https://orcid.org/0000-0001-5974-9428
Figures
Abstract
The SARS-CoV-2 pandemic has resulted in considerable mortality in hospital settings. Built environment surveillance can provide a non-invasive indicator of SARS-CoV-2 status in hospitals, but we have a limited understanding of SARS-CoV-2’s microbial co-associations in the built environment, including any potential co-occurrence dynamics with pathogenic and antimicrobial-resistant microorganisms. Here we examine the microbial communities on floors and elevator buttons across several locations in two major tertiary-care Ontario hospitals during a surge in SARS-CoV-2 cases in 2020. Total microbial community composition, prevalence and type of detected antimicrobial resistance genes, and virulence factor distributions were governed by sample source rather than SARS-CoV-2 detection status. Fifteen microorganisms were identified as indicator species associated with positive SARS-CoV-2 signal, including three opportunistic pathogens (i.e., two Corynebacterium sp. and a Sutterella sp). Key clinically relevant antimicrobial resistance genes showed varying prevalence across sites within the hospital, suggesting that our workflow could inform resistance burden in hospitals. Overall, these results indicate limited or only weak interactions between microbiome composition and SARS-CoV-2 detection status in the hospital built environment.
Citation: George NA, Bradford L, Hinz A, El Kadi M, Xing L, Doukhanine E, et al. (2025) SARS-CoV-2 detection is independent of microbiome composition on surfaces in a major Ontario hospital. PLoS One 20(12): e0326403. https://doi.org/10.1371/journal.pone.0326403
Editor: Dhammika Leshan Wannigama, Yamagata University Faculty of Medicine: Yamagata Daigaku Igakubu Daigakuin Igakukei Kenkyuka, JAPAN
Received: May 29, 2025; Accepted: November 18, 2025; Published: December 4, 2025
Copyright: © 2025 George et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All human-subtracted read datasets are publicly available from the SRA database under Biosample accessions SAMN46426143 - SAMN46426180.
Funding: This work was supported by The Ottawa Hospital Academic Medical Organization (TOHAMO) to DRMcF and CN, Alliance Grant # 554478 - 20 from the Natural Sciences and Engineering Research Council (NSERC) to RK, funding from the University of Ottawa COVID-19 Reintegration Taskforce from RK, and a Carleton University Rapid Response Research Grant to AW. LAH was supported by a Tier II Canada Research Chair. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: ED works for DNA Genotek that provided sampling swabs in-kind for this study in an unrestricted fashion. DNA Genotek had no control over the findings, interpretations, or conclusions published in this paper. MF was a consultant for ProofDx, a start up company creating a point of care diagnostic test for COVID-19; is an advisor for SIGNAL1, a start-up company deploying machine learned models to improve inpatient care; and has been an expert witness on content unrelated to this work. He also holds a provisional patent for a model that predicts acute dialysis needs.
Introduction
Since its identification in December 2019, severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2, the etiological agent of COVID-19) has caused a global pandemic, with over 778 million cases and associations with over 7 million deaths (as of October 2025, World Health Organization, https://covid19.who.int). Human-to-human transmission through aerosolization of the virus is the primary route of transmission [1]. Outbreaks of COVID-19 remain common in both hospitals and long-term care homes worldwide, leaving both patients and healthcare workers at high risk [2]. Monitoring of viral detection and burden in the built environment can provide information on disease burden and infection risk [3,4].
Recent research has detected SARS-CoV-2 on various hospital surfaces, with detection on floors being most frequent [5–8]. Detection frequency has been strongly associated with infection burden within the hospital [9]. Instances of fomite transmission are limited [1], and so built environment screening is primarily used as a proxy for SARS-CoV-2 burden. Some work has indicated built environment screening can be extended to explore SARS-CoV-2 interactions with other components of the human microbiome. SARS-CoV-2 detection on fomites in hospitals has been predicted by bacterial signatures [10], suggesting that some bacteria, and possibly other microorganisms, may stably co-occur with SARS-CoV-2 transmission and propagation. Some bacteria and other microbes can metabolize and transform a variety of chemicals in their environment, including disinfectants [11,12], potentially neutralizing or attenuating the effects of these agents against viruses.
Surface sampling for SARS-CoV-2 has advanced our understanding of where the virus can be detected within a built environment, and has yielded new surveillance methods for hospital COVID-19 cases [6,9]. There have been limited examinations of microbial co-associations with SARS-CoV-2 in hospital environments to date. One large-scale 16S rRNA amplicon sequencing survey examining these co-associations across patients, health care providers, and the built environment identified that floor samples carried the highest microbial biomass [10]. This survey also identified a single amplicon sequence variant, from the genus Rothia, that was strongly associated with SARS-CoV-2 detection. Pochtovyi and colleagues surveyed a Moscow hospital, swabbing floors, door handles, and electronics in several wards and the intensive care unit (ICU) to determine prevalence of six pathogens identified as co-occuring with SARS-CoV-2 infections [13]. Targeted PCR identified at least one of these six pathogens of interest from each built environment sample. Within this study, 16S rRNA amplicon sequencing identified more diverse communities on floors compared to other surfaces, and identified several lineages that were characteristic for the different units (e.g., ICU versus Respiratory Infections Department) [13]. These studies indicate microbiome co-occurrence with SARS-CoV-2 can be detected from the built environment. However, amplicon sequencing cannot determine microbiome features with clinical relevance, such as antimicrobial resistance genes (ARGs) and virulence factors.
