Responses of the Soil Bacterial Community, Resistome, and Mobilome to a Decade of Annual Exposure to Macrolide Antibiotics

ABSTRACT

Biosolids that are applied to agricultural soil as an organic fertilizer are frequently contaminated with pharmaceutical residues that have persisted during wastewater treatment and partitioned into the organic phase. Macrolide antibiotics, which serve as a critically important human medicine, have been detected within biosolids. To determine the impacts of macrolide antibiotics on soil bacteria, every year for a decade, a series of replicated field plots received an application of a mixture of erythromycin, clarithromycin, and azithromycin at a realistic (0.1 mg kg soil⁻¹) or an unrealistically high (10 mg kg soil⁻¹) dose or were left untreated. The effects of repeated antibiotic exposure on the soil bacterial community, resistome, mobilome, and integron gene cassette content were evaluated by 16S rRNA and integron gene cassette amplicon sequencing, as well as whole-metagenome sequencing. At the unrealistically high dose, the overall diversity of the resistome and mobilome was altered, as 21 clinically important antibiotic resistance genes predicted to encode resistance to 10 different antibiotic drug classes were increased and 20 mobile genetic element variants (tnpA, intI1, tnpAN, and IS91) were increased. In contrast, at the realistic dose, no effect was observed on the overall diversity of the soil bacterial community, resistome, mobilome, or integron gene cassette-carrying genes. Overall, these results suggest that macrolide antibiotics entrained into soil at concentrations anticipated with biosolid applications would not result in major changes to these endpoints.

IMPORTANCE Biosolids, produced from the treatment of sewage sludge, are rich in plant nutrients and are a valuable alternative to inorganic fertilizer when applied to agricultural soil. However, the use of biosolids in agriculture, which are frequently contaminated with pharmaceuticals, such as macrolide antibiotics, may pose a risk to human health by selecting for antibiotic resistance genes that could be transferred to plant-based food destined for human consumption. The consequences of long-term, repeated macrolide antibiotic exposure on the diversity of the soil bacterial community, resistome, and mobilome were evaluated. At unrealistically high concentrations, macrolide antibiotics alter the overall diversity of the resistome and mobilome, enriching for antibiotic resistance genes and mobile genetic elements of concern to human health. However, at realistic antibiotic concentrations, no effect on these endpoints was observed, suggesting that current biosolids land management practices are unlikely to pose a risk to human health due to macrolide antibiotic contamination alone.

INTRODUCTION

Antibiotics have the potential to enter the environment through many disparate pathways, including discharge from antibiotic manufacturing facilities and hospitals (1, 2), aquaculture (3), animal agriculture (4), and municipal sewage (5, 6). Macrolide antibiotics, among various other pharmaceuticals, are discharged into the environment via municipal sewage effluent and are removed inefficiently by most wastewater treatment processes (6–8). Macrolides that are persistent during the wastewater treatment process can sorb to organic matter and be recovered in the sewage sludge which, if subsequently treated, is termed biosolids (5, 9).

Biosolids are rich in plant nutrients and are applied widely to agricultural soil as a valued fertilizer, but usage is highly variable; almost all of the biosolids that are produced in Spain (80%, 2016) and Ireland (79%, 2017) are land applied (10), whereas only 55% (2004) are land applied in the United States (11). There are concerns that the long-term application of biosolids to agricultural soil could introduce pharmaceutical residues, including macrolide antibiotics, into the soil and select for resistant soil bacteria (12). Should this process be the case, macrolide-resistant bacteria and the associated resistance genes could be transferred to humans through the consumption of plant-based foods, such as fresh produce (13, 14).

To investigate the response of the soil bacterial community, resistome, and mobilome to long-term macrolide antibiotic exposure, a series of field plots were established in 2010 on the Agriculture and Agri-Food Canada research farm in London, ON, Canada. Every spring, these plots have received an annual application of a mixture of macrolide antibiotics (erythromycin, clarithromycin, and azithromycin) to result in a nominal dose of 0.1 mg kg⁻¹ soil (here referred to as the low dose) or 10 mg kg⁻¹ (here referred to as the high dose) of each antibiotic. The low dose is intended to simulate what could be entrained into soil with an application of biosolids, whereas the high dose is an unrealistic “effects” concentration, namely, a positive control (15, 16). The antibiotics were added directly to the soil to evaluate their impacts on native soil bacteria while avoiding the confounding effects of enteric bacteria, pharmaceuticals, and other chemical residues carried in biosolids (9). In previous reports from this experiment, several antibiotic resistance genes that were predicted to encode resistance to macrolides and other antibiotic drug classes and mobile genetic elements, such as class 1 integron-integrases (intI1), were increased in abundance in response to exposure to the high, but not the low, dose of macrolides (12). All three antibiotics were subjected to accelerated degradation when introduced into soil that had been exposed previously to the high dose of macrolides for 5 years and of erythromycin and clarithromycin in the soil exposed to the low dose, relative to control soil that was not exposed to antibiotics (17).

To further investigate if macrolide exposure of soil promoted antibiotic resistance or changed the bacterial community at the environmentally realistic low dose, and to elucidate the mechanisms of increased resistance observed at the effect-inducing unrealistically high dose, soil DNA was obtained from these same field plots that were either treated with macrolide antibiotics for 10 years or were left untreated. The 16S rRNA genes and integron gene cassettes were PCR amplified and sequenced, and the whole soil metagenome was sequenced to measure the effects of antibiotic exposure on the alpha (within-group) and beta (between-group) diversity of soil bacterial taxa, antibiotic resistance genes, and mobile genetic elements.

