Methylation by multiple type I restriction modification systems avoids influencing gene regulation in uropathogenic Escherichia coli

DNA methylation is a common epigenetic mark that influences transcriptional regulation, and therefore cellular phenotype, across all domains of life, extending also to bacterial virulence. Both orphan methyltransferases and those from restriction modification systems (RMSs) have been co-opted to regulate virulence epigenetically in many bacteria. However, the potential regulatory role of DNA methylation mediated by archetypal Type I systems in Escherichia coli has never been studied. We demonstrated that removal of DNA methylated mediated by three different Escherichia coli Type I RMSs in three distinct E. coli strains had no detectable effect on gene expression or growth in a screen of 1190 conditions. Additionally, deletion of the Type I RMS EcoUTI in UTI89, a prototypical cystitis strain of E. coli, which led to loss of methylation at >750 sites across the genome, had no detectable effect on virulence in a murine model of ascending urinary tract infection (UTI). Finally, introduction of two heterologous Type I RMSs into UTI89 also resulted in no detectable change in gene expression or growth phenotypes. These results stand in sharp contrast with many reports of RMSs regulating gene expression in other bacteria, leading us to propose the concept of “regulation avoidance” for these E. coli Type I RMSs. We hypothesize that regulation avoidance is a consequence of evolutionary adaptation of both the RMSs and the E. coli genome. Our results provide a clear and (currently) rare example of regulation avoidance for Type I RMSs in multiple strains of E. coli, further study of which may provide deeper insights into the evolution of gene regulation and horizontal gene transfer. Author summary DNA methylation is perhaps the most common epigenetic modification, and it is commonly associated with gene regulation (in nearly all organisms) and virulence (particularly well studied in bacteria). Regarding bacterial virulence, the current DNA methylation literature has focused primarily on orphan methyltransferases or phasevariable restriction modification systems (RMSs). Interestingly, no reports have studied the potential regulatory role of the first RMS discovered, the Type I RMS EcoKI. We used transcriptomics, Phenotype Microarrays, and a murine model of urinary tract infection to screen for functional consequences due to Type I methylation in three unrelated strains of E. coli. Remarkably, we found zero evidence for any epigenetic regulation mediated by these Type I RMSs. Thus, these Type I RMSs appear to function exclusively in host defense against incoming DNA (the canonical function of RMSs), while the methylation status of many hundreds of the corresponding recognition sites has no detectable impact on gene expression or any phenotypes. This led us to the concept of “regulation avoidance” by such DNA methyltransferases, which contrasts with the current literature on bacterial epigenetics. Our study hints at the existence of an entire class of regulation avoidant systems, which provides new perspectives on methylation-mediated gene regulation and bacterial genome evolution.


21
DNA methylation is a common epigenetic mark that influences transcriptional regulation, and 22 therefore cellular phenotype, across all domains of life, extending also to bacterial virulence. 23 Both orphan methyltransferases and those from restriction modification systems (RMSs) 24 have been co-opted to regulate virulence epigenetically in many bacteria. However, the 25 potential regulatory role of DNA methylation mediated by archetypal Type I systems in 26 Escherichia coli has never been studied. We demonstrated that removal of DNA methylated 27 mediated by three different Escherichia coli Type I RMSs in three distinct E. coli strains had 28 no detectable effect on gene expression or growth in a screen of 1190 conditions. 