Microbiome and resistome dynamics in different stages of commercial broiler production with restricted antimicrobial use

Abstract

Background

Antimicrobial use (AMU) in poultry production is central to curb the Antimicrobial Resistance (AMR) crisis. Institutional and market pressure led many commercial poultry operations to practice distinct levels of AMU restriction. On-farm data remains one of the main bottlenecks in understanding the impacts of AMU restriction at the farm level and across production systems. However, AMR dynamics in company-wide production chains remain largely unexplored, precluding improvement of AMU policies and stewardship.

Study aim

Here, we shotgun sequenced soil and litter samples from 26 poultry farms and carcass rinses from a processing plant to reconstruct the microbiome and resistome of two vertically integrated commercial poultry operations to explore their dynamics under AMU restriction.

Results

Shotgun sequencing revealed that litter microbiome and resistome changed significantly by production stage and company, reflecting management practices and possible effects of historical AMU. Meanwhile, broiler farms had increased detection of potential pathogens and AMR diversity. We found no evidence of farm-to-fork transmission. Effective biosecurity protocols largely maintained the separation between the internal and external environments of the poultry houses, except on two farms where breaches might have led to external spread of pathogens and AMR.

Conclusion

Our study highlights that AMR in commercial poultry system reflects the combined effect of production-stage, company practices, and environmental boundaries. Future studies should integrate quantitative AMR data and culture-based techniques with metagenomic findings to strengthen tracking and surveillance of AMR in poultry farm environments.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12866-025-04664-3.

Background

The US is among the world’s leading poultry producer and exporter [1], and domestic consumption and demand have increased in recent years [2]. US commercial poultry production adopted the vertically integrated system, where bird flocks are allocated to specific farms within the same company depending on production stage [3]. Pullet farms raise young chicks destined to become breeders. Once pullets mature, they are moved to breeder farms to produce fertilized eggs destined to become broiler chickens. These eggs are then sent to hatcheries, where chicks are hatched and then moved to broiler farms. Broiler farms raise the birds until they reach the market weight (approximately 3.5 to 10 lbs.) and/or market age (approximately 4.5 to 10 weeks), by which point the flock is then transported to the processing plant. Vertical integration ensures that every stage of poultry production is under the supervision of a single company, with consistent biosecurity and management protocols across the chain [4]. While this system allows for efficient and optimized processes, it also creates a potential pathway for transmission of pathogens [5] and antimicrobial resistance (AMR) across the production chain.

The rise of AMR in bacterial pathogens is a major One Health threat contributing to 5 million human deaths worldwide in 2019 [6]. Antimicrobial use (AMU) in humans, animals, and agriculture is a main contributor to this issue [7]. Food animal production accounts for approximately 73% of total global AMU [8]. Despite strict biosecurity and management practices, to prevent disease outbreaks [9], AMU in food animals is anticipated to rise 3% by 2030 in the US alone due to the rise in demand for animal protein [10]. Livestock farms provide an interface between animals, humans, and environmental microbiomes. AMR selected within animal settings can spread to the surrounding environments through biosecurity breaches and management lapses [11] promoting exchange of antimicrobial resistance genes (ARGs) across environments [12]. Thus, AMU in agriculture is key in tackling the AMR crisis.

Judicious AMU has been shown to affect levels of AMR in livestock. Decrease in AMR in UK livestock was related to decrease in antimicrobial sales [13], and decrease in prophylactic AMU was linked to decrease in resistance in Salmonella, Campylobacter and Escherichia coli from broiler production in Canada [14]. The US instated new guidelines for the approval of antimicrobials for animal use in 2003 [15] in efforts to tackle the effects of agricultural AMU in the rise of AMR. However, drugs approved before this date for use in poultry in feed or water for treatment purposes or growth promotion retained over the counter clearances until 2017, including compounds of interest for human medicine [9]. In 2017, the US Food and Drug Administration (FDA) completed the removal of growth promotion and over-the counter clearances from antibiotics important to human medicine, requiring veterinary oversight for their use in poultry [16].

Since then, broiler production in the US defaulted to No Antibiotics Important to Human Medicine (NAIHM) practices, where healthy flocks are raised without medically important antibiotics, although other classes of antimicrobials may still be used for disease control and production uses. Consumer pressure combined with marketing opportunities led to the increase of No Antibiotics Ever (NAE) operations, which raises flocks without any antibiotics whatsoever, precluding even the use of certain vaccine formulations and disease prevention/control reliant on compounds labeled as non-medically-important antibiotics by the FDA, such as bacitracin [9]. However, both NAIHM and NAE farms employ medically important antibiotics for disease treatment when necessary to safeguard flock health. In this case, NAE flocks lose the label and are sold as a conventional product. Past and sporadic AMU may lead to the accumulation and persistence of antimicrobial residues in the farm environment, exposing even NAE flocks to antimicrobials, contributing to AMR maintenance.

In practice, NAIHM and NAE farm operations reflect restricted rather than eliminated AMU. While the FDA reported a drop of 36% in sales and distribution of medically important antimicrobials since 2015 [17], more studies are necessary to investigate the effects of restricted use on AMR in broiler production. On-farm data is a bottleneck in understanding the dynamics of AMR and antimicrobial stewardship [18]. Although most US broiler production follows vertical integration, differences in company-specific farm management practices, flock genetics and rearing, treatment protocols and flock movement between farms create unique pockets where farm environments are subject to different pressures depending on company.

In this study, we used shotgun metagenomics to unveil the landscape of AMR in commercial broiler production under restricted AMU, examining its dynamics both across the production chain and within individual farms. We investigated the microbiome and resistome of poultry farms in two vertically integrated company complexes practicing restricted AMU (one NAE and one NAIHM), covering different stages of production (pullet, breeder, broiler and processing plant). To our knowledge, this is the first metagenomics study covering the entire production chain from integrated commercial broiler operations. Our study offers a rare insight into commercial production chains, providing valuable findings for designing effective risk assessment, management practices, and applicable policies.

