Using the Streptococcus mutans CRISPR-Cas9 System to Study Essential Genes

More than 95% of the 477 S. mutans strains for which whole or draft genome sequences are presently available carry at least one CRISPR-Cas system (Fig 1A). The Type II-A CRISPR-Cas system, which can be repurposed for CRISPRi, is the most prevalent across the species with 339/477 strains carrying a complete system (Fig 1A). For reprogramming of S. mutans UA159 Cas9 (SMu.1405c) into a CRISPRi system, we began by mutating two amino acid residues (D10 and H840) known to be required for the nuclease activity of S. pyogenes Cas9 [21] (Fig 1B). We were able to confirm the DNA-binding and repressing activity of dCas9_Smu by expressing a sgRNA against a constitutively expressed gfp gene [22] (Fig 1C). Next, the S. mutans dcas9 gene was cloned into pZX9 [23], such that expression could be activated by xylose. Prior to this, we discovered, using green fluorescent protein (GFP) as a readout for xylose-induced gene expression, that the P_xyl promoter was less active when glucose was the sole carbohydrate source compared to maltose (S1 Fig in S1 File). We suspect the different expression levels on glucose versus maltose could be due to carbohydrate catabolite repression and/or effects of certain phosphoenolpyruvate:sugar phosphotransferase system (PTS) components on P_xyl activity. Thus, we used the defined media ‘FMC’ containing maltose for all CRISPRi related experiments. To make the CRISPRi strain (Fig 1D), the sgRNA plasmid was integrated in single copy into the S. mutans genome in a strain that contained i) a gfp gene driven by the constitutively active P23 promoter and integrated into the chromosome, ii) the P_xyl-dcas9_Smu plasmid and iii) point mutations that prevent translation of SMu.1405c (cas9; start codon mutated to stop codon). S. mutans Cas9 must be mutated, otherwise endogenous cas9 expression in combination with sgRNAs would likely cause lethal chromosome damage; S. mutans is unable to repair the double-strand cleavage catalyzed by a functional Cas9.

Having developed the system, we moved to assessing its efficiency. Using the S. mutans CRISPRi technique, we observed titratable repression of GFP fluorescence (Fig 1E) and GFP protein levels (Fig 1F). Importantly, there were no obvious effects on the growth of S. mutans as increasing amounts of xylose were added (up to 0.025% w/v). Also of note, slightly higher amounts of xylose (0.025% vs. 0.00625%) were required to achieve similar levels of repression of GFP production in a strain containing a xylose-inducible S. pyogenes dcas9 (Fig 1E and 1F). We also observed slight growth impairments as increasing amounts of dCas9_Spy were produced (Fig 1E). Overall, the differences in performance of the two Cas9 constructs were minor. Next, we measured gene-silencing (absence of GFP fluorescence) at the single-cell level (S2 Fig in S1 File). The number of cells escaping CRISPRi-mediated gene knockdown (GFP positive) was very small, indicating that, at saturating amounts of xylose, the effects of the S. mutans-derived CRISPRi system are uniform across the population when the dCas9 protein is delivered on a multi-copy plasmid. The pZX9 plasmid that carries the xylose-inducible system uses the streptococcal replicon from pDL278. Without antibiotic pressure we have shown that pDL278 is stable in S. mutans for at least 50 generations [22].

RNA sequencing (RNA-seq) experiments were conducted to test the effects of the CRISPRi system on the S. mutans transcriptome. We compared the transcriptome of the CRISPRi strain without and with the addition of xylose (0.1%). With xylose, we observed an increase in dCas9 expression and a concomitant decrease in gfp expression compared to the no xylose condition (S1 Table). This was specific to the presence of sgRNA-gfp as we observed no reduction in gfp expression in a strain lacking the sgRNA and xylose added (no sgRNA in S1 Table). As has been noted by other authors [13,14] we measured a polar effect of CRISPRi on genes downstream of the target gene (msmK and dexB are downstream of the location of the gfp gene). Importantly, the effect of CRISPRi induction on the S. mutans transcriptome was minimal. We observed no S. mutans gene expression changes with a log₂FC >2, and only twelve genes with a log₂FC >1 (FDR <0.05). Furthermore, these twelve genes typically contained low read counts, indicating that the slight changes in gene expression could be attributable to noise. In the absence of CRISPRi, but with xylose (no sgRNA strain), no changes in S. mutans gene expression were observed; so 0.1% xylose has no discernable impact on the S. mutans transcriptome. From these RNA-seq results, we conclude that the CRISPRi system is highly specific and does not appreciably alter S. mutans gene expression other than targeted genes.