We wanted to explore the presence and prevalence of pathogenic indicators associated with the built environment microbiome in hospitals, and specifically connect those features to SARS-CoV-2 prevalence, to better understand microbiome interactions in these clinical spaces. We hypothesized that SARS-CoV-2 detection would correlate with specific microbial lineages across two regionally co-located hospitals, strengthening predictions of the conserved SARS-CoV-2-associated microbiome. We additionally expected to identify co-occurrence dynamics between SARS-CoV-2 and pathogen indicators, identifying key pathogens which may be able to opportunistically capitalize on SARS-CoV-2 outbreaks for further transmission, or which experience similar outbreak patterns. To assess these hypotheses, we conducted deep metagenomic sequencing on biomass collected from swabs of floors and elevator buttons to identify microbial community characteristics associated with SARS-CoV-2 detection and prevalence within a hospital environment. We further assessed the gene complement of antimicrobial resistance genes, viral prevalence, and virulence factors to determine prevalence and potential co-occurrence with SARS-CoV-2.
Materials and methods
Sample collection
Sample collection and SARS-CoV-2 detection was previously described [7]. In brief, samples were obtained over a 10-week period (September 28th, 2020 to December 6th, 2020) from two tertiary-care, academic hospital campuses from Ottawa, Ontario, Canada (Campus A and Campus B, totaling over 1100 beds), encompassing 8 different sites (ICU, elevators, parking garage, and both COVID and non-COVID wards). Surfaces were sampled with the P-208 Environmental Surface Collection Prototype kit (DNA Genotek, Ottawa, ON). This included swabbing a surface and then submerging the swab in a nucleic acid stabilization solution for subsequent storage at ambient temperature for up to one month. Before processing, control samples were generated through a 10−20 second exposure of a swab to the air of a location without any swabbing. The MagMAX Viral/Pathogen II (MVP II) Nucleic Acid Isolation Kit (Thermo Fisher Scientific, Waltham, MA) was used for total nucleic acid extraction. Extracted RNA was used as input to quantitative reverse-transcriptase polymerase chain reactions (RT-qPCRs) to detect SARS-CoV-2.
Samples for metagenomics were selected from this larger-scale study which specifically and exclusively examined the spatio-temporal dynamics of SARS-CoV-2 detection [7]. Samples were selected to allow a comparison of locations (wards, ICU, parking garage), site types (floors versus elevator buttons), and SARS-CoV-2 detection status as previously determined through RT-qPCR [7]. Of the 963 swabs, 63 were positive for SARS-CoV-2, 51 of which were from floors [7]. For floors, Ward 1 and the ICU from hospital B were selected as the SARS-CoV-2 positive samples, including their full time series (n = 15 total). Ward 4 from hospital B was selected as the SARS-CoV-2 negative ward, including its full time series of swabs (n = 10). There were only 12 positive elevator button swabs in total, limiting our ability to make geographic or replicate-based selections. We selected 6 positive samples from elevators in high-traffic areas and 6 negative samples from Ward 4, to allow comparison to the Ward 4 floor samples. Sample information is available in Supplemental Table 1.
Sequencing and quality control
Thirty-eight samples were submitted to the CAGEF DNA sequencing facility (Toronto, Canada). Libraries were constructed with the Illumina Nextera Flex kit and then converted into DNA nanoballs (DNBs) for sequencing on the MGI G400 using the MGI FCS PE150 sequencing kit. Adapter removal and quality filtering for sequenced reads was performed using Trimmomatic [14] and sickle [15].
Taxonomic assignment of reads
Assembly attempts were unsuccessful, yielding small numbers of short, low-coverage contigs, so we adopted a read-based approach. Quality-controlled reads were taxonomically assigned using Kraken2 [16] with the flags “--confidence” (set to 0.65) and “--report-zero-counts,” and the NCBI RefSeq Complete V205 500GB Kraken2 database [17], which is a substantially more complete database than the default distribution.
Bracken2 [18] was used to refine Kraken2’s estimates of reads assigned to each taxon and resolve potential misclassifications. Exploratory data analyses of Kraken2 and Bracken2 output were performed using PAVIAN [19].
Microbial community analyses
Microbial community analyses were conducted using the R Statistical Software v4.1.1 [20]. Data handling was accomplished using the phyloseq v.1.48.0 [21], tidyverse v.2.0.0 [22], and reshape2 v.1.4.4 [23] packages. Data was visualized using the dplyr v.1.1.4 [24] and ggplot2 v.3.5.1 [25] packages.
Taxonomic assignments for reads were imported from Bracken2 output files, and reads associated with the genus Homo were removed from all further analyses, in accordance with the ethics approval/REB associated with the original sampling [7]. Reads associated with the genus Homo were also removed from the raw read files deposited to the NCBI using the extract_kraken_reads.py script from the KrakenTools package [26]. The remaining Eukaryotic reads were very patchily distributed and primarily associated with plant matter (e.g., Brassica). All non-Homo Eukaryotic reads were subsequently removed from the datasets prior to microbial community composition analyses but were kept for data deposition to NCBI.
