The high diversity and abundance of the soil resistome in Alaskan tundra, temperate prairie and tropical ecosystems support the view that ARGs are naturally ubiquitous, and in widely different terrestrial ecosystems. The ARGs detected in native soils can potentially confer resistance to all major antibiotics used to treat humans and animals, such as beta-lactams (FEZ, Thin-B, and LRA, PER, TEM, and OXA genes), macrolides and lincosamides (erm genes), quinolones (qepA), aminoglycosides (aac and aph genes), and tetracyclines (tet genes). It is not surprising that ARGs naturally exist in native soils  because many antibiotics are produced by soil microorganisms, and indeed were the original source of pharmaceutical products . In accord with our observation, previous studies also identified divergent beta-lactamase resistance genes and a novel chloramphenicol resistance gene from undisturbed Alaskan soil [40, 41]. Vancomycin is regarded as the last line of defense against MRSA strains, but thirteen subtypes of vancomycin resistance genes were detected in these native soils, including vanH, vanA and vanX which are found in clinical pathogens S. aureus and vancomycin-resistant Enterococci . Similarly, D’Costa et al. detected the three vancomycin genes in 30,000-year-old permafrost sediments, and further analyses confirmed the similarity in structure and function between the ancient vanA and their modern variants . These results support the view that the primary function of ARGs in native environments is not antibiotic resistance, but other functions beneficial for microbial life [44–46]. Antibiotic concentrations are extremely low in native soils, so harboring antibiotic resistance genes is not necessary for their defense and would even be a burden for microbes, unless it has other beneficial functions. We observed significantly lower abundance of mobile element associated ARGs in Alaskan tundra soils compared to Midwest prairie and Amazon soils, perhaps due to the lower microbial diversity in Alaskan tundra soils. However, antibiotic resistance could become a primary function and be further transferred to pathogens when selection occurs. For example, some clinically relevant resistance genes have been acquired from organisms where their native function is not antibiotic resistance, and which may not even confer a resistance phenotype in their native context .
We defined background ARGs as those which were shared by all soils regardless of ecosystem type and geography. These ARGs were also found across various ecosystems in previous studies. Thirty of our background ARGs were detected in paddy soils ; ten were found in dryland (peanut) soils ; at least sixteen were observed in greenhouse soils  and ten were found in Antarctic soils . Most of the background ARGs are multidrug resistance genes with efflux pump as the dominant mechanism, a trait that has other natural physiological functions and found in the cytoplasmic membrane of all bacterial cells . For example, 11 of the background ARGs are involved in Mex-Opr efflux pump systems, and they are known to play a prominent role in the multidrug resistance of gram-negative bacteria such as Pseudomonas [49, 50]. The AcrAB efflux pump plays a physiologic role of pumping out bile acids, fatty acids, and various toxic compounds [51, 52]. bcrA confers resistance to bacitracin and is predicted to be responsible for energy coupling to the transport system . Thus, we argue that the background ARGs should be considered as a separate category, generally of low risk, when evaluating ARG risk in soil environments. However, this doesn’t mean that background ARGs are risk free, since some of them have been found in plasmids and can be enriched with anthropogenic activity. For example, macB can be easily acquired by mobile elements, and thus spread macrolide resistance . Background ARGs acrA, vanC, and mexF were found significantly enriched by the application of sewage sludge and chicken manure to soil . These results also imply that compared to the abundance of ARGs, assessment of ARG mobility may be more important for ARG risk evaluation since ARG transfer into pathogens is a primary risk factor.
The soil resistome profile had a significant geographic pattern, which was greater than land use change. The resistome profile of Amazon pasture soil was more similar to Midwest pasture soil, than Amazon rainforest soil was to Midwest pasture soil, indicating the above-ground vegetation’s influence on the soil resistome. The significant correlation (p < 0.01, R2 = 0.795) between ARG diversity and bacterial diversity (Additional file 1: Fig. S5) suggests that the lower bacterial diversity may explain the lower ARG diversity in tundra. We found significant correlation between the resistome profile and microbial community structure, indicating that the bacterial composition shapes the ARG profile in soil. This is consistent to previous studies which also found strong correlation between ARG profile and bacterial community structure in various environmental contexts [56–58]. Thus, the variation of soil resistomes at different geographical locations was probably related to the differing plant diversity, climate, and edaphic factors such as pH and soil organic matter [59–61] which will select different populations (and hence the ARGs they carry) or some ARGs by their alternative function.
It's well known that the introduction of selective or co-selective pressure by human activities is a primarily responsible for the enrichment of ARGs in soils. For example, irrigation with reclaimed water led to enrichment of 60 ARGs . Long-term application of pig manure significantly enhanced the abundance of tetL, tetB(P), tetO, tetW, sul1, ermB, and ermF as compared with inorganic fertilizers . However, it’s not clear whether the normal agricultural activities such as crop cultivation affect the soil resistome. In this study, no significant change was observed in either ARG diversity, resistome abundance, or abundances of mobile element associated ARGs after long-term continuous cultivation. However, the cultivated soils from the three Midwest sites tend to have similar resistome profiles which may be due to selection for similar adaptations of the bacterial community to grain cultivation. Cropping system type, fertilization and other soil management practices are generally thought to be other factors that can influence the soil resistome [10, 62]. In these study sites antibiotics and heavy metals were not used so external factors would not have provided for selection. norB was one of the two genes significantly changed after continuous cultivation. It confers resistance to multiple drugs but is also involved in soil denitrification. Both N fertilizer and/or manure applications were used at the temperate and tropical sites, which may have favored soil denitrification and thus enriched norB gene . Overall, our results suggest that previous standard cultivation and fertilization practices of US Midwest (primarily moldboard plow, inorganic N.P.K and low levels of manure, e.g. cattle grazing) did not increase the public health risk of ARGs in soil.