To begin to optimize the S. mutans CRISPRi system, we examined parameters that are important for functionality; the first of these was sgRNA specificity. sgRNA-Cas9 systems that operate with lower sequence stringency could be prone to off-target effects that occur when guide RNA and dCas9 bind to a site(s) in the genome that was not intended to be affected. To test this, we cloned an sgRNA with a complementary sequence for the gene atpB, which encodes the beta subunit of the F₁F_O proton-translocating ATPase. We modified the sgRNA-atpB sequence by introducing substitutions along the length of guide RNA (Fig 2A). The sgRNA plasmids were added to competent S. mutans and plated onto selective agar. We expected that transformation into S. mutans of the sgRNA-atpB plasmid lacking sequence modifications would not be tolerated, leading to a very low transformation efficiency. As expected, transformation of the sgRNA-atpB^T (T; targeting the template strand) occurred at very low efficiency, 1 x 10⁻⁵ (Fig 2B). Conversely, transformation efficiency in a Δcas9 (deletion of Smu.1405c) strain, the dead cas9 strain, or in a strain carrying a deletion of 20-bp of the atpB^T sequence from the sgRNA (sgRNAneg) was 100- to 1000-fold more efficient than with the intact atpB sgRNA (Fig 2B).

We next evaluated the effects of expressing sgRNAs with one, two or three base mutations within the atpB^T sequence. Single base-pair mismatches were tolerated at the 5’ end of the sgRNA (atpB_3, atpB_6, atpB_9 and atpB_12). Plasmids with mismatches closer to the PAM (atpB_15 and atpB_18) transformed with higher efficiency into S. mutans, indicative of a loss of sgRNA activity. All of the double and triple base mismatches reduced the activity of the sgRNAs, as these plasmids transformed S. mutans at levels comparable to the cas9-deficient strain (Fig 2B). These results suggest that near perfect sgRNA sequence complementarity is required to silence genes, and thus it should be possible to minimize off-target effects with careful design of sgRNAs.

Another important feature of CRISPR-Cas9 systems, both as a natural bacterial immune system and for CRISPRi, is the proto-spacer adjacent motif (PAM). This sequence is required for the Cas9 protein to recognize interference targets and is also critical during the acquisition of CRISPR spacers (for Type II-A systems) [21,24]. For a given bacterial strain, Type II-A PAM sequences are generated by aligning CRISPR spacers to phage genomes, extracting the sequence downstream of the spacer target, and then building a conserved sequence from these. The difficulty with taking this approach with S. mutans UA159 is that there are too few CRISPR spacers from which to build a reliable PAM sequence. Consequently, we compiled PAM sequences from a large (>450) collection of S. mutans genomes. After narrowing the data to strains that had Type II-A CRISPR-Cas systems and ≥5 spacers, PAM sequences were generated for 100 individual S. mutans strains using WebLogo (S3 Fig in S1 File). Per this bioinformatics analysis, a highly conserved PAM for Cas9_Smu was not evident. Rather, a significant number of S. mutans strains with Cas9 contained spacers that might require the 5’-NGG-3’ recognition sequence for phage DNA interference, and these spacer arrays grouped together by alignment (Fig 3A, yellow color). For other S. mutans strains with Cas9, PAM sequences were more random, but at least two other groups, 5’-TTN-3’ and 5’-NAAA-3’, existed (S3 Fig in S1 File). We observed no obvious grouping for S. mutans of the predicted PAM sequences by Cas9 protein identity (Fig 3A). We reasoned that PAM recognition might be protein sequence specific and thus expected that the different PAM groups predicted by bioinformatics would also group together. A lack of coherence between the predicted PAM sequences and Cas9 protein sequences suggests that the PAM sequences assembled via bioinformatics may not be the best indicator of the requirements for recognition for all S. mutans strains. So, to investigate DNA recognition in greater detail we developed an interference assay with GFP fluorescence as the readout. Using site-directed mutagenesis we incorporated assorted PAMs into a construct in which the P23 promoter could drive the expression of a gfp gene (Fig 3B). This assay confirmed that S. mutans dCas9 recognizes a 5’-NGG-3’ PAM (Fig 3C). With 5’-NAG-3’ we observed moderate CRISPRi activity with dCas9_Smu, although the results varied as a function of the DNA base present in the first position, with up to an 80% loss of GFP fluorescence. Weak binding of S. pyogenes Cas9 to a 5’-NAG-3’ PAM has been reported before [25–27]. Nuclease activity via 5’-NGA-3’ recognition has previously been reported for Cas9_Spy [26,27] but we were unable to observe CRISPRi activity with either dCas9 protein with a 5’-AGA-3’ PAM. No interference was observed with any of the other PAM sequence combinations tested.

