Four- to six-week-old broiler birds that were suspected of having colibacillosis were submitted to the Poultry Extension Team (PEX) at the Western College of Veterinary Medicine, University of Saskatchewan, as part of a routine analysis for the Saskatchewan broiler industry. In each case, the birds originated from local producers that had farms within a 2-h driving distance of Saskatoon. Members of the PEX performed necropsy on each bird and submitted organ samples for microbial isolation and diagnosis by MALDI-TOF mass spectrometry (Prairie Diagnostic Services¹). If E. coli infection in the birds was confirmed, the PEX sent samples of the heart, liver, and spleen, previously stored at 4°C, to the White lab. At the same time, the PEX requested from the same producers that they submit 3–4 healthy broilers from the flocks where the diseased birds had originated. The cecal contents from these birds were subsequently sent to the White lab.

Samples of the heart, liver, and spleen (diseased birds) or cecal contents (presumed healthy birds) were placed in 2 ml Eppendorf Safe-Lock tubes containing 1 ml of phosphate-buffered saline (PBS) and a 5-mm steel bead (Qiagen #69989) and were homogenized for 5 min at 30 Hz using a mixer mill (Retsch; MM400). Aliquots of the organ homogenates were spread on MacConkey Agar and colonies allowed to grow overnight at 37°C. Suspected E. coli colonies were confirmed via positive indole and negative Simmons-citrate biochemical tests. In total, we analyzed 56 disease-causing or “systemic” E. coli isolates and 38 healthy or “cecal” E. coli isolates, collected from 12 outbreaks over a 2-year period (Supplementary Table 1). E. coli isolates with visually distinct colony morphologies were selected for further analysis.

Bacterial isolates were stored in 20% glycerol at –80°C for long-term preservation. Isolates were inoculated from frozen stocks onto Tryptic soy agar (TSA; Becton Dickinson) and were incubated for 18 h at 37°C. A single colony was inoculated into 5 mL of LB broth and grown overnight at 37°C with shaking at 200 rpm. Overnight cultures were used for all subsequent experiments.

Overnight cultures were standardized to an optical density of 1.0 at 600 nm. Four microliter aliquots were spotted on Congo Red Agar (20 μg/mL of Congo Red dye in 1% tryptone, 0.5% yeast extract, 1.5% agar; pH 7.2–7.4), to visualize the rdar morphotype and curli production (Collinson et al., 1991) and onto Calcofluor White agar [200 mg/L fluorescent brightener #28; (Sigma-Aldrich, St. Louis, MO, United States) in 1% tryptone, 0.5% yeast extract, 1.5% agar] to visualize cellulose production (Solano et al., 2002).

Crystal violet assays were performed using the method of Merritt et al. (2011), with details outlined previously (Liu et al., 2018). For each isolate, 20 μL of standardized inoculum was used to inoculate wells of different 96-well plates, containing 180 μL of three biofilm media: (1) 1/2 tryptic soy broth (TSB), (2) M63 media (ammonium sulfate, potassium monophosphate, and ferrous sulfate), and (3) brain heart infusion (BHI) broth. Each isolate was inoculated into six replicate wells and grown at 28°C for 24 h; negative control wells contained 180 μL of broth and were not inoculated. After 24 h, liquid was removed and each plate was washed twice with 200 μL of PBS. Bacterial cells and extracellular biofilm matrix were fixed by adding 200 μL of methanol to each well and incubating at room temperature for 15 min. The methanol was removed, and plates were left to air dry for 30 min, prior to staining with 200 μL of 0.2% crystal violet for 5 min. Excess crystal violet was rinsed from each well using distilled water. Plates were left to air dry before addition of 160 μL of 33% glacial acetic acid. The absorbance of the resulting solution was measured at 570 nm using a spectrophotometer. The classification system developed by Stepanović et al. (2000) was used to describe biofilm formation by each strain. The cut-off OD (OD_c) was calculated as three standard deviations above the average OD of the negative control wells; this value was compared to the mean OD for each strain. The biofilm categories were strong (mean OD > 4*OD_c); moderate (4*OD_c > mean OD > 2* OD_c); weak (2* OD_c > mean OD > OD_c); or none (OD_c > mean OD).

