Monitoring within-farm transmission dynamics of antimicrobial-resistant Campylobacter in dairy cattle using broth microdilution and long-read whole genome sequencing

Campylobacter jejuni and Campylobacter coli are important foodborne zoonotic pathogens and cause for concern due to the increasing trend in antimicrobial resistance. A long-run surveillance study was conducted in animals from different age groups in five dairy cattle farms to investigate the within-farm diversity and transmission dynamics of resistant Campylobacter throughout time. The resistance phenotype of the circulating isolates (170 C. jejuni and 37 C. coli) was determined by broth microdilution and a selection of 56 isolates were whole genome sequenced using the Oxford-Nanopore long-fragment sequencing technology resulting in completely resolved and circularized genomes (both chromosomes and plasmids). C. jejuni was isolated from all farms while C. coli was isolated from only two farms, but resistance rates were higher in C. coli than in C. jejuni and in calves than in adult animals. Some genotypes (e.g. ST-48, gyrA_T86I/tet(O)/blaOXA-61 in farm F1; ST-12000, aadE-Cc/tet(O)/blaOXA-489 in F4) persisted throughout the study while others were only sporadically detected. Acquisition of extracellular genes from other isolates and intracellular mutational events were identified as the processes that led to the emergence of the resistant genotypes that spread within the herds. Monitoring with Oxford Nanopore Technologies sequencing helped to decipher the complex molecular epidemiology underlying the within-farm dissemination of resistant Campylobacter.

Campylobacteriosis is the main cause of human bacterial gastroenteritis in industrialized countries with the species Campylobacter jejuni and Campylobacter coli being the primary contributors. After poultry, cattle are considered a major source of Campylobacter transmission to humans through the consumption of contaminated food and/or water or by direct contact with the animals or their faeces 1 . Since Campylobacter infections are usually self-limited, antimicrobial therapy is only indicated in systemic and severe infections. Nevertheless, antimicrobial-resistant campylobacters are a matter of concern because they put infection treatment options at risk. Fluoroquinolones (FQs) and tetracyclines, often prescribed as empirical therapy for diarrhea 2 , have reduced efficacy due to the high levels of resistance. Macrolides have therefore become the antimicrobials of choice for laboratory-confirmed cases of severe campylobacteriosis 3,4 .
Our previous findings based on cross-sectional epidemiological studies carried out in the Basque Country (northern Spain) showed that dairy cattle represent an important reservoir for resistant Campylobacter 5,6 and that the prevalence of resistance to medically critically important antimicrobials, such as FQs, is increasing whereas resistance against macrolides remains low 6,7 . Acquisition of certain antimicrobial resistance traits has been associated with increased bacterial fitness (e.g. the C257T point mutation in the gyrA gene) 8 www.nature.com/scientificreports/ others lead to fitness decrease (e.g. the point mutation 2075G in the 23S rRNA gene) 10,11 . To some extent, this might explain the currently observed high rates of FQ resistance and the low rates of macrolide resistance in Campylobacter spp. from livestock 12 .
Several studies have reported variability in the faecal shedding patterns of Campylobacter associated with age and differences in the within-farm persistence of distinct genotypes [13][14][15] . However, the within-herd transmission dynamics of resistant genotypes have not been fully elucidated, and gaps still exist regarding the spread of antibiotic resistance. Considering the high correlation between resistance inference based on phenotype using minimum inhibitory concentrations (MIC) and genotype using whole genome sequencing (WGS) in Campylobacter 16 and the power of WGS for the in-depth genomic characterization of bacteria 17 , here, a longitudinal study was carried out in five dairy cattle farms for a period of 16 months to investigate whether certain resistant genotypes could be more adapted to long-term herd colonization. Phenotypic antimicrobial susceptibility testing and WGS were combined to characterize the isolates recovered from animals of different age groups (calves, heifers, and lactating cows) collected over time. To assure complete assemblies of circular bacterial genomes and plasmids and obtain accurate information about the location of the antimicrobial resistance genes (ARGs), long-read sequencing based on the Oxford Nanopore technology (ONT) was used. Understanding the complex epidemiology and dynamics of the acquisition and spread of antibiotic resistance in C. jejuni and C. coli within dairy cattle farms would aid in setting baselines for recommendations for on-farm management practices.

