The study collection analyzed here, consisted of 106 swine E. coli isolates recovered from 93 stool samples of pigs reared on 74 farms, managed by 35 different pig companies. The farms involved here, belong to 19 Spanish provinces of those Autonomous Communities (10 out of 17) with the highest census of pig livestock in 2020, according to data from the Ministry of Agriculture, Fisheries and Food (Government of Spain): Aragon, Cataluña, Castilla y León, Andalucía, Región de Murcia, Extremadura, Castilla-La Mancha, Galicia, Comunidad Valenciana, Comunidad Foral de Navarra. Three production stage groups were included: lactation, transition, and fattening (38, 49 and 6 samples, respectively). The samples, either fecal swabs or isolation plates of MacConkey agar (LMAC, Oxoid), were received at the Spanish Reference Laboratory of E. coli (LREC, USC), and tested for routine diagnosis of enteric colibacillosis between September 2020 and April 2022. A brief voluntary questionnaire was performed about the use of specific drugs (ceftiofur, enrofloxacin/marbofloxacin, neomycin, colistin, sulfonamides, florfenicol, tiamulin) for infectious treatment in the farms involved in this study.

Briefly, the confluent growth from the fecal swabs plated on lactose MacConkey agar (LMAC, Oxoid), and incubated at 37°C for 18–24 h, was subjected to polymerase chain reaction (PCR) to detect the presence of ETEC, STEC, and EPEC using specific genetic targets encoding toxins (LT, STa, STb, and Stx2e), intimin (Eae) and fimbriae (F4, F5, F6, F18, and F41) as previously detailed (García-Meniño et al., 2021) (Supplementary Table S1). Additionally, confluents were also screened for rbfO25 associated with the pandemic lineage B2-ST131 (Clermont et al., 2008) (Supplementary Table S2). From PCR-positive confluents, different E. coli-like colonies (up to 10) were selected and plated on tryptone soy agar (TSA, Oxoid), which were in turn individually analyzed by PCR. Per sample, all E. coli colonies with different virulence-gene profiles were stored at room temperature in nutrient broth (Difco) with 0.75% nutrient agar (Difco) for further characterization. Likewise, from plates with negative confluents for all virulence targets, one E. coli-like colony was tested to confirm the status of non-pathogenic E. coli and conserved. PCR amplification of the β-d-glucuronidase-encoding gene (uidA) was routinely used to confirm the identity of the species E. coli (Gómez-Duarte et al., 2010) (Supplementary Table S2), which was further confirmed at the Laboratorio de Sanidade Animal de Galicia (LASAPAGA, Lugo, Spain) by matrix-assisted laser desorption/ionization–time-of-flight mass spectrometry (MALDI-TOF; Bruker Daltonik, Bremen, Germany) after conventional culture. The phylogroup of the isolates was established according to the PCR-based method developed by Clermont et al. (2013, 2019) (Supplementary Table S3), which recognizes eight phylogroups belonging to E. coli sensu stricto (A, B1, B2, C, D, E, F, and G).

Minimal inhibitory concentrations (MICs) of the 106 E. coli isolates were determined at the LASAPAGA using the VITEK^® 2 (bioMérieux, Inc., Hazelwood, MO, United States) systems for conventional AST (bioMérieux, Spain) with AST-GN96 test kit cards, following the manufacturer’s recommendations. The following drugs and categories were included in the analysis: aminopenicillins (ampicillin); aminopenicillins in combination with beta-lactamase inhibitor (amoxicillin/clavulanic acid, ticarcillin/clavulanic acid); non-broad spectrum cephalosporins (cefalexin and cefalotin); broad-spectrum cephalosporins (cefoperazone, ceftiofur, and cefquinome); carbapenems (imipenem); fluoroquinolones (enrofloxacin, flumequine, and marbofloxacin); aminoglycosides (gentamicin and neomycin); tetracyclines (tetracycline); polymyxins (polymyxin B); sulfonamides, dihydrofolate reductase inhibitors, and combinations (trimethoprim, trimethoprim/sulphamethoxazole); amphenicoles (florfenicol). The florfenicol resistance was double tested for a representative group of 42 isolates using MIC and disk diffusion assays. In addition, and due to our previous data of high prevalence of resistance to chloramphenicol, susceptibility to this antimicrobial was determined by disk diffusion for the whole collection. AST results were interpreted following the recommendation of PRAN (AEMPS) and other European authorities (EMA, ESCMID, ECDC), which prioritizes the use of the clinical standard breakpoints established by the European Committee on Antimicrobial Susceptibility Testing (EUCAST 2022) when available. As an alternative, epidemiological cut-off values (ECOFF) or tentative ECOFF (TECOFF) were applied (Table 1). ECOFFs (and TECOFFs) distinguish microorganisms without (wild type) and with phenotypically detectable acquired resistance mechanisms (non-wild type, N-WT) to the agent in question. In addition, the veterinary microbiology laboratory standards from the Clinical & Laboratory Standards Institute (CLSI VET ED5:2020) (CLSI VET01SED5, 2020) and the veterinary recommendations from the Comité de l’antibiogramme de la Société Française de Microbiologie (CASFM VET2020) (CASFM Vétérinaire, 2020) were also visited (Table 1). Based on AST results, the isolates were classified as MDR if displayed resistance to a drug of ≥ 3 of the different antimicrobial categories (Magiorakos et al., 2012).