Microbial community compositions were compared using Principal Coordinates Analysis (PCoA) across SARS-CoV-2 detection status and location source (Elevator, Floor) using phyloseq’s ordinate function and a Bray-Curtis dissimilarity statistic [27]. The Shannon alpha diversity metric was calculated for all samples and compared across SARS-CoV-2 detection status and source [28]. Indicator species analyses were conducted using the package indicspecies v.1.7.15 [29] to identify key taxa specifically associated with SARS-CoV-2 positive or negative detection status.
Viral reads were subset and analyzed as a separate community, including PCoA, alpha diversity measures, and indicator species analysis as described for the full dataset above.
Virulence factor and antimicrobial resistance gene predictions
All reads were mapped to the core Comprehensive Antibiotic Resistance Database (CARD) v3.2.7 (downloaded June 12, 2023) [30] and core Virulence Factor Database (VFDB) (downloaded July 05, 2023) [31]. RGI v6.0.0 [30] was used for additional mapping to the core CARD. Predicted virulence factor and antimicrobial resistance gene prevalence and sample ordinations based on annotation categories were generated and visualized in R. Prevalence was normalized to reads per million to enable comparisons across samples.
Clinically relevant ARGs were selected (blaCTX-M, blaKPC, blaNDM-1, blaOXA, blaVIM, mecA, mecC, vanA, vanB), and annotations were manually harmonized to allow filtering of the datasets by this subset of ARGs. Kruskal-Wallis tests were used to determine if ARGs associated significantly across SARS-CoV-2 positivity status, sample source, or unit (ICU and Ward #) [32]. Pairwise comparisons between ARG abundance profiles were conducted using the Wilcoxon rank sum test with continuity correction [33], with a Benjamini-Hochberg adjustment to p-values to account for multiple tests [34].
Results and discussion
Metagenomes were generated for a total of 38 samples (of a total 963 swabs, 4.0%, Supplemental Table 1, [7]). Sample read counts ranged from 31,839,940–444,624,914 (median 104,749,332; control sample 3,902,008). Following taxonomic assignment of reads by Kraken2 and Bracken [16,18], all reads associated with the genus Homo were removed to meet study ethics requirements, removing 0.66–65.30% of the trimmed and quality-filtered reads (average 13.5%). Eukaryote-free datasets contained 1,909,955–39,486,112 reads classified to at least the domain level (median 7,344,826; control sample 590,963). The control sample was not expected to be aseptic, as it was an air swab rather than a sterile control.
Bacterial reads dominate Eukarya-free datasets
Samples varied in their bacterial, archaeal, and viral composition, with the Bacteria most abundant across all samples (Fig A in S1 Text). Samples were dominated by reads from four phyla: Actinomycetota (53.8%), Pseudomonadota (24.0%), Bacillota (16.7%), and Bacteroidota (4.4%) (Fig 1A). Reads from the other 34 detected phyla and reads unclassified at the phylum level comprised only 1.1% of the total reads (Fig 1A). Only the Cyanobacteria and Verrucomicrobiota were also present above 1% abundance in any individual sample, with Cyanobacteria above 1% in 10 and Verrucomicrobia above 1% in 1 sample(s) respectively. At the genus level, 21 genera had an abundance above 5% in at least one sample (Fig 1B), and 11 genera had an abundance above 1% across all samples. Notably, Cutibacterium dominated most samples, with other human-skin associated genera also strongly represented, including Pseudomonas, Corynebacterium, and Staphylococcus [35].
In A, all phyla present at or above 1% abundance in at least one sample are depicted as bars, with lower abundance phyla grouped into the “Others” category. In B, only genera present above 5% abundance in at least one sample (21 genera) are included. Bubbles are coloured by phylum and scaled by percent total abundance of the given sample. Sample names follow the convention UNIT_WEEK, as facets depict elevator vs. floor samples. Swabs positive for SARS-CoV-2 are colored red.
Viral communities showed higher variability across samples, with certain viruses identified sporadically at very high relative abundance (Fig 2). Phage orders with consistent presence across hospital samples included the Caudoviricetes (tailed viruses infecting bacteria and archaea [36]) and, at a lower abundance, the order Crassvirales (double-stranded DNA bacteriophages infecting bacteria in the phylum Bacteroidetes, and a recognized part of the human gut microbiome [37]) (Fig 2A). The order Zurhausenvirales (which includes the papillomaviridae) was the only human-associated viral order with consistent presence across samples (Fig 2A). The Zurhausenvirales may show this stable presence due to an equivalently stable level of infection and viral shedding across the human population in Ottawa (via warts), as papillomaviruses are not expected to cause acute illnesses requiring hospitalizations. Looking at specific viruses, three unclassified Caudoviricetes were consistently identified across samples, as was an uncultured human fecal virus (Fig 2B). Two papillomaviruses, one Betapapillomavirus and one human papillomavirus, were also consistently detected across samples, with the human papillomavirus showing an increased abundance in the Ward 6 elevator sample from week 2 (Fig 2B). These provide higher granularity to the trends observed at the order level.