Twenty-three new ARGs emerged in soil after converting rainforest to pasture. The grass vegetation (Urochloa brizantha, Urochloa decumbens, Panicum maximum) and/or cattle grazing, which includes their manures, may be responsible for the newly emerged ARGs in pasture soil by selecting different microbial populations and/or increasing their diversity (Additional file 1: Fig. S6) . Vancomycin resistance gene vanA and fosfomycin resistance gene fosB were enriched in pasture soil, corroborating similar results found by Yang et al. . Enterococcus, Bacillus, and Staphylococcus are important hosts of vanA and fosB, and are commonly found in animal manure [65–67]. We observed increased Enterococcus, Bacillus, and Staphylococcus in pasture soils (Additional file 1: Fig. S7), suggesting that enrichment of the two ARGs could be at least in part due to the increase of these taxa. Both enrichment and attenuation of ARGs were observed after land conversion, and we speculate that ARGs are enriched or attenuated following land use change because this selects for microbes that happen to be ARG carriers (or not) or for an alternative function of the ARG. For example, as noted above, the enriched norB in pasture soil may be because the grazing activity as well as the observed higher soil moisture in the rainy season (when the site was sampled) stimulated denitrification .
Growing evidence has shown that some ARGs in pathogens are acquired from environmental bacteria through horizontal gene transfer. For example, the CTX-M extended spectrum beta-lactamase originated from chromosomal genes of an environmental genus, Kluyvera , and the clinical vanA has been found in environmental Bacilli . Thus, we selected 12 ARGs which are clinically important and could transfer between bacteria  and assessed their sequence similarity to clinical ARGs in human pathogens. The ARGs we detected in Amazon soils are distinct from those found in human pathogens, implying that most ARGs in the natural soil resistome are not demonstrated as problematic. Only a few clinical-similar reads (> 90% amino acid identity) of our tested ARGs were observed, but further check with a target gene assembly method confirmed that most of them were aligned to conserved regions of these genes. For example, ampC codes clinically important cephalosporinases which cause resistance to cephalothin, cefazolin, cefoxitin, and most penicillins, but the assembled ampC in Amazon soils share less than 54% of similarity with those found in human pathogens. It’s noteworthy that most researchers used short-read based BLAST for ARG search, which provides a sensitive detection but will also recover non-functional pseudo genes or conserved domains. By contrast, ARG evaluation with assembled genes will miss some low abundance ARGs but should better reflect the presence, abundance and sequence similarity of potentially functional ARGs.
The wide co-occurrence of ARGs observed in this study is consistent with previous findings that soil ARGs commonly exist in clusters [17, 21, 28]. mdtA, mdtB, and mdtC co-occurred, consistent with the fact that they are parts of the MdtABC-TolC efflux pump system. Many background ARGs were found in ARG clusters, indicating that background ARGs can also be co-enriched when selective pressure for that population or trait emerges. For example, vanA, vanD, and vanX were found in the same cluster, and they were found co-enriched in soils during conversion of Amazon rainforest to pasture. The ARG cluster analysis could improve our understanding of the co-occurrence and spread of ARGs in the soil environment. However, the network analysis is based on a statistical method and may not reflect gene linkage or population level co-selection.
In comparison to traditional the qPCR method, the metagenomic approach is more comprehensive for assessing the entire known resistome. The ARGs-OAP approach used in this study can not only detect known ARGs but also identify ARG variants not yet recognized but might be the natural reservoir for emerging new subtypes since the Hidden Markov Models detect sequence diversity within the subclasses . It is noteworthy that some regulatory genes were contained in the SARG database as well as the widely used CARD database . It’s problematic as to whether regulatory genes should be counted as ARGs since they only control expression of ARGs. For example, vanR and vanS cannot confer resistance to vancomycin, but vanR can promote co-transcription of vanA, vanH, and vanX when activated by vanS . A high abundance of regulatory genes was detected and they differed in soils from the several ecosystems. We removed regulatory genes from our further analyses since the potential risk of ARGs is largely from the horizontal transfer of structural genes which code for functional proteins. In addition to the regulatory genes, some ARGs are components of a functional complex, for which an individual ARG cannot code antibiotic resistance without others. For example, many ARGs detected in our study are components of mdtABC-tolC, acrAB-tolC, and mexEF-oprN efflux complexes. Thus, the addition of ARGs belonging to a complex can inflate the total resistome abundance. Granted, soil as an important reservoir of ARGs; it harbors background ARGs that may or may not become problematic, probably harbors ARGs not yet emerged, and can harbor clinical ARGs, most likely to have entered soil from human or animal waste disposal. We recommend that more attention be paid to ARG genes or gene sets necessary for resistance function, for their status relative to common ARG backgrounds, for linkage to mobile genetic elements and their correspondence or linkage to host populations. Sequence similarity may or may not be indicative of potential ARG function but it is a strong indicator of whether the ARG source was from a known clinical resistance and detectable by methods targeting the clinical gene variant.