Methods

Sampling sites and methodology

We enrolled 2 pullet farms, 4 breeder farms, and 7 broiler farms per company from two vertically integrated poultry complexes, each operated by a different company in the southeastern region of the US. Company A raises NAE flocks, while Company B adopts NAIHM practices. No AMU had been reported by the individual poultry farmers for at least three months before sample collection. However, since we did not obtain authoritative reports on AMU by the companies’ overseeing veterinarians, all farms were considered restricted AMU for the purpose of this study. All farms are expected to adhere to biosecurity standards pertinent to each production stage to reduce the risk of infectious disease transmission and minimize any tracking of soil and litter between inside and outside the poultry house, including but not limited to, the use of protective gear to enter the poultry houses, disinfection stations at entry points, restricted access to essential personnel, cleaning of vehicles and equipment, and rodent control. The enrolled farms represent at least 10% of the poultry farms from each company complex for adequate representation of the production system. In total 26 farms were sampled: 4 pullet, 8 breeder and 14 broiler, and one flock from Company A was followed to the processing plant (Fig. 1A). At the time of sample collection, the age of birds varied across farm types: pullets ranged from 7 to 14 weeks, breeders from 27 to 45 weeks and broilers from 2 to 6 weeks. The approximate bird density also varied by farm type: pullet farms had a density of 4–8 birds/m², breeder farms 3.5–5.5 birds/m², and broiler farms 15–17 birds/m².

Fig. 1.

Sampling scheme across the production chain. A Number of farms sampled at each stage of the production chain/per company: 4 pullet farms, 8 breeder farms and 14 broiler farms were enrolled. The flock from one broiler farm was followed to the processing plant. B Sampling strategy at each farm and processing plant. For each farm, 10 litter samples and at least 3 soil samples were pooled to form a representative sample/farm. Similarly, 10 carcass rinses post-pick and post-chill were pooled to form a representative sample. C Then, all representative samples were shotgun sequenced. Note that for 1 pullet farm (FN2), 1 breeder farm (FN5) and 1 broiler farm (FN13), litter and soil were pooled in triplicate allowing greater resolution of the associated microbiome

This study was conducted from December 2021 to November 2023. Each farm typically had four to eight poultry houses, from which 2 were randomly selected for sampling. For each poultry house, two types of samples were collected (Fig. 1B): (1) Litter samples, representing the inside environment, consisting of a pool of 10 grab samples collected from various locations within the house; and (2) Soil samples, representing the outside environment, consisting of a pool of at least 3 topsoil samples from different locations surrounding the poultry house. Preference was given to soil in high traffic areas, such as near poultry house entrances to better capture the potential transit of contaminants between inside and outside. Then, for each type, samples from the 2 houses were pooled to create one representative litter and soil sample per farm (Fig. 1C). Furthermore, 1 broiler flock was followed to its respective processing plant. From this flock, 10 carcass rinses samples were collected from each post-pick and post-chill stage using 400 ml sterile water. The samples were then pooled to generate one representative sample for post-pick and post-chill respectively (Fig. 1C). All the collected samples were placed in ice and transported to the laboratory to be immediately stored at -80° C until further processing. Our study aims at capturing production-chain wide trends. Sample pooling decreases the number of representative samples to sequence, allowing us to maximize farm enrollment. In the case of carcass rinses, samples were pooled to allow the extraction of sufficient amount of DNA for sequencing. For 3 farms from Company A: FN2 (Pullet), FN5 (Breeder) and FN13 (Broiler), representative soil and litter samples were sub-sampled in triplicate and subjected to shotgun-sequencing as part of a pilot study. In total our study used 32 representative litter samples, 32 representative soil samples and 2 representative carcass-rinses for shotgun sequencing.

DNA extraction and shotgun metagenomic sequencing

For each representative litter and soil sample, DNA was extracted using Qiagen’s DNeasy PowerSoil Pro Kits^® following the manufacturer’s instructions. For carcass rinses, DNeasy PowerWater Kit^® was used according to the manufacturer’s instructions. The concentration and purity of extracted DNA were determined using a spectrophotometer, with absorbance measured at 260 nm. Purity was assessed by evaluating the A260/A280 and A260/A230 absorbance ratios.

DNA from each sample was shotgun sequenced by Novogene Co, Ltd. Briefly, DNA samples were quality checked to determine DNA concentration, using a Qubit fluorometer; sample integrity, using the Agilent 2100 bioanalyzer system, and sample purity by gel electrophoresis quantification. Samples were moved to library prep if total DNA amount ≥ 200 ng, concentration ≥ 10 ng/µL and A260/280 between 1.8 and 2.0. Illumina libraries were constructed using a VAHTS Universal DNA Library Prep Kit (Vazyme Biotech Co., Ltd). The library was checked with Qubit and real-time PCR for quantification, and with Agilent Bioanalyzer for size distribution detection. Quantified libraries were pair-end sequenced on an Illumina NovaSeq PE150 platform for a target of 20 million read pairs/sample. Overall, an average sequencing depth of 63.47 ± 29.04 M pairs/sample was achieved. Sequencing outcome details are listed in Supplemental Table 1.

