORF with low expression could not be assigned as early, middle, or late and is colored gray. (B) Sequence logos of sequences adjacent to aligned CR1, CR3, and CR2 spacers were generated to show PAM or PFS sequences. (C) A set of 15 type III BIMs with spacers against early-expressed phage genes was selected for reinfection in liquid culture to monitor growth. Curves show average optical density (OD₆₀₀) for uninfected, MOI 1, and MOI 100 cultures, performed in triplicate and plotted at every 10 min for 8 h. Raw OD data used to generate plots can be found in the Supplementary data spreadsheet. At least one BIM with a spacer for each early expressed ORF was tested, with the exception of ORFs 40 and 42 for which no BIMs were isolated. Three BIMs harboring a spacer for ORF31 were tested, as well as two each for ORFs 34 and 35. Wildtype controls are shown in black. (D) Reinfections of five type III BIMs containing spacers targeting either middle or late 2972 genes plotted the same as in panel (C). Tested BIMs contained spacers against ORFs 5, 6, 16, 18, and 28. Wildtype controls are shown in black.

We next examined the sequence context (protospacer flanking sequence, PFS) for each CR2 BIM-derived spacer. The PFS can influence interference and prevent autoimmunity in type III systems by modulating base pairing between the 5′ crRNA tag and target RNA (Supplementary Fig. S4A) [56–58]. No consistent signal was identified among type III-A BIMs (Fig. 2B). Several PFS sequences contained partial complementarity to the crRNA 5′ tag (up to three of the four key positions, Supplementary Fig. S4B and C), yet still conferred full immunity, suggesting that suppression may occur only with perfect complementarity in positions 2–5 as was observed in S. thermophilus [58] and Lactococcus lactis [57]. The type II BIMs showed canonical PAM motifs (CR1: NNAGAAW; CR3: WNGGNG) (Fig. 2B). Spacer length distributions of laboratory-generated BIMs were similar to those of natural spacers, with type II spacers averaging ∼30 nt and type III-A spacers ranging more broadly between 33 and 41nt (Supplementary Figs S2 and S5).

Previous reports have shown conflicting effects for spacers targeting late phage genes; work in our lab showed no growth delays following phage reinfection for four S. thermophilus BIMs with middle or late targeting spacers [13], but dormancy or growth arrest was observed for late targeting spacers in S. aureus [33]. We compared growth of our newly isolated BIMs targeting early, middle, and late genes in liquid culture under phage pressure. We inoculated the selected BIMs and WT strain (no immunity) with phage 2972 at MOIs of 1 and 100. All 22 tested BIMs, regardless of temporal class, displayed full resistance to reinfection and grew comparably to uninfected controls even at the highest MOI of 100 (Fig. 2C and D). In addition, a subset of BIMs was assayed on solid media and found to be fully resistant to plaquing (Supplementary Fig. S6).

To further compare functional outcomes across systems, we performed similar challenges on type II BIMs under identical conditions. CR1 (II-A) BIMs generally retained resistance across all MOIs, though two of nine BIMs showed growth collapse under high phage loads (Supplementary Fig. S7A). CR3 (II-A) BIMs were more variable—six of nine cultures lysed at high phage MOI despite resisting reinfection at lower titers (Supplementary Fig. S7B). To understand this variability, we explored how susceptible each of the tested BIMs was to phage escape. One explanation for growth collapse is the emergence of phage escaper mutations, which may be more readily attained against type II targeting because a single SNP in the PAM or seed region can abrogate interference [52, 59]. In contrast, type III systems require larger deletions or more substantial changes to eliminate transcript recognition [32, 60]. Consistent with these reports, we recovered hundreds of escaper phages from the 18 type II BIM cultures (Supplementary Fig. S7C and D) but recovered only a single escaper phage from the 22 type III BIM cultures despite multiple attempts. We randomly selected and sequenced 27 escaper phages from 10 different type II BIMs and in every case observed SNPs in the PAM and seed regions consistent with loss of type II interference (Supplementary Table S5). In striking contrast, the single type III escaper phage contained a 22-bp deletion overlapping with the majority of the protospacer, which lies in ORF6 encoding a capsid protein (Supplementary Table S6). We also sequenced phage escapers in three type III BIMs (targeting ORFs 6, 32, and 40) in a previously generated S. thermophilus strain, which only contains the type III system. Each of the three sequenced escapers also contained large deletions (11, 36, or 69 bp in length), which overlapped the majority of the respective protospacers (Supplementary Table S6), and the in-frame deletions may indicate that these are essential genes. Collectively, these results demonstrate that both natural and experimentally generated type III-A spacers show a reproducible preference for early phage genes, a pattern not shared by type II systems. Moreover, because middle and late spacers can still provide robust phage resistance, early gene enrichment is best explained by a bias in adaptation rather than a survival or growth-based selection for BIMs with early-gene spacers. Furthermore, the type III BIMs generated here showed a greater ability to resist phage escape with no evidence of growth delays even under the highest MOIs tested.

