From Farm to Fork: Persistence of Clinically Relevant Multidrug-Resistant and Copper-Tolerant Klebsiella pneumoniae Long after Colistin Withdrawal in Poultry Production

ABSTRACT Concerns about colistin-resistant bacteria in animal food-environmental-human ecosystems prompted the poultry sector to implement colistin restrictions and explore alternative trace metals/copper feed supplementation. The impact of these strategies on the selection and persistence of colistin-resistant Klebsiella pneumoniae in the whole poultry production chain needs clarification. We assessed colistin-resistant and copper-tolerant K. pneumoniae occurrence in chickens raised with inorganic and organic copper formulas from 1-day-old chicks to meat (7 farms from 2019 to 2020), after long-term colistin withdrawal (>2 years). Clonal diversity and K. pneumoniae adaptive features were characterized by cultural, molecular, and whole-genome-sequencing (WGS) approaches. Most chicken flocks (75%) carried K. pneumoniae at early and preslaughter stages, with a significant decrease (P < 0.05) in meat batches (17%) and sporadic water/feed contamination. High rates (>50%) of colistin-resistant/mcr-negative K. pneumoniae were observed among fecal samples, independently of feed. Most samples carried multidrug-resistant (90%) and copper-tolerant (81%; silA and pcoD positive and with a MICCuSO4 of ≥16 mM) isolates. WGS revealed accumulation of colistin resistance-associated mutations and F type multireplicon plasmids carrying antibiotic resistance and metal/copper tolerance genes. The K. pneumoniae population was polyclonal, with various lineages dispersed throughout poultry production. ST15-KL19, ST15-KL146, and ST392-KL27 and IncF plasmids were similar to those from global human clinical isolates, suggesting chicken production as a reservoir/source of clinically relevant K. pneumoniae lineages and genes with potential risk to humans through food and/or environmental exposure. Despite the limited mcr spread due to the long-term colistin ban, this action was ineffective in controlling colistin-resistant/mcr-negative K. pneumoniae, regardless of feed. This study provides crucial insights into the persistence of clinically relevant K. pneumoniae in the poultry production chain and highlights the need for continued surveillance and proactive food safety actions within a One Health perspective. IMPORTANCE The spread of bacteria resistant to last-resort antibiotics such as colistin throughout the food chain is a serious concern for public health. The poultry sector has responded by restricting colistin use and exploring alternative trace metals/copper feed supplements. However, it is unclear how and to which extent these changes impact the selection and persistence of clinically relevant Klebsiella pneumoniae throughout the poultry chain. We found a high occurrence of copper-tolerant and colistin-resistant/mcr-negative K. pneumoniae in chicken flocks, regardless of inorganic and organic copper formulas use and a long-term colistin ban. Despite the high K. pneumoniae isolate diversity, the occurrence of identical lineages and plasmids across samples and/or clinical isolates suggests poultry as a potential source of human K. pneumoniae exposure. This study highlights the need for continued surveillance and proactive farm-to-fork actions to mitigate the risks to public health, relevant for stakeholders involved in the food industry and policymakers tasked with regulating food safety.

Copper phenotypic assays were performed in 85% of K. pneumoniae isolates carrying or not carrying copper tolerance genes representative of different farms, flocks, KL type(s), and antibiotic resistance profiles ( Table 2). All K. pneumoniae isolates carrying the silA and pcoD genes (100%, n = 53/53) exhibited MICs to CuSO 4 of 16 to 32 mM (CuT phenotype, MIC CuSO4 $ 16 mM). These data contrast with those of most K. pneumoniae isolates without acquired copper tolerance silA and pcoD genes (n = 30/32), showing MICs to CuSO 4 of 2 to 12 mM ( Table 2). The MIC CuSO4 distributions were similar for isolates of the two feed sources (ITMF and OTMF).