Bubbles are scaled by normalized read count. In A, all viral orders identified in the data are included in the bubble plot. In B, only viruses identified at over 10,000 reads in at least one sample are depicted. Bubbles are coloured by viral order as in A. Facets represent samples that were negative or positive for SARS-CoV-2 detection in a previous study. Sample names follow the convention UNIT_WEEK_SOURCE, with positive SARS-CoV-2 samples colored red as in Fig 1.
Other viruses showed very high abundances in a single sample, including a Lymphocryptovirus humangamma4 (Epstein Barr Virus; Ward 1 Floor, week 2), a Streptococcus phage (ICU floor in week 6), and, at a lower abundance, the two papilloma viruses described above. It is notable that each of these specific viral spikes in abundance were identified from SARS-CoV-2 positive samples, though none of these were identified as indicator species for SARS-CoV-2 detection (see below). It seems likely these very high detection levels are related to specific patients within these locations shedding these viruses, but we did not have access to medical records to cross-check these detections against patient admissions. In the case of Epstein Barr and potentially the papilloma viruses, these detections may derive from a patient with active SARS-CoV-2 that has triggered a flare up of an older, latent infection under the immune assault of the SARS infection, a previously observed phenomenon [38].
The overall Shannon diversity was significantly lower in elevator samples relative to floor samples (Fig B and C in S1 Text). Shannon diversity across positive and negative SARS-CoV-2 detection status was not significantly different for either Floor or Elevator samples. For viral communities, there was no significant difference in Shannon diversity across Source or SARS-CoV-2 detection status (Fig D in S1 Text).
Sample composition clusters according to source and not SARS-CoV-2 detection status
Principal coordinates analysis (PCoA) on Eukaryote-free datasets showed grouping based on sample source (i.e., floor, elevator, or control), with limited overlap between floor and elevator samples (Fig 3A). SARS-CoV-2 detection status did not map to clear groups as defined by sample biodiversity. When PCoA groupings were examined by location (ICU, Wards, etc.), samples from the ICU grouped tightly, whereas samples from other sites were more widely spread out (Fig E in S1 Text). When only the viral subset of the datasets was used to generate the PCoA plot, a stronger degree of overlap was observed for floor and elevator samples (Fig 3B), with SARS-CoV-2 positive samples exhibiting more variation on the plot than SARS-CoV-2 negative samples, reflective of the sporadic detections of unique, high-abundance viruses in certain samples (Fig 2).
Points are coloured by source type (elevator, floor, and control (ctrl)), with SARS-CoV-2 detection status indicated by shape. Ellipses encompass SARS-CoV-2 positive (dashed) and negative (solid) samples within each site type.
Bacterial and viral indicator species based on SARS-CoV-2 detection status
While overall microbial community compositions were governed by sample source, multiple taxa showed strong associations with SARS-CoV-2 detection. Fifteen microorganisms showed positive association with SARS-CoV-2 detection, of which the most strongly correlated were a Sphingobacterium and a Lentilactobacillus species (Fig 4). Eight of the fifteen microorganisms are from the phylum Pseudomonadota. One archaeon, Halococcus hamelinensis, was identified as an indicator species for positive SARS-CoV-2 detection. No viruses showed positive correlation with SARS-CoV-2 detection. Unexpectedly, none of the 15 indicator species are obligate pathogens in humans and only a few are associated with the human microbiome. Three are opportunistic pathogens of humans: the two Corynebacterium sp. and the Sutterella sp. KLE1602. These three are also the only indicator species associated with the human microbiome, with Corynebacterium a common skin-associated lineage and Sutterella commonly found in the human gastrointestinal tract. Of the remaining twelve microorganisms, seven are typically associated with soil and sediment environments. We posit that the soil-associated lineages are not directly connected to SARS-CoV-2 prevalence, but rather serve as indicators for overall traffic in these areas and/or the cleanliness of a given surface (i.e., time elapsed between last cleaning and our swabbing). Under this model, we would expect common soil microbes to correlate with SARS-CoV-2 during times of high transmission/infection, as was true at the time of sampling. Two of the remaining five indicator species are psychrotolerant (Psychrobacter and Pontibacter arcticus) and one is halophilic and associated with hypersaline lakes (Halococcus hamelinensis). None of these would be expected to be identified from built environments impacted by human microbiomes, but none exhibit high abundance in the control sample, indicating they are a true signal of detection from our samples. Their prevalence and potential importance to predicting SARS-CoV-2 dynamics will require confirmation at other sites, or across a longer time frame. The Rickettsia identified is an endosymbiont of the black-legged tick, which could have medical associations in a hospital environment, but which we were not able to confirm given an absence of patient data. Lentilactobacillus parabuchneri is commonly found in dairy products, and thus its presence may be due to patient, staff, or visitor foodstuffs, which may have been ward specific. Lentilactobacillus parabuchneri exhibited more universal detection across all samples compared to the other organisms, just with higher abundance in SARS-CoV-2 positive samples.
