This study investigated the occurrence, phenotype, and genotype of ESBL-producing Escherichia coli isolates collected between January and August 2023. A total of 956 samples were analyzed, covering multiple sectors and ecological contexts. All sample collections were conducted exclusively in 2023, following harmonized procedures for traceability and reproducibility. Detailed descriptions of sample sources, collection methods, and locations are provided below.

Samples from Livestock and food production (n = 519) were collected from breeding farms, slaughterhouses, food processing plants, and retail establishments handling food of animal origin. The farm-level sampling included broiler chickens, fattening pigs, sheep, and beef cattle. At each site, environmental and animal samples (including feces, rectal swabs, and intestinal content) were collected by trained veterinary staff using sterile tools and following national biosafety guidelines. Slaughterhouse sampling targeted carcasses and surfaces in contact with animal products. Processing plant and retail samples focused on raw food items of animal origin, such as meat cuts and offal. These activities were coordinated with regional veterinary services under official monitoring programs.

Samples from wildlife (n = 301) were collected during regional passive surveillance activities for African Swine Fever (ASF) and Chronic Wasting Disease (CWD). The sampling was conducted by authorized wildlife officers and veterinarians across multiple sites in the Abruzzo and Molise regions. Animal carcasses found dead or killed in road accidents were examined. Samples included intestinal contents, fresh feces, and internal organ tissues. All samples were geo-referenced using handheld GPS devices and coded for traceability.

Samples were collected from companion animals, domestic dogs (n = 99) and cats (n = 18) admitted to veterinary clinics for diagnostic or preventive purposes. Rectal swabs and fecal samples were collected during routine examinations or upon hospitalization, always with informed consent from the pet owners. For one cat, a brain tissue sample was included, collected post-mortem for unrelated diagnostic purposes.

Human clinical E. coli isolates were obtained from blood and urine samples of 19 patients in a local hospital in the same geographic area. These samples were provided by the hospital’s microbiology laboratory under ethical approval, and all isolates had been previously confirmed as ESBL-positive through standard diagnostic workflows. As no information was available about the total number of tested patients or ESBL-negative isolates, human samples were excluded from statistical comparisons.

Proportions were calculated with exact binomial 95% confidence intervals (Clopper-Pearson method). Group comparisons used Pearson’s χ² test with post-hoc Fisher’s exact tests (Bonferroni-adjusted for multiple comparisons). Trend analysis employed the Cochran-Armitage test. Effect sizes included odds ratios (Baptista-Pike 95% CIs) and absolute risk differences. The Number Needed to Sample (NNS) was derived as the reciprocal of risk differences. All analyses were performed in R (v4.3.1) with α = 0.05. Additional methodological details are provided in Supplementary material 1.

Samples were collected using standardized protocols to ensure representativeness. All samples were transported in temperature-controlled containers at 4 ± 2°C and delivered to the laboratory within 2 h. Fecal samples (1 g ± 0.1 g) were homogenized in 9 ml of sterile buffered peptone water (BPW), while food samples (25 g ± 0.1 g) were homogenized in 225 ml of BPW.

Samples collected in buffered peptone water (BPW; Oxoid, United Kingdom) were incubated under aerobic conditions at 37 ± 1°C for 24 h. Subsequently, 10 μl of each enriched broth were plated onto MacConkey agar supplemented with 4 μg/ml cefotaxime (Sigma-Aldrich, Germany) to select for cefotaxime-resistant Escherichia coli. The inoculated plates were then incubated aerobically at 37 ± 1°C for 18 to 22 h.

Subsequently, 10 μl of each enriched broth were plated onto MacConkey agar supplemented with 4 μg/ml cefotaxime (Sigma-Aldrich, Germany) to select for cefotaxime-resistant Escherichia coli. The inoculated plates were then incubated aerobically at 37 ± 1°C for 18 to 22 h. Up to three presumptive E. coli colonies were subcultured onto fresh MacConkey Agar plates (4 μg/ml cefotaxime) and incubated as above. Colonies demonstrating consistent growth were considered potential ESBL-producing E. coli, confirmed colonies were stored in Microbank™ at −80°C.

