Extracellular vesicles drive stress-induced antibiotic resistance spread in soil

TL;DR

Mixtures of artificial sweeteners in soil increase antibiotic resistance by boosting extracellular vesicles that carry resistance genes.

Contribution

This study reveals that diverse sweeteners drive antibiotic resistance through stress-induced extracellular vesicles in soil.

Findings

Artificial sweetener mixtures enrich ARGs in extracellular vesicles, not in the bulk soil microbiome.

EVs originate from Pseudomonadota that upregulate vesicle-biogenesis genes under stress.

EVs transfer phenotypic resistance to recipient Escherichia coli.

Abstract

Antimicrobial resistance threatens millions of lives annually, yet its acceleration by non-antibiotic pollutants remains poorly understood. Artificial sweeteners, now ubiquitous in soils and waters, are known individually to promote conjugative transfer of resistance genes, but real environments contain complex mixtures whose collective impact is unknown. Extracellular vesicles (EVs) released by stressed bacteria serve as protected, long-range vectors for antibiotic resistance genes (ARGs), yet whether sweetener diversity modulates this pathway has never been tested. Here we show that increasing artificial-sweetener diversity dramatically enriches ARGs, virulence factors and mobile genetic elements inside soil-derived Evs, driving compositional shifts in 30.5% of EV-associated genera while leaving the bulk microbiome largely undisturbed. EVs originate from a small, fast-growing Pseudomonadota subset that upregulates vesicle-biogenesis genes in response to oxidative and membrane stress; these vesicles selectively package chromosomal resistance traits and transfer phenotypic resistance to recipient Escherichia coli. This stress-induced decoupling reveals EVs as rapid, hidden mediators of resistome mobilization that community-level surveys miss. By demonstrating that pollutant diversity itself drives resistance dissemination through nanoscale vectors, our findings establish EVs as a critical new indicator within the One Health framework and call for revised environmental risk models that account for chemical complexity rather than single-compound exposures.

Image 1

•Artificial sweetener mixtures promote antibiotic resistance spread in soil.•Resistance genes enrich in extracellular vesicles, not in bulk soil.•Sweeteners activate EV biogenesis genes in key responsive microbes.•Resistant bacteria increase and interact more, increasing EVs' antibiotic resistance profile.

Artificial sweetener mixtures promote antibiotic resistance spread in soil.

Resistance genes enrich in extracellular vesicles, not in bulk soil.

Sweeteners activate EV biogenesis genes in key responsive microbes.

Resistant bacteria increase and interact more, increasing EVs' antibiotic resistance profile.

Figures

Full text

Introduction

1

Antimicrobial resistance poses an escalating global health threat, with projections estimating up to 39 million deaths annually by 2050 without effective interventions [1]. A key mechanism underlying this threat is the horizontal transfer of antibiotic resistance genes (ARGs) to pathogenic bacteria, especially those also carrying virulence factor genes (VFGs)—a combination that enables pathogens to evade antibiotic treatments and heightens public health risks. Recent research has identified microbial extracellular vesicles (EVs) as novel, underappreciated vehicles for ARG dissemination [2,3]. These nanoscale, membrane-bound particles, secreted by bacteria in diverse habitats [4,5], can carry DNA, RNA, and proteins across species and even kingdoms [6,7]. EVs isolated from indoor dust, soil, and wastewater have been found to carry diverse ARGs and mobile genetic elements (MGEs) [5]. Metagenomic evidence from marine samples has revealed novel ARGs and transposons in EVs [8,9]. Unlike classical horizontal gene transfer (HGT) mechanisms, such as conjugation, transformation, and transduction, EVs may selectively package and protect specific bioactive molecules, preserving their activity even under harsh or spatially segregated conditions [6,[10], [11], [12]]. This enables genetic exchange in non-growing, physically separated, and even interkingdom contexts, dramatically expanding the ecological scope of gene flow and microbial adaptation [13,14].

While the overuse and misuse of antibiotics are well-established drivers of ARG proliferation, growing evidence highlights the role of nonantibiotic pollutants in promoting resistance [[15], [16], [17]]. Among these, artificial sweeteners (ASs) are of particular concern due to their large-scale production, chemical persistence, and environmental mobility [18]. Global AS consumption has risen steadily, reaching over 146,000 metric tons in 2016 [19], driven by dietary shifts and the demand for low- or zero-calorie products. Due to their poor degradability and strong environmental persistence, ASs are consistently detected in various ecosystems, including surface waters, effluents, and soils, often at concentrations exceeding 10 mg L^−1^ [20,21]. The surface soil environment has been confirmed as a critical reservoir for persistent ASs through quantitative monitoring. For example, Gan et al. reported that sucralose (SUC), saccharin (SAC), cyclamate (CYC), and acesulfame (ACE) are frequently detected in topsoil, with total residual concentrations spanning a wide range, from 0.01 to 1280 μg kg^−1^ [22]. These persistent residues exert selective pressure on microbial communities, reshaping their composition and functional potential. This selective pressure is known to promote HGT, thereby facilitating the mobilization of ARGs [[23], [24], [25]]. EVs are increasingly recognized as a key vector for HGT, mediating processes such as gene exchange [26,27], stress response [28], and interspecies communication [26,27]. However, to our knowledge, no studies have yet investigated whether ASs specifically utilize this EV-mediated pathway to promote ARG dissemination.

Currently, most research focuses on the effects of individual ASs, neglecting the ecological complexity of real-world environments, in which specific ASs are rarely found in isolation but often coexist at relatively low concentrations. Previous studies have shown that combined environmental stressors can synergistically promote the accumulation and dissemination of ARGs [28]. However, little is known about how EVs respond to such compound pressures.

