Antibiotic-resistant organisms establish reservoirs in new hospital built environments and are related to patient blood infection isolates.

Background: Healthcare-associated infections due to antibiotic-resistant organisms pose an acute and rising threat to critically ill and immunocompromised patients. To evaluate reservoirs of antibiotic-resistant organisms as a source of transmission to patients, we interrogated isolates from environmental surfaces, patient feces, and patient blood infections from an established and a newly built intensive care unit.
Methods: We used selective culture to recover 829 antibiotic-resistant organisms from 1594 environmental and 72 patient fecal samples, in addition to 81 isolates from blood cultures. We conducted antibiotic susceptibility testing and short- and long-read whole genome sequencing on recovered isolates.
Results: Antibiotic-resistant organism burden is highest in sink drains compared to other surfaces. Pseudomonas aeruginosa is the most frequently cultured organism from surfaces in both intensive care units. From whole genome sequencing, different lineages of P. aeruginosa dominate in each unit; one P. aeruginosa lineage of ST1894 is found in multiple sink drains in the new intensive care unit and 3.7% of blood isolates analyzed, suggesting movement of this clone between the environment and patients. Conclusions: These results highlight antibiotic-resistant organism reservoirs in hospital built environments as an important target for infection prevention in hospitalized patients.

Introduction

Healthcare-associated infections (HAIs) are a global challenge, posing a particularly acute threat in intensive care units (ICUs) where critically ill and immunocompromised patients are at elevated risk for infection during their stay[1,2]. Worldwide, HAIs are responsible for an estimated 2.5 million infections every year and are associated with increased morbidity, mortality, and healthcare costs[1,3-5]. The COVID-19 pandemic is associated with further expansion of hospitalized critically-ill individuals[6]. HAIs due to AROs in the ICU can be difficult to treat due to limited treatment options; available options are also associated with toxicity, are poorly tolerated by patients, and may exhibit negative interactions with other drugs[1,7,8].

Many studies and initiatives have focused on trying to limit HAIs through surveillance, prevention, and intervention[1,9,10]. Recent studies have used culture-independent metagenomic sequencing of hospital surfaces to generate an important catalog of the diversity and composition of their resident microbial communities[11-15]. However, metagenomic characterizations are limited in their ability to track viable, antibiotic-resistant strains and remain ambiguous to whether the taxa discovered on surfaces are environmental- or patient-derived, and/or associated with infections in patients. To better understand relationships between viable antibiotic-resistant organisms (ARO) in the built environment and critically-ill patients, we must determine 1) what hospital surfaces are acting as ARO reservoirs, i.e., surfaces where an organism can be cultured from multiple time points; 2) what are the spatial and temporal dynamics of reservoir colonization; and 3) whether viable ARO strains colonizing the hospital built environment can also be detected from human clinical infections.

There are multiple models proposed for ARO reservoir colonization and transmission in hospitals (Fig. 1a)[1,16-18]. A prominent model is that AROs are shed from colonized patients, frequently through fecal contamination, to surfaces, instruments, and shared equipment in patient rooms (Fig. 1a)[19,20]. High-touch hospital surfaces can act as intermediate ARO reservoirs, and transmission may occur from these reservoirs through patients, healthcare staff, and visitors[10,20-23]. Another model is that AROs are seeded from microbial communities which persistently colonize hospital built environments, particularly plumbing sources, where biofilms form and can act as a reservoir for potential pathogens (Fig. 1a)[24-27]. These models are not mutually exclusive. ARO reservoirs are likely dependent on a given facility’s history and modes of transmission likely interact within a hospital[28]. To better understand the colonization and transmission of AROs in the hospital built environment, we leveraged a unique opportunity to sample a newly-built stem cell transplant and oncology (SCT) ICU both before patient and staff occupancy and for one year after ICU establishment. This allowed us to identify and track persistent colonization of sink drains by AROs that began prior to patient and staff occupancy, a facet that has not been characterized in previous studies. As immunocompromised cancer patients demonstrate prolonged duration of ARO shedding and are at high risk of HAIs, the SCT ICU is a critical environment to study ARO surface colonization and transmission[29-32]. Additionally, we compared this new ICU environment (new ICU) with environmental samples from the established SCT ICU previously housing these patients and staff (old ICU). While previous studies have longitudinally tracked surface and patient samples within an ICU, they have been limited in their ability to discern the impact of the facility built environment from the population of patients and healthcare workers in the facility. Here, the same patients and healthcare providers transitioned between the old and new buildings across the study period, allowing for a direct comparison between their ARO communities.