Whole-genome analysis of K. pneumoniae poultry-associated isolates. The 20 sequenced K. pneumoniae isolates representing the most abundant KL type(s) were assigned  Table S2). Those clusters included isolates from different sample types from the same farm, i.e., feces of poultry and derived meat (n = 2 ST6406/farm C and n = 2 ST6405/farm A), as well as water and poultry feces (n = 2 ST11/farm H). Also, we detected clones persisting over time in different farms (n = 2 ST15/farms E and H; n = 2 ST631/farms B and D). The phylogenetic relationship of our genomes with others from different sources, regions, and time frames available at Pathogenwatch was explored (Fig. 6), revealing fewer than 10 allele differences in all but four of the lineages (ST147, ST525, ST631, ST6405). Of note, isolates belonging to ST15-KL19, ST15-KL146, ST392-KL27, and ST1537-KL64 lineages revealed ,21 SNPs with genomes from diverse origins (Fig. 6) (a threshold recently proposed for K. pneumoniae transmission in health care settings) (33). ST15 isolates were linked to genomes associated with human infections in the United Kingdom (ST15-KL19) and human infections and horses in Italy, France, and the United States (ST15-KL146). The ST392-KL27 isolate was linked to genomes from human infections and/or colonization in Spain, while ST1537-KL64 was detected in food products (chicken meat and salads) in France (Fig. 6).   WGS revealed a high load and diversity in antibiotic resistance and metal tolerance genes in comparison to virulence genes ( Fig. 5; Table S3). All isolates carried the chromosomal mrkABCDFHIJ cluster encoding type 3 fimbriae, whereas only ST15 isolates carried the virulence accessory genes kfuABC (ferric uptake system) and the kpiABCDEFG genes (pilus system). Regarding the genomic analysis of colistin resistance, we detected 96 different chromosomal mutations (mostly missense; n = 94/96) in 72% of the genes (n = 23/32) encoding proteins previously implicated in colistin resistance compared with the reference strain K. pneumoniae MGH 78578 (described in detail in Fig. 7; see also Fig. S1). From these, 52 distinct mutations were present in all ColR-Kp isolates (9 lineages) across five operons/ genes associated with colistin resistance mechanisms such as modifications of lipopolysaccharide (LPS)/lipid A (arnABDFT, crrAC, phoQ, pmrACD, mgrB), overexpression of efflux pumps (acrB), LPS loss (lpxA, msbA) or biosynthesis (yciM), and regulation (rstB) (Fig. 7). Diverse mutations implicated in colistin resistance were detected in seven of those genes (acrB, arnB, arnD, arnT, mgrB, phoQ, and pmrC), varying accordingly with the clone (Fig. 7). All but one of the isolates accumulated mutations in different gene clusters associated with colistin resistance (ranging from two to seven).

DISCUSSION
This study first showed the absence of mcr but a high occurrence and diversity of colistin-resistant (mcr-negative) K. pneumoniae isolates after .2 years of colistin withdrawal in intensive chicken farms, independent of the type of Cu supplementation used in the feed formulation. Furthermore, we demonstrated the persistence of particular K. pneumoniae clones throughout the whole poultry production chain and suggested poultry as a potential foodborne/environmental source of K. pneumoniae lineages with clinical relevance.
The high rates of intensively raised flocks positive for K. pneumoniae in the early and preslaughter stages, together with the detection of K. pneumoniae with or without standard fecal indicators in feed or water samples (e.g., Escherichia coli and/or Enterococcus detected in 9 of 14 water samples from all farms) (data not shown), suggest diverse contamination events (e.g., hatchery farm, poultry house cleanliness and biosecurity level, water, feed, inanimate surfaces, and/or human handlers) occurring early and frequently along the production chain (34,35). Environmental factors (e.g., type of litter, temperature, relative humidity, daily cycles, moment of sampling/age, season, and vacancy period) or the use of antimicrobial compounds (e.g., disinfectants, antibiotics, coccidiostats, or metals) are also known to have an impact on the composition of poultry gut microbiota (35)(36)(37). These factors could justify the differences found in fecal K. pneumoniae occurrence rates at the farm level in this study (82%, ranging from 25 to 100%) and compared to another recent study also using SCAi medium for bacterial recovery in chicken farms (26%) (38). However, despite the high rate of fecal samples carrying K. pneumoniae, resistant or not to colistin, our data revealed few positive chicken meat batches (17%), which is far below what was reported in studies from the United States (47%) and the European Union (60%) (39,40). Besides, Salmonella was not detected in any sample (data not shown). These results demonstrate reduced meat cross-contamination and the effectiveness of sanitary measures during animal transport and at the slaughterhouse, thus reducing the consumer's risk of exposure (12,37,41).