Here, we hypothesize that the increasing diversity of ASs in the environment may promote microbial adaptation by altering the production and molecular composition of EVs, thereby increasing the abundance of ARGs within EVs and posing potential health risks. To test this hypothesis, we selected seven ASs commonly found in the environment and constructed an exposure system with four levels of AS diversity—0, 1, 3, and 6 different types—keeping the overall concentration of AS constant across treatments. This study aims to (1) characterize ARG, VFG, and MGE profiles in EVs isolated from soils exposed to increasing AS diversity, (2) identify EV-producing microbial hosts and compare their genetic and functional traits with bulk soil microbiota under varying AS exposures, and (3) uncover the potential mechanisms through which AS diversity influences the antibiotic resistome in EVs. Our findings illuminate how AS diversity affects the antibiotic resistome of soil EVs and the associated health risks, offering a new perspective on the role of sweeteners in driving ARG dissemination via EVs and further enhancing the framework for assessing resistance risks in environmental settings.

Materials and methods

2

Soils and ASs

2.1

Soil samples were collected from farm fields in Lingnan Village, Ningbo City, Zhejiang Province, China (29.87° N, 121.54° E), sieved through a 2-mm mesh immediately after collection, and transported to the laboratory for further analysis. The soil had a pH of 7.8 and a total organic carbon concentration of 10.7 mg L^−1^ (Supplementary Table S1). Seven ASs were selected based on their high global consumption, frequent environmental detection, and persistence [22,29]. These were purchased from Shanghai Yuanye Bio-Technology Co., Ltd. (Shanghai, China) and included SAC, CYC, aspartame (ASP), ACE, SUC, neotame (NEO), and neohesperidin dihydrochalcone (NHDC).

Microcosm experiment

2.2

We conducted the experiment in 2 L circular plastic pots (13.5 cm diameter × 12 cm height), each containing 1 kg of substrate soil. Four levels of AS diversity were established: 0, 1, 3, and 6 AS types. Each treatment level included seven replicates (Supplementary Fig. S1). For the single-AS treatment, each replicate received one of the seven AS types (i.e., SAC, CYC, ASP, ACE, SUC, NEO, and NHDC). For the three-AS treatment, we created seven distinct combinations of ASs, ensuring that each AS type appeared in three different combinations. Similarly, for the level with six AS types, seven distinct combinations were generated, with each AS type appearing in six combinations. The total AS concentration was set at 1.2 mg kg^−1^ across all diversity levels, consistent with previously reported environmental concentrations [23,30]. For the single-AS treatment, AS was applied at 1.2 mg kg^−1^. For the three-AS treatment, each compound was applied at 0.4 mg kg^−1^ (1.2/3), and for the six-AS treatment, each compound was applied at 0.2 mg kg^−1^ (1.2/6). The ASs were thoroughly mixed with the soil substrate in the pots, and seven pots were prepared for each diversity level. Soil moisture was maintained at 60% field capacity, and the pots were cultured at room temperature for four weeks. After the experiment, the soil samples were collected and divided into three portions: one for physicochemical analysis, one for DNA extraction, and one for EV extraction.

Isolation of EVs from soil

2.3

EVs were extracted from soil samples following a previously established protocol [5]. Briefly, 500 g of soil was incubated in phosphate-buffered saline (PBS) at a 1:5 (w/v) ratio, with gentle rotation at 180 rpm for 8 h at 10 °C. The resulting mixture was then centrifuged at 8000×g for 20 min at 4 °C (CR22GII/Hitachi, Tokyo, Japan), followed by filtration through a series of mixed cellulose esters membrane filters with pore sizes of 8, 3, 0.8, 0.45, and 0.22 μm (Xinya, Shanghai, China). To concentrate the filtrate, tangential flow filtration was performed using a 100 kDa molecular weight cutoff (MWCO) filter (Pellicon 2 Cassette, Ultracel 100 kDa; Millipore, Billerica, MA, USA), reducing each sample volume to 200 mL. The pretreated filtrates were then centrifuged at 12,000×g for 45 min at 4 °C, followed by ultracentrifugation at 120,000×g for 90 min at 4 °C using an Optima XPN-80 ultracentrifuge (Beckman Coulter, Brea, CA, USA) to pellet the EVs.

Although no purification method guarantees absolute purity, we achieved relatively high-purity EVs by employing gradient density centrifugation, a technique that has proven effective for removing viruses and contaminating proteins [4,31]. We prepared a series of iodixanol (Sigma-Aldrich, St. Louis, MO, USA) density gradients using a buffer containing 3.6% (w/v) NaCl and 10 mmol L^−1^ HEPES (pH 7.5), resulting in gradients of 45%, 40%, 35%, 30%, 25%, 20%, and 15% iodixanol. The gradients were carefully layered into ultracentrifuge tubes, and the EV-enriched fraction was placed on top. Samples were then ultracentrifuged at 120,000×g for 16 h at 4 °C. Following centrifugation, we collected fractions with a density of 1.10–1.19 g mL^−1^, diluted them fivefold with PBS, and subjected them to a final round of ultracentrifugation at 120,000×g for 90 min at 4 °C to pellet the EVs. EV samples were treated with 2 U Turbo DNase (Invitrogen, Carlsbad, CA, USA) in 1 × DNase reaction buffer (Invitrogen) for 30 min to effectively remove free DNA.

Characterization of EVs from soil

2.4

The size distribution and total particle concentration of the EVs were analyzed using a nanoflow cytometer (N30, NanoFCM, Xiamen, Fujian, China). Prior to analysis, we diluted EVs with PBS to an appropriate concentration and performed three independent measurements for each sample. A silica nanosphere cocktail, used as the size reference, contained a mixture of nanospheres with diameters ranging from 68 to 155 nm. In addition, EV morphology was examined using transmission electron microscopy, following a previously established protocol [32,33]. In brief, EVs were fixed in 4% paraformaldehyde (Sigma-Aldrich) in 0.1 mol L^−1^ PBS (pH 7.4) and placed onto a copper mesh. Imaging was carried out using a transmission electron microscope (H-7650; Hitachi, Tokyo, Japan).