The generated reads were quality-checked with FastQC v0.12.1 [19]. The reads were clipped, trimmed and short reads were discarded using Timmomatic [20]. Reads that lost their pair after trimming were retained and used as single-end reads. Microbiome profiling was performed using MetaPhlan 4 v0.2 [21] and resistome profiling was performed using ARGs-OAP 3 v.3.2.3 [22] with adaptations. Briefly, reads were mapped to marker databases, then read mapping information was used by each software to reconstruct the abundances of each feature in the sample. In the case of ARGs-OAP, a naïve adjustment is applied to multi-component systems, where features marked as part of a n-component system have their abundance multiplied by a constant to correct biases in functional resistance estimates. The correction is applied to all features marked as part of a multi-component system, irrespective of the presence of the other components (completion). Our study aims at capturing the ARG abundances without making claims about functional resistance, which requires additional approaches. Thus, we opted to reverse the naïve adjustment of counts for multi-component systems and work directly with gene abundances to avoid unwanted distortions in individual ARG counts and downstream diversity analyses. Feature count tables, i.e., species level genomic bins (SGBs) from Metaphlan and ARG alleles for ARGs-OAP, alongside database metadata and sample metadata were used for diversity and statistical analysis in RStudio v4.4.1, running R v4.3.1.

Alpha diversity analysis was performed using the package phyloseq v1.44 [23]. Tukey HSD tests were used to compare alpha diversity indexes between sample groups. Beta diversity analysis was performed using the package vegan [24]. Given the nature of sequencing count data and the fact that Metaphlan outputs relative abundance profiles, compositional data analysis-friendly approaches were used [25]. Microbiomes and resistomes were compared using the robust Aitchison distance index and statistical significance was determined by PERMANOVA with 999 permutations complemented by analysis of homogeneity of variances by PERMADISP using the package vegan [24]. Distances were visualized with Principal Component Analysis (PCA) ordinations using MicroViz [26]. Differentially abundant features in groups and biomarkers (i.e. differentially enriched features associated with specific groups) were identified using linear discriminant analysis effect size (LEfSe) [27] using MicrobiomeMarker [28].

Thresholds for defining the core microbiome change depending on application, study design and underlying methodologies [29]. In this study, core microbiome was defined as taxa present in 90% of the samples within each group. This threshold was chosen to accommodate a small degree of variability in composition while being stringent, especially when group size is small, allowing one taxon to be absent from one sample/group at most. Moreover, it has been used in other broiler litter microbiome studies [30]. All significance levels were set to P ≤ 0.05. All plots were generated with the package ggplot2 [31]. Panels were constructed using Adobe Illustrator^®. All scripts for bioinformatic and statistical analysis and plot generation are deposited at: https://github.com/laurahuber/Poultry_Farm_Resistome_HuberLab.

Results

We used shotgun metagenomics to reconstruct the microbiome and resistome of representative litter and soil samples from 26 farms belonging to two poultry production companies (herein referred to as Company A and Company B), covering the entire production chain (pullet, breeder, broiler farms and processing plant, see Methods). We identified 274 species across 28 phyla, including archaeal and eukaryotic representatives, and 6,547 ARGs associated with AMR to 29 classes of antimicrobials (AMR types) across the poultry production chain.

Shifts in microbiome across production stages and company

Soil samples had no significant difference in observed richness, diversity indexes, or microbiome composition across farm types (Supp. Fig. 1B), indicating that the farm soils constitute a constant and stable ecosystem. Overall, the litter microbiome was composed of less phyla on average (6 ± 2%) and was dominated by Actinomycetota and Bacillota representing over 93% of the microbiome in all farm types (Fig. 2A), followed by Pseudomonadota (1.33 ± 1.16%) and Bacteroidota (2.07 ± 3.78%). Interestingly, the phylum Bacillota encompassed mostly bacteria related to chicken gut and was present in overall lower abundances in litter, although their presence increases as we move down the production stages, peaking in broiler (Fig. 2A).

Fig. 2.

Microbiome and resistome composition and comparisons across the production chain. A Microbiome and resistome composition of litter and soil samples across the production chain. Soil samples showed no significant difference in microbiome and resistome, whereas litter samples changed depending on production stage. Changes in litter microbiome per stage become more apparent when focusing on dominant genera (genera with ≥ 5% relative abundance). B PCoA ordinations of microbiome (by species – as MetaPhlan’s species-level genomic bins (SGB) classification level) and resistome (by gene – as gene accession number hit). Top row contains all litter and soil samples showing the clustering of soil and litter samples. Bottom row contains only litter samples showing the clustering by production stage and company. C Top microbiome markers (genus level) and resistome markers (AMR type level), differentially abundant features identified by Linear Discriminant Analysis Effect Size (LEfSe). Note that no markers were identified for pullet litter resistome

There were significant differences in the litter microbiome compositions (PERMANOVA, p = 0.001) across production stages (Fig. 2B). Pullet litter had significantly lower observed richness than breeder (p = 0.019), and lower Shannon and Simpson Diversity indexes than both breeder and broiler farms (p < 0.05) (Supp. Fig. 1B). Interestingly, litter samples from pullet and breeder farms from Company B had significantly higher observed richness than Company A (p = 0.0037 and p = 0.0045, respectively) without significant impact on Shannon and Simpson diversity indexes.

A closer look at the most dominant genera in each sample (genera with abundance > 5%), illustrates the differences in microbial composition between the production stages. Brevibacterium was a dominant genus in litter across all farm types, with 53.40 ± 4.56%, 31.56 ± 10.55% and 36.26 ± 14.83% average relative abundance in pullet, breeder and broiler samples, respectively. Genomes associated with the family Salinicoccus and Ruania were identified as pullet litter biomarkers by LEfSe analysis (Fig. 2C). The genus Brachybacterium was enriched in breeder litter and was the strongest biomarker for this stage. Company had a significant effect on the microbiome composition (PERMANOVA, p = 0.025). More evidently in pullet and breeder litter, which tended to cluster by company in the ordination plots (Fig. 2B). Moreover, Ruania was a dominant genus in Company A breeder litter, while Mycobacteriaceae genome (SGB66232) was dominant a dominant genus in Company B samples, corroborating the significant effect of Company in these stages. In contrast, Broiler litter had a consistent microbiome composition even across companies. It was distinct in the dominance of Staphylococcus (17.63 ± 11.66%), identified as a biomarker of this stage. Other noteworthy dominant genera were Corynebacterium (8/14 farms, 9.01 ± 3.18%), Mycobacteriaceae genome (SGB66232) (8/14 farms, 7.20 ± 2.12%) and Lactobacillus (7/24 farms, 7.23 ± 1.95%).

