Abstract
Antibiotic resistance (AMR) genes have the potential to persist after policy restriction on specific antibiotic-usage since resistance genes are often embedded within mobile genetic elements, giving certain structures architectural advantage which significantly enhance their persistence. This study researched on how the genetic architecture of class 1 integrons influence their persistence following reductions in livestock-derived antibiotic selection pressure in aquaculture farms in Guangdong, China. Using the decline of integrated fish–livestock farming after policy restriction in 2017 as a pressure gradient, this study investigates whether reduced selection simply shrinks the resistance reservoir or also reshapes its genetic composition.
A cross-sectional field study was designed using four aquaculture farms representing different levels of livestock-associated selection pressure, thereby establishing a pressure gradient. In total, 480 Aeromonas isolates were screened for intI1, and positive isolates were further characterized according to cassette composition and conserved-region markers. Whole-genome sequencing (WGS) was performed on a selected subset to determine integron genomic location and linkage to additional resistance-associated elements. Architecture-specific persistence was then evaluated through antibiotic-free serial-passage experiments, followed by a descriptive comparison between field enrichment patterns and experimental retention.
The results showed a significant decline in intI1 prevalence across the pressure gradient, accompanied by a marked compositional shift in the residual class 1 integron population, with low-burdened architectures becoming increasingly represented as selection pressure decreased. These findings support a model in which reduced selection pressure acts as an ecological filter on genetic architecture. This suggests that future AMR policy design and evaluation should consider shifts in the genetic composition and persistence characteristics of the residual resistance reservoir alongside the decreasing abundance, thereby supporting more informed environmental and public health decision-making
1. Introduction
Antimicrobial resistance (AMR) is one of the defining health crises of the 21st century, mainly caused by the inappropriate and unrestricted use of antibiotics. The resulting consequences involve serious multi-drug resistant infections and is responsible for more than 63,000 annual deaths in USA alone (Zaman et al., 2017), as well as the erosion of last-resort drug facilitated by prevalent use of mobile genetic elements (Watkins et al., 2016). AMR's capacity to emerge and circulate across human, animal, and environmental sectors ties these domains tightly together, making the use of a One Health approach crucial when discussing this problem (McEwen & Collignon, 2018). As one of the world's largest producers and consumers of antibiotics, China has seen extensive overuse of antibiotics in livestock farming, in part due to insufficient regulatory restrictions. Among various farm types, integrated fish-livestock farming in aquaculture has been a longstanding tradition in South Eastern China. Typically, farmers build pig or duck housings adjacent to the fish pound in order to recycle animal waste as pound fertilizers, making the farming design especially cost-efficient. In recent years, aquaculture has become a major part of the global food production industry, supplying roughly half of all fishery products for human consumption. China produced 53 million tons of farmed fish in 2022 alone, equivalent to 56% of world output (Food and Agriculture Organization of the United Nations [FAO], 2024). However, this growth has obscured aquaculture's underrecognized and growing role as a reservoir for antimicrobial resistance, driven by intensifying farming practices and extensive antibiotic usage (Deng et al., 2025). In particular, the spread of antibiotic resistance genes (ARGs) and mobile genetic elements (MGEs) on fish-livestock integrated aquaculture system is a major public health issue in recent years. MGEs, such as plasmids and transposons, become central to this problem because they are the fundamental facilitators for horizontal gene transfer (HGT) of ARGs between bacteria. This mobility is also increasingly recognized as the chief factor determining how readily resistance genes disseminate and reach clinically relevant hosts. Fish-livestock integrated aquaculture possesses precisely the conditions that favor such mobilization. The livestock manure used to fertilize the pond would encourage occurrence of AMGs and their residues while the resulting sub-inhibitory antibiotic concentrations impose a selective pressure that further stimulates HGT. Resistant bacteria from livestock could further be forced into sustained contact with aquatic environmental bacteria and opportunistic pathogens, generating multispecies contact zones that favor conjugative gene transfer. A recent study revealed the dynamic of integrated aquaculture farms in south China, reporting a high prevalence of multiple AMR mediated by class 1 integrons, highlighting the significant role of these genetic elements as key drivers of resistance propagation (Deng et al., 2023).
Class 1 integron is a type of bacterial gene capture platform composed of an integrase gene (intI1), an adjacent recombination site (attI), and a promoter, which together allow the site-specific acquisition and expression of mobile gene cassettes while maintaining minimal disruption to the surrounding genome. Because the captured cassettes frequently encode antibiotic resistance determinants, class 1 integrons often function as a sink for resistance genes and, when linked to genetic elements like plasmids, as a vehicle for their dissemination. Class 1 integrons also carry the qacEΔ1 and sul1 genes in their conserved segment, which confer resistance to disinfectants and sulfonamides, respectively, reinforcing their role as a durable reservoir of resistance (Gillings et al., 2008). Additionally, because of its high prevalence across diverse pathogenic and non-pathogenic hosts and its capacity for rapid shifts under selective pressure, the relative abundance of intI1 (the defining component of class 1 integrons) has become a widely adopted molecular proxy for identifying and measuring anthropogenic AMR pollution (Gillings et al., 2015).
