We have a collection of 477 S. mutans genomes that was curated by Vince Richards at Clemson University; sequencing files were downloaded from GenBank/NCBI and assembled. Genomes are available in. gbk format on the Github platform (https://github.com/theshieldslab/Streptococcus-mutans-CRISPR-Spacers-Analysis). For this study CRISPR spacers were isolated from 335 S. mutans genomes. Incomplete assemblies were evident in the other S. mutans genomes, likely because of Illumina based whole genome sequencing, leading to difficulty in identifying repeat regions (i.e., CRISPR spacer regions).

Identification and data curation of the CRISPR systems found in our clinical isolates was conducted at the University of Florida HiPerGator cluster computer with the CRISPRCasFinder (Couvin et al., 2018) version 4.2.17 pipeline. The resulting output files, containing the predicted spacers, were then processed and summarized in several files containing CRISPR system types and subtype, list of genes, repeat sequences, count of repeat sequences, and spacers sequences. The spacer sequences were then labeled with their respective isolate name, contig ID where the spacers was predicted, and an additional counter number. All CRISPR spacer sequences were then parsed into a FASTA format file, which is available on the Github platform.

Blastn was used to identify spacer targets from phage, integrative and conjugative elements (ICEs), plasmids, and S. mutans. Phage DNA sequences were downloaded from NCBI (ftp://ftp.ncbi.nih.gov/refseq/release/viral/). ICE sequences were downloaded from the ICE database ICEberg 2.0 (https://bioinfo-mml.sjtu.edu.cn/ICEberg2/download.html) (Liu et al., 2019). Plasmid sequences were downloaded from a database containing 10,892 complete and annotated plasmids (Brooks et al., 2019). In addition, known S. mutans plasmid sequences were compiled (available on the Github platform) (Zhou et al., 2001; Caufield et al., 2007; Rheinberg et al., 2013). Parameters for Blastn were set to 95% sequence coverage (qcovs) and 85% sequence identity (pident). These parameters are a reasonable threshold for identifying CRISPR targets but it is possible that CRISPR spacers meeting these criteria are not capable of DNA interference because of mismatches in the seed region, or because the foreign DNA has mismatches in the proto-spacer adjacent motif (PAM). Other parameters generated with the sequence results included the number of mismatches, the evalue and the subject strand (sstrand). All raw output data is available as. xlsx files on the Github platform.

The data mining for gene clusters was conducted in a series of steps, starting with Prokka annotations for each clinical isolate genome (Seemann, 2014). Since all isolates belong to the Streptococcus genus, annotations with Prokka were defined within Bacteria and Streptococci for better gene predictions. Multilocus sequence typing (MLST) was conducted with Roary (Page et al., 2015), which is ideal for close non-divergent genomes. The final step was to generate the phylogenomic tree from the genome alignment file created by Roary. A maximum likelihood tree was drawn with the Phyml (Guindon and Gascuel, 2003; Hordijk and Gascuel, 2005) algorithm and plotted in R (R: a language and environment for statistical computing) with the GGtree (Yu et al., 2017) package. This work also included custom bash command lines in order to generate accurate measurements of GC content and genome size for all the isolates in our bank.

Using CRISPRCasFinder (Couvin et al., 2018) across our genome database we identified a total of 8,078 CRISPR spacers of which 5,875 are unique sequences. Isolates with CRISPR arrays present carry an average of 23.9 spacers each, with strain smu174 carrying 129 spacers in total (on one array). By comparison, Streptococcus thermophilus strains carry an average of 33 spacers per cassette (Horvath et al., 2008). The majority of S. mutans strains carry one or two CRISPR arrays but there are rare instances of strains carrying 4 + (Figure 1A). Figure 1B shows a frequency distribution of CRISPR spacer load across all the isolates. CRISPR spacer acquisition is a balance between obtaining immunity against harmful sequences without diluting spacers (some which may provide inadequate immunity because of mutation) across the available Cas machinery (Garrett, 2021). Most CRISPR spacers were either 30-nt or 32–35-nt in length (Figure 1C), with these two groups likely arising from differences in Type I and Type II CRISPR spacer lengths. The average GC content of CRISPR spacers was 38.16%, which is above the average GC content of S. mutans strains (Figure 1D). This may indicate that foreign DNA encountered by S. mutans is on average of a higher GC content than the average S. mutans genome. By way of comparison, M102AD, an S. mutans infecting phage has GC content of 39.6%.

Having identified 8,078 spacers across our S. mutans genomes we next wanted to identify targets for these spacers. First, we characterized how many of these spacers target foreign DNA. This is the traditional target of CRISPR-Cas systems, and includes interference with phage, plasmid, integrative and conjugative elements (ICEs), and other bacterial DNA. In total, 1913 spacers (23.7%) mapped to these targets (Figure 2). Next, we will describe these targets in more detail for each sub-group of foreign DNA.