Next, we sought to explore the core microbiome (species present in > 90% of the samples) associated with each production stage and to what extent microbes were shared between the litter of different stages. Only 16 bacterial species were shared between the core microbiome of pullet, breeder, and broiler litter (Supp. Table 2, Supp. Fig. 2), most notably the Micrococcales: Brachybacterium and Brevibacterium, and genera Corynebacterium and Lactobacillus. Indeed, Micrococcales alone accounted for at least 46% of any given litter community. Almost all microorganisms shared exclusively between breeder and broiler litter were associated with the chicken gut microbiota, while pullet farms shared more members of Actinomycetota.

Shifts in resistome across production stages and company

The resistome in litter samples followed a similar trend to those observed in the microbiome. Noticeably, despite soil having significantly higher diversity indexes (p 0.05) across all farm types, average observed richness was significantly higher in the litter from breeder and broiler compared to their corresponding soil samples (Supp. Fig. 1C). This indicates that resistome compositions in these stages were more skewed, i.e., dominated by fewer ARGs.

Litter resistomes were marked by dominance of fewer AMR types, most noticeably ARGs associated with tetracycline and macrolides, lincosamides, and streptogramins B (MLS_B), all clinically relevant antimicrobials [32]. Compositions were significantly different between production stages (PERMANOVA, p = 0.001). AMR type distribution is more even in pullet litter with proportionally lower levels of tetracycline and MLS_B, and higher levels of mupirocin and sulfonamide, although no resistance biomarker was associated with this stage. Overall, tetracycline and aminoglycoside AMR types were biomarkers of breeder litter (Fig. 2), however the resistome of breeder and pullet farms is significantly different depending on the company (PERMANOVA, p = 0.028). Company A samples have increased levels of tetracycline and aminoglycoside resistance, dominated by ARG tet33. Meanwhile, Company B samples had higher levels of bacitracin, mupirocin, and MLS_B resistance, with ARGs bcrA and ileS, being most abundant (Supp. Fig. 3). Broiler farms had similar compositions irrespective of company (Fig. 2). Broiler litter was significantly enriched in bacitracin, MLS_B and quinolone AMR types, which were the strongest biomarkers for this group (Fig. 2C). Bacitracin AMR type had at least 4× higher relative abundance, driven by gene bcrA, when compared to other stages, while MLS_B was 1.7× more abundant, related mostly to InuA and ermC in some farms (Supp. Fig. 3). Additionally, broiler farms carried higher levels of quinolone related ARGs, rare in other litter samples and absent from most soil samples. Even though its average relative abundance is lower (4.08 ± 2.03%) in comparison with dominant AMR types, it is still an increase of 3.39×, 6.31×, and 4.64× over pullet litter, breeder litter, and soil, respectively.

Frequency and relative abundance of potentially pathogenic species

Five relevant pathogens for the poultry industry were identified in the microbiomes and investigated in more depth: Campylobacter jejuni and Campylobacter coli, foodborne pathogens; Clostridium perfringens, causative agent of necrotic enteritis and gangrenous dermatitis in poultry; and Escherichia coli and Escherichia fergusonii, which have potential to cause both poultry and foodborne infections. No Salmonella spp. were detected in the samples.

Potentially pathogenic species were mostly absent in soils and only in extremely low abundances when present (Fig. 3). Campylobacter spp. were not detected in soil, and E. coli was sparse, being present in only 1/4 of soil samples and in low abundances (0.07 ± 0.08%). C. perfringens was also observed in the soil of 3 farms, albeit in low abundances (0.16 ± 0.19%). Notably, the soils carrying both E. coli and C. perfringens belonged to farms with potential contamination events (discussed below). In contrast, litter samples had significantly higher prevalence of potentially pathogenic species (92.30%) than soil (26.92%), (p > 0.0001). Only two farms (FN1 and FN2) had no potentially pathogenic species present in the litter microbiome. Apart from those, E. coli was found in the litter samples of all farms, while E. fergusonii was more prevalent in broiler. C. perfringens is present in 57.14% of broiler farms, mostly from Company A, while C. jejuni and C. coli are present in 64.28% and 57.14% of the broiler farms, respectively, mostly from Company B (Fig. 3A). Overall, Company B had higher levels of potentially pathogenic bacteria detected across its production chain when compared to Company A (Fig. 3B). Despite the high prevalence, all species had low relative abundance. Combined, the Escherichia representatives were the most abundant across pullet, breeder, and broiler litter microbiomes (0.10 ± 0.12%, 0.23 ± 0.19%, and 0.70 ± 0.70%, respectively), followed by C. perfringens and Campylobacter representatives – most abundant in broiler litter (0.04 ± 0.08% and 0.04 ± 0.06, respectively).

Fig. 3.

