Abstract
BackgroundAntimicrobial resistance (AMR) in Enterobacterales is a global health threat. Capacity for individual-level surveillance remains limited in many countries, whilst population-level surveillance approaches could inform empiric antibiotic treatment guidelines.MethodsIn this exploratory study, a novel approach to population-level prediction of AMR in Enterobacterales clinical isolates using metagenomic (Illumina) profiling of pooled DNA extracts from human faecal samples was developed and tested. Taxonomic and AMR gene profiles were used to derive taxonomy-adjusted population-level AMR metrics. Bayesian modelling, and model comparison based on cross-validation, were used to evaluate the capacity of each metric to predict the number of resistant Enterobacterales invasive infections at a population-level, using available bloodstream/cerebrospinal fluid infection data.FindingsPopulation metagenomes comprised samples from 177, 157, and 156 individuals in Kenya, the UK, and Cambodia, respectively, collected between September 2014 and April 2016. Clinical data from independent populations included 910, 3356 and 197 bacterial isolates from blood/cerebrospinal fluid infections in Kenya, the UK and Cambodia, respectively (samples collected between January 2010 and May 2017). Enterobacterales were common colonisers and pathogens, and faecal taxonomic/AMR gene distributions and proportions of antimicrobial-resistant Enterobacterales infections differed by setting. A model including terms reflecting the metagenomic abundance of the commonest clinical Enterobacterales species, and of AMR genes known to either increase the minimum inhibitory concentration (MIC) or confer clinically-relevant resistance, had a higher predictive performance in determining population-level resistance in clinical Enterobacterales isolates compared to models considering only AMR gene information, only taxonomic information, or an intercept-only baseline model (difference in expected log predictive density compared to best model, estimated using leave-one-out cross-validation: intercept-only model = -223 [95% credible interval (CI): -330,-116]; model considering only AMR gene information = -186 [95% CI: -281,-91]; model considering only taxonomic information = -151 [95% CI: -232,-69]).InterpretationWhilst our findings are exploratory and require validation, intermittent metagenomics of pooled samples could represent an effective approach for AMR surveillance and to predict population-level AMR in clinical isolates, complementary to ongoing development of laboratory infrastructures processing individual samples.
Highlights
Antimicrobial resistance (AMR) is a global health emergency [1], in resource-limited settings, where effective microbiological services and antibiotics may be unavailable [2]
The results of the validation of our pooling approach, which consider metagenomic information from 30-sample pools and individual samples to assess whether pooled metagenomes are a fair representation of the individual metagenomes, are provided in the supplementary appendix
Bayesian model predictions expressed as percentages instead of counts are shown in appendix p17 for antibiotics with antibiotic susceptibility test (AST) results from >100 invasive infection isolates (i.e. 14 antibiotics in the United Kingdom (UK) and/or Kenya). In this exploratory study we suggest that metagenomic analysis of pooled extracts from individual faecal samples could be effective at predicting resistance in invasive Enterobacterales infections from different age groups and geographic settings at the population-level, if both AMR gene abundances and taxonomy metrics from the pooled metagenomes are considered
Summary
Antimicrobial resistance (AMR) is a global health emergency [1], in resource-limited settings, where effective microbiological services and antibiotics may be unavailable [2]. There has been significant investment in individual/patient-level surveillance, and an attempt to promote standardised collection, analysis and sharing of global AMR data, capturing both clinical and microbiological information [3]. Limitations of these approaches include developing and sustaining robust capacity in regions where AMR is most prevalent, and in obtaining systematic data even from countries with adequate infrastructure. To our knowledge, no study has used taxonomic and AMR gene profiles in pooled metagenomes to directly estimate AMR prevalence amongst clinical isolates in populations in the same setting.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
