Enterococcus faecalis strains with compromised CRISPR-Cas defense emerge under antibiotic selection for a CRISPR-targeted plasmid

Enterococcus faecalis is a Gram-positive bacterium that natively colonizes the human gastrointestinal tract and opportunistically causes life-threatening infections. Multidrug-resistant (MDR) E. faecalis strains have emerged that are replete with mobile genetic elements (MGEs). Non-MDR E. faecalis strains frequently possess CRISPR-Cas systems, which reduce the frequency of mobile genetic element (MGE) acquisition. We demonstrated in previous studies that E. faecalis populations can transiently maintain both a functional CRISPR-Cas system and a CRISPR-Cas target. In this study, we used serial passage and deep sequencing to analyze these populations. In the presence of antibiotic selection for the plasmid, mutants with compromised CRISPR-Cas defense and enhanced ability to acquire a second antibiotic resistance plasmid emerged. Conversely, in the absence of selection, the plasmid was lost from wild-type E. faecalis populations, but not E. faecalis populations that lacked the cas9 gene. Our results indicate that E. faecalis CRISPR-Cas can become compromised under antibiotic selection, generating populations with enhanced abilities to undergo horizontal gene transfer. Importance Enterococcus faecalis is a leading cause of hospital-acquired infections and disseminator of antibiotic resistance plasmids among Gram-positive bacteria. We have previously shown that E. faecalis strains with an active CRISPR-Cas system can prevent plasmid acquisition and thus limit the transmission of antibiotic resistance determinants. Yet, CRISPR-Cas was not a perfect barrier. In this study, we observed populations of E. faecalis with transient co-existence of CRISPR-Cas and one of its plasmid targets. Our experimental data demonstrate that antibiotic selection results in compromised E. faecalis CRISPR-Cas function, thereby facilitating the acquisition of additional resistance plasmids by E. faecalis.


35
Enterococcus faecalis is a Gram-positive bacterium that natively colonizes the human 36 gastrointestinal tract and opportunistically causes life-threatening infections (1-4). Antibiotic 37 resistance is a major concern for treatment of these infections. E. faecalis can acquire antibiotic 38 resistance through horizontal gene transfer (HGT), mediated primarily by plasmids, transposons, 39 and integrative conjugative elements (5-7). One of the most clinically relevant forms of HGT in E.

56
The CRISPR is transcribed and processed into small RNAs called crRNAs; one crRNA plated on agar selective for transconjugants, donors, and recipients. These data were previously 87 reported and demonstrated that CRISPR-Cas significantly reduces pAM714 acquisition by T11RF 88 (17). We randomly selected 6 T11RF(pAM714) and 6 T11RFΔcas9(pAM714) transconjugants for 89 further analysis in this study. The 6 T11RF(pAM714) transconjugants are referred to hereafter as 90 WT1-WT6, and the 6 T11RFΔcas9(pAM714) transconjugants are referred to as Δ1-Δ6. The 91 transconjugant colonies were each resuspended in brain heart infusion (BHI) broth. Samples were 92 removed from the colony resuspensions for serial dilution and plating to determine the percentage 93 of erythromycin-resistant cells, and for PCR and sequencing to determine the size and sequence 94 of the CRISPR3. These data are referred to as "Day 0." Next, each transconjugant colony 95 resuspension was split equally into BHI broth and BHI broth with erythromycin and passaged daily 96 for 14 days. At every passage, broth samples were removed to determine the percentage of 97 erythromycin-resistant cells and the size (by agarose gel electrophoresis) and sequence (by 98 Illumina sequencing) of the CRISPR3 amplicon. As a control, wild-type T11RF was also passaged 99 for 14 days in BHI broth.

101
At Day 0, the WT1-WT6 populations, except for WT4, were a mix of erythromycin-sensitive 102 (primarily) and erythromycin-resistant cells ( Figure S1A). The Δ1-Δ6 populations were comprised 103 of erythromycin-resistant cells ( Figure S1A). The CRISPR3 arrays of all populations were wild-104 type size (1.76 kb) based on PCR amplification and agarose gel electrophoresis analysis of To investigate whether mutations were present outside the CRISPR3 array, we performed whole 139 genome Illumina sequencing on select populations. We observed variation in the cas9 sequence 140 of the WT1, WT2, and WT3 Day 14 populations relative to the T11RF wild-type ( Table 3). All of 141 the mutations led to nonsynonymous changes ( Table 3). Some of the sequenced populations 142 possessed variations in one or more additional genes (Table S2). No mutations were identified in 143 the S6 protospacer or the PAM region of the repB gene in pAM714. However, we did identify a 144 mutation within repB, not associated with the protospacer or PAM, in the WT2 population (Table   145 S2). We observed no evidence of Tn917 movement from pAM714 into the T11 chromosome as 146 all reads overlapping the ends of Tn917 also overlapped the pAD1 reference sequence. erythromycin. We also analyzed the wild-type T11RF control passaged in BHI broth. We achieved 153 an average of 16 million reads for our amplicons (Table S3).