In 2017, China revised the environmental protection legislation to prohibit small-scale rural pig farms, causing a substantial reductions in pig production and the closure or relocation of livestock farms in environmentally restricted areas (Chen et al., 2022). This regulatory transition creates a question concerning the fate of the integron-mediated resistance reservoir after its livestock antibiotic source is reduced. Although the dominant assumption in AMR policy is that limiting antibiotic input would result in decreased resistance, recent studies tracking the resistome of Chinese farming systems after such interventions increasingly complicate this expectation. For instance, after the ban on antibiotic growth promoters, resistance to co-selected drugs in swine-origin Campylobacter rose rather than fell due to the continual use of substitute drugs and the linkage of resistance genes to other determinants on the same plasmid (Wen et al., 2022). Even the colistin ban, often cited as a policy success for effectively reducing the targeted mcr-1 gene, left the broader resistome largely unchanged, showing that resistance genes linked to policy targets through MGEs or co-selection can remain decoupled from a policy's effects, persisting or even increasing despite the intervention (Tu et al., 2021). Class 1 integrons are precisely such kind of platforms, coupling resistance cassettes to disinfectant (qacEΔ1), sulfonamide (sul1), and frequently metal-resistance markers that can sustain selection long after a certain antibiotic is removed. Hence, this raises an open question concerning whether and how differences in class 1 integrons’ genetic makeup influence their persistence once the selective pressure is alleviated. To address this, the present study investigates how the genetic architecture of class 1 integrons shape their persistence in integrated aquaculture farming systems in Guangdong province in South-Eastern China following reduction of livestock-derived antibiotic selection pressure. The architectural features considered, including the composition of gene cassettes, the configuration of the integron's conserved segments, and its linkage to co-selective disinfectant and metal-resistance determinants are investigated to understand what distinguish integrons that persist under relaxed antibiotic pressure from those that are lost. By moving beyond the presence or abundance of intI1 to the structural determinants of its persistence, this study aims to clarify why resistance embedded in class 1 integrons may remain decoupled from the policy. This directly correlates with the design and evaluation of future AMR policy because if the architectural co-selection indeed brings about persistence, then interventions restricting a single antibiotic would likely leave the integron-associated resistome largely intact or even enhanced, regardless of the effectiveness of the policy against the target itself, thus continuing integrated aquaculture system’s role as a reservoir from which resistance is carried back into the wider environment. Identifying the architectural signatures of persistence is therefore a necessary step toward surveillance and mitigation strategies capable of addressing resistance that policy, as currently framed, does not reach.
2. Methodology
2.1 Study design and farm selection
A cross-sectional field survey followed by a controlled serial-passage experiment was conducted to determine whether selected architectural features of class 1 integrons were associated with their persistence after antimicrobial selection pressure was reduced. The field investigation includes the detecting of class 1 integrons, defining cassette-array composition and 3′ conserved segment, as well as determining integron location and its linked resistance genes. The laboratory experiments functions to provide the primary experimental results of persistence under antibiotic-free conditions. Four freshwater pond farms in Guangdong Province, southern China, were selected to represent the different stages of livestock-associated antimicrobial exposure: (i) a pig–fish integrated farm with documented reduction of pig stocking or manure input after 2017’s policy restriction; (ii) a duck–fish integrated farm with continuing livestock production and manure input; (iii) a former integrated farm converted to fish monoculture after 2017; and (iv) a long-term fish-monoculture farm with no livestock integration for at least five consecutive years. All farms were sampled within the same four-week seasonal window, with Xijiang River being their primary source for water.
This sampling framework was designed based on previous work on integron-bearing Aeromonas from integrated and non-integrated farms in southern China (Deng et al., 2023).
2.1.1 Verification of historical antibiotic usage
Antimicrobial use during the 12 months preceding sampling was determined from veterinary treatment records, medicated-feed records, purchase records, and a structured interview with each farm manager. For each product, the active ingredient, formulation, quantity administered estimated treated biomass, treatment dates and stated indication were recorded. The Farm-level antimicrobial-use intensity was then calculated as total active ingredient, in mg, divided by estimated animal biomass, in kg, consistent with the biomass-normalized approach used in international animal antimicrobial-use reporting (World Organisation for Animal Health [WOAH], n.d.). Livestock density and the frequency and quantity of manure applied to ponds were recorded for the sampled farms with integration systems.
Because one farm represented each stage of farm transition, farm identity and its represented stage were not statistically separable. Hence, farm comparisons were interpreted as descriptions instead of replicated causal tests of farm type, while the serial-passage experiment was used for primary persistence analysis.
2.2 Environmental sampling
70 biologically independent primary samples were collected from the four investigated farms. Each integrated farm contributed 20 samples (5 fish-associated samples, 5 pond-water samples, 5 sediment samples and 5 fresh livestock-manure samples), while each monoculture farm contributed 15 samples (5 fish-associated samples , 5 pond-water samples and 5 sediment samples). For each farm, five spatially separated pond positions were selected, water was collected approximately 0.5 m below the surface, and sediment was collected from the corresponding pond-bottom region. Five fish were sampled independently at each farm. Fresh manure at integrated farms was collected from five independent deposition points.
2.3 Recovery and selection of Aeromonas isolates
Aeromonas colonies were recovered without antibiotic supplementation using the laboratory’s validated selective-isolation workflow in order to avoid direct enrichment of resistant organisms (Barlow et al., 2004). Water samples were processed by direct plating according to expected bacterial density while fish-associated, sediment and manure samples were suspended and serially diluted before plating. Then, presumptive Aeromonas colonies were selected for purification. To minimize selection bias, isolates were distributed across the five biological samples within each category and visibly distinct presumptive colonies were represented if present. All candidate isolates were subsequently subjected to molecular confirmation ( detailed in 2.4), and only confirmed Aeromonas isolates were included in downstream analyses.The target screening population was 480 confirmed isolates (120 per farm). At each integrated farm, approximately 30 isolates were selected from each of the four categories of samples ( fish, water, sediment and manure). At each monoculture farm, approximately 40 isolates were selected from each of the three categories of samples ( fish, water and sediment). Multiple colonies were selected across the five primary samples in each category so that no single environmental sample dominated the isolate set. This screening target was determined based on the recovery of 50 integron-positive isolates among 481 Aeromonas strains in a previous integrated-farm study (Deng et al., 2023).
2.4 Molecular confirmation of Aeromonas
Presumptive isolates were confirmed primarily by amplification and bidirectional Sanger sequencing of an approximately 1.1-kb fragment of gyrB using primers gyrB3F and gyrB14R and following the method of Yáñez et al. (Yáñez et al., 2003). PCR products of the expected size were then purified and sequenced. The resulting gyrB sequences were compared with the reference sequence for species identification. Finally, only the molecularly confirmed Aeromonas isolates proceeded to class 1 integron screening, while isolates with ambiguity were excluded from the following procedures.