Species identification was performed using the MALDI Biotyper system (Bruker Daltonics, Billerica, MA, USA) with the MALDI Biotyper Compass software version 4.1.

Antimicrobial susceptibility testing (AST) was performed on E. coli isolates grown on McConkey agar supplemented with cefotaxime to confirm ESBL production. Antimicrobial susceptibility testing (AST) was performed using the disk diffusion method in accordance with current EUCAST guidelines (EUCAST, s.d.). Epidemiological cut-off values (ECOFFs) established by EUCAST were applied to interpret the results in order to capture the presence of acquired resistance mechanisms such as ESBLs, as phenotypic breakpoints for some strains may not reflect resistance accurately in non-clinical isolates. Briefly, 0.5 McFarland suspensions of E. coli were inoculated onto Mueller-Hinton agar plates (Bio-Rad, France). Antibiotic disks were applied, and plates were incubated at 35 ± 1°C for 18 ± 2 h in aerobic conditions (Åhman et al., 2022). ESBL production was assessed using a combined disk diffusion test with cefotaxime/clavulanic acid (30 μg + 10 μg) and ceftazidime/clavulanic acid (30 μg + 10 μg) disks (ESBL disc kit, Liofilchem®, Roseto degli Abruzzi, Italy). E. coli ATCC 25922 and K. pneumoniae ATCC 700603 served as negative and positive controls, respectively. Inhibition zone diameters were measured, and an increase of ≥5 mm in the zone around the combination disks (ceftazidime/clavulanic acid or cefotaxime/clavulanic acid) compared to the corresponding single-antibiotic disks (ceftazidime or cefotaxime) was interpreted as ESBL production (Fröding et al., 2016).

ESBL-positive proportions were calculated with 95% confidence intervals using the Clopper-Pearson exact method. Between-group comparisons were performed using the Pearson chi-square test, followed by pairwise comparisons using the Fisher exact test and Bonferroni correction for multiple comparisons. Trend analysis was performed using the Cochran-Armitage test. Effect sizes were expressed as odds ratios (with 95% confidence intervals calculated according to the Baptista-Pike method) and absolute risk differences. The NNS was derived as the inverse of the risk difference. Logistic regression analysis was also performed to assess the association between different groups and ESBL positivity. All analyses were performed using R (v4.3.1) with a significance level of α = 0.05. Further methodological details are available in Supplementary material 1.

The samples were analyzed using R per statistical tests, including the Chi-square test to determine any differences between categories, the Pearson correlation test to analyze the relationship between the number of samples and the percentage of ESBL-positive, and the Z test to compare the categories. Human samples were excluded from the statistical analysis. This is because the human samples provided by the source had already tested positive for ESBL, without any information on the number of ESBL-negative samples. As a result, the percentage of negative samples in the human population could not be calculated, making it difficult to compare the human data with those of other categories.

Spatial analysis of samples was performed using QGIS and SaTScan to identify significant clusters of ESBL resistance. Data on cases, controls, and geographical coordinates were entered into SaTScan to perform spatial analysis and identify potential spatial clusters. The Bernoulli model in SaTScan was selected to detect clusters by comparing the observed and expected distributions of binary case–control data within scanning windows, identifying areas with a statistically significant excess of cases based on the Bernoulli probability model (Supplementary material 1).

