Antibiotic usage in shrimp farms in Bangladesh and its impact on resistant gene abundance and pond microbiomes

Shrimp aquaculture plays a crucial role in Bangladesh’s agrarian economy and contributes significantly to foreign exchange earnings, establishing the country as one of the leading global exporters of farmed shrimp^1,2. In 2022, shrimp exports generated USD 245 million, accounting for nearly 1.23% of the national foreign income³. The sector directly supports more than 1.15 million people in its farming stage, with an additional 5.2 million people indirectly engaged in its extensive supply chain¹. Given this economic importance, ensuring the security and sustainability of shrimp aquaculture matters not only from a national perspective but also in the interests of international market access, particularly in the global north, where food safety standards are stringent. However, due to higher dependency on antimicrobials, without which production may drastically fall, global shrimp aquaculture is increasingly considered a critical reservoir of antimicrobial resistance (AMR) and associated antimicrobial-resistant genes (ARGs)^4,5,6. The evolution and spread of AMR in shrimp aquaculture not only subvert the efficacy of disease control within farms but also cause extreme environmental and public health consequences by fueling the AMR pandemic worldwide. As a result, this emerging concern undermines long-term food security and puts the industry at odds with global food safety standards. Therefore, growing literature has raised serious concerns and argued for introducing a One Aquaculture approach, recognizing connections among aquaculture, human health, and the environment⁷.

Globally, 10,259 tonnes of antimicrobials were consumed by aquaculture annually, and it is predicted to rise to 13,600 tonnes in 2030⁸. Notably, the Asia-Pacific region accounts for 93.8% of the usage, while China alone consumes 57.9%. Such extensive use of antimicrobials has contributed to alarming levels of antimicrobial resistance. For instance, between 2000 and 2018, antimicrobial compounds used in aquaculture showed resistance rates of more than 50%⁹. Shrimp, at the level of antimicrobial intensity of consumption, ranks fourth among the major aquaculture species, with estimated consumption of 46 mg kg⁻¹, following catfish, trout, and tilapia, but ahead of salmon⁸, accounting for approximately 2.7% of total aquaculture antimicrobial⁵. Recent data from Asian countries indicate a severe and increasing AMR burden in shrimp aquaculture. For instance, 96.6% of Escherichia coli isolated from frozen shrimp imported into Saudi Arabia from China and Vietnam were found to be resistant to cephalothin, 92.7% to ampicillin, and 94.4% to multidrug, most likely due to prophylactic antibiotic use and the use of untreated waste in shrimp feed¹⁰. Bangladesh, where 23 different antimicrobials were reportedly being used in shrimp hatcheries¹¹, also witnessed waterborne bacteria resistance to tetracycline and ampicillin¹². Although the government of Bangladesh has banned the use of antibiotics in animal feeds, certain antibiotics, such as oxytetracycline and sulfadiazine, are permitted for the health management of aquatic animals, including shrimp, under veterinary prescription⁶. Despite this regulation, there is currently no clear guideline regarding the competent authority for prescribing, market monitoring and surveillance, and enforcement against misuse^13,14. Consequently, farmers can access veterinary antibiotics over the counter, and their use at shrimp farms is widespread, often without adherence to recommended dosages or withdrawal periods. Such practices contribute to the development and spread of antimicrobial resistance in shrimp aquaculture in Bangladesh.

The microbial imbalance in the wake of selective antibiotic pressure can disrupt natural microbial ecology, leading to dysbiosis and inadvertently favoring the survival and multiplication of opportunistic and pathogenic microbes^15,16. A growing body of literature has recently highlighted the prevalence of AMR in shrimp aquaculture in Bangladesh^17,18,19,20. However, despite the critical roles of microbiomes in maintaining healthy pond environments and of ARGs in spreading and persisting resistomes, there remains a significant lack of empirical evidence on how ABU and different culture systems influence microbial community composition and ARG patterns. To address this gap, this study aims to systematically examine antibiotic usage patterns in Bangladeshi shrimp farms and assess their impact on microbiome structure and the prevalence of ARGs. By presenting such interlinkages, the study aims to provide scientific evidence to inform sustainable shrimp farming management, AMR mitigation efforts, and public health protection, thereby contributing to the global discourse on responsible aquaculture within the OH paradigm.