Type III spacer bias for early genes occurs during spacer capture

We next examined spacer capture at the population level using a high-throughput assay in a range of S. thermophilus strains. These assays showed numerous new spacers without selecting individual surviving clones, and allowed us to compare type II and III spacers from the same pool of cells. The strains were challenged with phage 2972 at an MOI of 100, amplicon libraries of CRISPR arrays were prepared, and new spacers were identified as previously described [34, 46]. Similar to laboratory-generated BIMs, newly acquired type III spacers were particularly frequent at the first expressed gene, ORF31 and tapered over several downstream early genes, with low levels of adaptation across the remainder of the 2972 genome (Fig. 3A and B). This profile closely mirrored the pattern seen for BIM-derived spacers (Fig. 2A) and natural spacers (Fig. 1B). ORF31 is the first gene downstream from the promoter that drives early transcription immediately after infection (Fig. 3B). While some spacers were acquired within the promoter region of ORF31, the primary “hotspot” started ∼10–50 bps into the ORF31 coding sequence (Fig. 3B). Similar enrichment ∼10–50 bp downstream of start codons was observed previously for type III spacers acquired from the S. thermophilus genome [34], suggesting a shared mechanism. In contrast, type II (CR1 array) spacers were broadly distributed across the phage genome with no enrichment over early genes (Fig. 3A). Both type II and type III spacer profiles were similar at all time points (25 min after infection, lysis, and overnight; data not shown).

Figure 3.

Population-based mapping of newly acquired spacers during infection with phage 2972 reveals a type III-specific early gene hotspot. Select strains of DGCC7710 were infected with phage 2972 at an MOI of 100, grown until lysis or overnight, and then analyzed for new spacers integrated into the CR2 (type III) or CR1 (type II) arrays. (A) Full phage genome view of aligned spacers (protospacers). The phage gene map shows ORFs, with blue, red, and green bars indicating middle, late, and early expressed genes, respectively. The lysin ORF is shown in purple. Tracks show cumulative pile-up of spacers aligned to the top strand in purple and bottom strand in pink, normalized to total spacer counts. Inverted triangles show the positions of protospacers targeted by each of the four tested BIMs. Green triangle = protospacer for type III early BIM, targets ORF34; blue triangle = protospacer for type III middle BIM, targets ORF6; red triangle = protospacer for type III late BIM, targets ORF18; and black triangle = protospacer for the type II BIM, targets ORF5; all tracks show merged unique spacers from multiple replicates: WT type III (n = 14), WT type II (n = 13), ∆csm1-6 (n = 11), Early BIM (n = 9), Middle BIM (n = 9), Late BIM (n = 8), and type II BIM (n = 9). (B) Zoom of WT type III and ∆csm1-6 type III spacer profiles alongside an RNA-seq read depth profile (in blue) from 5 min post-infection. Spacers were enriched around ORF31, encoding a cro-like repressor gene, which lies immediately downstream of a promoter that drives early transcription. RPM = reads per million, SPM = spacers per million.

This high-throughput, population-based approach can capture adaptation events that ultimately prove auto-lethal or not protective against phage lysis. However, relative abundance of adaptation events could still be influenced by host cell growth and survival, so we followed up in a ∆csm1-6 strain, which has no type III interference (see Supplementary Fig. S1 for deleted genes, indicated by gray box). Notably, adaptation profiles were similar between the WT and ∆csm1-6 strains, showing the same strong preference for ORF31 and downstream early genes (Fig. 3A and B).