Fig. 1

ARO reservoir colonization models and sample processing scheme.

a Two models of reservoir colonization. Model 1 shows antibiotic-resistant organism (ARO) transmission from patients to hospital surfaces and then to other patients. Model 2 shows ARO transmission from environmental reservoirs to hospital surfaces to patients. b Sample collection time points and sample processing scheme from surface collections to WGS. In sample collection scheme, large circles represent months with small circles representing 2-week sampling within months. Purple indicates old intensive care unit (ICU) collections, green indicates new ICU collections, and pink indicates collections taken before patients enter the building in the new ICU. Icons labeled as such were acquired from nounproject.com, and other icons were used with permission from D’Souza, Potter et al.[82]. AST antibiotic susceptibility testing, MALDI-TOF matrix-assisted laser desorption/ionization-time of flight mass spectrometry.

To track ARO transmission events between patients and ICU surfaces, we collected remnant fecal samples from patients in the SCT ICU who had laboratory studies ordered on fecal samples and isolates from positive blood cultures ordered as part of routine clinical care during the same collection period. From this unique collection of environmental and patient samples, we used selective microbiologic culturing and whole-genome sequencing (WGS) to identify AROs, assess antibiotic resistance, and track strains across time and location.

We found ARO contaminants were rare on most ICU surfaces but prevalent in sink drains in both ICUs, with the old ICU having significantly higher ARO burden in sink drains than the new ICU. AR Enterobacterales, which are frequently associated with fecal contamination, were rarely found on surfaces. In both ICUs, Stenotrophomonas spp. and Pseudomonas spp. were the two most frequently collected genera; however, different lineages dominated each ICU. Stenotrophomonas maltophilia strains formed months-long reservoirs in sink drains in the new ICU with no evidence of strains association with bloodstream infections during our study time period. In contrast, Pseudomonas aeruginosa strains formed persistent reservoirs for most of the year in the new ICU in multiple sink drains and showed evidence of shared strains across environmental samples and patient blood cultures. These results provide evidence that sink drains in the healthcare environment can serve as ARO reservoirs that are associated with human clinical infections.

Methods

Sample collections and culturing

Environmental and fecal samples received a non-human subjects determination by the Institutional Review Board (IRB) of Washington University (201712083). Blood culture clinical isolate collection was reviewed and approved by IRB (201901053) and by the Siteman Cancer Center Protocol Review and Monitoring Committee. We received IRB approval and Siteman Cancer approval for clinical isolates from patients. The IRB granted a waiver of informed consent for the collection of these specimens because they had been collected as part of routine clinical care. We sampled 6 SCT ICU (old ICU) rooms 3 times over the course of 1 month in the old building from December 2017 – January 2018. At each time point, nine surfaces were sampled using Eswab collections (Copan) pre-moistened with molecular water: the foam dispenser, the gown and glove storage area, the bedside rail, the nursing call button, the room floor, the light switch, the computer, the in-room sink handles, and the in-room sink drain. Three swabs were held together to simultaneously sample each surface. We also collected 2 samples of 15 mL in-room sink water directly from the faucet: 1 sample was collected immediately after turning the faucet on, and 1 sample was collected after allowing the water to run for 2 min.

We sampled 6 SCT ICU (new ICU) rooms and communal SCT ICU areas every other week for 5 months and then every month for 1 year in the new building for a total of 21 samplings (Fig. 1b). Samples were collected twice during the first week of sample collections in the new ICU building: the first after construction terminal clean and the second after custodial terminal clean. Both time points collected were before patients and staff had entered the ICU. At each time point, the same nine patient room surfaces as described above were sampled plus an additional 3 surfaces: the sofa from the patient room, the bathroom toilet from adjoining bathroom, and the sink drain from the adjoining bathroom. We also collected 15 mL of in-room sink water and bathroom sink water. At each time point, we also sampled four communal surfaces: the housekeeping closet drain, the family area floor, the soiled utility room drain, and the vending machine. For each time point in both buildings, we obtained remnant de-identified fecal specimens that had been submitted to the clinical microbiology laboratory for C. difficile testing from patients in the same unit as surface swab collection.