Disease prevalence and associated factors in shrimp farms

Seventy-five percent of the farms surveyed were affected by at least one of the nine reported diseases (four viral, two bacterial, two fungal, and one parasitic), with their significantly higher occurrence in extensive and improved extensive farms (p < 0.05; KW), except for Early Mortality Syndrome (EMS) disease (Fig. 1A and ST1 in Supplementary File 1). Extensive farms exhibited the broadest pathogen spectrum, with high frequencies of WSSV (55.6%) and EMS (33.3%). Improved extensive farms showed a similar but somewhat reduced disease burden, with EMS (33.3%) and EUS (22.2%) as the most common cases. In contrast, semi-intensive farms exhibited a narrower disease profile, dominated by EMS (50%), and a higher proportion of “no disease” reports (50%). We observed that species combination had a significant impact on disease prevalence (χ² = 16.67; p = 0.0002). Ponds stocked solely with shrimp reported fewer disease types, mainly EMS (25%) and WSSV (25%), whereas polyculture systems showed more diverse disease profiles and higher prevalence (Fig. 1A). Shrimp–carp–prawn polyculture farms exhibited substantial EMS occurrence (50%) together with concurrent reports of EHP, tail rot, and WSSV. Conversely, shrimp–carp systems experienced lower disease prevalence.

While investigating the drivers of disease prevalence, farmers attributed most cases to poor water quality and climatic factors, including abrupt changes in temperature, erratic rainfall, and fluctuations in salinity (Fig. 1B). The causes of WSSV and EMS were believed to be poor seed quality and a lack of biosecurity, while the stocking of carp was alleged to have caused Epizootic Ulcerative Syndrome. However, statistical analyses showed that among the parameters, only water source had a significant impact (χ² = 0.034; p = 0.0073) on the occurrence of Monodon Baculovirus (MBV) disease (ST2 in Supplementary File 1).

Patterns of antibiotic usage (ABU) and source of information

Seven antibiotics, including amoxicillin, oxytetracycline, trimethoprim, ciprofloxacin, enrofloxacin, erythromycin, and kanamycin, were found to be used in shrimp farms. Amoxicillin and ciprofloxacin were the most commonly used antibiotics, accounting for 20.83% of farms, followed by enrofloxacin and erythromycin. Improved extensive farms showed the highest diversity and frequency of antibiotic use, with 8 of 9 farms using at least one antibiotic (Fig. 2A). These farms consumed six different compounds: erythromycin, ciprofloxacin, trimethoprim, enrofloxacin, oxytetracycline, and amoxicillin, making the systems the most significant contributors to overall antibiotic heterogeneity. Extensive farms displayed a similar but slightly narrower profile, with five antibiotics used across eight farms. By contrast, semi-intensive farms used only three antibiotics (amoxicillin, enrofloxacin, and oxytetracycline). When mapped against disease categories, antibiotic usage was reported in response to viral diseases such as WSSV, TSV, and IHHNV in extensive farms (Fig. 2B). Ciprofloxacin, enrofloxacin, and erythromycin were frequently administered following viral outbreaks, while trimethoprim and oxytetracycline were used for mixed bacterial, fungal, and unknown disease conditions. Improved-extensive farms showed comparable patterns, with amoxicillin, enrofloxacin, and erythromycin used to treat viral, bacterial, parasitic, and fungal diseases. Semi-intensive farms, however, used antibiotics only against EMS, EUS, or unknown infections. Notably, no semi-intensive farm used antibiotics during viral outbreaks, signaling comparatively more appropriate disease management. However, overall, we observed very weak correlations between farm-level disease prevalence and ABU (Pearson r = 0.07 for bacterial diseases vs. ABU; Pearson r = 0.13 for viral diseases vs. ABU; Pearson r = 0.15 for overall disease prevalence vs. ABU) (Fig. 2D), suggesting that antibiotics were often used irrespective of reported disease occurrence.