RESULTS

DNA from untreated control soil and soil exposed to a low (0.1 mg kg⁻¹) or high (10 mg kg⁻¹) dose of macrolide antibiotics was used to generate three sequence data sets. The 16S rRNA genes were sequenced to investigate the alpha and beta diversity of the soil bacterial community, whole-metagenomic DNA was sequenced to investigate the alpha and beta diversity of the resistome and mobilome in addition to the soil bacterial community, and integron gene cassettes were sequenced to investigate the alpha and beta diversity of integron gene cassette open reading frames (see Text S1 in the supplemental material for sequencing statistics).

Bacterial community diversity.

The diversity of bacterial taxa was assessed from 16S rRNA amplicon sequences and whole-metagenome sequences. From both data types, the alpha diversity of soil bacterial taxa, as measured by the Shannon diversity index and Simpson’s index (data not shown) and by Chao1 richness (Fig. 1a), and the beta diversity, as measured by composition (Fig. 1b), were not significantly affected by antibiotic exposure. However, several taxa were identified as differentially abundant in response to exposure. At the phylum level, the relative abundances of Cyanobacteria (Bonferroni-adjusted P < 0.001 and fold changes [W] = −5.7 ± 0.7) and Nitrospirae (Bonferroni-adjusted P = 0.03 and W = −3.2 ± 0.9) were decreased in the high-dosed soil, and the relative abundance of Patescibacteria was increased (P = 0.03, W = 3.2 ± 3.8) in the high-dosed soil, relative to the control soil (see Fig. S1 in the supplemental material). The decreased abundance of Cyanobacteria in the high-dosed soil was observed only in the whole-metagenome sequence analysis, while the decreased abundance of Nitrospirae and increased abundance of Patescibacteria were observed only in the 16S rRNA amplicon sequence analysis (Fig. S1). Within the differentially abundant bacterial phyla, the decreased relative abundance of Cyanobacteria was driven by decreases of Microcoleus vaginatus (P < 0.001, W = −5.2 ± 0.8) and Oscillatoria nigro-viridis (P < 0.001, W = −5.3 ± 0.3) in the high-dosed soil as detected by the whole-metagenome sequence analysis, while the increased relative abundance of Patescibacteria was driven by increases of an unknown taxon of the order Saccharimonadales in the high-dosed soil (P < 0.001, W = 31.3 ± 0.7) as detected by the 16S rRNA amplicon sequencing analysis (Table 1). At the species level, no taxa were differentially abundant within Nitrospirae. Other taxa that were differentially abundant at the species level, but were not members of differentially abundant phyla, are displayed in Table 1.

FIG 1.

Chao1 richness (a) and composition (b) of bacterial taxa in antibiotic-exposed (0.1 and 10 mg kg⁻¹) and untreated control soil. (a) Number of observed bacterial taxa (richness) from the whole-metagenome (n = 3) and 16S rRNA amplicon (n = 4) analyses within each treatment group. Horizontal lines connect samples for visual clarity. Statistically significant differences between treatments groups were assessed with a one-way ANOVA. ns, indicates that differences in richness were not significant. (b) Principal-component analysis (PCA) ordination plots of the CLR-transformed Aitchison distances of soil bacterial taxa. Percentages of total variance explained by each principal component (PC1 and PC2) are displayed in the axis titles. Shaded ellipses correspond to 95% confidence intervals of treatment groups. P and F values are from a PERMANOVA with 999 permutations.

TABLE 1.

Mean fold changes of differentially abundant soil bacterial taxa in response to macrolide antibiotic exposure^a

Differentially abundant bacterial taxon by antibiotic dose	Phylum	Mean fold change ± 95% CIb	Adjusted P valuec	Data typed
0.1 mg kg−1
Mycolicibacterium tusciae	Actinobacteria	−7.09 ± 0.09	1.59E-10	WMS
Sphingomonas sp. Leaf20	Proteobacteria	4.51 ± 0.51	7.17E-04	WMS
Lysobacter sp.	Proteobacteria	4.53 ± 3.81	1.52E-02	16S
Uncultivated soil bacterium clone C112	Acidobacteria	35.81 ± 0.74	1.99E-277	16S
Phenylobacterium sp.	Proteobacteria	37.03 ± 0.61	1.06E-296	16S
10 mg kg−1
Ramlibacter sp. Leaf400	Proteobacteria	−7.82 ± 0.18	5.89E-13	WMS
Arthrobacter sp. Leaf69	Actinomycetota	−5.56 ± 0.31	2.99E-06	WMS
Oscillatoria nigro-viridis	Cyanobacteria	−5.29 ± 0.28	1.34E-05	WMS
Arthrobacter globiformis	Actinomycetota	−5.23 ± 0.31	1.87E-05	WMS
Microcoleus vaginatus	Cyanobacteria	−5.22 ± 0.76	2.01E-05	WMS
MLE1-7 sp.	Proteobacteria	4.36 ± 3.63	3.29E-02	16S
Unknown bacterium of order Saccharimonadales	Patescibacteria	31.26 ± 0.71	3.91E-211	16S

^aMean fold changes are in response to macrolide antibiotic exposure at low (0.1 mg kg⁻¹) and high (10 mg kg⁻¹) doses. Differential abundance analysis was performed using ANCOM-BC for the 16S rRNA amplicon or whole-metagenome sequence analyses.

^bFold changes are reported with 95% Bonferroni-adjusted confidence intervals. CI, confidence interval.

^cAll P values are Bonferroni adjusted. No taxa were identified as differentially abundant from both types of data.

^dWMS, whole-metagenome sequences; 16S, 16S rRNA amplicons.

Resistome diversity.