EV-recipient bacteria co-culture assay

2.5

For the co-culture assay with a pure bacterial strain, antibiotic-sensitive Escherichia coli DH5α was selected as the recipient. A total of 100 μL of E. coli DH5α (∼10^5^ colony-forming units) was co-incubated with 100 μL of EVs (∼10^8^ particles), which had been pretreated with DNase I (1 U) and proteinase K (20 μL, 20 mg mL^−1^; Qiagen, Hilden, Germany) to eliminate free DNA and surface proteins. The mixture was incubated at 37 °C for 4 h. Subsequently, the culture volume was increased to 1 mL using fresh medium and incubated for an additional 4 h at 37 °C with shaking at 220 rpm. To allow for further growth, 9 mL of Luria–Bertani (LB) medium was added, and the culture was maintained overnight under the same conditions. Bacterial cells were then harvested by centrifugation (12,000 rpm, 1 min, 4 °C) (LX-185T100R; Haier Biomedical, Qingdao, China), washed, resuspended in 1 mL PBS, and subsequently cultured under both antibiotic-selective and nonselective conditions for downstream analyses.

For the co-culture assay with soil microbial communities, we first extracted total soil bacteria from 1 g of soil using iodixanol-based density gradient centrifugation. The extracted microbial suspension was then incubated with EVs at a dose equivalent to the EV content found in 1 g of soil (∼10^8^ particles). The co-culture procedure was identical to that used in the E. coli assay. After incubation, total DNA was extracted from the cultures and analyzed using high-throughput quantitative PCR (HT-qPCR; LightCycler 480 II; Roche Diagnostics, Mannheim, Germany) to assess changes in ARG abundance.

DNA extraction, metagenome sequencing, and quality filtration

2.6

We used the FastDNA Spin Kit (MP Biomedicals, Irvine, CA, USA) to extract soil DNA according to the manufacturer's protocol. Approximately 0.5 g of soil was weighed, lysed, purified, and eluted to yield total genomic DNA. For EV DNA extraction, a QIAamp® DNA Mini Kit (QIAGEN, Hilden, Germany) was used according to the manufacturer's instructions. The DNA concentration was quantified using a Qubit 3.0 Fluorometer (Thermo Fisher Scientific, Waltham, MA, USA), and its purity was assessed by electrophoresis on a 1.0% agarose gel. The extracted DNA was stored at −20 °C until further analysis.

Metagenomic analysis of the extracted DNA was performed by Majorbio Bio-Pharm Technology Co., Ltd. (Shanghai, China). Genomic DNA was processed using the NEBNext Ultra DNA Library Prep Kit for Illumina (New England Biolabs, Ipswich, MA, USA) according to the manufacturer's instructions. Sequencing was performed on the Illumina HiSeq4000 platform using the PE150 strategy. After quality control, approximately 8 Gb of trimmed 150-bp paired-end reads (sequencing depth ≈ 10 Gb) were generated per sample. Quality control included trimming with Fastp (v0.20.0) and filtering out reads containing >10% ambiguous nucleotides ('N') or >50% low-quality nucleotides (quality score <10) using Trimmomatic (v0.33).

Metagenomic assembly and bioinformatics analysis

2.7

High-quality metagenomic reads were assembled into contigs using MEGAHIT (v1.2.9) with default settings [34]. Contigs longer than 800 bp were processed to predict open reading frames with Prodigal (v2.6.3) [32], and a nonredundant gene catalog was generated with CD-HIT (v4.8.12) [35] using 95% identity and 90% coverage thresholds. The quality-controlled sequencing reads were then mapped to the nonredundant gene catalog using Bowtie2 (v2.4.2) [36], and gene abundance was estimated using SAMtools. To explore the microbial composition, nonredundant genes were annotated by searching the National Center for Biotechnology Information non-redundant protein database via BLASTp in DIAMOND with an e-value threshold of <1 × 10^−10^ [17]. The minimal doubling time for taxa was estimated by gRodon v.1 [37]. Kyoto Encyclopedia of Genes and Genomes (KEGG) functional annotations were performed using DIAMOND (v.2.0.14) against the KEGG database, with an e-value cutoff of 1 × 10^−10^ [38]. To identify ARGs, nonredundant genes were compared to the Structured ARGs (SARG) database using BLASTX with an e-value ≤1 × 10^−10^ and identity ≥80% [39]. Gene abundance was estimated using the ARGs-OAP 3.0 pipeline and reported as reads per kilobase per million mapped reads. PlasFlow (v1.1) was used to determine whether ARG-carrying contigs originated from plasmids or chromosomes [40]. High-risk ARGs were identified based on the list compiled by Zhang et al. [41]. VFGs were identified by searching the Virulence Factor Database (VFDB) [42]. For further MGE classification, we used the mobileOG-db80 database to assign labels to elements, including transposable elements (identified by ISfinder), integrative elements (e.g., integrases and transposases), and conjugative elements (identified by conjugation machinery markers). Candidate genes were annotated based on an e-value <10^−7^ and identity >80% [43]. Binning of metagenomic contigs was performed using MetaWRAP v.1.2.1, applying three binning methods: CONCOCT (v.0.4.0) [44], MaxBin (v.2.2.2) [45], and MetaBAT (v.2.12.1) [46]. The quality of the derived metagenome-assembled genomes (MAGs) was assessed using CheckM (v.1.0.12) [47], retaining only those with ≥50% completeness and ≤10% contamination for further analysis. To estimate the relative abundance of MAGs in each sample, quality-controlled reads were mapped back to all MAGs using Salmon. The resulting abundances were normalized to genome length and sequencing depth and expressed as transcripts per million. Taxonomic annotation and phylogenetic tree construction were conducted using GTDB-Tk (v.2.1.1) [48], with visualization and annotation performed in ChiPlot.

High-throughput qPCR for the detection of ARGs

2.8

To assess ARG profiles, HT-qPCR was performed on a SmartChip Real-Time PCR System (WaferGen Inc., Fremont, CA, USA), with each sample analyzed in six technical triplicates. Amplification conditions followed published protocols [49]. The system targeted 185 ARGs and the 16S rRNA gene (see the full list in Supplementary Table S2). Amplicon specificity was confirmed by melt curve analysis. Data was processed with SmartChip qPCR software (v2.7.0.1), applying a threshold cycle (Ct) value of 31 as the limit of detection. Gene abundance was estimated using the following formula:

$$[eqn]$$

where N represents the relative gene copy number and $[eqn]$ represents the standard PCR amplification efficiency.