Results
Antimicrobial resistance profiles distribution. Presence of Campylobacter spp., as determined by realtime PCR analysis of a loopful of bacterial growth on CASA® plates, was detected in 98.6% of the samples, 88.8% being PCR-positive to C. jejuni, C. coli or both. Whereas C. jejuni was isolated in all farms, C. coli was only found in two of them, i.e., in F4, where C. coli was the predominant species in all age groups, and in F5, where C. coli was only isolated from calves while C. jejuni was found in all age groups.
The different profiles of microbiological resistance resulting from the combination of resistance to antimicrobial agents with MICs above the ECOFF and their distribution within each farm are represented in  different profiles of resistance were observed in each farm along the study, diversity being higher in lactating cows and heifers compared to calves, and in F5 and F4 compared to the other farms. Two multidrug resistance (MDR) profiles (three or more classes of antimicrobial agents) were identified, i.e., CIP-NAL-STR-TET, present in C. jejuni isolates recovered from all farms except F2 and widespread in C. coli from F4, and ERY-CIP-NAL-STR-TET, only detected in C. coli recovered from calves in F5. The profile CIP-NAL-TET predominated in F1, F2 and F3. In F4, all but one resistant isolate were resistant to STR alone or in combination with TET and CIP/NAL. In this farm, the predominant profile of resistance in C. jejuni (5/6) was CIP-NAL-STR-TET while in C. coli 3 different profiles were identified. In F5, despite the low number of isolates recovered (only 5 samplings), diversity was the highest and included 4 C. jejuni profiles and the MDR C. coli profile ERY-CIP-NAL-STR-TET in calves.
WGS output and assembly results. A selection of 56 isolates from F1 and F4 (one per Campylobacter species, age group, and sampling time) was analysed by WGS to thoroughly compare the genomic profile of the strains circulating in these farms. ONT sequencing of these isolates provided a median of 23,393 reads per sample (IQR = 19,502-29,183) in a median of 302 Mb per sample (IQR = 280-314 Mb) corresponding to a median coverage of 157X (Q1-Q3 = 143-165 X) (Supplementary Dataset S1). Upon assembly, 53 isolates resulted in circularized chromosomes. In all cases, the chromosome size of the assembled draft genome corresponded to the expected size of C. jejuni (median = 1,740,992 bp, IQR = 1,690,866-1,743,662 bp) or C. coli (median = 1,661,122 bp, IQR = 1,660,938-1,681,616 bp) and G + C content (median = 30.43% and 31.49%, respectively).
Plasflow identified 14 plasmidic contigs and another 12 were proved compatible as plasmids after querying them with blastn. These twenty-six contigs corresponded to complete circular plasmids, 21 in C. jejuni and 5 in C. coli, that showed high degree of identity with several plasmids from GenBank (Supplementary Dataset S2). ProgressiveMAUVE multiple alignments identified 4 different plasmid structures with a high degree of synteny, and two of them carrying ARGs. Thus, ARG harbouring plasmids in C. jejuni (18 isolates from F1 and 1 isolate from F4) had an average size of ca. 44,650 bp and were highly similar among themselves. Specifically, a similar Tet-plasmid was found in C. jejuni of ST-48 and ST-19 types circulating in F1 whereas the plasmid in ST-45 (C0944) was slightly smaller (43,961 bp) but highly homologous. A slightly larger plasmid (45,150 bp) was found in a C. jejuni isolate (C0775) recovered from F4 ( Supplementary Fig. S2A). The only resistance gene present in C. jejuni plasmids was tet(O). The ARG-encoding plasmids in C. coli (3 isolates from F4), sized ca. 47,150 bp, were identical among themselves, and in addition to the tet(O) gene they harboured an aminoglycoside cluster (Δant(6)-Ia-sat4-aph(3')-IIIa) ( Supplementary Fig. S2B). Plasmids without ARGs were present in 2 C. jejuni isolates (sized ca. 4,365 bp) and 2 C. coli isolates (sized ca. 26,700 bp), all recovered from F4.