Further, we categorized the sources of information and influence on antibiotic selection into three main groups: drug shop owners, company agents, and fellow farmers. Extensive farms showed a high degree of dependence on informal networks: shop owners and fellow farmers were the primary influencers. In improved-extensive farms, both shop owners and company agents provided advice on treatment and prophylaxis (55.55%), often encouraging broad-spectrum antibiotic use even in the absence of confirmed diagnoses. Semi-intensive farms showed the most regulated pattern; every instance of antibiotic use was guided exclusively by company agents, and all applications were therapeutic rather than prophylactic.

The abundance of ARGs and their correlation with antibiotic usage

We detected 62 ARGs by qPCR (19 were below the detection limit in all samples) across 12 categories, with the most diverse group being tetracycline-resistant genes (12), followed by multidrug (9) and beta-lactam (8) resistant genes. The genes conferring resistance to fluoroquinolones were the most prevalent, with a higher average abundance (83.25 ± 31.86 × 10⁵/g soil) in extensive farms (Fig. 3). The significantly higher abundance of macrolides (12.07 ± 8.92 × 10⁵/g soil; p = 0.0008, KW test), multidrug (35.01 ± 31.39 × 10⁵/g soil; p = 0.043, KW test), and tetracycline (43.77 ± 41.92 × 10⁵/g soil; p = 0.021, KW test) resistant genes were recorded in improved extensive farms while the average abundance of sulphonamide resistant genes were significantly higher (9.49 ± 9.12 × 10⁵/g soil; p = 0.015, KW test) in extensive farms. The diaminopyrimidines-resistant gene dfrA1 was recorded (average abundance 3.78 ± 3.43 × 10⁴/g soil) only in samples collected from improved extensive farms. Improved extensive farms harbored the highest number of ARGs (51), followed by extensive (50) and semi-intensive (27) farms, while 19 genes were shared across the culture systems. Among these, the genes ampC, qnrA, sul1, and tetC were recorded in all samples, with the highest average abundance of qnrA (41.59 ± 22.94 ×10⁵/g soil), followed by tetC (24.64 ± 28.94 × 10⁵/g soil) and ampC (15.81 ± 5.54 × 10⁵/g soil). On contrary, tetS, tetD, folA, fosB, catB8, bacA, marR, aacC4, blaZ, lmrA, mdtA, and acrR were abundant only in one sample (ST3 in Supplementary File 1).

We observed a weak but statistically significant positive correlation between ABU (qualitative) and the total abundance of ARGs (Mantel statistic, r = 0.3398, p = 0.006), suggesting that increased antibiotic use may contribute to elevated ARG abundance in shrimp farms. Pearson correlation analysis further revealed associations between ARG abundance and the use of specific antibiotics. Notably, the use of oxytetracycline antibiotic was strongly correlated with tetracycline-resistant genes: tetC (r = 0.86, p < 0.001), tetR (r = 0.76, p < 0.001), and tet34 (r = 0.75, p < 0.001) (Fig. 4). Similarly, the use of trimethoprim antibiotics showed strong positive correlations with sulphonamide-resistant genes: sul1 (r = 0.97, p < 0.001), sul3 (r = 0.76, p < 0.001), sul2 (r = 0.75, p = 0.00028), and folA (r = 0.75, p = 0.00018). Moreover, the use of amoxicillin was positively correlated with beta-lactam resistant gene blaZ (r = 0.72, p = 0.0005), while kanamycin application was associated with aminoglycoside resistant genes, including aphA3 (r = 0.64, p = 0.0007), strA (r = 0.56, p = 0.005), and strB (r = 0.58, p = 0.003). Although erythromycin showed weak correlation with tetQ and gyrA, and enrofloxacin with aadA9 and catB8 genes, none of these associations were significant.

