Long-term evolution of antibiotic persistence in P. aeruginosa lung infections

Pathogenic bacteria respond to antibiotic pressure with the evolution of resistance but survival can also depend on their ability to tolerate antibiotic treatment, known as persistence. While a variety of resistance mechanisms and underlying genetics are well characterised in vitro and in vivo, the evolution of persistence, and how it interacts with resistance in situ is less well understood. We assayed for persistence and resistance with three clinically relevant antibiotics: meropenem, ciprofloxacin and tobramycin, in isolates of Pseudomonas aeruginosa from chronic cystic fibrosis lung infections spanning up to forty years of evolution. We find evidence that persistence is under positive selection in the lung and that it can particularly act as an evolutionary stepping stone to resistance. However, this pattern is not universal and depends on the bacterial clone type and antibiotic used, indicating an important role for antibiotic mode of action.


Introduction 42
43 Bacterial persistence is defined as the ability of cells to adopt a phenotype that tolerate and 44 survive exposure to a bactericidal drug concentration (1-3). It contributes to recalcitrant 45 chronic infections even when infecting strains are susceptible to antibiotic treatments (1-3). 46 We must better understand how bacteria respond and adapt to drugs in situ, to be able to use 47 them effectively. Most work has focused on resistance, but while a level of consensus has 48 been reached on how to define and quantify persistence (1), the extent to which it is under 49 positive selection in infections from antibiotic treatment and how persistence co-evolves with 50 resistance is still not well understood (4). To characterise long-term evolution of resistance 51 and persistence in infection, we studied isolates of Pseudomonas aeruginosa from chronic 52 lung infections of individuals with cystic fibrosis (CF). This infection system is ideal for 53 detecting selection on persistence because long-term sampling has occurred both of lineages 54 of transmissible strains (infecting for 40 years) and within single infections (up to 7 years)(5). 55 The collection of transmissible strains in particular allows us to look at evolutionary history 56 over decades. 57 58 In contrast to the antibiotic resistance phenotype, there is little evidence for a clear genetic 59 basis or the molecular mechanisms involved for the persistence trait. A specific gene 60 responsible for persistence has been found only for Escherichia coli, while a range of 61 regulatory and metabolic genes and mechanisms that contribute to or correlate with the 62 persistence phenotype have been identified in Pseudomonas aeruginosa, as well as other 63 was calculated as [P(X  events observed ) ~ pois(X; events expected ) < 0.05], where the expected 241 number of events were half of the total number of transitions from/to a given category. 242 243 Given the four different phenotypic categories of response strategies and transitions between 244 them, we can compare the number of transitions to and from each. We define a "stable" 245 phenotype as one that has few transitions from it, and "unstable" as one that has many 246 transitions from it. To test whether a phenotype was significantly more, or less, stable than 247 expected given random transitions between phenotypes, we calculated the probability of 248 finding the observed number of transitions from a phenotype compared to the expected 249 number, given as a quarter of the total transitions from any phenotype. We analysed all 250 isolates of DK1 and DK2 together to increase the sample size, and for each antibiotic 251 separately and all together (Table S4).   After exploring the evolution of persistence to an individual antibiotic we test if there are 299 patterns in isolate response across the three antibiotics. When following clonal lineages on 300 the DK1 and DK2 phylogenies we observe changes in the persistence phenotype going from 301 either low to high or high to low persistence. In pairwise comparisons between antibiotics we 302 find that persister phenotypes in response to meropenem and ciprofloxacin treatments are 303 frequently matched, so that an isolate with high persister under meropenem treatment is also 304 a high persister to ciprofloxacin, or a low persister under both antibiotic treatments. The null 305 model would predict that 50% of transitions would lead to the same persister phenotype and 306 50% would lead to unmatched persister phenotypes. However, we find that significantly 307 more often than expected by chance, a change in persistence to ciprofloxacin lead to the same 308 persister phenotype as for meropenem, and vice versa [P(X  27) ~ pois(X; 14.5) = 0.002]). 309 The tobramycin treatment does not show that changes in the persister phenotype more often 310 than expected by chance leads to a shared phenotype with either meropenem [P(X  17) ~ 311 pois(X; 12.5) = 0.13] or ciprofloxacin [P(X  9) ~ pois(X; 12.5) = 0.88]. 312

