The Crohn’s disease-associated Escherichia coli strain LF82 rely on SOS and stringent responses to survive, multiply and tolerate antibiotics within macrophages

Adherent Invasive Escherichia coli (AIEC) strains recovered from Crohn's disease lesions survive and multiply within macrophages. A reference strain for this pathovar, AIEC LF82, forms microcolonies within phagolysosomes, an environment that prevents commensal E. coli multiplication. Little is known about the LF82 intracellular growth status, and signals leading to macrophage intra-vacuolar multiplication. We used single-cell analysis, genetic dissection and mathematical models to monitor the growth status and cell cycle regulation of intracellular LF82. We found that within macrophages, bacteria may replicate or undergo non-growing phenotypic switches. This switch results from stringent response firing immediately after uptake by macrophages or at later stages, following genotoxic damage and SOS induction during intracellular replication. Importantly, non-growers resist treatment with various antibiotics. Thus, intracellular challenges induce AIEC LF82 phenotypic heterogeneity and non-growing bacteria that could provide a reservoir for antibiotic-tolerant bacteria responsible for relapsing infections.


Introduction 42
Adherent Invasive Escherichia coli (AIEC) strains recovered from Crohn's disease (CD) 43 lesions are able to adhere to and invade cultured intestinal epithelial cells and to survive 44 and multiply within macrophages (Darfeuille-Michaud et al, 1998;Glasser et al, 2001). 45 Attention around the potential role of AIEC in the pathophysiology of CD is growing 46 (Elhenawy et al, 2018); however much remains to be learned about the host-pathogen 47 interactions that govern AIEC infection biology. The diversity of virulence factors 48 displayed by multiple AIEC strains suggests that members of this pathovar have evolved 49 different strategies to colonize their hosts (Tawfik et al, 2014). AIEC ability to persist, 50 and in some cases replicate within macrophages is particularly intriguing. Previous 51 work performed with murine macrophage cell lines has revealed that the prototype 52 AIEC strain LF82, multiplies in a vacuole presenting the characteristics of a mature 53 phagolysosome (Bringer et al, 2006;Lapaquette et al, 2012). In such an environment, 54 AIEC should encounter acidic, oxidative, genotoxic and proteic stresses. Screening of 55 genes involved in LF82 fitness within macrophage has revealed that HtrA, DsbA, or Fis 56 proteins are required for optimum fitness, (Bringer et al., 2005;Bringer et al., 2007;57 Miquel et al., 2010). These observations confirmed that LF82 encounter stresses in the 58 phagolysosomes. The impact of these stresses on the survival and growth of LF82 inside 59 phagolysomes has not yet been investigated. 60 Studies on the bacterial cell cycle of few model organisms under well-controlled 61 laboratory conditions have revealed that to achieve accurate transmission of the genetic 62 information and optimal growth of the population, molecular processes must be 63 coordinated. (for reviews see Hajduk et al., 2016;Haeusser & Levin, 2008). When 64 growth conditions deteriorate, the cell cycle can be modified slightly, as in the case of 65 cell filamentation when genotoxic stress induces the SOS response, or more drastically 66 when sporulation is induced by nutrient deprivation (Jonas, 2014). Such cell cycle 67 alterations affect the entire population. However, under unperturbed conditions, a 68 subset of the population also appears to present a significantly reduced growth rate that 69 allows tolerance to antibiotic treatments. This small portion of the population, typically 70 1/10000 bacteria, is known as persisters (Wood et al., 2013;Lewis, 2010;Bigger, 1944). 71 Persisters have been detected for a number of bacteria. They can be found 72 spontaneously in normally growing or stationary phase populations, or they are induced 73 by exogeneous stresses or mutations. Significant increase of the proportion of S. 74 typhimurium persisters has been observed when these bacteria invade macrophages 75 (Helaine et al., 2014). Using a fluorescent reporter, it has been demonstrated that these 76 persisters were not multiplying prior to antibiotic addition. Recently, the same tool also 77 revealed the presence of non-growing mycobacteria inside macrophages (Mouton et al., 78 2016). Several mediators of persistence have been identified, with toxin-antitoxin 79 modules emerging as key players and the reduction of metabolic activities as the main 80 driver of persistence (Rycroft et  unclear, and their distinction in the context of a host-pathogen interaction is difficult 95 (Kim & Wood, 2017). 