A total of 583 unique antibiotic resistance genes were detected across the soil metagenomes. The high dose significantly increased the Chao1 richness of total antibiotic resistance genes in soil (Tukey’s all-pairs test, P < 0.05), but no effect was observed for the low dose (Fig. 2a) or for the Shannon diversity index or Simpson’s index (data not shown). Similarly, the principal-component analysis (PCA) analysis indicated that the composition of antibiotic resistance genes from high-dosed soil differed from that from the control soil, but not the low-dosed soil (permutational multivariate analysis of variance [PERMANOVA] pseudo-F = 1.49, P = 0.03) (Fig. 2d). These differences in composition were driven largely by 21 antibiotic resistance genes that significantly increased in relative abundance in the high-dosed soil (P < 0.05) (Fig. 3a). Only five antibiotic resistance genes were differentially abundant (two decreased, three increased) in the low-dosed soil. No single resistance gene was significantly increased or decreased by exposure to both low and high doses (i.e., no gradual concentration-dependent treatment effect was detected).

FIG 2.

Chao1 richness (a to c) and principal-component analysis (PCA) of the compositions (d to f) of antibiotic resistance genes, mobile genetic element variants, and integron gene cassette open reading frames in antibiotic-exposed (0.1 to 10 mg kg⁻¹) and untreated control soil. Richness was measured as the number of unique antibiotic resistance genes (n = 3) (a), mobile genetic element variants (n = 3) (b), and integron gene cassette open reading frames (n = 4) (c) detected within each treatment group. Horizontal lines connect samples for visual clarity. Statistically significant differences between treatment groups were assessed with a one-way ANOVA and Tukey’s all-pairs test post hoc. ns, indicates that differences in richness were not significant. PCA ordination plots of the CLR-transformed Aitchison distances of antibiotic resistance genes (d), mobile genetic element variants (e), and integron gene cassette open reading frames (f). Percentages of total variance explained by each principal component (PC1 and PC2) are displayed in the axis titles. Shaded ellipses correspond to 95% confidence intervals of treatment groups. P and F values are from a PERMANOVA with 999 permutations.

FIG 3.

(a and b) Mean fold changes (effect sizes) of antibiotic resistance gene relative abundances in antibiotic-exposed soil (0.1, 10 mg kg⁻¹) relative to untreated control soil (a) and (b) the antibiotic drug classes (b) to which resistance is predicted (right). The abbreviated name of the predicted antibiotic resistance gene (ARG) is listed in the middle. (a) Fold changes for both treatment groups (blue, low dose; pink, high dose) are shown only for the genes that were significantly differentially abundant (P < 0.05) within at least one treatment group relative to the control. The black vertical line at zero represents no difference in relative abundance compared with the control group. Horizontal lines intersecting with points are error bars, indicating the extent of Bonferroni-adjusted 95% confidence intervals of effect sizes. Points without error bars indicate that the confidence interval is smaller than the diameter of the point. Filled points represent genes whose abundances were significantly different from the unexposed control soil, and open points represent abundances that were not significantly different. (b) Antibiotic drug classes (and triclosan) listed on the x axis are abbreviated as follows: AMINO, aminoglycoside; CEPH, cephalosporin; PENAM, penam; TRICL, triclosan; PENEM, penem; MONO, monobactam; FLUORO, fluoroquinolone; SLFNM, sulfonamide; SLFON, sulfone; DIAMIN, diaminopyrimidine; ACRID, acridine dye; PHENI, phenicol; CARBA, carbapenem; MACRO, macrolide; TETRA, tetracycline.

The 21 antibiotic resistance genes that had increased relative abundances in the high-dosed soil were predicted to confer resistance to 10 different drug classes of antibiotics, especially aminoglycosides (n = 10) and diaminopyrimidines (n = 4), as well as the biocide triclosan (Fig. 3b). Only two of these increased antibiotic resistance genes were predicted to encode resistance to macrolides (mphE and mexQ). Sixteen of these antibiotic resistance genes were predicted to confer resistance to classes of antibiotics which, like macrolides, target the bacterial ribosome (aminoglycosides, phenicols, and tetracyclines). The gene that had the greatest increase in relative abundance in response to exposure to the high dose was the aminoglycoside resistance gene aph(3″)-Ib (P < 0.001, W = 22.9 ± 0.5).

The composition of several target drug class groups of antibiotic resistance genes (aminoglycoside, diaminopyrimidine, phenicol, tetracycline, lincosamide, and streptogramin), but not of macrolide resistance genes, was significantly altered in the high-dosed soil (P < 0.05) (Fig. 4). Of the three antibiotic resistance genes that had increased relative abundances in the low-dosed soil, two were predicted to encode resistance to macrolide antibiotics (mexL, W = 5.6 ± 0.2; mexP, W = 5.2 ± 0.2) and one was predicted to encode resistance to aminoglycosides [aac(6′)-IIa, W = 4.0 ± 0.4].

FIG 4.

Principal-component analysis (PCA) ordination plots of the CLR-transformed Aitchison distances of antibiotic resistance genes, grouped by their target drug classes. Only the ordination plots with statistically significant differences in composition between treatment groups are shown (using a PERMANOVA with 999 permutations). Percentages of total variance explained by each principal component (PC1 and PC2) are displayed in the axis titles. Shaded ellipses correspond to 95% confidence intervals of treatment groups.

Seven antibiotic resistance genes had significantly decreased relative abundances relative to the control soil, namely, five in the high dose and two in the low dose (P < 0.05). Interestingly, all seven of these resistance genes were predicted to encode beta-lactamases (Fig. 3). bla_SHV-71 (W = −24.1 ± 0.3), bla_SHV-165 (W = −11.0 ± 0.3), bla_CTX-M-117 (W = −7.6 ± 0.5), Escherichia coli ampC (W = −5.7 ± 0.4), and bla_PEDO-1 (W = −4.4 ± 0.2) were decreased in the high-dose treatment, and bla_TEM-1 (W = −4.9 ± 0.3) and bla_TEM-22 (W = −4.1 ± 0.8) were decreased in the low-dose treatment.

Mobilome diversity.