313
We observe the independent evolution of high persistence to all three antibiotics seven times, 314 and the same for resistance ( Fig. S1A   Given that resistance and persistence are under positive selection, we next show how these 364 traits might co-evolve within transmissible lineages. We assessed what the most stable 365 phenotype is, defined as one from which there are fewer than expected transitions to other 366 phenotypes (see methods), and the route taken to get there (Fig.4). The patterns we observe 367 are distinct for the different experimental antibiotic treatments. Meropenem data shows that 368 the resistance | low persistence phenotype is significantly more stable than expected by 369 chance (LR; P(X  8.25) ~ pois(X; 1) = 0.002). Resistance in combination with persistence is 370 the stable phenotype under ciprofloxacin treatment (HR; P(X  7.25) ~ pois(X; 0) < 0.001). 371 In contrast, the response to tobramycin shows that resistance alone, persistence alone, and the 372 combination of persistence and resistance are all equally stable phenotypes while the 373 susceptible | low persistence phenotype is significantly less stable than expected by chance 374 [LS; P(X  6.5) ~ pois(X; 12) = 0.034]; Table 1

21
We find that maximum bacterial density after 24 hours of growth is negatively correlated 418 with both persistence and resistance. This indicates that resistant and persister isolates either 419 grow slower or have an extended lag and/or stationary phase. Resistant isolates that are either 420 high or low persisters consistently have lower population densities (purple and yellow boxes 421 in Fig.  422 6; Table S7). While the high persister phenotype does occur in isolates of higher density 423 (grey boxes), the high persistent | resistant (HR) phenotype significantly has a lower density 424 in the presence of all three antibiotics (purple boxes). This pattern was in particular driven by 425 the DK1 clonal lineage (Fig. S7). There was, however, also a significant decrease in 426 maximum density with infection time despite the resistant or persistent phenotypes (Linear 427 regression; DK1: R 2 adj = 0.50, p < 0.001; DK2: R 2 adj = 0.21, p < 0.001). So the negative 428 correlation between resistance and high persistence with low density could be confounded by 429 the fact that resistant isolates tend to be sampled late in infection. Analysing the data 430 including both sampling time and population density to test for their effect on the evolution 431 of resistance and persistence, shows that lower population density in itself is associated with 432 an evolved antibiotic response (except for persistence to meropenem treatment in DK2; We investigated the evolution of antibiotic persistence and resistance during pathogen 446 evolution. We find that both persistence and resistance to three clinically relevant antibiotics, 447 meropenem, ciprofloxacin and tobramycin, are selected for during infection. However, we 448 observe differences in antibiotic response strategies between clone types and type of 449 antibiotic used. While persistence and resistance are equally likely to evolve first, persistence 450 can act as a stepping stone to gain resistance. In contrast, we did not find that resistance acted 451 as a stepping stone for persistence. As such, our in situ results are in line with recent findings 452 by experimental evolution of P. aeruginosa (10) and E. coli (21,22). In our collection of 453 23 isolates, we find both strategies take time to evolve; persistence evolves de novo after 7-19 454 years and clinical resistance after 12-20 years. This suggests that other mechanisms such as 455 biofilm production or limited diffusion of antibiotics within the CF lung must play a role for 456 the establishment of pathogens despite antibiotic treatment early in infection (13,40,41). 457