Recent studies suggest that banning colistin in food animal production has had an encouraging outcome by limiting mcr spread (11,12). Our results confirm this trend in poultry production by the absence of mcr in farms and meat isolates, although a high rate of samples carried ColR-Kp associated with a high diversity of chromosomal mutations, independently of the feed type used. This suggests the circulation of diverse colistin-resistant genotypes responding differently to the colistin ban. We detected mutations, alone or in Klebsiella pneumoniae genomes. Colored circles indicate the presence of a specific mutation, while each color represents the genes that are part of the same operon. Variants shown in bold were supported by the available literature as present in ColR-Kp and/or were predicted as deleterious by PROVEAN. Klebsiella pneumoniae isolates were grouped as colistin susceptible (S) when the MIC was #2 mg/mL or as colistin resistant (R) when the MIC was $2 mg/mL. When required, figures were minimally edited manually using Adobe Illustrator v25.3.1. Detailed resistance mechanisms: acrB, efflux pump; arnABDFT, lipid A modification with L-Ara4N addition; crrA, lipid A modification by upregulation of pmrAB/activation of the glycosyltransferase; crrC, connector protein; lpxA, inactivation of lipid A biosynthesis abolishing LPS synthesis; mgrB, inactivation of negative feedback regulator of the PhoP/PhoQ system; msbA, ABC transporter of lipid A; phoQ and pmrA, activation of LPS-modifying operation in the two-component systems; pmrC, lipid A modification with phosphoethanolamine; pmrD, connector protein; rstB, sensor protein; yciM, regulation of LPS biosynthesis. A, alanine; C, cysteine; D, aspartate; E, glutamate; G, glycine; I, isoleucine; K, lysine; L, leucine; M, methionine (start codon); N, asparagine; P, proline; R, arginine; S, serine; T, threonine; V, valine; Unk, unknown.
Colistin-Resistant K. pneumoniae in Chicken Production Microbiology Spectrum combination, in genes known to be involved in lipid A modifications and colistin resistance phenotypes (arnABDFT, phoQ, pmrACD, and mgrB), as described previously (42)(43)(44)(45)(46)(47)(48)(49)(50). However, a high variety of other nondescribed mutations were detected, supporting the urgent need for reliable genotypic-phenotypic correlations to explain colistin resistance mechanisms (51). Such a variety and frequency of chromosomal mutations in ColR-Kp isolates detected in different studies and environments (26,43,50,51) suggest that they could play a role in adaptive features other than colistin resistance (e.g., mgrB inactivation in environmental K. pneumoniae survival and transmission or phoPQ or pmrB mutations in chlorhexidine tolerance) (52)(53)(54). Also, some of these mutations do not confer significant biological cost (55,56), which may explain the occurrence of colistin-resistant mcr-negative K. pneumoniae isolates long after colistin withdrawal. Finally, other factors contributing to coselection (e.g., other antibiotics and growth promoters such as Cu) or maintenance (continuous contamination sources) of chromosomally mediated colistin-resistant strains in the livestock sector (26) could not be excluded.