Statistical analysis

2.9

We performed all analyses in R (v.4.2.1). Using the vegan package (v.2.6-4), we quantified species diversity in MAGs from microbial and EV samples as richness. Community dissimilarities (ARGs, the overall microbial community, and the ARG–VFG-containing microbial community) were computed using Bray–Curtis distances and visualized by principal coordinates analysis in the vegan package (v.2.6-4). PERMANOVA calculations were conducted to analyze the explanatory power of different treatments on the sample variance using the vegan package (v.2.6-4).

Additional statistical analyses were conducted in IBM SPSS (v.22.0), including nonparametric Kruskal–Wallis test, Tukey's Honestly Significant Difference test, and two-tailed unpaired Student's t-test, as appropriate. For differential analyses involving multiple comparisons (e.g., comparison of multiple genes or functional pathways between groups), the resulting p-values were adjusted using the Benjamini–Hochberg method to control the false discovery rate; features with adjusted p-values were less than 0.05. Changes in risk ARGs across treatment groups were examined using linear discriminant analysis of effect sizes, using an LDA score (log_10_) threshold of 2.

Correlations were constructed using linear fitting of the data with the ggplot2 package (v.3.5.0). Mantel test (999 permutations) in the vegan package (v.2.6-4) was used to analyze correlations between ARG abundance and the MGE/VFG distance matrix. Procrustes analysis in the vegan package (v.2.6-4) was used to determine the correlation between ARG communities and bacterial communities/functional features. A co-occurrence network of ARG-containing microbes was constructed using the ggClusterNet package (v.0.1.0) (Spearman's r > 0.8, adjusted p < 0.05), and we calculated network topological indices.

Results

3

AS diversity selectively alters the EV-associated microbial community and upregulates vesicle biogenesis genes

3.1

The community structure of soil and EV-associated DNA differed significantly (p < 0.05, Fig. 1a). Pseudomonadota was the dominant phylum in both EV potential hosts and the bulk microbial community, comprising 69.6% and 43.9% of their respective populations. However, the EV-associated genera were markedly distinct from those of the source microbiota. The top 10 genera accounted for 83.5% of EV DNA but only 11.3% of the source microbiota (Supplementary Fig. S2), suggesting a compositional shift and potential selective enrichment in EV-producing taxa. Genomic analyses further revealed that EV-producing bacteria had significantly larger genome sizes and lower guanine-cytosine (GC) content compared to the broader soil community (Fig. 1b and c).Fig. 1Community composition and functional traits of soil microorganisms and extracellular vesicles (EVs). a, Principal coordinates analysis (PCoA) based on Bray–Curtis distances showing antibiotic resistance gene (ARG) distribution patterns in soil microbial and EV samples under different artificial sweetener (AS) diversity treatments. Ellipses represent the 95% confidence intervals. b–c, Contig-derived estimates of genome fragment size (b) and guanine-cytosine (GC) content (c) for the overall soil microbiome and EV-associated microbial DNA. Error bars represent the standard deviation. Asterisks denote the significance level based on *t-*tests: ∗∗∗p < 0.001. d, Proportion of genera exhibiting significant (Sig, p < 0.05) versus non-significant (In_sig) changes in relative abundance in the soil microbiome and EV-associated communities across AS diversity levels. e, Heatmap showing the abundance of EV biogenesis-related genes identified in the overall soil microbiome and EV-associated microbial hosts under AS diversity exposure. Abundance is normalized across samples and presented as a z-score. f, Phylogenetic tree displaying EV-releasing phyla that exhibited significant increases in response to AS treatments. The tree was constructed using the Genome Taxonomy Database taxonomy and visualized with ChiPlot. The accompanying heatmap shows the relative abundance (z-score) of each genus across different samples. M_0, M_1, M_3, and M_6 refer to soil microbial DNA samples collected under treatments with 0, 1, 3, and 6 types of ASs, respectively, while E_0, E_1, E_3, and E_6 represent EV-associated DNA samples derived from the corresponding soil treatments.Fig. 1

Exposure to increased AS diversity had a limited effect on the bulk soil microbiome but substantially altered the taxonomic composition of EV-associated DNA. Among the 3037 genera detected in soil, only 1.2% (36 genera) showed significant changes across treatments. In contrast, 30.5% (874 out of 2860 genera) of EV-associated genera exhibited significant compositional shifts between groups (Fig. 1d), suggesting that EV-producing populations are more sensitive to AS-induced environmental pressures. Of the altered EV-associated genera, 72 showed consistent increases with rising AS diversity (Fig. 1f). Notably, the taxa with the highest relative DNA abundance in EVs at both the genus and species levels showed significant increases, particularly Pseudomonas (41.6% in EVs vs. 0.5% in soil) and Rheinheimera sp. LHK132 (4.9% vs. less than 0.1%), both of which were substantially enriched in EVs, with enrichment increasing with AS diversity. In addition, the abundance of EV biogenesis-related genes, curated from previous studies [50], increased with AS diversity, specifically in EV-hosting bacteria, but not in the overall microbiome (Fig. 1e). Furthermore, doubling time analysis showed that EV-hosting bacteria grew significantly faster with increased AS diversity, while the broader community showed no such response (Supplementary Fig. S3).