458
We find that individuals are typically colonized by antibiotic susceptible and non-persistent 459 phenotypes, and that strategies to evade the effect of antibiotics are, therefore, unsurprisingly 460 favored by selection in the lung ( Fig. 1 -3). The pattern of how these traits spread over time 461 differs, however, between the two transmissible lineages we examined. One of the lineages, 462 DK2, shows a significant increase in persistence over time in response to all three antibiotics, 463 while DK1 only shows increased persistence under tobramycin treatments, and significant 464 decrease in persistence under meropenem and ciprofloxacin treatments. The differences 465 between DK1 and DK2 may reflect topology of the phylogenies and ancestral state: the DK1 466 phylogeny has several distinct subclades that evolve independently (Fig. S1A). Some clades 467 maintain "wild-type-like" production of virulence factors after more than 30 years of 468 infection (e.g. the siderophore pyoverdine (42) and protease (30) An outstanding question is how specific persistence is to the type of antibiotic that cells are 478 exposed to. We found a significant overlap in persistence phenotypes between ciprofloxacin 479 and meropenem but not with tobramycin. In contrast, a recent study (10) shows correlations 480 of the persister phenotype between ciprofloxacin and tobramycin. This may be because the 481 authors select only a subset of strains from a lineage of susceptible strains with MIC values 482 below the clinical breakpoint and use much lower doses of antibiotics relative to the MIC 483 than in our assays (approximately 10-100 fold lower). Our results may reflect that 484 ciprofloxacin and meropenem were simultaneously used in the clinic for treatment, exerting 485 comparable selection pressures. Another explanation could be that despite selection for the 486 canonical antibiotic resistance genes, exposure to one antibiotic could select for genome wide 487 mutations in metabolic genes that can contribute to higher than MIC resistance to multiple 488 antibiotics (5, 44). Additionally, the CF lung is very complex and many other factors may be 489 selecting for persistence, such as evasion to the immune system and exposure to a stressful 490 oxidative environment (17). 491

492
We also found similarity between the response to ciprofloxacin and meropenem is that some 493 isolates from early infection with a high persistence phenotype exhibit a "revival" phenotype 494 to meropenem and/or ciprofloxacin. This is defined as no growth observed in the undiluted 495 cultures, however, once the antibiotics are diluted 10-fold (Fig. S5), growth is observed 496 (isolates shown as light red circles in Fig. S1). This phenotype indicates the presence of 497 persisters that are only detected when the antibiotic is removed from their environment. This 498 "revival" phenotype is only observed in the early isolates, and we speculate that this could be 499 a potential intermediary route to evolve high persistence. All other isolates classified as high 500 persister had high CFU counts from both undiluted and 10-fold diluted cultures. The 501 correlation between meropenem and ciprofloxacin could be shaped by the antibiotic killing 502 mode and bacterial resistance mechanisms: P. aeruginosa cells can tolerate ciprofloxacin 503 exposure by elongating without dividing, allowing cells to survive longer in the antibiotic 504 such that when it is removed they can divide and grow (45-47). Meropenem can be 505 enzymatically degraded by beta-lactamase activity (48). This detoxifies the environment from 506 meropenem at a faster pace than an antibiotic which is not, allowing cells to revive faster 507 (49). In contrast, tobramycin tends to be bactericidal to P. aeruginosa, it disrupts bacterial 508 cells membranes killing cells upon exposure to high concentrations (50), such that cells either 509 persist or die with no revival intermediary. This may also explain why we observe fewer high 510 persister isolates in the presence of tobramycin treatment (Fig. 6). 511