In phage-sensitive strains, lysis at 60 min post-infection will stop further spacer acquisition through host cell death. If spacer processing and integration is a slow process, only spacers arising from early events in phage infection may be detectable, biasing adaptation patterns. We therefore investigated adaptation profiles in one type II and three type III BIMs, which should all be immune to phage lysis, but allow infection to progress to different points. For the type II BIM, interference can start as soon as target DNA is detected (upon injection). The three type III BIMs were selected to have different starting points for interference, with spacers against early, middle, and late expressed targets. In wildtype S. thermophilus, early phage gene transcripts are detectable immediately after inoculation (0–5 min) and their abundance increases steadily until lysis. Middle/late transcripts are first expressed about 15 min after inoculation and quickly increase, becoming about 100× more abundant than early genes by the time lysis begins (Supplementary Fig. S8; [13]). In type III BIMs, these dynamics are altered by interference: early and middle type III BIMs begin reducing all phage transcripts by 5 min post-infection [13]. In contrast, a late type III BIM allows infection (and phage transcript accumulation) to continue until at least 45 min after inoculation. Despite these differences in infection dynamics, a clear and consistent bias toward ORF31 was observed for all four BIMs, including the late type III BIM (Fig. 3A). Together, these results indicate that type III adaptation, but not type II adaptation, is strongly driven to capture spacers from DNA underlying ORF31, and this bias is not shaped by the timing of interference nor absolute transcript abundance.

The use of BIMs also allowed us to test for primed adaptation, wherein the recognition of an existing target results in enhanced acquisition of new spacers around the targeted sequence [61–63]. Although we did not observe any enrichment suggestive of priming in BIMs, such as enrichment of spacers around the target (Fig. 3A), we did note that the proportion of phage-derived spacers changed. As described previously [34], a majority of new unique spacers in our assays arose from the host genome, for both WT and csmKO strains (Supplementary Tables S7 and S8). In all four BIM strains, the proportion of phage-derived spacers was 3–20 times higher than for the WT naïve strain, particularly the type II BIM (Supplementary Tables S7 and S8). However, the effect did not appear to be caused by priming since it was independent of array; for example, the type II BIM had more phage-derived spacers in both its type II array and type III array and likewise all type III BIMs had more phage-derived spacers in both arrays (Supplementary Tables S7 and S8). This increase may instead reflect pre-spacer substrate availability, with BIM cells exposed to phage DNA without lysis.

Type III preference for early transcribed genes is conserved across phages

The global trends found for natural spacers (Fig. 1) suggested that early genes could be targeted during spacer capture for other phages besides 2972. We selected three additional phages known to infect S. thermophilus strain DGCC7710 [64, 65]: another virus in the genus Brussowvirus (D3288), which is predicted to use pac-type genome packaging, and two phages in the genus Moineauvirus (D6037 and D5913) both predicted to use cos-type genome packaging. As with other phages of lactic acid bacteria, the overall genome organization is conserved in these phages, and temporal expression of gene ORFs was predicted by comparison with published time course data for three phages: 2972 [13], DT1 [10], and Sfi19 [22]. Infections were performed in the ∆csm1-6 background to avoid all interference-driven survival bias, and new spacers were sequenced from amplicon libraries. Although the total number of new, unique spacers varied among phages (Supplementary Table S9), we observed a consistent concentration of type III spacers over early genes (Fig. 4A–C), particularly downstream of early cro-like genes (Fig. 4D–F), mirroring profiles seen in phage 2972 (Fig. 3).

Figure 4.

