An ex vivo cystic fibrosis model recapitulates key clinical aspects of chronic Staphylococcus aureus infection

Staphylococcus aureus is the most prevalent organism isolated from the airways of people with cystic fibrosis (CF), predominantly early in life. Yet its role in the pathology of lung disease is poorly understood. In mice, and many experiments using cell lines, the bacterium invades cells or interstitium, and forms abscesses. This is at odds with the limited available clinical data: interstitial bacteria are rare in CF biopsies and abscesses are highly unusual. Bacteria instead appear to localize in mucus plugs in the lumens of bronchioles. We show that, in an established ex vivo model of CF infection comprising porcine bronchiolar tissue and synthetic mucus, S. aureus demonstrates clinically significant characteristics including colonization of the airway lumen, with preferential localization as multicellular aggregates in mucus, initiation of a small colony variant phenotype and increased antibiotic tolerance of tissue- associated aggregates. Tissue invasion and abscesses were not observed. Our results may inform ongoing debates relating to clinical responses to S. aureus in people with CF.


INTRODUCTION
Staphylococcus aureus is currently the most prevalent pathogen isolated from the airways of people with cystic fibrosis (CF) [1]. Historically, it is has been associated predominantly with paediatric pulmonary infection [2], yet the most recent data shows that ~50 % of even the oldest patients are now colonized [3]. The presence of S. aureus in the respiratory tract varies considerably geographically and over time, and although prevalence appears to decrease with age [4] MRSA is most prevalent between 10-30 year olds [3]. Determining the difference between colonization and infection is both important and difficult. Nasal carriage of S. aureus among children with CF is common: Stone et al. [5] reported that 52.4 % of the infants they studied harboured the organism. However, relatively high carriage rates in healthy children have also been recorded [6]. Further, expectorate is difficult to collect from infants, so samples are usually collected by oropharyngeal swab, but the presence of organisms in the upper respiratory tract is not always indicative of lower airway infection [7,8].
Moreover, the association of S. aureus with progressive lung disease -as measured by worsening lung function and the development of subsequent infection by the chronic CF pathogen Pseudomonas aeruginosa -is unclear [9]. S. aureus is able to rapidly adapt to and persist in the CF lung environment [10], and worsening lung condition has been associated with the formation of S. aureus small colony variants (SCVs) [11]. SCVs are known to remain in the lung longer than WT bacterial cells [12] and demonstrate increased antimicrobial resistance [13]. In studies of bronchoalveolar lavage (BAL) fluid from children aged 0-7 years, both Gangell et al. and Sagel et al. [14,15] found that positive S. aureus culture was linked to a higher degree of airway inflammation, as measured by increased neutrophil count and IL-8. In another study, lung damage (bronchiectasis) in early CF was recorded by CT scan and S. aureus was the most commonly isolated organism [16]. Conversely, in adult CF patients, S. aureus infections, in the absence of P. aeruginosa, are a marker of milder disease [1]. There is also an indication that methicillin-sensitive S. aureus may inhibit P. aeruginosa, thus delaying lung disease progression [17]. The evidence that early S. aureus infection worsens prognosis for CF patients remains conflicting and warrants further study [3,18].
The lack of understanding regarding the role of S. aureus in the development of lung disease in CF has led to debate over the use of anti-staphylococcal prophylaxis early in life. Studies by Ratjen et al. and Stutman et al. [19,20] linked the use of broad-spectrum antibiotics to an increase in P. aeruginosa isolation. However, a recent Cochrane review [21] showed no effect of anti-staphylococcal prophylaxis on P. aeruginosa colonization at 3-4 years. There was a suggested trend towards higher rates of P. aeruginosa at 4-6 years but as the studies reviewed did not last more than 6 years, conclusions about the long-term effects of prophylaxis could not be drawn. A pragmatic randomized controlled trial is currently in progress (http://www. cfstart. org. uk).
In addition to the lack of clarity over the clinical consequences of S. aureus colonization, there is also a gap in our mechanistic understanding of the microbiology of S. aureus in CF. The interaction of S. aureus with the pulmonary airway in CF, and its subsequent role in pathogenesis, is not clearly documented. CF epithelial cells have been shown to have an increased abundance of aGM1, a receptor that binds S. aureus and P. aeruginosa, compared with WT epithelia [22], and an in vitro study by Schwab et al. demonstrated that bacterial adherence to a bronchial epithelial cell line was significantly greater for CF S. aureus isolates than non-CF [23]. McKenney et al. also show aggregates of S. aureus are visible on the surfaces of bronchial sections from children with CF [24]. Yet, the only other study we found with direct evidence of S. aureus infection in human biopsy specimens, demonstrated that S. aureus did not adhere to the airway epithelium but was found aggregating within the mucus [25]. Furthermore, these authors cited several studies reporting an ability of S. aureus to bind mucins [26,27]. It is obvious that better understanding of just how S. aureus colonizes the airway in CF is required, as there are unambiguous data on the underlying mechanisms of pathogenesis and virulence, and the influence of S. aureus on subsequent infection by other micro-organisms.
Mice are the most commonly used animal model of pulmonary infection in CF. Mouse models have been used to identify virulence-related genes in pathogens and to test novel therapeutic agents aimed at reducing inflammation or infection [28]. However, mouse models present a number of challenges. In particular, mice do not develop spontaneous P. aeruginosa endobronchial infection, suppurative lung disease and mucus plugging of the airways, that are fundamental characteristics of human CF progression [29]. S. aureus does not appear to readily colonize the airways of mice or produce an inflammatory response, even when clinical strains are used, in the presence of mucus [30]. Furthermore, S. aureus forms severe abscess-like lesions in the mouse lung [31]. When lung abscess does occur in humans, S. aureus is the most commonly isolated organism [32]. However, lung abscess is rare in children [33] and even more so in people with CF: since the advent of neonatal screening and survival past infancy, abscesses are almost never observed in the CF population [34][35][36].
We previously developed a clinically relevant, high throughput model of chronic bacterial infection in CF. By combining sections of porcine bronchiole (obtained post-slaughter from a commercial abattoir) with a synthetic CF sputum medium (SCFM) [37] we have shown that we can study another key CF pathogen, P. aeruginosa, in a physicochemical environment similar to that present in vivo [38][39][40]. The use of CF-like growth medium combined with animal tissue with a greater structural, chemical and immunological similarity to human lung [41] means that our model is likely to facilitate a more human-like pathology of S. aureus than that observed in mouse models or in vitro studies. Using clinical reference strains and CF isolates of S. aureus, we can now demonstrate that in a CF-like environment, S. aureus shows non-invasive pathology. Bacteria grow as aggregates associated with the surface of bronchioles and in the surrounding artificial