512
We find that maximum bacterial density is negatively correlated with both persistence and 513 resistance. This indicates that persisters either grow slower or have an extended lag and/or 514 stationary phase; both traits that have been observed to contribute to persistence (51-55), and 515 a common phenotype in chronic infections (40). This is in accord with the notion that 516 persisters are dormant cells with reduced metabolism and/or non-dividing that are also found 517 in deep biofilm layers (13). It has also been shown that resistance often comes at a metabolic 518 cost that may lower growth rate (56, 57). Additionally, because later stages of infection, in 519 our study, are characterised by isolates that achieve smaller populations densities in vitro but 520 higher persister counts, this indicates that the fraction of persisters in the population will be 521 proportionally higher in these compared to isolates with a higher maximum density. 522 However, it is important to note that slow growth is a common phenomenon in P. aeruginosa 523 isolates from late stage infection and this may be influenced by factors other than antibiotic 524 response strategies. 525 526 Persistence is a complex polygenic trait, that may on one hand be under selection by 527 variables other than antibiotics, such as oxidative stress and host immune evasion, and on the 528 other hand also be affected by selection on other phenotypes such as growth rate. Therefore, 529 we here focus on the phenotype, and not the genotype, of persistence. While most work has 530 focused on resistance, we are with this work achieving a better understanding of the 531 correlation of persistence to resistance; how this is coupled to bacterial growth rate; and 532 associations with specific antibiotics. This is important because the discovery of new 533 antibiotics has stalled and our best approach is to use the already available drugs in novel 534 waysby reconsidering doses and combinations (58, 59). To do so we must continue to 535 explore how bacteria respond and adapt to drugs in situ. 536 537 In conclusion, our results indicate that persistence and resistance phenotypes rarely evolve 538 before a decade post infection. As expected, our results are consistent with resistance being 539 the primary strategy for surviving antibiotic treatment in late stage chronic infections. 540 Persistence, however, occurs earlier in infection than resistance, when it could be used as a 541 stepping stone to resistance and thereby contribute to the development of recalcitrant 542 infections. Our study highlights the importance of not generalising when drawing conclusions 543 from studies on single clones and/or single antibiotics to better understand persistence and 544 preventing its evolution as a means to mitigate resistance. We cannot assume every infection 545 follows the same course, but that each is influenced by the pathogenic clone type, antibiotic 546 treatment and the host environment. Critical to achieve this is the development of methods 547 for easy detection of persistence, which is not routinely screened for. These results could 548 have implications for early intervention treatment strategies to prevent evolution of 549 persistence and ultimately resistance, e.g. using lower doses of antibiotics and in specific 550 combinations, allowing us to prolong the use of antibiotics for treating infections (58,59).  Table S1: An overview of the non-lung, acute and chronic infecting isolates, and CFU and 557 MIC data. Data given as CFU undiluted (CFU1), diluted 10 fold (CFU10) and diluted 100 558 fold (CFU100) and the standard error for the replicates for each. Persistence phenotype listed 559 as high/low and high/low/revival. Time from first sampling of clone type to sampling of the 560 specific isolate given as time first. 561 562 Table S2: An overview of isolates of the transmissible clone types DK1 and DK2 and CFU 563 and MIC data. Data given as CFU undiluted (CFu1), diluted 10 fold (CFU10) and diluted 100 564 fold (CFU100) and the standard error for the replicates for each. Persistence phenotype listed 565 as high/low and high/low/revival. Persistence-resistance phenotype listed as LS, HS, HR. 566 time from first sampling of clone type to sampling of the specific isolate give as time first. 567 568     , IMG96 2007a, IMG180 1980, IMG73 2003, IMG190 1992, IMG198 2009b, IMG204 1992b, IMG198 1984, IMG72 1973, IMG76 1973, IMG75 1973, IMG74 2003, IMG205 2005b, IMG177 2008, IMG230 2008, IMG229 2008, IMG228 2008, IMG231 2012, IMG220 2012, IMG223 2012 , IMG187 1992, IMG196 1992, IMG193 1991, IMG179 1995, IMG195 2009, IMG184 2009,IMG95 1992a, IMG208 2009, IMG203 1973, IMG77 2009, 420 2012, 421 2006 , IMG96 2007a, IMG180 1980, IMG73 2003, IMG190 1992, IMG198 2009b, IMG204 1992b, IMG198 1984, IMG72 1973, IMG76 1973, IMG75 1973, IMG74 2003, IMG205 2005b, IMG177 2008, IMG230 2008, IMG229 2008, IMG228 2008