Cooperation between Different CRISPR-Cas Types Enables Adaptation in an RNA-Targeting System

CRISPR-Cas systems are immune systems that protect bacteria and archaea against their viruses, bacteriophages. Immunity is achieved through the acquisition of short DNA fragments from the viral invader’s genome.

phases, but the evidence for in trans-mediated spacer acquisition between CRISPR-Cas systems is still indirect, awaiting experimental verification.
The recently discovered type VI CRISPR-Cas systems (11,37,39) often lack Cas1/2 (4) and thus have been proposed to acquire spacers in trans (37,(39)(40)(41). However, as type VI loci exclusively target RNA, obtaining spacers from double-stranded DNA (dsDNA) (the modus operandi for most acquisition complexes) is not optimal, as only half of the potential spacers may functionally target mRNAs. Adaptation in type VI systems has yet to be experimentally demonstrated, and it is unknown if new spacers are acquired through the in trans adaptation model and, if so, how this affects their capacity to target RNA.
The genome of the fish pathogen Flavobacterium columnare has subtype II-C and VI-B CRISPR-Cas loci (Fig. 1). The II-C locus contains the adaptation proteins Cas1 and Cas2 and the endonuclease Cas9 that cleaves dsDNA (although some Cas9 variants also cleave single-stranded RNA [ssRNA] [44,45]). The II-C locus of F. columnare also contains genes encoding a predicted ParDE type II toxin-antitoxin system between cas9 and the repeat-spacer array, which may contribute to the immune response (46,47). II-C systems differ from other CRISPR-Cas systems by the expression of their array from repeat-encoded promoters instead of the leader sequence (44,48). Subtype VI-B of F. columnare encodes only Cas13b and does not contain adaptation genes, nor does it contain csx27 or csx28 that is often associated with type VI-B systems (4). Cas13 exclusively cleaves RNA and is guided to its target by its bound crRNA, similar to other interference complexes. However, once activated by this primary target, Cas13 becomes a promiscuous RNase that also cleaves noncomplementary phage and host transcripts, potentially leading to cellular dormancy or death (11,39,49).
Previously, we showed that natural isolates of F. columnare vary in their phage-targeting spacer contents in both loci, driving the evolution of sympatric phages (34). Only the phage mRNA strand was targeted by VI-B spacers (n = 15), while this was the case for only half of the II-C protospacers (34). While the accumulation of mRNA-targeting spacers in the VI-B locus was clear, it is unknown if this was due to a selective spacer acquisition process or from positive selection due to successful interference. To further characterize spacer acquisition in these loci, we cocultured F. columnare with a virulent dsDNA phage under laboratory conditions (Fig. 2). Our results revealed details of subtype VI-B and II-C adaptation in a native host as well as the first experimental evidence of in trans adaptation between CRISPR-Cas types despite differences in leader and repeat sequences.

RESULTS
Phage-induced acquisition in both loci. To study the adaptation phase of subtype II-C and VI-B loci in a controlled environment, we cultured F. columnare strain B185 (50,51) (GenBank accession no. NZ_CP010992.1) with its virulent dsDNA phage FCL-2 (accession no. NC_027125.1) (51) in liquid medium (bacterium plus phage [B1P] treatment) (Fig. 2). We also included treatments with UV-irradiated phage FCL-2 (bacteria plus UV-phage [B1UVP]) and a mixture of irradiated and nonirradiated phages (B1P1UVP). The addition of defective UV-treated phages has been shown to increase spacer acquisition in subtype II-A (52). The native II-C CRISPR array of F. columnare B185 contains four preexisting spacers targeting the genome of phage FCL-2 and four VI-B spacers with imperfect matches to FCL-2 (see Table S3A and B in the supplemental material). After a week of coculturing, we plated the liquid cultures to screen for expanded CRISPR arrays in individual colonies (total of 272 colonies screened) (Table S1). Only bacteria that were exposed to the phage mixture (B1P1UVP) had acquired new spacers (14 out of 85 colonies). We did not observe spacer acquisition in bacteria that were not exposed to phage (B) or that were exposed to either phage treatment alone (B1P or B1UVP) (Tables S1 and S2).
Next, we focused on CRISPR adaptation at the bacterial population level by amplifying by PCR the variable ends of both arrays directly from the same liquid cultures. In contrast to individual colony screening, we observed spacer acquisition in both the phage mixture treatment (B1P1UVP) (data not shown) and the phage-only treatment (B1P) (Fig. S1). The difference between the individual-colony screening and population-level screening likely stems from the higher sensitivity of the latter. All downstream analyses were done on the B1P samples, as this treatment is closer to a natural scenario (UV irradiation lowers phage infectivity by 2 orders of magnitude). Within each replicate of the B1P treatment (identified as replicates a to e), the relative rates FIG 2 Overview of the acquisition experiment. Spacer acquisition in F. columnare strain B185 (B) was investigated using a two-step growth method in diluted and undiluted media. Bacterium plus phage FCL-2 (B1P) treatment was accompanied by a phage mixture containing UV-treated phages (B1P1UVP) and UV-treated phages alone (B1UVP), all done in 5 replicates (a to e). Spacers were examined in individual colonies from all three phage-based treatments as well as by deep sequencing of the CRISPR arrays in the B1P treatment. "Failed amplification" refers to samples where the DNA yield from expanded arrays was insufficient for sequencing. of spacer acquisition were similar between the two arrays (Fig. S1). To identify the diversity and origin of the new CRISPR spacers, we deep sequenced the population-scale variable ends of the CRISPR arrays. We obtained data on new II-C spacers in replicates b and e and on VI-B spacers in replicates b, d, and e (other replicates did not allow for deep sequencing due to low DNA yields). PCR amplicons in the wild-type control cultures were also sequenced to ensure that the arrays had remained unchanged.