Escherichia coli biofilm formation and the measurement of disinfectant susceptibility was performed using the MBEC Assay^® biofilm inoculator 96-well plates (Innovotech Inc., Edmonton, AB, Canada) according to previously described method (Harrison et al., 2010). A standardized inoculum of 1.5 × 10⁷ CFU/mL was added to biofilm growth media to a final volume of 150 μL per well. The MBEC plate lid containing 96 polystyrene pegs was placed into the 96-well microtitre plate base, sealed with parafilm and incubated at 37°C for 24 h with slight rocking on a tilting platform shaker. After incubation, peg lids were removed from the base plate and washed twice with sterile PBS to remove loosely attached cells. Biofilm formation on pegs was quantified by viable cell counts following disruption of the biofilm by sonication for 30 min with a bath sonicator (Branson #3510, Canada). Dislodged biofilm cells were serially diluted and grown on Mueller Hinton (MH) agar plates (Liu et al., 2018; Sivaranjani et al., 2021).

The minimum inhibitory concentration (MIC) was determined by broth microdilution according to the Clinical and Laboratory Standards Institute (CLSI) standard for antimicrobial susceptibility testing (CLSI, 2020). Serial twofold dilutions of each disinfectant (Supplementary Table 2) were prepared in MH broth to a final volume of 90 μL per well. Ten μL aliquots of culture, representing 1.5 × 10⁷ bacterial cells, were added to each well, the plates were covered and incubated at 37°C with slight rocking. After 24 h of growth, the OD at 600 nm of each well was measured by an xMarkTM Microplate Absorbance Spectrophotometer (Bio-Rad, Mississauga, ON, Canada). The MIC was recorded as the concentration in the wells where bacterial growth was visibly inhibited. To determine the Minimum Bactericidal Concentration (MBC), 100 μl of aliquots from MIC wells were sub-cultured on MH agar and incubated for 24 h. The lowest concentration that resulted in no bacterial growth on MH agar plates was determined as the MBC.

Antimicrobial susceptibility of all strains was determined via broth microdilution using the Gram-negative panel for the MicroScan system (Beckman Coulter, Mississauga, ON, Canada), testing the following 26 agents: amikacin, amoxicillin-clavulanic acid, ampicillin-sulbactam, ampicillin, aztreonam, cefazolin, cefepime, cefotaxime, cefoxitin, ceftazidime, ceftriaxone, cefuroxime, cephalothin, ciprofloxacin, doripenem, ertapenem, gentamicin, imipenem, levofloxacin, meropenem, nitrofurantoin, piperacillin-tazobactam, tetracycline, tigecycline, tobramycin and trimethoprim-sulfamethoxazole. Chloramphenicol, colistin, and nalidixic acid testing was performed using agar dilution (Wiegand et al., 2008). For each antibiotic, petri plates were prepared containing LB agar (lysogeny broth, 1.5% agar) plus antibiotic ranging from 0.5 μg/mL to 32 μg/mL. Isolates were streaked on blood agar (Oxoid #CM0055) and incubated for 18 h at 37°C. A 0.5 MacFarland standard was made for each isolate and isolates were spotted onto antibiotic plates using a 96-pin microplate replicator. Plates were incubated at 35°C for 24 h before reading MIC values. For quality control, E. coli ATCC 25922, Staphylococcus aureus ATCC 29213 and Pseudomonas aeruginosa ATCC 27853 were included as appropriate.

The results were interpreted according to the CLSI Guidelines (CLSI, 2020). MIC breakpoints were as follows: ampicillin (≥32 μg/mL), cefazolin (≥8 μg/mL), cephalothin (≥32 μg/mL), ceftazidime (≥16 μg/mL), ceftriaxone (≥4 μg/mL), cefuroxime (≥32 μg/mL), cefotaxime (≥4 μg/mL), cefepime (≥16 μg/mL), amoxicillin + clavulanic acid (≥32 μg/mL), ampicillin + sulbactam (≥32 μg/mL), cefoxitin (≥32 μg/mL), aztreonam (≥16 μg/mL), gentamicin (≥8 μg/mL), tobramycin (≥16 μg/mL), nalidixic acid (≥64 μg/mL), tetracycline (≥16 μg/mL), colistin (≥4 μg/mL), chloramphenicol (≥32 μg/mL), ciprofloxacin (≥1 μg/mL), doripenem (≥4 μg/mL), ertapenem (≥2 μg/mL), imipenem (≥4 μg/mL), meropenem (≥4 μg/mL), nitrofurantoin (≥128 μg/mL).