A subset of 120 E. coli isolates was selected from among those that tested positive for ESBL production, using a stratified approach based on sample type (animal, food, human), geographic origin, and matrix. This strategy was adopted to ensure representative coverage of sources and to characterize the genetic diversity of resistance among ESBL-producing strains. Total genomic DNA was extracted from 120 Escherichia coli ESBL positive isolates using the Maxwell 16 Tissue DNA Purification Kit, following the standard protocol provided by the manufacturer. The concentration of the extracted DNA was measured using the Qubit DNA HS Assay (Thermo Fisher Scientific Inc., Waltham, MA, USA). Genomic DNA was sequenced using the Illumina NextSeq500 platform. Briefly, sequencing libraries were prepared using the Nextera XT kit (Illumina, San Diego, CA, USA) according to the manufacturer’s instructions and subsequently sequenced in paired-end mode (150 bp) using the NextSeq500/550 Mid Output v2 reagent cartridge. After demultiplexing and adapter removal, the quality of the reads was assessed using FastQC v0.11.5 (Andrews, 2010). Raw reads were further processed using Trimmomatic v0.36 (Bolger et al., 2014) with the following parameters: Leading: 25; Trailing: 25; Slidingwindow: 20:25. Genome scaffolds were assembled using SPAdes v3.11.1 with the following parameters: –k 21, 33, 55, 77; −careful (Bankevich et al., 2012). The quality of the assembled scaffolds was evaluated with QUAST v4.3 (Gurevich et al., 2013). The set of paired-end genome sequencing reads from E. coli obtained in this study was deposited in the Sequence Read Archive (SRA) and associated with Bioproject PRJNA1218927.

Genomic DNA from 120 high-quality E. coli isolates was subjected to whole-genome sequencing and analyzed using several bioinformatics tools. Plasmid incompatibility (Inc) groups and beta-lactamase genes, including ESBL genes, were identified using ABRicate v1.0.1 with 100% coverage criteria and the PlasmidFinder (Carattoli et al., 2014), NCBI (Feldgarden et al., 2020), and ResFinder (Zankari et al., 2012) databases (all updated February 21, 2023). MOB-recon module from MOB-suite v3.0.0 (Robertson and Nash, 2018) was used to predict plasmidic or chromosomal localization of ESBL genes using default settings. Flankophile (Thorn et al., 2024) was employed to analyze the synteny of blaCTX-M-1 and blaCTX-M-15 by comparing their flanking regions. We utilized single linkage clustering, which groups isolates iteratively: an isolate joins an existing cluster if its allelic distance to any member already within that cluster is less than or equal to the defined threshold. The threshold of 10 alleles corresponds to the default setting for this task template in SeqSphere+. We explored the additional cut-offs of 5 and 20 alleles to evaluate the stability of the clustering results across different thresholds. cgMLST data was used to generate a minimum spanning tree (MST) and UPGMA tree, which were visualized using iTOL v7 (Letunic and Bork, 2019). Data visualization of antimicrobial resistance gene presence was performed using RAWGraphs (Mauri et al., 2017). Novel MLST sequence types were submitted to EnteroBase for designation.

The analysis of 956 samples revealed the presence of ESBL-producing E. coli in 135 samples, corresponding to 14.12% of the total. The distribution of ESBL-producing E. coli varied among the host categories analyzed. In particular, the highest positivity rate was observed in companion animals, with 19 positive samples out of 117 tested (16.24%), followed by farm animals and environmental matrices with 76 positive samples (14.64%), and wildlife with 21 positive samples out of 301 (6.98%), according to antimicrobial susceptibility testing (AST) results (Supplementary material 1; Table 1).

The ESBL positivity among the categories, the percentage of positives and the 95% confidence intervals are as follows: among wild animals the percentage is 6.98%, with a confidence interval between 4.38 and 10.41%; for environmental and food matrices is 14.64%, with a confidence interval between 11.69 and 18.05%; finally, for companion animals, the positivity of ESBL is 16.24%, with a confidence interval ranging from 10.08 to 24.18%.

Statistical analysis shows significant differences between some of the categories compared.

Among non-human sample, the occurrence of ESBL-producing E. coli was significantly higher in environmental and food production samples (14.6%) and in companion animals (16.2%) compared to wildlife (7.0%). On pairwise comparison, environmental samples were significantly more positive than in wild animals (absolute risk difference: −7.66 percentage points; p = 0.0008; OR = 0.44, 95% CI: 0.26–0.73), with NNS of 13. Companion animals were also significantly more positive than in wildlife (risk difference: −9.26%; p = 0.0027; OR = 0.39, 95% CI: 0.20–0.75), with an NNS of 11. No significant difference was found between environmental and companion animals (p = 0.67; OR = 0.88; RD = −1.60%).

