Airborne antibiotic resistome and human health risk in railway stations during COVID-19 pandemic

Graphical abstract


Introduction
The ongoing global pandemic of coronavirus disease 2019 (COVID-19) poses a serious threat to public health (Huang et al., 2020;Platto et al., 2020). Long before the COVID-19 epidemic, antibiotic resistance was a global public health concern, and it remains so (Wang et al., 2022). Particularly, various antibiotics were used in pandemic disease treatments (Miranda et al., 2020). Antibiotics were administered to 72 % of hospitalized COVID-19 patients, and broad-spectrum antibiotics were prescribed at a high rate in reports to date (Rawson et al., 2020). For control of the pandemic, disinfectants are widely applied in public areas around cities since aerosol-based transmission is thought to be an important mode for COVID-19 MacIntyre and Wang 2020). Therefore, airborne-resistant bacteria and pathogens in public areas such as railway stations could be a great concern.
Public transit systems are one of the most common infrastructures, in which railway stations serve as crucial hubs for connecting various cities. According to a study on public transit systems around the world, airborne antibiotic resistance genes (ARGs) are common in cities (Danko et al., 2021). Another worldwide study has revealed that public transit airborne ARGs potentially source from commuters and the adjacent environment (Leung et al., 2021). Travelers bring their symbiotic microorganisms with them and exchange microorganisms within the public transit environment (Danko et al., 2021). High pedestrian traffic in a typical public transportation scenario can present public health problems since it facilitates the airborne transmission of microorganisms among commuters (Fujiyoshi et al., 2017). Therefore, during the COVID-19 pandemic, each country has taken travel-related control measures (Zhang and Tong 2021). China is a country with a large population and figures from 2017 show that the annual passenger traffic on Chinese railways can reach 3.039 billion . To control the spread of the virus, disinfectants are used in China in both indoor and outdoor spaces (Sills et al., 2020). Guangzhou is a cosmopolitan city of about 13 million people and it is one of the most densely populated cities in China, with nearly 1 million passengers per day at the railway stations (Lin et al., 2017). According to China Railway Guangzhou Group, the railway station staff has taken strict disinfection measures, disinfecting the air in public areas where passengers gather twice times daily and disinfecting the physical surfaces four times daily Zhao et al., 2022). Disinfectants might play an important role in promoting antibiotic-resistant bacteria (ARB) and ARGs (Li and Gu 2019). Thus, we are concerned that these measures may potentially promote the growth of airborne ARGs during the COVID-19 pandemic.
The objective of this study was to characterize the changes of airborne antibiotic resistome and microbiome from a metagenomic viewpoint during the COVID-19 pandemic and then examine the health risks associated with this. In this study, aerosol samples containing airborne bacteria and dust were collected from the three railway stations and corresponding outdoor areas in Guangzhou city, South China. Metagenomic sequencing (27 metagenomes) and culturing technique (164 cultured ARB flora) were employed to characterize airborne antibiotic resistome and microbiome from all air samples. Airborne ARGs from the railway stations were also assessed for possible sources by Bayesian source tracking. To the best of our knowledge, this study represents the first extensive airborne metagenomic study of public transit systems during the COVID-19 pandemic. The results from this study could raise awareness of antibiotic resistance in the public environment during the COVID-19 pandemic.

Railway Station
Air samples were collected from the three big railway stations in Guangzhou city including Central Railway Station, East Railway Station, and South Railway Station. The Central Railway Station is located in Yuexiu district, East Railway Station in Tianhe district and South Railway Station in Panyu district. Among them, the South Railway Station is a high-speed Railway Station. According to official statistics for 2021, Guangzhou's railway passenger stations sent 89.63 million people, with an average daily passenger throughput of 245,000. There are approximately 174,000, 35,000, and 30,000 passengers each day at the Guangzhou South, Central, and East Stations, respectively. Sampling campaign was conducted from late December 2019 to early January 2020 (before the pandemic outbreak), May 2020 (pandemic 1), and January 2021 (pandemic 2). All indoor samples were collected in waiting rooms, whereas outdoor samples were collected in station plazas.