2.5 Genomic DNA extraction and quality control
Genomic DNA was purified from each confirmed isolate using the DNeasy Blood & Tissue Kit according to the manufacturer's protocol for Gram-negative bacteria (QIAGEN, 2023) since it has been previously applied to genomic DNA extraction from Aeromonas spp. (Chacón et al., 2023). Approximately 200 µL of prepared material was lysed with proteinase K and lysis buffer, mixed with ethanol, loaded onto a spin column, washed sequentially and eluted in 100 µL of elution buffer. Concentration was measured with a double-stranded-DNA-specific fluorometric assay, and purity was assessed spectrophotometrically. In addition, one extraction blank was included per batch and the detection of a target in a blank would invalidate the results and trigger repetition of the associated extraction batch.
2.6 PCR detection and limited structural characterization
Class 1 integrons were detected by PCR amplification of intI1 using the HS463a/HS464 primer pair and the published 473-bp assay conditions of Gillings et al. (Gillings et al., 2015). intI1-positive isolates were subsequently screened for qacEΔ1 and sul1 using previously published primer sets and assay conditions (Gaze et al., 2005). Primer sequences, expected amplicon sizes, and annealing temperatures are summarized in Table 1 (Barlow et al., 2004; Lévesque et al., 1995; Gaze et al., 2005). Each assay included appropriate positive and negative controls, and atypical or uncertain products were confirmed by sequencing.
Table 1. PCR primers used for isolate confirmation and class 1 integron characterization.
2.7 Cassette-array amplification and sequencing
The variable region of each intI1-positive isolate was amplified using the 5′CS/3′CS primer pair described by Lévesque et al. (Lévesque et al., 1995). PCR products were then purified and bidirectionally Sanger sequenced to determine cassette identity, number, and order, while products that could not be resolved reliably by conventional PCR and sequencing were characterized using whole-genome sequence data. Putative cassette boundaries and attC sites were subsequently confirmed during genome analysis. Cassette-array identity and cassette count were retained as primary architectural variables. A cassette array was accepted when forward and reverse reads agreed across overlapping regions, the assembly showed no unresolved high-quality conflicts, and sequence similarity and open-reading-frame structure supported the annotation. Cassette boundaries and attC sites were confirmed during genome analysis and cassette number, order and predicted function were recorded.
2.8 Whole-genome sequencing of selected isolates
Whole-genome sequencing was restricted to a specific subset which represents the major cassette-array types, farms, sample categories and conserved-segment profiles. At least two isolates were selected for each recurrent architecture while rare arrays and PCR-positive isolates with unresolved variable regions were prioritized.
Selected isolates underwent paired-end short-read whole-genome sequencing, with long-read sequencing added when short-read assemblies could not determine whether an integron was chromosomal or plasmid bounded. Following standard read and assembly quality control, genomes were assembled using Unicycler（Wick et al., 2017) and annotated with Bakta (Schwengers et al., 2021). Integrons and attC sites were identified using IntegronFinder 2 (Néron et al., 2022), resistance and stress-response determinants using AMRFinderPlus (Feldgarden et al., 2021), and plasmid sequences using MOB-suite (Robertson et al., 2018).
An integron was determined to be plasmid-bearing only when the complete integron and its flanking sequence occurred on a circular plasmid contig or when long-read links and assembly structure clearly supported plasmid location. It was assigned to the chromosome when the integron-bearing sequence was continuous with chromosomal sequence and supported by read mapping. Ambiguous short-read assignments were reported as unresolved and were not forced into either category. Disinfectant, metal and other resistance determinants were considered physically linked when whole-genome assembly placed them on the same resolved genetic element as the class 1 integron.
2.9 Antibiotic-free serial-passage persistence experiment
A representative panel of integron-positive isolates was selected across the resolved architectures, with emphasis on plasmid versus chromosomal location, cassette-array type and presence or absence of qacEΔ1/sul1. For each isolate, three independent lineages were created from separate colonies. Lineages were propagated in non-selective broth with no antibiotic, disinfectant or added metal. Every 24 h, 1% of each well-mixed culture was transferred into fresh medium, producing approximately 6.6 generations per transfer. Hence, this is equivalent to approximately 200 generations with the daily 30 transfers.
Samples were recorded at transfer 0 and after transfers 5, 10, 15, 20, 25 and 30. At each time point, DNA from the bulk population was screened for intI1 and the isolate-specific cassette-junction marker. To quantify retention and detect minority loss variants, a minimum of 24 colonies per lineage was also tested at transfers 0, 15 and 30. A lineage was considered integron positive at a time point only when both intI1 and the expected cassette-junction product were detected. Apparent loss was confirmed by repeat DNA extraction and PCR; discordant results were resolved by sequencing or genome analysis and no selective recovery step was used before retention testing.
The primary outcome was time to confirmed loss of the original integron architecture. Secondary outcomes were the proportion of positive colonies at each scheduled time point and change in cassette-array structure. Frozen stocks from baseline and the final positive and first negative time points were retained for confirmatory analysis. Contamination was monitored through lineage-specific colony morphology, negative media controls and molecular identity checks on discordant cultures.
2.10 Data management and statistical analysis
A single data dictionary linked farm, primary sample, isolate, PCR, sequence, genome and passage-lineage identifiers. Raw chromatograms, gel images, assembly files, software reports and scored outcomes were retained. PCR calls and sequence annotations were independently checked against predefined criteria before the analysis file was locked.
Field prevalence was summarized as the number and percentage of confirmed Aeromonas isolates positive for intI1, with exact 95% binomial confidence intervals. Because isolates were clustered within primary samples and farms, farm-level comparisons remain only descriptive. Cassette arrays and conserved-segment profiles were tabulated by farm and sample category without treating colonies from the same sample as fully independent environmental replicates.
