This study included 29 commercial broiler farms from the Flanders region of Belgium. Farms were included based on convenience sampling, as they were recruited through outreach efforts facilitated by local veterinary poultry practices. Participation was on a voluntary basis and only conventional broiler (Ross 308) farms in Belgium were included. Per farm, one house was randomly selected for follow-up during one production cycle and samples were collected to determine prevalence of FQ non-susceptible E. coli on day 0, 3 ± 2 days, and 36 ± 3 days of that production round. Also, information was collected on farm size, hatchery, age of the parent stock, and AMU during that specific production round. AMU of the specific production cycle was collected by a form which was filled in by the farmer during the sampling period. Quantification of AMU was done through the calculation of the Treatment Incidence (TI) per 100 days, as described by Persoons et al. (20, 21). The TI is equal to the number of days animals are treated with antimicrobials in a theoretical period of 100 days. Sampling was conducted between October 2020 and December 2021.

Environmental swabs from the stable were collected on day 0 before the day-old chicks arrived, when bedding material was already placed and the feeding pans were filled. For the environmental swabs premoistened sponge swabs with 10 mL Buffered Pepton Water (BPW) were used (3 M, SLL10BPW, St-Paul, United States). Five locations were sampled in duplicate: the floor and boots in the anteroom, and the floor, drinking cups, and feeding pans in the house. An A4-sized floor area was sampled per swab, while three drinking cups, one feeding pan per swab, or both soles of a pair of boots were sampled per swab. These locations were chosen based on previous research, which identified these locations as high risk areas for contamination with bacteria, even after cleaning and disinfection (22). Also, drinking water samples were collected at the end of the drinking line in each house after cleaning the line with a disinfectant and running the water for around 3 min. The sample was sent to an external laboratory for bacteriological analysis.

On day 0, 3, and 36, chicks were sampled to screen for gastrointestinal carriage of FQ non-susceptible E. coli. On day 0, 30 day-old chicks were randomly collected before they were unloaded in the house. The chicks were transported to the laboratory in a cardboard transport box, followed by euthanasia through cervical dislocation and collection of the entire gastrointestinal tract and contents. On day 3 and 36, cloacal samples were collected from 30 randomly selected chicks using sterile rayon-tipped swabs (Vacutest Kima, Italy). Also, on day 36, three broilers were euthanized and feather samples were collected to screen for FQ residues. All collected samples were cooled directly, transported at a temperature of 2–7°C and processed within 4 h after collection.

To detect residues of FQ in the feathers of broilers, a validated UPHLC-MS/MS method was used (23). The feathers were screened for the presence of ciprofloxacin, enrofloxacin and flumequine residues. The residue concentration of the complete feather sample was determined by weighing around 500 mg of feathers, followed by cutting the feathers into smaller pieces and grinding. From every grinded feather sample, around 100 ± 1 mg was weighed and further processing took place as described in Ringenier et al. (23). The limit of quantification (LOQ) of the used method was 5 ng/g feather for enrofloxacin and ciprofloxacin, and 1 ng/g feather for flumequine.

In the day-old chicks, the entire gastrointestinal tracts were removed with sterile material, cut into smaller pieces, weighed and transferred into a stomacher bag. Subsequently, sterile BPW (Oxoid, Thermo Fisher Scientific, Merelbeke, Belgium) was added (in ratio of 9 mL of BPW per gram of intestines) and thoroughly mixed using a stomacher. Next, 100 μL of the mixture was inoculated on FQ selective (MacConkey agar Nr 3 (Oxoid, Thermo Fisher Scientific, Merelbeke, Belgium) supplemented with 0.25 mg/L enrofloxacin) and FQ non-selective (MacConkey agar Nr 3) media. These plates were aerobically incubated overnight at 37°C, as well as the intestine suspensions in BPW. If no growth was visible on the (supplemented) MacConkey agar plates, 100 μL of the enriched broth was inoculated on a MacConkey plate and/or a MacConkey plate supplemented with 0.25 mg/L enrofloxacin. The concentration of enrofloxacin in the supplemented plates was used to select for non-wild type isolates (0.25 mg/L is the epidemiological cut-off value (ECOFF) for enrofloxacin in E. coli). “FQ non-susceptible E. coli isolates” will be used in this article for FQ non-wild type isolates.

To the environmental swabs, 10 mL of sterile BPW was added immediately after transportation to the laboratory. After homogenization, using a stomacher, the swab suspensions were aerobically incubated for 24 h at 37°C. After incubation, 10 μL of this suspension was plated on FQ selective and FQ-non-selective media as described above and aerobically incubated overnight at 37°C.