Impact Statement
Chronic bacterial infection is one of the main causes of declining lung function and morbidity in people with cystic fibrosis (CF). Staphylococcus aureus is the most prevalent organism isolated from the airways of infants with CF. Yet its overall role in the pathogenesis of pulmonary disease is poorly understood. More accurate, high throughput models of CF lung infection are necessary to explore the interactions between S. aureus and host in this context. In vitro studies cannot accurately replicate chronic disease characteristics, such as the presence of small colony variants, and animal models also do not represent features of clinical disease. Mice, in particular, are not readily colonized by S. aureus and when an infection is established it leads to the formation of abscess, a phenomenon not observed in the human CF lung. Here we show an existing ex vivo pig lung model of CF infection supports the growth of S. aureus and that S. aureus demonstrates clinically significant characteristics, including appearance of small colony variants, increased antibiotic tolerance and preferential growth as aggregates in mucus.
sputum, without the formation of abscesses or other significant ultrastructural changes to tissue. In fact, we report a potential preference for aggregation in the surrounding mucus rather than growth associated with the bronchiolar surface. We also report the appearance of sub-populations of SCVs. Our work is consistent with S. aureus adopting a 'persister' rather than 'invader' strategy in CF. This stands in marked contrast to the pathology of this species in mouse and cell-culture infection models, and underlines the importance of selecting models that reflect host-pathogen interactions more closely, in order to reliably study the role of specific bacteria in clinical disease progression. A better understanding of S. aureus pathology in the highly idiosyncratic environment of the CF lungs will inform optimized clinical responses to positive S. aureus cultures.

Bacterial strains
USA300LAC Los Angeles County clone (JE2, BEI resources) was used as an example of a well-documented clinical strain. In total, 18 S. aureus strains isolated from nine people with CF at Hospital Universitari i Politecnic La Fe were included as exemplars of CF-associated strains. The strains used in this work were extracted for diagnostic purposes and sent to the microbiology service for routine analysis. Once analysed, and instead of being discarded as usual, the strains were grown in tryptic soy broth and kept in glycerin. We chose isolates that represented a range of patient clinical presentation and bacterial phenotypes (e.g. weak or strong biofilm formation as measured by in vitro attachment assays, and isolates from patients during periods of stable presentation and episodes of acute exacerbation) (Table S1, available in the online version of this article).