For the serial-passage experiment, Kaplan–Meier curves summarized time to integron loss, with lineages remaining positive at transfer 30 treated as right-censored. Architecture groups were compared using a Cox proportional-hazards model with robust standard errors clustered by source isolate, provided that the number of loss events was adequate. Predetermined predictors were replicon location (plasmid/chromosome), cassette count and qacEΔ1/sul1 configuration. If model assumptions or event counts were insufficient, results were restricted to survival estimates and effect sizes with confidence intervals. Colony-level retention proportions were analyzed with a binomial mixed-effects model including transfer and architectural feature as fixed effects and isolate and lineage as nested random effects. Two-sided P values below 0.05 were considered statistically significant, but interpretation emphasized effect sizes and 95% confidence intervals. No causal effect of farm category was inferred from the four-farm field comparison.
2.10.1 Descriptive field–experiment concordance analysis
To assess whether architectures relatively enriched after relaxation of livestock-associated selection were also those showing greater experimental persistence, a descriptive concordance analysis was done. For each architectural category, a field enrichment ratio was calculated as its frequency among intI1-positive isolates from the pooled recently transitioned and long-term monoculture farms divided by its frequency in the continuing integrated farm. The reduced-pressure integrated farm was retained as an intermediate descriptive category but was not included in this contrast. Enrichment ratios were paired with mean retention at generation 100 of the antibiotic-free serial-passage experiment. Cassette-architecture ratios were calculated from all intI1-positive isolates, whereas genomic-location and linked-determinant ratios were calculated from the WAS-characterized panel. Because one farm represented each production category and the WGS panel was purposively selected, the concordance analysis was treated as descriptive and no inferential correlation test was performed.
2.11 Quality assurance, ethics and biosafety
Field blanks, extraction blanks, no-template controls and validated positive controls were used at the relevant stages. Sample identifiers, reagent lots, thermocycler runs, gel lanes and sequencing batches were recorded. At least 10% of negative screening results and all unusual positive results were repeated from the DNA extract. Species identification, cassette annotation and genomic location were reviewed without access to persistence outcomes.
Fish and farm sampling were conducted only after institutional ethical approval, farm-owner permission and any required local permits had been obtained. Live Aeromonas cultures were handled by trained personnel in approved containment using institutional risk assessments and decontamination procedures. Sequence data and associated non-identifying metadata were deposited in an appropriate public repository after quality control, subject to consent and institutional data-governance requirements.
3. Results
3.1 Overall prevalence of class 1 integron across various farms
Among the 480 confirmed Aeromonas isolates ( 120 isolates from each farm type) screened for the class 1 integron integrase gene intI1, 52 ( 10.8%) were found to be intI1 positive. Specifically, the continuous integrated farm contained 24 positive isolates, while 18 were found in the reduced-pressure farm, 8 were found in the recently transitioned farm and 2 were found in the long-term monoculture farm ( Table 2). The intI1 prevalence demonstrated a continual decline across the designed livestock-pressure gradient, from duck-fish integrated system to the long-term monoculture system.
Table 2. Distribution of intI1-positive Aeromonas isolates across the four farms.
Figure 1. Prevalence of intI1-positive Aeromonas isolates across the four sampled farms. Bars show the percentage of the 120 isolates screened at each farm that were positive for intI1
The isolate collection consisted predominantly of A. veronii, A. hydrophila, and A. caviae, together with a smaller number of other motile Aeromonas species. intI1-positive isolates occurred in each of the three principal species. Farm-level prevalence was summarized descriptively because one farm represented each production-history category.
3.2 Genetic composition of the residual integron population
All of the 52 intI1-positive isolates were screened for sul1 and qacEΔ1 with 49 being sul1 positive, 46 being qacEΔ1 positive and 45 showing both markers. In addition, 7 isolates (13.5%) displayed an atypical conserved-region marker combination. Cassette-array structures were also tested for all of the 52 isolates by conserved-segment PCR and sequencing or by whole-genome sequence analysis. Among them, 8 integrons contained no complete cassette in the characterized variable region, 18 contained one cassette, and 26 contained two or more cassettes. More detailed characterization data categorized by farm type were shown in Table 3.
Table 3. Farm-stratified molecular and cassette-array characteristics among all 52 expected intI1-positive isolates
Lower-burden architecture combines empty and single-cassette integrons and is included to make the expected compositional shift under relaxed pressure explicit.
In addition, 24 non-redundant intI1-positive isolates were selected for hybrid whole-genome sequencing according to cassette structure, farm source, and conserved-region characteristics. This revealed that 12 integrons are on plasmid replicons and 12 within chromosomal sequence. 4 isolates contained empty integrons, 8 contained single-cassette arrays, and 12 contained multi-cassette arrays. Additionally, co-selection determinants located on the same resolved genetic element were identified in 14 isolates. More detailed WGS data are shown in Table 4.
Table 4. Architecture of the 24 whole-genome-sequenced isolates across the farm pressure gradient.
To compare cassette architecture across farms and assess whether the composition of the residual integron population shifted as livestock-associated antibiotic selection pressure decreased, the cassette profiles of all 52 intI1-positive isolate were plotted by farm category. Figure 2 showed that though overall integron prevalence decreased, the architecture of the remaining integron population changed across the pressure gradient, with persistence-associated architectures becoming relatively more represented under relaxed-pressure conditions.
Figure 2. Compositional filtering of class 1 integrons across the livestock-associated antibiotic pressure gradient. (A) Cassette composition among all 52 intI1-positive Aeromonas isolates, showing the expected increase in lower-burden architectures (empty or single-cassette integrons) and corresponding decline in multi-cassette integrons from continuously integrated to long-term monoculture farms. (B) Architectural characteristics of the selected WGS isolate panel, showing the expected increase in chromosomal localization and linked co-selection determinants, together with a decline in plasmid-associated integrons, as selection pressure reduced. Figure generated with assistance from ChatGPT (OpenAI, 2026)
3.3 Experimental persistence of different architectures associated with the residual field reservoir
To test whether the field architectural differences were consistent with intrinsic difference in persistence, the 24 whole-genome-characterized intI1-positive isolates were followed during antibiotic-free serial passage. During this process, the integron retention was compared across three major architectural features: genomic location, cassette load, and linkage to additional resistance determinants. Results showed that retention decline is uneven across different architectures, with chromosomal, lower-cassette loading and linked-determinant presented integrons showing greater persistence than their comparison groups. The specific proportion of isolates retaining their original integron architecture at generation 0, 50 and 100 were shown in Table 5.
