Increased survival of experimentally evolved antimicrobial peptide-resistant Staphylococcus aureus in an animal host

Antimicrobial peptides (AMPs) have been proposed as new class of antimicrobial drugs, following the increasing prevalence of bacteria resistant to antibiotics. Synthetic AMPs are functional analogues of highly evolutionarily conserved immune effectors in animals and plants, produced in response to microbial infection. Therefore, the proposed therapeutic use of AMPs bears the risk of ‘arming the enemy’: bacteria that evolve resistance to AMPs may be cross-resistant to immune effectors (AMPs) in their hosts. We used a panel of populations of Staphylococcus aureus that were experimentally selected for resistance to a suite of individual AMPs and antibiotics to investigate the ‘arming the enemy’ hypothesis. We tested whether the selected strains showed higher survival in an insect model (Tenebrio molitor) and cross-resistance against other antimicrobials in vitro. A population selected for resistance to the antimicrobial peptide iseganan showed increased in vivo survival, but was not more virulent. We suggest that increased survival of AMP-resistant bacteria almost certainly poses problems to immune-compromised hosts.


Introduction
Antimicrobial peptides (AMPs) are major components of immune defences in multicellular organisms (Tzou et al. 2002;Koprivnjak and Peschel 2011). Bacterial resistance to AMPs is rarely observed in environmental or clinical isolates, leading some to suggest that these molecules are 'resistance proof' (Zasloff 2002). Based on these observations and the decline in effectiveness of antibiotics, AMPs have been proposed as the basis of synthetic drugs that could be used to fight human infection (Zasloff 2002;Reddy et al. 2004;Giuliani et al. 2007).
In vitro experiments have now demonstrated that, contrary to earlier expectations, AMPs are not resistance proof. Resistance evolved at low cost within just a few hundred generations in Escherichia coli, Pseudomonas fuorescens (Perron et al. 2006), Salmonella enterica (Pr€ anting et al. 2008), Streptococcus pneumoniae  and Staphylococcus aureus (Dobson et al. 2013). Recent work has shown that pexiganan-resistant S. aureus can also be cross-resistant to human neutrophil defensin-1, a human AMP .
Importantly, the repertoire of AMP resistance mechanisms is limited (reviewed in Peschel and Sahl 2006;Koprivnjak and Peschel 2011;Gruenheid and Le Moual 2012). The fact that some microbes can utilise these few resistance mechanisms to inhabit AMP-rich environments, for example Pseudomonas aeruginosa in mucosal tissue (Gruenheid and Le Moual 2012), suggests that such mechanisms could be effective in conferring resistance to a broad range of AMPs. Thus, should therapeutic use of AMPs becomes a reality, providing selective conditions for the rise and spread of bacteria resistant to a broad spectrum of AMPs (cross-resistance), we risk 'arming the enemy' by equipping pathogens with tools for increased resistance to the immune system (Bell and Gouyon 2003;Buckling and Brockhurst 2005). To date, there has been no thorough test of the survival and virulence of a range of AMP-resistant bacteria in an animal host.
We recently demonstrated that fitness costs of AMP resistance are unlikely to constrain evolution in the grampositive bacterium S. aureus. However, simultaneous selection at the same intensity by a combination of two AMPs constrained resistance (Dobson et al. 2013). The experimentally evolved strains from this study provide a resource with which we can comparatively explore the potential limits of shifts in survival of both bacteria and host after infection. AMPs play a critical role in insect innate immune defences (Moon et al. 1994;Lemaitre and Hoffmann 2007;Chae et al. 2012), which are therefore well suited to investigate the potential problems of AMP cross-resistance. As a model host, we use the mealworm Tenebrio molitor. This beetle exhibits a long-lasting humoral antimicrobial immune response to persistent S. aureus infection (Haine et al. 2008a), which is largely dependent on induction of AMPs (Johnston et al. 2014). We therefore predicted that infection titters in this system are sensitive to variation in AMP resistance. We used this host to parameterize survival of AMP-resistant bacteria in vivo, which we complemented with in vitro assays of cross-resistance. We also assessed the effects of these populations on survival of infected hosts. This allowed us to address the potential risk of AMP therapy for the first time in an animal host.