The lengths of new spacers varied between 28 and 32 bp, with the majority at 30 bp in both loci (Fig. 3A). We divided the new spacers into unique and absolute sets. In the unique spacer data set, each spacer sequence was counted only once, whereas the absolute set allowed for repeated observations of the same spacer. Most unique II-C spacers targeted the phage genome, with a minority targeting the bacterial genome (Fig. 3B). New VI-B spacers targeted both genomes relatively evenly, although the replicates had some differences (Fig. 3B). All differences were accentuated when the spacer count was viewed as absolute (Fig. 3C). The distinct targeting preferences between the loci are likely explained by the negative selection of autoimmunity resulting from Cas9-mediated self-targeting (e.g., see reference 53). The lowered fitness due to self- (E) F. columnare conjugants on antibiotic agar plates. Cells were conjugated with either a plasmid containing a protospacer matching the most recent II-C spacer or a plasmid containing a nonmatching control sequence. All three replicate plates per treatment are shown, with one replicate enlarged. (F) Proportions of simulated and observed VI-B spacers that are identical to the II-C spacer pool. (G) PAM sequences next to II-C and VI-B phage targets (sequences from the nontarget strand; example from replicate e) (for all replicates, see Fig. S2A in the supplemental material). CRISPR-Cas Adaptation in an RNA-Targeting System ® targeting in II-C is further implied by the proportional reduction of self-targeting spacers in the absolute data set (Fig. 3C) compared to the unique set (Fig. 3B). The abundance of self-targeting VI-B spacers was surprising given that such spacers could induce dormancy and be negatively selected (49). The observed pattern therefore indicates that, at least during this 7-day experiment, self-targeting VI-B spacers were not necessarily harmful.
To confirm that II-C spacers yield functional interference, we performed a plasmid interference assay. We modified plasmid pAS43 (54) to include a protospacer that matches the leader-adjacent II-C spacer in the conjugation-capable F. columnare strain B245. As a control plasmid, we replaced the matching protospacer sequence with a nonmatching sequence. Using Escherichia coli as the donor, the efficiency of conjugation was roughly three times lower in the matching protospacer treatment than in the control treatment under antibiotic selection (P = 0.0016; t 4 = 7.62 [by an unpaired t test]) (Fig. 3D). Surviving colonies with the matching protospacer plasmid also had a reduced colony size (Fig. 3E). The II-C system in F. columnare seems to grant partial immunity to mobile genetic elements, leading to slower bacterial growth, at least when the CRISPR target is associated with positive selection (here, an antibiotic resistanceconferring plasmid). As speculated to occur with Listeria monocytogenes, small colonies may result from low Cas9 expression, allowing the persistence of a small number of plasmids in the cell (55).
To calculate the proportions of identical new spacers between the loci, we pooled unique phage-targeting spacers from both loci from all replicates. Remarkably, 58% of VI-B spacers were also found in the larger II-C spacer pool, suggesting that the loci share a predisposition toward a set of spacers. For statistical comparison, we simulated spacer acquisition across the phage genome by sampling random positions either freely or next to predicted PAM sites (59-NNNNNTAAA-39) (n = 1,528) (described in detail below). Both simulations were run 1,000 times, with each run sampling 430 spacer positions (the size of the VI-B spacer pool). We calculated the proportions of shared spacers between each run and the II-C spacers and found that on average, 6.3% (standard deviation [SD], 1.2%) of the random spacers and 29.6% (SD, 1.85%) of the PAM-adjacent spacers had a match in the subtype II-C spacer set (Fig. 3F). The similarity between the observed VI-B and II-C sets of spacers is significantly higher than that between either simulated pool and II-C spacers (P , 10 210 by a one-tailed test), which suggests that new spacers are sampled from a limited protospacer pool that is common to both loci.
Shared acquisition motifs suggest in trans adaptation. To further characterize the protospacers, we investigated PAM sequences for both loci. We used the guide-oriented approach that considers PAMs to be on the strand that matches a particular spacer's crRNA (56), i.e., on the nontarget strand. It is necessary to establish the direction of crRNA transcription before evaluating the PAM sequence as the transcription direction determines the resulting crRNA strandedness. Previous studies have shown that unlike other CRISPR-Cas systems, subtype II-C arrays are expressed from within the array toward the variable end using repeat-encoded promoter sequences (59-TAAAT-39) (44,48). The F. columnare II-C repeats lack this sequence but have a 59-TTG-39 motif 33 bp upstream of each spacer. As the sequence TTG is an established 233 promoter in Flavobacterium hibernum (57), this putative promoter may also drive II-C array transcription in F. columnare. This direction of transcription is also supported by our transcriptome data showing within-array transcripts that proceed toward the leader (albeit with only eight reads mapping on the array [ Fig. S3A]). The subtype VI-B array, on the other hand, shows an abundance of transcripts occurring in the leader-to-array direction (Fig. S3B), as typically observed for CRISPR-Cas systems (11,39), and has no putative repeat-encoded promoters.
In subtype II-C, the previously reported PAM (34,58) NNNNNTAAA was observed downstream of most (;63%) phage-targeting spacers ( Fig. 3G and Fig. S2A). In the few self-targeting subtype II-C spacers, the canonical PAM either was almost always absent or had extended or shortened N regions (Fig. S2B). This deprecation of the PAMs further suggests negative selection against functional self-targeting II-C spacers and indicates that both the nucleotide motif 59-TAAA-39 and the length of the N region in the PAM play a role in subtype II-C interference in F. columnare (Fig. S2B). The protospacers targeted by subtype VI-B had the same PAM sequence in the reverse complement (TTTANNNNN) and located upstream (Fig. 3G), with slightly less conservation (;50%) than the subtype II-C protospacers (Fig. S2B). Self-targeting VI-B spacers did not have deprecated PAMs similar to the self-targeting II-C spacers, suggesting that PAMs do not affect VI-B functionality.
When using the guide-oriented approach, the PAMs of II-C and VI-B targets are complementary to each other and on opposite sides of the protospacer. For both loci, the PAM-adjacent ends of the spacers are oriented toward the leader. Therefore, while the target interference motif (TIM) of the loci is seemingly different, the spacer acquisition motif (SAM) is shared by the loci. The term TIM is not strictly applicable to the studied RNA-targeting type VI systems, and their interference efficiency has instead been shown to be affected by short protospacer-flanking sites (PFSs) (11,39). Here, conserved PFSs were not discovered in the VI-B targets. Overall, these results suggest that the arrays likely share spacer acquisition machinery, as SAMs are generally hallmarks of specific acquisition complexes (12,14).