By PCR, the presence of mcr-1, mcr-2, mcr-3, mcr-4 and mcr-5 genes was screened within the collection as described elsewhere (García et al., 2018). Likewise, genetic identification of the ESBLs was performed using the SHV, CTX-M-1 and CTX-M-9 group-specific primers followed by amplicon sequencing (García-Meniño et al., 2018, 2021) (Supplementary Table S4).

Since the prevalent global ExPEC lineages implicated in human and animal infections, such ST131, belong to phylogroup B2 (Mora et al., 2013; Flament-Simon et al., 2020b), one isolate assigned by PCR to this phylogroup was further investigated by WGS. DNA was extracted and quantified as detailed in García-Meniño et al. (2019). Briefly, DNA was extracted with the DNeasey Blood and Tissue Kit (Qiagen, Hilen, Germany) according to the manufacturer’s instructions. After extraction, the DNA was quantified by an Invitrogen Qubit fluorimeter (Thermo Fisher Scientific, Massachusetts) and assessed for purity using a NanoDrop ND-1000 (Thermo Fisher Scientific, Massachusetts). DNA sequencing was performed using Illumina technology with a NovaSeq 6,000 S4PE150 XP system to obtain 150 bp paired-end reads at Eurofins Genomics (Eurofins Genomics GmbH, Konstanz, Germany), after a standard library preparation (unique dual indexing, that has distinct, unrelated index sequences for each of the i5 and i7 index reads). The quality of the paired-end Illumina reads was evaluated using FastQC. The reconstruction of the genome and in silico analysis was performed as described elsewhere (García-Meniño et al., 2019). Briefly, the raw reads were assembled with the VelvetOptimiser.pl. script implemented in the “on line” version of PLAsmid Constellation NETwork (PLACNETw). PLACNETw is a tool implemented for the reconstruction of plasmids from next-generation sequence pair-end. It allows the manual pruning of the graph representation.¹ The assembled contigs, with genomic size 5.0 Mbp (Supplementary Table S5), were analyzed using the bioinformatics tools of the Center for Genomic Epidemiology (CGE)² as specified, and applying the thresholds suggested by default when required (minimum identity of 95% and coverage of 60%): for the presence of acquired genes and or chromosomal mutations mediating antimicrobial resistance (ResFinder 4.1.), for identification of acquired virulence genes (VirulenceFinder 2.0), plasmid replicon types (PlasmidFinder 2.1/pMLST 2.0), and identification of clonotypes (CHTyper 1.0), and serotypes (SeroTypeFinder 2.0). For the phylogenetic typing, two different MLST (2.0) schemes were applied. Additionally, cgMLSTFinder1.2 was applied for the core genome multi-locus typing (cgMLST) from the raw reads of the isolates. The bacteria’s pathogenicity towards human hosts was predicted using PathogenFinder 1.1.

Pairwise comparisons were performed by a two-tailed Fisher’s exact probability test,³ namely: phylogroup distribution was correlated with production stage (lactation vs transition), with pathogenic status (commensal vs pathogenic group), and MDR. In addition, correlations between pathotypes and colonization factors with respect to the three stages of production, and differences between commensal and pathogenic isolates regarding antimicrobial resistance, mcr- and ESBL-production, were also compared. p values < 0.05 were considered statistically significant.