In addition to antibiotic resistance genes, the diversity of mobile genetic elements within the antibiotic-exposed and -unexposed soil metagenomes was investigated. Overall, 390 unique mobile genetic element variants were detected across the soil metagenomes, including several transposases and insertion sequence elements (e.g IS91 and IS26). As observed with antibiotic resistance genes, the Chao1 richness of mobile genetic elements was significantly increased in the high-dosed soil metagenome (Tukey’s all-pairs test, P = 0.001) (Fig. 2b), and the composition of mobile genetic elements was significantly affected by the high dose of macrolides (PERMANOVA pseudo-F = 1.95, P = 0.03) (Fig. 2e). No effect of antibiotic exposure was observed on the Shannon diversity index or Simpson’s index of mobile genetic elements (data not shown).

This altered composition of mobile genetic elements in the high-dosed soil was driven largely by 20 mobile genetic element variants with increased relative abundances in the high-dosed soil (P < 0.05) (Fig. 5). Of these 20 increased variants, 15 were identified as tnpA, 3 as intI1, 1 as IS91, and 1 as tnpAN. The maximum effect size of the mobile genetic element variants that were increased in the high dose was W = 25.7 ± 0.1 for intI1.1588 (P < 0.001).

FIG 5.

Mean fold changes (effect sizes) of mobile genetic element (MGE) variant relative abundances in antibiotic-exposed soil (0.1 and 10 mg kg⁻¹) relative to unexposed control soil, and the GenBank accession numbers of the host genomes of the variants. Fold changes for both treatment groups (blue, low dose; pink, high dose) are shown only for the mobile genetic elements that were significantly differentially abundant (P < 0.05) within at least one treatment group relative to the control. The black vertical line at zero represents no difference in relative abundance compared with the control group. Horizontal lines intersecting with points are error bars, indicating the extent of Bonferroni-adjusted 95% confidence intervals of effect sizes. Points without error bars indicate that the confidence interval is smaller than the diameter of the point. Filled points represent variants whose abundances were significantly different from the unexposed control soil, and open points represent abundances that were not significantly different.

The only mobile genetic element variant with an increased relative abundance in the low-dosed soil was identified as tnpA.2602 (W = 6.0 ± 0.3, P < 0.001) (Fig. 5). Of the three mobile genetic element variants that were decreased in the low-dosed soil (IS91.447, IS91.1007, tnpA.2490), IS91.447 was also decreased in the high-dosed soil (low dose, W = −5.8 ± 0.4; high dose, W = −5.8 ± 0.4).

Integron gene cassette open reading frame diversity.

Both the alpha diversity and beta diversity of integron gene cassette open reading frames were unaffected by antibiotic exposure (Fig. 2c; Fig. 2f). However, 370 of 35,808 open reading frames were identified in the treated soil as differentially abundant relative to the untreated control soil (P < 0.05), and of these open reading frames, more were differentially abundant in the high-dosed soil (n = 246) than the low-dosed soil (n = 144).

In total, 16 unique open reading frames (0.04%) were predicted to encode antibiotic resistance to rifamycin (n = 12) and aminoglycoside (n = 4) antibiotic drug classes by antibiotic inactivation. The relative abundance of one gene cassette-embedded antibiotic resistance gene was significantly increased in response to the high dose (W = 5.6, P < 0.001), and one cassette-embedded resistance gene was decreased in response to the low dose (W = −10.0, P < 0.001); both genes showed greatest DNA sequence identity (<75%) to arr-4, a rifamycin antibiotic resistance gene (see Table S1 in the supplemental material). No putative cassette-embedded antibiotic resistance genes were increased or decreased in response to both doses.

Integron gene cassette open reading frames were also assigned Clusters of Orthologous Genes (COG) functional categories to investigate if macrolide exposure changed the overall functional diversity of the cassette metagenome. Only 5,206 (15%) unique open reading frames could be assigned a functional category, and of those open reading frames, 2,053 (39%) were assigned a functional category other than “function unknown” (S) (see Fig. S2 in the supplemental material). The open reading frames that were assigned to functional categories E and K (E: amino acid transport and metabolism; K: transcription) had slightly increased relative abundances (W = 3.4 ± 1.1, P = 0.03) in the high-dosed soil relative to the control, and open reading frames assigned to categories D and J (D: cell cycle control, mitosis and meiosis; J: translation, ribosomal structure and biogenesis) had slightly decreased relative abundances (W = −3.6 ± 0.8, P = 0.02) in the high-dosed soil. However, only a few open reading frames were assigned to each of these categories (EK, n = 3; DJ, n = 2).

DISCUSSION

An environmentally realistic antibiotic exposure (low dose, 0.1 mg kg⁻¹) did not affect the overall diversity (evenness or richness [α] or change in composition [β]) of antibiotic resistance genes or mobile genetic elements (Fig. 2a, b, d, and e) detected in the soil metagenome. However, three antibiotic resistance genes were more abundant in the low-dosed group than those in the control (Fig. 3). Two of the three antibiotic resistance genes that were increased, namely, mexL and mexP, are associated with chromosomal efflux pumps which confer resistance to macrolide antibiotics in Pseudomonas spp. (18, 19), and their enrichment may be the result of a direct selection for macrolide antibiotic resistance. Only one mobile genetic element variant of tnpA—a transposase that is frequently associated with antibiotic resistance genes—was detected as more abundant in the low-dosed group than that in the control (20). The enrichment of only three antibiotic resistance genes and one mobile genetic element variant in the low-dosed soil is in clear contrast to the results of soil exposed to the high dose.

The high dose (10 mg kg⁻¹) changed the overall diversity of the resistome and mobilome, as an increased number of unique antibiotic resistance genes and mobile genetic element variants were observed (Fig. 2a and b). The composition of antibiotic resistance genes and mobile genetic element variants was also altered (Fig. 2d and e, Fig. 4). Of the 21 antibiotic resistance genes that were increased in the high-dosed soil, only 2 are predicted to encode resistance to macrolides, suggesting that coselection could be involved in the enrichment of antibiotic resistance genes at this dose. All of these antibiotic resistance genes have been reported previously in Gram-negative human pathogens (21), and many are known to be associated with mobile genetic elements.