Spacer target distributions are nonuniform and show possible II-C priming. We investigated which regions of the genomes are targeted by II-C and VI-B spacers. To do this, we examined both unique and absolute spacer counts separately. Despite the relatively even distribution of predicted PAM sequences (59-NNNNNTAAA-39) across the phage genome (Fig. 4), spacer targets were not uniformly spread. On the linear phage CRISPR-Cas Adaptation in an RNA-Targeting System ® genome, spacer targets were concentrated in the opposite end of the genome relative to the morphogenesis genes (59) and are therefore likely targeting the end of the genome that first enters the cell (Fig. 4). Furthermore, the two strands of the dsDNA phage genome were targeted unevenly, with stronger differences at the end of the genome (downstream of ;30 kbp), especially in the absolute spacer set. These distributions are not explained by uneven PAM distributions on the phage genome ( Fig. 4) or between the strands (the proportions of the 1,672 instances of the PAM sequence 59-TAAA-39 on the coding and template strands are 50.7% and 49.3%, respectively).
Intriguingly, the nonunique II-C spacers peak around a preexisting, fully matching II-C spacer on the phage genome ( Fig. 4A) (around 40 kb), a trend not visible in the unique spacer set (Fig. 4B). The congregation of new spacers around a preexisting one may result from CRISPR-Cas type II primed adaptation (24), where spacers are acquired from the free dsDNA ends produced by Cas9 cleavage (25). Furthermore, the distribution of new spacers around this protospacer resembles an asymmetric pattern that is characteristic of primed adaptation in subtype I-F CRISPR-Cas systems (21,22). It is unclear why the three additional phage-targeting II-C spacers (Table S3A) with canonical PAM sequences do not give rise to similar priming-like patterns (Fig. 4A). It is also unclear why the VI-B spacers do not form similar patterns, as the production of free dsDNA ends by Cas9 cleavage is expected to provide spacer material with no intrinsic bias toward any specific array.
Another possibility, not mutually exclusive with priming, is that targeting the terminal end of the linear phage genome that first enters the cell is predominant, as shown with the Staphylococcus pyogenes subtype II-A system (25). In phage FCL-2, the localization of the morphogenesis genes at the opposing end of the linear genome (59) suggests that the preferred targeted end is indeed the first to enter the cell. Natural samples of F. columnare also show consistent targeting of this end in other (genetically very similar) phage genomes (34). It is possible that primed adaptation and higher exposure to spacer sampling at this end of the phage genome cooccur, perhaps explaining why the three additional II-C spacers do not attract priming. In addition, if priming played no role in F. columnare adaptation, we would expect spacers to cluster at the very end of the genome instead of the currently observed priming-like peak approximately 10 kb from the end (Fig. 4A).
On the bacterial genome, VI-B spacers and the few II-C spacers congregated on two asymmetric hot spots (Fig. 4). One was centered on the subtype II-C CRISPR-Cas locus, while the other localized on the predicted origin of replication (oriC). Acquisition peaks on oriC were previously reported in self-targeting E. coli spacers, resulting from the frequent occurrence of double-stranded breaks in this region and the subsequent actions of the RecBCD dsDNA repair complex (60). The resulting prespacer substrates are overrepresented around oriC because this region is the first to be replicated and becomes enriched in a replicating cell. Acquisition peaks on CRISPR arrays were also documented previously (22,60) and were speculated to arise from frequent nicking of DNA during spacer acquisition, as nicking of the chromosome would also stall the replication fork, leading to increased spacer acquisition (60). Why spacers do not congregate on the F. columnare VI-B locus is unclear.
Most spacers are not targeting mRNAs. To investigate if the new spacers reflect the different target requirements of the loci (DNA versus RNA), we examined the ability of each spacer crRNA to bind mRNA by assessing their complementarity to predicted open reading frames (ORFs) from both target genomes. Spacers from all replicates within a locus were also pooled for an overall estimate. For statistical context, we calculated the probabilities of mRNA-targeting proportions in both loci with a binomial distribution that uses sample sizes matching those of the pooled spacers and assumed an equal probability of targeting each strand. Deviation from these expected distributions signals possible bias toward or away from mRNA targeting.
In the VI-B system, when considering unique spacers, the proportion of phage mRNA-targeting VI-B spacers was 0.46 (P = 0.074 by a one-tailed test), suggesting a weak or nonexistent bias toward non-mRNA-targeting spacers. VI-B mRNA targeting on the bacterial genome was rarer and significantly below expectation at 0.359 (P , 10 29 by a one-tailed test) (Fig. 5A). When considering the total spacer counts, the proportion of phage mRNA-targeting VI-B spacers rose to 0.53 (above expectation) (P = 0.0059 by a one-tailed test) (Fig. 5B), suggesting possible enrichment of mRNA-targeting spacers. On the bacterial genome, the mRNA-targeting proportions of absolute spacers were significantly below expectation at 0.348 (P = 10 210 by a one-tailed test) (Fig. 5B).
Subtype II-C mRNA-targeting proportions in the unique spacer set were the same for both target genomes (0.37 and 0.37). Given their vastly different sample sizes, the proportion of mRNA-targeting spacers on the phage genome was significantly below expectation (P , 10 210 by a one-tailed test), while self-targets were not clearly biased (P = 0.052 by a one-tailed test) (Fig. 5A). In the total spacer counts, the divergence of self-targeting II-C proportions may seem radical (Fig. 5B) but stem from the very small sample sizes, where a slight overrepresentation of a spacer may disproportionally skew the result by chance. Furthermore, the pooled II-C self-targeting spacers do not statistically deviate from expectation (P = 0.124 by a one-tailed test). The proportions of absolute phage-targeting II-C spacers were slightly lower than in unique counts, at 0.319, and deviated strongly from expectation (P , 10 210 ) (Fig. 5B).