Eswab specimens from surfaces, water samples and fecal samples were cultured the same day of sampling. Eswab specimens were vortexed and 90 µL of eluate was used for culture inoculation per plate/test condition. For fecal specimens, 90 µL of specimen was used for culture inoculation. For water samplings, 100 µL of vortexed water sample was used for culturing. All samples were inoculated to each of the following culture medium: Sheep’s blood agar (Hardy), VRE chromID (bioMerieux), Spectra MRSA (Remel), HardyCHROM ESBL (Hardy), MacConkey agar with cefotaxime (Hardy), Cetrimide agar (Hardy), and Sabouraud dextrose + chloramphenicol (Hardy). Plates were incubated at 35 °C in an air incubator and incubated up to 48 h prior to discard if no growth (up to 7 days for sabouraud dextrose + chloramphenicol). Two colonies of each colony morphotype were subcultured and identified using matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALD-TOF MS) with the VITEK MS system. All isolates recovered were stored at −80 °C in TSB with glycerol.

Isolates recovered from standard-of-care blood cultures during the same time frame of the surface sampling were recovered from frozen stocks in the clinical microbiology laboratory.

Antimicrobial susceptibility testing

Antimicrobial susceptibility testing (AST) was performed using Kirby Bauer disk diffusion, interpreted according to CLSI standards[33]. AST was performed on gram negative bacilli using ampicillin, cefazolin, cefotetan, ceftazidime, ceftriaxone, cefepime, meropenem, ciprofloxacin, levofloxacin, piperacillin-tazobactam, ceftolozane-tazobactam, ceftazidime-avibactam, ampicillin-sulbactam, trimethoprim-sulfamethoxazole, gentamicin, amikacin, fosfomycin, colistin, aztreonam, doxycycline, minocycline, and nitrofurantoin and antimicrobials were interpreted/reported as appropriate for the specific species. We also performed a carbapenamase inactivation assay on all Enterobacterales and Pseudomonas isolates that were resistant or intermediate to meropenem or imipenem.

Short read sequencing

Total genomic DNA was extracted from cultured isolates using the Bacteremia kit (Qiagen, Gernmantown, MD, USA) and DNA was quantified using the PicoGreen dsDNA assay (Thermo Fisher Scientific, Waltham, MA, USA). A total of 5 ng/µL was used as input for Illumina sequencing libraries with the Nextera kit (Illumina, San Diego, CA, USA). The libraries were pooled and sequenced on a NextSeq HighOutput platform (Illumina) to obtain 2x150bp reads. The reads were demultiplexed by barcode and had adapters removed with Trimmomatic[34]. Reads are available under BioProject PRJNA741123 (http://www.ncbi.nlm.nih.gov/bioproject/741123). Processed reads were assembled into draft genomes using SPAdes v3.11.0[35]. Assemblies were assessed for quality using Quast v3.2[36] and checkM v1.0.13[37]. Assemblies were considered to have passed quality standards if completeness was greater than 90% and contamination was below 5%. We used Prokka on the assembled genomes to identify and annotate open reading frames[38].

Long read sequencing

Isolates were streaked from frozen stocks onto LB agar and allowed to grow at 37 °C for 48 h prior to extraction. Lawns were scraped from plates into nuclease free water. Genomic DNA was extracted using the bacteremia kit (Qiagen, Gernmantown, MD, USA), with the modification of limiting the vortex step to 2 min to preserve DNA fragment length. A total of 1 ug DNA from each isolate was used as input for library preparation using the Oxford Nanopore ligation sequencing kit and native barcode expansion kits (Oxford Nanopore Technologies, Oxford Science Park, OX4 4DQ, UK). Libraries were pooled and sequenced on a MinION flow cell (Oxford Nanopore Technologies, Oxford Science Park, OX4 4DQ, UK). Raw reads were preprocessed using Filtlong v0.2.0[39] with parameters –min_length 1000 –keep-percent 95 –target_bases 650000000. Hybrid assemblies were created by assembling long read sequencing data in Flye v2.8.1[40] and polished with short reads from Illumina sequencing[41]. Assemblies were assessed for quality using Quast v3.2[36] and checkM v1.0.13[37]. Reads are available under BioProject PRJNA741123 (http://www.ncbi.nlm.nih.gov/bioproject/741123).