Distribution of genetic determinants of resistance in F1 and F4: both farms differed regarding their diversity and location. Screening for genetic determinants of resistance (GDRs) identified 7 acquired ARGs and point mutations in two other genes associated with resistance to antimicrobials representing 4 different classes (Fig. 2). The combination of GDRs detected in each isolate resulted in 13 different genotypic profiles of resistance (Supplementary Dataset S3). There was an overall very good concordance between susceptible pheno-and genotypes; the only discrepancies were 3 C. coli from F4 (C0698, C0700, and C0777) that were phenotypically resistant to STR (MIC = 8 mg/L) but did not carry any GDR associated with STR resistance, and one C. jejuni from F1 (C0980) that carried the tet(O) gene in a plasmid but was susceptible to TET (MIC ≤ 0.5 mg/L). Resistance to β-lactams was not phenotypically tested but different bla OXA genes were identified in 47 isolates; 5 C. jejuni isolates from F1 carried the bla OXA-184 gene and 42 isolates carried different bla OXA-61 -like gene alleles (bla OXA-61 and bla OXA-193 in 17 and 9 C. jejuni, respectively, and bla OXA-489 in 16 C. coli). In both farms, resistance to FQ was always associated with a SNP mutation (T86I) in the gyrA gene in C. jejuni and C. coli. Tetracycline resistance in C. jejuni recovered from F1 was always associated with the tet(O) gene, generally plasmid-encoded (n = 17), but also chromosomally-encoded (n = 3) or both (n = 1). In F4, TET-resistant C. jejuni carried the tet(O/32/O) gene in the chromosome (n = 3) or the tet(O) gene in a plasmid (n = 1), while in C. coli, TET-resistance was associated with a chromosomally-encoded tet(O) gene (n = 16), with 3 isolates also carrying a second copy of the gene in a plasmid. In C. jejuni, STR resistance was sporadic and associated with an SNP mutation in rpsL (K43R) in 2 isolates from F1 and coded by the ant (6) Genomic diversity characterization. MLST typing assigned the 56 isolates to 13 types, including two novel STs identified in C. coli from F4 (ST-12000, n = 15 isolates; ST-12001, n = 5) that were the result of new combinations of alleles and only differed between them in the sequence of the tkt gene. In F1, 29 C. jejuni isolates clustered into 8 MLST STs belonging to 6 clonal complexes (CC). In F4, the 7 C. jejuni sequenced were assigned to 5 STs (4 CC) and the 20 C. coli to the 2 novel STs (ST12000 and ST12001, both assigned to CC828). Only 2 STs were shared by both farms (Supplementary Fig. S3 www.nature.com/scientificreports/ C0690, C0911) shared an identical VF profile and were the only isolates that carried genes responsible for LOS synthesis and modification like Cj1135, Cj1136, Cj1137c, Cj1138, cstIII, hldE, neuA1, neuB1, neuC1, and wlaN. The pattern of VF presence/absence was used as a comparative genomic fingerprinting tool to subtype the isolates, showing large genetic diversity among the sequenced isolates. Although the VF profile was occasionally able to discriminate among STs, there was an overall good correlation between MLST types and VF profiles with isolates within the same ST clustering together based on the VF profile. The genomic relatedness among the strains was elucidated from the phylogenetic trees inferred from the core genome alignments (Supplementary Fig. S4). Although core genome characterization provided more precise subtyping, C. jejuni isolates belonging to the same ST and CC clustered together in the core genome SNP-based tree. In C. coli, however, ST12000 isolates formed two clusters. It was noteworthy that the same subclustering of ST-12000 C. coli isolates was also observed by VF profile analysis. One of the clusters included 3 C. coli isolates (C0698, C0700, and C0777) phenotypically resistant to STR (MIC STR = 8 mg/L) but without GDRs and a fully susceptible isolate (C0850); the other cluster included 11 isolates with the same phenotypic AMR profile and chromosomal ARG content. ARG transmission dynamics. Identical strains (same MLST, GDR, VF, and core genome SNP-based clustering) were isolated in animals from different age groups within each farm, and certain genomic subtypes predominated in each farm ( Fig. 2 and Supplementary Dataset S3). In F1, C. jejuni ST-48 was isolated throughout the study in the three age groups; it was the only type found in calves, represented 54.5% (6/11) of the isolates from heifers and was only occasionally found in lactating cows (2/9, 22.2%). All 17 ST-48 isolates were resistant to FQ as a result of a C257T SNP-mutation in the gyrA gene, all except one (C0906, calves, S10) harboured a tet(O) gene-encoding plasmid, and 2 other isolates (heifers, S04 and S08) differed by being STR-resistant due to an SNP-mutation in the rpsL gene. In heifers, 5 other isolates with 4 different genomic profiles were isolated, differing in the presence and/or location (plasmid and chromosomal) of the tet(O) gene. In lactating cows, genomic diversity was larger (9 isolates, 5 profiles). Isolates of ST-267 with identical ARG and VF profiles were obtained from lactating cows (S06 & S09) and heifers (S07).
