Humic acid’s sheltering effect paradoxically promotes antibiotic resistance gene dissemination during peroxymonosulfate water treatment

Abstract

Advanced oxidation processes (AOPs) utilizing peroxymonosulfate (PMS) are increasingly deployed for water disinfection, yet their impact on the dissemination of antibiotic resistance genes (ARGs) remains poorly understood. Here, we conducted field surveys in three aquaculture ponds in eastern China and employed a Transwell-based horizontal gene transfer (HRT) sorting assay, metagenomics, and transcriptomics to investigate the underlying mechanisms. We found that PMS-treated waters harbored significantly higher intracellular ARG burdens compared to untreated or chlorinated systems. Mechanistically, the ubiquitous humic acid (HA) acts as a selective radical scavenger, protecting bacteria from lethal oxidative damage while stimulating type IV pilus-mediated uptake of extracellular ARGs. This interaction redirects HGT from conjugation toward natural transformation. Crucially, this protective effect is oxidant-specific: HA effectively quenches PMS-derived radicals but provides no defense against direct electrophilic attack by chlorine. Furthermore, predation experiments using Caenorhabditis elegans demonstrate that this “sheltering effect” facilitates the accumulation of ARGs in nematodes, decoupling resistance acquisition from oxidative stress intensity. These findings highlight a critical risk pathway in which some water treatment chemicals can promote the spread of resistance, necessitating a reassessment of oxidant selection based on their specific chemical mechanisms.

In this study, the authors show that humic acid, a common organic substance, shields bacteria from peroxymonosulfate disinfection, inadvertently promoting the spread of antibiotic resistance genes via natural transformation in aquatic environments.

Introduction

With the urgent demand for clean water and growing concerns about global pandemics, large amounts of various oxidants are being produced and continuously released into aquatic environments^1,2. Among these, peroxymonosulfate (HSO₅^⁻, PMS), a solid-phase oxidant widely used in advanced oxidation processes (AOPs)^3,4, is increasingly used due to its high efficiency and environmental friendliness^5–9. A direct market search shows that PMS-based water treatment products, available as powders, tablets, and liquid formulations, are widely available on major e-commerce platforms (e.g., Amazon.com and 1688.com). More than a thousand distinct product listings are currently available (representative examples are shown in Fig. S1), with the lowest retail price as low as 0.76 UDS per kilogram (Fig. S1).

On the other hand, academic research on PMS has also drawn increasing attention, especially for the degradation of persistent organic pollutants (POPs) and the inactivation of pathogenic bacteria in various aquatic environments. As presented in Table S1, PMS-based treatments have proven effective across multiple water matrices, including aquaculture wastewater, surface water, municipal sewage, and industrial effluents, with successful validation from pilot to full scale. However, the current knowledge is still significantly incomplete. Although the toxicity of reaction byproducts is increasingly being evaluated, a systematic assessment of the broader ecological implications of PMS, particularly its impacts on non-target aquatic organisms, is still largely absent.

Recently, we surveyed three adjacent open-air aquaculture ponds in eastern China. The results showed that the burden of antibiotic resistance genes (ARGs) per bacterial cell in ponds treated with PMS was approximately three times higher than that in either untreated ponds or ponds treated with chlorinated lime. Based on a comprehensive field survey and laboratory analysis, it is highly probable that PMS use led to the unexpected increase in intracellular ARGs. Nevertheless, this observation is somewhat counterintuitive because PMS is known to degrade free extracellular DNA (eDNA) and inhibit microbial growth^3,6. This finding strongly motivated us to explore the impact of PMS use, along with other environmental factors, on the dissemination of ARGs, including changes in the frequency and mechanisms of horizontal gene transfer (HGT).

Current studies attribute enhanced HGT under environmental stress to a significant increase in cell membrane permeability^10–15, which facilitates the transmembrane transport of eDNA into the cytoplasm. Although this explanation appears reasonable, given the subsequent acceleration of natural transformation, it fails to address a critical issue: natural transformation is a systematic, relay-race-like process. After reviewing the literature, we hypothesized that the observed high burden of ARGs per cell may stem from the induction of natural competence under the unique oxidative stress conditions relevant to PMS. That is because the onset of competence can act as a versatile stress response, enabling the active uptake of exogenous eDNA and supporting self-DNA recombination and repair^16,17. Furthermore, this stress-induced competence status may inadvertently facilitate the integration of ARGs via natural transformation. In other words, exposure to PMS, probably in combination with other environmental factors, may trigger bacterial competence, thus protecting cells from DNA damage and promoting the acquisition of adaptive traits, including antibiotic resistance¹⁸.

To test the above hypothesis, several critical knowledge gaps must be addressed. First, natural transformation has been widely recognized as a key mechanism of HGT¹⁹. However, HGT also occurs via cell-to-cell conjugation^20–22. A significant methodological challenge lies in developing experimental approaches that account for both mechanisms rather than selectively focusing on one, especially under controlled laboratory conditions. Second, competence induction, a prerequisite for natural transformation, is a complex biological process that involves the active pilus uptake of eDNA, protection of single-stranded DNA (ssDNA), and subsequent integration via homologous recombination^23–25. Relying solely on increased cell permeability or higher ARG transfer frequency as evidence may lead to oversimplified conclusions. Third, our initial observations were conducted in aquaculture systems, environments shaped by both human activity and natural ecological dynamics. High organic loads, such as humic acids, and dense metazoan populations, such as nematodes, characterize these systems. Therefore, accurately simulating stress conditions in the laboratory requires accounting for environmental complexity.

In this study, we report that PMS treatment paradoxically promotes the dissemination of ARGs in aquaculture waters by redirecting HGT toward natural transformation. By developing a Transwell-based HGT Sorting and Evaluation Method (THSEM) and employing multiple biological and chemical analyses, we demonstrate that ubiquitous humic acid (HA) acts as a selective radical scavenger. This interaction creates a “sheltering effect” that protects bacteria from lethal oxidative damage while stimulating type IV pilus-mediated DNA uptake. Unlike chlorine, which causes indiscriminate membrane damage, PMS combined with HA specifically induces bacterial competence, decoupling resistance acquisition from oxidative stress intensity. Furthermore, we reveal that this sheltering effect facilitates the accumulation of ARGs in metazoans (e.g., Caenorhabditis elegans), highlighting a pathway for resistance accumulation in ecosystems. These findings provide critical insights into the environmental risks of PMS-based advanced oxidation processes and necessitate a reassessment of oxidant selection in water treatment strategies.

Results

ARG prevalence in field aquaculture water

The three sampled aquaculture water (AW) ponds were located in the Xiangshan district of Ningbo, Zhejiang Province, China (Fig. S2a), an area with over 20 years of intensive shrimp and clam farming. The three sampled AW ponds are all situated within the seawall and equipped with multiple pump-controlled water inlets and outlets that directly connect to the open-sea waterbody. As shown in Fig. 1a, these ponds share similar environmental conditions, with no significant differences in pH, salinity, and light intensity.

Fig. 1. ARG prevalences in field aquaculture waters and THSEM-based evaluation of horizontal gene transfer (HGT) mechanisms under different environmental stresses.

a Environmental parameters and pollutant concentrations were measured in water samples collected from three aquaculture ponds (AW) with different disinfectant application regimes. Pond 1 was treated with PMS (pink/red bars, AW 1_1–3), Pond 2 was nontreated (green bars, AW 2_1–3), and Pond 3 was treated with calcium hypochlorite (Ca(ClO)₂; blue bars, AW 3_1–3). Nine parameters were quantified: humic acid (HA), Fe³⁺, sulfadiazine, light intensity, pH, tetracycline, SO₄²⁻, Cl⁻, and NO₃⁻. Each bar represents one biological replicate (n = 3 per ponds, *=p < 0.05, **=p < 0.01, ***=p < 0.001). b Metagenomic profiling of ARG subtypes across the three ponds. Heatmap displays Log₂ (copies of ARG per cell) with the indicated color scale (4 to 1). ARG subtypes are listed on the right. c Workflow for isolation of indigenous recipient strains and construction of the transwell-based HGT segregation and evaluation model (THSEM). The lower schematic illustrates the THSEM apparatus: the upper chamber (donor + recipient #1) enables conjugative transfer, while the lower chamber (recipient #2) detects natural transformation via eDNA that passes through the filter membrane. d Effects of PMS and co-occurring environmental factors (HA, Fe³⁺, sulfadiazine) on conjugative transfer (red bars, upper chamber) versus natural transformation (blue bars, lower chamber) in the THSEM system: (d-1) Control, HA, PMS, and PMS/HA. (d-2) Fe³⁺, Fe³⁺/HA, Fe³⁺/PMS, and Fe³⁺/PMS/HA. (d-3) SD, SD/HA, SD/PMS, and SD/PMS/HA. Left panels: HGT frequencies (mean values). Right panels: representative colony images on selective plates. n = 3 biologically independent samples, *=p < 0.05, **=p < 0.01, ***=p < 0.001. All quantitative data are presented as mean ± standard deviation (SD) from at least three biologically independent replicates. Statistical significance was determined by one-way ANOVA followed by Tukey’s post-hoc multiple comparison test (for >2 groups). “ns” indicates not significant (p > 0.05). Components of this figure were created in BioRender. Zhou, F. (2026) https://BioRender.com/zqc8hte. Source data and exact p-values are provided in the Source Data file.