ETEC was the most prevalent pathotype (78.1%) found among the 64 diarrheagenic isolates, positive for enterotoxin-encoding genes (eltA and/or estA, and/or estB). The remaining isolates were assigned to STEC pathotype (7.8%) positive for stx2e (Shiga toxin); aEPEC (6.2%) positive for eae; and ETEC/STEC (4.7%) positive for both Shiga toxin and enterotoxin-encoding genes. Additionally, two isolates did not fit the definition of any pathotype since tested positive only for F41 fimbriae. Regarding the other intestinal colonization factors, F18 was the most prevalent fimbriae detected (40.6%), followed by F4 (25.0%). Overall, the most common virulence profiles determined (> 5 isolates) were: STa, STb, LT, F18 (23.4%); STa, STb, F4 (14.1%); STb (14.1%) and STb, LT, F4 (9.4%) (Supplementary Table S6, columns T–AF; Table 2). Pairwise comparisons of pathotypes and colonization factors for the three stages of production, only showed a statistically significant association (p < 0.05) for F18 fimbria with the fattening and transition isolates.

By the quadruplex PCR described by Clermont et al. (2013, 2019), 5 phylogroups (A, B1, C, D, and E) were identified. Of them, phylogroup A was the most prevalent (48.4%) followed by E (25%) and B1 (20.3%). Prevalence differences were observed by origin of isolation, specifically within the isolates recovered from lactating pigs, where B1 was more prevalent (43.7%), followed by A (25%) (Supplementary Table S6, column AK; Supplementary Table S7; Table 2).

The group of 42 E. coli negative for all virulence-associated genes with porcine diarrheagenic pathotypes exhibited 6 phylogroups (A, B1, B2, C, D, and E). Of them, B1 (38.1%) was the most prevalent, followed by A (30.9%) and C (19%) (Table 2). Pairwise comparisons regarding origin of isolation showed a significant difference (p < 0.05) in prevalence of phylogroup C (8% lactation vs 37.5% transition).

Overall, the commensal E. coli exhibited a similar phylogenetic structure to that of the pathogenic group, but with significant differences in the prevalence of B1, C, and E (38.1 vs 20.3%; 19 vs 1.6% and 7.1 vs 25%, respectively) (Supplementary Table S7). Besides, phylogroup B1 was significantly associated with the lactating group of isolates.

MIC values determined for the 106 E. coli isolates showed the highest levels of resistance (> 50%) to ampicillin, tetracycline, enrofloxacin (non-wild type, N-WT), flumequine (N-WT), and trimethoprim/sulfamethoxazole (Table 3). Noticeably, we found resistant/N-WT isolates for all antibiotics, except for imipenem (category A of the EMA). Broad-spectrum cephalosporins and polymyxin resistances (category B of the EMA) were exhibited by 16–21.7 and 8% of isolates, respectively. Additionally, disk diffusion assay revealed that 50.9% of the 106 E. coli isolates were chloramphenicol resistant. Lastly, the parallel assay for florfenicol (MICs and disk diffusion) performed for a representative group of 42 isolates showed moderate discrepancies (31% vs 38.1% of N-WT/resistant isolates, respectively), affecting AST results close to the cut-off/breakpoint values (MIC > 16 N-WT; disk-diffusion R < 19, respectively) (Table 3; Supplementary Table S6, columns AL–BY). Comparing the two groups of commensal isolates vs pathogenic E. coli, only two significant differences were observed, namely, against neomycin (16.7 vs 53.1, respectively) and polymyxin B (0 vs 12.5%, respectively; p < 0.05).

Of the 106 isolates, 83 (78.3%) showed resistance to at least one agent in ≥ 3 antimicrobial categories, which were defined as MDR (Magiorakos et al., 2012), with no significant difference between commensal vs pathogenic isolates. However, MDR to ≥ 6 antimicrobial categories, were significantly associated with pathogenic E. coli (48.4%) vs commensal (28.6%) isolates (p < 0.05) (Table 4). Regarding phylogroup distribution between the groups of MDR (≥ 3 antimicrobial categories) vs non-MDR, the phylogroup C appeared significantly associated with non-MDR isolates (4.8% vs 21.7%, respectively; p < 0.05) (Supplementary Table S7).

