henhaobio
Sign in

Nature CommunicationsMicrobiologyPeer-reviewed journal article

Genetic compatibility and ecological connectivity drive the dissemination of antibiotic resistance genes

David LundiD, Marcos Parras-MoltóiD, Juan Salvador Inda-DíaziD, Stefan EbmeyeriD, D. G. Joakim LarssoniD, Anna JohnningiD, Erik KristianssoniD

PDFDOIPreprintCode

In-depth summary

Claude Sonnet 4.6 · xhigh effort · 7 Nov 2026 · from the full text of version 1 (16 Oct 2024) · CC BY 4.0

Random forests using nucleotide-composition dissimilarity and metagenomic co-occurrence predict horizontal transfer of resistance genes between bacteria (general model AUROC 0.873); mismatch limits transfer, human and wastewater co-occurrence favors it.

The paper states that the flow of mobile resistance genes into and between pathogens is poorly understood, and that earlier studies mostly examined mobile genetic elements or biochemical functions rather than genome composition or ecological connectivity. It combines phylogeny-inferred transfers with composition dissimilarity and metagenomic co-occurrence in predictive models.

Key results

  1. ShownRandom forests separated observed ARG transfers (between hosts differing at order level or above) from randomized pairs with high accuracy, in a general model and in seven mechanism-specific models.

    General model (n=1,565): mean AUROC 0.873, sensitivity 0.814, specificity 0.780 over ten resampled iterations on a 70/30 split. Mechanism-specific mean AUROC ranged from 0.810 (APH) to 0.930 (Tet RPG). · Random forests can accurately predict the…; Inference on horizontally spread ARGs using…

  2. ShownGreater nucleotide-composition dissimilarity, between the two host genomes and between the ARG and its host genome, reduced the likelihood of transfer, most clearly between distant hosts.

    Genome-genome 5-mer Euclidean distance above about 0.025 contributed negatively. Observed versus randomized transfers differed at class and phylum level (p<10^-15, Wilcoxon) and at order level (p=0.012). The effect was strongest for tetracycline efflux pumps. · Random forests can accurately predict the…; High genetic incompatibility decreases the…

  3. ShownHost pairs that co-occur in metagenomes were more likely to have exchanged ARGs, with the strongest signal in human and wastewater microbiomes and with patterns that depend on the resistance mechanism.

    In the general model, human co-occurrence had the highest importance, followed by wastewater and animals. APH and class A, C, D beta-lactamases were linked most to human co-occurrence, AAC to wastewater. Of frequent pairs (at least 5 transfers), 63.3% co-occurred in at least one environment. · Random forests can accurately predict the…; Co-occurrence in human and wastewater…

  4. SuggestedCo-occurrence is environment specific, so no single environment explains all transfers. The authors conclude that most documented transfers probably occurred in human and wastewater microbiomes.

    23.8% of frequently transferring hosts co-occurring in human samples were not seen together in any wastewater metagenome; 3.3% in the reverse direction. For proposed origin species, connections in human 108, wastewater 81, animal 46, soil 24, water 23. · Co-occurrence in human and wastewater…; Discussion

  5. ShownComposition dissimilarity does not simply track taxonomy: some host pairs from different phyla are compositionally closer than many pairs within one phylum.

    Multidimensional scaling of 5-mer distributions from 7,609 ARG-carrying species showed low dissimilarity between Bacillota and Bacteroidota, and between Bacillota and Campylobacterota, among others. · High genetic incompatibility decreases the…; Inference on horizontally spread ARGs using…

Methods, data and reproducibility

Genomes were screened for ARGs with fARGene and gene trees were built per ARG class. Transfers were called at nodes whose descendants came from hosts differing at order level or above. Each transfer was described by 5-mer composition distances, genome size difference, OTU-based metagenomic co-occurrence in five environments, and Gram staining. Random forests were trained against permuted-leaf negatives.

  • Bacterial genomes from NCBI Assembly (downloaded 2022-04-04), screened for ARGs · bacteria · n = 867,318 genomes; 2,666,002 ARGs (60,773 unique protein sequences)
  • Horizontal ARG transfers identified from gene trees (1,565 after missing-value removal and downsampling, used in the general model) · bacteria · n = 6,276 transfers
  • Metagenomic samples from the Earth Microbiome Project and the Global Water Microbiome Consortium, used for co-occurrence in animal, human, soil, water and wastewater · bacterial communities · n = 24,417 downloaded; 20,816 retained
Methods & tools
fARGene v0.1, USEARCH v8.01445, mafft v7.458, FastTree v2.1.10, BLAST+ v2.10.1, CARD, randomForest v4.7-1.1, rfPermute v2.5.1, pROC v1.18, pdp v0.8.1, vegan v2.6-4, igraph v1.4.3
Software & environment
Software versions are given for the main tools (e.g. fARGene v0.1, mafft v7.458, FastTree v2.1.10, randomForest v4.7-1.1). No container or requirements file is mentioned.
Reproducibility gaps
The data statement only says all data are public and cited, with no URLs. The code statement covers scripts for the positive and negative datasets (also Zenodo doi 10.5281/zenodo.13827913), not necessarily the full analysis.
Limitations (5)
  • Only about 6,000 transfers were found despite almost one million genomes, which the authors call a clear underestimation. Too few observations existed to train models for Mph, Qnr and tetracycline-inactivating enzymes. · stated by the authors
  • Co-occurrence levels depend on the number of samples per environment and their heterogeneity. About half of the species pairs involved in transfer were not detectably co-occurring in any environment. · stated by the authors
  • The authors consider their results conservative, since low co-occurrence in soil and water may not hold for undiscovered resistance determinants or uncultured species. · stated by the authors
  • Only 33.82% of genomes could be assigned an OTU, and 3,419 of 6,276 transfers (54.48%) were dropped before modeling because of missing co-occurrence values. · noted in this summary
  • Negatives are random leaf pairs from the same ARG trees, and performance is measured on a random 30% split of the same data. No external validation set is reported. · noted in this summary

Read this if you study horizontal gene transfer of resistance genes and want evidence on how composition mismatch and host co-occurrence shape which bacteria exchange them.

Citations
90
FWCI
19.45
Field percentile
top 1%

Abstract

The dissemination of mobile antibiotic resistance genes (ARGs) via horizontal gene transfer is a significant threat to public health globally. The flow of ARGs into and between pathogens, however, remains poorly understood, limiting our ability to develop strategies for managing the antibiotic resistance crisis. Therefore, we aimed to identify genetic and ecological factors that are fundamental for successful horizontal ARG transfer. From the analysis of [~]1 million bacterial genomes and >20,000 metagenomes, we developed random forest models that could reliably predict horizontal ARG transfer between bacteria. Our results suggest that genetic incompatibility, measured as nucleotide composition dissimilarity, negatively influences the likelihood of transfer of ARGs between evolutionarily divergent bacteria. Conversely, environmental co-occurrence increased the likelihood, especially in humans and wastewater, in which several environment-specific dissemination patterns were observed. This study provides new ways to predict the spread of ARGs and provides new insights into the mechanisms governing this evolutionary process.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISForster '22Parras-Molt '25Lund '25Kittredge '21Kintses '18Yang '26
75% semantic matchGenomics
Samuel C. Forster, Junyan Liu, Nitin Kumar+10 more·bioRxiv·Jan 18, 2022