We constructed a library of sgRNAs targeting 253 potentially essential and growth supporting genes (Fig 4A). These genes (S2 Table) were selected from previously conducted Tn-seq experiments [10], Tn-seq performed as part of this study, and genes shown to be essential in previous studies (e.g. vicR/walR). We were unable to design sgRNAs for 21 putative essential or growth supporting genes, of which over half (12) were ribosomal proteins. These genes were small (average size of 275-bp) and generally lacked PAM sites that would give desirable sgRNA sequences (e.g. RNA folding, off-target effects, location in 5’ region of gene). During the first round of cloning (see Materials and Methods) we obtained correct sgRNA clones for 181 of 282 (64%) of the desired strains. After three rounds of cloning 259/282 (92%) of the sgRNAs were successfully cloned. We were unable to obtain sgRNA clones for 23 (8%) genes. After plasmid cloning in E. coli 10-beta, sequence-verified sgRNA plasmids were transformed into S. mutans Δcas9 P_xyl-dcas9_Smu. While it would be possible to clone the sgRNA plasmids directly into S. mutans, we wanted to minimize the number of passages in S. mutans as CRISPRi strains are known to acquire suppressors when guide RNAs are expressed in their hosts [14].

We measured the growth of each sgRNA strain in FMC-maltose either in the absence or presence of 0.1% xylose for 16 h using a Bioscreen C Automated Microbiology Growth Curve System. Afterwards, the area under the curve (AUC) was calculated for both conditions. Quantification of growth by AUC is useful as it combines growth phases (lag, exponential and stationary phases). For each strain, the AUC_xylose was divided by AUC_uninduced. Growth phenotypes were grouped per these ratios: >0.5, minor phenotype; 0.25–0.49, moderate growth defect; <0.25, strong growth defect. It is important to note that the differences between strains slightly above or below the ratio cut-off points are not substantial. Growth defects were observed in 194/253 (77%) of the strains expressing sgRNAs that were designed to target essential or growth supporting genes (Fig 4B; S2 Table). Overall, we observed a higher percentage of growth defects for strains with silenced essential genes versus growth-supporting genes, 82% and 62%, respectively (Fig 4B). When comparing between different Clusters of Orthologous Groups (COGs), the two lowest median growth defects were observed for silenced genes in the translation (J) and lipid metabolism and transport pathways (I) (Fig 4C; silenced genes in the category defense mechanisms (V) displayed strong growth defects but we are unsure of these genes functions). The two highest median growth defects were observed when sgRNAs targeted genes related to coenzyme transport and metabolism (H) and signal transduction mechanisms (T) (Fig 4C). Growth defects were not observed for the six sgRNAs that targeted non-essential genes (scrB, lacG, gtfB, bceA, gfp, gfpprom; see S2 Table). There was no significant relationship between the growth defect we observed and the position of the sgRNA within the gene or the size of the target gene (S4 Fig in S1 File).