In F4, C. jejuni was only sporadically isolated from lactating cows and heifers, each age group hosting different genomic subtypes. Conversely, C. coli was more widespread and included 2 MLST types, both detected in all age groups. However, the most prevalent type, ST-12000, predominated in calves and heifers and was detected throughout the study. ST-12001, associated with an MDR profile (CIP-NAL-STR-TET), was first detected in lactating cows during the second half of the study, and was only detected in calves and heifers at the last sampling.

Discussion
This longitudinal study designed to monitor the occurrence and within-farm transmission dynamics of AMR in C. jejuni and C. coli in dairy cattle showed a more widespread distribution of C. jejuni compared to C. coli but higher rates of resistance in C. coli than in C. jejuni, as described before 6,12,18 . When considering the different www.nature.com/scientificreports/ age groups analysed, resistance was higher in isolates recovered from calves compared to those obtained from heifers and lactating cows. In fact, all isolates from calves showed resistance to at least one antimicrobial class, likely reflecting age-related differences in management practices and antimicrobial consumption associated to the pathologies typically found in calves (mainly respiratory and digestive disorders). However, the diversity of phenotypic resistance profiles also varied between farms. Therefore, to investigate the molecular mechanisms of resistance and thoroughly compare the relationship of the circulating strains, we performed long-read WGS using ONT on a selection of 56 isolates from two farms: F4, which showed the largest diversity in Campylobacter species and AMR phenotypic profiles, and F1, where AMR profile diversity was lower but C. jejuni isolation frequency was the highest. Overall, 7 acquired resistance genes (aadE-Cc, ant(6)-Ia, aph(3')-III, bla OXA-184 , bla OXA-61-like , tet(O) and tet(O/32/O)) and SNP mutations in 2 other genes (gyrA and rpsL) were detected. bla OXA was the most widespread gene, but β-lactams were not included in the broth microdilution panel since this class of antimicrobials is not recommended for treating campylobacteriosis. Besides, the G57T transversion upstream of the promoter region of the bla OXA-61-like gene that allows the expression of the β-lactamase 19 was not analysed here. A previous study carried out in the Basque Country showed 38.6% of resistance to ampicillin and 42.9% of the isolates carrying the bla OXA-61-like gene had the G57T transversion at the promoter region 16 . For the remaining antimicrobials, there was a very good agreement between phenotypic resistance and the presence of the associated ARGs and/ or chromosomal point mutations, with only 4 isolates showing presumptive discrepancies. These included 3 C. coli isolates which were most likely misclassified as phenotypically resistant to STR based on the observed MIC (MIC STR = 8 mg/L, 1 dilution above the ECCOF value). These isolates did not carry any associated GDR and shared other genetic features (i.e., ST, core genome SNP-based clustering, VF profile, and absence of GDRs) among themselves and with a fully susceptible isolate, suggesting that the observed MIC for STR was the result of the variance associated to the inherent error for MIC methods (± 1 log 2 dilution). On the other hand, C. jejuni isolate C0980 was susceptible to TET (MIC < 0.5 mg/L) despite carrying the tet(O) gene in a plasmid. The presence of the tet(O) gene in phenotypically susceptible Campylobacter has been reported before 20,21 . Guernier-Cambert et al. 21 reported a P76L substitution to render a non-functional plasmid-borne tet(O) gene. Here, after sequencing isolate C0980 twice and performing an in-depth sequence analysis of the tet(O) gene (data not shown) we found in both cases a deletion in nucleotide position 1078 that would produce a frameshift and result in a non-functional gene thus explaining the phenotypic susceptibility to TET.
Fully functional TET-resistance encoding genes were found in all 40 phenotypically resistant isolates, with the tet(O) gene widespread in both species and the mosaic gene tet(O/32/O) restricted to 3 C. jejuni. The tet(O) gene was chromosomally encoded in C. coli, whereas in the majority of C. jejuni (19/22) it was located in pTet plasmids (Type 1) that also included several Type IV secretion system (T4SS) genes described to form the core genome of Tet-plasmids in Campylobacter 22 . The transfer of plasmid-encoded antibiotic resistance in Campylobacter has been frequently reported and has been regarded as the main mechanism of TET resistance in C. jejuni 23,24  MLST types belonging to clonal complexes considered as host-specific for cattle (e.g. CC42 and CC22) were detected, but most C. jejuni isolates were assigned to host-generalist clonal complexes (CC21; CC45; CC48; CC206), genotypes infectious to humans that are frequently recovered from multiple host species 26,27 . Among these were the 3 C. jejuni isolated from lactating cows and heifers in F1 that carried several genes responsible for LOS synthesis and modification. These included wlaN, which mediates phase-variation of LOS epitopes responsible for autoimmunity and associated with Guillain-Barré syndrome 28 , cstIII (sialyltransferase), responsible for the addition of sialic acid to LOS type C, and the N-acetylneuraminate biosynthesis genes (neuABC), as well as other putative outer-core sugar transferases of the LOS such as Cj1135, Cj1136, Cj1137c, and Cj1138. The 3 isolates harbouring these genes were classified into CC21, in agreement with previous descriptions of class C LOS in isolates of this clonal complex originating from humans and poultry 29 .
