Pooled E. coli analysis finds shared resistance signatures in South Africa

Multidrug-resistant Escherichia coli from human, food-animal and environmental sources in South Africa shared a large repertoire of antibiotic resistance genes, plasmid-associated features and mobile genetic element signatures, according to a district-level One Health study. Researchers assembled and shotgun-sequenced 20 culture-derived pools representing 539 isolates collected in uMgungundlovu District, KwaZulu-Natal, between August 2017 and February 2019. The analysis detected 247 distinct antibiotic resistance genes, or ARGs, across the pools. Of these, 86 were found in at least one pool from each sector, while 713 of 1,012 plasmid-associated features and 232 of 975 mobile genetic element, or MGE, features were similarly shared. Despite this extensive overlap, the study did not establish direct transmission, active mobility or physical linkage between individual ARGs and mobile elements.

The observational study treated 20 source-specific pools as its independent analytical units. Human material included 12 clinical isolates associated with bloodstream infections or urinary tract infections, divided into two pools. Animal sources contributed 328 pig-associated and 16 poultry-associated isolates, organized into pools representing retail meat, litter and longitudinal pig-fecal collections. The environmental collection comprised 37 soil and litter-fertilizer samples and 146 wastewater and surface-water samples, including influent, effluent, upstream and downstream river water, pig wastewater and chicken wastewater. Pool composition was unbalanced, ranging from three to 46 isolates, and the original farm-to-fork design made pig-associated groups disproportionately numerous. Only two pools represented humans, limiting replication and the strength of comparisons among sectors.

Each isolate was cultured before DNA from colonies within a predefined source group was combined, and the preparations were sequenced as paired-end libraries on an Illumina NextSeq 500 platform. The libraries produced more than 23 gigabytes of uncompressed data. Reads were quality-filtered, assembled into scaffolds and screened for genes predicted to encode resistance. The Comprehensive Antibiotic Resistance Database, ACLAME and MetaStorm database were used to annotate ARGs, plasmid-associated sequences and MGE-associated sequences, respectively. Because the sequencing input came from pooled E. coli cultures rather than untargeted environmental samples, the results describe culture-derived genomic profiles. They do not provide unbiased estimates of microbial-community abundance, isolate-specific carriage, strain prevalence or the distribution of resistance determinants among individual genomes.

Chromosomal determinants accounted for most of the classified resistance signal in every sector: 79.67% in animal pools, 81.07% in environmental pools and 85.82% in human pools. Intrinsic determinants contributed another 3.52%, 3.89% and 3.15%, respectively. The acquired fraction was smaller, representing 16.82% of animal profiles, 15.04% of environmental profiles and 11.03% of human profiles. Antibiotic inactivation and target protection were the largest contributors to the reported mechanism profiles across sectors, while target replacement accounted for 14.15% in animals, 14.18% in the environment and 17.13% in humans. Prominent acquired determinants included vgaC, msrB, tet(D), tet(34), qacH and several aminoglycoside-modifying, dihydrofolate reductase and sulfonamide-resistance genes. Detection indicates resistance potential, but sequence annotation alone cannot establish gene expression, phenotypic resistance or transferability.

Neither clustering nor ordination cleanly separated the sectors. Hierarchical analyses placed human, animal and environmental pools on overlapping branches, while pig-fecal time points, retail meat, litter, wastewater and soil samples were distributed across multiple subclusters. Principal coordinate analysis and nonmetric multidimensional scaling likewise showed extensive overlap. The two clinical pools were close to one another and to several environmental pools, and environmental groups largely occupied the same region as animal groups. A permutational multivariate analysis of variance detected a statistically significant but modest association between source sector and resistome composition, with pseudo-F = 1.585, R² = 0.157 and p = 0.022. Sector therefore accounted for 15.7% of compositional variation, leaving 84.3% unexplained by that classification.

Multivariate dispersion did not differ significantly among sectors in a PERMDISP analysis (F_2,17 = 2.498, p = 0.086), although the limited and uneven sample design limits the strength of that conclusion. Alpha diversity did not differ significantly among human, animal and environmental groups. Tests of ARG richness gave p = 0.940, Shannon diversity p = 0.820 and Simpson diversity p = 0.490, with all adjusted pairwise comparisons remaining nonsignificant. The shared signatures could reflect related bacterial lineages, common selective pressures, repeated independent acquisition from upstream sources, environmental persistence or genetic exchange. The pooled data cannot distinguish among those mechanisms or determine the direction of any movement between sectors. Proximity between pools in the analyses also does not establish epidemiological connectivity.