Next we evaluated the morphological phenotypes of eighty CRISPRi strains placed in the following functional categories: fatty acid biosynthesis/membrane biogenesis, peptidoglycan synthesis, glycolysis, DNA replication, transcription, translation, cell division, and ATP synthase activity. Using DAPI staining, we were able to visualize the bacterial chromosome, and with nile red we could detect the cell membrane. The cell morphology defects of essential gene-targeted strains were diverse (Fig 5). Defects included increased cell chaining, abnormally large cells, heterogeneously sized cells, cell lysis and anucleate cells (S5-S12 Figs in S1 File). The most commonly observed morphological phenotypes across all the pathways investigated were increased cell chaining (e.g. pfk, accD, dnaE, rs3, atpC) and heterogeneously sized cells (e.g. pgi, glmM, dnaG, eftS, atpF) (Fig 5A). It is important to note that there were strains without obvious growth defects that displayed major cell morphology defects (e.g. ftsX and glmS). Repression of genes involved in the glycolysis pathway led to significant growth defects and larger/misshapen cells (S5 Fig in S1 File). When pgm, encoding phosphoglucomutase, was repressed, cells appeared to be somewhat rod-shaped, probably associated with altered or deficient biosynthesis of cell wall polysaccharides. When genes involved in peptidoglycan biosynthesis were repressed, the most common phenotype was the formation of singular large round cells (S6 Fig in S1 File). The percentage of cells displaying this phenotype varied, with it being most common in a strain with silenced murM, which in S. pneumoniae encodes a protein that incorporates amino acids into the side-chain of peptidoglycan [28]. Repression of genes involved in the conversion of acetyl-CoA to malonyl-CoA (accACD and bccP), an important step in fatty acid biosynthesis, led to excessive chaining of cells (S7 Fig in S1 File). Repression of fatty acid biosynthetic (fab) genes yielded minor phenotypic changes, whereas silencing of genes downstream of the FAS II pathway (plsX and cdsA) led to more obvious morphological defects. As anticipated, when genes involved in DNA replication were repressed, not all cells contained nucleoids (S8 Fig in S1 File). In addition to anucleate cells, certain CRISPRi strains with repressed chromosome replication also showed severe morphological defects (e.g. dnaE, dnaG, priA). CRISPRi repression of translation-related genes increased cell chaining (S9 Fig in S1 File). When the ribosomal proteins rs3 (small ribosomal subunit) and rl11 (large ribosomal subunit) were repressed, we observed anucleate cells that lacked a well-defined cell envelope. CRISPRi silencing of genes encoding subunits of the F₁F_O-ATPase led to strong growth defects, increased cell chaining and variability in nile red staining (S10 Fig in S1 File). Repression of transcription-related genes also led to increased cell chaining and attenuated growth (S11 Fig in S1 File). The cell shape and size of rpoB-, rpoC-, and rpoZ-silenced CRISPRi strains were severely impacted. When genes involved in cell division were repressed, strains displayed heterogeneity in cell size and shape (S12 Fig in S1 File). Notably, with repression of divIC, ftsA, ftsL and ftsZ we observed similar morphological changes. Cells appeared to be rod-shaped, except that they bulged at the cell septum, probably consistent with an inability of these cells to complete septation. FtsZ is a tubulin homologue that forms a Z-ring at the midcell that is important for scaffolding and cytokinesis [29]. FtsA, DivIC and FtsL also localize to the midcell (at different stages) during cell division [29]. Repression of gbpB (glucan-binding protein B), an ortholog of the S. pneumoniae peptidoglycan hydrolase gene, pcsB, led to cell division defects. Although specific peptidoglycan hydrolase activity has not been observed for GbpB, the cell division defects we observed, along with the observation that GbpB can be localized to areas of active peptidoglycan synthesis [30], provide support for the idea that it may act enzymatically to affect the structure of the cell wall. Taken together, microscopic analysis of CRISPRi strains from different pathways has revealed generalized phenotypes of essential gene depletion, but also clear functional links between genes in established pathways.