Based on our field survey results, three ponds have different preferences for water treatment agents. Specifically, Pond #1 employs PMS twice a month for water and sediment treatment (Fig. S2b). Pond #2 refrains from using any water treatment chemicals. Whereas Pond #3 treats the water by adding calcium hypochlorite (i.e., Ca(ClO)₂) during each water-changing event, which occurs approximately once a season. After a systematic water quality analysis, we identified four key differences: humic acid (HA), ferric ion (Fe³⁺), sulfate ion (SO₄^2-), and sulfadiazine concentration (SD) (Table S2). Among the identified factors, the elevated sulfate ion concentration observed in Pond #1 was directly attributable to the application of PMS. Consequently, the sulfate ion was not treated as an independent variable in subsequent experimental assays.

Using metagenomic analysis, we evaluated 11 ARG classes and 20 subtypes (Fig. S2c). Pond #1 (i.e., sample 1) exhibited significantly higher ARG copies per cell, roughly 3.01-fold greater than those in the second and third ponds. The primary types of ARGs present in Pond #1 were β-lactam resistance genes (17.14%), aminoglycoside resistance genes (15.55%), and multidrug resistance genes (12.60%), among which sul1, bacA, and aadA were the top prevalent subtypes (Fig. 1b). Except for the abnormally high ARG copy per cell in Pond #1, they all exhibited characteristics typical of AW environments with respect to ARG type²⁶.

THSEM evaluation and indigenous recipient strains isolation

We established THSEM to segregate HGT between cell-to-cell conjugation and natural transformation using a transwell-based system (Fig. S3). The concept of THSEM was validated by using model type #1 recipient (Escherichia coli DH5α [pET-28a (Kan)]) and type #2 recipient (E. coli MG1655 (pBAD-ngo)) strains. In brief, a specific environmental stress (or a combination of multiple stresses) can be directly applied to the THSEM’s chambers. The upper chamber houses both donor and type #1 recipient cells, enabling cell-to-cell conjugation. A filter membrane physically separates the upper and lower chambers, allowing only cell debris, such as DNA molecules, to pass while preventing intact cells from crossing. In the lower chamber, only the type #2 recipient is inoculated; therefore, any new resistance acquired by the type #2 recipient can be attributed to natural transformation with high confidence. Notably, the eDNA released into the upper chamber can rapidly settle into the lower chamber, as we intentionally supplemented the system with 2% sodium chloride to accelerate DNA settling and facilitate its rapid crossing of the filter during translocation.

To effectively reflect ARG dissemination in those AW environments, we carefully isolated indigenous bacteria from field-collected water samples (Fig. 1c). First, we screened three strains with specific resistance, which could conjugate with the model donor strain (E. coli S17-1 λpir [RP4 (Ampr)]) to exhibit acquired ampicillin resistance (Ampr). Meanwhile, they still harbor a distinct resistance, allowing us to distinguish them from the donor (Figs. S4–7). One of the three isolations, Bacillus licheniformis A2.4 (Ery), was eventually selected as the type #1 recipient strain due to its inherent erythromycin resistance (Eryr), relatively stable cell-to-cell conjugation performance, and more distinguishable colonial morphology. For natural transformation, we isolated only one potential type #2 recipient strain, namely Acinetobacter baylyi A3.6 (Fig. S8), which meets our stringent criteria: it does not carry both Ampr/Eryr and undergoes natural transformation reliably with high genetic stability.

It should be noted that the environmental stresses used in THSEM are strictly aligned with real-world conditions. That includes applying commercial PMS at the suggested dosage (at a dosage of 2 g per 100 L with PMS equivalent concentration at about 10 ppm), adding the average HA/Fe³⁺/SD concentrations measured in the field samples (i.e., HA 20 mg/L, Fe³⁺ 50 µg/L, and SD 45 ng/L, respectively), and incubating all THSEM systems in a climate chamber fully adjusted to the actual site conditions for 24 hrs. The results of ARG dissemination patterns are presented in Fig. 1d. As shown, in the absence of PMS, HGT occurred exclusively in the upper chamber, indicating that cell-to-cell conjugation was the primary mechanism of HGT under no environmental stress.

After PMS was introduced, HGT was redirected to the lower chamber. Notably, the presence of HA increased transfer frequencies, suggesting that the combined exposure to PMS/HA stimulated the type #2 recipient to undergo substantially intensified natural transformation. The addition of Fe³⁺ in PMS slightly increased HGT in the upper chamber. However, we incline to attribute this elevated HGT in the upper chamber to natural transformation rather than conjugation. That is because divalent or trivalent cations can form a “cation bridging” between the eDNA and the filter membrane²⁷. As a result, the filter can capture eDNA and partially retain it in the upper chamber, despite the membrane pores being relatively larger. The addition of SD in PMS at 45 ng/L reduced HGT in the upper chamber from 10^–1.8 to 10^–4.4 (p < 0.001). In contrast, the lower chamber was not affected by SD, with a frequency of up to 10^–1.3. The most remarkable observation is that, compared with PMS alone, the combination of PMS/HA can further enhance the natural transformation in the lower chambers, regardless of the presence of Fe³⁺ and SD.

DEGs profiling: comparison between control, PMS, and PMS/HA

Herein, transcriptomic analyses were conducted to elucidate the molecular mechanisms underlying the enhanced natural transformation in the lower chamber. The microbial samples, including the control, PMS, and PMS/HA groups, were obtained under the same stress conditions as those in the aforementioned THSEM system and immediately subjected to RNA purification. The biological replicate samples exhibited high reproducibility (correlation coefficient: 0.94–1.0; Fig. S9).

Compared with the control group, the PMS group showed 56 differentially expressed genes (DEGs), including 27 upregulated and 29 downregulated genes (Fig. S10). Using Gene Ontology (GO) enrichment analysis, it was found that, under PMS stress, the respiratory chain and electron transport in the receptor bacteria were generally downregulated (e.g., the NarGHI complex). In contrast, the stress response transcriptional regulatory factors (e.g., DNA-binding transcription factor activity), natural transformation ability (e.g., establishment of competence for transformation), and membrane damage-related genes (e.g., undecaprenyl-diphosphatase activity) were significantly upregulated (Figs. 2a-1).

Fig. 2. Mechanisms underlying PMS- and PMS/HA-enhanced natural transformation in A. baylyi A3.6.

a GO enrichment analysis of DEGs identified by RNA-seq after exposure to PMS (10 ppm) versus control (a-1), and PMS/HA (20 mg/L HA) versus PMS (a-2). Bubble plots show enriched biological process and cellular component terms; x-axis, rich factor; bubble size, gene number; color, FDR. DEGs were defined as |log₂ fold change | ≥ 1 and FDR < 0.05. GO enrichment was performed using clusterProfiler (FDR < 0.05). b Relative abundance of the ampicillin-resistance gene blaTEM-1 on pUC19 after 24 h exposure to PMS or PMS/HA, quantified by qPCR and normalized to the untreated control (mean ± SD, n = 3 biologically independent samples, p < 0.05). c Schematic of natural transformation in competent A. baylyi A3.6, including type IV pili-mediated eDNA uptake, SSB-mediated ssDNA protection, and RecA-mediated homologous recombination. d Dose-dependent inhibition of twitching motility by tuspetinib. Twitching area was measured after 48 h incubation, with an IC₅₀ of 23.8 ± 0.4 μM (mean ± SD, n = 3 biologically independent samples). e Natural transformation frequency under PMS or PMS/HA stress with or without 50 μM tuspetinib (mean ± SD, n = 3 biologically independent samples, p < 0.05, p < 0.01, p < 0.001). f Representative TEM image of negatively stained PMS/HA-treated cells showing type IV pili binding eDNA; inset shows the enlarged pilus–eDNA interaction. The experiment was independently repeated 10 times with similar results. g Flow cytometric analysis of membrane potential using DiOC₂(3) staining. PerCP and FITC indicate red and green fluorescence, respectively, and quadrant values indicate the percentage of cells in each population. The gating strategies are shown in Fig. S22. h qRT-PCR analysis of competence-related genes: comGC, comA, ssb, and recA (mean ± SD, n = 3 biologically independent samples, p < 0.05, p < 0.01). Statistical significance in (b, e, and h) was assessed by one-way ANOVA with Tukey’s post-hoc test or two-tailed unpaired Student’s t test, as appropriate. ns, not significant (p > 0.05). Components of this figure were created in BioRender. Zhou, F. (2026) https://BioRender.com/b0s6kjg. Source data and exact p-values are provided in the Source Data file.

