Broad-spectrum profile of ARG abundance in tap water
In total, 181 ARG subtypes belonging to 16 ARG types were detected in at least one of the 25 tap water samples (Additional file 1 Table S4 and Additional file 1 Table S5). The ARG diversity (number of ARG subtypes) was in the range from 16 (sample S11, Macau, China) to 88 (sample S13, Guangdong province, China; Fig. 1). Overall, 14 tap water samples belonged to resistance level I (< 0.1 capc), 9 belonged to resistance level II (0.1~0.2 capc), and 2 belonged to resistance level III (> 0.2 capc; Additional file 1 Table S6). The highest ARG enrichment (resistance level III) was detected in samples S16 (4.3 × 10−1
capc, Henan province) and S04 (3.7 × 10−1
capc, Hebei province), which were both collected from northern China. This ARG enrichment was ~ 12-fold higher than that of the lowest S12 (Hainan province, southern China). Interestingly, Zhang et al. observed significantly high antibiotic emissions in the same regions, Henan and Hebei province of mainland China [28]. It has been shown previously that antibiotics released into the environment could contribute to the selection for antibiotic resistance [29]. The ARGs encoding resistance to bacitracin, multidrug, sulfonamide, beta-lactam, and aminoglycoside were more frequently detected in these tap water samples (Additional file 1 Table S7). As expected, these dominant ARGs were usually associated with antibiotics that have been extensively used in human or veterinary therapy [16]. Moreover, a moderate ARG level was observed in the drinking water of Singapore and the USA where tap water is considered to be direct potable water, and the similar ARG composition of their ARG compositions was revealed by cluster analysis (Additional file 1 Figure S2).
We also compared the top 8 ARG types in the drinking water samples (Fig. 2). Bacitracin-ARGs were found to be the most dominant in 68% of the tap water samples, with an abundance of 2.2 × 10−3–1.7 × 10−1
capc. Multidrug-ARGs, which encode resistance to multiple antimicrobial drugs, dominated in 20% of the tap water samples with an abundance of 2.7 × 10−3–2.4 × 10−1
capc. In contrast with previously reported data showing that the dominant ARG types in sewage, animal feces, and activated sludge were multidrug-ARGs, tetracycline-ARGs, and aminoglycoside-ARGs, respectively [16], we found that bacitracin-ARGs had the highest abundance in drinking water. A few previous studies also reported the prevalence of bacitracin-ARGs in freshwater [30,31,32]. Bacitracin resistance genes were once considered to be intrinsic to bacteria as they are widespread in 153 genera [21]. Diverse bacitracin resistant strains were isolated from deep glacial ice, such as Herminiimonas glaciei [30] and Dyadobacter hamtensis [31]. All 60 Aeromonas strains isolated from freshwater in India were resistant to bacitracin [32]. Moreover, significantly high levels of bacitracin-ARGs were observed in treated wastewater and Tibetan sediment, with levels up to 70% of total ARG abundance [6, 33]. Bacitracin is used for topical treatment of localized skin lesions, eye infections, and also for the prevention of wound infections. Additionally, its application in chicken feed has been approved by the US Food and Drug Administration, which could be greatly contributing to bacterial antibiotic resistance. Although bacitracin-ARGs were detected in these drinking water samples, whether these resistance genes could bring potential risks to public health still requires more systematic researches.
Among the 181 identified ARG subtypes, some were generalists existing in all samples (Additional file 1 Figure S3), for example, multidrug efflux protein, multidrug HAE1-family protein, multidrug mexF, beta-lactam TEM-2, macrolide macB, and beta-lactam TEM-15, etc. Their abundance accounted for 72.7% of the total ARGs identified from the drinking water samples (Fig. 3). In total, 65 ARG subtypes were specialists only detected in one tap water sample, for example, the chloramphenicol resistance gene floR was only detected in S04 (Hebei, China). Some of the ARG subtypes identified in this study were detected previously using traditional PCR-based techniques, such as ampC and mecA in municipal wastewater [34]; tetO, tetW, and tetQ in lagoons [35]; and sulI, sulII, and blaTEM in river and drinking water sources [36]. However, many of the identified ARG subtypes had not previously been revealed in water environments. Thus, the traditional PCR-based methods, limited by primers, could not provide comprehensive ARG profiles for drinking water samples.
To further explore the potential correlation among ARGs, network analysis was used. It revealed the ARG combinations of mexE-mexF-oprN and sul1-aadA-aadB in drinking water (modularity = 0.493; Additional file 1 Figure S4, Additional file 1 S3). These gene combinations were previously discovered on the whole genome of Pseudomonas aeruginosa and Salmonella enterica, respectively [37, 38]. Thus, the metagenomic based approach largely facilitated ARGs investigation over a larger spectrum without PCR bias and captured a more comprehensive picture of the correlation among ARG profiles in drinking water.