The enrichment of many antibiotic resistance genes in the high-dosed soil, but only very few genes in the low-dosed soil, is supported by a previous investigation of the same soil plots. After 5 years of annual macrolide antibiotic exposure, 10 antibiotic resistance genes were increased in the high-dosed soil as detected by quantitative PCR, but no genes were increased in the low-dosed soil (12). Five of these 10 antibiotic resistance gene targets [aph(3″)-Ib, mphE, sul1, sul2, and aac(3)] were also determined to be increased in the high-dosed soil using whole-metagenome sequence data from the present study. The coenrichment of macrolide and nonmacrolide antibiotic resistance in the high-dosed soil, but not in the low-dosed soil, is also supported by the separation of the high-dose group from the control and low-dose groups in the compositional ordination plots of antibiotic resistance genes when grouped by several nonmacrolide target drug classes (Fig. 4). The 20 mobile genetic element variants that were increased in the high-dosed soil (Fig. 5), which were mostly associated with transposases and class 1 integrons, may have played a role in facilitating the coselection of macrolide and nonmacrolide target drug classes of antibiotic resistance in the high-dose group.

Although variants of intI1, the integron-integrase gene associated with class 1 integrons, were enriched in the high-dosed soil, and although 8 of the 21 increased antibiotic resistance genes in the high-dosed soil [sul1, aac(3)-Ib, aadA, aadA15, aadA22, aadA24, dfrA15, and dfrA17] are known to be associated with class 1 integron gene cassettes (Fig. 3a) (22–24), the overall diversity of integron gene cassette open reading frames (richness, evenness, and composition) was unaffected by antibiotic exposure at either dose (Fig. 2c and f). Only one putative antibiotic resistance gene, homologous to aac(6′)-IIa (83% sequence identity), was enriched in the antibiotic-treated whole metagenome (low dose) and was also identified within integron gene cassettes. The absence of a detectable response of integron gene cassette open reading frame diversity to antibiotic exposure is most likely because the primer set that was used for gene cassette sequencing does not exclusively target class 1 integron gene cassettes (25), which are known to be biased toward encoding antibiotic resistance phenotypes (26).

The diversity of class 1 integron gene cassettes may have been altered in response to antibiotic exposure; however, if most of the amplicons that were generated from gene cassette PCR were derived from gene cassettes belonging to non-class 1 integron classes, the sequencing depth may have been insufficient to capture this treatment effect. This hypothesis is supported by the observed low recovery rate (0.04%) of antibiotic resistance genes among integron gene cassettes in this present study and by another evaluation of the similarly low recovery rate of antibiotic resistance genes by this same primer set when applied to environmental samples (27). The in silico and in vivo specificities of primer sets which exclusively target class 1 integrons have been evaluated recently for environmental samples, and these new primers will improve the resolution of the integron gene cassette metagenome in future studies of soil bacteria (26).

The overall diversity of the soil bacterial community was unaffected by antibiotic exposure at either dose; however, the relative abundances of several bacterial taxa were changed in response to either of the low or high doses. An analysis of soil bacterial taxa using 16S rRNA amplicon sequence data and whole-metagenome sequence data produced differing results, namely, the relative abundance of Cyanobacteria was decreased in the high dose soil by over 5-fold in the whole-metagenome analysis, but not in the 16S rRNA analysis. These differing results could be related to the reliance of 16S rRNA amplicon sequencing on a multicopy gene target, while MetaPhlAn excludes the use of multicopy marker genes for assigning taxonomy when at least 300 marker genes are available for a clade (28). In a comparison of these two methodologies when applied to human fecal samples, no bias was detected at the phylum level (29), although this comparison has not yet been performed for a complex environmental sample such as soil.

One of the cyanobacterial species that was decreased in the high-dosed soil, M. vaginatus, plays an important role in shaping the structure of the biocrust bacterial community, likely by engaging in a carbon-for-nitrogen exchange with nitrogen-fixing heterotrophs to colonize the surfaces of arid soils (30, 31). Cyanobacteria have recently been considered as indicator species for antibiotic pollution of aquatic ecosystems due to their sensitivity to several drug classes of antibiotics (32, 33), but this response is not uniform across all species and to all antibiotics (34, 35). The utility of cyanobacteria as indicator species for antibiotic pollution of terrestrial ecosystems, such as agricultural soil amended with biosolids, remains to be explored.

Biosolids contain numerous organic and inorganic contaminants (9, 15, 16). In the present study, we confined our interpretations to soil contaminated with only macrolide antibiotics, whereas a real-life biosolids application will contaminate soil with antibiotics belonging to multiple different drug classes and other organic pollutants. When introduced in combination with macrolides, these other chemicals could exert a selective pressure on antibiotic resistance genes that exceeds the realistic dose used in the present study. Sulfonamide and tetracycline resistance genes confer resistance to antibiotic drug classes which are detected commonly alongside macrolides in biosolids (9, 15, 16). Some of these genes (e.g., mexP and oprM) confer resistance to macrolides via cross-resistance (21), while others (e.g., sul1) could be mobilized along with macrolide resistance genes by mobile genetic elements via coselection (36). Antibiotics and other chemicals present in biosolids could also act in combination with macrolides to induce the “save our soul” (SOS) response in bacteria, thereby increasing gene cassette recombination (37) and the frequency of horizontal gene transfer (38) and potentially enriching for antibiotic resistance genes. Results from the present study were obtained with a single soil type and climate condition, and thus, the general applicability of the conclusions require verification with field experiments performed under very different conditions.