Genomic taxonomic identification

Following draft assembly, we determined taxonomic identification by ANI, MASH, and MLST. Species were determined if the genome had >75% aligned bases and >95% ANI with the type genome. Assembled genomes were considered to be the same genomospecies if they had >95% pairwise match but no >95% match with a type genome. We compared all assembled genomes against all assembled genomes and all type genomes using dnadiff[42]. If no species were identified, we used Mash to determine genera by comparing assembled genomes against all NCBI reference genomes[43]. After all phages were removed, genera were considered to be the same as the hit/hits with the highest identity. MLST was determined using mlst v2.4[44,45].

Phylogenetic analyses

To create core genome alignments, the gff files produced by Prokka were used as input in Roary[46]. Roary alignments were used to create an approximate maximum likelihood tree with FastTree[47]. Branch length precision was rounded to 0.0001 substitutions per site. The output newick files were visualized and annotated with isolate source data using ggtree (R)[48,49]. Roary pangenome sequences were further annotated using EggNOG v5.0[50].

Isolate groupings based on SNP pairwise distances

Snippy v4.4.3[51] was used to map forward and reverse reads for isolates to the type strain complete genome assembly and to call SNPs. To determine groups, we compared pairwise SNP distances between each isolate pairs of the same species. Isolates were grouped into perfectly reciprocal groups at every pairwise distance cutoff between isolates using igraph[52]. The SNP distance cutoff was set at the lowest SNP value where number of groups plateaued for many thousands of SNPs, indicating that the members of these groups are much more closely related to one another than other isolates.

Antibiotic-resistant gene identification and analyses

We identified acquired antibiotic resistance mutations against aminoglycosides, amphenicols, β-lactam, folate pathway inhibitors, fosfomycin, macrolides/lincosamides/streptogramins, quinolones, rifamycin, tetracycline, and vancomycin using ResFinder[53].

Bayesian phylogenetic analysis of molecular sequences using BEAST 2

Group 1 isolates were long-read sequenced and quality filtered as described above, and the core genome alignment was constructed as above. The core genome alignment was composed of 5964 core genes out of 6986 total genes, which we used as input genes for our time-measured phylogenetic analysis in BEAST v2.6.5[54]. The core genome alignment was converted to a Nexus file using MEGA X[55]. We used BEAUti v2.6.5 from the BEAST v2.6.5[54] software package to convert the Nexus file into a.xml file for input into BEAST. We chose to use the HKY site model because it allows for some flexibility in substitution rate for different types of substitutions, and catches most major biases[56]. We also used a strict clock model because our sequences are all from the same hospital within just over a year of each other, so we have no reason to suspect different substitution rates for different lineages[56]. Tip dates were determined as the number of days between each sample and the first sample collected. Model diagnostic information and parameter distribution were viewed using Tracer v1.7.2[57]. Individual trees were visualized using FigTree v1.4.4[58] and the consensus tree was visualized using DensiTree v2.2.7[59].

Statistics and reproducibility

Comparative statistics between old and new building samples were normalized by number of samplings. Generalized linear mixed models were used for significance testing, with Room and Week as random effects. In Fig. 2c, d, isolate frequencies were collapsed by Room and then averaged. Error bars indicate standard error. For all main text phylogenetic trees, branches with less than 80% bootstrap support were collapsed, and branches with 80–90% bootstrap support were labeled as such. Supplementary Figures containing phylogenetic trees (Supplementary Figs. 1c, d, 2, and 3a) have a minimum resolution of 0.00055.

Fig. 2

Variation in isolate collection location, identity, and timing across all sampling.

Error bars indicate standard error of intensive care unit (ICU) rooms. ** indicates generalized linear mixed-modeling (GLMM) p-value <0.01. a In-room and bathroom sink drains have significantly more isolates per collection than other surface locations in both the old and new ICU buildings (n = 566 surface isolates). Locations in light gray were not collected in old ICU. b Genus of matrix-assisted laser desorption/ionization-time of flight mass spectrometry (MALDI-TOF) species identification of all collected isolates in both the new and old ICU. Other Surface includes all other surfaces that are not in-room or bathroom sink drain. c Variation in number of isolates collected per bathroom or in-room sink drain sample collection by building (excludes fecal and communal samples, n = 429). d Variation in number of isolates per other surface sample collection by building (excludes sink drain, fecal, and communal samples, n = 137). e Variation in number of isolates per bathroom or in-room sink drain sample collection for all time points, n = 429. f Variation in number of isolates collected per other surface sample collection across all time points (excludes sink drain, fecal, and communal samples, n = 137). Gray bars indicate weeks with incomplete sampling of surfaces. BP before patient and staff move-in.