Bacteria and stressor selection
We designed a proof-of-principle study to test the effects of AMP resistance on host and pathogen. All bacterial strains were S. aureus that we had previously selected for resistance against antibiotics or antimicrobial peptides (Dobson et al. 2013 ; Table 1). Briefly, these cultures were grown from an isogenic ancestor in the absence of selection for 10 days by serial passage. They were then inoculated into media containing one of the AMPs pexiganan, iseganan or melittin. Procedural controls were the antibiotics vancomycin or streptomycin. Untreated controls were inoculated into unsupplemented media. All AMPs/antibiotics were suspended at a concentration sufficient to inhibit 50% growth of the ancestral bacteria. Five replicate populations were established per AMP/antibiotic, and cultures were grown by daily serial passage. Concentration of each respective AMP/antibiotic was doubled weekly. Populations were grown for up to 4 weeks, or until extinction shortly after the beginning of week 3 (vancomycin treatment only). The ancestral population was constitutively tetracycline resistant, allowing us to selectively recover them from T. molitor postinfection on media containing tetracycline. These populations demonstrated equivalent performance in fitness assays.

Bacteria preparation for infection and cross-resistance studies
Before infecting T. molitor, we pooled evolved populations from within treatments in equal proportion, for two reasons. First, 'public goods', in which one or a few clones in a bacterial population evolve traits that benefit the whole population, are known to mediate antibiotic resistance (Lee et al. 2010). AMPs can be inactivated by extracellular peptidases (Peschel and Sahl 2006;Koprivnjak and Peschel 2011), so population-level resistance, mediated, for example, by 'public goods', seems particularly likely for AMPresistant bacteria. Second, real infections are commonly caused by mixed populations, not isogenic strains, and this has also been shown in S. aureus (Balmer and Tanner 2011). We therefore sought to assay the population-level processes relevant to a real infection, which can establish the limits of phenotypic space of AMP-resistant bacteria equally well as equivalent experiments with isolated clones.
Populations were pooled by partially thawing glycerol stocks until 100 lL could be removed, which was inocu- Antibiotic-selected control. Selected in vitro by vancomycin for 14 days. Showed apparent increase in resistance then rapid extinction on days 17-18. lated into 5 mL LB and grown to stationary phase. 500 lL aliquots of these cultures were spun for 5 min at 4500 g. Pellets from each treatment were then pooled by communal resuspension in 1 mL tryptic soy broth (Sigma-Aldrich T8907, Munich, Germany) with glycerol (80% water: 20% glycerol v/v) and frozen at À90°C. Cultures of these pooled bacteria were prepared by directly inoculating 5 mL LB with a scrape of the pooled stock and grown for 24 h at 30°C with shaking. Twentyfour hour, cultures were pelleted by centrifugation at 4500 g for 5 min, then washed and resuspended in an equal volume of sterile PBS (NaCl 150 mM, Na 2 HPO 4 10 mM, pH 6.5).
To check for contamination and to quantify CFU in the inocula, 50 lL of each inoculated culture (diluted 9 10 À5 ) was plated with 20 sterile glass beads on LB (Sigma-Aldrich L2897) 1.5% agar containing tetracycline and amphotericin-B as previously. Plates were incubated for 24 h at 37°C and photographed. CFUs were automatically counted using OpenCFU (Geissmann 2013).

Tenebrio molitor culture and infection
Tenebrio molitor were purchased as final-instar larvae from a commercial supplier (www.livefoods.co.uk) and grown in rat chow (Harlan Laboratories, Shardlow, UK) in an insectary at 26°C AE 1 in a 12:12-h light/dark cycle. F 1 offspring of these beetles were used for infection experiments. Females were sexed as pupae and kept individually in rat chow, eclosion was recorded daily, and all beetles were virgins between 7 and 10 days old when used.
Pathogenicity and bacterial survival were assessed in T. molitor by injection with our experimentally evolved S. aureus. Beetles were surface-sterilized by swabbing the ventral abdomen with 80% EtOH, and loading a finepulled glass electrode (Narishige) with~10 5 CFU inoculum suspended in 5 lL PBS. Inocula were injected between the third and forth abdominal sternites into the haemocoel. Negative controls were injected with 5 lL sterile PBS.