The comparison between PMS and PMS/HA identified 126 DEGs, of which 38 were upregulated and 88 were downregulated (Fig. S10). GO enrichment highlighted several cellular functional transcription processes that underwent significant alleviation in the PMS/HA group, such as cellular respiration, respirasome, and plasma membrane respirasome. More importantly, the presence of HA further upregulated the transformation ability (e.g., the establishment of competence for transformation) and membrane transport proteins (e.g., the plasma membrane protein complex) (Fig. 2a-2). The above DEGs clearly depict a state in which the PMS triggers the bacterial oxidative stress response, and, simultaneously, naturally competent cells are initiated to carry out DNA repair via natural transformation.

eDNA availability under stress conditions

The availability of homologous eDNA provides the essential material basis for natural transformation, and its concentration should be considered first during any environmental stress tests. In this study, the pUC19 plasmid carrying the ampicillin resistance (Ampr) gene was used as the eDNA probe to explore how chemical stress affects the integrity of ARG. qPCR analysis showed that both 10 ppm PMS alone and its combination with 20 mg/L HA significantly compromised the integrity of the Ampr gene, reducing its amplification efficiency by 44-49% relative to the controls (fold change of 0.51–0.56; p < 0.05; Fig. 2b), corresponding to an estimated eDNA concentration of 1.37–1.50 ng/μL. The indigenous type #2 recipient strain (i.e., A. baylyi A3.6) was then used to assess the impact of changes in eDNA concentration on natural transformation frequency (Fig. S11). As shown, eDNA ranging from 0.1 to 5 ng/μL did not show a significant effect on the type #2 recipient’s natural transformation frequency. A significant increase was observed only at an eDNA concentration up to 10 ng/μL. However, such a high level of eDNA is relatively unrealistic in most natural or artificial environments^28,29.

Pilus-mediated uptake vs. transmembrane diffusion

In Fig. 2c, we illustrate the typical pathway by which a competent bacterial cell can access eDNA and integrate it into its chromosome^16,30. At the core of the eDNA uptake machinery is the extendable type IV pili (T4P), which actively mediates the initial binding of eDNA. According to recent research, T4P are essential for eDNA uptake in nearly all known competent bacterial species^24,31. On the other hand, numerous recent studies have reported that when a stress (e.g., chlorine and artificial sweeteners, etc.) disrupts the bacterial cell membrane barrier, non-specifically increasing its permeability, eDNA that was previously blocked can more easily enter the periplasm or even the cytoplasm via passive diffusion or nonspecific pores, thus accelerating natural transformation^10–15.

As a result, investigating whether the stimulated pilus or the disrupted membrane is the key promoting factor under PMS and PMS/HA stress becomes a priority. To effectively distinguish between the above two mechanisms, we first employed tuspetinib, a pili-formation inhibitor³², at different treatment levels, along with the twitching motility assay³³, to rule out the potential transformation mediated by the T4P pilus. According to the twitching assay, the half T4P inhibitory concentration of tuspetinib of our isolated A. baylyi A3.6 was determined to be 23.8 μM, and the optimal concentration for complete T4P inhibition was approximately 50 μM (Fig. 2d). The subsequent comparative analysis demonstrated a straightforward finding that in the absence of T4P inhibition (Fig. 2e), the competent strain can undergo highly active natural transformation under PMS and PMS/HA stresses. In contrast, when Tuspetinib was added at 50 μM to fully exclude the T4P uptake, only the PMS group showed a slightly higher transformation activity at 10^–4.16 (p > 0.05).

Directly visualizing competence pilus, especially while they are actively capturing eDNA, requires a rare spatiotemporal window that has, until now, seldom been captured. Due to PMS/HS-induced competence, our effort to capture the moment when T4P were pulling in eDNA was significantly facilitated. As shown in Fig. S13, following exposure to PMS, most surviving cells developed sparse T4P, yet exhibited minimal binding to eDNA. The most striking observation occurred under the coexistence of PMS/ HA, in which, on the TEM images, one can readily observe numerous T4P with their tips clearly bound to eDNA (Fig. 2f, Figs. S12–14), suggesting that a highly activated DNA uptake system was triggered at the onset of the combination stress.

Instead of using propidium iodide (PI) to cross-membrane stain apoptotic bacterial DNA, we applied 3,3’-Diethyloxacarbocyanine Iodide (DiOC₂(3)) to probe the cell membrane potential and directly reflect the degree of membrane healthiness under PMS oxidative stress (Fig. 2g)³⁴. Compared with both the control and HA groups, PMS stress significantly induced a red shift in the flow cytometric plot, with 88.8% of the cells located in the upper-right quadrant, indicating a highly disordered membrane potential and an almost non-selective, yet highly permeable membrane condition. Interestingly, when HA coexisted with PMS, the red shift was greatly alleviated; although it was not fully restored to the healthy lower-right zone, 68.7% of the cells regained a normal, functionalized membrane. The most critical observation here is the disparity in transformation outcomes (Fig. 2e). When using the pilus inhibitor, the PMS group exhibited only slight increases of 2.5-fold (p < 0.05) and 1.6-fold (statistically insignificant) compared to the controls and the PMS/HA group. In contrast, the PMS/HA group showed the highest transformation without adding the pilus inhibitor, approximately 33-fold (p < 0.001) and sixfold (p < 0.01) greater than the controls and the PMS group, respectively.

Profiling of competence-related genes

Based on the above results, pilus-mediated uptake appears to be the critical pathway for eDNA to translocate. However, natural transformation is a highly complex, relay-race-like process that involves several consecutive steps, including double-stranded DNA (dsDNA) binding, transmembrane transport, dsDNA unwinding, single-stranded DNA (ssDNA) protection, and homologous recombination^35,36. Thus, we selected four of the most crucial yet conservative gene expressions to profile and compare the natural transformation under different stress conditions, including ComGC (eDNA binding), ComA (regulatory), ssB (ssDNA binding), and recA (homologous recombination)^37–39. As shown in Fig. 2h-1–h-4, the PMS/HA group showed the highest expression levels for all genes. Particularly for ComGC, ComA, and ssB, whose expression levels exceeded those of the controls and the PMS group, with all differences statistically significant. The only exception occurred in recA, where the PMS and PMS/HA groups had highly similar expression levels, approximately 2.5-fold those of controls (p < 0.01).

No doubt, PMS stress induces competent cells to activate their transformation mechanism for eDNA uptake and subsequent processes. While it significantly disrupts the cell membrane, allowing a small proportion of eDNA to directly traverse, potentially in the form of intact plasmids, which may circumvent certain critical steps, such as ssDNA protection, that are essential for typical natural transformation. However, HA redefines the PMS stress. It not only protects the cell membrane but also infuses the pilus with greater vitality, leading to the highest natural transformation frequency. Thus, it is of great importance to investigate the role of HA in the context of PMS coexistence, as this combination of oxidative stress is highly likely to become increasingly prevalent in numerous circumstances.

Chemical Interactions between HA and PMS

To comparatively assess the oxidative capacity of the PMS and PMS/HA systems, methylene blue (MB) degradation was monitored and used as the oxidative probe (Fig. 3a, b, Fig. S15). While the PMS system achieved higher total degradation, the PMS/HA system exhibited accelerated initial kinetics within the first 10 min (Fig. 3b). This transient enhancement suggests that HA may initially act as a promoter for PMS activation before eventually becoming a substrate and potentially a radical scavenger.

Fig. 3. The Chemical Interactions between HA and PMS.

a UV-Vis absorption spectra of mixtures with different concentrations (0–10) MB / 20 HA (mg/L), and a standard curve was established at 595 nm. b MB degradation curves for the PMS/HA system and the PMS system (n = 3 biologically independent samples). c DMPO and d TEMP spin-trapping EPR spectra of the PMS/HA system. e Fourier transform infrared spectra of the reaction between PMS and HA at different time sequences and their 2D correlation spectroscopy (0:30 min). f Density functional theory (DFT) calculations on the electron gain and loss states of HA molecules. Source data and exact p-values are provided in the Source Data file.

To investigate the radical chemistry underlying this kinetic profile, EPR spectroscopy with spin-trapping was employed. The two systems exhibited fundamentally different radical evolution patterns (Fig. 3c, d, Fig. S16). In the PMS/HA system, a strong, transient signal corresponding to DMPO and organic complex (DMPO-X, a seven-line multiplet⁴⁰) surged within 5 min and decayed afterward, indicating the rapid generation and consumption of a specific, HA-derived radical intermediate. When tested with only PMS, a weaker but persistent combined signal from DMPO-•SO₄^- and DMPO-•OH adducts was observed, consistent with the conventional radical oxidation pathway. The characteristic signal of singlet oxygen (¹O₂, trapped by TEMP⁴¹) emerged after 5 min and accumulated steadily, confirming ¹O₂ as a major, long-lived oxidant regardless of HA’s presence.

Stepwise FTIR was used to track the molecular-level transformation of HA in the PMS/HA system (Fig. 3e). The spectral evolution reveals a clear oxidative degradation process. Among all signals, the emergence and growth of the S-O stretching at 1100 cm^-1 clearly prove the activation of PMS and the accumulation of its oxidation product, sulfate (SO₄^2-)⁴². Meanwhile, a continuous increase in the carbonyl (C = O) stretching band at 1720 cm^-1, assignable to carboxylic acids and quinones, confirms the progressive oxidation of the phenolic groups in HA. The latter decreases in bands at 1650 cm^-1 (aromatic C = C) and 1410 cm^–1 (phenolic O-H bending) reflect the degradation of the aromatic backbone and consumption of phenolic groups⁴³. The delayed appearance and growth of the aldehyde C-H stretching band at 2950 cm^–1 after 10 min provides direct evidence for the formation of formic acid, a small molecular acid resulting from advanced oxidative cleavage⁴³.