Next we wanted to employ CRISPRi as a tool to study poorly annotated essential genes in S. mutans. The gene SMu.471 and the intergenic region downstream of it were of interest to us as we noticed that this region did not tolerate transposon insertions (Fig 6A). After conducting BLASTn analysis of the intergenic region, we realized that the RNA component of RNase P, rnpB, is located here. The rnpB transcript, as measured by RNA-seq, was highly abundant, which is consistent with what has been observed in other bacteria [31]. RNase P is a ribonucleoprotein that catalyzes the maturation of the 5’ end of pre-tRNAs [32]. To characterize the gene SMu.471, we began with CRISPRi phenotyping, which showed that depletion leads to a strong growth defect (Fig 6B). Depletion of SMu.471 also led to cell morphology defects (Fig 6D). Cells were wider and larger than wild-type and some appeared to not be correctly dividing (Fig 6D). Although annotated as a hypothetical protein, BLASTp analysis showed that SMu.471 contains a GpsB/DivIVA domain. SMu.471 aligned best with GpsB proteins from other Gram-positive organisms (S14 Fig in S1 File), consistent with it being smaller (112 aa) than true DivIVA proteins. Of note, the S. mutans UA159 genome contains three proteins with DivIVA domains, SMu.471, SMu.557, and SMu.1260c (S14 Fig in S1 File). SMu.557 (271 aa) is larger than SMu.471, whereas SMu.1260c is the smallest protein of the three (59 aa) (S14 Fig in S1 File). GpsB proteins are smaller than DivIVA proteins because of a reduction in the length of the C-terminal domain (CTD). GpsB has been shown to interact with a diverse assortment of proteins involved in cell division, with penicillin-binding protein (PBP) interactions (predominately class A) being most common [33]. We tested the interaction of GpsB with PBPs, as well as with the cell elongation protein MreC [34–36], or the serine/threonine protein kinase PknB [34], using a bacterial-two-hybrid assay. We observed self-interaction of GpsB and a weak interaction with PBP1a, but no interaction was evident for either MreC or PknB (Fig 6E). Collectively, the CRISPRi growth defect, cell morphology changes and the interaction with PBP1a suggest that GpsB is a critical cell division protein in S. mutans.