Facets represent samples that were negative or positive for SARS-CoV-2 detection in a previous study [7]. Bubbles are scaled by abundance as normalized read counts, and coloured by phylum. Organisms are ranked by significance, with their individual p-value (p < 0.01) from the indicator species analysis noted after each organism’s name in brackets. Indicator species for negative SARS-CoV-2 detection status are presented in Fig F in S1 Text.
Thirty-six organisms, including one virus, were identified as indicator species associated with negative SARS-CoV-2 detection, demonstrating a negative correlation with SARS-CoV-2 detection (Fig F in S1 Text). The organisms most strongly negatively correlated with SARS-CoV-2 detection were Microbacterium liquefaciens, Salicibibacter halophilus, and Tsukamurella tyrosinosolvens (Fig F in S1 Text). We note that no members of the genus Rothia were identified as indicator species in this analysis, contrasting findings from an earlier survey of microbial co-associations with SARS-CoV-2 in a U.S. hospital [10]. This may be due to geographic differences in hospital microbiomes, or differences between study methodologies and analysis tools. Sixteen of thirty-six organisms are from the phylum Pseudomonadota, while ten are members of the Actinomycetota.
Virulence factor profiles
The distribution of virulence factors within the VFDB core dataset was examined across samples. A high proportional abundance of virulence factors was observed in a single elevator sample (Ward 3, week 2 (W3_2_ELE), Fig G in S1 Text). PCoA analysis revealed that, akin to microbial diversity, sample source – rather than SARS-CoV-2 detection status – governed the observed virulence factor profile distributions. Consistent with its higher overall virulence factor signal, Ward 3’s week 2 elevator sample had the most distinct virulence factor profile relative to other samples (Fig H in S1 Text, upper right).
Antimicrobial resistance genes
Antimicrobial resistance genes (ARGs) were detected by mapping reads to the Comprehensive Antimicrobial Resistance Database (CARD). In this analysis only, the control sample demonstrated an unexpectedly high proportional abundance of ARGs (Fig IA in S1 Text). The control sample had a much lower overall read count, which may have led to an over-inflation of predicted ARGs when normalized to counts per million reads. The control sample also showed a distinctly different microbial community composition across all metrics, and as such this enrichment in ARGs, if not an analysis artifact, is hypothesized to be due to contamination from DNA extraction and sequencing processing kits, where antimicrobial-resistant organisms were likely used to generate the relevant enzymes. None of the ARGs identified as clinically relevant for further analyses (see below) were detected in the control sample. We removed ARGs associated with the control sample from all samples’ predicted ARGs - resulting in a removal of 121 of the 1,830 identified ARGs (6.6%; Supplemental Figure 9B). Two elevator samples, Ward 3 week 2 and Ward 6 week 2, had the greatest proportional abundance of ARGs. PCoA analysis showed floor samples and elevator samples had distinct ARG profiles; however, floor samples contained higher intra-group similarity in ARG profiles relative to elevator samples, and elevator samples positive for SARS-CoV-2 detection specifically drove this intra-group variability (Fig J in S1 Text).
ARGs were categorized by CARD Resistance Mechanisms, which showed similar profiles across Source and SARS-CoV-2 detection variables. Antibiotic efflux was the most abundant resistance mechanism, with antibiotic inactivation and antibiotic target alteration also strongly represented (Fig K in S1 Text). By Drug Class, resistance to tetracycline, aminoglycoside, mupirocin-like, and peptide antibiotics showed elevated prevalence. The most abundant type of antibiotic resistance was multidrug resistances, where multiple antibiotic classes were targeted by the same resistance mechanism (Fig L in S1 Text). Between floors and elevators, there were differences in prevalence of certain antibiotics: rifamycin and glycopeptide resistance were elevated in floor samples, whereas fluoroquinolone antibiotic resistance was at a higher prevalence in elevator samples (Fig L in S1 Text). Given the importance of disinfection within the hospital built environment, we were interested in the prevalence of resistance to disinfecting agents, which was a stand-alone Drug Class as well as being included in several multidrug resistance Drug Classes. Resistance to disinfection agents represented 8.1% of ARG-associated reads, with an average prevalence of 19.35 reads per million reads (Table B in S1 Text). Elevator buttons carried a slightly higher prevalence than floors, but there were no significant differences by Source (Elevators: mean 20.8 + /- 17.8 rpm; Floors: mean 18.7 + /- 5.6 rpm; p = 0.69). Similarly, there was no significant difference between SARS-CoV-2 positive and negative samples (Positive: 20.3 + /- 14.4 rpm; Negative: mean 18.0 + /- 4.2 rpm; p = 0.48). Higher variance in the Elevators and SARS-CoV-2 positive statistics were driven by the Ward 3 week 2 elevator sample’s higher ARG prevalence.