Pathogen load across the production chain. A Relative abundance of potentially pathogenic species (E. coli, E. fergusonii, C. coli, C. jejuni and C. perfrigens) in litter and soil across production stages. Bottom plot shows the presence/absence of pathogenic species in the litter and soil of each farm. B Prevalence of each pathogen across farms from the same company in each production stage. Companies had a different profile of pathogens, with Company B having a higher load of pathogens overall, although broiler farms from Company A are more likely to carry C. perfringens while Campylobacter spp. are more common on broiler litter from Company B. Despite the high load of pathogens in litter, it is contained from contaminating the soil for the most part

Microbiome and resistome of pre- and post-processing carcasses

To assess the possibility of farm-to-fork transmission of pathogens and AMR, one broiler flock (farm FN13) was followed to the processing plant. The microbiome and resistome of representative samples of the carcass after defeathering (post-pick stage) and after the chilling step, when the carcass is ready for packaging (post-chill stage), were reconstructed. There was a higher richness than expected for these samples, with 180 species recovered from post-pick and 80 from post-chill, however, the composition was highly skewed with most of the microbiota at low abundances (< 0.1%) (Fig. 4A). The same is true for the resistome, where only 25/246 ARGs identified in the post-pick sample and 26/1,754 in the post-chill sample, had relative abundance above 0.1% (Fig. 4B).

Fig. 4.

Microbiome and Resistome composition of processing plant and associated farm samples. One flock (from farm FN13) was followed to the processing plant. Panel including comparisons of microbiome and resistome of litter samples from farm FN13, and the flocks post-chill and post-pick samples at the processing plant. A Microbiome composition of genera with abundance ≥ 1% in farm FN13’s litter, post-pick and post-chill carcass rinses. Venn diagram shows species shared between farm litter and carcass rinses. Litter and post-pick shares about half of the species as expected, since both are colonized with chicken gut microorganisms. More importantly, no microorganisms are shared with post-chill, indicating no farm-to-fork transmission. B Resistome composition in farm FN13 litter, post-pick and post-chill carcass rinses. Venn diagram shows ARGs shared between farm litter and carcasses. As expected, there is a higher number of shared features due to the HGT potential of ARGs across multiple microbial groups

The carcass microbiomes from the post-pick stage were dominated by the phylum Bacillota (76.59%), namely the Lachnospiraceae and Oscillospiraceae, followed by Pseudomonadota (14.69%), composed mostly of Helicobacteraceae spp., although both C. jejuni and C. coli were present at appreciable relative abundance (3.03%) (Supp. Fig. 4A). On the other hand, the post-chill (ready to package) samples were dominated by Pseudomonadota (89.56%) – mainly in the form of Pseudomonas spp., Buttiauxella spp., and Janthinobacterium spp. – with potential pathogenic Aeromonas spp. and Yersinia enterocolitica found in low levels (2.42% and 0.82%, respectively). Bacillota was the second most abundant phylum (10.22%), corresponding mainly to Carnobacterium spp. (9.63%).

The post-pick carcass and corresponding broiler litter samples shared 90 species, while the post-chill carcass shares no species with either sample type. The post-pick resistome had prevalence of tetracycline, MLS_B, and aminoglycoside AMR types, on the other hand, post-chill sample has a very distinct profile, dominated by multidrug resistance-related ARGs, followed by polymyxin, bacitracin, and MLS_B AMR types. The resistomes of the two processing stages had only 2.05% of the ARGs shared between the two. When only ARGs with abundance over 0.1% are considered, only 3 ARGs are shared between the two stages.

Microbiome and resistome from inside and outside the poultry house

Next, we sought to evaluate the microbiome and resistome overlap between the inside of the poultry house (litter samples), and the outside environment (soil samples) to assess the confinement and traffic of microorganisms and ARGs within the farm. Microbiome and resistome compositions of litter vs. soil across all farms were significantly different (PERMANOVA, p = 0.001), with all soil samples forming a distinct and isolated cluster (Fig. 2B). Soil microbiome composition encompassed 18 phyla and was dominated by Actinomycetota (57.91 ± 2.94%) and Pseudomonadota (23.68 ± 4.18%), followed by Bacillota (8.46 ± 4.72%) and Bacteroidota (3.68 ± 1.11%) with many soil samples carrying noticeable amounts of Cyanobacteriota and Deinococcota (Fig. 2A). Most soil diversity occurred at low abundances (< 1%). Soil dominant genera made up a smaller proportion of the community and shifted from sample to sample, with Microbacterium being the most frequent dominant genera. Soil had higher observed richness and diversity compared to litter samples across all farm types, being significantly higher in broiler by all alpha diversity metrics (p 0.05, Supp. Fig. 1A). There was little overlap between the microbial species identified in both environments with farms sharing on average 7.18 ± 4.52% (pullet), 3.92 ± 3.73% (breeder), and 2.60 ± 1.31% (broiler) of species between litter and soil. The litter dominant genera were rarely observed in soil, and at much lower abundances when present (Supp. Fig. 4C).

The composition of soil resistome was also consistent across all farms, regardless of production stage, mirroring the microbiome observations (Fig. 2A, Supp. Fig. 3). Genes associated with resistance to multiple antimicrobials, usually coding for efflux pumps, were the most dominant AMR type (multidrug AMR type) in the soil (34.45 ± 9.67%), while comprising only 8.20 ± 4.39% of the litter resistome on average. Soil had higher prevalence of Novobiocin (5.37 ± 3.18%) and rifamycin (8.17 ± 4.85%) compared to litter (1.08 ± 0.93% and 2.38 ± 2.49%, respectively). Unlike litter resistome which was dominated by clinically relevant ARGs, soil had a more even distribution with most samples carrying no dominant ARGs (relative abundance > 5%). In contrast with the microbiome, there was a larger proportion of ARGs shared between soil and litter environments: pullet (29.22 ± 7.38%), breeder (24.20 ± 9.55%), and broiler (17.24 ± 4.90%).