Bacterial community composition and diversity in shrimp farms

After denoising and clustering filtered reads, we obtained 16,096 bacterial ASVs (average length 417.47 ± 21.6 bp) with a total frequency of 549,302 across 24 samples. Taxonomic assignment classified these resulting ASVs into 58 phyla, among which Proteobacteria were the most abundant phylum with a relative abundance of 47.94%, followed by Firmicutes (13.85%), Actinobacteria (8.61%), Chloroflexi (7.58%), Acidobacteria (4.63%), Nitrospirae (4.11%), and Bacteroidetes (3.79%). We observed that antibiotic use had an insignificant impact on the relative abundance of bacteria, except for those affiliated with Elusimicrobia (p = 0.03); however, their contribution to the total abundance appeared negligible (ST4 in Supplementary File 1). On the contrary, among the phyla contributing ≥1% abundance (shown in Fig. 5), the relative abundance of Actinobacteria was significantly highest in semi-intensive farms (p = 0.045; KW test), while Bacteroidetes and Proteobacteria were in extensive (p = 0.025; KW test) and improved extensive (p = 0.004; KW test) farms, respectively. The relative abundance of Acidobacteria and Bacteroidetes differed significantly across water sources, with the highest abundance in farms fed pumped water (p = 0.032; KW test) and river water (p = 0.037; KW test), respectively.

Our study observed that culture systems had significant effects on Chao1 index (p = 0.021; KW test) and Shannon diversity (p = 0.037; KW test), with the highest values observed in extensive farms (Fig. 6A, B). In contrast, these indices showed no significant differences in antibiotic usage (Cao1: p = 0.493; Shannon: p = 0.420; KW test). The NMDS ordination plot based on the Bray-Curtis distance matrix showed a strong culture system-oriented separation at the bacterial community level (Fig. 6C). Pairwise PERMANOVA test also confirmed a significant difference in beta-diversity between extensive farms and semi-intensive farms (F = 1.131, p = 0.027). Water sources also had a possible significant effect on community composition (F = 1.132, p = 0.044; pairwise PERMANOVA), and a clear separation in bacterial communities between farms fed with pumped water and with river water was evident (Fig. 6D).

Gaining a deeper understanding of how increased antibiotic use in shrimp farming raises threats to the One Health Approach through contributing to microbial community shifts and elevated resistant gene abundance is essential to reduce selective pressures from antibiotic use. While national surveillance programs have not publicly reported any pathogens in shrimp in recent years, independent peer-reviewed studies and ongoing farm-level diagnostic services continue to detect pathogens such as WSSV, YHV, IHHNV, EUS, and TSV in Bangladesh^21,22. Our study revealed a higher prevalence of WSS in low-input, poorly managed, and open-water dependent extensive farms. Similarly, several studies also identified WSS as the most common shrimp disease over the last two decades and alleged poor water quality, lack of water treatment facilities, abrupt change in temperature in shallow ponds, and uncontrolled water systems as the main driving factors for its rapid spread^23,24,25,26. In contrast, semi-intensive farms were most affected by EMS, primarily due to the stocking of poor-quality seeds, which serve as a potential carrier for Vibrio parahaemolyticus and transmit from one pond to another^27,28.

To minimize economic losses caused by disease, imprudent antibiotic use is a common practice at every stage of shrimp production, from hatcheries to grow-out ponds⁵. A total of seven antibiotic categories were reported in this study, in line with previous findings¹³. Importantly, our results suggest that antibiotics are misused qualitatively, regardless of disease type or occurrence. Similar results were also documented for finfish aquaculture in Bangladesh^6,13. However, Ali et al.²⁹ reported fewer antibiotics than in our study and more responsible antibiotic use in shrimp farms. This shift in antibiotic use over time (2018-2023) suggests that increased disease frequency in recent days has driven greater demand and reliance on antibiotics in shrimp aquaculture. The extent of antibiotic misuse was reportedly higher on extensive and improved extensive farms, primarily due to higher disease prevalence stemming from inadequate farm management and limited biosecurity. In extensive systems, antibiotics are typically applied directly to pond water, whereas in semi-intensive farms, they are usually mixed with feed. The large water volumes in extensive farms often result in under-dosing, irregular application, and poor adherence to recommended withdrawal periods, leading to suboptimal efficacy and increasing the risk of antibiotic resistance.