Comparison of ARG profiles from drinking water and other environmental samples
To explore ARGs autochthonous to drinking water samples, the catalogue of ARGs in drinking water was further compared to 56 environmental samples from seven niches, i.e., sediment (n = 7), river water (n = 5), sewage (n = 4), treated wastewater (n = 4), activated sludge (AS, n = 13), anaerobic digestion sludge (ADS, n = 11), and feces and wastewater from livestock farm (n = 12). Basic information about the 56 environmental samples is summarized in the supporting information (Additional file 1 Table S9), for which 39 metagenomic data sets have been used in our previous studies on the antibiotic resistome [7, 16]. In total, 501 ARG subtypes belonging to 20 ARG types were detected in the 81 environmental samples. The abundance of ARGs followed the order of sediment < river water < drinking water < sewage treatment plant (STP) ADS < STP AS < STP effluent < STP influent < feces and wastewater from livestock farm (Fig. 4a). In total, 84% of drinking water samples had higher total ARG abundance than that in sediment and soil, and 8% of samples (S04 and S16) had more ARGs than STP AS and STP ADS. The ARG levels in sewage and feces and wastewater from livestock farm were within 1–2 orders of magnitude higher than drinking water. The 181 identified ARG subtypes in the drinking water samples were detected in other environmental ecosystems, with a percentage of 32–92% (Fig. 4a). The nine ARG subtypes prevalent in all the drinking water samples were also found in activated sludge, influent, and livestock farm samples; however, two subtypes (Beta-lactam TEM-2 and TEM-15) were absent from sediment, river water, STP ADS, and STP effluent (Fig. 4a). The beta-lactam resistance genes of TEM-2 and TEM-15 were only prevalent in 9–25% of AS, ADS, and effluent samples but occurred in all STP influent samples (Additional file 1 Table S10). The grouping patterns shown in the PCoA plot demonstrated that drinking water samples were clearly separated from sewage and feces and clustered more closely with river water (Fig. 4b), indicated by a higher similarity of ARG profiles with natural water environments.
The host of ARGs in tap water samples
There were 264 contigs assembled from drinking water metagenomes that carried ARGs. These ACCs were annotated as fragments of Acidovorax, Acinetobacter, Aeromonas, Methylobacterium, Methyloversatilis, Mycobacterium, Polaromonas, or Pseudomonas, except for those unclassified. Among them, 34.5% of the ACCs were identified as sequence fragments of Pseudomonas, frequently carrying multidrug-related ARGs with a percentage of 80% (Fig. 5). The multidrug resistance genes m
exF and hydrophobe_amphiphile efflux-1 (HAE1) family protein were frequently carried by Pseudomonas. Similarly to a previous study, 82% of the Pseudomonas aeruginosa strains isolated from a hospital wastewater treatment plant were resistant to multiple antimicrobial drugs [39]. P. aeruginosa is a notoriously difficult-to-treat pathogen that can cause severe disease and infections. In P. aeruginosa, the efflux mechanism for antibiotic resistance may pose a great challenge to antibiotic development [40]. Thus, the high frequency of ARGs carried by Pseudomonas in drinking water may increase the risk of infection and antibiotic ineffectiveness in human beings. Additionally, all the Methyloversatilis and Polaromonas contigs were observed to carry multidrug- and bacitracin-related ARGs, respectively. Moreover, two contigs were able to be assigned to species level. The contig of P. aeruginosa, S13_contig_64868 carried class A beta-lactamase resistance gene, and S08_contig_54033 annotated as contig of Hylemonella gracilis carried bacitracin undecaprenol kinase. The taxonomic annotation of ACCs greatly strengthens the identification of possible ARG hosts in drinking water samples.
The spatial distribution of bacterial community
Totally, 26,862 bacterial OTUs were observed in at least one of these tap water samples. Among them, the highest bacterial diversity across all samples, 5018 OTUs, was observed in Tibet tap water of mainland China, followed by 4302 OTUs in water from Xinjiang, and 4262 OTUs in water from Inner Mongolia in northwestern China. The lowest bacterial diversity was observed in Macau (122 OTUs), followed by California of the USA (210 OTUs), and Sichuan province of China (226 OTUs). Based on the OTUs and corresponding abundances, PCoA was performed to compare the spatial variations of bacterial community using weighted Unifrac distance, which considered both species abundance and phylogeny (Additional file 1 Figure S6). Overall, differences in phylogenetic diversity and abundance of OTUs were obvious across drinking water samples. Notably, the composition of bacterial communities was more similar among the triplicated samples collected from the same tap, triplicated DNA extractions from the same sample, DNA extractions using different kits, and triplicated PCRs using primers with different barcodes. Thus, in the present study, DNA extraction strategy and PCR amplifications using primers with different barcodes are not expected to be factors influencing the detection of microbial compositions in tap water samples.