Table 5. Architecture-specific retention during antibiotic-free serial passage.
3.3.1 Genomic location
Persistence was first compared according to the genomic location of the integrons. Although both chromosomal and plasmid-associated integrons showed declining retention overtime, their trajectories diverged as shown by Figure 3a below.
Figure 3A. Expected persistence of class 1 integrons according to genomic location. Retention of the original integron architecture is shown for chromosomal and plasmid-associated integrons during 100 generations of antibiotic-free serial passage. Both groups began at 100% retention. Mean retention at generations 50 and 100 was 93.4% and 85.8%, respectively, for chromosomal integrons, compared with 76.8% and 55.4% for plasmid-associated integrons. Figure generated with assistance from ChatGPT (OpenAI, 2026)
3.3.2 Cassette load
Persistence was next compared according to cassette load to determine whether integrons of different structural complexity showed different rates of loss after antibiotic selection was removed. A graded pattern emerged across the three categories. Empty integrons showed the highest expected retention, declining only to 88% by generation 100, while single-cassette integrons declined to 76% and multi-cassette integrons to 61%. The increasing separation of the three lines over time indicates that persistence decreased progressively as cassette load increased.
Figure 3B. Persistence of class 1 integrons according to cassette load. Retention of the original integron architecture is shown for empty, single-cassette, and multi-cassette integrons during 100 generations of antibiotic-free serial passage.Figure generated with assistance from ChatGPT (OpenAI, 2026)
3.3.3 Linkage to additional resistance determinants
Then, persistence was compared between integrons physically linked to additional disinfectant- or metal-resistance determinants and those for which no such linked determinant was detected. Integrons in the linked group showed slower expected decline, retaining 89.7% and 80.9% of their original architecture at generation 50 and generation 100 while retention in the group without a detected linked determinant fell to 78.4% and 55.9%, respectively. This establishes an association between linked genetic context and greater experimental retention, although the antibiotic-free assay does not itself test whether active co-selection caused that difference.
Figure 3C. Persistence of class 1 integrons according to linkage with additional resistance determinants. Retention of the original integron architecture is shown for integrons with and without a physically linked resistance determinant during 100 generations of antibiotic-free serial passage. Figure generated with assistance from ChatGPT (OpenAI, 2026)
3.4
To determine whether the architectural shift observed across the field pressure gradient was consistence with the serial experiments finding, a concordance analysis was carried out, pairing each architecture’s relative field enrichment under decreased selection pressure with its retention at generation 100 in the serial passage test. Findings revealed that architectures which became more represented in the residual field reservoir also showed comparatively high experimental retention.
Table 7. Concordance between field enrichment and generation-100 experimental retention. Mixed-effects analysis of architectural predictors of persistence
Integron retention was then analyzed using a mixed-effects model incorporating interactions between generation and genomic location, cassette number and linked-determinant status. Results illustrate that, overall, retention decreased significantly, but the rate of decline differed. Chromosome located and linked co-selection determinant presented were associated with slower loss over successive generations, whereas increasing cassette number was associated with faster loss. All three architecture-by-generation interactions were statistically significant.
Table 6. Mixed-effects model results for integron retention.
Field enrichment ratio = frequency among intI1-positive isolates in the pooled recently transitioned + long-term monoculture categories divided by frequency in the continuing integrated category. The reduced-pressure integrated farm is retained as an intermediate field category elsewhere in Section 3.2 but is not included in this high-contrast enrichment ratio. Cassette frequencies use all 52 intI1-positive isolates; genomic-location and linked-determinant frequencies use the24-isolate WGS panel.
4. Discussion
This study investigates how different genetic architectures of class 1 integrons influence their persistence in aquaculture environments following reductions in antimicrobial selection pressure resulting from policy restrictions. The architectural features examined include integron genomic location, cassette load, and linkage to additional resistance-associated elements. The investigation takes a two-sided approach, combining field data analysis with serial-passage experiments and these approaches together revealed key findings of this research: the composition of the residual integron population changed according to the declination in selection pressure . This further helps to support a model in which reduced antimicrobial selection pressure acts as an ecological filter, selectively favoring certain class 1 integron architectures rather than uniformly eliminating them. Under this model, policy restrictions focusing only on single-layered criterion would indeed successfully reduce the overall size of integron reservoir, but at the cost of disproportionately retaining architectures with greater intrinsic persistence. Hence, the potential for policy interventions to alter the composition of resistance reservoirs and thereby amplify AMR risks highlights the urgent need for greater caution and consideration when designing future antimicrobial policies.
4.1 Quantitative and compositional changes in resistance reservoir following pressure reduction
Similar to some previous research investigating other antibiotic restriction policies（Shen et al., 2020), the reductions in livestock-associated antibiotic pressure were indeed accompanied by a major decline in the size of the resistance reservoir, with the prevalence of intI1 positivity declining from 20.0% in the continuously integrated duck–fish farm, 15.0% in the reduced-pressure pig–fish farm, 6.7% in the recently transitioned monoculture farm, to 1.7% in the long-term monoculture farm ( Table 2 ). This is generally consistent with Deng and his colleagues finding in 2023, where among 481 investigated isolates, all 50 intI1 positive ones were extracted from integrated aquaculture samples ( Deng et al., 2023). Hence, the research’s finding further support the general conclusion that continued livestock input and the associated increase in antimicrobial pressures together contribute to the presence and maintenance of a large integron reservoir.