Airborne bacteria collection
Airborne bacteria samples were collected at each sampling point using a portable sampler (HighBioTrap, Beijing dBlue Tech, Inc., Beijing, China) at a flow rate was 1000 L/min (Zhang et al., 2019b). At each railway station, samples were collected from two representative public areas, including station plaza (distributing different transport flows, outdoor area) and waiting room (public enclosed spaces, indoor area). At each public area, three air samplers were used to collect airborne bacteria at human respiration level (1.5 m above the ground) in Nutrient Agar (NA) plates for 1 min. Three NA plates were used for sampling in each air sampler then colonies from which were pooled into one after subsequent incubation. After sampling, the plates were incubated for 24 h at 37 • C. The colonies were scraped from agar plates to 1 ml 20 % glycerine broth, and then the rest of the colonies were washed off the plates with 2 ml sterile phosphate-buffered saline (PBS) three times. The bacteria in PBS were harvested by using a centrifuge for 7 min at 7000 rpm, the pellet was transferred to 0.5 ml 20 % glycerine broth and combined with the scraped bacteria (Bai et al., 2022).

Dust collection
Since not all bacteria could be cultured, dust samples were collected to analyze the overall situation. In order to ensure that it is the dust that has accumulated from the deposition of suspended air, only dust that has settled on the surface of objects (e.g. window sills, billboards, etc.) is collected, and three parallel samples are collected randomly in the waiting room at each station during each sampling period. Dust samples were collected using clean sterile brushes and collected in sterile stool collectors from physical surfaces in waiting rooms. A 0.2 g sample of dust was used to extract each DNA.

Dust bacteria collection
Soil suspensions were made by vortexing 0.1 g of dust sample in 2 ml of 1 × PBS, vortex mixing for 5 min, and left to stand for 30 min. Fifty microliters of supernatant were then plated on NA plates with 16 µg ml − 1 Amphotericin B (which inhibits fungal growth) were incubated for 24 h at 37 • C. Culturable bacteria were collected as described above. Sample information and sample processing were presented in Fig. 1a.