Two-dimensional correlation spectroscopy (2D-COS) was applied to resolve the temporal sequence of these transformations (Fig. 3e), and the result depicted a consecutive three-stage process. The primary event is the decomposition of PMS, indicated by the early change in the S-O stretching vibration (1100 cm^–1). Then, the oxidation of the HA core structure is followed by an increase in carbonyl groups (1720 cm^–1) and changes in hydroxyl groups. The final stage involves the formation of small carboxylic acids, such as formic acid, as indicated by the late appearance of the 2950 cm^–1 band.

Identifying the electron transfer center and reactive sites

To complement the experimental findings and provide an electronic structure-level understanding of HA’s reactivity, we performed density functional theory (DFT) calculations on a representative soluble HA model structure (Fig. 3f). The calculated frontier molecular orbitals reveal the molecule’s inherent redox character. The HOMO energy level is –6.221 eV, and the LUMO level is –1.945 eV, yielding a HOMO-LUMO gap of 4.276 eV. The calculated substantial gap indicates good thermodynamic stability for HA. More importantly, the HOMO is primarily localized on the aromatic π-system and the oxygen atoms of the phenolic hydroxyl groups (Fig. 3f), indicating that these regions constitute the molecule’s primary electron-donating reservoir. This orbital distribution strongly suggests that the phenolic moieties are the most likely sites for initiating single-electron transfer to PMS, consistent with the initial rapid kinetics observed experimentally.

To further pinpoint the most vulnerable sites for electrophilic attack by PMS or its derived radicals, we conducted a condensed Fukui function analysis (Fig. 3f, Table S3). The oxygen atom (O17) of one phenolic hydroxyl group exhibits the highest nucleophilic Fukui index (f^- = 0.1038), followed by the oxygen of another phenol (O16, f^- = 0.0755). The aromatic carbon atoms directly bonded to these hydroxyl groups (C1 and C3) also show significantly elevated f^- values (0.0857 and 0.0828, respectively). It can be observed that the phenolic hydroxyl group and its ortho/para positions on the aromatic ring form an integrated, electron-affinity “hotspot”, which makes these positions the most likely sites for the initial oxidative attack.

Bacterial responses to PMS

To evaluate whether the observed protective effect of HA is specific to PMS or a universal phenomenon across different disinfectants, we used propidium iodide (PI) staining and flow cytometry to quantify the disruption of cell membrane integrity in A. baylyi A3.6 exposed to different disinfectants (i.e., PMS, calcium hypochlorite (Ca(Cl)O₂), hydrogen peroxide (H₂O₂), and potassium persulfate (PDS); disinfection efficacy optimized in Fig. S17) (Fig. 4a). At the commercially recommended doses tested, PMS alone induced PI positivity in 56.3% of cells. In contrast, coexistence with HA conferred significant, dose-dependent protection, as PI positivity decreased to 32.5% at 10 mg/L HA and further to 12.2% at 20 mg/L HA.

Fig. 4. Differential effects of HA on membrane damage and transcriptomic responses induced by different chemicals, and amplification of upward ARG transfer risk in the C. elegans food chain under PMS stress.

a PI staining and flow cytometry analysis of A. baylyi A3.6 membrane integrity after exposure to PMS (10 ppm), Ca(ClO)₂ (12.5 ppm available chlorine), H₂O₂ (1.5%), or PDS (10 ppm), alone or with HA (10 or 20 mg/L). Histograms show PI fluorescence intensity, and values indicate PI-positive cells with disrupted membranes. Gray indicates untreated control. The gating strategies are shown in Fig. S23–24. b Volcano plots showing DEGs between each chemical treatment and the corresponding chemical/HA treatment: (b-1) PMS vs PMS/HA (Up = 308, Down = 13). b-2 Ca(ClO)₂ vs Ca(ClO)₂/HA (Up = 138, Down = 6). (b-3) H₂O₂ vs H₂O₂/HA (Up = 28, Down = 5). b-4 PDS vs PDS/HA (Up = 112, Down = 37). Red, blue, and gray dots indicate significantly upregulated, downregulated, and non-significant genes, respectively; selected oxidative stress-related genes are labeled. DEGs were defined as |log₂FC | ≥ 1 and p < 0.05. c Schematic of the C. elegans L1-stage microcosm model. Approximately 200 worms were exposed in 200 mL M9 medium containing 10 ng/L antibiotics, SYTO 9-prestained ARG-harboring bacteria, PMS, and 20 mg/L HA. ARG accumulation was quantified by qPCR or visualized by LSCM. d qPCR quantification of ARG copy number per worm (log₂(Cq/Ct₁)). Data are mean ± SD (n = 3 biologically independent samples, ***=p < 0.001). e Quantification of intestinal green fluorescence intensity by LSCM and ImageJ analysis (mean ± SD, n > 7 biologically independent samples, p < 0.001). f Representative LSCM images of worms exposed to PMS concentrations of 0, 5, 10, 20, or 50 ppm with 20 mg/L HA, showing fluorescent bacteria localized mainly in the pharynx and digestive tract. Statistical significance was determined by one-way ANOVA with Tukey’s post hoc test or two-tailed unpaired Student’s t test, as appropriate; ns, not significant. Components of this figure were created in BioRender. Zhou, F. (2026) https://BioRender.com/31vo9s8. Source data and exact p-values are provided in the Source Data file.

In comparison, Ca(Cl)O₂ alone caused near-complete membrane damage (99.6% PI-positive). HA offered no meaningful protection, with PI-positivity remaining extremely high at 99.5% (10 mg/L HA) and 98.2% (20 mg/L HA). H₂O₂ showed responses similar to those of Ca(Cl)O₂, where H₂O₂ alone resulted in 60.2% PI-positive cells, and the addition of HA caused only marginal changes (59.1% at 10 mg/L and 59.0% at 20 mg/L). Likewise, PDS alone led to 55.8% PI-positive cells, with HA showing comparatively low protective effect (54.3% at 10 mg/L and 54.1% at 20 mg/L). These results demonstrate that HA’s protective effect against bacterial membrane damage is highly specific to PMS.

To explore the biological responses, we reperformed transcriptomic analysis and used volcano plots to visualize the distinct responses to the four chemicals (Fig. 4b-1 to b-4). Compared to PMS, Ca(Cl)O₂ exposure triggered 321 DEGs, of which 308 were upregulated (Fig. 4b-1 and b-2). These DEGs were highly enriched for oxidative stress response pathways, including key genes such as ahpF, mqsA, trxC, nemR, bhsA, and marR, indicating that Ca(Cl)O₂ can deliver a more severe, direct oxidative stress. In the presence of HA, the transcriptomic profile remained similar to that of Ca(Cl)O₂ alone (138 up, 6 down), and oxidative stress genes (mqsA, rclA, bhsA, trxC, soxS, ahpC, marR) were also significantly elevated. The interactions of H₂O₂ and PDS with HA were much more limited. Comparison between H₂O₂ alone and H₂O₂/HA yielded only 33 DEGs (28 upregulated, 5 downregulated) (Fig. 4b-3). The upregulated genes were primarily associated with amino acid metabolism and transport (argA, lysA, metF, metR, artJ), glycerol metabolism (glpK, glpD), and envelope stress (cpxP), rather than a broad oxidative stress response. Similarly, comparison between PDS alone and PDS/HA resulted in 149 DEGs (112 upregulated, 37 downregulated) (Fig. 4b-4). Upregulated genes were mainly involved in phosphate transport and starvation response (phoA, phoB, phoU, pstS), methionine metabolism (metR, metB), and cell wall-related functions (ldtC, yebE); downregulated genes were enriched in nitrogen metabolism and amino acid transport (astA, astC, asnA, asnB, amtB, glnK, argT). These modest, metabolism-oriented changes further confirm that HA cannot effectively alleviate the oxidative stress induced by H₂O₂ or PDS. Collectively, the transcriptomic results indicate that the significant alleviative interaction with HA is specific to PMS. In contrast, the other three exhibit only limited transcriptional modulation by HA.

The “Sheltering Effect” amplifies the upward transfer risk of ARGs

A critical environmental concern is whether a “sheltering effect,” in which HA protects bacteria and stimulates them to undergo natural transformation under oxidative stress, could also inadvertently facilitate the dissemination of ARGs through the food chain. To investigate this risk, we designed a model predation experiment using L1-stage Caenorhabditis elegans nematodes as primary consumers. Transformed type #2 recipient cells (harboring Ampr) were co-incubated with nematodes under the stress of PMS (0-50 ppm) and HA (20 mg/L) (Fig. 4d).