155
To identify specific CRISPR3 alleles in the amplicon sequencing, we created a pool of references 156 by sliding a contiguous and non-overlapping window of 96 bp in length along CRISPR3 reference, only. After mapping, the number of mapped reads per 96-bp reference was calculated and 165 normalized by the total number of mapped reads per all references. To further evaluate the 166 normalized reads per reference, we calculated z-score using mean and standard deviation from 167 the T11RF Day 1 and Day 14 mapping result. The higher the z-score, the more abundant reads 168 per reference. The significance of z-score was calculated using t-test with a degree of freedom of 169 483. By applying a p-value cutoff of 0.05, we identified the most abundant CRISPR3 alleles in 170 each population ( Figure 2 and Table 2). Of the five T11RF transconjugant populations analyzed, 171 all other than WT1 possessed at least one significantly enriched mutant CRISPR3 allele after 14 172 days of passage with antibiotic selection, and each of those alleles lacked S6 (Table 2).

174
One drawback for our method is that it is difficult to set a p-value cutoff. Here we used canonical 175 definition of 0.05. However, the normalized reads per reference for wild-type alleles is expected 176 to be much higher than that from mutant alleles, and in fact, it is much higher in control samples predict that more mutant alleles existed in Day 14 WT transconjugant samples, which is supported 180 by an examination of the normalized reads per reference plot ( Figure S4), but we are uncertain 181 about their significance as measured by p-value. Supporting our approach, control samples 182 yielded results largely as expected. In control samples (T11RF and the representative Δcas9 183 transconjugant), only wild type alleles were significantly abundant with p-value < 0.05. However,

184
we also observed uneven distribution among different 96-bp references representing wild-type 185 alleles, which resulted in a "loss of spacer16-17" call for the Day 14 Δcas9 transconjugant at the 186 p-value of 0.057. When looking at the normalized reads per reference heatmap ( Figure S4)

196
Cas. To do this, we utilized the pheromone-responsive plasmid pCF10 (43), which is not natively 197 targeted by the T11RF CRISPR3-Cas system. We previously demonstrated that pCF10 transfer 198 into wild-type and Δcas9 T11RF strains is equivalent (17). In this study, we modified pCF10 to be 199 targeted by the T11RF CRISPR3-Cas system. We generated three derivatives of E. faecalis 200 OG1SSp(pCF10), each with an insertion of a T11RF CRISPR3 spacer (S1, S6, or S7) and

201
CRISPR3 consensus PAM in the pCF10 uvrB gene ( Figure 3A). Disruption of uvrB does not 202 impact pCF10 conjugation (17). We then compared conjugation of wild-type pCF10 and these 203 derivatives into the control T11RF population that had been passaged in BHI broth for 14 days.

204
As expected, conjugation frequencies were significantly lower for all pCF10 derivatives bearing 205 CRISPR3 targets, as compared to the wild-type pCF10 ( Figure 3B).

207
We then evaluated pCF10 transfer into the WT5 Day 14 populations. For WT5 passaged without 208 erythromycin, we observed similar results to the T11RF control ( Figure 3B), demonstrating that 209 the CRISPR3-Cas system in this population is still functional. For WT5 passaged with 210 erythromycin, conjugation frequency of only the pCF10 derivative bearing S1 was reduced relative 211 to wild-type pCF10 ( Figure 3B). This is consistent with the deletion of S6 and S7 in this population 212 ( Figure 2), and demonstrates that the sequencing data are accurate and not the result of PCR 213 amplification bias. We note that we observed a ~3 log higher conjugation frequency of pCF10 into 214 this population compared to the WT5 population passaged without erythromycin ( Figure 3B). This respectively; raw transconjugant and donor numbers for Figure 2 are presented in Dataset S1).

222
We also evaluated pCF10 transfer into the WT4 Day 14 populations. WT4 was unique among the 223 6 WT transconjugants in that it maintained erythromycin resistance at high frequencies in the 224 absence of erythromycin selection ( Figure 1A). We identified a probable loss-of-function mutation 225 in the RuvC domain-encoding region of cas9. The WT4 populations did not interfere with any of 226 the three pCF10 derivatives bearing CRISPR3 targets ( Figure 3B). This is consistent with a loss 227 of CRISPR-Cas function in the WT4 populations conferred by a loss of function mutation in cas9.