Metagenomic sequencing and bioinformatic analysis
Total DNA of airborne culturable bacteria and dust samples was extracted using the DNeasy® PowerSoil® kit (QIAGEN, Germany) according to the manufacturer's instructions. The DNA sample was stored at − 80 • C for further analysis. All DNA samples were shipped on dry ice to the Novogene Bioinformatics Technology Co. ltd. for library preparation and paired-end 150-bp shotgun sequencing. The raw sequences were quality filtered using Kneaddata (v 0.7.4) with default parameters, using the human genome hg38 as a reference to removing human DAN sequences (https://github.com/biobakery/kneaddata). Kraken2 (v 2.0.8-beta) was used to provide taxonomic labels to metagenomic DNA sequences by referring to the k-mer index generated from GTDB complete genome database (v 1.7.0) (Wood and Salzberg 2014). Rapid characterizations of ARGs were performed with ARGs-OAP (v 2.3) by aligning with the structured database SARG (v 2.0) at the read-level (Yin et al., 2018).
SourceTracker compares the ARG profiles in the "sources" with the "sinks", using a Bayesian approach to determine the extent to which each source contributes to the sinks (Knights et al., 2011). A total of 61 metagenomic datasets were used as potential resistance group sources for ARG genes (Table S1). These source samples were selected to cover different ecological types, including human gut, skin and oral animal feces, soil, hospital sewage, and estuary in China. Quality filtering and resistome profiles (SARG) of the source datasets were processed using the same method described above. This was performed to take into account any potential geographically based heterogeneity in source resistomes. Metacompare computation pipeline was used to calculate the antibiotic resistome risk, which estimates the potential for ARGs to be associated with MGE genes and mobilize to potential human pathogens based on metagenomic data (Oh et al., 2018).
With the global expansion of microbiome sequencing, a global perspective is to link new microbiome samples to the existing microbiome sample space. The Microbiome Search Engine (MSE) enables fast search queries for the best match of microbiome samples by taxonomic similarity (Su et al., 2018). MIP (Microbial Index of Pathogenic bacteria) was calculated using Parallel-MeTA3 (https://github.com/qdu-bioi nfo/mip). The MNS (microbiome novelty score) was calculated by submitting the classification and abundance results of Parallel-META3 to the microbiome search engine webtool (Jing et al., 2017;Su et al., 2018).
The raw reads were deposited into the NCBI Sequence Read Archive (SRA) database (Accession Number: SRR19742207-SRR19742202).
The total DNA of antibiotic-resistant bacteria was extracted using the boiling method (Yamagishi et al., 2016). 500 µl of cells was resuspended twice in 500 µl of 1 × TE buffer and centrifuged at 10,000 g for 3 min to transfer the pellet to 100 µl of new 1 × TE buffer. Suspended samples were incubated at 100 • C for 10 min and immediately cooled on ice for 5 min. To characterize the taxonomic profile of the antibiotic-resistant bacteria, the V3-V4 region of the 16S rRNA gene was amplified with primer pairs 338F and 806R, 6-bp barcode unique to each sample was included in double-ended primers (Xiao et al., 2021). The following polymerase chain reaction (PCR) conditions were used: 95 • C for 5 min, followed by 28 cycles of 95 • C for 30 s, 55 • C for 30 s, and 72 • C for 30 s, and, finally, 72 • C for 5 min. Normalized equimolar concentrations of the PCR products were then pooled and sequenced using the Illumina Novaseq 6000 platform at Novogene Bioinformatics Technology. QIIME 2 (version 2020.11.0) pipeline was used for quality control of the sequencing data, including sequencing separation (Cutadapt) (Martin 2011). and denoising (DADA2) (Bolyen et al., 2019). Taxonomy was assigned against the SILVA database (Release 138). The raw reads were deposited into the NCBI Sequence Read Archive (SRA) database (Accession Number: SRR20828466-SRR19742202).

Statistical analysis
Data analysis was performed using Microsoft Excel (Version16.42), SPSS (Version 26.0), GraphPad Prism (Version 8.4.0), and R software (Version 4.1.3). The network visualization analysis was done by Gephi version 0.9.2 and Cytoscape version 3.7.1. Visualization of the phylogenetic tree completed by Interactive Tree Of Life (https://itol. embl. de). Visualization of gene structure comparisons was generated with EasyFigure (Version 2.2.2) (Sullivan et al., 2011). The distribution of strains enriched for each antibiotic was visualized using the 'HiveR' R package (https://github.com/bryanhanson/HiveR). The schematic diagrams or elements in the graphic abstract were drawn using BioRender (https://app.biorender.com/) with full publishing rights.

The overall profile of airborne antibiotic resistome
Metagenomic sequencing was performed to reveal airborne ARGs inside and outside the railway stations ( Fig. 1 and Figure S1). The flowchart of the analytical process for this study is shown in Fig. 1a, and detailed profiles of airborne antibiotic resistome and microbiomes are given in Fg. S1a, b. There were a total of 23 different ARG types among the airborne samples discovered at the railway stations. The proportion of ARGs in each sample showed that multidrug and macrolidelincosamide-streptogramin (MLS) were the two main classes of ARGs (Fg. S1a). Firmicutes, Actinobacteria, and Bacteroidetes were the most abundant phyla in the airborne cultivable bacteria (>56 %). The two most prevalent phyla in the dust samples included Proteobacteria, and Actinobacteria, representing>67.1 % of the bacterial community (Fg. S1b). The relative abundance of airborne ARGs increased slightly, but not significantly following the pandemic outbreak (Fig. 1b). However, Shannon diversity significantly declined after the outbreak (p < 0.05) (Fig. 1b). To explore the overall changes in airborne resistome at railway stations before and after the outbreak, we analysed ARGs that were shared between three railway stations during the same time period (core ARGs). The relative abundance of the core ARGs in airborne bacteria increased during the pandemic (p < 0.05). By counting core, sub-core (detected in at least two sites), and peripheral (detected in only one site) ARGs, we found a gradual convergence of ARGs subtypes in all air samples (airborne bacteria and dust) from different railway stations after control measures (Fg. S1c). Additionally, the core ARGs consisted of 16 types (ARGs-type detected in all datasets) and multidrug and MLS were still the major components of ARG types (Fig. 1c).

