29Antimicrobial resistance (AMR) is a global health threat, especially in low-/middle-income countries 30 (LMICs), where there is limited surveillance to inform empiric antibiotic treatment guidelines. 31Enterobacterales are amongst the most important causes of drug-resistant bacterial infections. We 32 developed a novel AMR surveillance approach for Enterobacterales by profiling pooled human faecal 33 metagenomes from three sites (n=563 individuals; Cambodia, Kenya, UK) to derive a taxonomy-34 adjusted AMR metric ("resistance potential") which could be used to predict the aggregate percentage 35 of resistant invasive Enterobacterales infections within each setting. Samples were sequenced 36 (Illumina); taxonomic and resistance gene profiling was performed using ResPipe. Data on organisms 37 causing bacteraemia and meningitis and antibiotic susceptibility test results from 2010-2017 were 38 collated for each site. Bayesian generalised linear models with a binomial likelihood were fitted to 39 determine the capacity of the resistance potential to predict AMR in Enterobacterales infections in 40 Antimicrobial resistance (AMR) is a global health emergency 1 , and imposes a particularly large 48 socioeconomic burden in resource-limited settings, where bacterial infections and several other 49 drivers of AMR commonly co-occur and effective antibiotics may be unavailable or unaffordable 2 . A 50 key pillar in AMR mitigation is the development of effective and standardised AMR surveillance, to 51 monitor trends, inform empiric treatment guidelines, identify emerging AMR threats, and monitor the 52 impact of interventions. There has been significant investment in surveillance capacity, such as by the 53 UK's Fleming Fund, and an attempt to promote standardised collection, analysis and sharing of global 54 AMR data with an emphasis on capturing clinical and microbiological information, encapsulated in 55 the WHO Global Antimicrobial Resistance Surveillance System (GLASS) 3 . However, limitations in 56 implementing GLASS include the time taken to develop robust infrastructural capacity to support data 57 collection in regions where AMR is most relevant or prevalent, and the difficulty in obtaining 58 systematic datasets even from enrolled countries with adequate infrastructure, especially outside 59 tertiary or University centres. Surveillance strategies which could bridge or complement the 60 implementation of approaches such as GLASS would be helpful. 61
62Colonisation with specific species and/or drug-resistant organisms, such as nasal colonisation with 63 Staphylococcus aureus 4 , or rectal colonisation with carbapenemase-producing Enterobacterales 5 , is 64 associated with risk of infection by these organisms. Metagenomic approaches are less biased than 65 targeted approaches which capture specific organism/resistance phenotypes of interest, and obviate 66 the need for culturing individual organisms. Resistance gene abundances and taxonomic distributions 67 in metagenomes are increasingly mined for a range of applicat...