However, in two broiler farms, FN13 and FN7, the soil microbiome and resistome are reminiscent of litter (Fig. 5). For one of these farms, FN13, all sampling and sequencing were performed in triplicate which adds robustness to our observations. Soil samples from these farms clustered towards the broiler litter samples in the microbiome and resistome ordination plots (Fig. 5B). Moreover, the soil microbiome composition at phylum and genus level in these farms mirrored that of their paired litter samples, while being different form other soils (Fig. 5A). Bacillota is a dominant phylum in broiler litter, 50.40 ± 21.74% relative abundance, compared to 5.14 ± 7.09% in soils. Soils from FN13 and FN7 have a distinct predominance of Bacillota, with 52.30% and 20.06% relative abundance, representing a 10× and 4× increase, respectively. Moreover, the soil of farm FN13 was dominated by the litter biomarkers Brevibacterium, Staphylococcus, Ligilactobacillus, Lactobacillus, and Mammalicoccus, and farm FN7 soils are dominated by the litter biomarkers Brevibacterium and Brachybacterium (Fig. 5C). Meanwhile, other broiler soils carried litter biomarkers only in low abundances. Moreover, farm FN13 and FN7 share 6.07× and 4.47× more microorganisms between litter and soil, than the average of other broiler farms, with a higher impact in soil microbiome composition. Altogether, these two farms shared 18% of the microbiome between litter and soil, while all other farms pooled shared 4.6% (Fig. 5D). The ARG composition of soil samples of the two farms are enriched in litter AMR type biomarkers (Fig. 5A and C). Most notably, bcrA, which dominated broiler litter, was detected at higher relative abundances in the soils of these two farms (FN13 = 21.48% and FN7 = 11.02%), compared to the soil average (5.59 ± 3.58). Additionally, the two farms have increased abundance of ARGs associated with quinolone resistance (FN13:5.34× and FN7:3.68×), another significantly enriched AMR type in broiler litter.

Fig. 5.

Evidence of contamination from litter to soil in two broiler farms. A Dominant genera and dominant ARGs compositions of litter and soil samples from broiler farms. Box and arrows highlight farms FN13 and FN7, where soil samples have a high abundance of genera and ARGs associated with litter. Bottom panel shows the relative abundance of soil and litter markers identified by LEfSe analysis in the litter and soil of each farm. Soils in farms FN13 and FN7 are enriched with litter biomarkers. B Ordination of broiler samples shows that soil samples from farms FN13 and FN7 cluster towards other litter samples and away from other soils. C Soil and litter microbiome and resistome markers in broiler farms identified by LEfSe (and used in panel A). D Shared features (Species for Microbiome and ARGs for Resistome) between litter and soil across non contaminated and contaminated soils (farms FN13 and FN7)

Discussion

In the US, several steps have been taken to curb the indiscriminate use of antimicrobials in poultry production as part of efforts to curb the spread of AMR, however the impact of such practices has not been extensively investigated. Several reports focus on the AMR profile of poultry feces or litter samples in specific production stages [33–38], but rarely consider the entire vertically integrated production chain. In our study, we focused on two vertically integrated companies to explore the microbiome and AMR dynamics across the poultry production chain and between the environments inside and outside the poultry houses in poultry operations with restricted AMU.

Litter microbiome is a complex matrix composed of microbiota from bedding, soil and bird excreta, alongside poultry house abiotic conditions [39–41]. We identified a core litter microbiome conserved across the production chain, dominated by Actinomycetota families: Micrococcales and Corynebacteriales which were previously associated with the litter environment [42]. Interestingly, genera associated with the chicken gut, such as Bacteroides, Lactobacillus or Faecalibacterium [43], did not dominate the microbiome despite the constant shedding via feces, indicating inefficient colonization of the litter environment. This dynamic suggests that core members of the Actinomycetota phylum could play a key role in the maintenance of AMR in litter by uptaking resistance factors via horizontal gene transfer from transient chicken gut microbial populations and reintroducing them to next flocks. Hence, proper litter management, including limited reuse between flocks, are essential in curbing the spread of AMR [44].

Overall microbiome and resistome compositions shift significantly across production stages, fitting the core-satellite distribution model, with most of the diversity and differences between groups in the satellite compartment [45]. Thus, differences in flock characteristics and management are important factors in shaping the litter microbiome. Pullet litter was the least diverse, likely reflecting the immature gut microbiota of younger birds [46], less fecal shedding due to lower stocking [47]. Conversely, the breeder stage had input of a mature gut microbiota [48] since birds were older, higher stocking density, and longest housing period, contributing to the development of a more diverse litter microbiome. Broiler litter microbiome was consistent across farms and companies, and had similar composition to that observed in a study with 37 commercial broiler flocks which also reported high consistency in broiler litter microbiome [30]. It was enriched in gut-related genera when compared to other stages. Broiler rearing is the most intensive production stage, with the highest bird density (15–17 birds/m²) and routine reuse of litter across multiple flock cycles as a cost-saving practice [49]. Longer litter use increases the proportion of intestinal bacteria in the litter bed microbiota [48], and alongside increased bird density, creates the conditions for the accumulation of gut microbes in the litter with impacts on pathogen and AMR persistence. Poultry litter is nutrient-rich rich and approximately 14 million tonnes of poultry litter are generated each year in the US [50]. Its high nutrient content and its wide availability make it a valuable fertilizer [50] and an important component of circular economy practices [51, 52]. However, litter application in cropland can facilitate the dissemination of pathogens and ARGs into the environment [51, 53, 54]. Moreover, the spread of antimicrobial resistance in poultry systems is complex, as air, water, and soil can serve as major environmental reservoirs and transmission pathways [44, 55].