In the context of antibiotic/colistin reduction and replacement, the use of diverse antimicrobial compounds may be expanded, making it difficult to understand the complexity of possible coselection events of the MDR strains. Some recent studies performed after colistin withdrawal revealed an increase in clinically relevant antibiotic resistance genes (e.g., bla CTX-M and bla NDM ) among food animal isolates (57,58). In this study, these genes were not detected, but we showed high rates of resistance to multiple antibiotics in K. pneumoniae isolates from poultry, independently of the sample selection strategy (with or without colistin supplementation) or between sample groups (poultry life stages and feed type). Resistance rates were higher for antibiotics (e.g., ciprofloxacin and tetracycline) frequently used for therapeutics at poultry farms (12,59), in contrast with other studies in poultry or other nonclinical environments (38,60) where resistance rates were low. Such differences in the available studies could reflect local variation in the usage of several antibiotic classes, the diversity of routes for K. pneumoniae dissemination within and beyond the production environment, and the circulating clonal lineages (38,60).
Other largely used feed ingredients with antimicrobial activity like metals/copper may also contribute to changes in poultry microbiota and/or even potentiate Cu-tolerant and MDR strain coselection (19,20,24), a factor unexplored for K. pneumoniae. Recent studies revealed correlations between metal environmental pollution (e.g., Cu at poultry farm samples) and an increase in antibiotic resistance genes (61,62). We have also detected the plasmid-acquired copper tolerance pco plus sil cluster in isolates dispersed in all farms and in most of the flocks. Furthermore, we describe for the first time in K. pneumoniae a correlation between the presence of pco and sil genes and the Cu tolerance phenotype (MIC $ 16 mM in anaerobiosis). These data suggest the adaptability of this species to stressful/unfavorable farming environments, as previously detected for other zoonotic bacteria such as the emergent MDR clones of Salmonella (63). According to our data, the feeding regime (ITMF or OTMF) does not seem to contribute to the expansion of MDR or ColR-Kp. These data suggest that the similar wholefarm environment (e.g., litter quality, ventilation system, and temperature control) and management practices (e.g., sanitation and disinfection protocols, waste, and water control) overlap with the feeding regime. Thus, studies evaluating environmental factors (e.g., pH) (64) and subinhibitory concentrations of heavy metals to maintain antibiotic-resistant bacteria in poultry production and other related environments are urgent (61,65,66).
It is of note that metal tolerance operons (pco plus sil) were mainly located in multireplicon F type (FII K plus FIB K ) plasmids carrying genes encoding resistance to several classes of antibiotics, including the critical extended-spectrum cephalosporins and/or fluoroquinolones, supporting the potential for diverse coselection events (67,68). These mosaic F type plasmids are common in K. pneumoniae populations from different sources (68)(69)(70)(71)(72)(73), suggesting a major role in both dissemination and persistence of antibiotic resistance and metal tolerance genes and K. pneumoniae adaptation to different niches. Besides, the similarity between plasmid backbones identified in poultry isolates and those described in K. pneumoniae collections from humans (see Tables S3 and S4 in the supplemental material) (72) suggests a common pool of shared plasmids between humans and eventually different animal species (70), which deserves to be further explored with comprehensive comparative plasmidome analysis. This is the first study tracking K. pneumoniae throughout the whole poultry production chain, enabling the identification of multiple transmission routes in flocks and the farm environment. The identification of closely related MDR K. pneumoniae isolates (e.g., ST11-KL106, ST15-KL146, ST631-KL109, ST6405-KL109, and ST6406/CG147-KL111) in multiple poultry stages and environments in the same or different farms suggests that food sector efforts should be made in improving the sanitary measures in the poultry houses, poultry workers, and the environment beyond (treatment of wastewaters and manure). The persistence of MDR K. pneumoniae clones in the poultry chain, which is according to previous studies conducted on farms (34,35,38) or at slaughterhouses (41), suggests the presence of adaptive environmental features other than antimicrobial resistance (e.g., biofilm formation). Most available studies assessing K. pneumoniae sources from a One Health perspective have included retail foods, livestock farms, or wastewater, while the poultry chain has been greatly understudied (27, 38-40, 59, 60, 74, 75). Despite the high diversity found, we highlight the commonality of KL types (KL64, KL102, KL105, and KL106) and lineages (ST15-KL19, ST15-KL146, and ST392-KL27) identified between poultry and human clinical isolates (this study) (28,76). Although the immunogenicity of different KL types is not well understood, particular capsular types have been shown to influence the selection of certain K. pneumoniae lineages (77), which might provide an advantage regarding host immunity (78,79). Finally, although human contamination cannot be excluded, our data suggest poultry as a reservoir and source of globally dispersed and human clinically relevant K. pneumoniae lineages with a potential risk for transmission to humans and other ecosystems (animals and the environment).