Overall, we conclude that multiyear exposure of soil bacteria to macrolide antibiotic concentrations, in the range of those found in municipal biosolids and in the absence of other antibiotics, are unlikely to significantly change the abundance or diversity of antibiotic resistance genes or mobile genetic elements of human health concern. Likewise, the composition of soil microbiomes will not be altered significantly. In contrast, contamination of soil by these antibiotics at the multiple mg soil⁻¹ concentrations will have profound impacts on the diversity of the bacterial resistome and mobilome.

MATERIALS AND METHODS

Field experiment.

Soil microplots were established at the Agriculture and Agri-Food Canada research farm in London, ON, Canada, as described in reference 17. Briefly, 12 2-m² fiberglass frames were placed into the ground and filled with a silt gray loam soil from the farm in 2010. Each spring for 10 years, these microplots were exposed to a dose of mixed macrolide antibiotics (erythromycin, clarithromycin, and azithromycin) to achieve nominal concentrations of 0.1 mg kg⁻¹ soil (n = 4) or 10 mg kg⁻¹ (n = 4) or were left unexposed as a negative control (n = 4). The 0.1-mg kg⁻¹ dose is representative of antibiotic concentrations that are at the upper limit of what could realistically result in soil from a land application of biosolids, when the biosolids contain 95th percentile concentrations of erythromycin (0.12 mg kg⁻¹, min = 0.002, max = 0.18), clarithromycin (0.17 mg kg⁻¹, min = 0.009, max = 0.62), and azithromycin (3.2 mg kg⁻¹, min = 0.008, max = 5.2) (15, 16) and the biosolids are applied at a rate of 1% to 10% (dw dw⁻¹ soil) and thereby reflect an antibiotic dilution factor of 10- to 100-fold. Stock solutions of erythromycin, clarithromycin, and azithromycin were prepared to 1 mg mL⁻¹ in 100% ethanol and stored at −20°C until used. Each June, antibiotics were mixed into 1 kg of soil obtained from each plot, and soil was reincorporated into the source microplots to a depth of 10 cm using a mechanized rototiller. Control plots received the same management. Soybean (Glycine max, var. Harosov) seeds were planted within 2 h after the antibiotics were added, and plots were maintained throughout the growing season by manual weeding only. In 2019, six 20-cm-deep soil core samples were obtained 30 days after the antibiotic application, pooled, and then sieved to a maximum particle size of 2 mm. Soil was stored at −20°C prior to the isolation of DNA used in the present study.

DNA isolation, PCR, and library preparation.

For targeted amplicon sequencing (16S rRNA, integron gene cassettes), 250 mg of soil was used as input for DNA isolation with the DNeasy PowerSoil kit (Qiagen, Hilden, Germany), and a final elution volume of 100 μL was used following the manufacturer’s protocol. For integron gene cassette amplicon sequencing only, total genomic DNA was isolated in technical duplicates which were processed separately. For whole-metagenome sequencing, 10 g of soil was used as input for DNA isolation with the DNeasy PowerMax soil kit (Qiagen), and a final elution volume of 2 mL was used following the manufacturer’s protocol. Quality (A260/A280) of the eluted DNA was determined using an ND1000 microspectrophotometer (NanoDrop Technologies, Wilmington, DE) and a Qubit double-stranded DNA (dsDNA) high-sensitivity (HS) assay kit was used to determine DNA concentration with a Qubit 4 Fluorometer (Thermo Fisher Scientific, Toronto, ON, Canada). DNA was stored at −20°C.

Next-generation sequencing.

The following three types of sequencing were performed on the total genomic DNA soil samples: 16S rRNA amplicon sequencing to investigate bacterial taxonomy; whole-metagenome sequencing to investigate the resistome and mobilome, as well as bacterial taxonomy; and integron gene cassette amplicon sequencing to investigate putative genes within environmental integron gene cassettes.

For 16S rRNA amplicon sequencing, 12.5 ng of total genomic DNA was used as the template for PCR amplification of the V3 and V4 regions of the bacterial 16S rRNA gene by following the Illumina workflow (https://support.illumina.com/content/dam/illumina-support/documents/documentation/chemistry_documentation/16s/16s-metagenomic-library-prep-guide-15044223-b.pdf) to generate 16S rRNA amplicons with overhang sequences (Table 2). The 16S rRNA amplicons were purified subsequently using AMPure XP magnetic beads (Beckman Coulter, Mississauga, ON, Canada), and unique dual-indices (i.e., sample-specific molecular barcodes) and sequencing adapters were then added to the overhang sequences using the Nextera XT Index kit v2 (Illumina, Vancouver, BC, Canada). The indexed DNA libraries were again purified using AMPure XP magnetic beads (Beckman Coulter), quantified using the Qubit dsDNA HS assay kit, and sized using the high-sensitivity DNA kit on a 2100 bioanalyzer (Agilent, Mississauga, ON, Canada). The indexed libraries were then normalized and pooled to a final concentration of 4 nM for 2 × 300-bp sequencing on a MiSeq instrument using the MiSeq reagent kit v3 (600-cycle; Illumina) with a 5% PhiX spike-in.

TABLE 2.

Information for 16S rRNA and integron gene cassette PCR amplification^a

Target	Primer	Sequence (5′ → 3′)	Amplicon size (bp)	Annealing temp (°C)	Referenceb
16S rRNA	Adapter + S-D-Bact-0341-b-S-17	TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGCCTACGGGNGGCWGCAG	∼460	55	68
	Adapter + S-D-Bact-0785-a-A-21	GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGACTACHVGGGTATCTAATCC
attC	Adapter + HS286	TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGTCSGCTKGARCGAMTTGTTAGVC	Variable	55	25
	Adapter + HS287	GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGCSGCTKANCTCVRRCGTTAGSC

^aDegenerate bases follow the IUPAC standard ambiguity code. Underlined text indicates the locus-specific binding sequence, while nonunderlined text indicates the location of Illumina adapter overhang sequences.