29 Additionally, deletion of the Type I RMS EcoUTI in UTI89, a prototypical cystitis strain of E. 30 coli, which led to loss of methylation at >750 sites across the genome, had no detectable 31 effect on virulence in a murine model of ascending urinary tract infection (UTI). Finally, 32 introduction of two heterologous Type I RMSs into UTI89 also resulted in no detectable 33 change in gene expression or growth phenotypes. These results stand in sharp contrast with 34 many reports of RMSs regulating gene expression in other bacteria, leading us to propose 35 the concept of "regulation avoidance" for these E. coli Type I RMSs. We hypothesize that 36 regulation avoidance is a consequence of evolutionary adaptation of both the RMSs and the 37 E. coli genome. Our results provide a clear and (currently) rare example of regulation 38 avoidance for Type I RMSs in multiple strains of E. coli, further study of which may provide 39 deeper insights into the evolution of gene regulation and horizontal gene transfer. thus allows a clonal population to generate phenotypic heterogeneity, which can serve 70 specific biological functions (6,7). Particularly for pathogens (and especially facultative 71 pathogens), epigenetics is one way to implement rapid responses to new environments such 72 as different host niches and immune pressures, which in turn may enhance survival and 73 virulence (8,9). 74 DNA methylation represents the most extensively studied epigenetic mechanism in all 75 domains of life (2,10-12). In eukaryotes, DNA methylation has been demonstrated to play an 76 important role in differentiation, development and disease (including cancer) (13,14). 77 Bacterial DNA methylation, in particular, has a special place in the collection of epigenetic 78 implementations, as its role "outside genetics" in mediating phage resistance was discovered 79 and characterized prior to the modern usage of the term epigenetics (15). 80 Bacterial DNA methylation occurs most commonly on adenine (N6-methyladenine, 6mA), but 81 also on cytosine (C5-methylcytosine, 5mC and N4-methylcytosine, 4mC) nucleotides (16). 82 The majority of known bacterial DNA methyltransferases belong to two general categories, 83 restriction modification system (RMS)-associated and orphan methyltransferases (4). RMSs 84 have been characterized as bacterial innate immune systems, using methylation of specific 85 motifs to differentiate between self and non-self DNA (15). RMSs can be found in 90% of all 86 sequenced prokaryotic genomes, with 80% possessing multiple systems, hinting at a diverse 87 reservoir of potential epigenetic information (11). RMSs are classified based on co-factor 88 requirement, subunit composition, cleavage pattern, and mechanism into 4 types (I to IV) 89 (17-20). Type I RMSs were the first discovered and are multi-subunit enzyme complexes 90 consisting of three proteins: a methyltransferase (typically denoted HsdM), an endonuclease 91 (HsdR), and a sequence recognition protein (HsdS). Type I RMSs recognise bipartite 92 sequences (for example, 5'-AAC(N 6 )GTGC-3', where N=A,T,G or C) with each HsdS subunit 93 possessing two highly variable target recognition domains (TRDs) which specify the two 94 halves of the bipartite motif (19). 95 Beyond immunity, RMS-associated methyltransferases are also known to regulate gene 96 expression. A dramatic example of such regulation is found in phasevariable Type I and 97 Type III RMSs. Phase variation of these RMSs results in rapid and reversible changes to 98 Therefore, DNA methylation can generally affect both gene expression and phenotypes, of 125 which virulence is an especially interesting case in pathogens. Interestingly, there has been 126 a relative dearth of such studies on the archetypal Type I RMSs, which are typically non-127 phase variable and retain both restriction and modification functions. We used single 128 molecule sequencing to characterize the Type I RMS in a uropathogenic Escherichia coli 129 strain, UTI89. We verified that this Type I RMS was functional for restriction of incoming 130 transformed DNA. Surprisingly, we could find no effect on gene expression or virulence 131 when this RMS was removed, despite >700 DNA bases changing in methylation state. 132 Similar results were found when we removed the Type I RMS from two other E. coli strains, 133 MG1655 and CFT073. Intrigued by these finding, we replaced the native UTI89 Type I RMS 134 with the Type I RMS from MG1655 and CFT073; in both cases, re-installing methylation at 135 ~400-600 distinct sites in the genome also had no effect on gene expression or on a broad 136 panel of bacterial growth phenotypes. These data suggest that, in E. coli, these canonical 137 Type I systems are purely host defense mechanisms devoid of any secondary regulatory 138 functions in host physiology and virulence. This is the first reported example of "regulation 139 avoidance" for any DNA methylation system in bacteria. The fact that three distinct 140 methylation specificities all demonstrate this regulation avoidance raises further speculations 141 about the evolution and plasticity of both the genome and the Type I RMSs in E. coli. 142