These misuses, in either quantitative or qualitative form, contribute to selective pressure on bacteria and thereby drive the formation of AMR mobile genetic elements³⁰. We revealed 62 ARGs in our study, substantially higher than the number reported in Bangladesh’s shrimp farms in 2021¹⁸. The increase in ARGs may result from differences in AMR screening protocols and sampling sites; nonetheless, it snapshots a temporal upward trend in ARGs prevalence within shrimp aquaculture. The pattern of ARGs observed in the current study appears to be less diverse and prevalent than those reported in China^{31,32,33,34,35}, Thailand³⁶, India^30,37, and Ecuador³⁸. The ubiquitous presence of genes conferring resistance to fluoroquinolones, sulphonamides, tetracyclines, and beta-lactams observed in our samples is consistent with previous findings across different aquaculture settings^{32,35,38,39,40,41,42}. A review on ARGs recorded in South Asian aquaculture identified blaTEM and blaSHV as the most frequently detected beta-lactam resistance genes⁴⁰. Interestingly, these were not detected in semi-intensive or improved extensive farms in our study, possibly because sulphonamide use was limited to only a few extensive farms. On the other hand, the highest abundance of fluoroquinolone-resistant genes in all systems is likely associated with the extensive consumption of fluoroquinolone antibiotics, particularly ciprofloxacin and enrofloxacin, in Bangladesh⁶. Moreover, we didn’t detect any macrolide-resistant genes nor find any evidence of macrolide use in semi-intensive farms. These all suggest that the prevalence of most ARGs is likely exclusive to the corresponding antibiotic administration. A lower abundance of ARGs in semi-intensive farms may be attributed to either more restrained qualitative and quantitative use of antibiotics or to increased use of probiotics, which may promote the proliferation and accumulation of ARGs^43,44. On the other hand, a higher abundance of ARGs in extensive and improved extensive farms was striking and is likely associated with integrated livestock farming and the use of untreated and polluted river water. Shi et al.³⁹ provided evidence that livestock farming and water fertilization with livestock manure significantly contributed to shaping the antibiotic resistome in water. This can be further validated by the presence of vanA, vanB, vanC, and vanG genes in extensive and improved extensive farms, despite no reported use of vancomycin in water. It is possible that these ARGs originated from the vancomycin consumption in livestock, highlighting the indirect impact of livestock practices on aquatic resistome development. Unlike semi-intensive farms, extensive farms primarily supply shrimp and finfish for local and domestic consumption². Therefore, extensive farms are more likely to provide a reservoir and medium for ARGs transfer, posing a serious threat to local public health.

While many ARGs were strongly correlated with corresponding antibiotic usage, for example, the use of oxytetracycline was tightly linked with tetracycline-resistance genes (tetC, tetR, tet34), and trimethoprim with sul1, sul2, sul3, not all resistance genes followed this pattern. The fact that some ARGs are not correlated strongly with antibiotic application suggests that other ecological or management-driven processes may strongly influence their presence and spread⁴⁵. This observation aligns with emerging evidence that aquaculture environments can act as reservoirs of ARGs even in the absence of direct antibiotic selection pressure. For instance, intrinsic resistance genes, efflux pump genes, or mobile genetic elements may be naturally present in environmental bacteria and persist due to horizontal gene transfer, biofilm formation, or co-selection driven by non-antibiotic stressors^42,45,46. Moreover, factors such as water quality variation, pollution (e.g., from heavy metals or other contaminants), farm biosecurity, and connectivity with other microbial reservoirs (e.g., terrestrial runoff, integrated farming systems) may shape the resistome independent of antibiotic use^39,47. Thus, while antibiotic use in farms clearly promotes selection of matching resistance genes, controlling AMR in shrimp aquaculture will likely require more than just reducing antibiotic input.

We did not observe any significant changes in microbial diversity and community composition associated with antibiotic usage. This may indicate a greater abundance of antibiotic-resistant bacteria in shrimp aquaculture settings; however, pond sediment, a complex and resilient environment, can also buffer against short-term changes from antibiotic pulses. Moreover, the resolution of 16S rRNA sequencing can be too coarse to detect subtle community-level changes: we therefore remain cautious in concluding the presence of resistant bacteria. The relative abundances of Proteobacteria and Bacteroidetes, which are believed to harbor ARGs in their genomes⁴⁸, were significantly higher in extensive and improved extensive farms. This might partially explain the elevated abundance of ARGs in improved and extensive farms. Proteobacteria, which play a more crucial role in the resistome than Bacteroidetes due to their intrinsic sensitivity to many antibiotics⁴⁹, were found to be more abundant in improved extensive farms, potentially explaining the higher ARG abundance compared to the extensive system. The relative abundances of these two bacterial phyla also increased in shrimp farms fed with river water, suggesting that river water may be highly contaminated with Proteobacteria and Bacteroidetes. On the contrary, many species within the phylum Actinobacteria, particularly Streptomyces, Actinomyces, Micromonospora, and Salinispora are known to be used as probiotics in aquaculture^50,51,52,53 and can reduce selective pressure by decomposing antibiotic residues⁵⁴. The significantly higher relative abundance of Actinobacteria observed in semi-intensive farms in our study is likely due to higher probiotic application and may partly explain the reduced ARG abundance in these ponds.

