We performed a prospective, single-centre, observational study (Bichat-Claude Bernard Hospital, Paris, France). The studies involving humans were approved by Ethical authorizations were obtained from the National Ethics Committee for the Protection of Persons Nord-Ouest (N°034/2018). The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study. No animal studies are presented in this manuscript. No potentially identifiable images or data are presented in this study. The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

All adult patients who underwent LT between 1 July 2018 and 30 June 2019 were included.

The included patients were followed up for 90 days. Surgical transplantation procedures and perioperative care, including postoperative and immunosuppressive management, were standardised for all patients according to our local protocol as previously described [9]. Cefazolin (or the antibiotic that was administered to the donor at harvest) was used as the standard antibiotic prophylaxis and was adapted to microbiological cultures obtained from bronchoalveolar lavage (BAL), which was systematically performed just after surgery. Antibiotic prophylaxis was stopped after 48 h in patients with negative cultures of postoperative BAL, as recommended [10]. IC was started [3 Million International Units (MIU) twice daily] in combination with intravenous antibiotic therapy in cases of P. aeruginosa or ESBL-PE pneumonia during the postoperative ICU stay, which were diagnosed from the recommendations for the standardisation of definitions of infections in cardiothoracic transplant recipients [11], as previously described elsewhere [4]. IC was used as secondary prophylaxis on the assumption that it could prevent recurrence of P. aeruginosa or ESBL-PE pneumonia, and intravenous antibiotic therapy was used to curatively treat the P. aeruginosa or ESBL-PE pneumonia episode according to the recommendations [12–14]. IC was continued for at least 90 days, regardless of whether the patient was still on mechanical ventilation. After 90 days, continuation of this treatment was left to the discretion of the physician in charge. The duration of intravenous antibiotic therapy was generally 7 days, but could be longer depending on the doctor’s decision.

We recorded patient characteristics at baseline (age, sex, body mass index, and aetiology of pulmonary disease), type of LT (i.e., single or double LT), rate of pneumonia, duration of mechanical ventilation, length of ICU stay, tracheostomy, time from LT and initiation of IC, duration of IC treatment within 3 months of LT, IC-related side effects, exposure to antibiotics within 3 months, specific lung graft complications (acute cellular rejection confirmed by histopathological evidence after transbronchial lung biopsies performed only in cases of suspicion and not systematically [15]; definite, probable or possible antibody-mediated rejection according to Levine et al [16]); and airway complications that were severe bronchial stenosis requiring balloon dilation or insertion of endobronchial stent and bronchial anastomosis dehiscence [17], ICU and mortality rates at 28 days and 90 days.

During the 90 days following LT, all respiratory samples [plugged telescoping catheter (PTC), BAL, bronchial aspirate (BA), sputum] were only taken during usual care (i.e., when pneumonia was suspected) during the postoperative ICU stay, conventional pulmonology hospitalisation and day hospital and systematically plated on a selective medium (SuperPolymyxinTM, Ellitech, Puteaux, France). There was no systematic respiratory sampling protocol to detect tracheobronchial colonization. Samples were incubated at 37°C for 48 h, in addition to the standard cultures. This selective medium allows for the detection of GNBs with reduced susceptibility to polymyxins, including colistin, regardless of the mechanism or level of resistance. All distinct colonies were studied. Identification was performed using mass spectrometry (Maldi Biotyper^®, Bruker Daltonics, Bremen, Germany). The susceptibility to antibiotics was determined using the disk diffusion method according to the recommendations of EUCAST. ¹ ESBL production was confirmed using the double-disk synergy test [18]. We distinguished GNB with natural resistance to colistin (Proteus spp., Providencia spp., Serratia spp., Morganella spp., Hafnia alvei) from GNB with acquired resistance to colistin. The minimum inhibitory concentration (MIC) of colistin was determined using microdilution (Umic, Biocentric, Bandol, France) for all strains naturally susceptible to colistin. Resistance to colistin was defined as an MIC >2 mg/L (EUCAST).