Type III adaptation preferentially targets early genes across phages with both pac and cos type genome packaging. (A–C) Phage genome-wide view of newly acquired spacers. The ∆csm1-6 strain of DGCC7710 was infected with phage D3288 (pac-type genome packaging) (A) or with phages D6037 (B) or D5913 (C) (both with cos-type genome packaging) at an MOI of 100, grown until lysis or overnight, then analyzed for new spacers integrated into the CR2 (type III) or CR1 (type II) arrays. Colored bars at top represent phage gene ORFs, with blue, red, and green bars indicating that a gene was predicted to express middle, late, or early in infection, respectively. Tracks show cumulative pile-up of spacers aligned to the top strand in purple and bottom strand in pink normalized to total spacer read counts. All tracks show merged unique spacers from multiple replicates: (A) D3288 n = 12, (B) D6037 n = 29, (C) D5913 n = 29. (D) Close view of new spacers and DNA read ends aligned near the cos linearization site for D5913. DNA read ends were generated by Oxford Nanopore sequencing of full-length phage linear genomic DNA isolated from mature viral particles. A 25 bp segment from the ends of each long read was extracted and aligned; black tracks show read pile-up of aligned these 25 bp ends. (E–G) Close view of new spacers that align near the predicted early transcriptional region. Data generation and display are as described for panels (A)–(C), but axes are adjusted to show the region between the predicted late expressed lysin gene (in purple) and predicted early expressed genes (in green). SPM = spacers per million.

At the start of infection, phages inject their genome in linear form, and these free DNA ends have been described as a favored substrate for type II CRISPR adaptation [66]. The cos phages D5913 and D6037 are predicted to have defined genome termini at their cos sites (e.g. similar to Sfi19 and Sfi21 [67]). For D5913, we sequenced the DNA ends of linear phage genome from mature virions and confirmed termini consistent with the predicted cos site, but did not observe enrichment of new spacers in the region (Fig. 4G). This may indicate that free DNA ends are not favored for spacer capture in our conditions or that other mechanisms prevent spacer capture, e.g. end protection proteins as observed for phages of S. aureus [68].

Overall, these results indicate a strong and conserved type III preference for genes downstream of early phage promoters. In contrast, promoters driving middle and late gene expression appear not to stimulate adaptation in the same manner, implying that features unique to the earliest transcriptional program preferentially promote type III adaptation.

Type III-A adaptation at the early-gene hotspot is transcription-dependent and driven by the early phage promoter

To recapitulate the early phage hotspot in an experimentally tractable system, we cloned the native early promoter and ORF31 of phage 2972 into plasmid pTRK882 [69]. This plasmid was transformed into the ∆csm1-6 strain, and adaptation in the type II and type III arrays was assessed by PCR. A clear adaptation hotspot formed around the cloned promoter and ORF31 that closely matched the shape of the hotspot observed during phage infection (Fig. 5A, top track). This hotspot was specific to the type III system, with no corresponding enrichment of type II spacers (Fig. 5A, lower track). We quantified unique spacers intersecting ORF31 and found that these spacers accounted for ~20% of all plasmid-derived spacers and ~2% of all unique spacers in the assay (plasmid or host chromosome derived; Fig. 5F and G).

Figure 5.

A transplanted early phage promoter is sufficient to recreate a type III adaption hotspot in a plasmid context. (A–E) ∆csm1-6 strain was transformed with plasmids bearing transplanted phage genes and analyzed for new spacers integrated into the CR2 (type III) or CR1 (type II) arrays. Tracks show cumulative pile-up of spacers aligned to the top strand in purple and bottom strand in pink. All tracks show unique spacers from a single representative replicate; three to six replicates were sequenced per plasmid construct with similar trends. Colored bars at top show the location of plasmid backbone ORFs in yellow, plasmid regulatory features in dark gray (hairpins of rho-independent transcriptional terminators, and single-stranded origins of replication, SSO), transplanted cro ORFs (ORF31) in teal, and mutations in red. A 2-kb scale bar is shown above each track; plasmids are all between 5 and 6 kb in length. (A) Wildtype phage 2972 cro (ORF31), with its native promoter, cloned into plasmid pTRK882. New spacers from the type III (CR2) and type II (CR1) arrays that target this plasmid are shown. (B) Same as panel (A) but the promoter-ORF31 cassette is inverted relative to the plasmid backbone. (C) Promoter disruption construct: the predicted −10 element in was changed from TAAAAT to GCCCCG to disrupt transcription. (D) Translation-disruption construct: the ORF31 start codon (ATG) was mutated to a TAA stop codon to prevent translation. (E) Spacer-hotspot relocation construct: an adaptation-inert phage gene (ORF9; capsid protein) was inserted between the cro native promoter and ORF31. SPM = spacers per million. (F) Quantification of spacers overlapping cro (ORF31) out of all unique plasmid-derived spacers. (G) Quantification of spacers overlapping cro (ORF31) out of all unique spacers detected in the assay, including genome-targeting spacers.

