Bacterial persisters in long-term infection: Emergence and fitness in a complex host environment

Abstract

Despite intensive antibiotic treatment, Pseudomonas aeruginosa often persists in the airways of cystic fibrosis (CF) patients for decades, and can do so without antibiotic resistance development. Using high-throughput screening assays of bacterial survival after treatment with high concentrations of ciprofloxacin, we have determined the prevalence of persisters in a large patient cohort using 460 longitudinal isolates of P. aeruginosa from 39 CF patients. Isolates were classed as high persister variants (Hip) if they regrew following antibiotic treatment in at least 75% of the experimental replicates. Strain genomic data, isolate phenotyping, and patient treatment records were integrated in a lineage-based analysis of persister formation and clinical impact. In total, 19% of the isolates were classified as Hip and Hip emergence increased over lineage colonization time within 22 Hip+ patients. Most Hip+ lineages produced multiple Hip isolates, but few Hip+ lineages were dominated by Hip. While we observed no strong signal of adaptive genetic convergence within Hip isolates, they generally emerged in parallel or following the development of ciprofloxacin resistance and slowed growth. Transient lineages were majority Hip-, while strains that persisted over a clinically diagnosed 'eradication' period were majority Hip+. Patients received indistinguishable treatment regimens before Hip emergence, but Hip+ patients overall were treated significantly more than Hip- patients, signaling repeated treatment failure. When subjected to in vivo-similar antibiotic dosing, a Hip isolate survived better than a non-Hip in a structured biofilm environment. In sum, the Hip phenotype appears to substantially contribute to long-term establishment of a lineage in the CF lung environment. Our results argue against the existence of a single dominant molecular mechanism underlying bacterial antibiotic persistence. We instead show that many routes, both phenotypic and genetic, are available for persister formation and consequent increases in strain fitness and treatment failure in CF airways.

Fig 1. High-throughput screening approach for isolates with a high persister (Hip) phenotype.

(A) A large collection of Pseudomonas aeruginosa clinical isolates were grown to stationary phase in quadruplicate wells for two biological replicate (BR) experiments (each isolate tested 8 times in total). Each isolate was treated with 100 μg/ml of ciprofloxacin for 24 hours, while growth was assessed by plating on LB agar. Following antibiotic treatment, cultures were diluted then plated on agar, at which point survival was assessed. Each isolate was given a persister score based on consistent replicate survival following treatment. Isolates for which 3–4 replicates survived for each BR were given a score of 3–4, respectively, and were considered high persisters (Hip). Isolates with a respective score of 0–2 were considered low persisters (Lop). (B) Score distribution of P. aeruginosa Hip (blue) and Lop (grey) isolates against ciprofloxacin. (C) Traditional time-kill assays were performed for three Hip (HipIso 1–3, solid blue lines) and three Lop isolates (LopIso 1–3, solid gray lines) from the same patient to validate the high throughput screen. Colony forming units (CFU) per ml were determined following treatment with 100 μg/ml of ciprofloxacin. Data are the mean of 3 independent biological cultures, bars represent SD. The dashed black line represents the limit of detection. Three independent ciprofloxacin-exposed colonies were regrown from each of the residual cultures of the persister assays for the Hip strains and then treated with the same persister assay method (RGHip 1–3, blue dashed lines), showing the same killing kinetics as the first assay. (D) We present the final distribution of Hip versus Lop isolates via pie chart.

Fig 2. High-persisters in the multi-trait landscape.

(A) Lop (grey circle) and Hip (blue diamonds were analyzed via principle component analysis with respect to their similarity with other infection-linked traits: growth rate (GR_LB), adhesion, and ciprofloxacin MIC (cip). 446 isolates with complete trait sets were included. Hip isolates do not consistently cluster with any one additional trait. Each symbol represents a P. aeruginosa isolate. The first Hip isolates from a lineage (FirstHip, yellow triangles) were highlighted as Hip variants with mitigated effects of other accumulating mutations within the lineage to improve cross-lineage comparison. In each case, FirstHip and the remaining Hip isolates shift to various degrees from ‘naïve’ towards ‘adapted’ levels given the particular Hip dataset. We illustrate this using data ellipse enclosing samples approximately within the first standard deviation (t distribution, 68% of the set) for isolate sets characterized as FirstHip (yellow ellipse), and the remaining Hips (blue ellipse). (B) We visualized the association between lineages that produced Hips versus resistant isolates and/or slow growing isolates (identified by the minimum growth rate of lineage isolates falling below 70% of the P. aeruginosa PAO1 growth rate based on a 45 minute generation time in LB in microtiter wells). Association between variables is illustrated by a mosaic plot (multi-way contingency table visualization) where color indicates significant deviation from the expected frequency of lineages in each cell under trait independence using Pearson’s chi-squared test. Presence of isolates with a given trait in a lineage is indicated by ‘Yes’, absence by ‘No’.