One hundred milliliter of the collected water samples was filtered through a 0.45 μm filter. The filter was placed on non-supplemented Rapid E. coli (Bio-Rad) agar plates and Rapid E. coli agar plates supplemented with 0.25 mg/L enrofloxacin to screen for FQ non-susceptible E. coli. Plates were aerobically incubated at 37°C for 24 h. The colonies were counted and the number of E. coli isolates was determined per 100 mL.

Cloacal swabs were weighed before and after collecting feces to determine the amount of feces collected. Three milliliter of BPW was added and the swab was thoroughly vortexed. Next, 100 μL of the feces suspension was used to make 10 fold serial dilutions (from 10⁻¹ to 10⁻⁴) in Phosphate Buffered Salt solution (PBS). Of every dilution, 100 μL was inoculated on FQ selective and FQ non-selective media. These plates were aerobically incubated overnight at 37°C, as were the swabs containing BPW broth for microbial enrichment. The concentration of FQ non-susceptible E. coli and total E. coli in the cloacal samples were determined (CFU/gram feces) based on the counts on plates with 20–200 CFU and taking into account the weight of the feces on the swab (ranging between 0.01 and 0.34 gram). The proportion of CFU grown on FQ selective over FQ non-selective MacConkey plates allowed to determine the prevalence of FQ non-susceptible E. coli strains in a specific cloacal sample. Identification of the isolates at the species level were done by MALDI-TOF mass spectrometry, as previously described in Vereecke et al. (24). When no growth was visible on the FQ selective or the non-selective MacConkey plates, 100 μL of the overnight enriched broth was inoculated on a second MacConkey plate (supplemented with enrofloxacin) and aerobically incubated overnight at 37°C. When colonies were detected after enrichment only, it was assumed that the concentration was below the detection limit of direct plating. For analysis, the value of such samples was set at the detection limit (for day-old chicks this was 10 CFU/mL BPW and for cloacal swabs 100 CFU/mL BPW) and corrected for the amount of feces on the swabs, as described previously (25).

Of every farm from each sampling point, around 10 randomly selected FQ non-susceptible isolates originating from 10 different animal samples were stored at −20°C. From each positive environmental sample, one FQ non-susceptible E. coli isolate was stored. Due to practical and financial constraints, only a selection of isolates was further analyzed.

For the determination of the minimum inhibitory concentration (MIC), five farms were included based on the results of the prevalence of FQ non-susceptible E. coli in the environment and in day-old chicks. Only isolates of day 0 and 36 were included for MIC determination, which resulted in a selection of 92 E. coli isolates obtained from FQ supplemented plates. Antimicrobial susceptibility testing was performed using a Sensititre EU Surveillance E. coli EUVSEC Plate (Trek Diagnostic Systems, Thermofisher Scientific, Merelbeke, Belgium) according to the manufacturer’s guidelines. The protocol was followed as described in Ibrahim et al. (26). In brief, the Sensititre plate was aerobically incubated at 35°C for 18–24 h. The MIC was determined as the lowest concentration at which no visible growth was observed. The quality control strain used was E. coli ATCC 25922. The ECOFFs from the European Committee on Antimicrobial Susceptibility Testing were used for interpretation (27).

Based on MIC results, a further selection of E. coli isolates was made for whole genome sequencing (WGS). Isolates with different resistance profiles from the same sample time point within a farm were included to undergo WGS, resulting in a selection of 66 FQ non-susceptible E. coli isolates, originating from four farms. DNA extraction was done using the GenElute™ Bacterial Genomic DNA Kit from Sigma-Aldrich (Merck, Overijse, Belgium) following the manufacturer’s protocol. The quantity and purity of the DNA was determined spectrophotometrically using a NanoDrop (Thermo Scientific, Waltham, MA, United States) and the quantity using fluorometry (Qubit 4, Invitrogen, Carlsbad, CA). A DNA-library was made using the Nextera XT library kit (Illumina). Library fragmentation length and library DNA concentration were evaluated using the TapeStation 4,200 with the HS D5000 ScreenTape and Reagent kits (Agilent Technologies, Santa Clara, CA) and a dsDNA HS assay kit for the Qubit 4 fluorometer, respectively. All libraries were sequenced on a MiSeq instrument (Illumina, San Diego, CA) using the MiSeq V3 chemistry, as described by the manufacturer’s protocol, for the production of 2 × 250 bp paired-end reads. Pre-processing and quality control of the sequenced data were performed using the Shiga-toxin producing E. coli pipeline described by Bogaerts et al. (28). NCBI AMRFinder+ (v3.10.18) was used to screen the assembled contigs for genes and mutations associated with AMR using the 2021-12-21.1 database. This database includes genomic markers commonly associated with FQ resistance, such as the often plasmid-encoded qnr and qepA genes, as well as chromosomal mutations in for example gyrA and parC. Additionally, the assemblies were screened for the presence of virulence-associated genes using blastn v2.6.0 and the VirulenceFinder E. coli database (accessed on July 23, 2023). Sequence typing was performed using the MLST and cgMLST schemes obtained from EnteroBase (29) (accessed on July 7th 2024) as described previously (28). Additional SNP-based clustering was performed for sequence types (STs) with multiple isolates using SnapperDB v1.0.6 (30), of which the selected reference genomes are listed in Supplementary Table S1. Isolates were considered to be clusters if their pairwise SNP distances were 5 or less (31).