143
Uropathogenic Escherichia coli UTI89 possesses a functional Type I

145
The clinical cystitis strain UTI89 encodes a Type I RMS at the hsdSMR locus (59,60). 146 PacBio single molecule real time (SMRT) sequencing identified 3 methylation motifs in 147 UTI89: 5'-GATC-3', 5'-CCWGG-3', and 5'-CCA(N 7 )CTTC-3' (S1 Table). The first two motifs 148 are well known in E. coli as the methylation sites of orphan methyltransferases Dam and 149 Dcm, respectively (12,61). The third motif, a bipartite N6-methyladenine (m6A) motif, is 150 typical for a Type I RMS (19). Deletion of the hsdSMR locus in UTI89 resulted in loss of 151 adenine methylation only at the 5'-CCA(N 7 )CTTC-3' motif, confirming the specificity of this 152 RMS (S1 Table). 153 To verify that the UTI89 hsdSMR locus was functional for restriction of incoming DNA, we 154 performed a classic plasmid transformation efficiency assay (62). A plasmid containing the 155 5'-CCA(N 7 )CTTC-3' motif was isolated from the UTI89ΔhsdSMR strain (which does not 156 methylate this motif); this plasmid was transformed less efficiently into wild type (wt) UTI89 157 than into an otherwise isogenic UTI89ΔhsdSMR strain ( Fig 1A). Moreover, as seen with 158 other Type I RMSs, addition of a second motif further reduced the efficiency of 159 transformation into wt UTI89 (Type I RMSs cleave DNA when translocation is stalled due to 160 collision between adjacent Type I complexes (63)). When plasmids were isolated from wt 161 UTI89 (which methylates the 5'-CCA(N 7 )CTTC-3' motif), no difference in efficiency was seen 162 between transformations into UTI89 and UTI89ΔhsdSMR ( Fig 1A). Thus, uropathogenic E. shown). We then tested virulence using a transurethral murine model for ascending urinary 173 tract infection (UTI) (66). Competitive co-infections in 6-8 week old C3H/HeN mice using 174 equal mixtures of wild type UTI89 and UTI89ΔhsdSMR showed no competitive advantage 175 for either strain at 1 or 7 days post-infection (dpi) in either bladders or kidneys (Fig 1B). 176 Comparison of bacterial loads from single infections also showed no significant difference 177 between the two strains (S1 Fig). toxins, antibiotics, inhibitors, and other conditions. Relative to wt UTI89, the UTI89ΔhsdSMR 198 strain had gained resistance to the aminoglycoside antibiotic paromomycin (PM12, wells 199 C01-02), which was consistent with the use of a kanamycin resistance cassette to knock out 200 the hsdSMR locus. (Fig 1D, S4 Table). UTI89ΔhsdSMR had also gained resistance to the 201 formazan dye Iodonitrotetrazolium (INT) violet (PM19, well D07). There are four wells 202 containing different concentrations of INT on PM19 (D05 -D08); well D07 is the second-203 lowest concentration, and the other three wells were not called as different from wt UTI89. In 204 particular, at the two highest concentrations (wells D05 and D06) there was very little 205 difference in the growth curves between UTI89 and UTI89ΔhsdSMR. We therefore suspect 206 that this was a false positive phenotype (Fig 1D, S4 Table). We thus find, strikingly, no 207 phenotypic difference attributable to the loss of methylation. 