We found that breeder and broiler litters were associated with a distinct resistome profile, all containing high levels of resistance to medically relevant antimicrobials despite restricted AMU. Both NAIHM and NAE farms will use antibiotics to safeguard flock health in the case of infection outbreaks. Since our farms reported no AMU three months prior to sampling, these findings likely reflect historical AMU. Studies reported enrichment of ARGs related to poultry production for five years after AMU [56]. Antimicrobial compounds can form lingering residues [57], driving the maintenance of otherwise fitness-impairing resistance factors [58], even fixing resistance phenotypes in the population [59]. For example, company is major factor in the abundance of tetracycline, MLSb and aminoglycosides AMR types in breeder litter, pointing towards company specific AMU history as a driver of resistome composition. However, since both resistome and microbiome clustered by company, it is possible that differences are explained in part by taxa fluctuations related to other company-specific factors such as diet [60] and flock genetics [61]. Broiler litter carried distinctively prominent levels of MLS_B, bacitracin and, surprisingly, quinolone related ARGs. Bacitracin is not considered important for human medicine in US and is often employed to prevent and control necrotic enteritis and other diseases in NAIHM farms [18, 62], however this practice is not permitted in the NAE farms. Surprisingly the NAE farms had similar levels of bacitracin resistance to the NAIHM farms. Bacitracin’s short half-life [57] dampers the effects of residue pollution due to historical AMU and is unlikely to sustain prolonged selective pressure after use. It is important to note that the gene bcrA accounts for the majority of bacitracin AMR type abundance. Yet, resistance is conferred by the operon bcrABD and requires the presence of both bcrA and bcrB for the phenotype [63]. We found that the overrepresentation of bcrA over bcrB in the reconstructed resistomes reflects a bias towards bcrA sequences over bcrB in the database [22]. Thus, the bcrA gene abundance is a rough proxy and cannot alone indicate functional resistance. On the other hand, quinolones are a critically important class of antimicrobials in human medicine [15] and its use has been banned in poultry since 2005 due to the association with the increase in human fluoroquinolone-resistant Campylobacter infections [64]. Therefore, in both cases increased AMR levels in NAE broiler farms cannot be attributed to direct antimicrobial pressure from past use nor present, highlighting the need to uncover alternative mechanisms of persistence, such as co-selection [65] including by other environmental contaminants [66]. Moreover, United States Department of Agriculture (USDA) NAE label regulations only apply directly to broiler flocks, and it is unclear how this label extends to previous production stages. Antimicrobial residues can be transmitted vertically from breeder hens to broiler chicks [67], jeopardizing NAE flocks without technically violating the label policies. Moreover, the USDA reported the detection of antimicrobial residues in 20% of NAE cattle carcass [68], emphasizing the importance of accounting for alternative routes of antibiotic exposure. Ongoing studies are focusing on combining microbiome and resistome to unveil maintenance mechanisms of AMR reservoirs, characterize vectors and mobile elements and AMR transmission routes within and across communities.

Potentially pathogenic bacteria were detected in increasing prevalence and relative abundance as we moved down the production chain, with broiler farms having higher prevalence and relative abundances. Company B (NAIHM) had a higher prevalence of potential pathogens across the production chain than Company A (NAE), which might reflect stronger biosecurity practices of the latter since NAE operations rely heavily on them for disease prevention and flock health [9]. All the potentially pathogenic species tracked were rarely detected in the soil, pointing to the poultry house as their source environment. NAE broiler farms had higher prevalence of C. perfringens which is expected since they do not use the approved antimicrobials for prevention and control of necrotic enteritis which are routine in NAIHM farms. However, this prevalence might not represent a risk since not all strains of C. perfringens are virulent and our analysis did not attempt to perform strain annotation. We identified high prevalence of Campylobacter in Company B farms, especially in broiler. This finding is corroborated by work on Campylobacter prevalence on the same company using culture isolation and whole-genome sequencing approaches [69], which showed that 61% of the poultry houses tested positive with an increase odds-ration of Campylobacter presence in broiler farms. Escherichia spp. was the most frequent and abundant species recovered, being present in the litter of 24/26 farms and in higher abundances in broiler. These estimates align with data on E. coli isolation performed on both companies (Manuscript under review) that showed widespread prevalence and increase in broiler. This study validates the concern for the increase of potentially pathogenic species in general, showing an enrichment of avian pathogenic E. coli in broiler. Interestingly, Escherichia fergusonii, an emerging zoonotic pathogen [70–72] associated with extraintestinal infection in poultry [71], was detected in litter of 2/8 breeder farms and predominantly in broiler litter samples (10/14 farms). Recently, E. fergusonii isolated from food animals has also emerged as reservoir of AMR [73]. E. fergusonii presence poses a concern not only due to its pathogenic potential but also because of its role in dissemination of AMR, highlighting the need for its continued monitoring, although its detection across healthy broiler farms raises the question of whether it might be part of a normal microbiota, as suggested by similar observations in Canada [74].To our knowledge, there are no reports on E. fergusonii from integrated broiler production in the US, but additional approaches such as pathogen-specific amplification and isolation efforts are necessary to confirm our initial observations. Our study did not detect any Salmonella spp. in any of the metagenomes. Longitudinal sampling of Company A detected Salmonella sporadically in the farm environment but did not recover any isolates from surfaces inside the poultry houses [75]. Moreover, Salmonella was also detected in a small proportion of pullet, breeder and broiler farm environments of Company B, including surfaces inside the poultry house [76]. Pathogen detection from shotgun metagenomics is challenging. Because pathogens are usually present in low abundances they are often at the edge of detection thresholds, i.e. abundances so low that they cannot be reliably differentiated from false positives [77]. The marker based approach used in this study for microbial profiling is very recalcitrant to false positives, offering a trade-off between recall and precision [21, 78–80], reported to lead to lower but precise detection of pathogens in metagenomes, such as low abundances of Salmonella in both synthetic and real datasets [80, 81]. Nonetheless, interpretation of low abundance data requires caution, thus we offered supporting data obtained from pathogen isolation work to complement our findings. Our analysis of potentially pathogenic species failed to detect elusive Salmonella spp. presence, as expected [80], but accurately captured the overall trends also captured by culture-dependent studies.