229
Analysis of E. faecalis genomes identifies a strongly supported instance of in situ CRISPR-

230
Cas compromise. Our data demonstrate that loss-of-function mutations in CRISPR-Cas arise 231 that promote plasmid acquisition (for WT4) or plasmid maintenance (for WT2-3 and WT5-6) in E. 232 faecalis. We used genome data available for E. faecalis to identify potential instances where this 233 may have occurred in nature. We focused on published MDR E. faecalis strains that possess Cas9 active sites that we previously experimentally confirmed in E. faecalis (17). We conclude 243 that CRISPR3-Cas function is likely to be compromised in this strain, which requires experimental 244 confirmation.

248
In this study we investigated the fates of E. faecalis transconjugants that acquired a CRISPR-249 targeted plasmid, using serial passage with and without antibiotic selection for the plasmid.

251
We observed that 5 of 6 wild-type T11RF(pAM714) transconjugants lost the plasmid over the 252 course of antibiotic-free serial passage, while T11RFΔcas9(pAM714) transconjugants maintained 253 the plasmid. This is indicative of active CRISPR-Cas defense in the T11RF(pAM714) 254 transconjugant populations. An important caveat is that erythromycin is a bacteriostatic antibiotic, 255 and we did not restreak transconjugant colonies before starting the serial passage experiment, 256 therefore we cannot be certain that plasmid-free T11RF surviving on the transconjugant selective 257 agar were not carried over with the initial transconjugant colonies that were picked. At least two 258 non-mutually exclusive explanations for our observations are possible. The first is that T11RF 259 cells lacking pAM714 (either carried over or sporadically arising) outcompete T11RF(pAM714) 260 cells. This competitive effect is due to the growth defect of cells that simultaneously possess 261 CRISPR-Cas and one of its plasmid targets, as we have previously reported (15,16). The second 262 is that CRISPR-Cas 'scrubs' pAM714 from T11RF over the course of antibiotic-free serial 263 passage. We point to Figure S1, where the WT4 colony stands out as containing essentially all

286
Our work occurs in the broader context of work on "self-targeting" spacers in CRISPR-Cas 287 systems, which have been studied for some time across different types of CRISPR-Cas in 288 different species (48). Self-targeting spacers typically refer to those with sequence 289 complementarity to targets within the host genome. This phenomenon was mostly observed 290 during phage infection and plasmid invasion (49), but self-targeting spacers against non-mobile 291 elements have also been reported (50). Different hypotheses have been posited for their 292 maintenance in microbial genomes, including alteration of target sites via DNA repair (51-53), 293 large deletion of the target sites (54), alteration of the PAM sequence (55), loss of spacers (16, 56) or mutations in cas genes (54). The "self-targeting" outcome is dependent on the fitness cost 295 and environmental selective pressure (57). The conflict between active CRISPR-Cas and "self-296 targeting" spacers plays a role in shaping bacterial evolution, including altering the population-297 level genetic diversity (58-60), remodeling of pathogenicity islands (61) and modulation of 298 metabolic pathways (62). Our work sought to investigate how this conflict can be resolved in E.

299
faecalis, as entry and maintenance of pAM714 into wild-type T11RF creates a "self-targeting" 300 situation that theoretically must be resolved to prevent persistent stress from DNA damage at the 301 targeted pAM714 repB site. In essence, this is the basis of CRISPR-Cas gene editing, and in a 302 previous study we took advantage of this property to implement CRISPR-Cas9 gene editing in E.

303
faecalis (15). Here, a recombination template to repair DNA damage (in this case, damage to 304 pAM714 repB) was not provided to the cells. We observed (in general) that the "self-targeting"

305
CRISPR spacer was retained and the target lost in populations passaged without antibiotic 306 selection for the target. On the other hand, CRISPR spacer and/or potentially overall CRISPR-

307
Cas function was lost, and the target retained, in populations passaged with antibiotic selection 308 for the target. In other studies, these "self-targeting" spacers have been engineered to promote 309 the loss of genomic islands and other mobile elements in bacteria (54, 61, 63). Our results 310 demonstrate that selection for the targeted element impedes this process, which is relevant to the 311 design and implementation of CRISPR-based antimicrobials that "cure" E. faecalis of antibiotic 312 resistance genes (19, 64). Our results suggest that these systems will need to be introduced and 313 utilized either pre-or post-antibiotic therapy, not during.