We next examined the distribution of specific clinically relevant ARGs known to confer resistance to beta-lactams, carbapenems, and glycopeptides: blaCTX-M, blaKPC, blaNDM-1, blaOXA, blaVIM, mecA, mecC, vanA, and vanB. The relationship between ARG abundance and Unit was not significant for these ARGs of interest (Table C in S1 Text). A Kruskal-Wallis test examining ARG abundance across SARS-CoV-2 detection status was not significant (p = 0.31). Comparison of the selected ARGs’ abundances across sample source (elevator button vs. floor) was significant (p < 0.0092), driven by vanA showing a significantly higher prevalence on floors compared to elevator (p < 1.74 x10-5; Fig M and Table C in S1 Text). This indicates ARG burden is not correlated with SARS-CoV-2 incidence across Units, and that floors likely remain the most effective sampling location for hospital-based built environment surveillance.
A Kruskal-Wallis test based on ARG abundance across samples identified a significant difference (p < 2.2x10-16) by abundance of these ARGs. A pairwise Wilcoxon rank sum test using a Benjamini-Hochberg correction for multiple tests identified four distinct groups based on ARG abundance profiles: blaNDM-1, blaVIM, mecC, and vanB (Group 1, very low abundance); blaCTX-M and blaKPC (Group 2, low abundance); vanA alone (Group 3, sporadic high abundance), and blaOXA with mecA (Group 4, high abundance) (Fig 5, Table D in S1 Text).
A: ARGs of interest abundance across hospital Units, organized by groups identified through a pairwise Wilcoxon rank sum test: Group 1 – low abundance; Group 2 – very low abundance; Group 3 – sporadic high abundance; and Group 4 – high abundance. Box plots indicate the first and third quartile of results with the thick horizontal bar indicating the median. Whiskers identify the range of the data, with outlier data points plotted. B: Bubble plot showing ARGs of interest abundance across samples. Sample names follow the convention UNIT_WEEK, as facets depict elevator vs. floor samples. Swabs positive for SARS-CoV-2 are colored red.
Among the clinically relevant ARGs of interest, the most abundant were blaOXA, mecA, and vanA. blaOXA and mecA were identified at all seven Units, while vanA was specifically detected in the ICU, Ward 1, and Ward 4. The three higher-abundance ARGs profiles changed in abundance across Units and across weeks (Fig 5), indicating this built-environment monitoring protocol does show variation in the ARG burden on hospital surfaces. Additional work is required to determine if this protocol provides insight into the ARG burden within the patients on these wards, which would extend the utility of built environment surveillance.
Conclusions
Metagenomic sequencing of swabs from hospital floors and elevator buttons identified a diverse microbial community. Bacterial and archaeal community membership was more stable across samples than the viral community, which was defined by sporadic viruses with elevated abundances. These included viruses with microbial hosts as well as human viruses with latent infection stages that may have flared during SARS-CoV-2 infections (e.g., papilloma viruses and the Epstein Barr virus). Community diversity was more strongly defined by sample source (floors vs. elevator buttons) than the underlying SARS-CoV-2 detection status (positive or negative) across all variables examined. This is despite a high case load of SARS-CoV-2 patients during the window of sampling, and SARS-CoV-2 detection signal correlating with the underlying patient load within a Unit [7].
We hypothesized that SARS-CoV-2 detection would correlate with specific microbial lineages. Indicator species analysis based on SARS-CoV-2 detection identified a handful of strongly correlated microbes, including seven soil-associated lineages and three human-microbiome-associated opportunistic pathogens. Given most of the identified organisms do not have expected roles in pathogenicity or the human microbiome (12 of 15), this analysis serves as a prediction of potential key lineages that will need mechanistic explanations and confirmation of interaction with SARS-CoV-2 from additional studies. Only one virus was identified as an indicator from this analysis, an unclassified virus from an unclassified phylum, that makes is impossible to infer the host of this virus. These data suggest that, while the overall microbial community is defined by the environment and other external forces, some organisms are stably coincident with SARS-CoV-2. This was, overall, a weaker signal for microbiome/SARS-CoV-2 co-occurrence than we had expected to identify, but is in line with findings from other hospital surveys that found very few, or weak signals for co-associations of the microbiome [10,13]. Notably, we did not identify any of the same lineages as co-occurring with SARS-CoV-2 as previous studies [10,13], suggesting a general lack of universal co-association between SARS-CoV-2 and the microbiome deposited on surfaces. This has important implications for future surveillance efforts.
We also expected to identify cooccurrence dynamics between SARS-CoV-2 and pathogen indicators, which was similarly inconclusive. Antimicrobial resistance patterns were similarly governed by sample source rather than SARS-CoV-2 detection status, with higher prevalence of resistance mechanisms targeting tetracycline, aminoglycoside, and fluoroquinolone antibiotics, with a strong signal for resistance to multiple drug classes. We did not have access to antibiotic use statistics for the hospitals during the swabbing time period, but note that these three antibiotic classes cover four of the top ten prescribed antibiotics in Canadian hospitals between 2019–2023 [39], making their use within this environment highly likely.
Key ARGs of interest were identified based on clinical relevance, which showed four broad categories of abundance, from consistently very low or low detection levels to consistently high. One ARG, vanA, showed higher variability in its abundance across Units and a statistically significant difference between its detection on floors versus elevator buttons. vanA is part of a gene cluster that is responsible for high-level vancomycin resistance in Enterococcus species (such as Enterococcus faecium and Enterococcus faecalis). Vancomycin resistant Enterococcus (VRE) are considered environmentally entrenched in hospitals – the presence of vanA was expected, but its variable prevalence across units and time (Fig 5) indicates built environment swabbing may allow identification of target areas to reduce VRE in hospitals [40].