Although we did not assess the AMR profiles of source water prior to pond application, our findings indicate that water type may indirectly influence resistome configuration by altering baseline microbial communities. River water, particularly when untreated, may introduce diverse environmental bacteria and anthropogenic contaminants originating from upstream agricultural, livestock, or municipal activities, potentially enriching bacterial groups known to harbor ARGs^6,55. In contrast, pumped groundwater harbors comparatively lower concentrations of contaminants and less anthropogenic influence. Therefore, pumped water showed a significantly higher relative abundance of non-pathogenic bacterial taxa, including Acidobacteria. Such differences in community composition can influence horizontal gene transfer dynamics, biofilm formation, and microbial interactions that facilitate ARG persistence, even in the absence of direct antibiotic pressure^5,42. Importantly, while antibiotic usage did not significantly affect microbial diversity, culture system and water source significantly structured beta diversity. This suggests that ecological drivers, including pond management and water origin, may play a stronger role in shaping microbial community configuration than antibiotic exposure alone. Therefore, water source should be considered a potential indirect contributor to AMR ecology in shrimp aquaculture, acting through microbiome restructuring.

We acknowledge certain limitations in this study, including the absence of quantitative data on antibiotic use, which prevents the actual selective pressure from being gauged. These constraints reflect challenges related to farm accessibility, limited record-keeping, and farmers’ reluctance to disclose detailed antibiotic usage. As a result, the present work, as an initial exploratory assessment rather than sector-wide quantification, suggests that future research prioritize quantitative antibiotic-use measurements. The study was also limited by the relatively small number of farms surveyed, which may influence statistical power and the robustness of inter-group comparisons. Nonetheless, our results confidently indicated that shrimp farms serve as critical reservoirs for ARGs, and we therefore suggest several strategies to mitigate AMR in shrimp aquaculture. First, robust antibiotic stewardship is essential, including clearer regulations on therapeutic use, restricted over-the-counter access, stringent market monitoring, and defined prescribing authority. Second, farm management improvements, such as enhanced biosecurity, controlled water sources, regular pond sediment management, and optimized stocking densities, can reduce disease prevalence and consequently the need for antibiotics. Third, training and extension services should be strengthened to equip farmers with knowledge on disease diagnosis, responsible antibiotic use, and alternative disease-control strategies, including probiotics and phage therapy. Finally, continuous surveillance of both antibiotic usage and resistance genes should be established at the regional and national levels to monitor trends and inform policy.

Farms, farming practices, social data collection, and sampling protocol

The study was conducted in the shrimp-producing hub of Bangladesh (Khulna, Bagerhat, and Satkhira districts), ensuring coverage of different culture systems (as detailed in Fig. 7). The primary categories are extensive, improved-extensive, and semi-intensive culture systems characterized by variation in farming infrastructure, stocking density, feed and water management, and species selection⁵⁶. Across the surveyed shrimp farms, management practices varied widely as shown in ST5 in Supplementary File 1. Pond numbers ranged from 1 to 5 per farm, with target pond areas spanning 0.22–2.33 ha and depths from 3 to 6 ft. Water sources included river water, treated river water, and pumped groundwater, with semi-intensive farms predominantly relying on treated and pumped water within a closed water circuit. Stocking densities were less than two shrimp per m² in extensive farms, 3–8 shrimp per m² in improved-extensive systems, and 9–17 shrimp per m² in semi-intensive farms. Feeding practices also reflected system intensity: extensive farms used either supplementary feeds or no feeding; improved-extensive farms used mixed pellet and supplementary diets; and semi-intensive farms relied on commercial pellets with higher feeding rates and frequencies. The use of probiotics was more common in higher-intensity systems, reflecting more structured health-management protocols.