315
Overall, we posit that the interplay of CRISPR-Cas, plasmids, and antibiotic selection should be 316 further investigated to understand the role of CRISPR-Cas in the antibiotic resistance crisis. against pAM714 in a mouse intestinal colonization model where OG1SSp(pAM714) donors and 324 T11RF or T11RFΔcas9 recipients were introduced (18). CRISPR-Cas defense against pAM714 325 appeared robust -no T11RF(pAM714) transconjugants stably colonized the mouse intestine 326 above our limit of detection (<10 2 CFU/g feces). Conversely, T11RFΔcas9(pAM714) 327 transconjugants were stably present in most mice (up to ~10 7 CFU/g feces). The key question we 328 did not answer in that study is whether T11RF (pAM714) Table 1. E. faecalis strains were grown in Brain Heart Infusion (BHI) 336 broth or on agar plates at 37°C unless otherwise noted. Antibiotics were used for E. faecalis at 337 the following concentrations: erythromycin, 50 μg/mL; chloramphenicol, 15 μg/mL; streptomycin, 338 500 μg/mL; spectinomycin, 500 μg/mL; rifampicin, 50 μg/mL; fusidic acid, 25 μg/mL; tetracycline, 339 10 μg/mL. Escherichia coli strains used for plasmid propagation and were grown in lysogeny broth 340 (LB) broth or on agar plates at 37°C. Chloramphenicol was used at 15 μg/mL for E. coli. PCR was 341 performed using Taq (New England Biolabs) or Phusion (Fisher Scientific) polymerases. Primer 342 sequences used are in Table S1. Routine Sanger sequencing was carried out at the Generation of pCF10 derivatives. To insert the T11 CRISPR3 spacer 1 (S1), S6, and S7 347 sequences and CRISPR3 PAM (TTGTA) into pCF10, 47 bp and 39 bp single stranded DNA oligos 348 were annealed to each other to generate dsDNA with restriction enzyme overhangs for BamHI 349 and PstI. The annealed oligos were ligated into the pLT06 derivative pWH107 that includes 350 sequence from pCF10 uvrB, to insert these sequences into the uvrB gene of pCF10 by 351 homologous recombination. A knock-in protocol was performed as previously described (17).

357
After 18 h incubation, the conjugation mixture was scraped from the plate using 2 mL 1X PBS  (Table S1). The PCR products were purified using the Thermo 382 Scientific PCR purification kit (Thermo Scientific). Genomic DNA was isolated using the phenol-

396
Genomics Workbench. The basic variants and fixed ploidy variants were combined for each 397 sequencing sample and subjected to manual inspection. The variants that were detected in the T11 genome from all samples were inferred to be variants in our parent T11 stock and were 399 manually removed. The variants that were detected in pAD1 genome from all transconjugant 400 samples were inferred to be variants in our pAM714 stock, hence were also manually removed.

401
Next, variants within the CRISPR3 array were removed as we analyzed CRISPR3 alleles using a 402 different approach (amplicon deep sequencing; see below). All variants detected from all 403 populations were manually checked for coverage depth to eliminate the detection bias. The

404
variants detected in all samples are shown in Table S2. To detect the insertion site of Tn917, the 405 mapped reads on reference M11180 were inspected. Reads immediately adjacent to the 5' and is from S0 to S21 while the selection of spacer[y] is from S1 to ST. Additionally, we assume Sequencing reads were first clipped into 75 bp fragments to enhance mapping efficiency, allowing 429 for retrieval of maximal sequence information. The clipped reads were used to map to the pool of 430 96-bp references using stringent mapping parameters in CLC Genomics Workbench (Qiagen).

431
The stringent mapping parameters require 100% of each mapped read to be ≥95% identical to 432 one unique reference. Thus, the sequencing reads from different CRISPR alleles will be 433 distinguished. The number of mapped reads per each reference was calculated and normalized 434 by the total number of mapped reads to all references, generating normalized reads per reference.

435
The total number of mapped reads to pools of wild type references and mutant references were 436 summarized in Table S3.

438
To visualize the normalized reads per reference, a heatmap was generated ( Figure S4). On the 439 heatmap, each row name represents 5'-Sx (x is from 0 to 20) and each column name represents

446
To evaluate the statistical significance of normalized mapped reads, we calculated z-score using 447 the mean and standard deviation calculated from T11RF control. Combining the normalized reads 448 per reference from Day 1 and Day 14 T11RF control, we obtained an average normalized reads used to calculated z-score for normalized reads per reference for all samples. To assess the 451 significance, the z-score is further transformed into p-value. A p-value of 0.05 was used as a 452 significant cutoff.

454
To visualize the p-value and derive mutant alleles, a heatmap was plotted using the same method