Phage . Adaptive immunity against bacteriophage is a notable feature of CRISPR-Cas systems across bacteria. Consistent with this concept S. mutans CRISPR spacers mapped to 1,137 phage sequences, of which 531 were unique matches. A total of sixteen distinct phage species are targeted by S. mutans CRISPR spacers (Supplementary Table S1). For four of these phages S. mutans is the known host, whereas for the other phages the hosts are other Firmicutes bacteria; Staphylococci, Enterococci, Lactobacilli and Streptococci. For these phages, typically only one spacer aligned against a conserved target. We anticipate that rather than infecting S. mutans, these phage genes are present either as prophage elements or as orphan genes in S. mutans genomes. The four S. mutans infecting phage that S. mutans has acquired spacer immunity against are known as M102, M102AD, ɸAPCM01 and smHBZ8 (Delisle and Rostkowski, 1993; Delisle et al., 2012; Dalmasso et al., 2015; Zaken et al., 2021). These are the only S. mutans phage that have been genome sequenced and all belong to the Siphoviridae family of double-stranded DNA viruses; they are lytic phage. At least two other S. mutans phage, e10 and f1, are known to infect S. mutans but these strains have not been sequenced (Delisle and Rostkowski, 1993). The four sequenced S. mutans phage are highly related, particularly the first 20 Kbp (Figure 3). However, they each contain a variable region consisting of small open reading frames with unknown functions at the 3’ end of the genome (Figure 3). Spacers were mapped and viewed with a genome browser to determine if there is any bias in spacer acquisition to specific phage genes. Overall, it appears that immunity to the variable region is less commonly acquired, and this was most evident for M102 and M102AD.

Isolation of S. mutans phage is rare, For example, ɸAPCM01 was the only S. mutans phage isolated from 85 saliva samples (Dalmasso et al., 2015) and smHBZ8 was the single S. mutans phage isolated from 254 samples (diverse sources; saliva, dental sewage, cariogenic dentin, extracted teeth, and dental plaque) (Zaken et al., 2021). To understand how important CRISPR-Cas is in explaining the low frequency of infective phage isolation we calculated the frequency of phage immunity spacers across our genome database. In total, 64% of S. mutans strains that carry CRISPR-Cas systems have phage targeting spacers (against M102AD or M102 or ɸAPCM01). The high frequency of CRISPR-Cas systems and anti-phage spacers would likely limit the efficacy of phage therapy in S. mutans.

Plasmids . CRISPR spacers are also naturally acquired against plasmid DNA (Palmer and Gilmore, 2010; Touchon et al., 2011). Acquisition of spacers against plasmids was very infrequently identified for S. mutans. In a database containing ∼12,000 sequenced plasmids (Brooks et al., 2019) only two spacers mapped. A spacer from smu333 mapped to pLF25067 carried by Lactobacillus fermentum (Aryantini et al., 2017) and a spacer from smu408 mapped to pSZ4 from Staphylococcus warneri (Cheng et al., 2013, 1). One strain, smu441, carries a spacer that maps (88% coverage) against a small 5.6 kbp ‘cryptic’ plasmid that is carried by ∼5% of S. mutans strains (Zhou et al., 2001, 140; Rheinberg et al., 2013).

With the observation that S. mutans has gained immunity against inter-species transfer of mobile genetic elements we next wanted to explore if CRISPR spacers have been acquired more generally against S. mutans strain DNA. To do this, we searched for CRISPR spacer hits within genes encoded by S. mutans across all the genomes included in this study. A total of sixty-three annotated genes are the target of CRISPR spacers (Supplementary Table S2). In addition, six other genes with functional predictions (Supplementary Table S2) are also the target of CRISPR spacers. Lastly, S. mutans hypothetical proteins were also matched to S. mutans CRISPR spacer sequences. From this set of genes (excluding hypotheticals), 22 reside in the core genome (>99% of genomes), 4 are soft core genes (95–99% of genomes), 17 are shell genes (15–95% of genomes), and 20 are cloud genes (<15% of genomes) (Supplementary Table S2). CRISPR targets against S. mutans core genes are over-represented given that core genes only account for 5% of the S. mutans pangenome. This is compared with cloud genes which make up 85% of the pangenome. Notable S. mutans gene CRISPR targets include the clp system (Kajfasz et al., 2009), the gtfC glucosyltransferase (Hanada and Kuramitsu, 1988), the cell surface antigen spaP (Crowley et al., 1999), carbohydrate utilization systems (i.e., fruA and lacF) (Burne et al., 1999; Zeng et al., 2010) and several genes involved in DNA replication, repair or modification (i.e., dpnM, parE, radD, recF, repA, smc, ssb and topB). Several of the CRISPR spacers that target the S. mutans pangenome are likely targeting mobile DNA elements. For example, spacers against tyrosine recombinases (i.e., xerC and xerD), conjugal transfer protein traG, and the Int-Tn transposase from the ICE Tn916. To visualize the extent of CRISPR spacer acquisition against S. mutans genomes, spacers were mapped to two strains, UA159 and NN2025 (Figures 5A,B). The number of spacers that aligned within 10 kb regions varied, with notable hotspots against the genomes of both UA159 and NG8. These hotspot regions were TnSmu1 and repeat regions within the promoters of clpP (protein homeostasis), rexB (putative exonuclease) and a hypothetical protein.