This study identified key microbial community members associated with SARS-CoV-2 detection on hospital built environment surfaces and demonstrated the ability to detect both punctuated viral signatures and varying ARG prevalence. Future work will determine the correlations between this detection capacity and the associated disease and/or resistance burden within the hospital to explore the potential of built environmental surveillance as an additional mechanism for monitoring outbreaks and other health risks developing in a unit or over time.
An important limitation to our study is that we lacked human-level clinical data. For example, in the location where we found a high abundance of herpesvirales we are not able to report whether there was a patient hospitalized with a related infection at that point in time – this disconnect does limit the conclusions we are able to derive from this survey. Future studies that include not only built environment data but also clinical data are needed to better contextualize and understand the associations we observed in our study.
Supporting information
S1 Text. PDF file containing supplemental tables A-D and supplemental Figures A-M.
https://doi.org/10.1371/journal.pone.0326403.s001
(PDF)
Acknowledgments
We thank Dr. Morgan Langille’s lab for providing us advanced access to the NCBI RefSeq Complete V205 500GB Kraken2 database.
References
- 1. Meister TL, Dreismeier M, Blanco EV, Brüggemann Y, Heinen N, Kampf G, et al. Low Risk of Severe Acute Respiratory Syndrome Coronavirus 2 Transmission by Fomites: A Clinical Observational Study in Highly Infectious Coronavirus Disease 2019 Patients. J Infect Dis. 2022;226(9):1608–15. pmid:35512326
- 2. Abbas M, Robalo Nunes T, Martischang R, Zingg W, Iten A, Pittet D, et al. Nosocomial transmission and outbreaks of coronavirus disease 2019: the need to protect both patients and healthcare workers. Antimicrob Resist Infect Control. 2021;10(1):7. pmid:33407833
- 3. Siddiqui H, Hicks AMA, Hinz A, Ray P, Johnstone J, MacFadden DR, et al. Environmental surveillance of SARS-CoV-2 for outbreak detection in hospital: A single centre prospective study. Journal of Clinical Virology Plus. 2025;5(1):100199.
- 4. Ray P, Lim B, Zorcic K, Johnstone J, Hinz A, Hicks AMA, et al. Environmental surveillance of SARS-CoV-2 for COVID-19 outbreak detection in hospital: a single-centre prospective study. Infect Prev Pract. 2024;6(2):100353. pmid:38516392
- 5. Fralick M, Burella M, Hinz A, Mejbel HS, Guttman DS, Xing L, et al. The spatial and temporal distribution of SARS-CoV-2 from the built environment of COVID-19 patient rooms: A multicentre prospective study. PLoS One. 2023;18(3):e0282489. pmid:36913370
- 6. Fralick M, Nott C, Moggridge J, Castellani L, Raudanskis A, Guttman DS, et al. Detection of Covid-19 Outbreaks Using Built Environment Testing for SARS-CoV-2. NEJM Evid. 2023;2(3):EVIDoa2200203. pmid:38320044
- 7. Hinz A, Xing L, Doukhanine E, Hug LA, Kassen R, Ormeci B, et al. SARS-CoV-2 detection from the built environment and wastewater and its use for hospital surveillance. FACETS. 2022;7:82–97.
- 8. Guo Z-D, Wang Z-Y, Zhang S-F, Li X, Li L, Li C, et al. Aerosol and Surface Distribution of Severe Acute Respiratory Syndrome Coronavirus 2 in Hospital Wards, Wuhan, China, 2020. Emerg Infect Dis. 2020;26(7):1583–91. pmid:32275497
- 9. Fralick M, Moggridge JA, Wiebe M, Castellani L, McGeer A, Feenstra B, et al. SARS-CoV-2 burden on the floor was associated with COVID-19 cases and outbreaks in two acute care hospitals: a prospective cohort study. Infect Control Hosp Epidemiol. 2024;45(11):1–5. pmid:39301606
- 10. Marotz C, Belda-Ferre P, Ali F, Das P, Huang S, Cantrell K, et al. SARS-CoV-2 detection status associates with bacterial community composition in patients and the hospital environment. Microbiome. 2021;9(1):132. pmid:34103074
- 11. Belter B, McCarlie SJ, Boucher-van Jaarsveld CE, Bragg RR. Investigation into the Metabolism of Quaternary Ammonium Compound Disinfectants by Bacteria. Microb Drug Resist. 2022;28(8):841–8. pmid:35759372
- 12. Oh S, Kurt Z, Tsementzi D, Weigand MR, Kim M, Hatt JK, et al. Microbial community degradation of widely used quaternary ammonium disinfectants. Appl Environ Microbiol. 2014;80(19):5892–900. pmid:24951783
- 13. Pochtovyi AA, Vasina DV, Kustova DD, Divisenko EV, Kuznetsova NA, Burgasova OA, et al. Contamination of Hospital Surfaces with Bacterial Pathogens under the Current COVID-19 Outbreak. Int J Environ Res Public Health. 2021;18(17):9042. pmid:34501634
- 14. Bolger AM, Lohse M, Usadel B. Trimmomatic: a flexible trimmer for Illumina sequence data. Bioinformatics. 2014;30(15):2114–20. pmid:24695404
- 15. Joshi N, Fass J. Sickle: A sliding-window, adaptive, quality-based trimming tool for FastQ files (Version 1.33) [Software]. 2011. p. 2011. https://github.com/najoshi/sickle
- 16. Wood DE, Lu J, Langmead B. Improved metagenomic analysis with Kraken 2. Genome Biol. 2019;20(1):257. pmid:31779668
- 17. Wright RJ, Comeau AM, Langille MGI. From defaults to databases: parameter and database choice dramatically impact the performance of metagenomic taxonomic classification tools. Microb Genom. 2023;9(3):000949. pmid:36867161
- 18. Lu J, Breitwieser FP, Thielen P, Salzberg SL. Bracken: estimating species abundance in metagenomics data. PeerJ Computer Science. 2017;3:e104.