Media and culture conditions for growth in the ex vivo pig lung (EVPL) model
For use in the lung model, bacterial stocks were grown overnight at 37 °C on lysogeny broth (LB) agar. SCFM was prepared according to Palmer et al. [37] with the modification that we removed glucose and supplemented with 20 µg ml −1 ampicillin. Our previous work suggested that glucose facilitated the growth of endogenous bacteria present in the lungs and that ampicillin helped to limit the growth of any resident bacteria in the lung that remained after sterilization [38]. We selected a concentration of ampicillin that provided the best possible coverage against endogenous populations but was sub-inhibitory for the S. aureus strains used (confirmed by standard MIC testing at the time of collection).
The EVPL model was adapted from our group's previous work [38,39], which in turn built on prior use of pig lungs for non-CF studies [42]. Briefly, lungs were collected from a local butcher (Steve Quigley and Sons, Cubbington, Warwickshire, UK) as soon as possible following abattoir delivery, and processed immediately upon arrival at the laboratory. Previous antibiotic administration history of the pigs is not known, but use of antibiotics as growth promoters is banned in the EU, so use is restricted to prophylaxis and mass medication of herds only when infection is suspected. Approximately 5 mm 2 sections of bronchiolar tissue were dissected from the lung under sterile conditions. During dissection the sections were washed three times with a 1 : 1 mix of RPMI 1640 and Dulbecco's Modified Eagle Medium (DMEM) (Sigma-Aldrich), supplemented with 50 µg ml −1 ampicillin, and then rinsed once in sterile SCFM without ampicillin supplementation. Bronchiolar samples were transferred to a clean petri dish and sterilized under a UV lamp for 5 min. Previous work has shown that tissue damage as a result of UV sterilization is minimal (not visible using light microscopy) [39]. Samples were transferred to 24-well tissue culture plates: each well contained 400 µl SCFM supplemented with 0.8% w/v agarose to form a soft pad. Tissues were inoculated with the appropriate strain of bacteria using a sterile hypodermic needle (29G). The tip of the needle was lightly touched to the surface of the chosen S. aureus colony, and then used to pierce the surface of bronchiolar tissue for inoculation. A sterile needle was used for mock infection in uninfected controls. Then, 500 µl of SCFM+ 20 µg ml −1 ampicillin was added to each well and the plate was sealed with a Breath-Easier membrane (Diversified Biotech). Plates were incubated at 37 °C for up to 7 days and refreshed with 300 µl of SCFM+ 20 µg ml −1 ampicillin at 48 h. Following incubation, tissue was rinsed in 1 ml PBS to remove loosely adhering cells and processed for colony counts and microscopy.

Bacterial load assay
Tissue samples used to assay total bacterial numbers in tissue-associated biofilm were homogenized individually in 1 ml PBS in reinforced metal bead tubes (Fisherbrand) using a FASTPREP 24 5G homogenizer (MP Biomedicals) for 40 s at 4.0 m s −1 . Homogenates were serially diluted and aliquots were plated on LB and mannitol salt (MSA) agar (Oxoid, Thermo Scientific) to obtain single colonies. MSA agar is selective for Staphylococcus and was used to determine colony numbers of test strains and ensure that negative controls did not contain any S. aureus already resident in the lung tissue prior to experimentation. Nonselective LB agar was used to check for contamination or growth of endogenous populations. When the bacterial load in SCFM surrounding the lung tissue was measured, 300 µl of SCFM was taken from each well at the same time as tissue processing and transferred to a sterile Eppendorf tube. Samples were vortexed, serially diluted and plated in the same way as tissue homogenate.
LB plates were incubated aerobically at 37 °C for 18-24 h and MSA plates were incubated in at 37 °C 5 % CO 2 for 24 h and then for an additional 48 h at room temperature in ambient CO 2 to allow visualization of slow-growing colonies. Total numbers of S. aureus recovered from the complete sample were calculated from tissue and SCFM colony numbers combined.

Microscopy
Replica 5 mm 2 bronchiolar tissue samples, dissected and inoculated as above, were preserved for haematoxylin and eosin staining. Samples were placed in individual tissue processing cassettes (Simport) and fixed overnight in 10 % neutral buffered formalin (NBF) at approximately 20 times w/v. Longitudinal, step sections (100 µm) were prepared for histopathology by an external service (University of Manchester). If samples required storage before transport, they were placed in 70 % ethanol and kept at 4 °C for up to 7 days. Sections were transported in PBS with residual NBF. In addition, 10 µl samples were taken from the SCFM surrounding lung tissue, diluted 1 : 10 in PBS and 5 µl drops prepared for Gram stain.
Microscopy was conducted with an Axio scope.A1 light microscope (Carl Zeiss) with AxioCam ERc5S digital camera and images processed in Zen Pro 2.1 Blue edition.

Colony identification
Colonies that were able to grow on MSA plates and ferment mannitol (indicated by yellow pigmentation of colony and surrounding clear zone) were regarded as S. aureus. Colony morphology was recorded by digital photography. The identity of colonies that were weak fermenters or small in size was initially confirmed using hydrogen peroxide to test for the production of catalase and then a staphylase assay (Oxoid). If SCVs were suspected, colonies were recovered by aerobic incubation overnight on Columbia Blood Agar and morphology was compared to the originally inoculated parent strain. PCR of large and small colonies from MSA plates was used to verify taxonomic identity. First, we performed colony PCR using Staphylococcus-specific primers [43]. Once genus was confirmed, amplification of the 16 S-23S rRNA intergenic spacer region and Sanger sequencing was used to confirm that SCVs were S. aureus (primers 5′-TGCCAAGGCATC-CACCG-3′ and 5′-GGCTGGATCACCTCCTT-3′).