Graphing of data and statistical analyses were performed using GraphPad Prism software v.8.0.2 (San Diego, CA, United States). Data from the disinfectant susceptibility assays were logarithmically transformed and analyzed using the Kruskal–Wallis test with post hoc analysis via Dunn’s multiple tests. Statistical differences in biofilm biomass, as measured by crystal violet staining, were determined using an ordinary-one way ANOVA test [F = 14.28, p > 0.0001]. Distribution of systemic isolates was analyzed using a chi-square test to determine whether the frequency of positive biofilm formation in each media was statistically significant [chi-square statistic: 15.74, df = 2, p > 0.0001].

Genomic DNA was extracted from each E. coli isolate using the GenElute Bacterial Genomic DNA Extraction Kit (Sigma-Aldrich, St. Louis, MO, United States), according to the manufacturer’s instructions. Purity of the DNA was assessed using a NanoDrop spectrophotometer (Thermo Fisher, Ottawa, ON, Canada) and quantified using a Qubit Fluorometer (Thermo Fisher Scientific, Ottawa, ON, Canada).

All E. coli isolates were sequenced on a Nanopore MinION according to the protocol developed by Lohman et al. (2018), with the amount of ligase added to each step calculated based on the expected number of DNA molecules available for sequencing. Small fragments were removed from the genomic DNA eluate by adding 0.4x magnetic beads to each sample (NucleoMag NGS beads, Macherey-Nagel, Allentown, PA, United States). DNA fragments larger than 500 bp were bound, washed in 80% ethanol and eluted in 1 mM Tris buffer, pH 8.0. DNA was stored at –20°C for up to 1 week before sequencing.

Up to 200 fmol of purified and size-selected DNA was added to a Lobind Eppendorf tube. Distilled water was added to a final volume of 30 μL. Nicks and gaps in the DNA were repaired using FFPE DNA Repair Mix (#M6630; New England Biolabs, Whitby, ON, Canada) and polyA-tails were added to DNA fragments using the Ultra II End Repair/dA-Tailing Module (#M7645; New England Biolabs). Nanopore barcodes (#NBD-104; Oxford Nanopore Technologies, Oxford, United Kingdom) were ligated to DNA fragments using Blunt/TA Master Mix (New England Biolabs). Multiplexed samples were pooled into a single Eppendorf tube and the reaction was adjusted to 1 M NaCl. DNA repair enzymes and excess DNA barcodes were removed using a 0.2X magnetic bead clean-up. Beads were washed twice with 80% EtOH and the DNA was eluted in 10 mM Tris, pH 8.0. Sequencing adapters (AMII; Oxford Nanopore Technologies EXP-NBD104) were ligated to the ends of DNA fragments using Quick T4 DNA ligase (New England Biolabs) before a final bead clean-up (1:1 ratio of beads to sample) was performed to remove T4 DNA ligase and excess sequencing adapters from the final purified DNA library. Beads were washed by incubating two times in long chain fragment buffer and resuspending by flicking. DNA was eluted in 1mM Tris Buffer, pH 8.0. The final library was prepared and loaded onto the sequencer as described in the SQK-LSK109 protocol. Sequencing was performed using R.9.4.1 MinION flow cells. Sequencing runs were terminated after approximately 24 h.

Base-calling and de-barcoding of Nanopore reads was performed using MinKNOW (version 2.0) and Guppy (version 2.1.1) software (Oxford Nanopore Technologies). Read quality and lengths for each sequencing run were determined and visualized using the NanoPack software (version 1.35.4) developed by De Coster et al. (2018). Additional read trimming was performed using Porechop to remove any barcodes missed by Guppy (Wick, 2017). Reads were filtered for quality using Nanofilt to ensure that only reads with a Q score of 10 or greater were included in draft assemblies (–q 10) (De Coster et al., 2018).