With the hotspot reconstituted, we generated a series of mutations to define features required to stimulate type III adaptation. First, we inverted the promoter and ORF31 cassette and confirmed that the profile was not dependent on its orientation within the plasmid backbone (Fig. 5B). We next disrupted transcription by mutating the predicted −10 TATA box element (TAAAAT to CGGGGC). This abolished the hotspot and reduced spacer uptake across ORF31 toward plasmid background levels (Fig. 5C). In contrast, interrupting ORF31 translation by mutating the start codon (ATG to TAA Stop codon) did not eliminate spacer enrichment; instead enrichment was enhanced and >50% of all unique plasmid-targeting spacers overlapped ORF31, with many more aligning downstream, extending until the plasmid-encoded rho-independent transcription terminator (Fig. 5D and F). This indicated that translation is not required for hotspot formation, and may inhibit spacer uptake. The ratio of plasmid-derived to genome-derived spacers was also enhanced for this stop codon construct (Supplementary Table S10), and we found that the large peak of spacers over the start codon mutant ORF31 accounted for ~22% of all unique spacers in the assay (Fig. 5G). Because ORF31 is small (210 bp) and plasmid copy number is predicted to be up to 85 copies per cell [70], this indicates that a feature constituting <1% of sequence space was the site of >20% of spacer capture events.

To test whether the downstream coding sequence itself was important, we inserted an “adaptation inert” gene (ORF9; capsid protein) between the phage promoter and ORF31. We considered this ORF “adaptation inert” as type III spacer enrichment was not observed over ORF9 during phage infection (Fig. 3A). However, in the plasmid context, a pronounced peak of type III spacers appeared near the beginning of ORF9, within the same ∼10–50 bp window previously observed for ORF31 (Fig. 5E). Together, these results show that high level type III-A adaptation at this site requires transcription from the early gene promoter and may be influenced by post or co-transcriptional processes like translation.

Natural type III spacer acquisition hotspots also map to early phage promoters

Following our experimental observation that the cro-like gene promoter, rather than the gene itself, was responsible for driving high levels of adaptation, we returned to the natural spacer dataset and focused on the genes and promoters in the early gene region. Interestingly, we noted that all phage genomes encoded one or two cro-like genes downstream from lysin, implying some conserved function in lytic phages.

In a subset of phages targeted by S. thermophilus spacers, a cro-like gene was the first ORF downstream from lysin (n = 108) (Fig. 6A-i), while for others (n = 96) a set of up to 14 ORFs was positioned between lysin and the cro-like gene (Fig. 6A-ii). These ORFs were generally less conserved than the rest of the genome, consistent with the designation of this area as a “variable region” in previous reports [11, 17]. The variable regions included anti-CRISPR genes known to inhibit type I, II, and III systems [17, 18, 71] and other genes of unknown function. All of the cro-like genes and all variable regions were preceded by a promoter with −10 and −35 regions perfectly matching the σ^A consensus sequence of S. thermophilus (Fig. 6B). This promoter drives transcription of host “housekeeping” genes like 16S and 23S rRNA, which are also well expressed (Supplementary Fig. S8). Long AT-rich Up elements were also apparent, known to promote expression through interaction with the C-terminal domain of RpoA (Fig. 6B) [72]. Even promoters internal to variable regions, such as that preceding AcrIIIA2 [18] (Fig. 6B, promoter “d”) followed this same “housekeeping” promoter and Up element structure, suggesting similar expression in the phage life cycle. To look at spacer trends in the “cro-first” versus “variable-first” phages, we parsed genomes based on whether their early gene layout matched that of Fig. 6A-i or 6A-ii and changed their genome coordinates such that the start codon of the first-encoded cro-like gene was at position 15 kb. Mapping natural type III spacers to these new meta-phage categories revealed spacer enrichment was not tied to cro-like genes but rather appeared to coincide with whichever genes were most proximal to lysin (Fig. 6C). From this finding, we believe the canonical promoters upstream from cro-like genes or variable region genes create favorable conditions for type III-A spacer uptake in S. thermophilus.