Determination of antibiotic susceptibility by microdilution
The MIC of flucloxacillin against a pair of representative MSSA isolates (FQ128 and FQ142) and the standard MSSA control strain, ATCC29213, was determined according to the British Society of Antimicrobial Chemotherapy guidelines for microdilution [44] before examining antibiotic tolerance in the EVPL model. All isolates were from sputum samples collected from people with CF during stable presentation. All three strains demonstrated an MIC of <0.0625 mg l −1 and were sensitive to flucloxacillin, MIC breakpoint for resistance is 0.12 mg l −1 (BSAC Guidelines).

Antibiotic tolerance in the EVPL model
Infected lung bronchioles incubated for 48 h as described above. At 48 h 'pre-dose' tissue samples were homogenized and assayed as above. Biological replicates (four bronchiolar sections for each strain and condition) were challenged with antibiotic: tissue samples were washed in PBS then placed in a new 24-well plate on agar pads. Thereafter, 500 µl SCFM alone was added for controls and 500 µl SCFM +either 0.25 mg l −1 or 5 mg l −1 flucloxacillin for antibiotic tolerance test samples. Concentrations were calculated from the recorded broth MIC of <0.0625 mg l −1 for the strains tested and the clinical target pharmacokinetic (PK) parameters. The PK parameter for beta lactam efficacy is T >MIC with a target of 50 % of dosing interval at a free plasma concentration 4-5×MIC [45]. Then, 5 mg l −1 was included to account for potential protein binding, which is 95 % for flucloxacillin [46]. Plates were incubated at 37 °C. At 4 and 24 h post-dose, treated tissues were washed in PBS to prevent any antibiotic carryover, and all samples homogenized and plated as described above. The bacterial loads in non-antibiotic and antibiotic treated tissues were compared.

Statistical analysis
All data were analysed using RStudio (Mac OS X 10.6+version 1.0.153) (RStudio, https://www. rstudio. com) using linear models followed by ANOVA and post-hoc Tukey HSD tests where appropriate, or using Kruskall-Wallis tests, as specified in the text.

Clinical strains of S. aureus are able to colonize EPVL
Data taken from preliminary experiments (Fig. 1) showed that bacterial load (c.f.u.) recovered at 48 h was similar for a large range of S. aureus CF clinical isolates, and comparable to Los Angeles County clone of USA300LAC, a methicillinresistant (MRSA) clinical isolate known to form biofilm [47]. Table S1 shows details of CF isolates. Samples were taken from people presenting either with acute exacerbation or in a stable condition (defined as chronic isolates). Isolates were grouped randomly and tested on three separate occasions. In each case, lung samples were dissected from a single lung. Differences in growth were distinguishable between strains (ANOVA. F 2,40 =1.9, P=0.02) but post hoc analyses showed no significant difference between any CF isolate and USA300LAC (RStudio, script available on request). The median values for S. aureus load recovered at 48 h were 1.64×10 5 , 3.64×10 5 and 6.1×10 5 c.f.u. respectively for lungs 1-3. Growth on Staphylococcal selective media (MSA) was only observed on one of the uninfected controls (Fig. 1a).

S. aureus growth in association with bronchiolar tissue depends on strain and lung but trends are consistent over time
To verify the reproducibility of S. aureus numbers recovered from bronchiolar tissue and the ability to observe differences between strains, triplicate samples of USA300LAC and two clinical strains were inoculated into three further lungs taken from different pigs (Fig. 2) and incubation was extended to 7 days. Clinical isolate FQ151 was compared with FQ184, these strains were chosen as an appropriate example pair as they were both isolated from patients with similar demographics, both MRSAs that form biofilm, and had both demonstrated comparable rates of recovery at 48 h from the same lung (Fig. 1). However, FQ184 was isolated during an acute disease exacerbation and FQ151 during stable presentation (representing chronic infection or colonization) (Table  S1). It was therefore hypothesized, that by comparing these two strains, differences between exacerbation-associated strains and those apparently persisting asymptomatically may be observed, and, if so, could be further investigated with a larger selection of paired isolates. Fig. 2 shows that USA300LAC was again able to establish in the lung and mean S. aureus load recovered at 48 h was 6.8×10 6 c.f.u. Mean yields for clinical strains at 48 h were 3.8×10 6 and 5.5×10 5 c.f.u. for FQ184 and FQ151, respectively. By 7 days no colonies were recovered from tissue taken from one of the three lungs for either clinical strain, and counts for clinical strains in a second lung were below or close to the lower limit of detection. Average c.f.u. counts at day 7 were 4.2×10 3 for USA300LAC, 7.9×10 3 for FQ184 and 8.1×10 3 for FQ151. The data were analysed by ANOVA to test for differences between strains, lungs and days, and interactions between day*lung and strain*lung. Total bacterial load was significantly different once again between strains (F 2,40 =12.0, P <0.001) and between lungs (F 2,40 =6.03, P=0.005). There was a main effect of day (F 1,40 =155, P <0.001) and the magnitude of the drop between 48 h and 7 days did not vary significantly between lungs (day*lung interaction F 2,40 =9.5, P=0.110) or between strains (strain*day interaction F 2,40 =1.73, P=0.190). Different lungs did, however, affect the growth of the strains differently (lung*strain interaction F 4,40 =2.62, P=0.049).