Fig 3. High-persister incidence patterns from a lineage-based perspective.

(A) Lineages were classed according to several nested characteristics: transient versus non-transient lineages, Hip presence, presence of multiple Hips, continuous periods of isolated Hips, and lineage-initiating Hips. Lineages representing each combination of traits are shown on the left (Hip blue diamonds, Lop grey circles), while characteristic sets are identified and enumerated for the entire collection on the right. (B) The continuous lineage count of Hip- lineages (grey circles) versus Hip+ lineages (blue diamonds) for the prior year of colonization is plotted, while the accumulating count of Hip+ lineages from time 0 is shown by black diamonds. (C) A survival curve shows the probability of no Hip emergence over time since first detection of a lineage, based on time from first isolate to first Hip isolate for each lineage. The step function is shown in dark gray with confidence band in light gray. Lineages which never produce a Hip isolate are censored at time of last isolate (black crosses). (D) Transient lineages (lineages of shorter than 2 years duration, less than 50% of total patient infection length, and which are followed by the appearance of a new lineage) are significantly associated with lineages lacking Hips (Hip-), while non-transient lineages are associated with the presence of Hips based on Pearson's chi-squared test (via a mosaic plot visualizing a multi-way contingency table). Transience-unclassifiable lineages of shorter than 2 years’ duration at the end of a patient’s collection period are shown for context (‘New’).

Fig 4. High-persister incidence versus patient treatment.

(A-C) Differences in treatment between Hip- patients, Hip+ patients, and Hip+ patients prior to their first Hip were evaluated using clinical records of treatment with antibiotics active against Pseudomonas aeruginosa. We focus on fluoroquinolone treatments, which includes ciprofloxacin (used regularly for nearly all patients) and ofloxacin. (A) The degree of fluoroquinolone treatment was determined by summing the length of treatment periods and dividing by the length of each patient’s monitoring period (to normalize data across patients). (B) The longest continuous treatment period with ciprofloxacin was determined by finding the longest continuous span of months where a drug was prescribed. (C) The number of months where 3 or more P. aeruginosa-targeting antibiotics were prescribed divided by monitoring timespan was used to assess elevated drug use. (D-E) The top 4 P. aeruginosa-targeting antibiotic classes prescribed to the patients include aminoglycosides (top drug: tobramycin), fluoroquinolones (top drug: ciprofloxacin), macrolides (top drug: azithromycin) and polymyxins (top drug: colistin). We evaluated the months these drugs were prescribed in Hip+ patients in the period before their first Hip was isolated. (D) We evaluated the length of the last treatment (in months prescribed) preceding the month of Hip emergence. (E). The time between this last treatment month to the month of Hip emergence (a value of 0 means the drug was used in a patient the month prior to Hip emergence) is shown. (F) Hip presence is evaluated in the lineages of 20 patients assessed previously in a study of strain persistence over their maximum ‘eradication’ period, the MEP (at least 6 months of P. aeruginosa-negative sputum) via a mosaic plot. The majority of lineages which persisted through apparent eradication periods are Hip+. Statistics: Panels A-C show boxplots for Hip+ versus Hip- patients (Hip- N = 17, Hip+ N = 22) where difference in population distributions is tested via an unpaired Wilcoxon test of the means for each Hip+ population versus the Hip- population. Panels D-E show boxplots for data from all Hip+ patients (N = 22). Panel F evaluates associations between groups based on Pearson's chi-squared test.

Fig 5. Fitness comparison of Lop and Hip isolates in biofilm conditions.

(A) A representative Lop and Hip isolate with similar characteristics (Lop: cip MIC 1.0 μg/ml; growth rate 0.62 hr^-1; time since first detection 4.28 years. Hip: cip MIC 0.75 μg/ml; growth rate 0.57 hr^-1; time since first detection 5.49 years) were differentially tagged with CFP (Lop) or YFP (Hip). Tagged isolates were cocultured and allowed to form biofilms in a flow-cell model for 72 hours. Mixed biofilms were treated for 24 hours with ciprofloxacin (4 μg/ml). Propidium iodine (PI) was added to visualise dead cells (red). (B) Biomass was quantified for each population. Significant differences in biomass following treatment were determined using unpaired t-test (*** p <0.001).