Intensive care unit sinks are persistently colonized with multidrug resistant bacteria and mobilizable, resistance-conferring plasmids

Abstract

Contamination of hospital sinks with microbial pathogens presents a serious potential threat to patients, but our understanding of sink colonization dynamics is largely based on infection outbreaks. Here, we investigate the colonization patterns of multidrug-resistant organisms (MDROs) in intensive care unit sinks and water from two hospitals in the USA and Pakistan collected over 27 months of prospective sampling. Using culture-based methods, we recovered 822 bacterial isolates representing 104 unique species and genomospecies. Genomic analyses revealed long-term colonization by Pseudomonas spp. and Serratia marcescens strains across multiple rooms. Nanopore sequencing uncovered examples of long-term persistence of resistance-conferring plasmids in unrelated hosts. These data indicate that antibiotic resistance (AR) in Pseudomonas spp. is maintained both by strain colonization and horizontal gene transfer (HGT), while HGT maintains AR within Acinetobacter spp. and Enterobacterales, independent of colonization. These results emphasize the importance of proactive, genomic-focused surveillance of built environments to mitigate MDRO spread. IMPORTANCE Hospital sinks are frequently linked to outbreaks of antibiotic-resistant bacteria. Here, we used whole-genome sequencing to track the long-term colonization patterns in intensive care unit (ICU) sinks and water from two hospitals in the USA and Pakistan collected over 27 months of prospective sampling. We analyzed 822 bacterial genomes, representing over 100 different species. We identified long-term contamination by opportunistic pathogens, as well as transient appearance of other common pathogens. We found that bacteria recovered from the ICU had more antibiotic resistance genes (ARGs) in their genomes compared to matched community spaces. We also found that many of these ARGs are harbored on mobilizable plasmids, which were found shared in the genomes of unrelated bacteria. Overall, this study provides an in-depth view of contamination patterns for common nosocomial pathogens and identifies specific targets for surveillance.

Keywords: antimicrobial resistance; genomic epidemiology; horizontal gene transfer; hospital surveillance; plasmid ecology; whole-genome sequencing.

Fig 1

Study overview. (A) Overview of collection scheme. Created with BioRender.com. (B) Most commonly recovered species from both this study and D’Souza et al., colored by order. (C) Number of isolates recovered per room, per time point. Faceted by environment and colored by country. Due to a local holiday, Month 4 samples were not collected from PAK sites. (D) Number of isolates collected per room, per collection. Faceted by environment and colored by country.

Fig 2

Overview of strains identified in commonly recovered isolates. Rows represent single strains, and diamonds represent a time point where that strain was identified. Lines drawn between collections of the same strain to highlight persistence. The time between both studies, where samples were not collected, is grayed out.

Fig 3

(A) Maximum likelihood phylogenetic tree of Pseudomonas aeruginosa isolates, generated from core genome alignment. Metadata are colored as rings (from inside out): Strain, Study month, Room class, and Surface. (B) Space and time information for P. aeruginosa isolates. Squares represent at least one isolate was recovered at that time point, and rows represent different rooms. Squares colored by strain identity. Gray blocks indicate no sampling was performed at those time points. (C) Maximum likelihood phylogenetic tree of Acinetobacter johnsonii isolates. (D) Maximum likelihood phylogenetic tree of A. junii isolates. (E) Space and time information for Acinetobacter isolates. Shape represents species, and as in panel (C), the x-axis represents study month and the y-axis represents room. Gray blocks indicate no sampling was performed at those time points.

Fig 4

Overall ARG content of genomes. (A, B) Scatterplot showing the number of ARGs and number of ARG classes per genome for all isolates (as predicted by AMRFinder), with small amount of jitter added for visibility. *** = P < 0.001, * = P < 0.05 by Wilcoxon rank-sum test. (C, D) Stacked bar plots showing absolute number of isolates collected from each genus in USA/PAK, colored by room class. Bars to the left represent genera found more in ICUs, and bars to the right represent genera found more in HOME/WORK rooms. Each bar is annotated with heatmaps showing the mean number of ARGs (red/orange) and ARG classes (green/blue) in each genus. *** = P < 0.001, ** = P < 0.01, * = P < 0.05 by pairwise Wilcoxon tests with Benjamini-Hochberg adjustment. Individual group sizes and ARG counts are noted in Supplementary Data. (E) Balloon plot showing the average number of times each ARG appears in a genome from each genus (num_ARG_appearances/num_genomes), colored by ARG class. For visibility, only the top 35 ARGs found in the data set are shown. ARG, antibiotic resistance gene; ICU, intensive care unit.

Fig 5

HGT events within the sink environment. (A) Phylogenetic cladogram of all 822 genomes in this study, generated using GTDB-Tk and RAxML. The outer ring is colored by taxonomic order, and the lines connecting each node represent shared genomic space between those two genomes by BLAST alignment (>5 kbp in length, >99% identity, and >95% coverage). Lines are colors blue if the two genomes are the same species, yellow if they are different species, and pink if they are different species and the shared genomic spaces encodes an ESBL or carbapenemase. (B) Network showing the 11 clusters of plasmid sequences identified using nanopore sequencing and sequence alignment. Each node represents a single contig, colored by genus. An edge connecting two nodes represents a significant BLAST alignment between those two contigs (>5 kbp in length, >99% identity, and >95% coverage, <10% difference in contig size, at least one contig is circularized). (C) Balloon plot showing the most commonly shared ARGs different taxonomic levels. For each ARG encoded on a shared genomic space, the number of unique taxa combinations sharing that ARG was counted at different taxonomic levels. For visibility, only the most commonly-shared ARGs are shown. (D) Nucleotide alignment of plasmid Clusters 10, 7, and 2. Gray blocks show BLAST matches of >99% ID and >5 kb, with SNV counts on the left, and the ANI (based on SNVs) noted on the right. ORFs are colored by function (pink = ARG, orange = MGE, teal = other). ANI, average nucleotide identity; ARG, antibiotic resistance gene; ESBL, extended-spectrum β-lactamase; HGT, horizontal gene transfer; SNV, single-nucleotide variant.