After PMS/HA exposure, nematodes were dissected, and their digestive glands were isolated for qPCR analysis targeting the Ampr gene. Strikingly, the qPCR results (Fig. 4e, Fig. S18) revealed that all PMS treatments (5, 10, 20, and 50 ppm) significantly and indistinguishably increased the Ampr load per nematode by about 2-fold compared to the control (p < 0.001). This result demonstrates that in the presence of HA, the uptake and accumulation of ARG-harboring bacteria by nematodes become decoupled from the applied oxidative stress level, directly evidencing the risk-amplifying role of the sheltering effect. To corroborate the qPCR findings and visualize the spatial distribution of accumulated bacteria, we used laser-scanning confocal microscopy (LSCM). The transformed bacteria were pre-stained with the nucleic acid dye SYTO 9 before the predation experiment. The LSCM images (Fig. 4f, g) confirm that surviving nematodes from all stress groups harbored fluorescent bacteria within their bodies, primarily localized along the pharynx and throughout the digestive tract.

Discussion

This study reveals a critical environmental issue associated with PMS use in water treatment. While intended to disinfect, PMS, when combined with ubiquitous HA, can inadvertently promote the dissemination of ARGs. Field surveys of aquaculture ponds showed that PMS-treated waters harbored several-fold higher intracellular ARG burden than untreated or chlorine-treated systems. Through the new THSEM system, we demonstrated that PMS/HA stress specifically redirects HGT toward natural transformation rather than conjugation.

Based on our chemical analyses, the interplay between PMS and HA is a dynamic, bidirectional process that dictates the system’s oxidative fate. Our spectroscopic and computational analyses (Fig. 3) reveal that PMS actively oxidizes phenolic moieties on HA into quinones and carboxylic acids. Meantime, the structural evolution of HA significantly modulates PMS activation. Initially, electron-rich phenolic groups in HA act as electron donors, accelerating PMS activation. However, as the reaction proceeds, HA functions as a potent radical scavenger, effectively quenching aggressive free radicals (e.g., •SO₄^- and •OH) while favoring the generation of longer-lived, less lethal species, particularly ¹O₂.

This mutual interaction creates a buffered oxidative condition, in which HA degradation consumes oxidative potential, while the altered radical spectrum spares bacteria from lethal damage. As a result, a permissive environment is established in which bacteria, shielded from death, activate competence-related pathways, enhance type IV pilus-mediated DNA uptake, and ultimately increase the natural transformation frequencies of ARGs. Comparative analysis with calcium hypochlorite highlighted the oxidant-specific nature of this phenomenon. In brief, HA offers effective protection against radical-based oxidation of PMS, yet it provides no measurable defense against direct electrophilic attack by chlorine, highlighting that the chemical mechanism determines environmental risk.

It is essential to distinguish between the induction of natural transformation and the selection of antibiotic-resistant transformants, because in many real-world conditions, background residual antibiotics inevitably coexist. Our experimental results indicate that the addition of sulfadiazine reduced HGT in the upper chamber, suggesting that, at environmentally relevant concentrations (e.g., sulfadiazine at 45 ng/L), it primarily exerts a straightforward inhibitory effect (killing donor or recipient cells), thereby reducing the overall opportunity for bacterial conjugation (Fig. 1d-3).

However, in the lower chamber, our study demonstrates that PMS/HA stress promotes natural transformation both with and without antibiotic selection pressure to approximately the same extent (Fig. 1d-3). No doubt, the primary driver of this outcome is oxidative stress-induced competence, a physiological SOS response wherein competent bacteria take up eDNA to mediate chromosomal repair, thereby facilitating the natural transformation of ARGs. This finding indicates that the “sheltering effect” facilitates the HGT of genetic material regardless of the immediate presence of antibiotics. It should be noted that while background residual antibiotics do not directly trigger activation of the bacterial DNA uptake machinery, they may impose co-selection pressure that drives the enrichment of surviving transformants, worsening the condition and requiring further investigation.

We also found that the environmental implications can further extend to ecosystem-level consequences. Our model predation experiment demonstrated that metazoans, such as nematodes, can use the space in their digestive tracts to decouple the intensity of oxidative stress, enabling consistent amplification of ARG accumulation in aquatic animals. This finding reveals a previously unrecognized risk pathway in which water treatment chemicals, particularly those that operate via radical pathways such as PMS-based advanced oxidation processes, may inadvertently create selective conditions that favor ARG-harboring bacteria and facilitate their entry into food webs.

The study argues that environmental risk assessment must evolve beyond evaluating direct antimicrobial efficacy. It is equally important to consider how chemical-biological interactions in complex matrices, such as HA-rich aquatic environments, can generate unintended evolutionary pressures. As PMS use expands globally, these insights provide an urgent scientific basis for reevaluating water treatment strategies and developing more comprehensive frameworks that account for the potential of disinfectants to shape microbial evolution and the dissemination of antibiotic resistance in unexpected ways.

Methods

Bacterial strains and culture media

The donor strain, E. coli S17-1 λpir [RP4 (Ampr)], which was independently isolated and screened in our laboratory, carries the self-transmissible plasmid RP4. Indigenous recipient strains used in the transwell-based HGT segregation and evaluation model (THSEM) were isolated from the three field-collected AW samples described in Fig. S1. All strains were routinely cultured in LB broth or on LB agar plates at 37 °C unless otherwise specified. Antibiotics were added when required at the following concentrations: ampicillin (Amp, 100 μg/mL), erythromycin (Ery, 100 μg/mL).

Recipient #1 (conjugative transfer, upper chamber)

Recipient #1 was selected as Bacillus licheniformis A2.4 (Eryr). To screen for suitable type #1 recipients, AW samples were first spread on plain LB agar plates without any antibiotics (Fig. S4). 100 single colonies were picked and tested for antibiotic susceptibility to ampicillin, kanamycin, sulfadiazine, neomycin, erythromycin, tetracycline, streptomycin, gentamicin, and chloramphenicol. Three candidate recipient #1 strains were identified: KA1.5 (gentamicin- and kanamycin-resistant), A1.4 (erythromycin-resistant), and A2.4 (erythromycin-resistant). These strains were then streaked for purification, and single colonies were picked and preserved. The three candidate strains were further tested for MICs to confirm their ampicillin sensitivity. Subsequently, individual colonies were used to assess HGT by performing biparental mating with the model donor E. coli S17-1 λpir [RP4 (Ampr)]. The mating mixtures were plated on LB agar containing both Amp (100 μg/mL) and Ery (100 μg/mL) to select transconjugants that had acquired the RP4 plasmid (Ampr) while retaining their native Eryr marker. This dual-selection strategy ensured that the recipient could be unambiguously distinguished from the donor. The three candidate strains that met these criteria were further purified by repeated streaking and identified by 16S rRNA gene sequencing (Figs. S12–14). B. licheniformis A2.4 was ultimately chosen as the final recipient #1 strain because of its stable conjugative performance, easily distinguishable colony morphology, and high genetic stability under THSEM conditions.

Recipient #2 (natural transformation, lower chamber)

Recipient strain #2 was selected as A. baylyi A3.6. To isolate a suitable type #2 recipient, field AW samples were screened for indigenous strains capable of natural transformation but lacking pre-existing Ampr or Eryr markers. Briefly, colonies grown on plain LB agar were tested for their ability to take up naked RP4 plasmid DNA (carrying Ampr) in a standard natural transformation assay. Transformants were selected on LB + Amp plates, and only strains that (i) did not exhibit intrinsic Ampr or Eryr resistance before transformation and (ii) showed reproducible, high-frequency natural transformation were retained. After monoclonal purification, MIC testing, and species identification by 16S rRNA sequencing (Fig. S15), A. baylyi A3.6 was selected as the final type #2 recipient because it fulfilled all stringent criteria: absence of both Ampr/Eryr markers and reliable, genetically stable natural transformation under the environmental stresses applied in the THSEM system.

Both recipient strains were stored as 50% (v/v) glycerol stocks at -80 °C and freshly streaked from frozen stocks for each experiment. All subsequent THSEM assays used these verified indigenous recipients to ensure ecological relevance to the original aquaculture water environment.

Environmental parameters and indicator analysis of AW samples

Study site and sample collection

Field sampling was conducted in September 2021 in shrimp aquaculture ponds located near Ningbo, Zhejiang Province. The sampling sites represented typical intensive aquaculture systems for Litopenaeus vannamei cultivation in the coastal region of eastern China. Three shrimp ponds with distinct water-management regimes were selected for this study: Pond A (121°29′23.65″E, 29°24′12.57″N) received periodic PMS treatment, Pond B was untreated (121°30′54.64″E, 29°27′41.49″N) was maintained without chemical treatment, and Pond C (121°31′4.50″E, 29°27′27.49″N) received periodic calcium hypochlorite treatment. Both PMS and Ca(ClO)₂ were applied according to the manufacturers’ recommended dosages during routine pond management.

For each pond, three independent biological replicates (A1-A3, B1-B3, and C1-C3) were collected from spatially separated locations within each pond to account for spatial heterogeneity. Surface water samples were collected and immediately transported to the laboratory on ice. Upon arrival, samples were filtered through 0.45 μm polyethersulfone membranes and stored at 4 °C until further analysis. All samples were processed within 4 h after collection.