^bReference for primer sequence.

For whole-metagenome sequencing, total genomic DNA was sent to The Centre for Applied Genomics (TCAG) at SickKids Hospital (Toronto, ON, Canada) for Nextera XT library preparation and Illumina high-throughput sequencing. Only three of four biological replicates of total genomic DNA (nine samples in total, n = 3) were used to prepare DNA libraries for whole-metagenome sequencing using the Nextera XT DNA library preparation kit and following the manufacturer’s protocol. The DNA libraries were indexed using the Nextera XT index kit v2 after the tagmentation step. Finally, the indexed libraries were diluted to the same DNA concentration and then pooled to a final concentration of 2.15 nM for 2 × 150-bp sequencing across two duplicated lanes on a NovaSeq 6000 instrument using an SP flow cell (Illumina, Vancouver, BC, Canada).

For integron gene cassette amplicon sequencing, integron gene cassettes were PCR amplified using primers HS286 and HS287 (25) with 33- and 34-bp Illumina adapter overhang sequences attached to the 5′ ends (Table 2). The purpose of these 5′ ends was to extend the distance between the tagmentation site and the desired gene cassette sequence since as many as 50 bp from each distal end of the amplicon could be expected to be lost due to transposome activity during library preparation. The gene cassette PCR primers anneal to highly conserved GTTRRRY motifs within the attC sites of gene cassettes. Total genomic DNA was diluted 10-fold in Tris-EDTA buffer and used as the template DNA for five technical replicates of 25-μL PCRs (125 μL total). Integron gene cassettes were amplified under the following thermocycler conditions: 94°C for 3 min; 35 cycles of 94°C for 30 s, 55°C for 1 min, and 72°C for 2 min 30 s; and 72°C for 5 min. Each PCR was comprised of 2 μL of diluted template DNA, 0.5 units of Q5 high-fidelity DNA polymerase (New England BioLabs, Mississauga, ON), 0.2 μL of 25 mM deoxynucleoside triphosphates (dNTPs), 5 μL of 5× Q5 reaction buffer, and 1.13 μL of each 10 μM forward and reverse primer. Technical replicates were pooled, and the combined PCR product was purified using the Geneaid GenepHlow PCR cleanup kit (FroggaBio, Concorde, ON, Canada) and was eluted in 25 μL of nuclease-free water. DNA concentrations of the cleaned PCR products were determined using the Qubit dsDNA HS assay kit. The Nextera XT DNA library preparation kit was used to prepare DNA libraries of the gene cassette amplicons by following the manufacturer’s protocol. The DNA libraries were indexed using the Nextera XT index kit v2 following the tagmentation step. The indexed DNA libraries were purified, quantified, and sized as described for 16S rRNA amplicon sequencing, except using a 0.8× bead:sample ratio (vol/vol) for purification. The indexed libraries were pooled and sent to TCAG, where the pooled library was denatured and diluted to achieve a final concentration of 15 pM prior to 2 × 125-bp sequencing in one lane (shared with other samples) on a HiSeq 2500 instrument (Illumina) using a high-throughput flow cell.

Sequence data analysis.

For all sequence data sets, the quality of the demultiplexed reads was assessed using FastQC (v0.11.8) and MultiQC (v1.7) and then reassessed after adapter removal and trimming (when applicable) (39, 40).

To remove primer sequences from 16S rRNA amplicon reads, cutPrimers (downloaded 9 September 2019) was used while allowing 3 errors in either forward or reverse primers sequences and keeping reads at >250 bp (41). To remove low-quality bases from the resulting amplicon reads, Trimmomatic (v0.36) was used in paired-end mode with the following parameters: the first 25 bases of each read were dropped, sliding window trimming was performed where a window of 4 bp would be trimmed if the average quality of the window had a quality score (Q-score) of <15, and trimmed reads with a length <25 bp were discarded (42). Finally, the DADA2 denoise-paired plugin within QIIME 2 (v2019.10) was used with default options and without further truncation of the 5′ and 3′ ends to correct sequencing errors in the trimmed 16S rRNA amplicon reads, remove chimeric sequences, merge paired reads, and then establish a set of unique amplicon sequence variants and obtain their read counts (43, 44). To assign taxonomy to the amplicon sequence variants, naive Bayes classification implemented in the QIIME 2 feature-classifier plugin was used, after first being trained on SILVA (v132) 16S rRNA gene reference sequences (45, 46).

Cutadapt (v2.8) was used to remove adapter sequences from the 3′ ends of whole-metagenome sequence reads (47). Trimmomatic was run in paired-end mode to remove low-quality leading and trailing bases from the adapter-trimmed reads (Q-score of <20), and remaining reads with a length <100 bp were discarded. MetaPhlAn3 (v3.0.7) was used to assign taxonomy to whole-metagenome reads using the “very-sensitive” algorithm of Bowtie 2, ignoring eukaryotes and archaea, and profiling the bacterial taxonomy of the metagenomes using relative abundances with estimations of the number of reads coming from each bacterial taxon (48).

To identify whole-metagenome reads that corresponded to antibiotic resistance genes, the metagenomic reads were mapped to two Comprehensive Antibiotic Resistance Database (CARD) databases as follows: the CARD “canonical” database (v3.0.8) of phenotypically confirmed antibiotic resistance genes and the CARD prevalence, resistomes, and variants (v3.0.7) database of in silico-predicted resistance genes, derived from genomic data of 82 human pathogens (21). The metagenomic reads were mapped to these databases using Bowtie 2 implemented by the CARD resistance gene identifier (CARD-RGI) in metagenomics mode (v5.1.0) (21). To identify whole-metagenome reads that corresponded to mobile genetic elements, the metagenomic reads were mapped to a database of known mobile genetic elements (downloaded online from https://github.com/KatariinaParnanen/MobileGeneticElementDatabase on 24 May 2021) (49). The metagenomic reads were mapped to this database using Bowtie 2 (v2.4.2) with end-to-end searching and using the predefined very-sensitive search algorithm, except for allowing a maximum of one mismatch in the seed alignment (50). All sequences corresponding to variants of the biocide resistance gene qacEΔ1, which is a part of the 3′-conserved sequence of clinical class 1 integrons (51), were removed manually from the mobile genetic element database. Each variant in the database was referenced here by the name of the mobile genetic element, followed by the record number in the FASTA file of the database, separated by a period. Fold coverages of antibiotic resistance genes and mobile genetic elements were computed by dividing the total number of mapped base pairs by the length of the reference sequence and were used as relative abundances for subsequent analyses.