Distribution of Antibiotic-Resistant bacteria and horizontally acquired ARGs
Seven antibiotics were used to identify antibiotic resistance bacteria (ARB) of all airborne samples (including 27 samples), and 16S rRNA sequencing was used to analyze the microbiome. Overall, 89 % of medium (164 of 189) had a growth of ARB. Over 41 % (11 of 27) of them developed ARB on all seven-antibiotic media, and 9 of these 11 samples were collected during the pandemic (Fig. 2a, Fg. S2a). At the phylum level, Actinobacteria, Firmicutes, and Proteobacteria were the dominant ARB in the air environment of railway stations (Fig. 2a). At the genus level, Acinetobacter, Paenibacillus, Staphylococcus, Pseudomonas, Lysobacter, and Microbacterium were found to be the top six ARB in relative abundance (Fig. 2b). In addition, there were no significant differences in Shannon index by ARB from different antibiotic-selective culture media (Fg. S2b). We used the Microbial Index of Pathogenic Bacteria (MIP) to describe potential disease risks as well as susceptible human organs (Sun et al., 2022). The MIP values increased following the outbreak of the COVID-19 pandemic (Fg. S2c). Notably, the ARB enriched by ciprofloxacin (CIP), kanamycin (KAN), and meropenem (MEM) had a lower potential disease risk than those enriched by the rest antibiotics (oneway ANOVA, p < 0.05).
To further elucidate the relationship between the airborne bacteria and horizontally acquired ARGs, we used ResFinder annotations for all samples. Resfinder predicted resistance phenotypes, showing mainly macrolide resistance and aminoglycoside resistance phenotypes (Table S4). ResFinder identified a total of 260 horizontally acquired ARGs in the airborne bacteria and dust samples. The hosts that horizontally acquired ARGs are shown by the network in Fig. 2c. Staphylococcus and Acinetobacter were the major hosts of all detected classes of horizontally acquired ARGs. Most of them are commensals of human skin (Staphylococcus epidermidis, Staphylococcus cohnii, Staphylococcus arlettae, Acinetobacter baumannii, Acinetobacter schindleri, and Acinetobacter johnsonii, etc.), but equally, most of them were also human opportunistic pathogens. In the airborne bacteria, nine and four different classes of ARGs were observed in Staphylococcus and Acinetobacter, respectively. In dust, eight and three different classes of ARGs were present in Staphylococcus and Acinetobacter, respectively. The types of horizontally acquired ARGs found in the air samples were similar to those found in the dust, indicating that there is a potential transfer of ARGs between airborne bacteria and adjacent surface environment (dust). We also quantified the relative abundance of horizontally acquired ARGs from Staphylococcus and Acinetobacter in various samples to emphasize the impact of horizontally acquired ARGs during the COVID-19 pandemic. Results showed an increase in the relative abundance of horizontally acquired ARGs during the pandemic ( Figure S3). Significantly, the relative abundance of acquired methicillin resistance gene (mecA) and fosfomycin resistance gene (fosB) in Staphylococcus was also found to increase ( Figure S4). Overall, antibiotic resistomes in the air samples of railway stations during the pandemic were featured with higher horizontal transfer and were hosted more by human pathogens than those prior to the pandemic.
A total of 85 acquired ARGs were identified by ResFinder, mainly carried by Staphylococcus, Acinetobacter, and Enterobacter (Fig. 4b).
Macrolide, beta-lactam and phenicol resistance genes were the main classes of horizontally acquired ARGs. These bacteria that carry ARGs also carry many virulence factors (Fig. 4c). In addition, we also quantified relative abundance of these MAGs, and the results revealed that their relative abundance significantly increased during the pandemic (Fig. 4d).