Environmental parameters

pH was measured on-site using a calibrated portable pH meter. Anions (SO₄^2-, Cl^-, and NO₃^-) were quantified by ion chromatography (IC) after filtration through a 0.22 μm membrane.

Fe³⁺ and Mn²⁺ concentrations were determined by an atomic absorption spectrometer (AAS, SP-3803AA, Shanghai Spectrum, China): The instrument was equipped with hollow-cathode lamps for the respective elements, with lamp currents set according to the manufacturer’s recommendations. Fe and Mn were determined by flame atomic absorption spectrometry (FA). All reagents used were of analytical grade or higher, and ultrapure water was used throughout. Single-element standard stock solutions (1000 mg/L, TMRM®, China) were serially diluted with 1% HNO₃ to prepare calibration curves (Fig. S20). Samples were pretreated before analysis: 25 mL of the water sample was acidified with 1% concentrated HNO₃ to pH <2, diluted to a final volume of 50 mL, and stored at 4 °C. Before measurement, the samples were further diluted to an appropriate concentration with ultrapure water as needed. For the analytical conditions, flame AAS was performed using an air-acetylene flame, with gas flow rates optimized for maximum sensitivity. No matrix modifier was added. Calibration curves were established using the external standard method, with correlation coefficients (R²) exceeding 0.995 in all cases. Each sample was measured in triplicate, and results are presented as mean ± standard deviation. Full-process blanks were included throughout the experiment to correct for background signals.

HA was measured by UV-Vis spectrophotometry following standard extraction and calibration with commercial humic acid standards (Fig. S19). Light intensity (W/m²) was recorded on-site at the water surface using a photometer.

Sulfadiazine (SD) and tetracycline (TET) were measured using the following process. AW samples were filtered (0.7 µm glass fiber), and 500 mL of filtrate was subjected to solid-phase extraction. An HLB cartridge was activated sequentially with 12 mL each of pure water, methanol, and n-hexane. After pH adjustment, the sample was loaded at 25 mL/min. The cartridge was then eluted with 20 mL of methanol. The eluate was concentrated to near dryness under nitrogen and reconstituted in 10 mL acetonitrile. Chromatographic analysis was performed on an Agilent UPLC-1290 system equipped with a Waters BSH UPLC-C18 column (30 °C). Separation used a mobile phase of 0.1% formic acid, 1 mM ammonium acetate, and acetonitrile, with gradient elution at a flow rate of 0.2 mL/min. The injection volume was 10 µL. Detection was performed using a triple quadrupole tandem mass spectrometer (Shimadzu MSTQ-8050) with electrospray ionization in positive mode. The ion source temperature was 550 °C, the electrospray voltage was 5000 V, the auxiliary gas pressure was 65 Psi, the curtain gas pressure was 27 Psi, and the collision gas was nitrogen. The ion transfer tube was maintained at 560 °C. Multiple reaction monitoring was used with optimized parent/daughter ion pairs and collision energies for each antibiotic.

Concentrations below the limit of detection were reported as “n.d.” (not detected). All measurements were performed in technical duplicates, and the values presented in Table S1 represent the three replicates per pond.

Statistical analyses

Statistical analyses were performed using GraphPad Prism version 10.4.0.621, and data were presented as means ± SEM. For two paired samples, significance was determined by a nonparametric Wilcoxon test. To analyze differences among more than two independent groups, statistical analysis was performed using one-way or two-way analysis of variance (ANOVA), and significance was determined using Tukey’s multiple-comparison test or the Kruskal-Wallis test for nonparametric data, with Dunn’s multiple-comparison test (results were considered significant if the P value was less than 0.05). Sample-to-sample reproducibility was evaluated using Pearson correlation and principal component analysis (PCA) with the R packages Corrplot and Factoextra, respectively (correlation coefficients ranged from 0.94 to 1.0 across biological replicates).

Metagenomic sequencing

The raw metagenomic sequencing data were trimmed and filtered with FASTP v0.23.2 with default parameters. Then, the trimmed paired-end reads were analyzed using ARGs-OAP 3.2 with an alignment length cutoff of 75%, an alignment e-value cutoff of 10⁻⁷, and an alignment identity of 80%. Based on the interpretation of the comparability and biological relevance of the results, the ARG abundance was calculated as the number of ARG copies per cell using the following equation:

$$\mathrm{ARG}\mathrm{copy}\mathrm{per}\mathrm{cell}=\frac{{\sum }_{i=1}^n{R}_{iARGlikereads}/L_{iARGreferencesequence}}{\frac{1}{30}\times {\sum }_{k=1}^{30}{\sum }_{j=1}^m{R}_{jkmarker-likereads}/L_{jkmakergene}}$$ 1

where $R_{iARGlikereads}$ is the sum of aligned lengths of all reads assigned to the i_th ARG reference sequence; $L_{iARGreferencesequence}$ is the nucleotide sequence length of the i_th reference sequence in the ARG database; $R_{jkmarker-likereads}$ is the sum of aligned lengths of all reads that mapped to the j_th reference sequence in the k_th marker gene of essential bacterial (or prokaryotic) single-copy marker genes, where 30 genes are used in this study; $L_{jkmakergene}$ is the nucleotide sequence length of the j_th reference sequence in the k_th marker gene cluster⁴⁴.

Transwell-based HGT sorting system & evaluation methods (THSEM)

To evaluate the contribution ratio of various HGT pathways in the dissemination of ARGs, a Transwell-based HGT Sorting System & Evaluation Methods (THSEM) was established. The basic principle involves placing a transwell insert into a 24-well plate, with the upper and lower chambers filled with 400 μL and 1 mL, respectively, of a disinfection system containing the same disinfectant concentration. The two chambers are separated by a polycarbonate (PC) membrane with a pore size of 0.2 μm.

Thus, the THSEM system enables the physical separation of bacteria and the eDNA they release (Fig. 1a). Similar transwell-based approaches have previously been used to separate intercellular contact from extracellular signaling⁴⁵. Specifically, the upper chamber was inoculated with an equal concentration (1 × 10⁸ CFU/mL) of donor bacteria (E. coli S17-1 λpir [RP4 (Ampr)]) and indigenous recipient #1 (B. licheniformis A2.4). As these two strains can fully interact, conjugation can occur. The lower chamber was inoculated with indigenous recipient #2 (A. baylyi A3.6) at 5×10⁸ CFU/mL. Since the pore size of the PC membrane is insufficient to allow bacterial passage between chambers, but eDNA or cellular debris released from bacteria in the upper chamber can pass through the 0.2 μm pores, natural transformation can proceed in the lower chamber¹². All bacterial strains were guaranteed to be free of phage contamination through monoclonal sequencing and the double-layer agar plaque assay⁴⁶. Furthermore, all reaction solutions and culture media underwent a 30-min nucleic acid enzyme treatment (DNase I and RNase A, each at 5 μg/mL) followed by high-temperature sterilization to eliminate the possibility of phage transduction. Worth noting when using THSEM, the cations inherently present in the LB medium, particularly those from the 2% sodium chloride, promote DNA precipitation and translocation across membranes, thereby effectively minimizing eDNA in the upper chamber. The above setup selectively facilitated natural transformation in the lower chamber, where we inoculated exclusively with the type #2 recipient (A. baylyi A3.6).

After 24 h of incubation at 37 °C, bacteria from the upper and lower chambers were plated onto LB solid plates containing the corresponding antibiotics. Colony counts were obtained after overnight incubation at 37 °C, allowing quantification of transconjugants (recipient #1) and transformants (recipient #2). Ten single colonies from each antibiotic-resistant plate were selected for bacterial PCR and 1% agarose gel electrophoresis to confirm the presence of ARGs. The transfer frequency of ARGs via each pathway was finally calculated according to Eq. (2). Each THSEM experiment included negative and positive controls to ensure the absence of cross-contamination between chambers and proper system operation.

$$Frequency=\frac{transformant\left(CFU/mL\right)}{recipient1\#or2\mathrm{\#}colonies\left(CFU/mL\right)}$$ 2

The system has been validated using model strains. The settings in the model THSEM are as follows: The donor strain, E. coli S17-1 λpir [pUC19 (Ampr)], which was independently isolated and screened in our laboratory, carries the self-transmissible plasmid pUC19. This plasmid confers resistance to ampicillin (Amp) and streptomycin (Str). Recipient #1, E. coli DH5α [pET-28a (Kan)], provides resistance to kanamycin (Kan). To confer a controllable natural competence phenotype, the core DNA uptake genes from Neisseria gonorrhoeae (comP, pilE, pilF, pilQ, comE, comA) were codon-optimized, synthesized, and cloned downstream of the arabinose-inducible pBAD promoter in the chloramphenicol-resistant vector pBAD33. The resulting plasmid, pBAD33-ngo-comp, was transformed into E. coli MG1655 via heat shock, and transformants were selected on LB agar containing 34 µg/mL chloramphenicol. The resulting engineered strain, Recipient # 2, E. coli MG1655/pBAD33-ngo-comp (MG1655 (pBAD-ngo)), was verified by colony PCR and sequencing. The donor strain was cultured in LB liquid medium supplemented with 100 mg/L Amp and 10 mg/L Str, while recipient #1 was grown in LB broth containing 50 mg/L Kan. Based on the THSEM of model strains, we have successfully explored the effects of PMS/HA on model strains in our previous study (Fig. S3).