AS diversity selectively enriches ARGs in EVs

3.2

We next evaluated how AS diversity affects antibiotic resistance. EVs were successfully isolated using density gradient centrifugation, reaching concentrations of up to 10^8^ particles per gram of soil (Fig. 2a), suggesting their abundance and potential involvement in the spread of resistance within soil ecosystems. Metagenomic profiling based on the SARG database [39] identified 107 ARG subtypes (against 17 antibiotic classes) in microbial DNA and 105 ARG subtypes in EV-associated DNA (Fig. 2b and Supplementary Fig. S4). ARG composition differed markedly between microbial cells and EVs (PERMANOVA, p < 0.001, Fig. 2c). Multidrug resistance genes were the most prevalent, accounting for 54.5% of microbial ARGs and 63.2% of EV-associated ARGs. Resistance genes for vancomycin (11.7%) and bacitracin (7.8%) were more abundant in the microbial fraction, while bacitracin (22.8%) and beta-lactam (4.9%) resistance genes were more abundant in EVs. Notably, we detected 10 ARG subtypes exclusively in EVs, including aminoglycoside (aac3-II, ant2-I), beta-lactamases (metallo-beta-lactamase, NDM-1, TEM-171, PER-2), chloramphenicol (floR), multidrug (mexH), polymyxin (mcr-4), and a quinolone resistance gene (qnrB) (Supplementary Fig. S5). We proposed three possible origins of these EV-specific ARGs. First, they might have originated from low-abundance bacteria within the native community, whose genomic DNA fell below the detection threshold, yet whose ARGs were concentrated and captured within the EVs they released. Second, the lipid bilayer of EVs likely protected these ARGs from environmental nuclease degradation—a fate that might otherwise have occurred if they existed as unstable extracellular DNA. Finally, we could not exclude a nonlocal origin, where EVs from external sources (e.g., airborne microorganisms) could have adsorbed onto the sampled matrix rather than being produced by the local community.Fig. 2Profiles of antibiotic resistance genes (ARGs) in soil microorganisms and extracellular vesicles (EVs). a, Transmission electron microscopy images and particle size distribution of EVs isolated from soil. Calibration was performed using standard reference particles ranging from 68 to 155 nm b, Composition of ARGs in EVs (n = 105), categorized by ARG type. The outer ring represents major ARG types, while the inner ring indicates subtypes. Genes conferring resistance to macrolide–lincosamide–streptogramin are labeled as MLS. c, Principal coordinates analysis (PCoA) based on Bray–Curtis distances, showing distribution patterns of ARGs in soil microbial and EV samples under different artificial sweetener (AS) diversity treatments. d, Correlations between AS diversity and the relative abundance (expressed in transcripts per million [tpm]; top) and diversity (bottom) of ARGs in EVs and soil microbiota. e, Heatmap of significantly altered ARGs in EVs and soil under AS diversity exposure. Significantly altered ARGs were determined by analysis of variance with false discovery rate (FDR) correction. f, Relative abundance of microbial features associated with antibiotic resistance, based on Kyoto Encyclopedia of Genes and Genomes (KEGG) level C annotation (expressed in tpm). ROS: reactive oxygen species; ABC: ATP-binding cassette transporters. M_0, M_1, M_3, and M_6 refer to soil microbial DNA samples collected under treatments with 0, 1, 3, and 6 types of ASs, respectively, while E_0, E_1, E_3, and E_6 represent EV-associated DNA samples derived from the corresponding soil treatments.Fig. 2

Across AS treatments, microbial ARG diversity and abundance remained relatively unchanged (Fig. 2d). However, the relative abundance of ARGs in EVs increased significantly with AS diversity (Fig. 2d, p < 0.05, R^2^ = 0.26, despite minimal changes in overall microbial composition, indicating a decoupling between community structure and EV-associated ARG content. Of the 105 ARGs detected, 36 (34.3%) exhibited significant changes in EVs in response to AS exposure, with bacitracin-resistant genes showing a notable increase as AS diversity increased (Fig. 2e). We also detected 59 high-risk ARGs (Rank I–II), with the multidrug efflux gene mdtK significantly elevated in the three-AS group and ANT2_Ia (aminoglycoside resistance) and sul1 (sulfonamide resistance) markedly increased under the six-AS condition (Supplementary Fig. S6).

Functional analyses of EV-associated DNA revealed that increased AS diversity was accompanied by a higher representation of pathways related to antibiotic resistance (Fig. 2f). Genes involved in DNA repair mechanisms—including base excision repair, homologous recombination, mismatch repair, and nucleotide excision repair—were significantly more abundant in EVs under high-AS conditions (p < 0.05). Similarly, membrane-associated functions, such as ATP-binding cassette (ABC) transporters, bacterial secretion systems, flagellar assembly, and lipopolysaccharide biosynthesis, were enriched at the DNA level. Pathways associated with quorum sensing and SOS responses were also significantly enriched (p < 0.05).

ASs diversity enriches EV-associated VFGs and MGEs and promotes phenotypic transfer

3.3

Next, we examined the potential pathogenicity and mobility of ARGs under varying AS diversity. In EVs, both the diversity (p < 0.01, R^2^ = 0.32) and abundance (p < 0.01, R^2^ = 0.25) of VFGs increased significantly with increased AS diversity (Fig. 3a). In contrast, microbial VFG diversity decreased significantly (p < 0.05, R^2^ = 0.16), while overall abundance remained unchanged (p > 0.05, Fig. 3a). A strong positive correlation was observed between VFG and ARG abundance across both compartments (p < 0.001, R^2^ = 0.53), with stronger associations in EVs (Fig. 3b and Supplementary Fig. S7a). In EVs, the VFG types most strongly associated with ARGs included adherence, antimicrobial activity, and biofilm formation, whereas in microbes, they included antimicrobial activity, effector delivery systems, and motility (Fig. 3c).Fig. 3Potential pathogenicity and mobility of antibiotic resistance genes (ARGs) in soil microorganisms and extracellular vesicles (EVs). a, Correlation between artificial sweetener (AS) diversity and the relative abundance (expressed in transcripts per million [tpm]) of virulence factor genes (VFGs) and mobile genetic elements (MGEs) in EVs and soil microbes. R^2^ and p values are derived from simple linear regression analysis. b, Correlations between the relative abundances (expressed in tpm) of VFGs (left) and MGEs (right) and ARGs in EV samples. c, Correlations between EV ARG profiles and VFG types. d, Correlations between EV ARG profiles and MGE types. Edge width represents Mantel's r value, and edge color indicates statistical significance. Pairwise correlations of these variables were shown with a color gradient corresponding to Pearson's correlation coefficient. Asterisks denote the significance level: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. e, Schematic illustration of the experimental setup and total relative abundance of ARG types in soil bacteria treated with EVs derived from the six-ASs (D6) treatment, compared to the control (CK). Genes conferring resistance to macrolide–lincosamide–streptogramin are labeled as MLSB. f, Schematic illustration of the experiment and the number of Escherichia coli DH5α colonies that grew on cefotaxime-supplemented plates following EV treatment. The asterisk denotes the significance level: ∗∗p < 0.01 (based on t-test).Fig. 3