The WGS data analysis workflow includes taxonomic classification of the trimmed reads using Kraken 2 v2.0.7 and a custom database containing all NCBI RefSeq genome entries (database accessed 2nd November 2021) annotated as “complete genome” with accession prefixes NC, NW, AC, NG, NT, NS, and NZ of the following taxonomic groups: archaea, bacteria, fungi, human, protozoa, and viruses (32). For nine strains, the proportion of reads classified to an unexpected species (i.e., not E. coli) was greater than 5%, which is the failure threshold enforced in the workflow (28). For these datasets, the majority of reads were classified as E. fergusonii. Additional verification was performed by constructing a core genome MLST phylogeny containing all in-house strains, complemented with 50 randomly selected complete E. coli and randomly selected complete E. fergusonii genomes from NCBI. In the resulting topology, the nine strains classified as E. fergusonii by Kraken 2 clustered with the E. fergusonii genomes obtained from NCBI, indicating that this taxonomic identification is likely to be correct.

For descriptive statistics, IBM SPSS Statistics 29.0 (IBM, New York, United States) was used. To investigate the relationship between the proportion of FQ non-susceptible E. coli at the end of the production cycle and total TI of the production cycle, age of the parent stock, capacity of the stable, proportion of positive day-old chicks, and sum of FQ positive environmental locations; a backwards stepwise regression model was conducted, using p < 0.1 for inclusion in the model, resulting in a univariable linear regression model. An independent t-test was conducted to determine if there was a significant difference in proportion of FQ non-susceptible E. coli on day 3 between the flocks that were treated with lincomycin and spectinomycin and the flocks that did not receive such treatment. An independent samples Kruskal-Wallis test was performed to compare the average number of day-old chicks carrying FQ non-susceptible E. coli per farm across different hatcheries.

On 72.4% (21/29) of the farms, the production cycle started with a course of antimicrobials (Figure 1), of which 95.5% used lincomycin in combination with spectinomycin. Five farms started using antibiotics some days after the start of the production cycle. Only three farms did not administer antimicrobials during the entire production period (Supplementary Table S3). The average TI_DDDvet was 4.18 [min 0 - max 33.72], meaning that broilers in this study received antimicrobials during on average 4.18% of their lifespan, which equals to 1.8 treatment days within a production cycle of 42 days. Only one farm reported using a FQ (flumequine) during the observed production cycle. The concentration of flumequine in the feathers from this farm ranged from 50 to 600 ng/g feather (Supplementary Table S4). In the feather samples of the other farms, no FQ concentrations above LOQ could be detected, except for one farm where very low levels of flumequine above LOQ were found, ranging from 1.2 to 3.7 ng/g feather. According to the AMU records, only lincomycin in combination with spectinomycin was administered on this farm in the first 3 days of the production cycle.

On 14 farms, at least one FQ non-susceptible E. coli isolate could be obtained from the environment before the day-old chicks entered the stable (Table 1; Supplementary Table S5). On 79.3% (23/29) of the farms, FQ non-susceptible E. coli were detected in the gastro-intestinal tracts from day-old chicks before entering the stable (Table 2). Of these positive flocks, the percentage of day-old chicks carrying FQ non-susceptible E. coli was on average 55.3% [min 6.7 - max 100] (Figure 2). There was no statistically significant difference in the percentage of positive day-old chicks between the different hatcheries (p = 0.38), indicating that there was no specific hatchery providing a significant higher or lower number of day-old chicks carrying FQ non-susceptible E. coli.