Inter-species CRISPR targeting was also visualized using network analysis (Figure 5C). This analysis showed that a minority of S. mutans strains (ca. 60 strains; red nodes in Figure 5C) were accounting for the majority of CRISPR spacer connections between S. mutans strains. These strains carry CRISPR spacers against highly conserved (core) S. mutans genes (Supplementary Table S2). We reasoned that this group of highly connected strains could actually be carrying self-targeting CRISPR spacers against highly conserved genes. Analysis of self-targeting revealed that ninety-four S. mutans strains carry self-targeting CRISPR spacers (Supplementary Table S3) and most of the highly connected nodes in the network analysis are strains that self-target. All of these strains contain the Cas genes required for functional CRISPR-Cas systems but it is unknown if these systems are inactivated through point mutations. Acquisition of self-targeting spacers would likely be detrimental and create pressure for inactivation of the CRISPR-Cas systems. Six out of the ninety-four self-targeting strains contain anti-CRISPR (Acr) proteins that could also account for a loss of functionality (Supplementary Table S3). In summary, S. mutans strains carry spacers that target S. mutans genes although the relative amount of intra-species-targeting versus only self-targeting is currently unknown.

Analysis of S. mutans CRISPR spacers has shown potential immunity acquisition against phage, plasmid, ICE, and S. mutans DNA. Acquisition of spacers against these types of DNA should act to limit the rate of HGT in S. mutans. The association between CRISPR-Cas and its impact on HGT in bacteria is not clear. For example, P. aeruginosa genomes with CRISPR-Cas are significantly smaller than those lacking CRISPR-Cas which suggests that CRISPR-Cas impedes HGT (Wheatley and MacLean, 2021). However, a similar trend was not found for Staphylococcus aureus or Acinetobacter baumannii (Pursey et al., 2022). Here, we are using CRISPR spacer load as a measure of CRISPR-Cas activity. It has limitations because we cannot determine if the CRISPR arrays/systems are functional or defective. However, it is the most appropriate measure because it is unfeasible to test for genomic differences between strains with and without CRISPR-Cas systems because the majority of S. mutans strains have at least one CRISPR-Cas system. CRISPR spacer load, the number of CRISPR spacers per genome, has been used as a measure of CRISPR activity by others (Gophna et al., 2015). For our analysis, S. mutans strains were grouped by the total number of spacers they have acquired and for each group the median size of the genome was computed (Figure 6). We found that increased CRISPR spacers loads were associated with marginally increased genome sizes. Strains with 1–20 CRISPR spacers had a median genome size of 2.01 Mbp and strains with 40–60 CRISPR spacers had a median genome size of 2.05 Mbp (two-tailed t-test p < 0.01). Our analysis shows that increased CRISPR spacer acquisition does not lead to reduced genome size in S. mutans.

By limiting HGT CRISPR-Cas could also affect the GC content of host genomes by reducing the acquisition of higher or lower GC foreign DNA. The average GC content across all S. mutans CRISPR spacers was 38.2% (Figure 1D) which may indicate that on average foreign DNA elements have higher GC content than the average S. mutans genome (36.8%). Phages that infect S. mutans, M102AD, M102, and SMHBZ8 have GC content of 39.6%, 39.21%, and 38.8% respectively (van der Ploeg, 2007; Delisle et al., 2012; Zaken et al., 2021). The GC content of S. mutans genomes was consistent for strains with 1–20 (36.80%), 20–40 (36.82%) and 40–60 (36.79%) CRISPR spacers (Figure 7). A significant difference in GC content was observed for strains with 60–80 (36.68%) CRISPR spacers compared with the 1–20 and 20–40 groupings (two-tailed t-test p <0.05). Although only nominally lower than the other groups, this might indicate a reduction in the acquisition of higher GC foreign DNA among strains with 60–80 CRISPR spacers.

To gain a greater insight into the impact of CRISPR-Cas systems on HGT we wanted to examine how CRISPR spacer load alters the size and distribution of the S. mutans pangenome. In our analysis including 335 S. mutans genomes, the pangenome size was calculated to include 15,062 genes. There were 822 genes in the core genome (>99% of genomes), 496 soft core genes (95–99% of genomes), 982 shell genes (15–95% of genomes), and 12,762 cloud genes (<15% of genomes). Next, pangenome analysis was conducted on two groups, one with a low CRISPR spacer load (1–20 spacers, average 9.36; 163 genomes) and another with a high CRISPR spacer load (21–129 spacers, average 37.93; 169 genomes). For both groups we generated pie charts which show the percentage of genes located in the different gene clusters (core; soft core; shell; cloud) (Figure 8). The low and high CRISPR spacer load groups have a similar percentage of genes located within the cloud genes gene cluster. Cloud genes are those that are most likely acquired via HGT. By this measure there is no limiting of HGT by CRISPR-Cas systems for S. mutans strains with either low or high CRISPR spacer loads.