Plasmid- and MGE-associated matches were widespread across the sampled groups. Human-only detection was uncommon: one plasmid-associated feature was found exclusively in human pools, along with 14 MGE-associated features. In contrast, 285 plasmid-associated features occurred in animal and environmental pools but not in human pools, and 267 MGE-associated features had the same animal-environment distribution. The animal and environmental sectors also shared 59 ARGs exclusively between them, the largest pairwise-only ARG intersection. These intersections are consistent with overlapping material and resistance features among livestock, manure, litter, soil and wastewater systems, but database matches do not demonstrate that the sequences belonged to complete plasmids, transposons or integrons. The analysis did not establish whether an ARG and an MGE were located on the same contig, plasmid, chromosome or bacterial strain, nor did it show that a mobile element was active or capable of transfer.

An exploratory random forest model provided weak evidence that source sector could be predicted from resistance profiles. Its best repeated five-fold cross-validated accuracy was 63.1%, while a tree-matched observed accuracy of 60.5% was compared with a mean permuted-label accuracy of 47.3%. The resulting permutation p value of 0.053 did not meet the study’s threshold for statistical significance. Mean testing accuracy across repeated partitions was 56.2%, with a standard deviation of 15.7 percentage points, and mean out-of-bag error was 47.5%. Although 102 variable ARG predictors were assessed for only 20 pools, and correlated resistance features complicate interpretation, recurrent rankings identified candidates for future investigation. dfrA12 appeared among the top 35 predictors in all 100 runs; Ecol_emrE, mdtN and tet(D) each appeared in 99%, followed by pmrF in 98%. The rankings are internal to the fitted models and do not demonstrate epidemiological relevance or causal source attribution.

MetaCompare produced comparative resistome-risk scores from co-occurrence patterns among ARGs, MGEs and sequences associated with human pathogens. The unweighted mean was 23.26% for animal pools, 19.13% for environmental pools and 19.04% for human pools. Individual pools varied more widely: the January pig-fecal group reached 37%, while chicken wastewater reached 29.16%, compared with 22% for the urinary-tract-infection pool and 16% for the bloodstream-infection pool. Because MetaCompare was designed for assembled metagenomic datasets and used relatively permissive sequence thresholds in this analysis, the scores were treated as within-dataset prioritization measures. They are not probabilities of clinical harm, evidence of active gene mobility or confirmation that any pool was a transmission hub. Variation in sequencing depth, assembly completeness, the number and composition of contributing isolates, and partial sequence recovery can all affect such estimates.

Overall, the results show a broadly shared but heterogeneous E. coli resistome across the sampled human, animal and environmental interfaces, superimposed on modest sector-associated differences. The predominance of chromosomal and intrinsic determinants indicates that the genetic background of E. coli strongly shaped the profiles, while the shared acquired ARGs and mobile-associated features identify targets for integrated surveillance. Resolving whether the similarities arose from clonal dissemination, horizontal gene transfer, common sources or similar local pressures will require balanced longitudinal sampling, isolate-resolved whole-genome sequencing, phenotypic susceptibility testing, phylogenomics, contig-level ARG–MGE analysis and long-read reconstruction of plasmids. Until those steps are completed, the study’s strongest conclusion is shared resistance potential across One Health sectors rather than confirmed transmission.

Subject of Research: One Health metagenomic comparison of resistance and mobile-element signatures in multidrug-resistant Escherichia coli

Article Title: Shared resistome and mobilome signatures across One Health Escherichia coli sources

Article References: Nnah, E. P., Ismail, A., Abia, A. L. K., Essack, S., & Amoako, D. G. (2026). Shared resistome and mobilome signatures across One Health Escherichia coli sources. Frontiers in Bioinformatics, 6, Article 1930054. https://doi.org/10.3389/fbinf.2026.1930054

Image Credits: AI Generated

DOI: 10.3389/fbinf.2026.1930054

Keywords: Escherichia coli, antimicrobial resistance, One Health, resistome, mobile genetic elements, shotgun metagenomics, South Africa, plasmid signatures, environmental surveillance, food animals, Shared, mobilome