- 19. Breitwieser FP, Salzberg SL. Pavian: interactive analysis of metagenomics data for microbiome studies and pathogen identification. Bioinformatics. 2020;36(4):1303–4. pmid:31553437
- 20.
R Core T. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2024.
- 21. McMurdie PJ, Holmes S. phyloseq: an R package for reproducible interactive analysis and graphics of microbiome census data. PLoS One. 2013;8(4):e61217. pmid:23630581
- 22. Wickham H, Averick M, Bryan J, Chang W, McGowan L, François R, et al. Welcome to the Tidyverse. JOSS. 2019;4(43):1686.
- 23. Wickham H. Reshaping Data with thereshapePackage. J Stat Soft. 2007;21(12).
- 24.
Wickham H, François R, Henry L, Müller K, Vaughan D. dplyr: A grammar of data manipulation. 2023.
- 25.
Wickham H. ggplot2: Elegant graphics for data analysis. New York: Springer-Verlag; 2016.
- 26. Lu J, Rincon N, Wood DE, Breitwieser FP, Pockrandt C, Langmead B, et al. Metagenome analysis using the Kraken software suite. Nat Protoc. 2022;17(12):2815–39. pmid:36171387
- 27. Bray JR, Curtis JT. An Ordination of the Upland Forest Communities of Southern Wisconsin. Ecological Monographs. 1957;27(4):325–49.
- 28. Shannon CE. A Mathematical Theory of Communication. Bell System Technical Journal. 1948;27(3):379–423.
- 29. De Cáceres M, Legendre P. Associations between species and groups of sites: indices and statistical inference. Ecology. 2009;90(12):3566–74. pmid:20120823
- 30. Alcock BP, Huynh W, Chalil R, Smith KW, Raphenya AR, Wlodarski MA, et al. CARD 2023: expanded curation, support for machine learning, and resistome prediction at the Comprehensive Antibiotic Resistance Database. Nucleic Acids Research. 2023;51: D690–D699.
- 31. Chen L, Yang J, Yu J, Yao Z, Sun L, Shen Y, et al. VFDB: a reference database for bacterial virulence factors. Nucleic Acids Res. 2005;33(Database issue):D325-8. pmid:15608208
- 32. McKight PE, Najab J. Kruskal‐Wallis Test. The Corsini Encyclopedia of Psychology. Wiley; 2010. 1–1.
- 33. Haynes W. Wilcoxon Rank Sum Test. Encyclopedia of Systems Biology. Springer New York; 2013. p. 2354–5.
- 34. Haynes W. Benjamini–Hochberg Method. Encyclopedia of Systems Biology. Springer New York; 2013. p. 78–78.
- 35. Byrd AL, Belkaid Y, Segre JA. The human skin microbiome. Nat Rev Microbiol. 2018;16(3):143–55. pmid:29332945
- 36. Zhu Y, Shang J, Peng C, Sun Y. Phage family classification under Caudoviricetes: A review of current tools using the latest ICTV classification framework. Front Microbiol. 2022;13:1032186. pmid:36590402
- 37. Smith L, Goldobina E, Govi B, Shkoporov AN. Bacteriophages of the Order Crassvirales: What Do We Currently Know about This Keystone Component of the Human Gut Virome?. Biomolecules. 2023;13(4):584. pmid:37189332
- 38. Bernal KDE, Whitehurst CB. Incidence of Epstein-Barr virus reactivation is elevated in COVID-19 patients. Virus Res. 2023;334:199157. pmid:37364815
- 39.
Canadian Antimicrobial Resistance Surveillance System. National antimicrobial prescribing survey. Canadian Antimicrobial Resistance Surveillance System; 2024. https://health-infobase.canada.ca/carss/amu/results.html?ind=02
- 40. Tan C, Linkenheld-Struk A, Williams V, Kozak R, Dhabaan G, Maze Dit Mieusement L, et al. The role of the environment in transmission of vancomycin-resistant Enterococcus: A proof-of-concept study. Antimicrob Steward Healthc Epidemiol. 2022;2(1):e178. pmid:36386008