Figure 6.

Target biases in S. thermophilus natural spacer dataset. (A) Overview of gene organization downstream from the lysin gene ORF in S. thermophilus lytic phages without (i) or with (ii) a variable region. In ∼50% of phages, cro is the first encoded early gene with its own promoter (a), with around half of these encoding a second cro with an independent promoter (b). The remaining ∼50% of phages encode a variable region with an upstream promoter (c), and often internal promoters (d). (B) Alignment of representative cro and variable region-promoters, with promoter elements indicated. Early gene promoters exhibit canonical “housekeeping-like” architecture with consensus σA −35 and −10 elements as well as an A/T-rich Up element. (C) Alignment of all phage genomes with the start codon of the first cro positioned at 15 kb reveals that CR2 (type III-A) adaptation enrichment extends into which ever region lies immediately downstream of lysin; when variable regions precede cro, enrichment extends upstream of cro. This enrichment is prominent for CR2 (type III-A) and absent from CR1 (type II-A) profiles.

Discussion

Despite conservation in adaptation machinery (Cas1 and Cas2), recent work has shown that CRISPR–Cas subtypes differ in their dependency on PAMs and on the spacers they acquire, with different preferences for broad phage or plasmid features and regions [32–34, 39]. The dependence of spacer acquisition on PAM identity in DNA-targeting systems indicates that spacer selection by adaptation complexes can co-evolve with spacer (i.e. crRNA) utilization by the interference complex. The regional spacer uptake preferences hint that co-evolution may also guide adaptation toward spacer substrates likely to support the system’s mechanism of interference. For DNA-recognizing systems, capture of PAM-adjacent DNA fragments by Cas1-2 should provide a set of competent spacers against phage genomes or plasmids. However, for most RNA-recognizing systems, the relationship between adaptation and interference appears more complex. DNA spacers must arise from transcribed regions if the crRNA is to bind target RNA. Furthermore, given that transcription of phage genes is temporally staggered over infection, the transcribed regions must be targeted early enough that phage anti-CRISPRs or other phage-encoded factors have not yet irreversibly compromised the host cell. In this work, we examined spacer acquisition by the RNA-targeting, type III-A system of S. thermophilus during active viral infection and compared it directly to the co-occurring DNA targeting, type II systems. We showed that type III-A adaptation, but not type II, was consistently biased toward early-expressed phage genes in a transcription-dependent manner. Although previous studies have noted a similar early-gene enrichment, those observations were attributed to survivorship bias [32]. Here, we show that the bias is an inherent feature of type III-A adaptation and requires transcription (Fig. 5).

Why target early genes?

Targeting early-expressed phage genes should provide an RNA targeting immune system maximum time to recognize and respond to infection. In the case of late-targeting spacers, phage infection may be all-but complete before the corresponding crRNA-complementary target mRNA is transcribed and recognized by type III effectors [13]. Additionally, early phage gene products frequently play decisive roles in host takeover and may need to be neutralized immediately [73, 74]. A clear example is the recently identified AcrIIIA2, an anti-CRISPR protein encoded in the variable region of several S. thermophilus phage genomes with a canonical σ^A promoter (promoter “d” in Fig. 6). This Acr forms a complex with host enolase and prevents target recognition by type III-A crRNPs, thereby abolishing immunity [18]. Early expression of this Acr could only be countered by spacers targeting transcripts expressed equally early. Therefore, targeting of early-expressed genes, including those encoded in phage variable regions [18], may have proven a highly beneficial “strategy” over evolutionary time scales.

Mechanisms favoring early-gene adaptation

Our results show that type III-A adaptation occurs preferentially at sites downstream of phage early gene promoters, and our plasmid-based experiments indicated that transcription was required for this preference. Previous work from our lab has also shown adaptation at highly transcribed chromosomal genes (rRNA and tRNA) and sites of RNAP stalling (such as riboswitches and predicted RNA/DNA secondary structures). Work in a heterologous system utilizing Staphylococcus species also showed type III-A spacer capture concentrated around rRNA and tRNA gene clusters [39], indicating that the preferences we observed are likely not unique to S. thermophilus. However, the mechanism underlying the association between type III adaptation and transcription remains unclear; below we propose several hypotheses (Fig. 7).