Potential ARG dissemination risks
According to Fig. 4c, the phenicol-resistance gene cat86 and the macrolide-resistance gene mph(C) were the two most annotated horizontally acquired ARGs. We looked into the genomic locus of the phenicol-resistance gene cat and the macrolide-resistance gene mph(C) to assess their mobility potential. They were distributed across multiple MAGs and occurred in different species. It is worth noting that these genes were found to have the potential for cross-species transfer in this study (Fig. 5 ab). The mph(C) gene appeared not only in Staphylococcus spp. but also in Macrococcus bohemicus, the cat gene was found in Bacillus spp. and in Cytobacillus kochii. Mobile genetic elements (MGEs) were found in the vicinity of these genes such as transposon Tn552 DNAinvertase Bin3 and transporter YycB.
Notably, The microbiome novelty score (MNS) ≥ 0.12 (the cutoff for novel microbiomes) was all observed in the dust samples collected during the epidemic. The overall MNS (Fig. 5c) values for the samples during the pandemic were significantly higher than those prior to the pandemic (one-way ANOVA, P < 0.05). Simultaneously, overall MIP values were found to be increased following the pandemic outbreak (Fig. 5d). Moreover, we assessed the potential for ARGs to be disseminated into human pathogens by the MetaCompare pipeline to explore their AMR risks (Fig. 5f). The overall risk score for the airborne bacteria during the pandemic was significantly higher than that before the pandemic (one-way ANOVA, P < 0.05).

Sources of air resistome in public transport
In order to estimate the relative contribution of different sources to the resistance in the air of public transport sites, Bayesian source tracking was applied by using the public datasets as a putative source. Different ecotypes have unique microbial taxa and ARG composition, the total 61 source samples from China in this study, including human skin, animal feces, human gut, human oral, sludge, hospital sewage, and estuary. As illustrated in Fig. 5e, human skin and estuary were significantly dominant contributors (Average: 52 % and 24 %, respectively). Human oral (Average: 0.56 %) and human gut (Average: 0.69 %) were associated with the least source contribution in all samples. In addition, antibiotic resistance associated with human skin had a higher contribution to indoor air than to outdoor air (Fig. 5e). In addition to using resistome of public data from different environments as "source", we also performed a separate SourceTracker analysis using dust collected at  railwaystions. The average contribution of dust samples to antibiotic resistomes of the indoor air bacteria reached 93 % ( Figure S5). Resistome risk of dust samples in this study was compared with public data (Fig. 5f), and results showed that the risk value for dust prior to the outbreak in railway stations was similar to those for environmental samples like estuaries. Notably, the risk value for the dust during the pandemic was significantly higher than that for estuary samples (oneway ANOVA, P < 0.05).
We used the Microbiomes Search Engine 2 (MSE2: https://mse.ac. cn) platform to search query microbiomes in the global data space based on the taxonomic similarity across the microbiome (Jing et al., 2021). In the terms of source distribution, human-associated sources were the most dominant contributor reaching 38.52 % (Skin: 25.93 %, Gut: 1.11 %, Nose: 0.74 %, Oral: 5.19 %, and other human body sites: 3.70 %), followed by building (17.04 %) (Table S3). Moreover, using global microbiome databases, we produced a bird's-eye view based on the microbiome similarities (Fig. 6). Results demonstrated the high similarity (0.85-0.94) between air samples in Guangzhou, China, and those from global samples via the MSE2 search engine (Table S3).