208 Loss of Type I methylation in two other E. coli strains also has no effect on 209 gene expression or growth phenotypes 210 We next asked whether the lack of any detectable gene expression or growth phenotype 211 changes was unique to UTI89 and the EcoUTI methylation system. E. coli strain MG1655 is 212 a well-studied, lab-adapted K12 strain, while CFT073 is a commonly used pyelonephritis 213 strain; both also carry a single Type I RMS (EcoKI and EcoCFTI, respectively). EcoKI 214 methylates the 5'-AAC(N 6 )GTGC-3' motif, while EcoCFTI is predicted to methylate the 5'-215 GAG(N 7 )GTCA-3' motif (59,73). Plasmid transformation efficiency assays confirmed that 216 both EcoKI and EcoCFTI also were functional as restriction systems for incoming DNA, prophage genes when tested by qRT-PCR; we therefore consider the expression changes in 224 these four genes an artefact of that single strain, possibly due to a second site mutation 225 during generation of the knockout (data not shown). The CFT073ΔhsdSMR strain had 226 limited changes in gene expression relative to the wild type CFT073, including the hsdSMR 227 genes, and flanking gene yjiW (polar effect). Targeted qRT-PCR on malK and lamB genes 228 was unable to validate the changes seen in RNA-seq, confirming these as false positives 229 ( Fig 2B, S4B Fig, S2 and S3 Tables). 230 Using the Biolog PM, we again found no phenotypic differences between either (i) MG1655 231 and MG1655ΔhsdSMR or (ii) CFT073 and CFT073ΔhsdSMR; except for gained antibiotic 232 resistance due to kanamycin resistance cassette used to knock out the hsdSMR locus (S4 233 Switching Type I methylation systems in UTI89 also does not affect gene 237 expression or any growth phenotypes 238 Type I RMSs can be highly polymorphic among different strains of the same species. In 239 many cases, this is thought to be due to both mutation and recombination, possibly driven by 240 diversifying selection (74,75). Furthermore, the whole hsdSMR locus need not be 241 recombined; Type I RMSs are classified into 5 families (designated A through E) based on 242 the similarity of the hsdM and hsdR genes, enabling intra-family genetic complementation 243 with divergent hsdS genes (19). As hsdS encodes the specificity determinant, mutation or 244 recombination of just the hsdS gene is sufficient to alter the methylation and restriction 245 specificity of the entire system (35,76). The RMSs in UTI89 and MG1655 are both 246 subclassified as Type IA and demonstrate this latter relationship; the hsdM and hsdR genes 247 in these two strains have very similar sequences (99% identical), while the hsdS genes are 248 only 45.4% identical and direct distinct specificities ( Fig 3A). The Type I RMS in CFT073, on 249 the other hand, is a Type IB RMS with <40% identity to the Type IA system in all three genes 250 ( Fig 3A). 251 To analyze the effect of changing Type I methylation specificity, we created two derivatives 252 of UTI89: one where the hsdS gene was replaced by the MG1655 hsdS allele (UTI89 253 hsdS MG1655 ), and one where the entire hsdSMR locus was replaced by the EcoCFTI locus 254 (UTI89 hsdSMR CFT073 ). As a control, we also re-inserted the UTI89 hsdS allele (UTI89 255 hsdS UTI89 ) using the same cloning strategy as that used to make UTI89 hsdS MG1655 (i.e. 256 UTI89 hsdS MG1655 and UTI89 hsdS UTI89 share the same parental strains and have undergone 257 the same cloning steps). PacBio SMRT sequencing confirmed that plasmids isolated from 258 each of these strains methylated only the expected motif (data not shown). Plasmid 259 transformation efficiency assays also showed that all 3 "methylation-switch" strains 260 generated indeed had functional Type I RMSs with specificities as expected based on the 261 encoded hsdS allele (i.e. UTI89 hsdS MG1655 , UTI89 hsdSMR CFT073 , and UTI89 hsdS UTI89 262 restricted only plasmids carrying the EcoKI, EcoCFTI, and EcoUTI motif respectively, in a 263 methylation-dependent manner) (Fig 3B). 264 These strains introduce two different Type I methylation specificities in the UTI89 genome, 265 accounting for 641 (MG1655) and 423 (CFT073) predicted newly methylated sites (Fig 4A,  266 S5 Table). All together, the three RMSs we examined account for 1818 distinct methylation 267 sites that would vary in methylation status among the strains we studied. The number of 268 intergenic methylation sites in UTI89, which are expected to be more likely to affect gene 269 regulation (46,77-80), are comparable for the three Type I RMSs (31-40 intergenic sites) (S5 270   Table). Using RNA-seq, we again found no gene expression changes for any of the 271 "methylation-switch" strains compared with the parental wt UTI89, except for one 272 hypothetical gene that was close to both the FDR and fold-change cutoffs (Fig 4C and 4D between UTI89 and UTI89ΔhsdSMR was a false positive ( Fig 1D, S4 Table). 