The diversity and abundance of MGEs in microorganisms did not change significantly after the addition of ASs (Fig. 3a). Although the types of MGEs within EVs remained largely unchanged, increased AS diversity was associated with a significant rise in MGE abundance in EVs (p < 0.01, R^2^ = 0.29, Fig. 3a). ARG and MGE abundances were strongly correlated (p < 0.01, R^2^ = 0.53 in microbes; p < 0.01, R^2^ = 0.41 in EVs), with the correlation being stronger in microbial communities than in EVs (Fig. 3b and Supplementary Fig. S7b). Within EVs, ARG profiles were significantly associated with eight types of MGEs, particularly ISRj1, istA, and qacEΔ1-insertion sequences and integron-associated genes involved in gene mobility and resistance (Fig. 3d). Several of these (e.g., ISRj1, qacEΔ1) also correlated with ARG profiles in the microbial fraction (Supplementary Fig. S8). These findings suggest that MGEs partially contribute to ARG enrichment in EVs, but other mechanisms may also be involved. To explore these alternative mechanisms, we assessed the genomic origin of ARGs using PlasFlow. The majority were predicted to be chromosomally derived (82.3%), with only minor contributions from plasmids (4.6%) and unclassified sources (13.1%) (Supplementary Fig. S9a). Notably, the proportion of chromosomally derived ARGs in EVs increased significantly with AS diversity—a trend not observed in microbial DNA (Supplementary Fig. S9b).

We further evaluated whether high AS diversity enhances the potential for ARG dissemination via EVs by assessing both genotypic and phenotypic responses in recipient microbial communities. Genotypically, EVs extracted from soils treated with either a six-AS mixture or no ASs (control) were separately incubated with a total soil-derived bacterial community. EVs from the six-AS treatment significantly increased the abundance of ARGs in the recipient community, particularly those conferring multidrug resistance (p < 0.01) and aminoglycoside resistance (p < 0.05; Fig. 3e), compared to the control. At the phenotypic level, co-incubation with EVs from the six-AS treatment resulted in a significantly higher proportion of E. coli colonies surviving on cefotaxime plates than with EVs from the control (p < 0.01; Fig. 3f). HT-qPCR analysis showed that recipient E. coli cells exposed to AS-derived EVs exhibited significantly higher expression of cefotaxime-related ARGs than the control (t-test, p < 0.05; Supplementary Fig. S10).

Contribution of the community, genetics, and function of the microbial community to the enrichment of EV ARGs

3.4

To investigate how microbial traits contribute to ARG enrichment in EVs, we reconstructed 264 MAGs (Fig. 4a). The taxa richness in microbes remained unchanged with increased AS diversity, whereas the taxa richness associated with EVs increased (Supplementary Fig. S11). Based on the significantly higher relative abundance in EVs versus bulk soil samples (p < 0.05), MAGs were classified as either EV-enriched (67 MAGs) or microbiome-enriched (181 MAGs). The EV-enriched MAGs recovered in this study were exclusively affiliated with Pseudomonadota and harbored both ARGs and VFGs, consistent with their dominance in EV-associated DNA. This pattern suggests an elevated potential for both antibiotic resistance and pathogenicity among EV-associated microbes. Notably, 97.0% (65 out of 67) of EV-enriched MAGs also carried MGEs, highlighting their potential for HGT. These high-risk MAGs—defined by the co-occurrence of ARGs, VFGs, and MGEs—were primarily associated with resistance to multidrug, polymyxin, and β-lactam antibiotics (Fig. 4b). While Pseudomonadota also dominated the microbiome-enriched MAGs, high-risk MAGs comprised only 22.2% (40 out of 181), and the prevalent ARGs in this group conferred resistance to multidrug, polymyxin, and tetracycline (Supplementary Fig. S12). Furthermore, we identified four gene clusters co-localizing ARGs and MGEs in the EV-enriched group, while only one such cluster was found in the M-enriched group (Supplementary Fig. S13).Fig. 4Characteristics of metagenome-assembled genomes (MAGs) with emphasis on antibiotic resistance genes (ARGs)- and virulence factor genes (VFGs)-containing genomes. a, Phylogenetic tree of all MAGs recovered from the samples. From inner to outer circles, tracks represent MAG relative abundance, and the number of associated ARGs, mobile genetic elements (MGEs), VFGs, and genes involved in ARGs mobilization mechanisms. M_0, M_1, M_3, and M_6 refer to soil microbial DNA samples collected under treatments with 0, 1, 3, and 6 types of artificial sweeteners (ASs), respectively, while E_0, E_1, E_3, and E_6 represent EV-associated DNA samples derived from the corresponding soil treatments. b, UpSet plot illustrating the number of MAGs containing ARGs, VFGs, and MGEs in EV-enriched samples. Lower left bars represent the number of MAGs individually carrying ARGs, VFGs, or MGEs. Upper bars show the number of MAGs co-harboring multiple gene types, with filled circles in the matrix below indicating the specific combinations. The blue bar represents MAGs co-harboring all three gene types, while the gray bar represents MAGs carrying other combinations. The pie chart and associated percentage indicate that all MAGs (100.00%) simultaneously carrying ARGs, VFGs, and MGEs belong to the phylum Pseudomonadota. The panel on the right displays the types and relative proportions of ARGs found in the subset of MAGs. Genes conferring resistance to macrolide–lincosamide–streptogramin are labeled as MLS. c, Average number of ARGs, MGEs, and VFGs per MAG in soil microbial and EV samples. d, Relative abundance of microbial features associated with complex carbohydrate degradation and antibiotic resistance in soil microbial and EV MAGs. The bubble size represents the average number of genes per MAG for that feature category. ROS: reactive oxygen species.Fig. 4