Cutadapt was used to remove adapter sequences from the 3′ ends of the integron gene cassette sequence reads. To preserve primer-binding sites at the ends of the reads for downstream filtering, no quality-based trimming was performed. Integron gene cassette reads were assembled into contigs using MEGAHIT (v1.2.9) with default options (52).

Assembled gene cassette contigs from all samples were combined into one file for downstream filtering of gene cassettes. First, the boundaries of gene cassettes from assembled contigs were identified based upon the presence of two highly conserved motifs within integron gene cassettes (i.e., attC sites); BBDuk (v38.90) within BBTools was used identify assembled contigs that did not contain the terminal 9 bp of both of these motifs, and they were discarded from further analysis (53). Next, Prokka (v1.14.6) was used to identify open reading frames within the curated gene cassette contigs, which were then clustered at 97% identity using CD-HIT (v4.8.1) to obtain a set of unique open reading frames (54, 55).

To identify possible antibiotic resistance genes within integron gene cassette sequences, the open reading frames were aligned against the CARD canonical protein homolog database using an implementation of BLAST within CARD-RGI, including all “loose” hits and running in low-quality mode. To further potentiate the discovery of novel antibiotic resistance genes, the open reading frames were compared to high-confidence (group I) antibiotic resistance profile hidden Markov models using hmmer (v3.1b2) as implemented by Meta-MARC (downloaded online from https://github.com/lakinsm/meta-marc on 13 February 2021) (56, 57). Integron gene cassette open reading frames were classified as antibiotic resistance genes based upon high-confidence alignments as follows: CARD-RGI hits with bit scores of ≥150 and Meta-MARC hits with bit scores of ≥200.

To predict general functions of gene cassettes, the integron gene cassette open reading frames were compared to orthologous groups in the eggNOG database (v5.0) and assigned one or more Cluster of Orthologous Groups (COG) functional categories using the Web implementation of eggNOG mapper (v2.0) (58).

Sequence reads from each sample were mapped to all unique integron gene cassette open reading frames and were normalized by open reading frame length and sequencing effort to obtain fold coverages using BBMap (v38.90) within BBTools (53). These fold coverages were used as relative abundances for subsequent analyses.

Statistical analysis and data visualization.

All statistical tests were performed using statistical packages for Python (v3.9.2) unless otherwise stated and used a P value of <0.05 as a cutoff (59). Data visualizations were generated using plotly (v4.14.3) and matplotlib (v3.4.1) packages (60, 61) implemented in Python and using ggplot2 (v3.3.5) implemented in R (v4.1.2) (62, 63).

Alpha diversity (within-group diversity) was estimated using the Chao1 richness estimator, Shannon diversity index, and Simpson’s index with the scikit-bio package (v0.5.6) (64). A Shapiro-Wilk test was used to assess the normality of alpha diversity, followed by a one-way analysis of variance (ANOVA) for parametric data or a Kruskal-Wallis test for nonparametric data to test if differences in alpha diversity between treatment groups were statistically significant, as implemented by the SciPy package (v.1.6.2) (65).

To assess beta diversity, a center log ratio (CLR) transformation of the relative abundances of genes and taxa was used to obtain CLR-transformed relative abundances. Next, a principal-component analysis (PCA) of the CLR-transformed Aitchison distances—the Euclidean distances between CLR-transformed relative abundances—was performed to investigate differences in the composition of bacterial taxa, antibiotic resistance genes, mobile genetic elements, and integron gene cassette open reading frames between treatment groups. CLR-transformed Aitchison distances are a statistically robust approach for comparing the proportions of genes and taxa between different samples within the same high-throughput sequencing data set (66). To avoid an error due to calculating the logarithm of zero (where zero reads were obtained for a taxon, gene, or open reading frame from a sample), a pseudocount of 0.5 was added to all relative abundances prior to CLR transformation. A log base 2 transformation was performed so that differences in CLR-transformed Aitchison distances represent fold changes. The scikit-learn package (v0.24.1) was used to perform the PCA, and a permutational multivariate ANOVA (PERMANOVA) was used to determine if differences in the dispersion between treatment groups were statistically significant using the sci-kit bio package with 999 permutations.

Statistically significant differences in the abundances of bacterial taxa, antibiotic resistance genes, mobile genetic elements, or gene cassette COG functional categories between groups were determined using analysis of composition of microbiomes with bias correction (ANCOM-BC; v1.2.0) with Bonferroni correction as implemented in R (67). The results of these differential abundance analyses were reported as effect sizes (fold changes [W]) ± Bonferroni-adjusted 95% confidence intervals. A one-way ANOVA was used to test if differences in the numbers of merged 16S rRNA amplicon reads between treatment groups were statistically significant as implemented by the SciPy package.

Data availability.

All biological sequence data are accessible on the NCBI server under BioProject identifier (ID) PRJNA804921. Nucleotide sequences for 16S rRNA amplicon sequence data were submitted to the SRA under SRR17965709 to SRR17965720, whole-metagenome sequence data under SRR17965547 to SRR17965555, and integron gene cassette amplicon sequence data under SRR17966098 to SRR17966121.