278 As a final functional test, we used several in vitro and in vivo assays for virulence. Changing 279 the methylation system had no effect on growth rate in rich or minimal media or on virulence-280 related assays such as motility and biofilm formation (S7 -S9 Figs). Competitive co-281 infections with UTI89 and UTI89 hsdS MG1655 again revealed no competitive advantage for 282 either strain, irrespective of the time-point or organ tested (Fig 4B). 283 To test whether UTI89 was unique in its ability to tolerate changes in methylation, we 284 created a similar "methylation-switch" in MG1655 by inserting the UTI89 hsdS allele. We 285 again found no differences between wt MG1655 and MG1655 hsdS UTI89 in the Biolog PM 286 phenotype screen (S4 Table,

291
The EcoKI restriction modification system (RMS) was identified over a half century ago due 292 to its role in the specific restriction of exogenous (phage) DNA (73). EcoKI is thus the 293 prototypical example of a RMS, but subsequent studies also demonstrated that, particularly 294 for E. coli but also for many other bacteria, this was an archetypal system as well. It was 295 thus designated "Type I" in the RMS classification that followed. As more RMSs were 296 discovered, they were initially assumed to play similar roles in host defense and horizontal 297 gene transfer (81,82). In bacteria, one of the early indications that methylation could affect 298 transcription arose from the study of regulation of RMSs themselves, some of which utilize 299 methylation as a readout of expression level, effectively a type of product-inhibition feedback 300 (83). The general observation that DNA methylation, particularly that mediated by orphan 301 methyltransferases (such as Dam or CcrM, which may themselves have evolved from 302 RMSs), could alter transcription (44,48,84), led to the discovery of a broad suite of 303 associated roles in DNA metabolism, cell physiology, and virulence (12,51,55,85-87). An 304 active research community continues to describe and characterize the additional roles that 305 DNA methylation (from both restriction and orphan systems) play in bacteria (4,39,88), some 306 of which can be very specific (67,69). We now show that, in contrast with expectations from 307 this literature, the archetypal EcoKI Type I RMS has nearly zero impact on gene regulation 308 or any phenotype among more than 1000 growth conditions tested. These results generalize 309 to two additional Type I RMSs from pathogenic strains of E. coli. We therefore conclude that 310 these Type I RMSs have a strictly limited role in the classical defense against incoming 311 foreign DNA, and the associated methylation of hundreds of adenines throughout the 312 genome specifically avoids any consequential regulation of gene expression. 313 The paradigm of "one gene, one function" made the initial discovery that restriction systems 314 could play a distinct epigenetic role in gene regulation somewhat surprising. Now, the idea 315 that methylation can affect the interaction with DNA binding proteins is quite clear (12), and 316 orphan methyltransferases were thought to have preserved DNA methylation for such non-317 defense (i.e. regulatory) functions (11). Perhaps the strongest evidence that gene regulation 318 is a biologically selected function for RMSs comes from the discovery of phasevarions, in 319 which phase variable expression (through recombination or simple sequence repeat 320 variation) leads to rapid switching between RMSs with different methylation specificities, 321 leading to differential regulation of genes that have particularly been shown to impact 322 regulation avoidance when introduced into UTI89, and the UTI89 EcoUTI RMS has 360 regulation avoidance when introduced into MG1655. We strongly suspect that regulation 361 avoidance is not limited to the specific E. coli strains and Type I systems studied here, 362 despite the fact that many RMSs have been shown to not have this property (65,67-69). 