Conclusions. This study constitutes the first comprehensive analysis of K. pneumoniae from a poultry farm-to-fork perspective. Our data revealed, independently of the Cu feed type, a high rate of chicken fecal samples carrying a high diversity of K. pneumoniae isolates, including MDR, Cu-tolerant, and colistin-resistant/mcr-negative strains, long after colistin withdrawal. Furthermore, we propose that poultry may serve as a reservoir and source of human clinically relevant lineages and mosaic F type plasmids (with antibiotic resistance and metal tolerance genes), which could pose a potential risk of human exposure through the food chain (e.g., poultry carcasses and occupational exposure) and/or environmental release (e.g., farm or slaughterhouse wastewaters). Therefore, it is crucial to further explore the drivers contributing to the circulation and persistence of MDR K. pneumoniae in poultry production. This will help to identify novel mitigation strategies addressing producers' efforts, consumer concerns, and the EU farm-to-fork goals in the fight against antimicrobial resistance and toward improving food safety and environmental sustainability.

MATERIALS AND METHODS
Sampling strategy at the chicken farm and slaughterhouse processing plant. Our pilot study involved seven Portuguese intensive farming-based chicken farms with similar conventional indoor and floor-raised production systems, comprising all the key practices and requirements in compliance with EU legislation, as indicated by the operators (80). To ensure proper management, the farms were selected based on having grow-out poultry houses with similar conditions and individual feed silos, allowing for the division of each flock upon arrival (1-day-old chicks) into two groups, each receiving a different feed type (OTMF or ITMF). In all farms, colistin had been banned since January 2018, while copper was routinely used as an additive in poultry feed. The current inorganic formulation feed (ITMF) was supplemented with inorganic sources of Cu, and the organic formulation (OTMF) was supplemented with organic forms, both previously evidenced as capable of maintaining broiler performance under commercial conditions (23). Both mineral supplements were added to the same commercial starter and grower feeds adapted to different periods of the broiler's life, with copper concentrations decreasing 42% in OTMF versus ITMF feeds (both far below the maximum dose of 25 mg/kg of Cu, according to EU Regulation 2018/1039).
Eighteen chicken flocks (10,000 to 64,000 animals from each house; Ross 308 strain) fed with ITMF (n = 9) or OTMF (n = 9) were sampled at each of the seven farms between October 2019 and November 2020 (two farms were sampled twice in different seasons). Personal protective equipment such as gloves, boots, and coveralls were used. Pooled fresh chicken feces (50 g; total n = 34) were collected with sterile spatulas/shovels from the floor by walking in a zig-zag pattern in two separated poultry houses (ITMF or OTMF) in each farm. The collection period was at 2 to 3 days of life (P1 stage; n = 18 samples) and the day before slaughter when the chickens were 28 to 30 days old (P2 stage; n = 16 samples; 2 samples missing during 2020 COVID-19 lockdown). Environmental samples, including feed (50 g collected at the silo that supplied the feeding lines; n = 18) and water (1 L collected at the faucet that supplied the drinking water lines; n = 14), were also collected inside grow-out houses at each farm at the P1 stage (Fig. 8).