RNA-Seq transcriptome analysis

A. baylyi A3.6 was routinely cultured in LB broth at 37 °C with shaking (180 rpm) until the mid-logarithmic growth phase (OD₆₀₀ ≈ 0.6–0.8). Cells were harvested by centrifugation (5000 × g, 5 min, 4 °C) and resuspended in fresh LB medium to a final density of approximately 10⁸ CFU/mL. The bacterial suspension was then divided into treatment groups and exposed to the different conditions (PMS (10 ppm) / PMS + HA (20 mg/L) / CaClO₂ (12.5 ppm available chlorine) / CaClO₂ + HA (20 mg/L)) for either 5 min (short-term) or 60 min (long-term) at 37 °C:

Untreated cells (blank/control) served as the reference group. Each treatment group was performed in three independent biological replicates. After exposure, cells were immediately centrifuged (8000 × g, 2 min, 4 °C), washed once with ice-cold PBS, and snap-frozen in liquid nitrogen. Total RNA was extracted using the Bacterial RNA Extraction Kit (Magen, China) according to the manufacturer’s protocol, followed by DNase I treatment to remove contaminating genomic DNA. RNA concentration and purity were determined using a NanoDrop 2000 spectrophotometer (A₂₆₀/A₂₈₀ ratio 1.9–2.1), and RNA integrity was assessed by 1% agarose gel electrophoresis and Agilent 2100 Bioanalyzer (RIN ≥ 7.0). Only high-quality RNA samples were used for downstream analysis. Qualified RNA samples were sent to Paisenno Biotechnology (Shanghai, China) for strand-specific library preparation and sequencing. Briefly, ribosomal RNA was depleted using a bacterial rRNA removal kit, followed by fragmentation, cDNA synthesis, adapter ligation, and PCR enrichment. Libraries were sequenced on an Illumina NovaSeq 6000 platform with paired-end 150 bp reads, targeting approximately 6 Gb clean data per sample.

Raw FASTQ reads were quality-checked with FastQC (v0.11.9). Adapter sequences and low-quality bases (Phred score <20) were trimmed using Trimmomatic (v0.39). Clean reads were mapped to the reference genome of A. baylyi ADP1 (GenBank: NC_005966.1) using HISAT2 (v2.2.1) with default parameters. Gene-level read counts were quantified with featureCounts (Subread v2.0.3). Differentially expressed genes (DEGs) were identified using DESeq2 (v1.34.0) in R. Genes with |log₂ fold change | ≥ 1 and adjusted P-value (FDR) < 0.05 were considered significantly differentially expressed. Gene Ontology (GO) enrichment analysis of DEGs was performed with the R package clusterProfiler (v4.2.2) using the A. baylyi GO annotation database. Enriched GO terms in the categories “Biological Process”, “Cellular Component”, and “Molecular Function” were considered significant at FDR < 0.05. All statistical analyses and visualizations (volcano plots, bubble plots, heatmaps) were conducted in R (v4.2.2).

Observation of type IV Pili by transmission electron microscopy (TEM)

The morphology of type IV pili in recipient #2 was examined using TEM (Hitachi HT-7800). Briefly, healthy bacterial colonies were inoculated into LB broth without antibiotics in a 24-well plate and cultured at 37 °C to the mid-logarithmic phase⁴⁷. The test groups were treated by directly adding the following solutions onto the bacterial lawns: (1) Control (sterile 0.9% NaCl); (2) PMS (10 ppm); (3) PMS (10 ppm) co-supplemented with HA (20 mg/L). After 30 min of treatment at room temperature, the naturally dispersed, turbid supernatant from each well was carefully collected. The bacterial suspensions were then negatively stained with 0.9% NaCl on carbon-coated copper grids. Grids were air-dried and visualized using a TEM operated at 100 kV. For each group, at least 10 random fields of view were examined to assess the presence and morphology of type IV pili.

Natural transformation frequency assay

A. baylyi A3.6 was streaked on LB agar plates and incubated overnight at 37 °C to obtain fresh colonies. Colonies were gently resuspended in ice-cold sterile PBS by pipetting (avoiding scraping to minimize cell damage), and the density was adjusted to 1  × 10⁸ CFU/mL using a spectrophotometer (OD₆₀₀). One hundred microliters of this standardized cell suspension (containing approximately 10⁷ cells) were aliquoted into 1.5 mL microcentrifuge tubes. Purified pUC19 plasmid DNA was added to each aliquot to achieve final eDNA concentrations of 0 (no-DNA control), 0.1, 1, 2.5, 5, 10, or 100 ng/μL in the 100 μL transformation mixture⁴⁸. The plasmid was diluted in sterile nuclease-free water as needed, and the added volume was kept ≤10 μL to avoid significant dilution of the cell suspension. Tubes were gently mixed and incubated at 37 °C for 1 h to allow natural transformation⁴⁹. Following incubation, serial dilutions of the transformation mixtures were prepared and plated in parallel on non-selective LB agar plates (to determine total viable recipient cells) and on LB agar plates supplemented with ampicillin (100 μg/mL) (to enumerate antibiotic-resistant transformants). Plates were incubated at 30 °C for 24–48 h. Transformation frequency was calculated as Eq.( 2).

Degradation of ARGs (pUC19) by PMS/HA and subsequent quantification by qPCR

pUC19 plasmid DNA was used as the model ARG template (Sangon, China, Fig. S21). The reaction system contained 10 ng/μL pUC19, 10 ppm PMS, and 20 mg/L HA in nuclease-free water. After incubation under the designed reaction conditions, samples were collected at predetermined time points and immediately quenched with excess sodium thiosulfate to terminate PMS oxidation. Residual DNA was purified or appropriately diluted to minimize qPCR inhibition by HA and salts. The concentration of remaining ARGs was quantified by qPCR using primers targeting the pUC19-associated ARG fragment (Table S4). The residual ARG abundance was calculated from Cp values and expressed as DNA copy number or relative concentration relative to the initial sample.

Twitching motility assay

A. baylyi A3.6 was streaked on LB agar plates and incubated overnight at 37 °C to obtain fresh colonies. Colonies were gently resuspended in ice-cold sterile PBS by pipetting (avoiding scraping to minimize cell damage), and the density was adjusted to 1 × 10⁸ CFU/mL using a spectrophotometer (OD₆₀₀). Twitching motility plates were prepared by supplementing molten LB agar (1% w/v agar, cooled to 50 °C) with Tuspetinib (dissolved in DMSO; final DMSO concentration ≤0.5% v/v) at final concentrations of 0, 1.5, 3, 6, 12.5, 25, 50, and 100 μM. Control plates contained the same volume of DMSO vehicle. 1 μL of the standardized bacterial suspension was spotted onto the center of each plate and allowed to dry briefly under laminar flow. Plates were incubated at 37 °C for 48 h in a humidified chamber to prevent agar drying.

After incubation, the subsurface twitching zone (halo) surrounding the inoculation site was photographed under a stereomicroscope. The twitching area (cm²) was quantified using ImageJ software by outlining the outer edge of the visible twitching halo and calculating the enclosed area. Background “negative” twitching level (the dotted line in Fig. 2d) was obtained by coating 1 μL onto 220 rpm cultured plate for 24 h. On these culture plates, the twitching phenomenon disappeared completely. Dose-response curves were generated by plotting twitching area against log₁₀ [Tuspetinib] and fitted with a four-parameter logistic (Hill) equation using GraphPad Prism. The half-maximal inhibitory concentration (IC₅₀) was calculated to be 23.8 ± 0.4 μM (mean ± SD; n = 3 independent biological replicates, each with three technical replicates).

RT-qPCR assay

All the samples in the experiment were extracted using bacterial RNA extraction kits (Magen, China). RNA integrity and purity were tested by gel electrophoresis and Nanodrop. Subsequently, reverse transcription of the RNA was performed using the PrimeScript RT Reagent Kit (Takara, Japan) to obtain complementary DNA (cDNA). The cDNA extraction procedure was carried out according to the instructions. The target gene was then quantified by RT-qPCR (LightCycler480, Roche, Switzerland) with 16S rRNA as the internal reference gene, and relative quantification was calculated using the 2^ct method. The primers are shown in Table S3. A total volume of 20 μL qPCR mixtures consisting of 10 μL TB Green Premix Ex Taq, 0.8 μL of each primer, and 8.4 μL template DNA were used during the qPCR assay.