Carriage of ESBL genes was determined in 23 isolates, 13 commensal (31%) and 10 pathogenic (15.6%) E. coli (p > 0.05), originating from the three stages of production (14 lactation, 8 transition, and 1 from fattening). These 23 isolates were phenotypically predicted as ESBL by the VITEK^® 2 system, with resistant MICs for all non-broad spectrum cephalosporins (23 isolates) and for the broad-spectrum cephalosporins: ceftiofur (23 isolates), cefoperazone (17 isolates), and cefquinome (17 isolates). The ESBL-typing by sequencing revealed the presence of SHV and CTX-M genes in 6 and 17 isolates, respectively. Most of the CTX-M positive sequences were classified as group 1 (CTX-M-1, 2 isolates; CTX-M-15, 2 isolates; CTX-M-32, 7 isolates; CTX-M-55, 1 isolate; one could not be typed), and 4 belonged to CTX-M of group 9 (CTX-M-14) (Table 4; Supplementary Table S6, columns CB–CG).

The 106 isolates were also screened for the presence of mcr-1 to 5 genes, which determined that 12 isolates (11.3%) were mcr carriers (8 mcr-1 and 4 mcr-4), all conforming the ETEC pathotype. These mcr-positive isolates were recovered from the lactation and transition subgroups (3 and 9 isolates, respectively). Interestingly, 2 mcr-1 and 1 mcr-4 isolates were also ESBL-carriers (CTX-M-32 and CTX-M14, respectively). Phenotypically, 8 out of the 12 mcr-positive isolates exhibited resistance to polymyxins (MICs > 2), and all 32 isolates, positive for ESBL and/or mcr, exhibited MDR (Table 4; Supplementary Table S6, columns AL–CG). Notably, 10 ESBL isolates (2 commensal and 8 pathogenic E. coli) displayed MDR to ≥ 8 categories, which were recovered from the three stages of production (3 lactation, 6 transition and 1 fattening) (Supplementary Table S6, columns BZ–CG).

Supplementary Table S8 compares AST results obtained by applying breakpoints/cut-off values from different standards (EUCAST 2022, TECOFF 2022, CLSI VET ED5:2020, and CASFM VET2020). Clinical breakpoints are missing within the three standards for veterinary antibiotics such as cefquinome, marbofloxacin, or florfenicol.

The PCR screening of rfbO25, performed to presumptively detect the pandemic ST131 clonal group of phylogroup B2, gave no positive result. However, this phylogroup was determined in one commensal isolate, which was further investigated by WGS to determine its virulence and potential zoonotic profile. Table 5 shows the analysis of the assembled contigs using the bioinformatics tools of the CGE (see footnote 2). Briefly, SeroTypeFinder predicted O4:H5 antigens, MLST tools assigned the sequence type (ST)12 of the Achtman 7-gene scheme and core genome ST (cgST) 44799 based on the Enterobase database, and CHTyper predicted the clonotype (CH)13–223, so that the clonal group eventually assigned to this isolate was O4:H5-B2-ST12 (CH13-223). The resistome analysis revealed the absence of chromosomal mutations but the presence of encoded mechanisms of acquired antibiotic resistance for beta-lactams (bla_TEM), aminoglycosides (aph(3″)-Ib, aph(6)-Id), sulphonamides (sul2) and tetracycline tet(A), which correlated with the phenotypic expression of resistance against ampicillin and tetracycline. VirulenceFinder revealed virulence traits associated with ExPEC lineages. In fact, the virulence profile predicted for this isolate conforms the status of extraintestinal pathogenic E. coli (ExPEC status ≥ 2 of these five virulence traits: papA and/or papC, afa/dra, sfa/foc, iutA, kpsM II) (Johnson et al., 2003), and the status of uropathogenic E. coli (UPEC status ≥3 of these four traits: chuA, fyuA, vat and yfcV) (Spurbeck et al., 2012). Furthermore, this genome was predicted as human pathogen (probability 88%) when analyzed with PathogenFinder. Regarding plasmid replicon types, PlasmidFinder identified an IncQ1 with 100% identity but with a coverage of 529/796. The PLACNET genome reconstruction (Supplementary Figure S1) showed a MOBQ relaxase within a 977 kb chromosomal contig, implying the presence of a potential integrative and conjugative element (ICE). Besides, a plasmid replication initiator protein (RIP) was identified in a node of 4.6 kb linked to plasmid and chromosomal references.