363 Therefore, future studies will be required to characterize the regulatory role of other RMSs, 364 potentially leading to general principles that may help explain when and why some RMSs 365 have regulation avoidance (possibly only in some strains). At this point, we suspect that 366 Type I systems, due to their longer recognition sequence and overall fewer methylated sites, 367 will be more likely to be regulation avoidant than other RMSs. consistent theme from such studies is that some details of the regulation can be rationalized 390 using a scenario where a new RMS is introduced into a bacterium: methylation is expressed 391 first, while restriction is expressed only with a delay, thus preventing immediate degradation 392 of the chromosome (94). This is similar to the requirements expected for introduction of a 393 new toxin-antitoxin system (requiring antitoxin expression before toxin expression) (95), and 394 further suggests that the introduction of a new RMS may be a regular occurrence. Of note, 395 we have emphasized the evolutionary time scale here to contrast with phasevarions, which 396 switch RMS specificity on a time scale much shorter than that required for genomic 397

adaptation. 398
In summary, we have conclusively demonstrated that, for 3 different strains of E. coli, the 399 Type I RMSs have very nearly zero effect on gene expression and no effect on any 400 phenotype tested besides their originally described function of defense against incoming 401 foreign DNA. We introduce the concept of "regulation avoidance" to highlight the contrast of from plasmid pKD4 or pKD3, respectively (97). Primers incorporated, at their 5'-end, 50bp of 431 homology to the genomic locus being knocked out. The resultant PCR product was 432 transformed into cells expressing λ-Red recombinase from vector pKM208, recovered at 433 37°C for 2 hours with shaking followed by 2 hours without shaking, then plated onto the 434 appropriate antibiotic at 37°C overnight. 435 Strains containing a seamless marker-free replacement of a gene with a different allele were 436 generated using a previously described negative selection strategy (98). Two successive 437 rounds of λ-red recombinase mediated homologous recombination were performed, using 438 the protocol described above, to first replace the wild type allele with a dual positive-negative 439 selection cassette (amplified from plasmids pSLC217 or pSLC246 instead of pKD4 or 440 pKD3). A second round of recombination was then performed using a PCR product with the 441 desired allele instead of a resistance cassette, and the second selection done on M9 442 supplemented with 0.2% rhamnose. Correct clones were confirmed by PCR and by 443 washed twice with sterile water followed by a final wash with sterile 10% glycerol and 456 resuspension in 1/100 of the original culture volume of 10% glycerol; these were then stored 457 at -80°C in 50 µl aliquots. Plasmids with 1 or 2 copies of the bipartite Type I motif were 458 extracted from the appropriate wild type or Type I RMS mutants to obtain methylated or 459 unmethylated preparations. Plasmids were extracted using the Hybrid-Q plasmid miniprep kit 460 (GeneAll, South Korea) according to the manufacturer's recommended protocol. Competent 461 cells were transformed with 100ng of each plasmid using 1mm electroporation cuvettes in a 462 GenePulser XCELL system, at 400 Ω resistance, 25µF capacitance, and 1700V output 463 voltage (Bio-Rad, Singapore). Cells were recovered in 1 ml of prewarmed LB at 37°C for 1 464 hour with shaking and plated on selective (chloramphenicol) and non-selective (LB) plates. 465 Transformation efficiency was calculated by dividing the cfu/ml obtained on selective by that 466 on non-selective plates per unit amount of plasmid DNA. 467

Mouse infections 468
In vivo infections were performed using a murine transurethral model of urinary tract infection 469 (66). Briefly, bacterial strains were grown in Type I pili-inducing conditions by two passages in 470 LB broth at 37°C for 24 hours without shaking; a 1:1000 dilution was made from the first to the 471 second passage. Cells were then harvested by centrifugation and resuspended in sterile cold 472 PBS to OD 600 = 1. Type I piliation for each strain was evaluated by a hemagglutination assay 473 and Type I phase assay as described previously (99). A 1:2 dilution of the PBS suspension 474 (final OD 600 = 0.5) was then used as the inoculum. 