Rough draft genomes were assembled using Unicycler (version 0.4.8) (Wick et al., 2017). A summary of the characteristics of the 96 E. coli strains selected for draft genome assembly, as well as sequencing information can be found in Supplementary Table 1. Each isolate tested was considered distinct due to having a unique combination of biofilm and AMR profiles.

The phylogroup of each E. coli strain was determined using the Clermontyping tool², a web-based platform that predicts phylogroups based on an in silico quadruplex PCR to screen the whole genome FASTA sequence of an isolate for the arpA, chuA, yjaA, TspE4.C2 and TrpAgpC genes (Beghain et al., 2018).

We screened E. coli draft assemblies for plasmid sequences using PlasFlow, a program that uses a neural network to differentiate plasmid from chromosome sequences and has a higher accuracy than BLAST-based programs such as Plasmid Finder (Krawczyk et al., 2018). The draft chromosome and plasmid sequences were screened for antimicrobial resistance genes using the ResFinder database (ResFinder 4.0) from the Centre for Disease Epidemiology (Bortolaia et al., 2020). Genes were identified and used to generate presence and absence data for all 94 E. coli isolates. In the end, 28 known AMR genes were identified as being present in at least one isolate, encoding for resistance to tetracycline (tetA, tetB), sulfonamides (sul1, sul2), trimethoprim (dfraS), aminoglycosides [aph(6)-Id, aph(3′′), aph(4)-I1a, aph (3′), aadA1, aac(2), aac(3)-IV, aac(3)-VI, aac(3)-2d], and beta-lactam antibiotics (bla_TEM–18, bla_TEM–176, bla_TEM–105, bla_CMY–2, bla_CMY–138, bla_CMY–14, bla_CMY–149, bla_CMY–15, bla_CMY–16, bla_CMY–155, bla_CMY–122, bla_CMY–4, bla_OXA–1). The mdfA gene encodes a multi-drug efflux pump.

A total of 94 E. coli strains were isolated from birds with colibacillosis (n = 56; systemic E. coli) and from presumed healthy birds of the same flock (n = 38; cecal E. coli). These E. coli strains represented 12 outbreaks on Saskatchewan farms between 2019 and 2020. Antimicrobial susceptibility testing was performed on all 94 isolates against 27 different antimicrobials: the results are summarized in Table 1. Resistance to Tetracycline (29.8%), Ampicillin (24.6%), Gentamicin (23.4%), and Tobramycin (19.1%) was the most observed. In contrast, there was no resistance detected to colistin, chloramphenicol, or the carbapenems. 43 isolates were susceptible to all 27 tested antimicrobials, while 32 were resistant to three or more drugs, and nine isolates were resistant to drugs from six different classes (Supplementary Table 1). Overall, the systemic isolates were significantly more drug resistant than the cecal isolates; 71% of systemic E. coli were resistant to at least one drug compared to only 29% of cecal E. coli (Supplementary Table 1). Tetracycline resistance was common in both groups, but resistance to ampicillin, gentamycin, and tobramycin was 20–30% higher in systemic isolates (Table 1). Systemic isolates also had higher MIC values for first and third generation cephalosporins, compared to cecal isolates (Table 1). Resistance to cefoxitin was found exclusively in systemic isolates, while Cefepime resistance was exclusively identified in cecal isolates.

Biofilm formation is associated with bacterial persistence, and we wanted to determine if this physiological state was correlated with antibiotic resistance. All E. coli isolates were screened for biofilm-forming ability in three different liquid media: nutrient rich BHI broth versus relatively nutrient poor ½ TSB and M63 media. Crystal violet staining was used to quantify and classify the level of biofilm. Overall, the systemic isolates generated significantly more biomass in nutrient rich BHI broth, as compared to the other two media (Figure 1A), whereas the cecal isolates did not show a clear preference for any one media type (Figure 1B).