Based on available records of antibiotic usage and disease prevalence over the last 12 months, 24 shrimp farms were selected for social data and molecular sample collection. The distribution of farms (nine extensive farms, nine improved extensive farms, and six semi-intensive farms) was proportional to the density of each farming system in the region. To avoid data ghosting on antibiotic usage, only registered and reliable farms were purposively selected with the help of government officials. Taking the near-harvesting stage into consideration, an interview was conducted in October 2023 to collect data on farming practices (culture systems, species, seeds, feeding, and water management), disease prevalence (disease occurrence and farmers’ perceptions of causation), and management applications (antibiotic usage, purpose of usage, source of advice, and probiotic usage) for the entire production cycle. The study received ethical approval for human involvement from the “Ethical Standard of Research Committee” of Bangladesh Agricultural University (Reference No: 977/BAURES/ESRC/FISH/31; Date: September 10, 2023). Informed consent was obtained from all farmers, ensuring that no individual information that could affect their marketing and social values would be published or disclosed.

Disease occurrence in shrimp farms was documented using a combination of farmer-reported information (indigenous knowledge) and diagnostic reports provided by private companies, NGOs, and research institutes^6,14. In Bangladesh, disease diagnosis at the farm level is supported by multiple actors. Veterinary and aquaculture drug companies frequently offer on-site diagnostic services (to increase trust and legitimacy), including evaluation of clinical signs and, in some cases, laboratory-based PCR testing through their existing facilities. In addition to industry-based services, the Bangladesh Fisheries Research Institute and several NGOs (e.g., WorldFish) also provide diagnostic support to farmers, typically through laboratory-based assessments⁵⁷. However, consistent with existing national regulations, none of these agencies is authorized to prescribe any antibiotics, except the Department of Fisheries. Furthermore, farmers increasingly receive training from the Department of Fisheries and nearby universities, enabling them to identify common disease symptoms through visual observations, behavioural changes, and mortality patterns.

During the social data collection survey, sediment samples were collected from three locations within each farm to ensure representativeness. At each sampling point, surface sediment (0–5 cm depth) was collected using a sterile stainless-steel corer (5 cm diameter), following methods described by Bashar et al.⁵⁸. The upper 5 cm was specifically targeted because this layer represents the most biologically active sediment zone and directly interacts with pond water and benthic microbial processes. The three subsamples from each farm were placed in sterile polyethylene bags, transported on ice to the laboratory, and homogenized thoroughly under sterile conditions to form one composite sample per farm. Homogenization was performed to minimize microscale spatial heterogeneity within ponds and obtain a representative sediment profile. The homogenized composite samples were transferred into sterile containers and stored at −30 °C until DNA extraction.

DNA extraction, library preparation, and high-throughput sequencing

Genomic DNA was extracted from 0.28 g homogenized sediment samples using DNeasy PowerSoil kit (Qiagen, Germany) following the manufacturer’s protocol. The sequencing library was prepared using a two-step PCR method as described by Chan et al.⁵⁹. Primers (341 F: CCTAYGGGRBGCASCAG and 806 R: GGACTACNNGGGTATCTAAT) with the Illumina overhang and sequencing adapters were used to amplify the V3-V4 hypervariable region of bacterial 16S rRNA gene⁶⁰. The 20 μL PCR mixture included 0.5 μL Taq Polymerase (5U/μL), 3 μL 10X buffer, 0.5 μL of each primer, 3 μL dNTPs (2 nM), 2 μL DNA template, and 10.5 μL nuclease-free water. Amplification was initiated with a pre-denaturation step at 95 °C for 3 min, followed by 35 cycles of PCR (denaturation at 98 °C for 30 s, annealing at 50 °C for 45 s, extension at 72 °C for 30 s), and a final extension step at 72 °C for 5 min. The first PCR products, cleaned with the DNA Clean and Concentrator kit (D4010, Zymo Research), were used as templates for the 2nd PCR, where sample-specific i7 and i5 dual indexes were ligated according to the Nextera XT library preparation protocol. The PCR recipe and thermal conditions for the 2nd PCR were obtained from Bashar et al.⁵⁸. After cleanup, all amplicons were pooled together in an equimolar concentration, and the pooled library was submitted for pair-end (2 × 250 bp) Nova-seq sequencing.