96 97 In the present work, we analyzed growth characteristics of the prototype AIEC strain 98 LF82 in THP1 monocyte-derived macrophages. We observed that stresses within 99 macrophages induce a profound bacterial response that leads to the formation of non-100 growing and antibiotic-tolerant LF82 bacteria at a high rate through the successive 101 induction of stringent and SOS responses. A portion of non-growing LF82 produced 102 within macrophages is tolerant to antibiotics and presents a survival advantage. Our 103 work revealed that internalization within phagolysosomes curbs bacterial 104 multiplication, and frequent escape from the replicative cycle toward non-growing 105 state(s) is a way to improve long-term survival in the host. 106 107 Results 108 109 Inside macrophages, LF82 population size increases despite extensive death. 110 We used THP1 monocyte-derived into macrophages to monitor the population size of 111 LF82 bacteria over a 24 h period post infection (P.I.) ( Figure 1A) Figure S1A). Using direct ex vivo Live and Dead labeling, it has been 120 previously proposed that 80% of LF82 present at 24h within macrophages were alive 121 (Lapaquette et al., 2012). We observed that this method slightly underestimates dead 122 bacteria inside macrophages because of a weak propidium iodide (PI) labeling 123 (Supplementary Figure S2A). Live and dead assay performed immediately after 124 macrophage lysis revealed a nearly constant proportion of dead LF82 in the population 125 (20 -30%) at 1 h, 12 h, 18h and 24 h post-infection ( Figure 1A). To estimate the speed of 126 dead bacteria disappearing in macrophages, we observed the elimination of heat-killed 127 bacteria by THP1 macrophages. Dead LF82 disappeared exponentially with a decay rate 128 of 0.6 h -1 and a half-life of 1.4 h (Supplementary Figure S2B); therefore dead LF82 129 observed at 12h, 18h or 24 h did not correspond to the accumulation over infection 130 period but rather to the bacteria killed in the last 3 hours before observations. This 131 finding led to consider that LF82 must be under stress attack by macrophages at all 132 times during infection. 133 134 LF82 is under attack by macrophages. 135 Using RT-qPCR, we measured the expression of genes induced by the acid (asr, ydeO and 136 frc), oxidative (soxS and ykgB), and SOS (sulA) responses, and the responses to 137 membrane alteration (pspB), the lack of Mg2+ (mgrB), the lack of phosphate (phoB), 138 general efflux pump (emrK) and the gltT tRNA gene that is repressed by the stringent 139 response ( Figure 1B). Every response pathway was induced inside the macrophage. The 140 induction of acid and the oxidative responses was already high at 1 h P.I., while the SOS 141 response, the response to membrane alterations and to the lack of Mg2+, were peaking 142 at 6 h P.I.. The expression of the gltT tRNA was strongly repressed at 1 h P.I. indicating 143 that stringent response is on early in the infection. To test the impact of stress responses on the ability of LF82 to colonize macrophage, we 148 constructed deletion mutants of several key regulators of E. coli stress pathways and 149 analyzed their survival. Deletion of the acid stress regulators evgA-evgS, phoP and ydeO 150 significantly impacted the ability of LF82 to survive and multiply within macrophages to 151 a level comparable to or even below that of a K12-C600 E. coli ( Figure 1C). Similar 152 observations were obtained with the rpoS (general stress response), recA (SOS 153 response), soxS (oxidative stress) and pspA and htrA (envelope damages) deletion 154 mutants. The ppGpp0 strain, relA spoT deletions, impaired in the stringent response 155 reporting a lack of nutrients, is the most impacted strain; less than 5% of the initial 156 population survived a 24 h period within macrophages. These observations confirmed 157 that LF82 encounter severe stresses in the macrophage environment and that its ability 158 to repair stress mediated injuries will determine survival. 159

SOS and stringent responses severely impacted LF82 survival 161