To classify the level of biofilm formation for individual isolates, we used the formula described by Stepanović et al. (2000) (see section “Materials and Methods”). Isolates were categorized as positive if they formed strong or moderate biofilms, while isolates that produced none or weak biofilms were categorized as negative. 72% of systemic isolates were positive for biofilms in BHI broth, as compared to only 55% of cecal isolates (Figure 2). In contrast, 53% of cecal isolates were biofilm positive in M63 media, as compared to only 23% of systemic isolates (Figure 2). In general, systemic isolates were significantly worse at forming biofilms in nutrient poor media, with 28% positive in ½ TSB and 23% positive in M63 media (X², p < 0.005) (Figure 2). In contrast, a roughly equal proportion of cecal isolates formed biofilms in M63 (53%) as BHI broth (55%), suggesting that these strains are equally capable of forming biofilms in nutrient-poor and nutrient-rich conditions (Figure 2). Approximately 30% of cecal isolates formed no biofilm in each media tested (Figure 2).

We also screened all E. coli isolates for curli and cellulose production since these polymers are central to biofilm formation in the host and the environment (White et al., 2011; MacKenzie et al., 2017; Chuppava et al., 2019). Sixty-two isolates produced both curli and cellulose, distributed as 73% (41/56) of systemic isolates and 55% (21/38) of cecal isolates (Supplementary Table 1). Twenty isolates produced only curli, consisting of 10 systemic and 10 cecal isolates. Nine isolates produced only cellulose, consisting of five systemic and four cecal isolates, and five cecal isolates were the only strains negative for both curli and cellulose. Taken together, the results indicated that systemic isolates had a slight increase in biofilm phenotypes as compared to cecal isolates.

We performed additional screening of 55 isolates that were classified as strong biofilm formers in BHI, M63 and ½ TSB media (Supplementary Table 1). We tested the ability of strains to adhere and form biofilms on polystyrene pegs in MBEC biofilm inoculator plates. Rather than measure biomass by staining, we focused on live cell counts, looking for strains with cell counts greater than 10⁷ CFU, which we considered an appropriate cut off for good biofilm formation, based on previous MBEC assays (Liu et al., 2018). In total, 21 isolates formed biofilms with cell counts above 10⁷ CFU (Figure 3; stars). consisting of 11 systemic and 10 cecal E. coli.

We further screened 12 of these isolates (7 systemic + 5 cecal) for resistance to four commercial disinfectants (Supplementary Table 2). Each disinfectant was able to kill planktonic cells of all E. coli isolates at low concentrations, ranging from 0.0016-0.0031% for Virocid, 0.125-0.25% for Virkon, 1.9–3.8 ppm for the quaternary ammonium disinfectant DDAC and 0.031–0.063% for H₂O₂ (Supplementary Table 3). The MBC values of all four disinfectants were either equal or one serial dilution higher than the corresponding MIC values. We deemed these results to be somewhat misleading because MIC and MBC determination requires exposure to test antimicrobials for 24 h, whereas most common disinfectants require contact times of only minutes to achieve bactericidal effects. Therefore, we repeated disinfectant screening on planktonic and biofilm cells of each of the 12 E. coli strains with a 30 min exposure. The starting cell numbers for each strain ranged between 10⁶ and 10⁸ CFU (Supplementary Table 4). Virkon and Virocid disinfectants exhibited effective bactericidal activity against both cell types of all 12 isolates even at low concentrations (Table 2 and Supplementary Figures 1, 2). The concentration of Virocid required to achieve full bactericidal effect was 1/2- to 1/32- fold lower than the suppliers’ recommended concentration, and for Virkon it was reduced approximately by 1/2- to 1/8- fold (Table 2 and Supplementary Table 2). DDAC was effective at killing planktonic cells of the 12 E. coli isolates with bactericidal concentrations ranging from 4.7 to 9.4 ppm (Table 2). However, biofilm cells of all 12 isolates had 2- to 64-fold increased resistance to DDAC, and one of the systemic isolates was not killed at the manufacturer recommended concentration of 300 ppm (Table 2 and Supplementary Figure 3). Higher concentrations of DDAC were required to eradicate the biofilms of three systemic isolates as compared to the five cecal isolates (Table 2). The results with H₂O₂ were similar to DDAC, as planktonic cells of all tested isolates were susceptible to low concentrations, but biofilms cells had 4- to 64-fold increased resistance (Table 2 and Supplementary Figure 4). Again, 4 systemic isolates were more resistant and required 4% H₂O₂ for the complete eradication of biofilms, whereas all five cecal isolates were susceptible to concentrations of 2% or less (Table 2 and Supplementary Figure 4). To determine if these trends based on isolate source were consistent, we would need to analyze more E. coli strains.