A significant increasing trend in ESBL positivity was detected along the gradient from wild animals to environmental samples to companion animals, consistent with progressive exposure to selective pressures or anthropogenic influence (Cochran-Armitage test, Z = 3.24, p = 0.0012).

Finally, logistic regression also confirmed that both companion animal (OR = 2.36, 95% CI: 1.30–4.31, p = 0.005) and environmental samples (OR = 2.10, 95% CI: 1.25–3.55, p = 0.005) had higher odds of ESBL positivity relative to wildlife, once more in favor of the observed associations. (Supplementary material 1; Table 1).

SaTScan’s space-scan statistics identified spatial areas with observed occurrences that are higher than those expected by chance for Bernoulli-distributed variables. In particular, a cluster of ESBL-positive strains in the northeastern region was identified, with an observed to expected case ratio of 1.72. This cluster demonstrated a 5.34-fold increased probability of ESBL-producing E. coli isolation compared to the regional average (p < 0.001), indicating a significantly higher concentration of positive cases within this area (Supplementary material 1; Figure 1).

Whole-genome sequencing (WGS) was performed on 120 E. coli strains phenotypically confirmed as extended-spectrum ESBL producers. Multilocus sequence typing (MLST) identified 58 sequence types (STs), including three novel STs (Supplementary material 1). While most STs were represented by a single strain, some exhibited higher occurrence. Notably, ST-131 was the most frequent among human isolates (n = 12/19), ST-10 was frequently identified among isolates from dogs, poultry, and cattle (n = 10/64), although not exclusively dominant in each category. Additionally, ST-206 was identified in eight isolates from swine, representing the most frequent ST within that host group.

To further resolve the genetic relatedness among these isolates, core genome MLST (cgMLST) and cluster analysis were conducted (Supplementary Figure 1). Pairwise comparisons revealed a maximum allelic distance of 2,421 alleles, out of 2,513, examined between any two strains. Cluster analysis, employing allelic difference cut-offs of 5, 10, and 20 alleles, yielded similar numbers of clusters (16, 16, and 14, respectively), with the majority of clusters comprising only two strains. At the default threshold of 10 alleles, the largest cluster encompassed five strains isolated from diverse hosts, including two wild animals, two dogs, and one human, with pairwise allelic distances within this cluster ranging from 4 to 14. Interestingly, another cluster linked a human isolate with an isolate from a wild boar, separated by just 8 alleles. The remaining clusters were composed of strains originating from the same host species. Despite the predominance of ST-131 among human isolates, pairwise comparisons within this ST revealed substantial genetic diversity, with allelic distances generally exceeding 50 alleles. This high degree of genetic variation, even within the STs, underscores the significant diversity within the ESBL-producing E. coli population in Abruzzo. Among the sequence types identified in more than one host species, ST69 was found in human, wild boar, dog, porcupine, and deer isolates. While most ST were specific for a single isolation source, one of the most common STs, ST-10 contained strains from cattle, sheep and dogs. Within this ST we examined inter-source genomic diversity, which revealed considerable genetic distances between isolates from different hosts (maximum intra-ST distance: 992 alleles). The minimum pairwise cgMLST distance between ST-10 isolates from dogs and cattle was 294 alleles, whereas minimum distances involving poultry were substantially larger (562 alleles for dog-poultry; 591 alleles for poultry-cattle). Even higher maximum pairwise distances were observed between ESBL-producing E. coli strains isolated from different source types and assigned to different STs (>2,000 alleles).

To further explore the distribution of genetic variation, we examined the intra-host type diversity using cgMLST pairwise allelic differences (Supplementary material 3). Substantial genetic variation was evident within multiple host populations, with maximum pairwise differences exceeding 2,100 alleles in isolates from cattle, dogs, poultry, swine, wild boar, other wild animals, and sheep. Furthermore, multiple MLST Sequence Types (STs) were identified within most host groups, reinforcing the genetic heterogeneity observed within these populations. This finding demonstrates that the high overall genetic diversity of ESBL-producing E. coli observed in the Abruzzo region is not concentrated within a single host reservoir but is broadly distributed across the various animal sources investigated.

