An omics-based framework for assessing the health risk of antimicrobial resistance genes

Abstract

Antibiotic resistance genes (ARGs) are widespread among bacteria. However, not all ARGs pose serious threats to public health, highlighting the importance of identifying those that are high-risk. Here, we developed an 'omics-based' framework to evaluate ARG risk considering human-associated-enrichment, gene mobility, and host pathogenicity. Our framework classifies human-associated, mobile ARGs (3.6% of all ARGs) as the highest risk, which we further differentiate as 'current threats' (Rank I; 3%) - already present among pathogens - and 'future threats' (Rank II; 0.6%) - novel resistance emerging from non-pathogens. Our framework identified 73 'current threat' ARG families. Of these, 35 were among the 37 high-risk ARGs proposed by the World Health Organization and other literature; the remaining 38 were significantly enriched in hospital plasmids. By evaluating all pathogen genomes released since framework construction, we confirmed that ARGs that recently transferred into pathogens were significantly enriched in Rank II ('future threats'). Lastly, we applied the framework to gut microbiome genomes from fecal microbiota transplantation donors. We found that although ARGs were widespread (73% of genomes), only 8.9% of genomes contained high-risk ARGs. Our framework provides an easy-to-implement approach to identify current and future antimicrobial resistance threats, with potential clinical applications including reducing risk of microbiome-based interventions.

Fig. 1. Survey into the distribution of total antibiotics and antibiotic resistance genes (ARGs) in diverse environments to inform the development of the risk framework.

a The total concentrations of total antibiotics (filled circles for ng/L and open circles for ng/g) in different habitats from 30 studies (Supplementary Data 1). Each circle represents a sample, with color representing its habitat. b The abundance of total ARGs and abundance of ARGs in different habitats. Each circle represents a metagenomic sample, with color representing its habitat. The abundance of ARGs was normalized into copy of ARGs per cell (ARGs-OAP, more details in “Methods” section). c PCA plotting ARG composition clustered 854 metagenomic samples from seven global eco-habitats into undisturbed natural environments (green panel covering permafrost, soil, sediment, and water) and human associated environments (red panel covering wastewater treatment plants (WWTPs), animal feces and human feces). Samples were clustered into three habitats (from undisturbed natural habitats, to WWTPs, to feces) along the primary axis of “human-association”, with its color representing its habitat. Data in a are presented as mean values (center) and 25%, 75% percentiles (bounds of box). Data in b and c are presented as median values (center) and 25%, 75% percentiles (bounds of box). The minima and maxima represent the range of the data.

Fig. 2. A conceptual model of the evolution and emergence of antibiotic resistance genes (ARGs) and a framework for assessing ARG risk to human health.

a We designed a conceptual model to demonstrate the evolution and emergence of ARGs accelerated by selective agents of antibiotics. b Based on this model, we further designed a framework to reflect the natural progression of the evolution and emergence of ARGs and the risk ranks of all ARGs reference sequences of the initial set in the Structured ARG Database v1.0. The color represents the rank of ARGs (Ranks I–IV). The relevant data was obtained by searching the ARGs in all available bacterial genomes and plasmids from NCBI, mobile genetic element (MGE) databases, and 854 global metagenomes using the ARGs Online Searching Platform. HGT horizontal gene transfer, ARGs antibiotic resistance genes.

Fig. 3. The risk ranks of 37 known high risk ARG families identified by World Health Organization (WHO) and literature review (highlighted in dark purple) (referred to as “listed” ARGs), 73 Rank I ARG families of the initial set, and 19 Rank II ARG families of the initial set.

“Listed” (WHO-listed) represented Rank I ARGs that were reported to cause problems in hospitals and/or have been widespread on mobile genetic elements by previous literatures (highlighted in dark purple). “Unlisted” represented Rank I ARG families that were not reported in previous studies as high risk ARGs. “Homologs in Rank I” represented Rank II ARGs with homologous Rank I ARGs of the same gene family (highlighted in light purple). ARGs antibiotic resistance genes.

Fig. 4. Validation of the framework in this study by three clinical datasets collected after the initial analysis between 02/20/2019 to 02/10/2021 as the training datasets were collected on 02/19/2019.

a The content of three datasets. Dataset1 contains the chromosomal genomes and plasmids collected from opportunistic pathogen isolates cultured from a tertiary hospital after 02/20/2019. Dataset1 covers a total of 94 taxa (2,347 chromosomal genomes and 5,910 plasmids). Dataset2 contains 116 metagenomes collected after 02/20/2019 from a tertiary hospital. Dataset3 contains all NCBI pathogen genomes submitted in the last 2 years (02/20/2019 to 02/10/2021). Dataset3 covers a total of 362 taxa (42,481 genomes). ARG sequences of the initial set were searched in three testing datasets using the same criteria described before (90% aa similarity over 80% aa hit length). b the percentage of unique ARGs of different Ranks in hospital plasmids. Asterisk of box represent the p value for fisher exact test. n = 16 (No. of unique ARGs). c The mean abundance of ARGs of different Ranks in 116 hospital metagenomes. Quantiles (represented as bars), mean (represented as points), and standard error (represented as error bars) were computed for all ARGs in an ARG Rank. Asterisk of box represent the p value for one way anova test. n = 1393 (No. of unique ARGs detected in hospital metagenomes). Adjustments were made for multiple comparisons. Data are presented as mean values ± SE (error bars), median values (center of box), and 25%, 75% percentiles (bounds of box). The minima and maxima represent the range of the data. d The percentage of unique ARGs that were found in pathogens after 02/20/2019 and were not detected in pathogens before 02/20/2019. Asterisk of box represent the p value for fisher exact test. n = 76 (No. of unique ARGs newly present in pathogens). Adjustments were made for multiple comparisons. The color represents the rank of ARGs (Rank I on WHO list, Rank I not on WHO list, Rank II-IV). ARGs antibiotic resistance genes.