S. aureus is visible as aggregates at the airwaytissue interface but does not appear to invade tissue
Monitoring bacterial load within the model by c.f.u. count shows that the model can maintain a S. aureus population at 48 h. In addition, it was important to investigate whether the consistent decline in c.f.u. counts by 7 days was due to the inability of the model to support S. aureus or whether adaptation of S. aureus within the model prevented sufficient monitoring, for example if cells were growing in strongly adherent biofilm that had not been washed off, or as SCV or persister cells that were difficult to recover on media. Therefore, samples were histologically examined to gain a better understanding of growth characteristics in the model. Replicate samples of the inoculated lung tissue (5 mm 2 , dissected from bronchi, as for original model) were washed in PBS and prepared for microscopy, by longitudinal, step section (100 µm), at 2 and 7 days postinfection. Histopathological staining (Figs 3 and 4) showed bronchial mucosa, represented by mature cartilage. In infected samples there is clear evidence of polymorphonuclear leukocytes and organisms, including, predominantly Gram-positive bacilli. Polymorphonuclear leukocytes and organisms are largely absent from uninfected controls with the exception of a small number of rod-shape organisms (Fig. 3b). Gram-positive bacilli aggregate primarily at the tissue-airway interface of infected lungs and their absence from the uninfected lung, (Figs 3a, b and 4a, b) strongly suggests that they are the S. aureus deliberately inoculated into test samples. Lung tissue remained largely intact across all tested isolates, even at 7 days. There was some disruption  (Table S1) and numbers are designated as supplied. Analysis of untransformed data by ANOVA showed statistically significant differences in the bacterial load attained by the different strains (F 7,32 =1.90, P=0.02).
of tissue integrity at the surface, illustrated by the presence of fibrinous material (highlighted with *). None of the samples show any evidence of abscess, which would present as a clearly defined, bordered structure. S. aureus is present only at the tissue-airway interface and does not appear to demonstrate tight association with tissue in the form of a mature biofilm, although more specific staining for biofilm would be required to confirm this. Replica images confirmed histological presentation (Figs S1-S4).
Evidence of other organisms in some samples, for example rod-shaped cells in the tissue infected with FQ184 at 48 h (arrow, Fig. 3f, ×100 magnification), is likely to be endogenous populations present prior to dissection (healthy lungs are not sterile) or contaminants not removed during sterilization; they are also present in the uninfected control (arrow, Fig. 3b). Interestingly, when other bacteria are present, there appear to be distinct ecological niches within the lung environment with Gram-positive bacilli clustered primarily at or near the surface, and rods present embedded within smooth muscle and connective tissue.
This supports much of the current understanding of the structural architecture of polymicrobial communities and it would be useful to study this further especially as the presence of other bacteria may affect the ability of S. aureus to survive and adapt within the model. By day 7, cells in clusters of FQ151 were small and poorly stained, and thus difficult to image (Fig. 4d). This may indicate the cells are no longer viable and explain the lack of colonies recovered on plates (Fig. 2). Although inconclusive, the images appear to suggest that there is still a presence of bacterial cocci in association with the lung tissue at 7 days post-inoculation, despite poor recovery of colonies on selective media. It highlights that sputum sampling and colony count alone cannot reliably determine the presence of S. aureus populations that may be viable but non-culturable. It would be useful to further investigate, perhaps with the use of Staphylococcus aureus specific fluorescent probes or live dead staining in order to confirm the identity and viability of the cells.

S. aureus aggregates in artificial sputum surrounding lung tissue
Initial experiments (Figs 1 and 2) only recorded bacterial burden recovered from washed tissue pieces. The histology suggests at least some S. aureus may be tightly associated with the tissue in the model and not removed during homogenization. In addition, there is evidence in the literature to suggest S. aureus may preferentially bind to mucus plugs in the airways of the CF lung, rather than associating with the tissue surface [25]. Both are possible explanations as to why, despite presumably being adapted to the lung environment, c.f.u. loads associated specifically with the bronchiolar tissue showed such a marked drop by day 7.
A subset of clinical strains, two MRSA and two MSSA (methicillin-sensitive) strains, were chosen to investigate growth in the surrounding SCFM, and compared with USA300LAC (Fig. 5). When bacteria were also recovered from the surrounding SCFM and enumerated, the total c.f.u. count from the sample (tissue plus surrounding SCFM) was higher than the c.f.u. count for tissue only (Fig. 5a). Although  'mucus plugs' do not form in the model, aggregates of S. aureus cells were visible in Gram stains of surrounding SCFM at day seven for all strains tested (Fig. S5) and by day 7, a greater proportion of the total bacterial population was consistently found in the surrounding SCFM than in tissue-associated biofilm (Fig. 5b). The four biofilm forming strains all showed a significant increase in the proportion of recoverable cells in the SCFM between 48 h and day 7 as measured by the Kruskal-Wallis test (FQ128 χ 2 =7.44, FQ140 χ 2 =11, FQ151 χ 2 =7.1, USA300LAC χ 2 =12.3; all P-values ≤0.008). FQ142 did not show a significant increase in SCFM proportion between the two time points (Kruskal-Wallis test, χ 2 =1.2, P=0.22). FQ142 is not a biofilm former (Table S1), and so likely to have a higher proportion already in the SCFM by 48 h. SCFM is a minimal media and it is significant that there are viable cells are maintained in the SCFM for 7 days with limited nutrient replenishment, suggesting the addition of lung tissue to the model is an important improvement.