As a final comparison of the E. coli strains analyzed, we divided each group of isolates from outbreak to outbreak. This perhaps has more biological relevance than comparing the large groups of systemic and cecal strains to each other, since in almost all cases outbreaks occurred on different poultry farms. We identified 20 unique antibiotic resistance profiles across 12 Saskatchewan colibacillosis outbreaks (Supplementary Table 1). The most common resistance pattern was ampicillin + gentamicin + tobramycin, which was identified in isolates from 7 of 12 outbreaks. In all but three outbreaks, we identified systemic E. coli isolates that were resistant to at least four drugs. While the cecal isolates were generally susceptible, we identified at least one drug resistant cecal isolate in half of the outbreaks. In two outbreaks, cecal isolates were identified that were resistant to more than four drugs. This comparison again revealed an overall trend for systemic E. coli isolates to be more antibiotic resistant than cecal isolates, but there was a lot of diversity and no absolute patterns.

For each outbreak, disease-causing E. coli were isolated from up to three diseased birds and cecal E. coli were isolated from up to three healthy birds from the same flocks. Strain profiles based on the combination of curli, cellulose and AMR phenotype screening, indicated that the 94 isolates did not consist of groups of identical clones (Supplementary Table 1).

We determined the phylogroup of each E. coli strain using the Clermontyping tool (see footnote 2). Systemic isolates were primarily distributed into phylogroups G (32.8%), A (20.7%), and D (18.9%) (Supplementary Table 1). This was consistent with extraintestinal pathogenic E. coli, with phylogroups G and D known to contain many pathogenic E. coli strains (Clermont et al., 2000, 2019). In contrast, the majority of Saskatchewan cecal strains were distributed in phylogroups A (60.5%) and B1 (15.8%) (Supplementary Table 1), which are phylogroups known to typically contain commensal and environmental E. coli isolates (Duriez et al., 2001). The wide genetic diversity identified in the pool of systemic isolates suggests that transmissible elements, such as plasmids, may play a role in infection (Sola-Gines et al., 2015).

The E. coli Nanopore draft assemblies were analyzed for plasmid sequences using PlasFlow (Krawczyk et al., 2018). We found that the 56 systemic isolates possessed an average of four plasmids (range from 0 to 9), while 38 cecal isolates possessed an average of 1 plasmid (range from 0 to 4) (Supplementary Figure 5). Next, we screened draft chromosome and plasmid sequences for antimicrobial resistance genes to see if we could identify genes unique to either group of isolates. We used the publicly available ResFinder 4.0 database from the Centre for Disease Epidemiology (Bortolaia et al., 2020). Due of the high error rates of Nanopore sequencing, and a lack of any polishing or consensus step in the Unicycler assembly program, we did not search for AMR acquired via point mutations.

The chromosomal and plasmid DNA sequences for each isolate were screened for the presence of 28 known AMR genes. We identified few chromosomal antimicrobial resistance genes in our E. coli population, except for mdfA, an outer membrane transporter that confers resistance to quaternary ammonium compounds, the active ingredient in a common commercial barn disinfectant. A small number of systemic isolates had chromosomal TetA genes, and four cecal isolates had chromosomal aminoglycoside or sulfonamide genes (Figures 4A,B). The remaining AMR genes identified in both groups were present on plasmids (Figures 4C,D). Over 25% of systemic isolates possessed plasmid-encoded tetracycline [tet(A)], beta-lactam (blaTEM-1B), and aminoglycoside [aac(3)-2d, aph(3”)] resistance genes (Figure 4C). Plasmid-mediated aminoglycoside resistance [aph(6)-Id, aac(3)-IV] was the most common type found in cecal isolates, at 9% abundance (Figure 4D). Almost half (41%) of cecal isolates lacked any plasmid-encoded AMR genes, compared to only 10% of systemic isolates.