Having shown that CRISPRi allows for the functional study of essential genes, we next focused our attention on a pathway that is thought to be of major importance to S. mutans, rhamnose-glucose polysaccharide (RGP) biosynthesis. Certain streptococcal species lack cell wall teichoic acids (WTAs) and instead produce RGPs that are covalently linked to the peptidoglycan [37]. RGP biosynthesis begins with a four-step pathway encoded by the genes rmlABCD that converts glucose-1-phosphate into dTDP-L-rhamnose. The last gene in the pathway, rmlD, resides in an operon (SMu.824-835) that contains genes required for additional RGP biosynthesis steps and transport of the carbohydrates through the cell membrane. By our Tn-seq analysis, most of the genes involved in producing RGP are either essential or required for the growth of S. mutans UA159 (Fig 7A) [10]. L-rhamnose biosynthetic pathways are also essential in S. pyogenes [38] and Mycobacterium smegmatis [39]. Several of these genes have, however, been mutated in S. mutans, including rmlBC [40], rmlD [38], rgpE [30,41], rgpF [41,42], and rgpG [43]. Inconsistencies between our observations and others could be related to the nature of our Tn-seq selection and/or the presence of undetected suppressor mutations that bypass essentiality in strains that carry RGP gene deletions in other studies. Use of CRISPRi to study RGP function facilitates the rapid evaluation of depleting each gene product under conditions that allow for selective pressure to be applied only when xylose is provided in the medium. By CRISPRi, down-regulation of every gene directly involved in RGP biosynthesis displayed severe growth defects when silenced with sgRNAs (Fig 7B; S13 Fig in S1 File). In the invertebrate infection model organism Galleria mellonella, strains with silenced rmlB, rgpC, and rgpF genes displayed attenuated virulence compared to wild-type S. mutans UA159 and a dcas9 strain (Fig 7C). For this proof-of-concept study, sgRNAs were expressed in an S. mutans strain where the endogenous Cas9 had been converted to dCas9. Using this approach, no xylose is required and instead genes are repressed by the constitutively produced dCas9 (Fig 1C). A strain with silenced rmlD was not attenuated. However, we suspect this strain acquired suppressor mutation(s) as it originally grew very poorly, but after some passage it was able to grow to a final OD₆₀₀ similar to the parental strain. Having shown that RGP biosynthesis genes are required for growth in laboratory media and during infection, we next investigated their importance to S. mutans cell biology. Consistent with previous observations [30,43], cellular morphologies were exceptionally aberrant, compared to the wild type (S13 Fig in S1 File). Cells were large, not consistently ovococcoid, and formed longer chains than normal. Although resolutions were limiting, it also appeared that certain large cells contained multiple nucleoids by DAPI strain (S15 Fig in S1 File). To gain greater ultrastructural insights, we investigated cell biology phenotypes of rmlB, rmlD, rgpC, and rgpF silenced strains with transmission electron microscopy (TEM). In these strains, division site positioning was irregular and not always in the mid-point of cells (Fig 7D). There were instances where the division septum was unable to traverse the cell (rmlB, rmlD, rgpF) or cells had double septa (rgpF) (Fig 7D). Accordingly, large cells with incomplete septation appeared to have multiple unequally sized compartments, which likely has a deleterious effect on chromosome segregation, and might explain the unusual patterns of DAPI staining observed during fluorescence microscopy (S15 Fig in S1 File). In rgpC-silenced cells, the equatorial ring, an outgrowth of cell wall material that is normally duplicated and separates as cells divide, was abnormally thick and in some cases was ‘peeling’ back from the cell (Fig 7D). Measurements from the CRISPRi strain micrographs using ImageJ confirmed that they were wider than wild-type cells (Fig 7E).

Strains of Streptococcus mutans were cultured in Brain Heart Infusion (BHI) broth (Difco) or chemically defined medium (FMC) [56] supplemented with 0.5% maltose. Streptococci were grown in a 5% CO₂ aerobic environment at 37°C, unless stated otherwise. Strains of Escherichia coli were routinely grown in lysogeny broth (LB) with slight modifications (Lennox LB; 10 g/L tryptone, 5 g/L yeast extract and 5 g/L NaCl) at 37°C with aeration. Antibiotics were added to growth media at the following concentrations: kanamycin (1.0 mg/ml for S. mutans, 50 μg/ml for E. coli), erythromycin (10 μg/ml for S. mutans, 300 μg/ml for E. coli), spectinomycin (1.0 mg/ml for S. mutans, 50 μg/ml for E. coli) and ampicillin (100 μg/ml for E. coli). Strains and plasmids are listed in S3 Table.

The 477 S. mutans genomes available in the National Center for Biotechnology Information (NCBI) GenBank in June 2018 were mined using CRISPRone [57] to determine the occurrence and diversity of CRISPR-Cas systems in this species. CRISPR arrays were also identified using CRISPRone and these were used to predict the PAM sequence based on spacer-protospacer matches using CRISPRTarget [58]. Afterwards, consensus PAM sequences for each strain were built using the WebLogo server [59].