On days 3 and 36, FQ non-susceptible E. coli were present at 100% of the farms (and in 94.0 and 99.4% of the broiler samples, respectively) (Figure 2). On days 0, 3 and 36, the proportion of FQ non-susceptible E. coli in the total E. coli population (only taking into account samples that contained E. coli) was 28.3%, 26.7, and 30.6%, respectively (Figure 3; Supplementary Table S6). The total TI of the production cycle was significantly associated with the proportion of FQ non-susceptible E. coli at the end of the production cycle (r = 0.42, 95% CI [0.06, 0.68], p = 0.03) (Figure 4). There was no significant difference in mean proportion of FQ non-susceptible E. coli at day 3 between the flocks treated with lincomycin and spectinomycin and the flocks that did not receive such treatment, with a mean difference of 4.87% (95% CI [−10.38, 20.10%], p = 0.52).

Of the 92 selected isolates (obtained from FQ supplemented plates) which underwent MIC determination, 63.6% was multidrug-resistant when applying ECOFF values (resistant against 3 or more antimicrobial classes) (Supplementary Table S7). Of the 66 isolates selected for WGS, nine isolates which were previously identified as E. coli by MALDI-TOF MS, were identified as Escherichia fergusonii based on WGS data (Supplementary Table S8). Of the remaining 57 E. coli isolates, 14 isolates originated from environmental samples, 22 isolates from day-old chicks and 21 isolates from broilers at day 36. The concordance between the detected genetic traits and the observed phenotypic resistance, known as the phenotypic-genotypic relationship, ranged from 93.0% for beta-lactam antibiotics to 100% for sulfonamides, trimethoprim and quinolones. About 80.7% of the isolates carried resistance genes against beta-lactam antimicrobials and 71.9% against aminoglycosides. Only one isolate, from a broiler at day 36, carried an ESBL gene (bla_CTX-M-55). WGS detected genes and/or mutations associated with three or more classes of antibiotics in over 86% of the isolates (Supplementary Table S8).

Of the FQ non-susceptible isolates, 61.4% carried point mutations in the Quinolone Resistance-Determining Region (QRDR) and 38.6% carried qnr resistance genes (Figure 5). No other FQ resistance traits were detected. The proportions of qnr genes and QRDR mutations per sampling time point are listed in Table 3. All isolates with triple or quadruple QRDR mutations (n = 11) had ciprofloxacin MIC values > 4 mg/L (Supplementary Table S8). Of the 14 strains carrying qnr genes, two carried an additional AMR gene on the corresponding contig, being bla_CTX-M-55 and tet(A) (Supplementary Table S8). All E. fergusonii isolates (n = 9), consisting of two clusters of identical isolates carried the gyrA S83L mutation. One cluster of E. fergusonii isolates (n = 7) additionally carried the parE I355T mutation.

The 57 analyzed E. coli isolates carried on average 13 virulence-associated genes (VAGs) (SD ± 5.13). Of the 57 E. coli isolates, most common VAGs were terC (93.0%), nlpI (87.7%), iss (86.0%), traT (77.2%), sitA (75.4%), hlyF (64.9%), csgA (56.1%), iutA (56.1%), iucC (54.4%), and lpfA (52.6%). Other genes were present in less than 50% of the isolates (Supplementary Table S8).

There was a large variation in the STs of FQ non-susceptible E. coli, both within and between farms (Figure 5). Three isolates were assigned to different novel STs (i.e., STs not yet introduced in the EnteroBase database). On farm 10 (ST1196) and 25 (ST40), FQ non-susceptible E. coli isolates collected on day 0 from the environment, showed identical cgMLST profiles and <5 pairwise SNP differences to isolates obtained from cloacal swabs of broilers on the same farm (Figure 5; Supplementary Table S6). In addition, on farms 6 (ST1485) and 10 (ST1146), isolates from day-old chicks clustered together with those from the broiler at the end of the production cycle. On farm 6, five ST1485 isolates exhibited ≤3 pairwise SNP differences, while on farm 10, two ST1146 isolates, for which SNP analysis was not possible due to the lack of a suitable reference genome, had identical cgMLST profiles. Moreover, isolates from different day-old chicks within the same farm clustered together, which was the case on farm 25 (three ST1485 isolates with 0 pairwise SNP differences), farm 22 (three ST69 isolates with 0–4 pairwise SNP differences), and farm 6 (four ST1485 isolates with 1–4 pairwise SNP differences) (Supplementary Table S9). For ST5451, five isolates collected from chicks on day 0 shared identical cgMLST profiles. However, no suitable reference genome was available on EnteroBase, and SNP analysis was not performed.