Tissueassociated populations demonstrate phenotypes consistent with chronic infection
The data collected by cell enumeration (Figs 1, 2 and 5), for growth and localization of S. aureus in the model, suggest the existence of subpopulations present during chronic infection. These likely comprise aggregates of cells in the mucus and a smaller percentage of tissue or biofilm-associated cells, which may be slow-growing SCVs or non-growing persisters. If this is the case it would be consistent with hypotheses in the literature regarding S. aureus infection [48], and we would anticipate that there would be SCVs present in the model and the 'persister' subpopulation would demonstrate increased antibiotic tolerance.
Small colonies (<50 % diameter of usual S. aureus colonies) were observed on MSA plates after incubation for at least 48 h in 5 % CO 2 at 37 °C for samples taken 7 days post-inoculation. Example colonies were photographed (Fig. S6) and some had weak catalase and coagulase results when tested, weak catalase is indicative of deficient haem production and is associated with SCVs, as is weak coagulase. SCVs were confirmed as S. aureus spp. using genus-specific primers (STaG) [43] followed by sequencing of the 16 s-23S intergenic region (Table S2).

Antibiotic tolerance of tissue-associated aggregates
Slow-growing or biofilm bacterial populations may be tolerant to antibiotics and implicated in the persistence of chronic infection. Tolerance of tissue-associated aggregates to flucloxacillin was assessed for a pair of clinically isolated MSSAs that were susceptible to flucloxacillin in a standard MIC microdilution and had the same MIC as the control strain ATCC29213 (<0.0625 mg l −1 ). They were isolated from people who presented asymptomatically (without exacerbation) (Table S1). Lung tissue was infected with S. aureus and incubated as described above. At 48 h, antibiotic test samples were challenged with a single dose of either 0.25 mg l −1 or 5 mg l −1 (to account for potential protein binding) flucloxacillin; growth control samples were maintained in SCFM alone and uninfected lungs used as a negative control. Tissue was processed at 4 and 24 h post-dose and total S. aureus load recorded as before (Fig. 6a). As in previous experiments, uninfected control tissues showed no growth on MSA plates. Clinically, bactericidal activity of an antibiotic is regarded as a reduction of 99.9 % (>3 log 10 decrease) in the c.f.u. ml -1 of the original sample. The data presented in Fig. 6b demonstrates that a mean decrease of >3 log 10 was only achieved against FQ142 when challenged with the higher concentration of 5 mg l −1 flucloxacillin for 24 h, despite all three strains demonstrating susceptibility in a standard MIC test. For ATCC29213 there was no significant effect of dose (ANOVA, f 2,21 =0.2 P=0.8) or dosing interval (ANOVA, f 2,21 =1.9 P=0.2) on c.f.u. ml -1 . Although there was an overall significant effect of dose on c.f.u. ml -1 for FQ128 (ANOVA, f 2,21 =5.8, P=0.01) and FQ142 (ANOVA, f 2,21 =7.5, P=0.004), post hoc analysis showed this was not significant for 0.25 mg l −1 (FQ128, P=0.8 and FQ142, P=0.2) and a mean decrease in c.f.u. ml -1 of >3 log 10 was not achieved for either strain at this concentration (Fig. 6b). Dosing interval was found to be significant for FQ128 (f 2,21 =7.2, P=0.004) but not for FQ142 (f 2,21 =0.03, P=0.97). The fact that the required bactericidal target was not reached for the control strain, ATCC29213, and that the response of the two clinical isolates varied, shows that standard MIC testing may not reliably predict clinical response to antibiotic regimes. There was no significant decrease in bacterial recovery between pre-dose and 24 h post-dose when challenged with 0.25 mg l −1 flucloxacillin and FQ142 showed a potential to grow under these conditions, suggesting sub-inhibitory flucloxacillin concentrations could lead to selection for resistance. Furthermore, although 5 mg l −1 flucloxacillin had a significant effect on c.f.u. ml -1 for both clinical strains, it failed to cause a mean decrease in FQ128 c.f.u. ml -1 of >3 log 10 , suggesting that some strains could tolerate clinically relevant concentrations even when antibiotic protein binding is accounted for. This was a very small sample set, and further investigation is warranted, but, it indicates that maintaining plasma concentrations above the target of 4×MIC may be necessary for effective treatment of S. aureus infection in CF.