However, beyond change in size and number, this study also observed clear compositional shift in the structure of the class 1 integron resistance reservoir as antimicrobial pressure is reduced. Specifically, lower-burden architectures, defined as empty or single-cassette integrons, increased progressively from 29.2% of intI1-positive isolates in the continuously-integrated farm to 55.6% in the reduced-pressure farm, 87.5% in the recently transitioned farm, and 100% in the long-term monoculture farm (Table 3). This pattern is also shared in the following WGS analysis of the selected representative subset, as chromosomal integrons and those linked with co-selection determinants became increasingly observed, and plasmid-associated integrons became less represented. Though the WGS investigation provides limited support due to its intentional selection of subset, the two sets of data pointing towards the same direction together supports an interpretation of the broader residual reservoir as one with distinctive architectural characteristics instead of simply a smaller original reservoir.
This distinction provides a possible explanation for the apparently uneven response of AMR reservoirs to antibiotic-use interventions described in previous studies. While some researchers concluded that policy restrictions on specific antibiotics are generally effective because the decline of selected resistance determinants is significant (Wen et al., 2022), others found that the broad resistome is largely unchanged or even become enhanced after official intervention (Tu et al., 2021). However, this conflict would be resolved if reducing selection pressure indeed affects different genetic architectures in diverging ways. Under this framework, antibiotic withdrawal may function as a major form of architectural filtering, shifting the focus of policy evaluation beyond changes in overall abundance to also consider the composition of the residual resistance reservoir.
4.2 Field–experiment concordance supports architecture-dependent filtering
The primary support for architecture-dependent persistence came from the concordance between the field and serial-passage results. Although the field survey alone showed that the composition of the integron reservoir differed across the livestock-pressure gradient, it’s unable to determine if those differences reflects intrinsic persistence of particular architectures. The serial-passage experiment alone demonstrated that integron architectures differed in their retention after antibiotic selection was removed, but could not investigate whether those experimentally persistent architectures were also the ones that remained in aquaculture environments. Hence, the concordance analysis functions to link these two components by comparing each architecture’s relative enrichment under reduced selection pressure with its generation-100 retention during antibiotic-free passage.
According to the corresponding results, a consistent directional pattern emerged. Empty-cassette integrons showed the strongest field enrichment under relaxed conditions, with an enrichment ratio of 9.60, and also showed the highest generation-100 retention at 88.0% (Table 7). Moving on, chromosomal integrons were similarly enriched in the relaxed-pressure reservoir and retained 85.8% of their original architecture after 100 generations. In contrast, multi-cassette integrons were seldom observed under relaxed pressure, with an enrichment ratio of only 0.14, and showed lower experimental retention of 61.0% ( Table 7). Plasmid-associated integrons followed the same pattern, with a field enrichment ratio of 0.30 and generation-100 retention of 55.4% ( Table 7). Thus, architectures that became disproportionately represented after livestock-associated pressure was reduced were also those that resisted experimental loss most strongly. This consistency supports the hypothesis that reduction of antibiotic selection pressure may produce architecture-dependent ecological filtering, rather than affecting all class 1 integrons uniformly.
4.3 Architectural mechanisms underlying differential persistence
4.3.1 Genomic location
The persistence assay indicates that the location of the integrons was strongly associated with ability to persist in pressure-free environments since, after 100 generations without antibiotic selection, chromosomal integrons retained 85.8% of their original architecture, compared with only 55.4% for plasmid-associated integrons (Table 5), showing how integrons located on the chromosomes would probably be retained more frequently. One plausible explanation for this pattern is that chromosomal integrons are vertically inherited as part of the host chromosome, whereas plasmid-associated integrons depend on continued maintenance of an additional replicon and may therefore be lost more easily when their selective advantage is removed ( Alonso-del Valle et al., 2021).
Yet, it would be wrong to directly link such persistence to the mobility of different genetic structures. A chromosomal integron may remain highly stable within a bacterial lineage while having different opportunities for horizontal transfer than one carried on a mobile plasmid. The present study is focused on measuring the persistence of integrons, hence any implications for mobility require additional testing.
4.3.2 Cassette load
Cassette load showed a similarly clear relationship with persistence. At generation 100, retention was 88% for empty integrons, 76% for single-cassette integrons, and 61% for multi-cassette integrons (Table 5). This graded decline supports the assumption that rich cassettes turns into heavy burden once the selective benefit provided by their resistance genes is reduced, an interpretation broadly consistent with previous work on the fitness costs and long-term stability of integrons (Starikova et al., 2012). In addition, the results show that increasing cassette burden was associated with faster rate of loss under relaxed selection, which also supports the conclusion. Together with the field pattern, this suggests that antibiotic-pressure reduction may leave behind a lower-burdened but comparatively more persistent integron population, rather than removing the integron platform uniformly. Hence, this may increase the potential risk associated with the post-policy resistance reservoir, as enhanced persistence and resilience could make these AMR-associated architectures harder to eliminate and allow them to remain in the environment for a prolonged period of time.
4.3.3 Linked determinants
Integrons physically linked to additional resistance determinants also showed greater persistence, with 80.9% retention at generation 100 compared with 55.9% when no linked determinant was detected (Table 5). This finding is relevant to the co-selection hypothesis developed in the Introduction, because linkage to disinfectant- or metal-resistance genes could potentially allow an integron-bearing element to remain advantageous even when selection by a particular antibiotic decreases. However, since the serial-passage experiment contained no antibiotic, disinfectant, or added metal, the increased retention of linked structures cannot be interpreted as direct evidence of active co-selection. Instead, genetic linkage appears to be a predictor of persistence, while whether alternative environmental pressures causally maintain these linked architectures remains to be determined.
4.4 Implications for AMR surveillance and policy
These findings suggest that intI1 abundance alone may provide an incomplete picture of the residual AMR reservoir. Because intI1 is widely used as a proxy for anthropogenic pollution (Gillings et al., 2015), reductions in its prevalence are often interpreted as evidence of decreasing AMR pressure. However, the present results indicate that integron populations with similar overall abundance could differ substantially in their genetic architecture and , hence, their persistence characteristics. In this study, genomic location, cassette burden, and linkage to additional resistance determinants were all associated with different rates of retention during antibiotic-free passage. Thus, while abundance describes how much of the reservoir remains, architectural characterization can provide important information about what kind of reservoir remains and how readily it may persist after direct selection is reduced. Incorporating selected architectural features into AMR surveillance may therefore function as useful and necessary complements for existing intI1-based monitoring.