7-8 week old female C3H/HeN mice 475 (InVivos, Singapore) were anaesthetized using isoflurane and 50 µl of inoculum (OD 600 = 0.5, 476 ~1-2 x 10 7 cfu/50 µl) was transurethrally instilled into the bladder using a syringe fitted with a 477 30 gauge needle covered with a polyethylene catheter (Product #427401, Thermo fisher 478 scientific, USA). At specified times, mice were sacrificed and bladders and kidneys were 479 harvested aseptically and homogenized in 1 ml and 0.8 ml of sterile PBS, respectively. Ten-480 fold serial dilutions were plated on appropriate selective plates to quantify bacterial loads. For 481 co-infections, the inoculum consisted of a 1:1 mixture of two strains with an antibiotic 482 resistance cassette inserted at the phage HK022 attachment site attP (100), each at 1-2 x 10 7 483 CFU/50 µl; otherwise, the procedure was identical to that described for the single infections incubated for the specified times at 26°C or 37°C in a humidified chamber. Plates were washed 561 once with water and stained with Crystal Violet for 30 mins. Excess stain was removed by 562 washing thrice with water. Residual stain was then dissolved in 200 µl of 50% ethanol, with 563 care taken to avoid disturbing the biofilm. The amount of biofilm was quantified by measuring 564 OD at 590nm using a Sunrise 96 well microplate absorbance reader (Tecan, Switzerland). The 565 biofilm produced by each test strain was normalized to that of the corresponding wild type 566 strain. 567

568
Growth curves for bacterial strains were measured using the Bioscreen C instrument 569 (Bioscreen, Finland). Strains were grown to log phase (OD 600 = 0.4 -0.5) by sub-culturing 570 overnight bacterial culture 1:100 in LB at 37°C and resuspended in sterile PBS to OD 600 = 0.4. 571 5 µl of this normalized bacterial suspension was then inoculated into 145 µl of the desired 572 growth media (LB (rich) or M9 (minimal)) in triplicates. Plates were incubated at 37°C and 573 OD 600 was measured every 15 mins for 20 hours.  Recipient cells were wild type UTI89 and the isogenic ΔhsdSMR mutant, as indicated by the 591 labels below the x-axis. Unmethylated and methylated plasmid preparations were used to 592 transform each strain, as indicated by the legend at the top right. An unpaired t-test was 593 used to identify significant differences between plasmids with 0, 1 and 2 methylation sites for 594 both preparations and strains; * p <0.05, n = 3 biological replicates. Data represents mean ± 595 standard deviation (s.d.) of log transformed values. (B) Competitive index (CI) for in vivo co-596 infections with wt UTI89 and UTI89ΔhsdSMR. Bladder and kidney pairs (as indicated on the 597 x-axis) were aseptically harvested at 1 or 7 days post infection (dpi), as indicated by the 598 labels below the x-axis, and plated on appropriate selective plates for calculation of CI. A 599 Wilcoxon signed rank test was used to test for a significant difference of log CI from 0; * p 600 Each plate is represented as a 12x8 grid of growth curves (red (wt), green (ΔhsdSMR), and 609 yellow (overlap) on a gray background). Each growth curve plots growth (measured 610 colorimetrically) (y-axis) against time (x-axis). Wells representing conditions where a height 611 difference was observed between the strains in both replicates are boxed in black, and wells 612 which also have a quality score >150 are considered significant and boxed in red. n = 2 613 biological replicates. 614 The percentage identity between hsd genes from different systems is indicated; green >99% 631 and black <50%. (B) Transformation efficiency assay using plasmids bearing 0, 1, or 2 632 copies (as indicated on the x-axis) of the MG1655 (5'-AAC(N 6 )GTGC-3'), CFT073 (5'-633 GAG(N 7 )GTCA-3') and UTI89 (5'-CCA(N 7 )CTTC-3') Type I RMS motifs. Recipient cells were 634 UTI89 hsdS MG1655 , UTI89 hsdSMR CFT073 and UTI89 hsdS UTI89 , as indicated by the labels 635 below the x-axis. Unmethylated and methylated plasmid preparations were used to 636 transform each strain, as indicated by the legend at the top right. An unpaired t-test was 637 used to identify significant differences between plasmids with 0, 1 and 2 methylation sites for 638 both preparations and strains; * p <0.05, n = 3 biological replicates. Data represents mean ± 639 s.d. of log transformed values. 640