Cell membrane potential analysis

Cell membrane potential was assessed using DiOC₂(3) staining and flow cytometry. Cells from the Control, PMS, HA, and PMS/HA groups were harvested after the respective treatments, washed twice with PBS, and 1 × 10⁶ cells were resuspended in 500 μL of phenol red-free culture medium containing 40 nM DiOC₂(3). The suspensions were incubated at 37 °C in the dark for 30 min. After incubation, cells were washed once with PBS, resuspended in PBS, and supplemented with 1 μg/mL propidium iodide (PI) to exclude dead cells. Samples were immediately analyzed by the FongCyte 3 flow cytometer (Challenbio, China) equipped with a 488 nm laser. DiOC₂(3) green fluorescence was collected in the FITC channel (488/525 nm) and PI signal in the PerCP channel (488/680 nm), with a minimum of 10,000 events acquired per tube. Data were processed using FlowJo software: debris and doublets were excluded by FSC/SSC gating, live cells were defined as the PI-negative population, and the proportion of DiOC₂(3) high cells (high mitochondrial membrane potential) was quantified by quadrant gating. All experiments were performed in at least three independent replicates.

Process of DFT calculation

The calculations were performed at the B3LYP/6-31 G(d) theoretical level, combined with the SMD water solvent model to simulate the actual reaction environment; structural optimization and electronic structure calculations were completed with Gaussian 16, Fukui function and spin density analysis were performed with Multiwfn, and visualization was done with IQmol.

Process of FT-IR experiment

During the experiment, first, 4 reaction solutions containing HA were prepared, each with a volume of 50 mL and an initial HA concentration of 20 mg/L, and placed in beakers. Subsequently, 10 ppm PMS was added to each reaction system to initiate the reaction, which was carried out under simulated sunlight irradiation using a Xenon lamp (PLS-SXE300, PerfectLight) at 1 W/m². Samples were taken at 0, 5, 10, and 30 min after the start of the reaction, with 10 mL of the reaction solution taken each time and transferred to a brown sample bottle. After sampling, the samples were immediately placed in liquid nitrogen for quick freezing to terminate the reaction, then transferred to a –80 °C ultra-low temperature refrigerator for storage, and the entire dehydration process was protected from light. Afterward, the samples were dehydrated using freeze-drying technology (EYELA-FDU-2100). After the samples were freeze-dried, the FT-IR spectra of the samples at different reaction times were determined using a Thermo Scientific Nicolet iS50 Fourier transform infrared spectrometer combined with the KBr pellet method to analyze the characteristics of changes in the HA functional group structure during the reaction⁵⁰.

Process of MB degradation

A 5 mg/L MB solution and a 5 mg/L MB + 20 mg/L HA mixed solution were preemptively prepared. 5 mM PMS was added to initiate the reaction, and the reaction was carried out under simulated sunlight irradiation using a Xenon lamp (PLS-SXE300, PerfectLight) (light intensity = 1 W/m²). Samples were collected at 0, 20 s, 2 min, 5 min, 10 min, 20 min, 30 min, 60 min, 90 min, and 120 min after the start of the reaction. The absorbance at 595 nm was measured by UV-Vis spectroscopy (Shimadzu UV-2600), and the degradation curves for the different groups were recorded (see Fig. S15).

Flow cytometric analysis of cell membrane permeability

A. baylyi A3.6 was treated with 10 ppm PMS/12 ppm Calcium hypochlorite for 0.5 h. After washing 3 times (8000 rpm, 1 min), samples were stained with propidium iodide (PI; 30 μM, 15 min, dark). FongCyte 3 flow cytometer (Challenbio, China) was used to quantify membrane damage. PI fluorescence was detected using a 610/20 nm filter. At least 1,000,000 events were recorded per sample. The percentage of PI-positive cells was determined by comparing with unstained and untreated controls. All experiments included three biological replicates.

Minimum inhibitory concentration (MIC) assay

The MIC of Amp against recipients #1 and #2 was determined by broth microdilution. Briefly, a mid-logarithmic phase bacterial culture was standardized to approximately 1 × 10⁶ CFU/mL and exposed to a series of Amp concentrations (0, 0.01, 0.05, 0.1, 0.2, 0.5, 1, 5, 10, 25, 50, 100 μg/mL) in a 96-well plate. The plate was incubated statically at 37°C for 16-20 h. The MIC was defined as the lowest concentration that prevented visible bacterial growth. All experiments were performed in triplicate.

Nematode cultivation and physiological indicator assessment

The wild-type Caenorhabditis elegans N2 strain was procured from GCC and routinely maintained at 20 °C on nematode growth medium (NGM) plates seeded with E. coli OP50. To evaluate the stress response of PMS on invertebrates, the experiment was conducted in two stages. In the first stage, 2000 synchronized L1-stage nematodes were co-incubated with 10⁸ CFU/mL antibiotic-resistant bacteria (ARB) in 300 mL of 0.9 % (w/v) NaCl solution at 20 °C with constant shaking at 200 rpm for 24 h until the nematodes reached the L3 stage. In the second stage, the reaction system was prepared according to the conditions described in section 2.5, supplemented with 20 mg/L HA to simulate advanced water treatment conditions, and exposed to 1.5 h of illumination to activate PMS. After an additional 24 h of culture, two 50 mL aliquots were withdrawn and centrifuged at 2000 rpm for 1 min to obtain the nematode-ARB mixture. One aliquot of the mixture was placed on sterile NGM plates and incubated for 2 h; the number of surviving nematodes was counted under a stereomicroscope (NSZ818, Novel, Ningbo) to calculate the mortality rate. The second aliquot was fixed with methanol, and the body length of individual nematodes (n > 20) was measured under a microscope.

Accumulation of ARGs in nematodes

The accumulation of ARGs within nematodes was determined by qPCR quantification and intestinal fluorescence intensity analysis. Two 50 mL aliquots of the reaction suspension (as described in Nematode Cultivation and Physiological Indicator Assessment) were collected. One aliquot was centrifuged at 2000 rpm for 1 min; the pellet was subjected to DNA extraction followed by qPCR amplification of the nematode reference gene tba-1 to quantify the number of nematodes, while the supernatant was extracted for DNA and analyzed by qPCR targeting the Ampr gene to determine extracellular ARG abundance (Fig. S18). The second 50 mL aliquot was directly subjected to total DNA extraction to measure the overall ARG content. The amount of ARGs sheltered per nematode was calculated using the following equation:

$${\mathrm{ARG}}_{\mathrm{worms}}=\frac{{\mathrm{ARG}}_{\mathrm{total}}-{\mathrm{ARG}}_{\mathrm{supernatant}}}{\mathrm{Number}\mathrm{of}\mathrm{worms}}$$ 3

To visualize the sheltering of ARGs (or ARB) inside nematodes more intuitively, ARB were stained with SYTO™ 9 Green Fluorescent Nucleic Acid Stain (Thermo Fisher Scientific). Adult nematodes together with the fluorescently labeled ARB were introduced into the reaction system, with PMS exposure limited to 30 min to allow observation of intact nematodes. Whole nematodes were imaged using a confocal fluorescence microscope under identical exposure intensity and duration. Average fluorescence intensity within the nematode intestine was quantified using Imaris software.

Empirical foundation of PMS applications in aquaculture and water treatment (literature search and study selection)

A comprehensive literature search was conducted using Web of Science, Scopus, and Google Scholar to assess the practical applications of PMS in water treatment, with a particular focus on aquaculture. The search covered publications up to early 2025, using combinations of keywords such as (“peroxymonosulfate” OR “PMS” OR “HSO₅^-”) AND (“aquaculture” OR “aquaculture wastewater” OR “fish farming” OR “shrimp” OR “disinfection” OR “pathogen” OR “pilot” OR “economic”). Initial searches yielded more than 1,100 records. After duplicate removal and preliminary screening of titles and abstracts, 178 potentially relevant articles were selected for full-text evaluation. Ultimately, 30 high-relevance studies were included in this table (Table S1). These studies were systematically categorized into five thematic groups in Supplementary Table S1: (1) PMS applications in aquaculture water treatment (refs. ^1–6); (2) PMS applications in surface water and groundwater treatment (refs. ^{7–9,51–53}); (3) PMS applications in municipal and industrial wastewater treatment (refs. ^10–14,54); (4) pilot-scale or real aquaculture system applications (refs. ^15–20); and (5) techno-economic assessments and life-cycle evaluations of PMS-based processes (refs. ^21–26). This curated collection of 30 studies provides a solid empirical foundation for the growing practical relevance and application potential of PMS in aquaculture and broader water-treatment contexts.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

F. Z., D. L., and H. G. designed research; F. Z., S. P., C. M., S. W., and Y. Y. performed research; X. S., D. L., H. G., and J. X. analyzed data; H. G., D. Z., C. L., J. Z, and J. G wrote the paper.

Peer review

Peer review information

Nature Communications thanks Hanmin Zhang who co-reviewed this paper with Mengbo Cao, and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the Zhejiang Provincial Natural Science Foundation of China (LMS26E090008 to D.L.), the National Natural Science Foundation of China (52070103 to H.G.), and the Ningbo Youth Leading Talent Project (2024QL038 to H.G.).

Data availability

The metagenomic and RNA-Seq transcriptomic data generated in this study have been deposited in the NCBI Sequence Read Archive (SRA) under accession number PRJNA1347934 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1347934). All other data supporting the findings of this study are provided in the Source Data file. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76727-6.