Core genome analysis, MLST types (ST and CC) and the repertoire of GDRs and VFs were used to characterize and compare isolates recovered from the two farms over the 16-month study period. The persistence of certain genotypes was observed throughout the study while other strains were only sporadically detected. In F1, whereas ST-48 seemed to be adapted to calves outcompeting all other strains, in heifers and lactating cows it circulated simultaneously with several other strains suggesting that multiple source contamination events were more frequent in older animals than in calves. ST-48 isolates clustered together based on core genome SNP analysis and their VF profile and carried a similar repertoire of GDRs. All but one harboured a nearly identical Tet-plasmid that was also similar to that found in an isolate of a different MLST type (ST-19) recovered from lactating cows at the end of the study (S10). Homology searches using blastn showed a high percentage of identity between these and other plasmids found in GenBank, including those previously found in C. jejuni from sheep and cattle from the same region 16 . These findings suggest that these represent ubiquitous pTet Campylobacter plasmids known to transfer resistance to tetracycline.
In F4, C. jejuni was only sporadically isolated from heifers and lactating cows and no genotypes were shared between both age groups. Conversely, C. coli was widespread, but MLST diversity was lower; just 2 STs (ST-12000 and ST-12001) that only differed between them in the tkt allele and were closely related (different glnA alleles) to other widespread STs like ST-1055 and ST-827, respectively. The lower genetic diversity of C. coli of ruminant www.nature.com/scientificreports/ origin compared to C. jejuni or C. coli isolated from swine or poultry has been reported previously 30,31 . However, both genotypes were new and never described before, indicating that diversity is greater than reported. All resistant C. coli shared a similar repertoire of chromosomally encoded GDRs, and major differences were observed in 3 ST-12000 C. coli isolates due to a 47,150 bp plasmid that harboured additional ARGs (tet(O)-Δant(6)-

Ia-sat4-aph(3')-IIIa).
Considering only chromosomally encoded GDRs, the only difference between both STs was the presence of a FQ-resistance SNP mutation (C257T) in the gyrA gene of all ST-12001 isolates that was absent in ST-12000 isolates. In F4, the FQ-resistant genotype (ST-12001) was detected during the second half of the study, first in lactating cows (S07, S08, S11) and later (S12) in heifers and calves. In addition to providing FQ-resistance, this SNP in the gyrA gene, known to be involved in DNA supercoiling 32 , delivers enhanced fitness during commensal colonization 8 . In this sense, FQ resistance in C. jejuni has been associated with an increase in virulence and the ability to form viable biofilms in oxygen-rich environments 9 . This might play a critical role in the persistence of FQ-resistant isolates in the animal reservoirs even in the absence of antibiotic selection pressure thus explaining the increasing trend in FQ resistance in Campylobacter. The emergence of the FQ-R genotype (ST-12001) and the plasmid-harbouring ST-12000 isolates occurred in both cases near the end of the study and, therefore, it was not possible to monitor their possible spread and persistence in the farm. Additional samplings would have been needed to see whether the FQ-R genotype turned out to be better adapted and able to out-compete the FQ-S through local clonal expansion and to monitor possible further dissemination of the C. coli plasmid to other isolates.
In conclusion, this longitudinal study illustrated the within-farm diversity and transmission dynamics of resistant Campylobacter throughout time in dairy cattle farms, enabling the detection of changes in the AMR profiles and the emergence of different genotypes over time. Both persistent and sporadic strains were detected, and examples of processes that led to the emergence of novel genomic types that could spread within the herds were proposed, i.e., acquisition of extracellular genes from other isolates and intracellular mutational events. Several genotypes were simultaneously present within each herd but in some instances certain genotypes seemed to be more adapted and persisted for up to 16 months (e.g. ST-48 in F1 and ST-12000 in F4). Long-read whole genome sequencing-based surveillance helped to decipher the complex epidemiology underlying the withinfarm dissemination of resistant Campylobacter.

