Effect of host-mimicking medium and biofilm growth on the ability of colistin to kill Pseudomonas aeruginosa

In vivo biofilms cause recalcitrant infections with extensive and unpredictable antibiotic tolerance. Here, we demonstrate increased tolerance of colistin by Pseudomonas aeruginosa when grown in medium that mimics cystic fibrosis (CF) sputum versus standard medium in in vitro biofilm assays, and drastically increased tolerance when grown in an ex vivo CF model versus the in vitro assay. We used colistin conjugated to the fluorescent dye BODIPY to assess the penetration of the antibiotic into ex vivo biofilms and showed that poor penetration partly explains the high doses of drug necessary to kill bacteria in these biofilms. The ability of antibiotics to penetrate the biofilm matrix is key to their clinical success, but hard to measure. Our results demonstrate both the importance of reduced entry into the matrix in in vivo-like biofilm, and the tractability of using a fluorescent tag and benchtop fluorimeter to assess antibiotic entry into biofilms. This method could be a relatively quick, cheap and useful addition to diagnostic and drug development pipelines, allowing the assessment of drug entry into biofilms, in in vivo-like conditions, prior to more detailed tests of biofilm killing.

*Author for correspondence: f.harrison@warwick.ac.uk ^T hese authors contributed equally to the manuscript  Open symbols denote colonies of endogenous bacteria, closed symbols colonies identifiable as P.
aeruginosa. Circles: lungs 1; squares: lung 2, diamonds: lung 3. ANOVA showed there was no significant interaction between lung and strain at either day, in total or area-standardised data (see Table S1). Raw data, R code and full results of ANOVA analyses of these data are supplied in Document S1. recovered from individual tissue sections. ANOVA was used to test for effects of strain, treatment (no, subinhibitory or bactericidal concentration) and their interaction. The residual mean square from the ANOVA was used to conduct planned pairwise t-tests to compare the mean c.f.u. from biofilm exposed to BODIPY-colistin with the c.f.u. from biofilms of the same strain that were not treated with BODIPYcolistin. A significant drop in c.f.u. was observed only for SED8 treated with the highest concentration (denoted with *; t 18 = 6.23, p < 0.001; c.f.u. was approx. 20% of that observed in untreated biofilms), all other comparisons were not significant. Full data and statistical analysis are supplied in Document S1. period was the same as for the tissue sections exposed to BODIPY-colistin to allow for comparable degradation of the signal due to time or temperature in both experimental samples and calibration samples.
The best fit was calculated in DataGraph 4.5.1 (Visual Data Tools Inc.) with equal weighting for all data points and R 2 was 0.99. Raw data is supplied in Document S1.

Fig. S5 Recovery of initial dose of BODIPY-colistin, as measured by fluorimetry of biofilm
homogenate and surrounding SCFM after 18h exposure. Each symbol is one tissue section. Box shows 1 st and 3 rd quartile with median line, whiskers show interquartile range, asterisk shows outlier. Note that tissues + biofilms were exposed to BODIPY-colistin in a total volume of 1 ml SCFM, therefore concentrations correspond to total µg present. Background fluorescence from the lung tissue in the absence of either bacteria or BODIPY-colistin was very low. Raw data is supplied in Document S1.
Table S1 Reproducibility of biofilm loads on ex vivo tissue. The data in Figure S2 were analysed using ANOVA to test for effects of lung, strain and their interaction on bacterial load, and by linear mixed-effects models to calculate the variance in each species 'bacterial load within and between lungs. This latter information was used to calculate the intraclass correlation coefficientthis is simply the proportion of total variance explained by lung and is a commonlyused measure of repeatability. The larger this value, the greater the between-lung variance relative to the within-lung variance, i.e. a larger value means the data from replica lungs in the same treatment group are more similar, and that there is less noise present in the data due to random variation ("error"). The ICC is bounded between 0 and 1 and is conceptually similar to the commonly-used Pearson's correlation coefficient. In the table, "Interaction" records results for the lung*strain interaction term in ANOVAs. "ICC" refers to the intra-class correlation coefficient. Raw data, R code and full results of analyses are supplied in Document S1.