The histogram in Fig. 6 shows the percentages of the top 10 bacterial classes (accounting for a total abundance of 84.8%) that were most abundant within tap water. The most abundant bacterial classes were Alphaproteobacteria, followed by Betaproteobacteria, Gammaproteobacteria, and Actinobacteria. Significantly, the cluster patterns revealed by weighted UniFrac tree (Fig. 6) showed that drinking water sample collected from Macau (S11) closely resembles the sample from California in the USA (S25) based on bacterial community analysis. The Macau and California samples had low OTU diversities of 122 and 210 and had significantly high percentages of Alphaproteobacteria at 99.6 and 90.2%, respectively. As shown in Fig. 3, ten OTUs were widely spread with a detection frequency of 80% (20 out of 25 samples), accounting for 16.8% of the total abundance. To further explore the shared, specialist, and generalist bacteria in tap water samples, all of the OTUs that could be annotated at genus level by SSU SILVA database were summarized to obtain the matrix of percentages at genus level in all tap water samples. About 42% of 16S rRNA gene sequences were annotated at genus level (686 genera). Nine genera, Sphingomonas, Pseudomonas, Mycobacterium, Acinetobacter, Hyphomicrobium, Planctomyces, Sediminibacterium, Legionella, and Rhodobacter were present in all tap water samples, and the average abundance was 0.7–14.9%. There were 36 generalist genera (occurrence observed in at least 80% of samples) found to be widely present in tap water samples, accounting for total percentage as 84.9%. The most dominant genus, Sphingomonas spp., is known to be aerobic and able to form biofilms. Sphingomonas spp. are often investigated in oligotrophic environments [41] and reported to be present in drinking water distribution systems [10, 42, 43]. However, none of the ACCs identified in the present study were annotated as Sphingomonas (Fig. 5
), indicating their low frequency of carrying ARGs in drinking water. The dominant genera, Pseudomonas (10.0%), Mycobacterium (3.4%) and Acinetobacter (3.3%), were found to carry aminoglycoside, bacitracin, multidrug, and sulfonamide resistance genes from metagenomic analysis of drinking water samples (Fig. 5
). Some Pseudomonas spp. (e.g., P. aeruginosa), Mycobacterium spp. (e.g., M. avium), and Acinetobacter spp. (e.g., A. baumannii) have been considered as opportunistic pathogens, and their capacity of thriving in drinking water supply systems could increase the risks of the exposure and spread of ARGs in drinking water. Thus, the dominance of these ARG-carrying bacteria observed in drinking water at the user end should receive more attention, and their potential negative effects merit further study.
The horizontal gene transfer potential for ARGs among bacterial population
Previous metagenomic analysis-based study revealed the exchange of ARGs between clinical pathogens and environmental bacteria [44], illustrating that ARGs could be transferred between different environments via specific bacteria, especially pathogens. Another study explored the potential HGT frequency among bacterial populations by using Procrustes analysis, and a low HGT frequency of resistomes (M
2 ≤ 0.5) was observed among soil bacteria [12]. Because drinking water exerts direct exposure to human beings, there is a pressing demand to determine the likelihood of the transmission of ARGs to specific bacteria and the probable hosts of ARGs. Here, network analysis together with Procrustes analysis was applied to explore the correlation between drinking water resistomes and bacterial population (Fig. 7). The modularity index of 0.786 indicated that the formed ARG-bacteria network had a modular structure [45], while positive correlations were frequently found in ARG–ARG and bacteria–bacteria pairings. No significant correlation was observed between ARGs and bacterial population using network analysis. To further validate the absence of significant correlation between ARGs and bacterial population, Procrustes analysis was used based on a one-way ANOVA test with Tukey post-hoc tests [12]. Similarly, the result of Procrustes analysis (M
2 = 0.606, P < 0.01) showed that the ARG content did not correlate with bacterial community, different from the strong correlation observed in soil samples, suggesting the potential of HGT of ARGs in drinking water microbiota [12]. The possible HGT of ARGs indicated by the correlation-based Procrustes analysis in drinking water microbiota may enhance the risks to human health. Thus, further observation and additional analyses are required to study the putative HGT of ARGs in drinking water systems.