The minimal phage mRNA targeting in subtype II-C spacers was unexpected, as Cas9 functionality is indifferent to which strand is being targeted. One potential explanation would be a strongly uneven distribution of PAM sequences on the coding versus template strands, but this is not the case in either genome: the proportions of PAM sequences (59-TAAA-39) in annotated coding sequences from B185 are 51.2%/ 48.8% (coding/template) (n = 81,490), and the proportions are 51.6%/48.4% for FCL-2 (n = 1,528). Another possibility is that competition between RNA and DNA targets for Cas9 association would impact subtype II-C immunity negatively.
The lack of new VI-B self-mRNA-targeting spacers is expected due to their possible dormancy-inducing effects (49). This is also in agreement with the lack of preexisting self-mRNA-targeting VI-B spacers in F. columnare B185 (n = 7) ( Fig. 4 and Table S3B). However, this counterselection is much weaker than that against self-targeting II-C spacers ( Fig. 3B and C), which argues for a lesser effect of VI-B autoimmunity in this experimental setting. Saturation of phage mRNA-targeting VI-B spacers was not observed here, which is likely due to the short time span of the experiment but may also be affected by the preexisting, fully matching II-C spacers (Table S3A). If the preexisting spacers already provide some level of CRISPR-Cas-based immunity, as suggested by our plasmid interference assay, the need to acquire new ones is decreased, and their selective advantage may be weaker.
Overall, the reasons for the unexpectedly low levels of mRNA-targeting spacers in both CRISPR loci remain speculative. In this analysis, we relied on predicted ORFs and do not take into account possible antisense transcription or intergenic transcripts; the number of RNA-targeting spacers may therefore be underestimated. In contrast, in our previous study on natural F. columnare isolates, there was a saturation of phage mRNA-targeting VI-B spacers (n = 19; P = 1.9 Â 10 26 by a one-tailed binomial test) (including the preexisting four spacers in the wild-type array of B185 [Table S3B]). Our new results suggest that this was likely the result of positive selection toward mRNAtargeting spacers. Interference by the VI-B system in nature is also supported by the negative selection of targets: none of the preexisting VI-B spacers in strain B185 are complete matches with their protospacers on the phage genome, and VI-B protospacers accumulate mutations over time (34). Future studies are needed to assess how RNA-based VI-B immunity evolves and what constitutes functional targets for VI-B interference in native systems.
The VI-B array uses II-C Cas1 in trans. Identical spacer acquisition efficiencies, similar protospacer localization patterns, and shared PAM sequences supported the hypothesis that the F. columnare subtype II-C and VI-B CRISPR-Cas loci share spacer acquisition machinery. To verify this, we deleted cas1 from the subtype II-C locus from F. columnare strain B245. Using the resulting B245 Dcas1 strain, we performed a prolonged adaptation experiment with 10 phage-exposed (phage V156) and 2 bacteriumonly replicates. As a control, we used a "reversion wild-type" (rev-wt) strain that resulted from an alternative outcome of the mutation process and contains an intact cas1 (see Materials and Methods). We did not observe spacer acquisition in either CRISPR array in any of the 12 Dcas1 cultures (Fig. S4), all of which survived until the end of the experiment. Of all 12 rev-wt cultures, only 2 phage-exposed replicates survived, both showing spacer acquisition in their II-C or VI-B arrays (Fig. S4). These results demonstrate the dependence of the VI-B locus on the II-C acquisition machinery.
The inability of B245 rev-wt to survive in the presence of phages is in contrast with the previous experiment performed with strain B185 and phage FCL-2. This difference may partly stem from the experimental setups, as the bacterial strains B245 and B185 were cultivated in different media and for 3 weeks and 1 week, respectively (see Materials and Methods). Also, strain B245 has four preexisting, fully matching phage mRNA-targeting spacers in the variable end of the VI-B arrays (Table S3C and D), whereas B185 has only incomplete VI-B spacer matches with its target, all located in the array's conserved end. It is therefore possible that the B245 phage-targeting VI-B spacers induce dormancy during the CRISPR-Cas response (49), whereas B185's lack of fully matching preexisting spacers does not. If this is the case, the survival of B245 Dcas1 is surprising as the inability to acquire new spacers should not abrogate interference using preexisting ones. These strains also belong to different genetic groups, and thus, their differences are not limited to spacer content. Further experiments to explore the effect of spacer diversity and the role of different cas genes and other host defenses in F. columnare are required.
Comparison of II-C/VI-B leaders and repeats across species. In F. columnare, subtypes VI-B and II-C have similarly sized repeats (36 bp), with sequence similarity mostly in the leader-distal ends (59-GTTGT-39) (Fig. 1). The lack of similarity in the leader-adjacent end of the repeat is surprising given the importance of this region during spacer insertion, at least in the S. thermophilus subtype II-A system (26). The F. columnare leaders also lack similarity in the leader-repeat junctions but share similarly positioned poly (A) and poly(T) regions up to 100 bp from the junction (Fig. 1). To see if II-C and VI-B CRISPR-Cas loci display similar patterns in other species, we extracted the repeats and leaders of nine species that carry these loci (lacking cas1 and cas2 in their VI-B locus). II-C and VI-B repeats shared the same leader-distal motifs across the species, but F. columnare was the only species with similarly sized II-C and VI-B repeats (II-C repeats in the other species were on average 11 bp longer) (Fig. S5). Leaders, clustered by Cas1 similarity for meaningful alignment, had leader-distal poly(A) and poly(T) motifs similar to those of F. columnare (Fig. S6), which may act as binding or signaling sites for spacer integration. As the generation of new repeats is based on a precise ruler mechanism in the acquisition complex (15,27,61), it remains unknown how the other species with differing repeat lengths generated their subtype VI-B arrays. It is therefore possible that all or some of the subtype VI-B CRISPR-Cas systems in these species are inactive and that F. columnare is one of the few or the only species engaging in cross talk between VI-B and II-C systems. Indeed, subtype II-C adaptation was recently shown in one of these species, Riemerella anatipestifer, but its cooccurring VI-B locus (denoted an orphan array) did not acquire new spacers (62).