Figure 7.

Conceptual model of transcription-linked events that may promote type III spacer acquisition. The observed association between type III spacer capture and transcription may arise from direct or indirect coupling of Cas1–Cas2 activity to host RNA polymerase (RNAP) or RNAP-associated states, including RNAP occupancy, RNAP stalling, transcription-induced negative supercoiling, R-loops and RNA:DNA hybrids, exposure to single-stranded DNA, translation-coupled effects, and transcription–replication collisions (see text for more detailed explanation).

One parsimonious model is that the type III Cas1–Cas2 adaptation complex directly interacts with the RNAP holoenzyme (or some iteration thereof, e.g. through an associated sigma factor or elongation factor). This would result in general recruitment of adaptation machinery to transcribed regions and would bias spacer selection toward areas with the highest RNAP occupancy (e.g. sites of RNAP stalling and highly expressed loci like chromosomal rRNA and tRNA genes). We note that not all regions with high apparent RNA expression are heavily sampled by adaptation: by late infection, transcripts from the phage middle/late promoter were far more abundant than early transcripts, but spacers were sporadic over this region (Supplementary Fig. S8 and Fig. 3). However, a different RNAP holoenzyme is likely active at middle/late phage promoters, one which incorporates phage-encoded transcription and/or sigma factors [75, 76] and this may underlie adaptation differences at the early versus late phage promoter.

Alternatively, adaptation may be indirectly linked to transcription. For example, high levels of transcription can result in R-loop formation as nascent transcripts invade negatively supercoiled DNA in the wake of RNAP. These R-loops comprise stable RNA:DNA hybrids with a free ssDNA non-template strand [77, 78]. There is precedence for the integration of RNA:DNA hybrids into CRISPR arrays in the reverse transcriptase (RT)–Cas1 fused adaptation systems [9, 79], and even a classical Cas1–Cas2 adaptation complex harbors endonuclease activities targeting both DNA (Cas1 [80]) and RNA (Cas2 [81, 82]). Thus, some of the biochemical requirements for using RNA:DNA hybrids as spacers could be fulfilled by the Cas1–Cas2 complex. However, no RT domains are predicted for the S. thermophilus type III-A Cas1–Cas2 complex, and at this time it is not clear whether an RNA:DNA hybrid can be integrated into the CRISPR array. In addition, the displaced ssDNA strand of R-loops can form secondary structures with spans of paired bases (hairpin stems) bounded by unpaired ends [83]. These partially double stranded DNAs are reminiscent of the 3′ overhangs generated during pre-spacer processing [84] and if cleaved correctly, would provide ideal spacer substrates [85]. We have previously suggested that these structures may be direct adaptation substrates for type III systems [34], and our data presented here leave open this hypothesis.

High levels of transcription can also result in genomic instability from transcription–replication collisions [86, 87] and from R-loop formation [88, 89]. The predicted origin of replication of phage 2972 (and most Streptococcus phages) [11] is located within the early genes, downstream of the replication module. Little is known about replication of Streptococcus phages, however the similarly organized Bacillus phage SPP1 (and most studied Lambdoid phages) encodes a replication origin (oriL), which drives several rounds of theta replication before switching to rolling circle replication (RCR) [90, 91]. RCR initiates from a DNA break caused by a stalled replication fork at a second, AT-rich origin (oriR) [91]. If 2972 and related phages replicate via a similar mechanism, theta replication could cause head-on collisions with early gene transcription (later resolved by co-directional RCR) and/or the AT-rich region upstream of early genes could function as oriR sites, with RCR-initiating DNA breaks. In both scenarios, the DNA breaks could provide free DNA ends for spacer generation similar to that observed for other CRISPR–Cas systems [39, 92, 93]. However, these free DNA ends should be available to both type II and type III adaptation, yet we only observed an early-gene bias for type III spacers, suggesting adaptation is more specific than free DNA ends.

Finally, it should be noted that these hypotheses are not mutually exclusive and multiple modes of adaptation may exist. For example, association of Cas1-2 with RNAP increases its exposure to transcription-dependent R-loops; or the more canonical Cas1-2 function of sampling free DNA ends could be a retained ancestral state of the type III system, with extra functionalities evolving to fine-tune adaptation toward transcribed substrates.