Quantifying infection
Staphylococcus aureus were recovered from T. molitor 24 h after infection, coincident with peak AMP expression (Johnston et al. 2014). Controls were injected with sterile PBS to identify any tetracycline-resistant contaminants.
Bacteria were recovered using a 'perfusion bleed' assay, refined from that of (Haine et al. 2008b). The beetle cuticle was surface-rinsed for~45-60 s in 80% EtOH, to remove cuticular contaminants. Genitalia were everted by gently squeezing the abdomen and swabbed with 80% EtOH, and a small incision was made with a scalpel. A 30-gauge needle was inserted through the plural membrane exposed between the abdomen and thorax on the side of the body laterally opposite to the genital incision and pushed into the abdominal haemocoel. 500 lL sterile PBS was pushed through the needle, washing haemolymph out of the abdomen and into a sterile collection tube via the genital incision. 50 lL of perfused haemolymph samples was plated with 20 sterile glass beads on LB 1.5% agar plates containing tetracycline and amphotericin-B to select the experimental strains from the background insect flora. Plates were incubated and colonies counted as previously. CFU counts were then extrapolated to estimates per beetle. If bacteria grew at uncountably high densities, the sample was censored and assigned the maximum number of CFUs that could be counted by our assay (13 000).

Pathogenesis assays
Beetles were injected 8 days after eclosion (50 individuals per treatment) as in the bacterial survival assay, and mortality of each individual was checked at least every 2 days. Mortality was recorded when beetles ceased to respond to mechanical stimulation.

Cross-resistance assays
Cultures for in vitro cross-resistance assays were grown from pooled stocks in M€ uller-Hinton broth (Sigma-Aldrich 70192, 5 lg mL À1 tetracycline, 5.6 lg mL À1 amphotericin-B) for 18 h at 30°C with shaking, to a density of 5 9 10 6 CFU mL À1 . We tested resistance of our evolved populations to the antimicrobial peptides pexiganan (courtesy of Michael Zasloff, Georgetown University, USA), melittin (Sigma-Aldrich M2272) and a 50:50 combination of the two (PGML). Unfortunately, we were unable to obtain sufficient iseganan for a fully reciprocal design. Resistance of naive bacteria to these antimicrobials (crossresistance) was determined by dose-response assays in 100 lL volumes per well in sterile 96-well microtitre plates, using a twofold dilution series from 64 to 0.125 lg mL À1 and an additional well of unsupplemented M€ uller-Hinton broth. 10 lL of culture was added to each well and OD 595 was measured every hour for 6 h, allowing for exponential growth. MIC was determined as the first concentration in which no growth was detected after 6 h.

Data analysis
PBS-injected controls revealed no contamination and so were excluded from further analyses. We performed two analyses of CFU counts. All data were analysed in R version 3.0.2 (R Core Team 2013). To account for censored data from overgrown plates, we used the MCMCglmm package (Hadfield 2010). This innovative package uses Markov chain Monte Carlo techniques to fit generalized linear models in a Bayesian framework, with a wide range of potential error distributions. These distributions include the censored Poisson distribution, which we used to account for censored data from our CFU counts. This model was fitted with 1.3 million iterations, thinning interval of 1000 and a burn-in of 300 000 generating posterior coefficients for each treatment group with 95% confidence intervals. To verify these results, we additionally fitted a standard frequentist GLM with a quasi-Poisson distribution after excluding the censored samples, using the R base GLM function. Post hoc multiple comparisons of frequentist GLM were performed using the general linear hypothesis testing (glht) function from the R multcomp package. The two approaches yielded congruent results, so we present results from the more powerful MCMCglmm.
Survival of beetles in pathogenesis assays was censored 21 days after eclosion. The data were analysed with an accelerated failure time (AFT) survival model with a Weibull distribution, using the survival package (Therneau 2013).