F. columnare CRISPR-Cas model. Based on our observations and previous biochemical studies, we propose a model for F. columnare spacer acquisition and interference. During acquisition, new spacers are captured by the Cas1/2 acquisition complex (possibly aided by Cas9) in a specific conformation relative to the PAM (63,64). This prespacer is inserted into the array (15,(64)(65)(66), with the PAM end of the (proto)spacer facing the leader (6,67,68). Leader recognition is probably mediated by shared motifs in the subtype II-C and VI-B leaders and repeats that are distal from the leader-repeat junctions. During interference, spacers produce a downstream interference-PAM (TIM) in subtype II-C, as is generally observed in other type II systems (69). Since the arrays are transcribed in opposite directions with respect to the variable end of the array ( Fig. 1 and Fig. S3), the TIM location is switched for subtype VI-B spacers during interference. Acquisition events relying on an identical SAM can therefore lead to two complementary interference patterns, dictated by the array in which the spacer is inserted (Fig. 6).

DISCUSSION
We demonstrate that a native subtype VI-B CRISPR-Cas locus acquires spacers from a virulent dsDNA phage and from the host genome. Despite differences in repeat and leader sequences, shared characteristics of the newly acquired spacers with the coadapting subtype II-C locus support the model of in trans spacer acquisition in the subtype VI-B locus, which was confirmed by analysis of a cas1 deletion mutant. This arrangement provides the RNA-targeting VI-B array with an abundance of spacers that do not target mRNA. Other species with similar CRISPR-Cas loci have more divergent repeat sequences, suggesting that in trans acquisition may not occur in these species or may utilize a highly plastic acquisition complex. This study also highlights F. columnare as one of the few known species that have been demonstrated to acquire spacers from virulent phages under laboratory conditions (70).
Our conjugation assay suggested functional yet limited interference efficiency of the II-C locus against plasmids, with a 3-fold reduction in conjugation efficiency. Similarly, our previous study (34) showed that preexisting spacers do not grant F. columnare efficient phage immunity. These results may be explained by suboptimal experimental conditions (i.e., high nutrient/phage concentrations) for testing interference. Another possibility for low efficiency against phages may be phage-encoded anti-CRISPR proteins (71). On the other hand, the lack of self-targeting II-C spacers in this study suggests effective autoimmunity.
While our plasmid assay did not address VI-B interference, its role in actively targeting mRNA can be deduced by contrasting naturally acquired VI-B spacers with those acquired in the laboratory: both mRNA-and DNA-targeting VI-B spacers were acquired in the current study, but only mRNA-targeting spacers are observed in natural samples of F. columnare (34), suggesting positive selection of such spacers under natural conditions. Also, none of the preexisting VI-B spacers of F. columnare B185 (see Table S3B in the supplemental material) are perfect matches with their targets, suggesting that phages have previously escaped VI-B targeting, as also suggested by multiple natural isolates (34).
We expect future studies to reveal the extent of cross talk in type VI systems and the biochemical basis for the plasticity of the acquisition machinery when interacting with different leader and repeat sequences. Possible priming in the II-C locus should also be studied in detail, as should the interference stage of both loci, including the possible dormancy-generating effects of the VI-B locus.

MATERIALS AND METHODS
Bacterial cultures and sampling. Two strains were used to test for CRISPR spacer acquisition: Flavobacterium columnare B185 and B245 (34). Strain B185 was used as the host in the main spacer acquisition assays, and strain B245 was used in the cas1 deletion mutant experiments. Strain B185 was revived from a freezer stock in a Shieh medium culture grown overnight (72). Five replicates of 5-ml cultures were then inoculated in 0.1Â Shieh medium with the culture grown overnight to produce an initial concentration of 10 4 CFU/ml. Phage FCL-2 (34) was added to the samples at a multiplicity of infection (MOI) of 1. For the phage UV treatment (52), phage FCL-2 was exposed to UV light for 5 min on a petri dish (5,000 mJ) (UV Stratalinker 1800; Stratagene), lowering the infectivity by roughly 2 orders of magnitude (data not shown). In the bacterium plus phage plus UV-phage treatments, wild-type phages were mixed with an equal volume of UV-treated phages from the same stock and dilution. Five bacteriumonly and bacterium plus UV-phage cultures were established as controls. The cultures were incubated under agitation (120 rpm) at room temperature for 5 days, after which they were transferred (1:100) to FIG 6 Model for subtype II-C and VI-B interference in F. columnare. In this scenario, the leader-adjacent spacer of the II-C array is also present in the VI-B array. However, due to opposite transcription directions, the crRNA is reversed and complementary between the loci. Mapping the crRNA onto the phage genome using the guide-oriented approach (56) reveals a downstream TIM for II-C and an upstream "TIM" for VI-B (orange letters on the phage genome), although the SAM is the same for both. All sequences shown here correspond to actual genomic sequences (except the VI-B spacer, which is the same as the II-C spacer for illustration purposes). The relative orders of spacers/repeats in crRNA are derived from previous studies (39,48). 5 ml of 1Â Shieh medium. After 2 days (day 7 from the beginning of the experiment), cell debris had sedimented at the bottom of the tubes. We sampled 1 ml of the clear phase for living planktonic cells and extracted DNA using a DNeasy blood and tissue kit (Qiagen). The cultures were also plated on Shieh agar plates with 10-fold dilutions to estimate CFU per milliliter and to perform colony PCR.
Colony PCR. The variable ends of subtype II-C and VI-B CRISPR arrays were PCR amplified from colonies obtained under all conditions (total of 281 colonies) at day 7 (see Table S1A in the supplemental material) using primers C1_B185_F and C1_B185_R and primers C2_B185_F and C2_B185_R (Table S2A). Colonies with expanded arrays were recultured on Shieh agar plates, and the variable ends of the CRISPR arrays were rechecked. Colonies that still showed expanded arrays were grown in liquid medium, the DNA was extracted with a blood and tissue kit (Qiagen), and the variable ends were sequenced with the Sanger method using a BigDye Terminator v3.1 cycle sequencing kit (Applied Biosystems) and the automated 3130xl genetic analyzer (Applied Biosystems) (Table S1B).