Because of their known potential to influence growth and cell cycle parameters we 162 explored the stringent and SOS responses in more details. RecA is the main inducer of 163 the SOS response, which activates nearly 100 genes involved in DNA repair and many 164 others with unrelated or unknown functions, but it is also a crucial to correct DNA 165 lesions by homologous recombination and translesion synthesis (Kreuzer, 2013). In 166 addition to recA deletion, we constructed deletions of sulA (division inhibitor) and a 167 mutation in lexA (lexAind-), which blocks SOS induction in MG1655 E. coli and reduces 168 viability in the presence of mitomycin C for LF82 and MG1655 (Supplementary Figure  169   S3A). We observed that the deletion of each SOS gene significantly decreased the 170 survival of LF82 within macrophages ( Figure 1C). Inside the macrophage, survival of the 171 ppGpp0 strain was dramatically impacted. However, this mutant also presented a strong 172 growth defect in liquid culture that complicates interpretation of the macrophage 173 results. To study the impact of the stringent response on LF82 survival and induced 174 antibiotic tolerance, we constructed deletion mutants that might have partial stringent 175 response phenotypes; deletion of dksA, encoding a protein linking the stringent 176 response to transcription (Sharma & Chatterji, 2010); and deletion of the polyphosphate 177 kinase and exopolyphosphatase ppk and ppx (Rao & Kornberg, 1999). As expected, the 178 dksA and ppk-ppx deletions had a much less dramatic effect on LF82 growth and 179 survival within macrophages than the relA-spoT mutant; nevertheless, the dksA 180 mutation significantly impacted the number of live bacteria recovered at 24 h P.I. 181 ( Figure 1C). We investigated the ability of LF82 to survive within macrophages when 182 both stringent and SOS responses were altered. We chose to combine dksA deletion with 183 recA deletion or lexAind-mutation. These strains presented a survival defect comparable 184 to that of the single dksA mutant ( Figure 1C). These observations demonstrate that 185 surviving LF82 simultaneously or successively require SOS and stringent responses.  showed that only 20% of the population has performed more than 1 division at 6 h P.I. 216 From these observations, we can estimate that the highest generation rate of LF82 217 within macrophages is ≈ 0.5 doubling /h between 6 and 20 h P.I.. Interestingly, FD also 218 revealed that approximately 4% of the population did not divide or divided fewer than 2 219 times intracellularly in 24 h ( Figure 3A and Supplementary Figure S4B). By contrast 220 among the small amount of K12-C600 bacteria that survived for 24 h in the macrophage, 221 60% of K12-C600 bacteria underwent fewer than 2 divisions and less than 10% 222 underwent 5 divisions (Supplementary Figure S4D). Second, we used TIMER to refine 223 these observations; it provides an instantaneous evaluation of the generation time 224 during the infection kinetics ( Figure 3B and S4A and S4B). TIMER indicated that at 18 h 225 P.I., 18% of the LF82 population was not actively dividing, supporting the existence of a 226 non-growing or slow-growing subpopulation. Since they require dilution of fluorescent 227 proteins both TIMER and FD are poorly informative about the first hours of the 228 infection. Therefore, we used a GFP fusion with the septal ring protein FtsZ to monitor 229 division in the individual bacterium ( Figure 3C). In LB, exponentially growing LF82 230 frequently presented the FtsZ ring (70% of the population), but stationary phase LF82 231 rarely presented the FtsZ ring (<2% of the population, Figure 3D). Following infection of 232 macrophages with the stationary phase culture of LF82 ftsZ-gfp, we observed that 5% 233 (+/-2) and 40% (+/-12) of the population, respectively, presented a FtsZ ring at 1 h and 234 24 h P.I. (Figure 3C  we observed significant increases in the proportion of green LF82 (non-growing) inside 265 macrophages ( Figure 4B). These observations suggest that the sub-population of non-266 growing bacteria largely overlaps with that of persisters, where protection from 267 antibiotics may also confer enhanced tolerance to intracellular stresses. 268 269 Tolerance to antibiotics is enhanced for LF82 compared with K12-C600 E. coli. 270 We compared the number of LF82 and a non-pathogenic K12-C600 laboratory strain 271 with tolerance to ciprofloxacin following brief (1 h) or long (24 h) passages in 272 macrophages. After a brief passage in macrophages, the proportion of LF82 that were 273 tolerant to ciprofloxacin was significantly higher for LF82 than K12-C600 ( Figure 4C). 274 Interestingly, even if the absolute number of ciprofloxacin-tolerant K12-C600 was 275 largely reduced compared with LF82, their proportions among bacteria that survived 24 276 h inside macrophages were comparable ( Figure 4C). These findings demonstrate that 277 the number of antibiotic-tolerant bacteria formed in response to macrophage attack is 278 reinforced for LF82 compared with the laboratory strain. 279 280 Tolerance to antibiotics is a transient state. 281 We next evaluated whether the antibiotic tolerance was a stable or transient phenotype. 