Selected Staphylococcus aureus resistance in Tenebrio molitor
There was significant treatment-specific variation in survival of selected S. aureus populations 24 h after inoculation into T. molitor ( Fig. 1A and B, Table 2). Of the AMP-selected lines, iseganan-selected bacteria showed higher survival in the host than the ancestor (p MCMC = 0.002), and pexiganan-selected bacteria showed a trend towards higher survival (p MCMC = 0.052). Melittinselected bacteria did not survive significantly better than the ancestral population, but there was significantly more variance in this group than in the rest of the data set together (F-test, F = 2.48, df = 8, 64, P < 0.05), suggesting a more diverse mix of resistant and nonresistant clones in this inoculum than in those in other treatments. Relative to the ancestral population, vancomycin-selected and unselected lines showed no differences (p MCMC > 0.1). Streptomycin-selected bacteria showed increased survival compared with the ancestral strain as well as the unselected control strain (p MCMC = 0.02). Results from a frequentist GLM, along with Tukey's multiple comparisons, are presented in Table S1.1 and S1.2.

Pathogenicity of selected Staphylococcus aureus to Tenebrio molitor
Bacterial selection line had significant effects on the pathogenicity of some strains (Fig. 2). Our iseganan-selected bacteria became less pathogenic streptomycin-and pexiga-nan-selected and unselected controls (Table 3). Comparisons between beetles injected with PBS to beetles infected with bacteria revealed the same pattern of differences as The box and whisker plots present the density of each bacterial population in T. molitor as the average log 10 CFU per beetle, showing medians, first and third quartiles and 5th and 95th percentiles. (B) MCMC techniques were used to fit a generalized linear model to the CFU data, because certain plates were uncountable and were therefore right-censored. The MCMCglmm used a censored Poisson distribution to estimate differences between treatments. Solid line at 0 represents the intercept (ancestral population), AE95% confidence intervals represented by dashed lines. The points represent coefficients of each treatment AE95% confidence intervals. Iseganan (p MCMC = 0.002), streptomycin (p MCMC = 0.02) and pexiganan (p MCMC = 0.052) showed significantly more immunoresistance than the ancestor strain. Frequentist GLM and multiple comparisons are presented in supplementary material.
seen in the iseganan group, indicating that treatment by iseganan reduced pathogenicity of the population. Pexiganan-selected and unselected bacteria were also more pathogenic than the vancomycin-selected population.

Assessment of cross-resistance in vitro
We did not find strong evidence of S. aureus cross-resistance to AMPs through in vitro assays (Fig. 3). Relative to   Table 3.
the ancestor, melittin-and pexiganan-selected bacteria showed increased resistance to the compounds that they had been selected by. Melittin-selected bacteria additionally showed increased resistance to PGML. Iseganan-selected bacteria showed decreased resistance to melittin, pexiganan and PGML, suggesting a cost of iseganan resistance in this context. All other bacteria showed equal pexiganan, melittin and PGML susceptibility as the ancestor.