Raw chicken meat samples (n = 18 batches; recovered after slaughter and air chilling) of the same flocks were collected after slaughter (P3 stage) in the poultry production slaughterhouses, immediately before distribution for retail sale. Each meat sample included approximately 50 g of neck skin cut with a sterile scalpel from a pool of 10 carcasses from the same batch (each batch corresponded to one flock from the same farm slaughtered at the same time) (Fig. 8).
All previous solid and water samples were collected in sterile plastic bags or containers, transported at 4°C, and processed on the same day at the laboratory. Subsequent sample processing was performed by culture approaches, as described in the following sections.
Screening of K. pneumoniae and colistin-resistant K. pneumoniae. K. pneumoniae was selectively recovered, directly from the sample and after enrichment, in the Simmons citrate agar plates with 1% inositol (SCAi), a medium that selectively favors the growth of Klebsiella in potential Escherichia coli-rich samples (see Fig. S2 in the supplemental material). A common initial step consisted of weighing a portion of 25 g of tested solid samples (P1, P2, and P3 stages) or the 0.45-mm filter from filtration of 500 mL of water samples (P1 stage), suspended in 225 mL of buffered peptone water (BPW) supplemented with 3.5 mg/L of colistin. The direct culture method included spreading an aliquot of 100 mL of the BPW plus colistin after 1 h at room temperature (resuscitation step) on SCAi supplemented or not with colistin (3.5 mg/L). The enrichment approach involved the same procedure, but after a previous incubation of BPW plus colistin at 37°C for 16 to 18 h. All the SCAi plates were incubated at 37°C for 48 h. One to five colonies of each presumptive morphotype were selected for identification. Isolate identification was performed by matrix-assisted laser desorption ionizationtime of flight mass spectrometry (MALDI-TOF MS) (MALDI-TOF Vitek MS, bioMérieux, France) and by PCR for K. pneumoniae (81). In all the identified isolates, screening of colistin resistance genes (mcr-1 to mcr-5 and mcr-6 to mcr-9) was assessed by two multiplex PCRs, as previously reported (82,83). The MIC for colistin was determined by the reference broth microdilution (84). An estimation (in CFU per gram) of colistin-resistant K. pneumoniae in the poultry samples directly plated on SCAi plus colistin (see the procedure described above) was performed after counting typical Klebsiella colonies, species identification, and determination of colistin MIC.
Phenotypic and genotypic characterization of K. pneumoniae. Relatedness between isolates from different samples was inferred by Fourier transform infrared (FT-IR) spectroscopy with attenuated total reflectance (ATR) using a PerkinElmer Spectrum Two instrument. After growth under standardized culture conditions (37°C, 18 h), a colony was directly deposited on the ATR accessory of the FT-IR instrument and air dried. Spectra were acquired under standardized conditions (4,000 to 600 cm 21 , 4 cm 21 resolution, and 16 scan coadditions). The region corresponding to polysaccharides (1,200 to 900 cm 21 ) was then compared between each other and with those from an in-house spectral database (allowing identification of 30 KL type(s) from wellcharacterized international K. pneumoniae clones), as previously described (76,85). FT-IR-based assignments were confirmed using PCR of the wzi gene and further sequencing at Eurofins Genomics (https://www .eurofinsgenomics.eu/) to infer the KL type(s) at BIGSdb (https://bigsdb.pasteur.fr/klebsiella/) (86).