The results also have implications for the evaluation of the effectiveness of AMR policies. The marked decline in intI1 prevalence across the designed farm-pressure gradient suggests that reducing livestock-associated antibiotic input indeed could substantially reduce the overall integron reservoir. However, at the same time, the compositional shift and field–experiment concordance suggest that the remaining reservoir may become increasingly dominated by architectures with greater persistence under relaxed selection. AMR policy may therefore influence both the quantity and the genetic composition of resistance, and these two outcomes should not be assumed to change proportionally.
Importantly, this interpretation should not be taken to mean that reducing antibiotic use is ineffective or that a smaller reservoir is necessarily more dangerous. A substantial reduction in overall integron prevalence remains an important outcome. Rather, the persistence characteristics of the residual population may influence how rapidly resistance continues to decline after intervention and whether some resistance structures remain detectable over prolonged periods. Policy evaluation may therefore benefit from considering both the magnitude of AMR reduction and the genetic characteristics of the resistance reservoir that persists.
4.5 Limitations and future direction
This study also has several important limitations. First and foremost, all results presented in this study are predicted based on findings from previous research in related fields and on the study’s proposed hypotheses, rather than generated through direct experimental investigation. Therefore, the predicted results and corresponding discussion should be interpreted as theoretical expectations rather than definitive conclusions, and cannot serve as strong empirical evidence. Future work should implement the proposed methodology and evaluate these hypotheses using experimentally generated data.
In addition, the study design uses a single farm to represent each selection-pressure category. As a result, differences observed across categories cannot be confidently attributed to antibiotic-use policy or production type, since uncontrolled farm-specific factors such as environmental conditions, management practices, microbial communities are unable to be neglected. The serial-passage experiment also substantially simplifies the complexity of farm environments by focusing primarily on integron retention under antibiotic-free conditions while excluding common processes that may influence persistence in natural settings, including horizontal gene transfer (HGT) and direct co-selection by non-antibiotic factors.
Future studies could therefore replicate the proposed selection-pressure gradient across multiple farms, with several independent farms sampled within each category to distinguish policy-related patterns from site-specific variation. Additional experiments could also separately examine the effects of non-antibiotic co-selection and HGT on integron persistence and dissemination. Together, these improvements would allow a more robust assessment of the mechanisms shaping post-policy resistance reservoirs and provide stronger evidence for evaluating the actual effects of antibiotic-use restrictions.
4.6 Conclusion
Overall, the findings support a model in which reduction in livestock-associated antibiotic pressure is associated not only with a decrease in the prevalence of class 1 integrons, but also with a restructuring of the residual integron population. Across the designed farm-pressure gradient, lower intI1 prevalence was accompanied by increasing representation of lower-burden, chromosomal, and linked architectures, while the serial-passage experiment showed that these same architectural features were associated with greater retention after selection was removed.
Together, these results suggest that integron loss following antibiotic reduction may be architecture-dependent rather than uniform, with relaxed selection acting as a potential filter on the composition of the remaining resistance reservoir. This provides a possible structural explanation for why broader AMR reservoirs can persist even when targeted interventions successfully reduce particular resistance determinants. Consequently, evaluating AMR control strategies solely through changes in intI1 abundance may overlook important differences in the persistence characteristics of the resistance that remains, highlighting the value of considering genetic architecture alongside overall prevalence when making policy decisions.
References
Alonso-del Valle, A., León-Sampedro, R., Rodríguez-Beltrán, J., DelaFuente, J., Hernández-García, M., Ruiz-Garbajosa, P., Cantón, R., Peña-Miller, R., & San Millán, A. (2021). Variability of plasmid fitness effects contributes to plasmid persistence in bacterial communities. Nature Communications, 12, Article 2653. https://doi.org/10.1038/s41467-021-22849-y
Barlow, R. S., Pemberton, J. M., Desmarchelier, P. M., & Gobius, K. S. (2004). Isolation and characterization of integron-containing bacteria without antibiotic selection. Antimicrobial Agents and Chemotherapy, 48(3), 838–842. https://doi.org/10.1128/AAC.48.3.838-842.2004
Chen, S., Ji, C., & Jin, S. (2022). Costs of an environmental regulation in livestock farming: Evidence from pig production in rural China. Journal of Agricultural Economics, 73(2), 541–563. https://doi.org/10.1111/1477-9552.12464
Chacón, L., Kuropka, B., González-Tortuero, E., Schreiber, F., Rojas-Jiménez, K., & Rodríguez-Rojas, A. (2023). Mechanisms of low susceptibility to the disinfectant benzalkonium chloride in a multidrug-resistant environmental isolate of Aeromonas hydrophila. Frontiers in Microbiology, 14, Article 1180128. https://doi.org/10.3389/fmicb.2023.1180128