Growth and localization of S. aureus in EPVL
There is an apparent contradiction between the usual pathology of S. aureus in human CF samples (non-invasive) and in standard laboratory models using cell-culture or mouse lung infections. To address this gap between the clinical norm and standard lab models, we grew a biobank of clinical isolates in an ex vivo model of CF lung infection, which carefully mimics human tissue structure and the chemistry of CF lung secretions. S. aureus grew with reasonable consistency, both associated with lung tissue, at the airway surface, and as bacterial aggregates in artificial sputum (SCFM) around the tissue. S. aureus bioburden was maintained in EPVL over 7 days, although tissue-associated c.f.u. numbers were diminished over this period and in general the bacteria showed a preference for growth as aggregates suspended in the SCFM.
Defined biofilm was not investigated in histological samples, although there was evidence of potential biofilm architecture and increased structure in bacterial populations for some samples by 7 days compared to 48 h. Interestingly, we have recently reported that P. aeruginosa does demonstrate extensive biofilm growth in the EPVL that can be observed using the same histological staining techniques [40]. This is consistent with the available clinical evidence where extensive biofilms have not been described for S. aureus in the CF lung [49], but small, biofilm-like aggregates are reported in the sputum and on the surface of airways [24,25,50,51]. Consistent with growth as aggregates in vivo, there is evidence that S. aureus switches off the global regulator agr in CF [48].
We did not observe interstitial invasion by S. aureus. Generally, interstitial bacteria are rarely observed in biopsy samples from people with CF (even if significant interstitial inflammation is observed) and bacteria in the airways are confined in luminal mucus plugs or occasionally attached to defined foci of epithelial erosion; this lack of tissue invasion is consistent with the rarity of bacteremia in people with CF [24,25,[52][53][54][55][56][57]. CF lung disease thus presents as bronchiectasis with mucous plugging of small airways. In contrast, microscopy images presented in studies of murine pulmonary S. aureus infections typically show abscesses (a cavitating, pus-filled lesion within the tissue with a defined border: see Fig. 2 in [31] and Fig. 2 in [58], for examples) and lead to chronic pneumonia [59]. Clearly, there are significant differences in S. aureus pathology in mouse models versus the highly specialized environment of CF airways as approximated by the ex vivo pig lung model.
In light of this, it is significant that our results suggest preferential localization of S. aureus in the SCFM surrounding the tissue sections, as opposed to epithelial surface attachment, and we did not observe the appearance of abscess-like structures. S. aureus can cause lung abscess in humans -in fact it is one of the most common bacteria isolated from abscesses [60] -but abscesses are almost never observed in people with CF [56,57]. A specific Panton-Valentine Leukocidin positive strain of MRSA was associated with invasive, cavitating lung lesions in a CF centre in the USA in the mid 2000s [61], but this is unusual. Other than this specific outbreak, a literature search revealed only five cases of abscess among people with CF, only two of which were associated with S. aureus: one patient was co-colonized by P. aeruginosa and the other with P. denitrificans [34,62]. It should be noted that abscess formation is associated with the presence of neutrophils, our model uses post-mortem tissue and, therefore, the effect of the host immune response is not observable. In addition, tissue samples in our model are taken only from the bronchi wall and so may not capture abscess formation. However, given the rarity of clinical evidence for invasive S. aureus infection in people with CF, induction of abscess formation may not Fig. 6. Recovery of S. aureus clinical isolates from EPVL following flucloxacillin challenge. Each S. aureus isolate was inoculated into replica bronchiolar sections of tissue from the same pair of lungs, and incubated as previously described for 48 h. Four replica samples were processed for counting of bacterial load, prior to antibiotic challenge (0 h, pre-dose). Remaining samples were moved to SCFM containing either no antibiotic (circles), 0.25 mg l −1 (triangles) or 5 mg l −1 (diamonds) flucloxacillin and incubated for 4 or 24 h. Uninfected tissue pieces were used as control and processed as infected replica, no colonies were observed on recovery plates for uninfected controls and so c.f.u. was recorded as at the limit of detection (stars). Lines represent mean bacterial load. All isolates were classified as sensitive by standard antibiotic susceptibility testing (MIC<0.0625 mg l −1 ). Bactericidal activity, (>3 log 10 in c.f.u.) was only observed against FQ142 when challenged with 5 mg l −1 flucloxacillin for 24 h (Fig. 6b, highlighted). Antibiotic dose had no significant effect for ATCC29213 (ANOVA, f 2,21 =0.2 P=0.8) for clinical strains there was an overall effect of dose: ANOVA, f 2,21 =5.8, P=0.01 and f 2,21 =7.5, P=0.004 for FQ128 and FQ142, respectively. However, post hoc analysis revealed there was no significance for either strain when exposed to 0.25 mg l −1 flucloxacillin (FQ128, P=0.8 and FQ142, P=0.2).
be clinically important in a CF model. Clinical observations are consistent with a study in which mucus hypersecretion was induced in cultured primary nasal epithelia cells to better mimic conditions in the CF lung: mucus presence led to S. aureus cells moving away from the epithelial cell surface and growing in the mucus [25] -as observed in our model.