Translation may suppress adaptation at transcribed regions

Interestingly, preventing translation of the cro-like ORF31 amplified the spacer bias in our plasmid experiment (Fig. 5). We speculate that this unexpected effect may be related to our previous observation that spacers accumulated around tRNAs and rRNAs, which are transcribed but not translated [34]. Likewise, for most riboswitches in the attenuated state, transcription initiates but does not proceed to translation due either to termination or occlusion of translation initiation [94]. It is not clear how the ribosome would reduce spacer capture from transcribed DNA, but we can imagine several possibilities related to the models described above and diagrammed in Fig. 7. Coupling between transcription and translation has been described for a range of prokaryotes and ribosome engagement of the nascent RNA reduces pausing of RNAP and aids transcription processivity [95–97]. Without the ribosome, more frequent RNAP pausing and backtracking can promote replisome-RNAP collisions, leading to dsDNA breaks [98–100]. Additionally, if the adaptation machinery is directly recruited to the RNAP holoenzyme, slower progress of RNAP would mean longer residence time and more opportunity for spacer capture. Alternatively, when nascent RNA is not engaged with a ribosome, it may be more free to form R-loops [101, 102], yielding more RNA:DNA hybrids and more non-template ssDNA. Notably, when the cro gene with native promoter was cloned into the host chromosome instead of a plasmid, no spacer hotspot was observed (data not shown). This suggests that transcription-activated adaptation may depend on the DNA context, for example topology, DNA binding proteins, or replication dynamics. Detailed follow-up experiments are needed to determine the mechanism that links spacer uptake with transcription.

Unique risks and advantages of type III immunity

We show that type III BIMs confer immunity at high phage titers where type II BIMs often fail, likely due to phage evading immunity through mutation. Whereas single PAM/seed SNPs allow escape from type II immunity [52, 59], type III escape requires large deletions, which are rarely tolerated (Supplementary Tables S5 and S6) [32, 60]. Consistent with this, it was reported that a single type III spacer can cause extinction of an infecting phage [60], while multiple type II spacers are required for similar phage clearance [103]. Despite this potent immunity, S. thermophilus type III arrays are significantly shorter than type I/II arrays (Supplementary Fig. S2), implying that adaptation events are rare and/or lost due to counter-selection. This may be a result of greater promiscuity for type III systems: (i) there are almost no PFS constraints compared to strict PAM requirements (Supplementary Fig. S4); (ii) the possibility for G:U interactions in RNA:RNA pairing allows for more base pairing partners in RNA-targeting versus DNA-targeting systems; (iii) every sixth nucleotide of the crRNA is dispensable for pairing in Class 1 systems, allowing broader sequence acceptance [57, 58]; and (iv) mismatches and short indels that block type II immunity are well tolerated by type III complexes [104]. Collectively, these features increase the probability of off-target autoimmunity, even for phage-derived spacers. Thus, type III adaptation likely reflects a balance between the benefits of phage defense and the costs of self-targeting. These forces could explain why type III systems are widespread in S. thermophilus, but tend to have short arrays and more closely guided spacer capture.

In summary, our work demonstrates that type III-A CRISPR adaptation in S. thermophilus is fine-tuned toward early-expressed phage genes in a transcription-dependent manner. This bias appears to be a distinct feature of type III systems, likely shaped by co-evolution between adaptation and interference to maximize immunity. By preferentially sampling early transcripts, type III-A systems can counteract phage takeover mechanisms, including early-expressed anti-CRISPR proteins and host-manipulating factors, and confer rapid, robust immunity even under high phage load. However, this fine tuning may come at the cost of increased autoimmunity as early phage promoters closely align with host promoters. This, combined with an increased potential for off-target recognition of host transcripts, may explain why natural S. thermophilus type III arrays harbor fewer spacers, even when the type III system confers a more robust, escape-resistant immunity compared to co-occurring type II systems.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

This work was supported by the National Institutes of Health (R35GM118160 to M.P.T., R35GM118140 to B.R.G.)

Data availability

Sequence data were deposited in the NCBI Sequence Read Archive under the BioProject ID PRJNA1379557.