Discussion
As COVID-19 poses a serious threat to global health, there are concerns about the potential ecological impact of high concentrations and doses of disinfectants and antibiotics used to curb the spread of the virus (Bandyopadhyay and Samanta 2020;Chen et al., 2021b). A special report released by the Centers for Disease Control and Prevention (CDC) in 2022 claims that the COVID-19 pandemic has erased years of progress in curbing antimicrobial resistance in the United States (https://www. medscape.com/viewarticle/978163#vp_1). However, the utility of metagenomic surveys in public transit bioaerosols during the pandemic remains underexplored. This study represents, to the best of our knowledge, the first attempt to reveal airborne microbiome and resistome in public transit centers during the pandemic.
Our results indicated that the diversity of airborne ARGs decreased following the pandemic, but the relative abundance of the core airborne ARGs increased, associated with the dominance of multidrug ARG types. Multidrug resistance has been an abundant type of ARGs in air samples such as hospitals (Jia et al., 2020;Wu et al., 2022). The multidrug efflux pump mechanisms can be attributed to its involvement in multiple bacterial functions as well, not just antibiotic efflux (Jauregi et al., 2021;Piddock 2006). Multidrug resistance may be a normal state for most bacteria to overcome environmental stress (Wright 2007). In addition, the significant reduction in ARG diversity may have been caused in part by the high use of disinfectants during the COVID-19 pandemic. Previous studies have reported that chlorination effectively lowered the amount of ARG subtypes (Lin et al., 2016;Ma et al., 2022). It has been discovered that chlorine disinfection could enrich specific types of ARGs and effectively reduce the number of unique ARGs (Jia et al., 2015;Ma et al., 2022).
It has been cautioned that control measures during the pandemic could pose environmental and public health risks by accelerating the spread of antimicrobial resistance (Chen et al., 2021a;Getahun et al., 2020;Sills et al., 2021). Antimicrobial resistance may develop as a result of the widespread use of biocidal agents in the clinical environment as well as in public environment and personal hygiene (Hsu 2020;O'Neill, 2016). In this study, this speculation was verified by enriching ARB from several samples. Selective culture of antibiotics-based enrichment allows us a clear insight into the airborne microbiome with real resistance phenotypes. It was observed that multidrug resistance (seven antibiotics) was increased in the air samples following the outbreak. We consider that the control measures during the pandemic are selective for more antibiotic-resistant organisms, just as previously reported hospital cleaning measures (e.g. disinfection) had the same effect (Chng et al., 2020). Increasing use of disinfectants might allow microorganisms in the environment to select for broad-spectrum resistance features . Due to the biofilm lifestyle, pathogenic microorganisms showed more resistance to the disinfectants (Frieri et al., 2017). It also showed that disinfectants can facilitate the exchange of ARGs between bacteria .
On the other hand, our data show that many pathogens with virulence factors (for example, Acinetobacter and Pseudomonas) carry antibiotic resistance, ARGs and horizontally acquired ARGs became abundant during the pandemic. Previous studies have supported this finding, claiming that ARB is more resistant to chlorine disinfection than antibiotic-sensitive bacteria (Khan et al., 2016) and that disinfectant- resistant bacteria may also be resistant to antibiotics and carry multiple ARGs (Zhang et al., 2019a). The disinfectant exposure can induce a range of cellular responses, and these effects ultimately lead to the enhanced transformation of ARGs (Guo et al., 2015;Zhang et al., 2017). Another troubling issue is that pathogenic bacteria continue to evolve sophisticated genetic systems to acquire and regulate endless mechanisms of antibiotic resistance (Gootz 2010). The highquality MAGs obtained by assembly provide a clearer resource for the distribution of strains and the diversity of ARG boxes in the atmospheric environment of public transit centers. We found that many pathogen species carry large amounts of ARGs. For instance, Enterobacter cloacae (ARGs, n = 22) is a clinically significant pathogen capable of producing a variety of infections such as pneumonia, urinary tract infections, and sepsis (Annavajhala et al., 2019;Wisplinghoff et al., 2004). Another rare human pathogen belonging to the Enterobacteriaceae family is Leclercia adecarboxylata (n = 16) . In addition, our analysis highlights the presence of clinically important ARG combinations in the air of public transit centers (for example, oxazolidinones, phenicol, and fosfomycin resistance). This is a worrying result considering the potential use of fosfomycin in the treatment of MRSA infections (Chng et al., 2020;Falagas et al., 2009). Oxazolidinones are also considered to be the last resort for the control of clinical infections caused by multidrug-resistant gram-positive pathogens (Du et al., 2019). In particular, we also found the prevalence of horizontally acquired ARGs and the possibility of cross-species transfer of genes.