Chronic infection phenotypes of S. aureus in EPVL
The tissue-associated subpopulation exhibited enhanced tolerance to flucloxacillin compared with standard susceptibility assay results. This is likely due to physiological cues from the lung environment and the appearance of a subpopulation of SCVs [11,13,[63][64][65] but may also be influenced by the presence of biofilm structure matrix if it is present (which could usefully be explored in future work).
Small colony variants emerged once S. aureus isolates were passaged through the EPVL. The phenotypic development of SCVs has been linked to chronic CF infections [63,64]. The impact of SCVs on virulence and persistence is difficult to study in vitro, as they rapidly revert to normal colony morphology when attempts are made to culture them [51], and it usually requires the creation of artificially generated, stable SCVs which may not fully represent those generated by the in vivo environment [66]. Notwithstanding these limitations, studies have linked the SCV phenotype with attenuated virulence [67] and a dampened host cytokine response [68,69] -both changes that are consistent with the adoption of persistence strategy. The ability of the ex vivo lung model to cue SCV emergence is likely to be a significant strength of the model, permitting a rapid means to generate this phenotype and allowing more extensive future research into the effects of SCVs. It will be important, however, to tease apart the relative contributions of the presence of tissue and the presence of sub-inhibitory levels of ampicillin on SCV emergence. Since completing this work, we have begun a more extensive investigation of methods to reduce the presence of endogenous bacteria from the lung tissue, and we have found that expedited access to lungs after slaughter allows us to obtain tissue which requires less rigorous control measures. Future experiments using EVPL with different or no antibiotic supplementation will allow us to explore the impacts of tissue and antibiotic presence on microbial physiology.

Population heterogenicity of S. aureus in vivo and in EVPL
The evidence outlined above is consistent with S. aureus growing as phenotypically heterogeneous populations in CF -comprising tissue-associated, biofilm-like aggregates and mucus-embedded aggregates, with the potential presence of SCVs. Persistence, antibiotic tolerance, gene expression and ability to be cultured in vitro are likely to differ between these subpopulations, and understanding the differences within heterogenous populations could improve understanding of virulence mechanisms. If this is the case, the ex vivo pig lung model will be valuable in future studies, for the generation of heterogenous populations that are hard to reproduce in animal and traditional in vitro models.

Strengths and limitations of the EVPL model
Our data reinforce the notion that host-specific pathology is an important area of future study, with particular significance for S. aureus. Modelling S. aureus infection, in any tissue other than human biopsy, is likely to present differences and challenges. It is abundantly clear that both mouse and in vitro models fail to capture key features of pathology observed from human clinical data that are revealed by our EVPL model. Consequently, EVPL could fill an important gap in the toolkit for answering crucial questions about the microbiology and clinical impact of S. aureus in CF.
There are, of course, some areas where the model could be optimized for future use. Primarily, variability in S. aureus bacterial load was considerable. We inoculated tissue sections from single colonies, and not a standardized broth culture, as preliminary work with P. aeruginosa in an earlier version of the model showed no significant variation in the number of cells inoculated from single colonies, or the cell numbers recorded at 48 h [39]. Additionally, in CF there is unlikely to be a large burden of colonizing bacteria [70]. Given the differences in the way that P. aeruginosa and S. aureus grow in the model, starting inoculum may be a more significant determinant of the growth and survival of S. aureus and standardizing cell numbers could be important for future developments, for example assessing the impact of antimicrobial agents on viable bacterial burden. More broadly, better data on the variability in S. aureus aggregate size between foci of infection within CF lungs, and between patients, would be useful for understanding S. aureus pathology. In a study by Hirschhausen et al. [59], adaptive changes differed in patients infected with the same S. aureus clonal lineage, indicating that individual host factors had an impact on adaption. This is reminiscent of our results, which show significant variability between lungs taken from different pigs (Fig. 2).
We also stress, given the reported intraspecific phenotypic and genetic diversity of bacteria within and between people with CF, and given observed differences in the growth and localization of a pair of isolates taken from stable infection and acute exacerbation (Figs 2-4), future work could usefully explore a larger collection of isolates in more depth.

CONCLUSION
The ex vivo lung model is designed to maximize physicochemical similarity to chronically infected human CF airways. It is also high throughput and inexpensive, and because it uses post-consumer waste from the meat industry, it presents no ethical concerns [38]. Our results from S. aureus infection of the ex vivo model reveal aspects of chronic CF pathology, which are not captured by existing 'gold standard' animal models: aggregation of bacterial cells associated with tissue but also (perhaps predominantly) in mucus; the development of SCVs; and an increase in antibiotic tolerance. The potential preferential binding of S. aureus to mucus, and lack of tissue invasion, is a phenomenon that may have clinical relevance, especially given current questions about the role of S. aureus as a CF pathogen and the consequences of trying to remove it.

Funding information
The work was funded by a Medical Research Council New Investigator Research Grant to FH (grant number MR/R001898/1) and by the University of Warwick via AHH's final-year research project. N.E.H. is funded by a PhD studentship from the BBSRC Midlands Integrative Biosciences Training Partnership (MIBTP). The Histology Facility equipment used in this study was purchased with grants from The University of Manchester.