Discussion
Our studies show increased survival of iseganan-resistant S. aureus in vivo, consistent with the hypothesis of AMP therapy having the potential to 'arm the enemy'. This result associates evolved resistance to a single AMP, as would be the case in a topical application, with enhanced survival in the face of an AMP-dependent immune response. Our streptomycin-selected procedural control showed somewhat increased survival in the beetle. As streptomycin interferes with bacterial ribosomeswhich to our knowledge are not affected by insect immune systemswithout further functional studies, this result can only be attributed to increased 'vigour' after selection (Kawecki et al. 2012). Nevertheless, the iseganan-selected bacteria survived in the host on average 3.7 times better (adjusted P < 0.005) than the streptomycin-selected bacteria (Fig. 1, Table  S1.2.), which can be most parsimoniously attributed to differences arising from selection for iseganan resistance. Further work is required to attribute these effects to specific mechanisms.
We observed some congruence between in vivo survival of bacteria in this study and in vitro resistance in our previous (Dobson et al. 2013), but this pattern must be interpreted with caution. Iseganan-selected bacteria were highly iseganan-resistant in vitro and also survived well in T. molitor. Melittin-selected bacteria showed striking interpopulation variability in resistance in vitro, and considerable variation in the numbers of bacteria recovered from beetles. However, our pexiganan-resistant bacteria, which performed poorly under selection in vitro, survived somewhat better in vivo. Additionally, the observed increase in survival in the host could not be clearly predicted from the in vitro cross-resistance data presented here. This highlights that in vivo testing is crucial to explore the potential of cross-resistance in a natural context (Martinez 2008). It would be of value in future to assess how many mutations arose in iseganan-selected strains, and whether genetic variants associated with improved survival in the host are also associated with iseganan resistance. As there are no data on the number of generations that each strain grew for under selection, we cannot estimate the number of mutations that arose in each of our populations, and whether these were equivalent across treatments. However, these strains were all able to grow to roughly equivalent density (optical density, 595 nm) in their respective growth media, and there were no differences between the fitness index (calculated from growth rate) of populations selected for resistance to iseganan or streptomycin (Dobson et al. 2013), indicating that effects in the iseganan-selected bacteria are not likely due to accrual of more mutations than other populations.
Increased in vivo survival of AMP-selected bacteria raises concerns about the eligibility of AMPs as therapeutic antimicrobial or immunomodulatory drugs (Hancock 2000;Zasloff 2002;Easton et al. 2009;Yeung et al. 2011). Our in vivo data augment the proposition that clinical use of AMPs could select widespread resistance to the immune system.  demonstrated in vitro cross-resistance to humanneutrophil defensin-1 in pexiganan-resistant S. aureus, although this effect was not universal amongst resistant clones. However, Pr€ anting et al. (2008) were unable to identify cross-resistance to a panel of AMPs and antibiotics in an isogenic Salmonella enterica sbaA mutant resistant to the porcine AMP PR-39 (Pr€ anting et al. 2008). Collectively, these results and ours support the notion that cross-resistance and immunoresistance of AMP-resistant bacteria is context-dependent : it is possible, but not a universal feature of AMP-resistant clones. Also, in our study, the in vitro assay was less informative than our in vivo assay. This highlights the importance of using robust in

Melittin
Pexiganan PGML vivo models to parameterize performance of resistant microbes. Further work is required to determine the genetic bases of broad-spectrum AMP resistance in these bacteria. It was surprising that iseganan-selected bacteria showed higher survival in vivo but were less pathogenic than our unselected, streptomycin-and pexigananselected bacteria. Despite this disparity, any antimicrobial therapy that confers an advantage to infectious bacteria in vivo is clearly undesirable. One possibility is that the iseganan-selected bacteria have evolved mutations that facilitate better survival in the host mediated by reduced activation of the immune system (Atilano et al. 2011). An additional possibility is that bacteria at high-density negatively regulate virulence traits to maximize transmission (Antia et al. 1994) or that mortality is too crude a measure of pathology to quantify costs to the host of infection with AMP-resistant bacteria. Our design, using mixed populations rather than individual clones, does not allow us to make causal links between resistance and pathogenicity of the selected strains because individual clones within the populations may cause different phenotypes. It is beyond the scope of the present manuscript to determine the mechanistic basis of the reported phenotypes, but future characterization of isogenized strains from these populations could be leveraged to test of the role of AMPs in mediating trade-offs between resistance and pathogenicity. As subcutaneous S. aureus must putrefy its host for transmission, such a trade-off would indicate stabilizing selection on AMP resistance.
In medical settings, the processes described above may be exacerbated. AMPs are an important part of the repertoire of mammalian immune defences (Peschel and Sahl 2006;Kraus and Peschel 2008;Koprivnjak and Peschel 2011;Gruenheid and Le Moual 2012). Our finding, that bacteria that have evolved resistance to antimicrobial peptides can survive better in vivo, complements the finding of  that pexiganan-resistant S. aureus can be cross-resistant in vitro to the human AMP HNP-1. The feasibility of AMPs as future antimicrobial therapies is therefore seriously compromised: resistant nosocomial infections could pose serious hazards, particularly for immune-compromised patients.
In summary, we have shown that an opportunistic pathogen can be fitter in a host after selection for antimicrobial peptide resistance. To our knowledge, this is the first study linking in vitro antimicrobial peptide resistance to in vivo survival. The therapeutic use of a single AMP against infection could 'arm the enemy' (Bell and Gouyon 2003) with cross-resistance to AMPs, including those produced as part of human immune responses.