We found that the resistome of indoor air and surrounding dust overlapped, consistent with previous studies (Prussin and Marr 2015). Dust resuspension increases the concentration of microorganisms in the air (Ding et al., 2020). The air microbiome and resistance group in the railway stations appear to have great similarity to human skin by resistome and microbiome source tracking (Leung et al., 2021). This is not surprising because the shedding of skin cells or the aerosolization of saliva strongly influences the airborne bacteria in indoor environments. A large number of skin sources in the station air may be the result of direct skin shedding and particle resuspension (Hospodsky et al., 2012;Miletto and Lindow 2015). We must consider the interplay between the indoor air of public transport hubs and the microorganisms of travelers. Therefore, we need to assess the risk of airborne resistome in a public transport hub. We found the overall Metacompare risk score of samples during the pandemic was higher than that prior to the pandemic outbreak. Indoor dust samples collected during the pandemic generated the highest level of risk score (29.58). The highest level of risk score was found even higher than the highest risk score reported by Wu for hospital PM 2.5 samples in Guangzhou (25.60) . In addition, the MNS was significantly higher in samples during the pandemic compared to those before the outbreak. It is thus vital to further monitor the air environment in order to assess the potential human health risks.
We found changes in antibiotic resistance before and during the pandemic at the Guangzhou railway stations, this could have global impact. Using the Microbiome Search Engine, a global microbiome network with 177,022 microbiome samples, we found that the air microbiome from the Guangzhou railway stations was similar to microbiomes from every continent in the world (meta-Storms similarity > 0.85). Globally, disinfectants play a very important role in the prevention of COVID-19. A survey from the US showed that 60 % of respondents reported cleaning or disinfecting their homes more frequently than before the outbreak (Gharpure et al., 2020). Moreover, extensive, and rigorous sanitary disinfection of buildings using chlorine-based disinfectants is common practice in many countries during the pandemic (Chu et al., 2021). Therefore, what happened at the Guangzhou railway stations could also happen in other settings.

Environmental implications
The results from this study showed that control measures during the pandemic decreased airborne ARGs diversity, enriched specific types of ARGs, and increased multi-antibiotic resistance in airborne bacteria, thus elevating human health risks. Although the investigation was carried out only in one city in China, it still has global implications. COVID-19 has become a significant threat to global health. A recent study, which examined antibiotic concentrations in human urine, showed that the general population in several regions of China had an increased intake of antibiotics during the pandemic (Yu et al., 2022). Heavy use of disinfectants around cities and wide use of antibiotics in the treatment of COVID-19 disease clearly affect development of antibiotic resistance and subsequently put great pressure on the public health system. Since the world is well connected via transport, trade, and air movement, bacteria can be transported globally (Meng et al., 2019;Zhu et al., 2021). Therefore, antibiotic resistance could become a major challenge for humanity in the post-epidemic era.
Although we were fortunate to have collected pre-outbreak samples from railway stations during routine sampling, there are still some limitations to this study. Due to the requirements for outbreak control in public places and difficulty in site access, the collected aerosol samples were not enough for direct sequencing. Thus dust samples were applied to understand the microbial composition that could not be cultivated. Further investigations with more samplings are needed to explore resistome of aerosols in the future.