282 We used the macrophage lysis procedure to recover LF82 with induced persistence for 1 283 h in the macrophage; then, we either challenged them immediately with ciprofloxacin or 284 allowed them to recover in LB for 1 h or 2 h before antibiotic challenge. When bacteria 285 were cultured for 1 h in LB, the frequency of tolerant bacteria was decreased in 286 comparison to bacteria that were immediately treated with the antibiotic; however, this 287 number was still higher than that of bacteria that had not infected macrophages. Two 288 hours in LB was sufficient to cause a comparable frequency of ciprofloxacin-tolerant 289 LF82 to that of bacteria that had not encountered macrophages ( Figure 4D). These 290 observations show that when the environment is no longer stressful, antibiotic-tolerant, 291 non-growing LF82 rapidly switch back to a replicative mode. 292

294
Characterization of non-growing LF82. 295 Both FD and TIMER revealed slightly more non growing LF82 (4% and 18% 296 respectively, figure 3A) than antibiotic-tolerant LF82 after macrophage lysis (0.5% at 1 297 h P.I. or 5% at 24 h P.I. , Figure 4A). This finding raised the possibility that persisters 298 only form a portion of the non-growing population. To quantify this proportion, we 299 infected macrophages with TIMER-tagged LF82, lysed the macrophages and allowed 300 bacterial growth on a LB-agarose pad under the microscope at 37°C. Seventy percent of 301 the LF82 bacteria recovered quickly from the challenge and formed microcolonies, but 302 approximately 30% of them never divided ( Figure 4E). These non-cultivable LF82 303 presented either non-growing or growing TIMER fluorescence ( Figure 4F). The presence 304 of non-cultivable LF82 among the bacteria with non-growing TIMER fluorescence 305 explains the difference between fluorescence and antibiotic assays. 306

SOS and stringent responses influence antibiotic tolerance. 308
Among mutants that affected LF82 survival ( Figure 1C), only the recA, relA spot and dksA 309 deletions negatively impacted the number of LF82 that were tolerant to a 3h 310 ciprofloxacin treatment ( Figure 5A). The impact of the recA deletion might be 311 misinterpreted because ciprofloxacin alters DNA and limits resuscitation of recA 312 persisters. Therefore, we repeated the tolerance assay with cefotaxime for the following 313 SOS mutants: recA (impaired for DNA lesion repair and SOS induction), lexAind-(unable 314 to induce SOS) and sulA (unable to block cell division). In vitro, SOS mutants did not 315 present defect for cefotaxime tolerance (Supplementary Figure S3B). However, these 316 mutants exhibited a decreased tolerance to antibiotics when persisters were induced by 317 a pretreatment with subinhibitory concentrations of ciprofloxacin (Supplementary 318 Figure S3C). This finding is in good agreement with previous reports (Dörr et al., 2009), 319 and it confirms that SOS induction favors the production of persisters. We analyzed 320 cefotaxime tolerance of these SOS mutants after a 1 h or 20 h period within 321 macrophages ( Figure 5B and 5C). We observed a significant reduction of the proportion 322 of recA and sulA mutants that were tolerant to cefotaxime treatment after a 20 h passage 323 in the macrophage ( Figure 5C) but no effect on bacteria that remained only 1 h in 324 macrophages ( Figure 5B). The lexAind-mutation did not change the number of tolerant 325 bacteria in these conditions. We also analyzed the dksA mutant in these assays; 326 surprisingly, it behaved differently than the recA and sulA mutants: we observed a 327 significant reduction of the proportion of cefotaxime tolerant bacteria following a 1 h 328 passage within macrophages ( Figure 5B Knowing that SOS and stringent responses influence the production of antibiotic 341 tolerant LF82 after a passage within macrophage we examined whether they also 342 contributed to LF82 cell cycle control, i.e. production of non-growing, replicative or dead 343 LF82. We used the FD assay to measure the number of non-growing LF82 in the relA-344 spoT and recA mutants. FD revealed that the non-grower number was dramatically 345 reduced in the relA-spoT mutant (<1%) ( Figure 5D). This suggests that in the absence of 346 stringent response LF82 cannot immediately curb its cell cycle upon phagocytosis. By 347 contrast the number of non-growers was unchanged for the recA mutant ( Figure 5E and fraction of persisters are relevant to future fundamental studies, but also to devising 370 therapeutic strategies involving AIEC or other intracellular pathogens. In order to 371 explore the mechanistic bases of the infection kinetics, we have used a mathematical 