Population-level CRISPR deep sequencing. Total DNA was extracted from 1-ml liquid samples using a DNeasy blood and tissue kit (Qiagen). The variable ends of subtype II-C and VI-B CRISPR loci (C1 and C2, respectively) from the phage and bacterium-only treatments were amplified with DreamTaq (Thermo Fisher Scientific), with one primer binding to the leader sequence (C2_B185_F and C1_B185_R) and one binding to the second (C1_B185_F) or third (C2_B185_R) spacer (Table S2A). The PCR protocol to amplify the CRISPR array associated with subtype II-C was as follows: 95°C for 3 min; 32 cycles of 95°C for 30 s, 59°C for 30 s, and 72°C for 1 min; and 72°C for 15 min. The PCR protocol to amplify the CRISPR array of subtype VI-B was as follows: 95°C for 3 min; 30 cycles of 95°C for 30 s, 60.2°C for 30 s, and 72°C for 1 min; and 72°C for 15 min. For subtype II-C, additional MgCl 2 (4 mM) was required for amplification. To minimize PCR bias, four separate PCRs were performed for each of the five replicates, which were then pooled and cleaned using a Qiagen MinElute reaction cleanup kit. The resulting 10-ml samples were run on a 2% agarose gel (4.6 V/cm) for 2 h 45 min in Tris-acetate-EDTA (TAE) buffer. The expected PCR product sizes for the wild-type array (subtype II-C, 181 bp; subtype VI-B, 223 bp) as well as for the expanded arrays with one new repeat-spacer unit (subtype II-C, 246 bp; subtype VI-B, 289 bp) were extracted from the phage treatment using X-tracta (Sigma) extraction tools and a gel extraction kit (Qiagen) using MinElute columns (Qiagen). In the bacterium-only treatments, the wild-type PCR products from both loci were extracted as controls. All extractions underwent two gel purification rounds to reduce contamination.
For deep sequencing, we used the previously published pipeline for multiplexed Ion Torrent sequencing (4). The extracted PCR products from type VI-B were reamplified using Maxima Hot Start Taq DNA polymerase (Thermo Fisher Scientific), the primers M13-B185_223bp_C2F and P1-B185_223bp_C2R, as well as the IonA_bc_M13 primer that contained multiplexing barcodes and an M13 adaptor (Table S2A) (73). The following PCR program was used: 95°C for 5 min; 20 cycles of 94°C for 45 s, 53°C for 1 min, and 72°C for 1 min; and 72°C for 5 min. As we were unable to obtain amplification of the type II-C array with the abovedescribed three-primer PCR approach, the PCR was split into two stages. First, type II-C PCR products were amplified using the primer pair M13-B185_C1_F3/P1-B185_181bp_C1R and the DreamTaq enzyme with added MgCl 2 (4 mM) using the following program: 95°C for 5 min; 31 cycles of 95°C for 30 s, 59°C for 30 s, and 72°C for 1 min; and 72°C for 15 min. The reaction mixture was then purified with Agencourt AMPureXP (Beckman Coulter) and reamplified using primers IonA-bc_M13 (73) and P1-B185_181bp_C1R in a reaction identical to the three-primer reaction used for type VI-B. Finally, all samples were purified using AMPureXP and quantified with a Qubit fluorometer and a Qubit dsDNA HS kit. Equimolar amounts (5 ng) of PCR products were pooled for sequencing. Pooled PCR products were sequenced after emulsion PCR with the Ion OneTouch system and Ion OT2 400 kit (Life Technologies) on Ion 314 chips with the Ion PGM (personal genome machine) sequencing 400 kit (Life Technologies), according to the manufacturer's instructions.
Deep sequencing data preparation. Nontrimmed reads were obtained from PGM and sorted by their barcodes. Trimmomatic 0.36 (74) was used for quality control with the following parameters: SLIDINGWINDOW:3:21, MINLEN:100, and TRAILING:23. Spacers were extracted from the trimmed reads using a custom Python script that extracted 27-to 32-bp spacers between an intact first repeat and the first 4 nucleotides of the subsequent repeat. Due to short reads, a fully intact second repeat was usually not available. To obtain unique spacers, each spacer pool was clustered with CD-hit-est (75) using a clustering threshold of 0.8 and a word size of 5. The resulting spacers before and after filtering are listed in Table S2B.
Mapping. The spacers were mapped to the previously published phage genome (59) using Bowtie 2.0 (76) with the following custom configuration: -very-sensitive-local -score-min G,10,8. The physical ends of the linear phage genome had been determined previously using a combination of next-generation sequencing and Sanger sequencing (59). Unmapped spacers were then mapped onto the bacterial genome, and the remaining unmapped reads (most probably due to poor sequence quality) were discarded. On average, 7.9% and 1.14% of unique spacers were discarded in the type II-C and VI-B loci, respectively. The genomes were divided into bins that span 3% of the respective genomes, and the protospacer count of each bin was calculated with a custom Python script. The resulting protospacer distribution on both genomes was illustrated using the ggplot2 package in RStudio 1.1.463 (R 3.5.3). The predicted origin of replication (oriC) region of F. columnare was determined using DoriC 10.0.0 (77).
PAM sequences. With the aim of comparing PAM sequences between the loci, we report all PAM sequences based on flanking DNA (instead of RNA) regions due to the large proportion of only-DNA-targeting type VI-B spacers. Similarly, we follow the guide-oriented approach for both loci, as is common with DNA-targeting systems (56) (PAM is depicted on the noncomplementary strand with respect to the crRNA). All PAM sequences were extracted using custom Python scripts and shown using WebLogo (78).
Spacer pool identity level analysis. To calculate the proportion of subtype VI-B spacers that are identical to II-C spacers, unique VI-B spacers from all replicates (b, d, and e) were first pooled. CD-hit-est was used to extract unique spacers from the pooled spacer set (clustering threshold of 0.8). The process was repeated for the type II-C spacers from replicates b and e. Next, the pooled unique spacer sets from both loci were compared to each other with CD-hit-est-2D using a similarity threshold of 0.9. The number of resulting clusters was then used to calculate the proportion of VI-B spacers that had a match in the type II-C spacer pool. Simulated sets of 430 spacers (the number of unique VI-B spacers pooled from all three replicates) were generated by sampling the phage genome either at random positions or from the 1,672 predicted PAM sites . Sampling was repeated 1,000 times, and similarities of the simulated spacers were compared with the type II-C spacer pool using CD-hit-est-2D as described above. A normal distribution was fit onto the simulated distributions using the R function fitdistr from the MASS package. The probability of the observed similarity, given the null hypothesis, between subtype II-C and VI-B spacers (0.58) was measured using the function pnorm from the upper tails of the distributions.