Xylose-inducible plasmids were cloned by following a protocol developed by Xie et al [23]. Three plasmids were generated that contained gfp, dcas9_Smu and dcas9_Spy inserts cloned into the pZX9 plasmid. For each plasmid, two linear fragments were created by PCR using pZX9 or S. mutans UA159 genomic DNA as templates. dcas9_Spy was amplified from a plasmid obtained from Addgene (Addgene #44249). dcas9_Smu was amplified from the strain described in the supplemental material. The gfp gene was amplified from previously constructed gfp-reporter strains [22]. The PCR fragments are designed so that they have overlapping complementary regions that facilitate concatemer formation during a 2^nd PCR. The products generated by this 2^nd PCR are transformed directly into S. mutans. Primers pZX9seqF and pZX9seqR were used to amplify gene inserts followed by Sanger sequencing. For cloning of the sgRNA plasmid (pPM::sgRNA), a synthetic DNA fragment was generated (Integrated DNA Technologies) that contained the following components sequentially: 1) a P_veg promoter [22], 2) a 20-nt sgRNA targeting gfp, 3) a 42-nt dCas9 binding handle sequence, and 4) an S. pyogenes terminator. Restriction enzyme recognition sequences for SacI and SphI were included at the 5’ and 3’ ends of the sgRNA construct respectively. The synthetic DNA fragment was subsequently cloned into the pPMZ plasmid which integrates at the phnA-mtlA locus of S. mutans UA159. This plasmid was also used to generate plasmids containing sgRNAs that target other areas of the S. mutans genome. Protocols for cloning of additional strains are provided in S1 File. Oligonucleotides used in this study are listed in S4 Table.

To test the CRISPRi system and compare the activity of dCas9_Smu to dCas9_Spy we used a GFP based assay. CRISPRi strains were cultured overnight, grown to an OD₆₀₀ = 0.5, and then diluted 1:100 into FMC-maltose with or without xylose. For these assays the concentration of xylose was serially diluted. In triplicates, 200 μL samples were added to clear-bottomed, black-sided, 96-well microtiter plates. After loading of the cultures, the 96-well plate was placed in a Synergy HT microtiter plate reader (BioTek). The fluorescence and optical density (absorbance at 600 nm) were measured at 30-min intervals using Gen5 software (BioTek). The fluorescence settings were excitation at 485 nm, emission at 525 nm, and a sensitivity of 65. Data readings were collected, and the background fluorescence or OD600 was subtracted prior to data visualization using GraphPad Prism (v7) software (GraphPad Software). A slightly modified version of this assay was used for PAM comparisons (S1 File). The protocol for Western blotting of GFP and dCas9_Spy is available in S1 File.

sgRNA sequences (20-nt) were selected with CRISPy-web [60]. With this tool we uploaded the S. mutans UA159 GenBank file (NC_004350.2↗) and selected essential genes. For each gene, the tool designs sgRNAs with predicted off-target matches (13-bp upstream of the PAM; off-targets containing 0, 1, 2, or 3-bp mismatches). To minimize off-target effects we only selected sgRNAs with two or more mismatches, which we have shown to diminish sgRNA activity (Fig 2B). We also aimed to select sgRNA sequences as close to the start of each gene as possible. For each sgRNA we predicted the secondary structure (20-nt sequence and 42-nt dCas9 binding handle) with the RNAfold program provided by ViennaRNA [61]. Only sgRNAs that maintained the hairpin structure of the dCas9 binding handle were selected for cloning. sgRNA sequences were cloned using the Q5 Site-Directed Mutagenesis Kit (New England Biolabs). Primers were designed using the NEBaseChanger tool (v.1.2.9), substituting the sgRNA-gfp 20-nt sequence for whichever sgRNA we were cloning. Using the designed primers, the pPM::sgRNA plasmid was amplified and then added to a mix of Kinase-Ligase-DpnI for 5 min at room temperature. This mixture (5 μL) was then transformed into chemically competent E. coli 10-beta with selection on LB-kanamycin. For each sgRNA plasmid transformation, two independent clones were selected and cultured overnight. With each culture we made a glycerol stock for storage at -80°C, and purified plasmid with the QIAprep Spin Miniprep Kit (Qiagen). Each plasmid was Sanger sequenced using the pRCS1seqR primer (S4 Table).

sgRNA sequence specificity was tested using a plasmid interference assay. A 20-nt sequence complementary to the template strand of the atpB gene (SMu.1528) was introduced into the pPM::sgRNA plasmid with the Q5 Site-Directed Mutagenesis Kit. Using the same kit, single, double, or triple base-pair mismatches were introduced into the 20-nt atpB sgRNA sequence. These plasmids were transformed into competent S. mutans and cells were serially diluted onto BHI agar, or BHI agar containing kanamycin. The percentage of cells that were transformed was calculated from the colony forming units present on both types of agar.