Antibiotic susceptibility profiles were determined by disc diffusion using 14 antibiotics (amoxicillin plus clavulanic acid, 30 mg; amikacin, 30 mg; aztreonam, 30 mg; cefepime, 30 mg; cefotaxime, 5 mg; cefoxitin, 30 mg; ceftazidime, 10 mg; chloramphenicol, 30 mg; ciprofloxacin, 5 mg; gentamicin, 10 mg; meropenem, 10 mg; sulfamethoxazole, 300 mg; tetracycline, 30 mg; trimethoprim, 5 mg). The interpretation was Colistin-Resistant K. pneumoniae in Chicken Production Microbiology Spectrum performed using the guidelines of the European Committee of Antimicrobial Susceptibility Testing (EUCAST) (87) and, when this was not possible, by the Clinical and Laboratory Standards Institute (CLSI) guidelines (88). Escherichia coli ATCC 25922 was used as the control strain. MDR was considered when the isolates were resistant to three or more antibiotics of different families (in addition to ampicillin, to which all K. pneumoniae are intrinsically resistant). All the isolates were screened for the silA and pcoD Cu tolerance genes by PCR, as previously reported (63). Copper susceptibility was studied in representative isolates (representing the diverse farms, flocks, feeds, and genomic backgrounds) by the agar dilution method (MIC CuSO4 , 0.5 to 36 mM; anaerobiosis), as previously published (63).
The assemblies were annotated using the RAST server (94). Genome assemblies were then uploaded to Pathogenwatch v2.3.1 (https://pathogen.watch/) to determine species, capsular polysaccharide (K) and lipopolysaccharide O locus types and serotypes (29), MLSTs (95), and cgMLSTs. Pathogenwatch uses the calculated pairwise SNP distances between the genomes based on a concatenated alignment of 1,972 core genes to infer a neighbor-joining tree for phylogenetic analysis (96). Closely related genomes and the associated metadata (country, source of isolation, and collection date) were retrieved from all public genomes available at Pathogenwatch, after cgMLST single linkage clustering and selection of those with less than 10 allele differences (threshold = 10). Neighbor-joining trees were edited using iToL (97).
Snippy v3.2-dev (https://github.com/tseemann/snippy) was used to identify substitutions (SNPs) and insertions/deletions (indels) in genes (n = 32) putatively associated with colistin resistance (43,51,98), by aligning the raw read data from each isolate against the reference genome Klebsiella pneumoniae MGH 78578 (GenBank accession number CP000647). All gene mutations were further manually confirmed in the assembled genomes using the Geneious Prime 2023.0.1 software (http://www.geneious .com/). The deleterious effect of detected mutations on protein function was further assessed using the PROVEAN protein variation effect analyzer (99) or data from previous publications. We used the standard PROVEAN scores of less than or equal to 22.5 for a deleterious effect and greater than 22.5 for a neutral effect on protein function.
In all 20 WGS-selected isolates, plasmid replicon typing was determined using PlasmidFinder (103, 104) and pMLST v2.0 (104) from the Centre for Genomic and Epidemiology (http://www.genomicepidemiology.org). IncFII K plasmids were further characterized according to reference 105 (https://pubmlst.org/organisms/plasmid -mlst). To confirm the location of metal tolerance genes and reconstruct putative plasmids based on draft assemblies, we used the MOB-recon tool v3.0.0 from the MOB-suite package (106,107). The metal tolerance genes were considered part of a specific plasmid when identified by MOB-recon or when located on the same contig as the replicon/incompatibility determinant.
Statistical analysis. Differences in occurrence, antimicrobial resistance, copper tolerance, and bacterial count among K. pneumoniae isolates and flocks fed ITMF and OTMF as well as among P1, P2, and P3 stages were analyzed by Fisher's exact and Wilcoxon matched-pairs signed rank tests (a = 0.05) using Prism software, version 8.

ACKNOWLEDGMENTS
We thank Lina Cavaco, Yang Wang, Alessandra Carattoli, and Maria Borowiak for providing the mcr-positive controls, the pregraduate students Beatriz Pereira (FCNAUP), Carina Baptista (FFUP), Eulália Monteiro (FCNAUP), and Sofia Paiva (FCNAUP) for technical support, and the staff of the participating farms and slaughterhouses for their kind cooperation. We thank the Institut Pasteur teams for the curation and maintenance of BIGSdb-Pasteur databases at http://bigsdb.pasteur.fr/.