The goal of in vitro AST is to inform clinicians whether an antimicrobial is appropriate for the infection caused by a specific isolate. Furthermore, there is no successful option of empiric treatment for E. coli due to their heterogeneous MIC profile, so optimization to minimize resistance selection should be based on AST for each clinical case (Cortés et al., 2010; García-Meniño et al., 2019, 2021; Vilaró et al., 2022). However, we report here the lack of standardized breakpoints for many antimicrobials of common use in livestock. Currently, in the EUCAST reference method, there are no clinical breakpoints for cefalothin, cefoperazone, ceftiofur, flumequine, enrofloxacin, neomycin, tetracycline, and florfenicol. For these, we reviewed ECOFF (and TECOFF) values. Since cefquinome, marbofloxacin had neither clinical breakpoints nor cut-off values in EUCAST (nor in CLSI), we took them from CASFM VET2020, which were also applied for florfenicol-disk assay interpretation (Table 1). It is of note that the last version of CLSI and CASFM veterinary standards was reviewed in 2020. The comparison of results using the different standards (Supplementary Table S8), showed a moderate to high correlation between clinical breakpoints and cut-off values, except for enrofloxacin, neomycin, and florfenicol. The discrepancies observed for enrofloxacin and neomycin are explained by lower ECOFF values compared to CASFM VET2020 (> 0.125 vs > 2 and > 8 vs > 16, respectively). This observation is consistent with the fact that ECOFF values define a microorganism as non-wild type (N-WT) for a species by the phenotypical detection of an acquired or mutational resistance mechanism to the drug in question, while clinical breakpoints define a microorganism as resistant by a level of antimicrobial activity associated with a high likelihood of therapeutic failure. Nevertheless, it is outstanding the correlation obtained for flumequine applying CASFM VET and ECOFF (64.2% resistant and 76.4% N-WT, respectively), in comparison with data on enrofloxacin (31.1% resistant and 76.4% N-WT, respectively). Besides, resistance prevalence to marbofloxacin is expected to correlate with enrofloxacin (Vilaró et al., 2022). In our previous study (2005–2017), where the CLSI2020 clinical breakpoints were applied, a high percentage of isolates implicated in colibacillosis exhibited quinolone/fluoroquinolone resistance: 82.3% to nalidixic acid, 56.5% to ciprofloxacin and 48.4% to levofloxacin. To note, 93.2% of the farms involved in the present study reported the use of fluoroquinolones, and 44.6% reported the use of neomycin. Although applying ECOFF values would greatly restrict the therapeutic possibilities of neomycin and enrofloxacin, the high percentage of N-WT determined in both (> 70% commensal and pathogenic E. coli for enrofloxacin; 16.7% commensal and 53.2% of pathogenic E. coli for neomycin) would indicate an overuse, which might compromise therapeutic use soon. The ATS analysis revealed another critical point, which is the fixed range of drug concentrations provided by commercial kits for MIC determination. This greatly limits the flexibility of standard applications, as shown in Supplementary Table S8 for cefoperazone, and trimethoprim/sulfamethoxazole.

Regarding phenicols, we conducted a comparative florfenicol AST using MIC and disk diffusion in 42 isolates. In addition, susceptibility to chloramphenicol was determined in the entire collection by disk diffusion. By applying the clinical EUCAST breakpoint for chloramphenicol, we observed a similar high resistance prevalence (around 50%) compared to the previous study, with no discrepancies within standards. However, contradictory differences were observed in florfenicol resistance (50% “R” vs 31% N-WT, according to CLSI VET 2020 and ECOFF values, respectively). Phenicols are broad-spectrum antimicrobials used in veterinary medicine, although chloramphenicol was banned in EU in 1994 for its use in food-producing animals due to its toxicity and side effects for humans. The resistance level to chloramphenicol maintained over time (observed in this and previous study) can be probably due to different molecular mechanisms conferring resistance to both non-fluorinated (e.g., chloramphenicol) and fluorinated (e.g., florfenicol). Besides, genes encoding phenicol resistance (catA1, cmlA, and floR) are often carried by plasmids together with other resistance genes. Co-selection under the selective pressure of non-phenicol antimicrobial drugs has been reported (Poirel et al., 2018). Florfenicol in pigs is indicated for the treatment of swine respiratory disease caused by Actinobacillus pleuropneumoniae and Pasteurella multocida. The CLSI 2020 references the breakpoint for S. enterica subsp. enterica serovar Choleraesuis “R” > =16, and the suggested ECOFF value for E. coli to detect an N-WT is > 16.