Strains of S. mutans carrying sgRNA cassettes were grown overnight in BHI with the appropriate antibiotics in a microaerophilic environment at 37°C. The following day the bacterial cultures were washed once in FMC-maltose, and then diluted 1:1000 into FMC-maltose or FMC-maltose containing 0.025% xylose. Triplicates of 300 μL were added to a Bioscreen C 10 x 10 well plate for each condition. OD₆₀₀ was measured every 30 min for 16 h with a Bioscreen C Automated Microbiology Growth Curve Analysis System (Growth Curves USA).

Overnight cultures of CRISPRi strains were diluted 1:100 into 200 μL FMC-maltose containing 0.025% xylose, and then incubated at 37°C in a 5% CO₂ incubator for 16 h. After incubation, 1 μL of 1 mg/mL nile red (ThermoFisher Scientific) was added to each sample and incubated for 5 min at room temperature (RT) in darkness. Next, 1 μL of 1 mg/mL DAPI (4′,6-diamidino-2-phenylindole; ThermoFisher Scientific) was added to each sample and incubated at RT in darkness for a further 5 min. Samples were centrifuged for 5 min at 4,000 x g and suspended in ProLong Live Antifade Reagent (ThermoFisher Scientific). For imaging, 2.5 μL of stained cells were placed on PBS (pH 7.4; 137 mM NaCl, 2.7 mM KCl, 8 mM Na₂HPO₄, 2 mM KH₂PO₄) agarose pads [62] before sealing with a coverslip. Images were captured with a Zeiss Axiovert 200M microscope equipped with a Zeiss AxioCam MRm camera using a 100X oil objective.

CRISPRi strains were grown as described for fluorescence microscopy. Afterwards cells were rinsed with 0.1M sodium cacodylate buffer and then fixed in 3% glutaraldehyde overnight at 4˚C. The following day, cells were treated with 1.5% osmium tetroxide in the dark for 1 h at 4˚C. Next the cells were mixed in equal parts with 5% agarose in PBS, collected by centrifugation at 2000 x g, and cooled to 4°C. Small chunks of the bacterial pellet plus agarose were then dehydrated in ethanol via the following steps: 30%, 50%, 70%, 80%, 90% each for 15 min, 99% 10 min, and then absolute ethanol 2 x 10 min. The dehydrated cells were embedded in an epoxy resin, sectioned and stained with uranyl acetate and lead citrate. Microscopy was conducted using a Hitachi H7600 transmission electron microscope.

Galleria mellonella killing assays were performed as described previously [63]. Insects in the final instar larval stage were purchased from Vanderhorst, Inc. (St. Marys, OH), stored at 4°C in the dark and used within 7 days of shipment. Groups of 15 larvae, ranging in weight from 200 to 300 mg and with no signs of melanization, were selected at random for subsequent infection. Insects were injected with 1x10⁷ CFU of bacterial inoculum, prepared from overnight cultures that had been washed and resuspended in 0.9% saline solution, into the hemocoel of each larva via the last left proleg using a 25-μl Hamilton syringe. Fifteen larvae were injected for each condition. The number of viable cells injected into the larvae was verified by colony counts on BHI plates. As a negative control, one group of G. mellonella were injected with UA159 that was heat killed at 80°C for 30 minutes. After injection, insects were kept at 37°C and survival at different time intervals was recorded. Larvae were scored as dead when thigmotaxis was absent. Kaplan-Meier killing curves were plotted, and estimations of differences in survival were compared using a log rank test. All data were analyzed with GraphPad Prism, version 7.0c software. Experiments were repeated three times.