Methods
Study design and sample collection. Five commercial farms (designated F1, F2, F3, F4, and F5) in the Basque Country (northern Spain) representative of the style of farming in the region were enrolled in the study. All farms were closed production systems where replacement heifers originated from the same farm. Two of the farms (F2 and F4) raised their heifer replacements off-site in two different breeding centres. Farms were located in the three counties of the Basque Country, and the distance between farms ranged from 15 to 25 km for those located within the same county (i.e., F3-F4 and F1-F2, respectively) and up to 160 km (F4-F5). Further details on general information about farm characteristics, management practices, vaccine programs, and antimicrobial drug use were reported elsewhere 33 .
The sampling strategy, planned as a year-long study with monthly samplings, was interrupted by the COVID-19 pandemic and extended from February 2019 to October 2020. Rectal faeces were collected from calves (1-5 months-old), heifers (young cows that have not yet given birth to a calf, 6-22 months-old approx.), and lactating cows (animals in the milking period at sampling) during 12 samplings carried out over 16 months; one of the farms (F5) dropped out after 5 samplings due to operational changes. At each sampling time and farm, five animals were randomly selected within each age group and rectal faeces (minimum of 5 g) were collected with a gloved hand, and faeces of the 5 animals of the same age group were analysed in a single 25 g pool. In farms F2 and F4, heifers were at the breeding centre at several sampling dates (5 samplings in F2 and 2 in F4) and faeces could not be collected. Thus, a total of 760 rectal faecal samples were collected and analysed in 152 pools.
Sample collection was carried out by veterinary practitioners strictly following Spanish ethical guidelines and animal welfare regulations (Real Decreto 53/2013). The collection of this material, being considered a routine veterinary practice, did not require the approval of the Ethics Committee for Animal Experimentation. Informed oral consent was obtained from the farm owners at the time of sample collection. All methods were performed in accordance with the relevant guidelines and regulations and complied with ARRIVE guidelines 34 .

Isolation and identification of Campylobacter species. For the isolation of thermophilic
Campylobacter spp., 25 g of pooled rectal faecal samples were diluted 1:10 in Preston broth, homogenized, and incubated for 18 ± 2 h at 42 ℃ for enrichment. Suspensions (0.1 mL) were then subcultured onto a Chromogenic-Campylobacter Selective Agar (CASA® Agar, Biomerieux) and incubated at 42 ℃ in a microaerobic atmosphere (5% O 2 , 10% CO 2 , 85% N 2 ) for 48-72 h. To confirm the presumptive Campylobacter and identify the species present, 10 individual colonies were tested using a multiplex real-time PCR targeting C. jejuni mapA gene and C. coli ceuE gene 35 . When neither C. jejuni nor C. coli could be identified by colony picking, DNA was extracted from a loopful of bacterial culture in CASA® agar (InstaGene, BioRad, CA, USA) and screened for the presence of Campylobacter spp. by a real-time PCR that amplifies the 16S rRNA gene of Campylobacter genus 36 . A maximum of 2 C. jejuni and 2 C. coli colonies per sample were stored for further characterization.
Antimicrobial susceptibility test determination by broth microdilution. When available, 2 isolates per plate and Campylobacter species (i.e., C. jejuni and C. coli) were selected and tested to assess antimicrobial susceptibility. Minimum inhibitory concentrations (MIC) were determined by broth microdilution using Sensititre® EUCAMP2 Susceptibility Plates (ThermoFisher Scientific, Waltham, MA, USA) containing twofold serial dilutions of six antimicrobial agents: gentamicin (GEN), streptomycin (STR), tetracycline (TET),  48 was used to perform core genome analyses and construct phylogenetic trees based on the chromosomal genomes to identify structural and point variations (SNPs), with defaults parameters and specifying -r ! parameter to randomly select the reference from the set of genomes analysed. The resulting trees were visualised with iTOL 49 .
Statistical analysis. Differences in AMR prevalence between the different Campylobacter species, age groups, and/or farms were evaluated with multivariate logistic regressions. Adjusted odds ratios (OR adj ) were used as the measure of association between positivity and the explanatory variables and were expressed with their confidence interval at 95% (95% CI). Differences were considered statistically significant if p < 0.05. Analyses were conducted using statistical software Stata/IC version 16.1 (StataCorp LP, College Station, TX, USA).

Data availability
Raw sequencing data of the 56 strains analysed in this study are available at NCBI Sequence Read Archive (SRA) database under accession numbers as detailed in Supplementary Dataset S5, associated with the BioProject PRJNA 932719.