The high prevalence of MDR to ≥ 3 antimicrobial categories (> 75%) observed in our study is of concern. Furthermore, no differences were observed between the groups of commensal and pathogenic isolates, which would indicate that the therapeutic administration is exerting selective and permanent pressure on the commensal population (Zeineldin et al., 2019). On the other hand, the finding of MDR to ≥ 6 categories associated with pathogenic isolates, still implies a greater selective pressure on commensals which might incorporate resistance genes or could be displaced by resistant bacteria with the successful stabilization in the porcine microbiota. Strict control measures are necessary to avoid transmission (MDR bacteria or genes) to humans via food, or to the environment through animal wastes.

Since the E. coli phylogroups are not randomly distributed, the phylogroup assignment described by Clermont et al. (2013); Clermont et al. (2019) is a simple and valuable method to analyze population changes, and even for clinical purposes. Different factors define the phylogeny, such as environment, physiology, diet, and maturation of the digestive tract, host status, or virulence-gene carriage of the bacteria (Escobar-Paramo et al., 2006; Cortés et al., 2010; Clermont et al., 2013). Thus, we apply this method in the diagnostic routine to monitor the evolution of pathogenic clones. The structure of E. coli in pigs is quite homogeneous worldwide, and within commensal pig E. coli, the predominance of phylogroup A, followed by B1 has been reported by different authors (Bok et al., 2013; Reid et al., 2017). Here, we found a similar phylogenetic population structure both in commensal and pathogenic isolates, which were assigned to A, B1, C, D, and E (additionally, one commensal isolate showed phylogroup B2). However, significant differences were detected in the prevalence of certain phylogroups associated with commensal (B1 and C) or pathogenic (E). Bok et al. (2020) reported wider diversity, including in addition the detection of F, or B2. The B2 phylogroup was determined by Bok et al. (2020) in 4.5% of piglets and 2.4% of sows. B2 is a phylogroup rarely reported in domestic pigs, which in our previous surveys was found in 3.6% of 499 pig isolates, associated with the pandemic clonal group ST131. In the present study, we did not recover any B2-ST131 isolate but the phylogroup B2 was determined in one commensal E. coli. The in silico characterization revealed the presence of clone O4:H5-B2-ST12 (CH13-223). The virulome analysis showed the carriage of various traits associated with extraintestinal virulence and, notably, its virulence profile conforms both, the ExPEC and UPEC status. Besides, it was predicted as human pathogen. ST12 is one of the global ExPEC lineages according to Manges et al. (2019), which is frequently implicated in extraintestinal disease, in both humans and domestic animals (mainly dogs, cats, and pigs) (Spindola et al., 2018; Flament-Simon et al., 2020b; Kidsley et al., 2020; Carvalho et al., 2021). The finding of this B2 clone, likewise our previously reported ST131 mcr-1 isolates (García-Meniño et al., 2018), deserves special attention. It would be indicative of swine as a potential reservoir of ExPEC not only for livestock but for humans too (Zhu et al., 2017; Bok et al., 2020; Flament-Simon et al., 2020b). Extraintestinal infections by E. coli such as urinary tract infections (UTI) also cause severe losses in the swine industry, mainly linked to antibiotic therapy, early disposal of breeding sows, or acute death when severely affected (Spindola et al., 2018). Therefore, specific clones such as O4:H5-B2-ST12 (CH13-223), or O25b:H4-B2-ST131 (CH40-374/161) of virotype D5 (García-Meniño et al., 2018; Flament-Simon et al., 2020a), should be monitored under a “One-Health” perspective.