To determine the colistin resistance mechanism, whole genome sequencing (WGS) of Enterobacterales and P. aeruginosa isolates with acquired resistance to colistin was performed on each isolate of the same species with identical antibiotic susceptibility per patient. WGS was performed on a MiniSeq system (Illumina, San Diego, United States) with paired-end reads and read lengths of 150 bases. Libraries were prepared using the Nextera DNA Sample Preparation Kit from Illumina. Reads from Illumina sequencing were used for whole genome analyses. Read quality was assessed using FastQC v0.11.8. and Trim Galore v0.4.5 was used for quality and adapter trimming. Trim Galore was set up to trim basecalls with a Phred quality score inferior to 30, and reads less than 50 bases long were withdrawn. MetaPhlAn2 v2.6.0 [19] was used to verify the identifications of isolates and identify putative cross-contaminations. Reads were assembled using SPAdes v3.11.1 [20]. The quality of the assemblies was examined using QUAST v5.0.2 [21]. Gene annotation was performed using Prokka v1.13.3 [22].

The sequence type (ST) of the isolates was determined using CGE MLST software [23]. Diamond [24] was used to identify all of the antibiotic resistance genes by aligning all genomes against the AMRFinder database (version 2019-04-29). To obtain a reference genome, Enterobacter cloacae and Klebsiella aerogenes strains were downloaded from the Genome Taxonomy Database (GTDB) [25], and the closeness of the strain was tested using Mash [26]. For Escherichia coli, strains from the same phylogroup were downloaded from the EnteroBase database [27]. We used the closest strain to avoid SNPs linked to evolution. For P. aeruginosa, PAO1 genes were downloaded from NCBI. Using CD-HIT v4.7 [28], interesting genes (phoP, phoQ, pmrA, pmrB and mgrB/yobG for Enterobacterales and phoP, phoQ, pmrA, pmrB, parR, parS, colR, colS, cprR and cprS for P. aeruginosa) were searched in the genomes of our strains. Polymorphisms in these genes were determined using ClustalOmega v1.2.4 [29] for alignment against the reference genome downloaded from NCBI and a Python script for specific SNP detection. The impact of mutations detected was assessed using SIFT [30], PROVEAN [31] and Polyphen-3 [32]. We considered a deleterious effect for the mutation if two of these software packages predicted a deleterious effect.

Data are presented as medians and interquartile ranges for continuous variables and as frequencies and percentages for categorical variables.

We compared the baseline characteristics of patients and outcomes according to their exposure to inhaled colistin using Fisher’s exact or Wilcoxon tests, as appropriate.

The incidence rate of colistin-resistant GNB emergence in the respiratory tract in the 90 days following LT was estimated by pooling the GNB with acquired or natural resistance to colistin. To control for the immortal-time bias induced by direct comparison of exposed and unexposed patients to IC treatment, we applied the statistical methodology described and developed by Suissa [33, 34]. The immortal-time bias is the bias induced by the period before exposure to IC treatment in patients who will be exposed at a given time. During this unexposed period, a bias occurs because no events may be observed under exposure. According to this approach, comparisons are made between exposed and unexposed person-times, unlike subjects. Following the methods described in a previous study [33], we defined the observation period from LT to 90 days for surviving patients and death for the other patients. Exposure and non-exposure times to IC were identified for each patient, and the emergence of resistance to colistin, if applicable. Incidence rates under exposed and unexposed periods, their ratios and 95% confidence intervals (CI) were estimated using Poisson log-linear regressions.

All tests were 2-sieded, with a type-I error of 0.05. Analyses were performed using R software, version 4.0.5 (Copyright (C) 2021 The R Foundation for Statistical Computing).