Proportions of mRNA-targeting spacers. The ability of each spacer to target an ORF's transcript was determined by two rules: (i) the crRNA of the spacer must be complementary to the coding strand, and (ii) the protospacer must be fully contained within an ORF. Intergenic spacers were excluded from the analysis. The number of mRNA-targeting spacers was divided by the total number of ORF-targeting spacers in the sample to obtain the proportion of mRNA-targeting spacers for each replicate separately. We also performed the analysis for pooled spacers from the replicates. Pools were made on the basis of locus and target, resulting in four pools (II-C phage, II-C self, VI-B phage, and VI-B self).
The mRNA proportions from pooled spacers were compared to a null hypothesis that assumed an equal chance of a spacer targeting both strands. Since intergenic spacers were excluded from the analysis, a binomial distribution could be used to construct the model. Separate distributions were created for each spacer pool to account for different numbers of spacers in each. The observed values in both loci were then compared to their respective distributions to yield direct P values on their probability given the null hypothesis described above (one-tailed test using the pbinom function in R). This analysis was done separately for the unique and absolute spacer counts.
RNA-seq and transcription direction. The direction of CRISPR array transcription was determined from bacterial RNA sequencing (RNA-seq) data. F. columnare B185 was grown without phages in 10 ml of Shieh medium at 24°C with constant shaking at 150 rpm. Next, 24-h-grown cultures (optical density [OD] of 0.166 to 0.203) were centrifuged (5,000 rpm) and stored in RNAlater (Qiagen), until RNA was extracted with an Ambion MicrobExpress mRNA purification kit. RNA quality was verified using an Agilent Bioanalyzer 2100 RNA nanochip, and samples with RNA integrity values of .9.5 were selected for library preparation (Ion Total RNA-seq kit v2). The cDNA was sequenced with Ion Torrent using a 318 chip (v2) and an internal ERCC (External RNA Controls Consortium) spike-in control, after ensuring cDNA quality (Agilent Bioanalyzer 2100 DNA high-sensitivity chip). Our results support previous reports showing type VI-B crRNA transcription from the leader end (39) and type II-C transcription toward the leader end (starting from within each repeat) or the leader-distal end of the array (48) (Fig. S3). The RNA samples for this analysis were taken from three pooled cultures of F. columnare B185 in the absence of phages, showing that these arrays are expressed constitutively albeit at a very low level (reads mapping on arrays at 167.8 and 986.6 reads per million in II-C and VI-B, respectively). cas1 deletion mutant. Not all F. columnare strains accept plasmids by conjugation (54). As strain B185 was unable to receive plasmids via conjugation or electroporation (data not shown), we used strain B245 (34) to create the cas1 deletion mutant Dcas1. Due to its competence in plasmid conjugation, strain B245 was also used in the plasmid interference assay described below. The 2.1-kbp region upstream of cas1 was amplified by PCR using Phusion DNA polymerase (New England BioLabs) and primers 2322 (adding a KpnI site) and 2323 (adding a BamHI site). The PCR product was digested with KpnI and BamHI and ligated into the plasmid pMS75 (79) that had been digested with the same enzymes, to produce pRC30. A 495-bp region downstream of cas1 was PCR amplified using primers 2324 (adding a BamHI site) and 2325 (adding a SalI site) ( Table S2A). The PCR product was digested with BamHI and SalI and ligated into pRC30 that had been digested with the same enzymes, to generate pRC32. The plasmid pRC32 was transferred from E. coli S17-1lpir into F. columnare strain B245 by conjugation. One milliliter of a culture of the recombinant E. coli strain grown overnight was inoculated into 9 ml of LB containing 100 mg/ml ampicillin and incubated with shaking at 37°C until the OD at 600 nm (OD 600 ) reached 0.6. Similarly, 5 ml of a culture of F. columnare B245 grown overnight was inoculated into 25 ml fresh TYES (tryptone yeast extract salts) broth (80) and incubated at 28°C with shaking, until the OD 600 reached 0.6. The E. coli and F. columnare cells were centrifuged separately at 5,000 rpm for 15 min, and the pellets were washed with 10 ml of TYES medium and centrifuged at 5,000 rpm for 10 min. The E. coli and F. columnare cell pellets were each suspended in 0.8 ml of TYES medium, mixed, and centrifuged at 7,000 rpm for 3 min. Excess medium was removed, and the mixed pellet was suspended and spotted on FCGM (Flavobacterium columnare growth medium) agar medium (80). After incubation at 30°C for 24 h, cells were scraped off the plate and suspended in 1.5 ml of TYES medium. Next, 100-ml aliquots were spread on TYES agar containing 1 mg/ml tobramycin and 5 mg/ml tetracycline and incubated at 30°C for 72 h. The resulting tetracycline-resistant colonies were streaked for isolation, inoculated into TYES liquid medium without tetracycline, and incubated overnight at 25°C with shaking to allow plasmid loss. The cells were plated on TYES medium containing 10% sucrose, incubated at room temperature (20°C) to select for the lack of sucrose toxicity, and streaked for isolation using the same selection. The cas1 deletion was screened by PCR and gel electrophoresis and verified with Sanger sequencing using primers Cas1_F and Cas1_R. All primers used are listed in Table S2A.
Strains used in the new spacer acquisition experiment were Dcas1 and B245_rev (rev-wt). The latter is a reversion mutant where the integrated plasmid was lost by recombination in a manner that regenerated the wild-type sequence. Phage V156 (34) was used to trigger spacer acquisition. The infectivity of phage V156 against the two hosts (Dcas1 and rev-wt) was measured with a standard double-layer method, where 300 ml of a turbid bacterial culture was mixed with 3 ml of TYES medium with 0.1% mucin and 1% agar. The mixture was poured onto a TYES agar plate, and 10 ml of phage V156 dilutions (10 21 to 10 27 ) were spotted in duplicates onto the solidified agar. The number of plaques was counted after 2 days from the 10 27 dilutions. Phage titers (PFU per milliliter) were 1.9 Â 10 10 and 3.7 Â 10 10 on the Dcas1 and rev-wt strains, respectively.