Regarding the present collection of 64 pathogenic isolates, we found here a higher phylogenetic diversity, with a shift in prevalences compared to previous data. While the prior study (2005–2017) on pathogenic E. coli clearly showed the predominance of phylogroup A (85.5%), followed by E (12.4%) and B1 (2.1%), the present collection (2020–2022) exhibited a different distribution: phylogroup A (48.4%), followed by E (25%), B1 (20.3%), D (4.7%) and C (1.6%). These changes in prevalence, and therefore, in the predominant clones involved in swine colibacillosis, were already observed by us in Spain between the periods 2005–2017 and 1986–1991 (García-Meniño et al., 2021). This fact might be explained by the selective pressure derived from different antimicrobial therapies and vaccination programs applied over time.

An outstanding finding in the present study, is the important decrease in colistin-resistant mcr-bearing isolates, from 76.9% within 186 PWD isolates of the period 2005–2017 to 11.3% (present study, within 106 E. coli; p < 0.05). Besides, we have found mcr genes exclusively associated with pathogenic isolates (11.3% vs 0% of commensals; p value < 0.05), and the result is still significant if we compare data only from the group of diarrheagenic isolates: 12 isolates from 64 E. coli, 18.7% (2020–2022) vs 143 out of 186, 76.9% (2005–2017). Differently to our first study, where we observed co-occurrence of mcr-1/mcr-4, mcr-1/mcr-5, and mcr-4/mcr-5 in a significant number of pathogenic pig isolates (García et al., 2018; García-Meniño et al., 2019), the 12 mcr-positive isolates recovered in the period 2020–2022, show single mcr carriage (mcr-1 or mcr-4). Ours would be a unique study, analyzing not only mcr-1 but also other prevalent mcr genes in swine, in both commensal and pathogenic E. coli. The mandatory monitoring in the EU under Commission Implementing Decision (EU) 2020/1729 [Commission Implementing Decision (EU), 2020] is based on phenotypic susceptibility only, and it does not discriminate between different colistin resistance mechanisms. From our point of view, molecular testing is required to confirm the underlying mechanisms of resistance and to gain understanding on the epidemiology of mcr-positive E. coli. Besides, if we apply the standardized phenotypic assays and current clinical cut-off (> 2 mg/l), many mcr-positive E. coli (especially those of the mcr-1 type) remain undetected (García-Meniño et al., 2020). This would be the case of 4 out of the 12 mcr-1 E. coli in the present study. Another limitation of the Decision (EU) 2020/1729 [Commission Implementing Decision (EU), 2020], is that surveys only refer to indicator commensal E. coli. We suggest the monitoring of pathogenic (invasive) E. coli in surveillance programs to accurately determine the persistence of colistin resistance. Our findings on the decrease in colistin-resistant mcr-bearing isolates would reinforce the conclusions of longitudinal studies on the mcr-1 abundance in food-producing pigs, and its relationship with the use of polymyxins (Wang et al., 2020; Miguela-Villoldo et al., 2022). Thus, Miguela-Villoldo et al. (2022) reported the decreasing trend of colistin resistance associated with mcr-1 gene in Spain, since the EMA and AEMPS strategies were applied in 2016 to reduce colistin use in animals. Likewise, Wang et al., 2020 showed a substantial reduction in sales of colistin in animals (2015–2018), after the withdrawal of colistin as growth promoter in China in 2017, which was rapidly followed by a significant reduction in the prevalence of mcr-1 in both the animal and human sectors. Despite the decrease of antibiotic selective pressure due to the colistin consumption reduction, the effect of co-selection by other antimicrobials, such as broad-spectrum cephalosporins, could be causing the persistence of colistin resistance, as recognized in the joint report of the ECDC, EMA and the European Food Safety Authority (EFSA; JIACRA III) (European Centre for Disease Prevention and Control (ECDC); European Food Safety Authority (EFSA); European Medicines Agency (EMA), 2021). In the present study, 23 of the 106 isolates (21.7%) showed resistance to ceftiofur, and the 23 were positive for ESBL genes, without significant difference in commensal vs pathogenic groups (31% vs 15.6%; p > 0.05). The AST analysis of our previous study (2005–2017) determined that 17 (9%) isolates were phenotypically identified as ESBL-producers (including 11 mcr-positive isolates), which would mean a significantly lower figure to present data (12 of 64, 18.7% of diarrheagenic E. coli; p < 0.05) (García-Meniño et al., 2021). The resistant prevalences showed here, both for commensal and pathogenic strains, might be highlighting an increasing use of cephalosporins, in agreement with the voluntary survey, in which 47 (63.5%) out of 74 farms reported the use of ceftiofur.