# Mapping the underlying drivers of resistome risk across diverse environments

## TL;DR

This study maps how environmental factors influence antimicrobial resistance risk using the MetaCompare 2.0 framework across 12 environments.

## Contribution

Expands MetaCompare to distinguish ecological and human health resistome risks and identifies key factors influencing these scores.

## Key findings

- ERR and HHRR scores were highest in transient, human-linked environments like wastewater and the gut.
- Taxonomic diversity was inversely correlated with resistome risk, supporting the niche occupation hypothesis.
- ARG relative abundance only correlated with risk in high-diversity, low-coverage environments due to assembly limitations.

## Abstract

Understanding the drivers of antimicrobial resistance (AMR) across the One Health spectrum is crucial for controlling its spread. The MetaCompare framework, which assesses “resistome risk” based on antibiotic resistance gene (ARG) co-occurrence patterns on metagenomic contigs, has been expanded to distinguish between “ecological resistome risk” (ERR) and “human health resistome risk” (HHRR) scores across anthropogenic gradients. However, comprehensive surveys are still needed to untangle the biological (e.g., ARG relative abundance), ecological (e.g., taxonomic diversity), and technical (e.g., coverage) factors influencing these risk scores. Here, we analyzed 1,326 metagenomes from 12 key environments using the MetaCompare 2.0 pipeline to map global ERR and HHRR landscapes, identifying significant factors modulating risk scores through network analysis, machine learning, and multivariate regression models.

ERR and HHRR scores varied significantly across environments and were highly correlated (ρ = 0.73, p < 2e-16), indicating shared underlying drivers. Transient environments closely linked to human activity, such as wastewaters and the human gut, produced the highest ERR and HHRR scores, while stable environments like sediments, soils, and activated sludge yielded the lowest. These patterns corresponded directly with taxonomic diversity, where more diverse ecosystems exhibited lower risk scores, supporting the hypothesis that niche occupation may act as an ecological barrier to ARG invasion. In contrast, scores were positively correlated with sul1 and crAssphage, further confirming that transient, low-diversity environments have higher resistome risks, although they did not fully account for risk variability across all environments. ARG relative abundance correlated with risk scores, but only in high-diversity, low-coverage environments due to poor assembly quality and an inability to resolve ARG flanking regions. The ARGs contributing to ERR and HHRR scores were largely aligned with existing ARG risk ranking frameworks.

This study demonstrated how the MetaCompare 2.0 pipeline can effectively disentangle complex relationships between ARG abundance, composition, and environmental context. Although robust across diverse environments, the framework’s ability to detect ARGs and their co-occurrences may be limited in high-diversity, low-coverage samples, such as soils and sediments. Finally, we provide a series of recommendations for appropriate use cases for MetaCompare 2.0.