372 model ( Figure 6A) to fit the observed changes in CFU and persister counts during 24 h 373 for LF82, K12-C600 and the stringent response mutant LF82dksA ( Figure 6B). The 374 model is based on the following biologically-informed hypotheses (illustrated in Figure  375 6A and detailed in the Supplementary text 1). i) Reproduction: the population of 376 replicating bacteria B has a constant net growth rate (birth minus death rate) δ 1 , which 377 is either 0 or negative during a lag phase of duration λ, and β>0 otherwise. ii) A stress-378 induced death rate, δ2(S) that increases with stress (S). We assume that stress S, possibly   Figure 4). The most notable 394 quantitative difference between strains is that K12-C600 displayed a lag phase of more 395 than 13 h, twice as long as LF82 and LF82dksA. A consequence of this difference is that 396 when K12-C600 bacteria start actively duplicating, stress has already built up. Together 397 with K12-C600's enhanced sensitivity to stress, this curbs the population expansion, 398 resulting in a lower overall growth within macrophages. In the LF82dksA mutant, 399 growth is instead impaired by increased initial mortality (whose rate δ1 has been 400 estimated by PI measures, Supplementary Figure S2C), presumably related to stringent 401 response failure. With respect to the other strains, moreover, LF82 is advantaged at 402 later times -when the SOS response becomes important -thanks to reduced stress-403 induced death rate. Rate of persistence production for LF82 and LF82 dksA (0.08 and 404 0.002 h -1 , respectively) is estimated to be higher than for K12-C600 (0.001 h -1 ) , 405 supporting the notion that AIEC strains within macrophages turn to persisters at an 406 enhanced rate, but less so if their stringent response is impaired. The model allows 407 testing changes in infection dynamics for 'virtual mutants' LF82*, obtained by varying kp, 408 λ and dmax -the parameters that quantitatively differ between LF82 and E. coli K12-C600 409 ( Figure 6C). The total population overshoot is enhanced when lag phase is shorter and 410 the effect of stress less acute, but damped when persisters production is more frequent. Live and Dead assay indicating that LF82 progeny has a significant chance to be killed 441 and destroyed by the macrophage. By contrast, some macrophages contained growing 442 LF82 and ultimately acquired more than 50 bacteria in one or several compartments. 443 The live and dead assay confirmed that LF82 was frequently killed by macrophages. in the second phase a phenotypic switch to non-growing LF82 will eventually result in a 455 sizeable increase of the persister population. We thus understand the survival of LF82 as 456 a consequence of its ability to adapt to harsh phagolysosome environment both at the 457 entry of the macrophage, by induction of stress responses and particularly the stringent 458 response, and during exponential expansion by SOS response. LF82 advantage over K12-459 C600 would reside in its ability to exit from the lag phase to perform a few rounds of 460 replication/division before stress becomes too strong ( Figure 6). Strategically, early 461 onset of growth is compensated by production of persistent bacteria, which endows the 462 pathogenic strain with long-term survival in spite of rapid exploitation of the 463 macrophage environment. We have not yet identified LF82 specific regulons, genes or 464 mutations that allow this transition to take place.  Table S1) were constructed using the recombineering 553 method as described in (Demarre et al., 2017). Plasmids are described in Supplementary 554 Table S2.
where the half-saturation stress value S1/2 and the sensitivity parameter a are assumed 612 to be identical for all strains. Here stress is an effective variable quantifying the effect of 613 crowding on growth within macrophages, and could correspond both to density-614 dependent reduction of bacterial growth rate (e.g. due to resource depletion), and to the 615 progressive buildup of macrophage-induced killing. 616 617

Fit of the infection kinetics data 618
Parameters providing the best fit of eqs. (1) to the times series of CFUs and persisters 619 have been obtained by a weighted least-square distance minimization using the python 620 differential evolution algorithm. We used a two-step approach to the fit which allowed us to 621 establish first a subset of 7 parameters (λ and kp for each strain and β) that shape the lag 622 and exponential phases of growth. Subsequently, we fixed β and the λs, and fitted the 623 remaining parameters. Details of the fitting procedure are found in the SI, and the 624 results of the fit in Table 1