Seventy patients underwent LT during the study period and were included in the present study. No patient was lost to follow-up. The baseline characteristics of the patients are presented in Table 1. Patients were primarily transplanted for chronic obstructive pulmonary disease (COPD) (31.4%) and interstitial lung diseases (ILD) (57.1%), with a median age of 59 [52–63] years and a male/female ratio of 1.80. Three patients (two patients who did not receive IC and one patient who received IC) had been colonized by P. aeruginosa prior to lung transplantation. No patient had a history of IC treatment prior to lung transplantation.

Twenty-seven (38.6%) patients received IC (Figure 1). The median time between LT and the initiation of IC was 16 [10–32] days, with a median duration of exposure of 74 [60–80] days. Patients receiving IC experienced important morbidity during the postoperative course in the ICU with a longer duration of mechanical ventilation (25 [4–45] vs. 2 [1–5] days, p = 0.0004), more tracheostomies (59.3% vs. 18.6%, p = 0.0007), and longer ICU length of stay (41 [24–68] vs. 14 [11–23] days, p = 0.0002) than patients who did not receive IC. The rate of GNB pneumonia was higher in patients who received IC than in those who did not (100% vs. 55.8%, p = 0.00001), whereas the rate of Gram-positive cocci pneumonia was similar (18.5% vs. 16.3%, p = 1). Patients who received IC had similar rates of acute cellular rejection (40.7% vs. 18.6%, p = 0.06), antibody-mediated rejection (63% vs. 48.8%, p = 0.33), severe bronchial stenosis (29.6% vs. 20.9%, p = 0.57), and bronchial anastomosis dehiscence (7.4% vs. 9.3%, p = 1) versus patients who did not receive IC. The mortality rates at Day 28 and Day 90 were not significantly different between the patients with and without IC (3.7% vs. 9.3%, p = 0.64 and 11.1% vs. 16.3%, p = 0.73) (Table 1).

No significant IC-related adverse events were observed, particularly no bronchospasm or acute kidney injury.

A total of 867 respiratory samples were screened for the presence of GNB resistant to colistin, with 393 (45.3%) BAL, 38 (4.4%) PTC, 420 (48.4%) BA and 16 (1.9%) sputum samples.

The median number of samples collected per patient was 11 [9–16]. Patients receiving IC had more samples than patients without IC (13 [10.5–18.5] vs. 11 [8–14], p = 0.009).

The MIC to colistin of the 28 GNB isolates with acquired resistance was between 4 and 16 mg/L.

WGS was performed on one isolate of the same species with identical antibiotic susceptibility per patient for Enterobacterales and P. aeruginosa strains. The results are described in Supplementary Table S1. The 7 isolates of E. cloacae isolated from the same patient were ESBL-producing and carried bla _CTX-M-15. No known mcr genes (mcr-1 to mcr-10) were detected. A missense mutation in the pmrA gene coding for a two-component system (PmrAB) associated with colistin resistance was identified in the E. coli and K. aerogenes strains. For both strains, the mutation was localized in G53, which is an amino acid hot spot in PmrA [35]. The mutation was G53C for E. coli and G53R for K. aerogenes. Both mutations were predicted to impact the protein and were described previously [35]. No mutation was found in the pmrB, phoP, phoQ or mgrB genes. No mutation was found in pmrA, pmrB, phoP, phoQ or mgrB genes in the E. cloacae strain. This strain belongs to cluster VIII, which is not known to have a heteroresistance phenotype to colistin [36]. Among the eight sequenced P. aeruginosa strains, a missense mutation was identified in the pmrB gene (P175S) in one strain and in the parS gene (V216A) in another strain. These mutations have not been described, but they were predicted to impact the protein, and both genes are associated with colistin resistance [37]. No mutation was found in the pmrA, phoP, phoQ, parR, colR, colS, cprR, or cprS genes. For the 6 remaining P. aeruginosa strains, no mutation was found in the 10 studied genes.

Antibiotic exposure is represented as the total number of days over the 3-month follow-up period (Table 3). Overall, patients receiving inhaled colistin had greater exposure to antibiotics than patients without colistin, with 11/13 antibiotics more used in patients receiving IC.