For the spacer acquisition experiment, both strains were grown overnight in single 5-ml cultures to reach an OD 600 of 0.45. Ten phage-infected cultures in 0.1Â TYES medium as well as two bacterium-only cultures were started from these cultures grown overnight. The spacer acquisition protocol differed from the one used with B185, as follows: medium TYES (versus Shieh medium), volume of 1 ml (versus 5 ml), and duration of growth in 0.1Â medium of 3 weeks (versus 5 days). The protocol was prolonged because follow-up experiments showed that longer incubation in diluted medium resulted in more efficient spacer acquisition. The medium or its volume has no detectable effect on the acquisition efficiency (data not shown) and was changed for practical reasons. After 3 weeks in diluted medium (0.1Â), the cultures were transferred to regular TYES medium (1Â), and the CRISPR arrays were PCR amplified after 2 days. All 12 phage-containing cultures of the Dcas1 strain showed growth in 1Â medium, in contrast to only 2 of 12 replicates of the rev-wt cultures. The variable ends of both CRISPR arrays in all surviving cultures were PCR amplified using primers B245_C1_F and F_col_C1_R (B245 II-C array) and primers F_col_C2_F and B245_C2_R (B245 VI-B array) (Table S2A and Fig. S4).
Plasmid interference assay. A derivative of plasmid pAS43 (54) (with a removed KpnI restriction site, not relevant to this experiment) was used as a template for adding a protospacer sequence followed by the putative PAM sequence 59-NNNNNTAAA-39. Two plasmids were constructed: one with a protospacer matching the most recent II-C spacer (next to the leader sequence) of F. columnare strain B245 (GGTAATTTTAAAACAAATGAGTATGTACGAACTGCTAAA [the PAM sequence is underlined]) and one with a nonmatching control sequence (ATCAGATCTAATCTCTATGTCAATGTATGAACTGCTAAA [the PAM sequence is underlined]). The insertions were added to the plasmid at an intergenic multiple-cloningsite region using the Q5 site-directed mutagenesis kit (New England BioLabs) and the oligonucleotide pair pAS43_protoSDM_F/R (matching protospacer) or pAS43_protoSDM_neg_F/R (nonmatching protospacer) ( Table S2A). The initial PCR was done using Phusion polymerase with 3% dimethyl sulfoxide (DMSO) using the following program: 98°C for 10 s and 24 cycles of 98°C for 1 s, 60°C for 7 s, 72°C for 12 min, and 72°C for 15 min. The amplified product was inspected on an agarose gel and subjected to downstream KLD (kinase, ligase and DpnI mix) treatment according to the mutagenesis kit instructions. The resulting plasmid was transformed into chemically competent E. coli strain S17-1lpir. Successful transformation and insertion were verified by colony PCR and Sanger sequencing (primer pair pAS43_MCS2_seq_F/R) (Table S2A).
In the quantitative conjugation assay, the plasmids were transferred into F. columnare strain B245. Both the donor and the recipient cells were first grown overnight from a freezer stock. The turbid cultures were then transferred to fresh medium (LB for E. coli and TYES medium for F. columnare). E. coli was grown to an OD 600 of 0.95, and the recipient F. columnare strain was grown to an OD 600 of 0.414. The conjugation protocol was performed similarly to the cas1 deletion (described above), and the final donor/recipient mixtures were spotted on 0.45-mm filters resting on FCGM plates. After 24 h of growth at 30°C, the cells were scraped from the filters and resuspended in 2 ml TYES liquid medium. Next, 150 ml of the mixed cultures was plated on double-antibiotic selection plates without dilution (cefoxitin for conjugant selection and tobramycin for eliminating E. coli) as well as on tobramycin-only plates with 10 25 and 10 26 dilutions for determining the number of potential recipient cells in each replicate (F. columnare is naturally resistant to tobramycin). Colonies were counted and plates were photographed after 72 h at 25°C. The efficiency of conjugation was determined by dividing the number of F. columnare conjugants (double-selection plate) by the number of potential recipients (tobramycin plates) in each replicate.
Comparison of type II-C and VI-B leaders and repeats in other species. Microbial species that carry intact subtype II-C (containing cas1, cas2, cas9, and a CRISPR array) and VI-B (containing cas13b and a CRISPR array) loci were identified using CRISPRCasdb (81), and one strain per species was selected for further analysis (Flavobacterium branchiophilum was excluded from the analysis due to the absence of cas1 in the subtype II-C locus). From each strain, repeat sequences from both loci were aligned with Geneious aligner (global alignment with free gaps, 51% cost matrix, gap open penalty of 12, and gap extension penalty of 3). Next, 200-bp leader sequences were extracted downstream of the expected variable end of the array. The variable ends were primarily determined by the repeats' direction compared to F. columnare strain B185. These initial predictions were supported by the presence of possible degenerate terminal repeats and alignment scores of both flanking regions compared to the leader of F. columnare B185. To allow the comparison of the highly divergent leaders, the strains were first clustered based on their Cas1 protein (Jukes-Cantor, unweighted pair group method using average linkages [UPGMA]). The leaders from both loci within the resulting six clusters (most of which contained single species) were aligned with Geneious aligner (global alignment, 65% cost matrix, gap open penalty of 17, and gap extension penalty 12). We used strict alignment rules to emphasize the importance of the distance of possible motifs from the repeat-leader junction. Analyses and alignments were performed with Geneious 9.1.8.
Data availability. The B245 genome and raw reads from the spacer acquisition experiment have been submitted to GenBank under accession numbers CP071008 and SAMN18022999, respectively. The genomes of F. columnare strain B185 (accession no. NZ_CP010992.1) and phage FCL-2 (accession no. NC _027125.1) are available in GenBank. Custom code and instructions for analyzing the raw data and reproducing all figures in the manuscript are in GitHub at https://github.com/vihoikka/spacerAQ_vh.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.