Bioinformatics

The Quantitative Insights into Microbial Ecology (QIIME2: version 2023.9) pipeline was used to process raw pair-end FASTQ files. Primers and adapter sequences were trimmed using the Cutadapt plugin. After denoising and merging reads, chimeras were removed, and reads were clustered into high-resolution Amplicon Variant Sequencing (ASVs) using the DADA2 algorithm. For taxonomic annotations with the SILVA 138.1 reference database, we used a trained Naïve Bayes classifier via the q2-feature-classifier plugin. Using the qiime2R function in R, we imported the artifact files into the R environment for diversity analyses. Using the Phyloseq package, we analyzed alpha diversity and species richness across samples. Nonmetric Multidimensional Scaling (NMDS) ordinations based on the Bray-Curtis dissimilarity index were plotted in R using the vegan function to examine the effects of culture systems and water management on variation in bacterial communities. PERMANOVA analysis was performed using the qiime diversity plugin to assess whether community variation was statistically significant. All scripts associated with these analyses were adopted from our previous studies⁶¹.

Determination of ARGs abundance through quantitative PCR

Based on the classes of antibiotics used, previously reported ARGs in the aquatic environment^42,62,63, and public health priorities⁶⁴, we selected a total of 81 ARGs for quantification using the AriaMX qPCR system (Agilent Technologies, G8830A, California, USA). These included seven aminoglycosides resistance encoded genes, four amphenicols resistance encoded genes, eight beta-lactam resistance encoded genes, one diaminopyrimidines resistance encoded gene, three fluroquinolones resistance encoded genes, 11 MLSB (Macrolides, Lincosamides, and Streptogramin B) resistance encoded genes, 12 multidrug resistance encoded genes, five sulphonamide resistance encoded genes, 14 tetracycline resistance encoded genes, six vancomycin resistance encoded genes, and 10 other resistance encoded genes. Due to the widespread use of macrolides in aquaculture, their diverse mechanisms of action, and broad host range, we considered macrolides separately out of the broader MLSB category. All ARGs with their primer sequences and primer-specific annealing temperatures were obtained from previous studies^18,42,62,63 and are outlined in ST6 (Supplementary File 1). Serially diluted, highly purified DNA resulting from the amplification of each gene by conventional PCR was used to construct the standard curves⁶⁵. The 20 μL qPCR assays, containing 8 μL SYBR green master mix, were run in triplicate, and each run included a no-template control. The thermal profile for qPCR included hot start at 95 °C for 2 min, 35 cycles of denaturation at 92 °C for 30 s, and primer annealing for 30 s. Melt curve analysis was performed from 60 °C to 95 °C, and assays lacking an amplification efficiency of 89–103% and a regression coefficient of < 0.98 were discarded. The absolute abundance of each gene was calculated from average CT values using the AriaMx Software (Version: 2.1), considering the CT threshold for detection limit was 31.

Statistical analysis

All plots, representing disease prevalence, the pattern of antibiotic usage, and the average abundance of ARGs across different culture systems, were created in R (version: 4.3.3) using ggplot2 and factoextra packages. To confirm if the variations in average abundance of ARG categories across different culture systems were statistically significant, we performed ANOVA, followed by a Tukey test using broom function. To visualize farmers’ perceptions of disease causation, linkages were established between reported diseases and farmers’ responses in Gephi (version 0.10.0), with response frequencies as edge weights. A correlation heatmap between the class of antibiotic usage and the absolute abundance of ARGs was created with pheatmap package. A nonparametric Kruskal-Wallis test, followed by Dunn’s post hoc comparisons with Bonferroni correction, was used to assess statistical significance in alpha diversity and species richness across different culture and water circulation systems.

Sequence accession

All raw sequences (pair-end fastq files) generated in the study were submitted to the Sequence Read Archive (SRA) database in NCBI and can be accessed under the Shrimp-AMR BioProject (accession no. PRJNA1241319).