To investigate diversity at the farm level, we examined intra-farm cgMLST pairwise allelic differences (Supplementary material 3). Notably, substantial genetic diversity was present even within the confines of single farms. Several poultry, cattle, and swine farms (Farms 1, 3, 4, 8, 9) exhibited high maximum pairwise differences, often exceeding 2,300 alleles, indicating the co-circulation of highly divergent ESBL-producing E. coli lineages. This intra-farm diversity was also reflected in the detection of multiple MLST STs (up to 5 per farm). While some farms showed more limited diversity (e.g., Farm 2, Farm 7) or contained identical isolates (minimum distance of 0 on Farms 1, 2, 3, 5, 8), the overall data underscores that significant genetic heterogeneity can exist within individual farm environments in the Abruzzo region.

Analysis of β-lactamase genes, including ESBL genes and AmpC enzymes, revealed that most strains harbored multiple bla genes (Figure 2; Table 2), although ESBL genes were never found in duplicate unless in combination with AmpC genes. The majority of strains (76/120) carried ESBL genes belonging to the bla_CTX-M class, followed by bla_SHV-12 (30/120). Within the bla_CTX-M class, the most prevalent variants were bla_CTX-M-1 (30/120) and bla_CTX-M-15 (29/120). Notably, bla_CTX-M-1 was the dominant variant in swine isolates (13/15), whereas bla_CTX-M-15 was frequently identified in human isolates (12/19). Additionally, bla_SHV-12 was prevalent in poultry isolates (15/21). In wild animals, both bla_SHV and bla_CTX-M genes were detected, with bla_CTX-M-15 predominating. Interestingly, specific ESBL gene variants were often found in strains isolated from the same host species (Figure 3A), even when those strains were phylogenetically distant, as demonstrated by cgMLST. This distribution of bla genes could therefore suggest potential host-specific adaptation and dissemination patterns of ESBL genes among E. coli strains in Abruzzo.

Figure 3B illustrates the distribution of ESBL genes in strains carrying specific plasmid incompatibility (Inc) groups. The data reveal complex plasmid carriage, with individual isolates often harboring multiple plasmids or hybrid plasmids encompassing more than one Inc. group. ESBL genes frequently co-occurred with IncF plasmids, which were the most commonly identified plasmid type among our ESBL-producing E. coli isolates. All bla_CTX-M genes were associated with the presence of multiple Inc. groups, most commonly including IncF. bla_SHV-12 showed a similar pattern, frequently found alongside IncF and IncI, often in combination. Several ESBL-producing strains lacked detectable Inc. groups. Due to the frequent assembly of ESBL genes, in particular bla_SHV-12 and bla_CTX-M-14, within short contigs using short-read sequencing, we were unable to reliably determine the specific plasmid localization of these genes. Table 3 shows the distribution of the main ESBL genes in the different plasmid replicons.

For ESBL genes located on longer contigs with at least 1,500 bp of flanking sequence, we analyzed gene synteny using Flankophile. This analysis included bla_CTX-M-1 from 9 E. coli strains and bla_CTX-M-15 from 15 E. coli strains. A distance tree based on flanking region sequences (Figure 4) revealed three main clades: two corresponding to bla_CTX-M-1 and one to bla_CTX-M-15. These clades were further subdivided into branches or smaller clusters, often containing strains from the same host. Notably, we identified a bla_CTX-M-15 cluster containing strains from diverse hosts (e.g., dog, wild boar, human, and cattle) exhibiting identical gene synteny despite significant genomic divergence by cgMLST. This suggests potential horizontal gene transfer of the ESBL genes via mobile elements rather than clonal expansion. Intriguingly, gene location prediction indicated that bla_CTX-M-15 in six of the seven strains within this cluster was likely chromosomally located. These findings highlight the complex dynamics of ESBL gene dissemination, involving both clonal expansion and horizontal gene transfer, and suggests the potential for a stable chromosomal integration of these resistance genes.