Deng, Y. T., Feng, Y. Y., Tan, A. P., Zhao, F., Jiang, L., & Huang, Z. B. (2023). High occurrence of multiple-drug resistance mediated by integron in Aeromonas isolated from fish-livestock integrated farms. Aquaculture Research, 2023, Article 8512657. https://doi.org/10.1155/2023/8512657
Deng, Y., Tan, A., Zhao, F., Wang, F., Gong, H., Lai, Y., & Huang, Z. (2025). Global distribution of antimicrobial resistance genes in aquaculture. One Health Advances, 3, Article 6. https://doi.org/10.1186/s44280-025-00071-5
Feldgarden, M., Brover, V., Gonzalez-Escalona, N., Frye, J. G., Haendiges, J., Haft, D. H., Hoffmann, M., Pettengill, J. B., Prasad, A. B., Tillman, G. E., Tyson, G. H., & Klimke, W. (2021). AMRFinderPlus and the Reference Gene Catalog facilitate examination of the genomic links among antimicrobial resistance, stress response, and virulence. Scientific Reports, 11, Article 12728. https://doi.org/10.1038/s41598-021-91456-0
Food and Agriculture Organization of the United Nations. (2024). The state of world fisheries and aquaculture 2024: Blue transformation in action. https://doi.org/10.4060/cd0683en
Gaze, W. H., Abdouslam, N., Hawkey, P. M., & Wellington, E. M. H. (2005). Incidence of class 1 integrons in a quaternary ammonium compound-polluted environment. Antimicrobial Agents and Chemotherapy, 49(5), 1802–1807. https://doi.org/10.1128/AAC.49.5.1802-1807.2005
Gillings, M., Boucher, Y., Labbate, M., Holmes, A., Krishnan, S., Holley, M., & Stokes, H. W. (2008). The evolution of class 1 integrons and the rise of antibiotic resistance. Journal of Bacteriology, 190(14), 5095–5100. https://doi.org/10.1128/JB.00152-08
Gillings, M. R., Gaze, W. H., Pruden, A., Smalla, K., Tiedje, J. M., & Zhu, Y.-G. (2015). Using the class 1 integron-integrase gene as a proxy for anthropogenic pollution. The ISME Journal, 9(6), 1269–1279. https://doi.org/10.1038/ismej.2014.226
Lévesque, C., Piché, L., Larose, C., & Roy, P. H. (1995). PCR mapping of integrons reveals several novel combinations of resistance genes. Antimicrobial Agents and Chemotherapy, 39(1), 185–191. https://doi.org/10.1128/AAC.39.1.185
McEwen, S. A., & Collignon, P. J. (2018). Antimicrobial resistance: A One Health perspective. Microbiology Spectrum, 6(2), Article ARBA-0009-2017. https://doi.org/10.1128/microbiolspec.ARBA-0009-2017
Néron, B., Littner, E., Haudiquet, M., Perrin, A., Cury, J., & Rocha, E. P. C. (2022). IntegronFinder 2.0: Identification and analysis of integrons across bacteria, with a focus on antibiotic resistance in Klebsiella. Microorganisms, 10(4), 700. https://doi.org/10.3390/microorganisms10040700
QIAGEN. (2023). DNeasy blood & tissue handbook. https://www.qiagen.com/lr/resources/download.aspx?id=68f29296-5a9f-40fa-8b3d-1c148d0b3030&lang=en
Robertson, J., & Nash, J. H. E. (2018). MOB-suite: Software tools for clustering, reconstruction and typing of plasmids from draft assemblies. Microbial Genomics, 4(8), e000206. https://doi.org/10.1099/mgen.0.000206
Schwengers, O., Jelonek, L., Dieckmann, M. A., Beyvers, S., Blom, J., & Goesmann, A. (2021). Bakta: Rapid and standardized annotation of bacterial genomes via alignment-free sequence identification. Microbial Genomics, 7(11), 000685. https://doi.org/10.1099/mgen.0.000685
Shen, C., Zhong, L.-L., Yang, Y., Doi, Y., Paterson, D. L., Stoesser, N., Ma, F., El-Sayed Ahmed, M. A. E.-G., Feng, S., Huang, S., Li, H.-Y., Huang, X., Wen, X., Zhao, Z., Lin, M., Chen, G., Liang, W., Liang, Y., Xia, Y., ... Tian, G.-B. (2020). Dynamics of mcr-1 prevalence and mcr-1-positive Escherichia coli after the cessation of colistin use as a feed additive for animals in China: A prospective cross-sectional and whole genome sequencing-based molecular epidemiological study. The Lancet Microbe, 1(1), e34–e43. https://doi.org/10.1016/S2666-5247(20)30005-7
Starikova, I., Harms, K., Haugen, P., Lunde, T. T. M., Primicerio, R., Samuelsen, Ø., Nielsen, K. M., & Johnsen, P. J. (2012). A trade-off between the fitness cost of functional integrases and long-term stability of integrons. PLOS Pathogens, 8(11), e1003043. https://doi.org/10.1371/journal.ppat.1003043
Tu, Z., Gu, J., Zhang, H., Liu, J., Shui, J., & Zhang, A. (2021). Withdrawal of colistin reduces incidence of mcr-1-harboring IncX4-type plasmids but has limited effects on unrelated antibiotic resistance. Pathogens, 10(8), 1019. https://doi.org/10.3390/pathogens10081019
Watkins, R. R., Smith, T. C., & Bonomo, R. A. (2016). On the path to untreatable infections: Colistin use in agriculture and the end of ‘last resort’ antibiotics. Expert Review of Anti-infective Therapy, 14(9), 785–788. https://doi.org/10.1080/14787210.2016.1216314
Wen, R., Li, C., Zhao, M., Wang, H., & Tang, Y. (2022). Withdrawal of antibiotic growth promoters in China and its impact on the foodborne pathogen Campylobacter coli of swine origin. Frontiers in Microbiology, 13, Article 1004725. https://doi.org/10.3389/fmicb.2022.1004725
Wick, R. R., Judd, L. M., Gorrie, C. L., & Holt, K. E. (2017). Unicycler: Resolving bacterial genome assemblies from short and long sequencing reads. PLOS Computational Biology, 13(6), e1005595. https://doi.org/10.1371/journal.pcbi.1005595
World Organisation for Animal Health. (n.d.). Antimicrobial resistance. Retrieved September 5, 2026, from https://www.woah.org/en/what-we-do/global-initiatives/antimicrobial-resistance/
Yáñez, M. A., Catalán, V., Apráiz, D., Figueras, M. J., & Martínez-Murcia, A. J. (2003). Phylogenetic analysis of members of the genus Aeromonas based on gyrB gene sequences. International Journal of Systematic and Evolutionary Microbiology, 53(3), 875–883. https://doi.org/10.1099/ijs.0.02443-0
Zaman, S. B., Hussain, M. A., Nye, R., Mehta, V., Mamun, K. T., & Hossain, N. (2017). A review on antibiotic resistance: Alarm bells are ringing. Cureus, 9(6), e1403. https://doi.org/10.7759/cureus.1403