Importantly, EV-enriched MAGs contained significantly more ARGs, VFGs, and MGEs than microbiome-enriched MAGs (p < 0.05) (Fig. 4c). These MAGs were also enriched in genes involved in DNA replication and repair, membrane transport, quorum sensing, and responses to oxidative stress (p < 0.05) (Fig. 4d). Consistently, the abundance of these stress-related genes in EVs increased with AS diversity (Supplementary Fig. S14). In addition, EV-enriched MAGs exhibited a higher prevalence of genes involved in the degradation of complex polysaccharides (e.g., cellulose, chitin, hemicellulose, starch, and lignin), indicating broader metabolic capabilities. The abundance of carbohydrate-degradation genes in EVs also rose in response to increasing AS diversity (Supplementary Fig. S15).

Increased AS diversity significantly altered the composition of ARG-bearing taxa in both soil and EV-associated communities (p < 0.05; Supplementary Fig. S16), suggesting that AS complexity can reshape resistant microbial assemblages. The relative abundance of ARG-containing taxa in EV-enriched MAGs increased with AS diversity (p < 0.05; Supplementary Fig. S17a). A similar trend was observed based on read mapping, in which the cumulative abundance of ARG-annotated sequences in EVs increased significantly with AS complexity (p < 0.05; Supplementary Fig. S17b). Co-occurrence network analysis further revealed that AS diversity strengthened positive correlations among ARG-bearing EV-enriched MAGs, whereas the correlation structure among microbiome-enriched MAGs remained largely unchanged (Fig. S14c and Supplementary Fig. S18).

We next sought to identify key microbial strains to examine whether they served as critical nodes for resistance dissemination. Seventeen EV-enriched MAGs showed consistent increases in relative abundance with rises in AS diversity, primarily from the genera Pararheinheimera and Pseudomonas_E (Supplementary Fig. S19). The relative abundance of these genera strongly associated with EV-derived ARG levels (Supplementary Fig. S20a). Procrustes analysis confirmed a significant concordance between the community structure of Pararheinheimera and Pseudomonas_E and the overall ARG profile (Supplementary Fig. S20b), suggesting that these taxa may play key roles in shaping resistance gene dynamics under complex AS exposure.

Discussion

4

This study demonstrates that multiple ASs significantly increase the risk of ARG dissemination in soil via EVs, a frequently overlooked conduit. We found that increased AS diversity significantly elevated the overall ARG abundance in soil-derived EVs, notably enriching high-risk ARGs. Importantly, EVs from high-AS diversity treatments increased the resistance potential of recipient bacteria at both the genotypic and phenotypic levels. Mechanistically, changes in the functional attributes and community composition of EV-producing host bacteria following multiple AS exposures may have contributed to the elevated ARG abundance observed in EVs. Our findings shift the focus from single-pollutant effects to the interactive impacts of pollutant diversity, highlighting the urgent need to consider the compounded effects of multiple exposures and the underestimated role of EVs as key conduits for the dissemination of resistance in contaminated soils.

We found that microbial communities and their associated EVs responded differently to changes in AS diversity, and we explored several potential mechanisms underlying this divergence. First, this divergence may reflect the distinct selection pressures acting on the intracellular versus extracellular ARG pools. At the community level, ARG abundance is largely governed by total microbial biomass and the fitness of host microbes [51]; in contrast, ARGs encapsulated within EVs constitute a dynamically regulated and mobilizable gene reservoir, whose release depends on the physiological state of the producer rather than population size. Second, EVs may serve as rapid responders or early-warning carriers under environmental stress. They can react to perturbations by disseminating adaptive traits, such as ARGs, faster and more sensitively than the slower ecological succession of microbial communities. A similar temporal decoupling has been observed in Cryptococcus neoformans, where subinhibitory fluconazole altered EV production without affecting resistance gene expression [52]. Finally, our data provide a direct mechanistic basis for this decoupling: EVs mainly originate from a small subset of AS-responsive microbes. Although these microbes are of low relative abundance, their strong stress-induced responses are disproportionately reflected in the EV pool, thereby masking broader community-level stability.

Here, we attempt to decipher the elevated abundance of ARGs in EVs by examining genetic traits and community compositional shifts. From a genetic perspective, EV-producing hosts are characterized by abundant ARG-related features. These include genes involved in oxidative stress response, DNA repair, membrane remodeling, and biofilm formation—all associated with ARG generation and release [[53], [54], [55], [56], [57], [58]]. Furthermore, these bacteria tend to possess larger genomes and a lower GC content than the broader microbial community. Larger genomes typically harbor more functional genes related to resilience and resistance, supporting the acquisition and maintenance of ARGs [59,60]. In addition, low-GC-content genomes are more prone to recombination and often carry MGEs such as integrons, transposons, and plasmids—key reservoirs for ARGs [61,62]. Moreover, these ARG-bearing hosts are more easily activated. We observed that EV-producing bacteria were enriched in genes related to carbohydrate degradation, with many of these taxa belonging to the Pseudomonadota phylum, known for its metabolic versatility and for including previously identified AS degraders [63,64]. The structural complexity of AS compounds [65] may thus favor the activation of these specialized degraders under high AS diversity. Taken together, under multiple AS exposures, EV-producing hosts with large genomes, low GC content, and strong stress-response capacities are likely to be preferentially activated. During this activation, the same genetic elements that confer resistance and adaptability to the hosts may be selectively incorporated into EVs, providing a mechanistic explanation for the elevated ARG abundance observed in EV fractions under high AS diversity. It is important to note that while these genetic features may be associated with ARG enrichment, such correlations do not necessarily imply causality. The observed relationships may also be influenced by ecological niche preference, in which microbes with larger genomes and lower GC content are better suited to specific environmental conditions, such as the presence of pollutants. This could indirectly enhance their ability to harbor and transfer ARGs in response to these environmental pressures.

Altered community composition may also help explain the elevated ARG abundance observed in EVs. All identified EV-producing taxa were ARG carriers, and their relative abundance increased with AS diversity. The enrichment of resistant bacteria likely led to the release of more EVs carrying resistance-associated cargo. Notably, interactions among ARG-harboring bacteria intensified under higher AS diversity. Because EVs are critical vehicles for intercellular communication and HGT [6,33], we propose that these bacteria may actively package ARGs into EVs, facilitating their dissemination across microbial populations. Such enhanced EV-mediated exchange could indirectly contribute to the elevated abundance of ARGs observed within EV fractions. Previous studies have highlighted that AS exposure can reshape the community structure of resistant bacteria [66,67]. Our findings extend this understanding by showing that greater AS diversity not only restructures the resistant community but also increases ARG abundance within EVs.

Next, we identified key microbial targets driving ARG enrichment. All high-risk MAGs within the EV-enriched group belonged to the phylum Pseudomonadota, a well-known dominant source of EVs in environmental settings [9,68,69]. This overlap highlights the potential role of Pseudomonadota as a key conduit for the mobilization of resistance factors via EVs. Among these, genera such as Pararheinheimera and Pseudomonas_E showed strong positive correlations between their abundance and ARG content. Pararheinheimera is considered the ancestral host of the bla (PER) resistance gene, later mobilized by insertion sequences and transferred to clinical pathogens through MGEs [70]. Pseudomonas_E, a clinically relevant genus, harbors diverse ARGs and significantly contributes to antibiotic resistance dissemination in the environment [71]. Taken together, AS diversity reshapes microbial EV production, enriching ARG-carrying taxa and potentially enhancing the spread of resistance. Notably, specific EV-producing taxa, such as Pararheinheimera and Pseudomonas_E, may serve as critical transmission hubs that facilitate the spread of ARGs.

The enrichment of EV-associated ARGs has important implications within the One Health framework. Owing to their nanoscale size and colloidal nature, EVs display higher mobility than intact cells in porous media [72], facilitating long-range gene dissemination beyond the original contamination source. Their lipid bilayer confers strong protection against enzymatic degradation (e.g., DNases) [10], enhancing environmental persistence relative to naked extracellular DNA. Together, these features make EVs stable and efficient vectors for ARG transfer in the environment. Moreover, our data (Fig. 3e and f) demonstrated that EVs isolated from AS-treated soils successfully transferred phenotypic resistance to pathogenic recipients, confirming their potential as mobile reservoirs bridging environmental and clinical resistomes. As EVs can also carry virulence factors [73], their co-dissemination of ARGs and VFGs may intensify infection severity and reduce therapeutic efficacy. While further quantification of exposure routes is needed, these findings underscore the need to integrate EV monitoring into One Health antimicrobial resistance surveillance frameworks.

While we acknowledge that the following aspects of this work have limitations, they serve to open new avenues and highlight important directions for future research. First, the total AS concentration applied (1.2 mg L^−1^) was higher than typical field levels, representing a high-end yet environmentally relevant scenario [22]. This concentration was selected to elicit stable microbial responses under controlled conditions and to reveal potential ecological mechanisms that may also occur at lower environmental exposures. Second, while our data suggest a physiological, stress-induced packaging mechanism, we cannot exclude the possibility that the higher EV yield is also partly a consequence of the increased biomass of these specialized EV-releasing taxa. Third, although clear correlations were established between AS exposure, EV dynamics, and ARG enrichment, the proposed mechanisms remain inferential and are based mainly on metagenomic evidence. The lack of direct molecular validation (e.g., gene knockout or pathway inhibition) limits causal interpretation. Future studies integrating targeted genetic and proteomic analyses will be essential to verify these mechanisms and provide deeper functional insights.

Conclusion

5

In summary, our findings suggest that greater AS diversity may progressively intensify health risks through EV-mediated mechanisms. This includes increased enrichment of ARGs, VFGs, and MGEs in EV threats that may be overlooked by analyses focused solely on the bulk microbial community. These EVs were primarily produced by a small subset of soil microbes that responded strongly to AS diversity, showing increased abundance and elevated expression of EV biogenesis genes. This pattern suggests a potential mechanism in which AS-induced oxidative stress responses lead to membrane structural changes, thereby promoting EV release and increasing the likelihood of ARG packaging. Notably, most EV-producing microbes were antibiotic-resistant bacteria characterized by larger genomes and lower GC content; their increased abundance and intensified interactions under AS exposure likely constitute another potential mechanism underlying the elevated ARG levels observed in EVs. In addition, key taxa, such as members of the Pseudomonadota phylum, are pivotal contributors to EV production and facilitate ARG dissemination. Given the ubiquity of EVs in environmental matrices and their role as concealed vectors of ARG dissemination, future risk assessment frameworks must integrate EV-based indicators. This approach will be critical for developing more accurate and proactive early warning systems to address the growing threat of antimicrobial resistance.

CRediT authorship contribution statement

Yi-Fei Qin: Writing – original draft, Methodology, Investigation, Conceptualization. Wan-Rong Zhang: Methodology, Investigation. Lu Wang: Methodology, Investigation. Yi-Fei Wang: Investigation. Da Lin: Methodology. Tian-Gui Cai: Investigation. Hong-Zhe Li: Writing – review & editing. Qian-Sheng Huang: Writing – review & editing, Supervision. Matthias C. Rillig: Writing – review & editing. Dong Zhu: Writing – review & editing, Supervision.

Data availability

The datasets generated during the current study are available from the NCBI Sequence Read Archive repository (accession no. PRJNA1290030).

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.