We did not observe farm-to-fork contamination as no microorganisms and minimal number of ARGs were shared between the farm environment and processed carcass rinses. Our results point to the processing plant as the main source of post-chill carcass microbiome and resistome, including Janthinobacterium spp., associated with meat spoilage [82] and potential foodborne pathogens A. hydrophila and Y. enterolitica., while shared ARGs likely correspond to common intrinsic genes as some ARGs are widespread [83]. Indeed, taxa shared between litter and post-pick carcass rinses, mostly associated with the chicken gut such as Clostridia [84] and potential pathogens C. coli and C. jejuni, were successfully eliminated from the processed post-chill stage, in line with previous studies showing decreased odds-ratio of Campylobacter presence in post-chill samples from processing plants [69]. Thus, it simultaneously highlights the efficacy of carcass processing and the need for strict biosecurity measures when managing meat products to avoid contamination. However, farm-to-fork contamination is not ruled out. Our sampling efforts were limited (one processing plant, one time point), and low abundance organisms can evade detection in shotgun sequencing efforts. We showed that our study had a limited ability to detect Salmonella in farm samples, and although this pathogen was not detected in the carcass samples, a previous study detected Salmonella spp. across multiple production stages of this same company, with the highest prevalence reported in transport trucks and the processing plant [76]. Furthermore, our findings rely on pooling all sampled carcasses for each group together for sequencing, effectively eliminating all variations present in the original samples. Further studies with true biological replicates are needed to validate our initial findings.

Overall, the environment inside the poultry house was contained from the outside, indicating limited spread of microbes and AMR from the poultry houses to the outside environment and vice versa. Our results corroborate the efficacy of AMR containment of biosecurity protocols, as reported elsewhere [85, 86]. However, we did detect evidence of microbial transfer from within poultry houses to the surrounding environment in two farms, including potentially pathogenic organisms and ARGs. Since soil samples came from high traffic areas, there is high likelihood that the movement of poultry farm workers served as a vector for contamination of outside environment, as previously reported [87]. Thus, we show that biosecurity breaches can lead to sufficient contaminant load to change the microbiome and resistome profiles of surrounding soil, highlighting the need for biosecurity enforcement to mitigate the spread of AMR. Beyond structural biosecurity, commercial broiler operations rely on multiple biosecurity practices that contribute to limiting microbial and ARGs dissemination in the poultry system. These include routine cleaning and disinfection, water sanitation, proper ventilation, cleaning between flocks, appropriate bird density, all-in-all-out system, and litter management practices [88–92], although these were not assessed in our study.

Limitations

One of the major limitations of our study was the lack of antimicrobial use records, precluding the establishment of a direct link between our findings and AMU events. On-farm usage data is one of the main bottlenecks in antimicrobial stewardship efforts [18]. Additionally, other variables that might potentially shape the AMR in poultry production such as nutrition, farm management practices, farm demographics, and socio-geographical factors need to be incorporated. Increased transparency is indispensable for accurate understanding of the impacts of AMR guidelines on the AMR burden of broiler production. Initiatives such as the National Antimicrobial Resistance Monitoring System and funding for initiatives to support data collection in US animal agriculture [93] will ameliorate this issue. Additionally, future studies focusing on longitudinal antimicrobial residue quantification are needed to elucidate the mechanisms of historic AMU in AMR maintenance. Our study is based on shotgun sequencing of representative pooled samples which can reduce the final intra-group variability and exacerbate statistical group differences. Although a sharp increase in group size was favored to recuperate statistical power, in depth studies with true biological replicates are necessary to increase the power of microbiome and resistome analysis. While metagenomics provides valuable insights into community dynamics, the inconsistent detection of low abundance organisms and inability to generate absolute load information, hinders its surveillance capabilities [94]. Thus, quantitative methods, culturing and isolation remain crucial for studying AMR dynamics and must complement metagenomic studies. Finally, our study is cross-sectional and cannot, in isolation, provide insight into the temporal changes of the AMR landscape in broiler production. Long term longitudinal studies are necessary to assess the effects of AMU guidance and regulations in broiler production.

Conclusions

This study offers a novel perspective on the dynamic and interconnected nature of the microbiome and resistome throughout different stages of vertically integrated poultry production systems, extending to both the inside environment of poultry houses and the surrounding soil. Our findings describe shifts in microbiome and resistome profiles along the production chain showing the persistence of high abundance of AMR related to medically important antimicrobials despite restricted AMU. We show that microbiome and resistome are linked to company specific factors, highlighting the importance of management strategies in shaping AMR diversity in broiler production. Moreover, our findings provide empirical evidence of the importance of farm practices in curbing the diversity and spread of AMR, even in companies with restricted AMU. Beyond descriptive surveys, effective AMR data needs to apprehend processes of maintenance and transmission to generate effective risk assessment and interventions. Here, we present the first snapshot of the AMR and microbiome landscape across vertically integrated systems. On going studies are focusing on combining the two datasets to identify AMR reservoirs, vectors and routes of AMR movement across environments with a focus on the role of horizontal gene transfer in shuttling AMR into bacteria of interest; using qPCR to quantify AMR in this sample set to identify AMR absolute enrichment across the production chain; and tracking model pathogens across the production chain using comparative genomics.

Authors’ contributions

PPG, ALRSP, LH: Conceptualization, Writing – original draft, Writing – review & editing draft, Data Processing, Analysis, Visualization. PPG, RG, AL, CH, YG, MB, KM: Sample Collection, Laboratory Analysis. LH: Supervision. All authors reviewed and approved the final manuscript.

Funding

This project was supported by the United States Department of Agriculture (USDA) under Grant No. G00017930, and by intramural funding from the Alabama Agricultural Experiment Station at Auburn University.

Data availability

All Illumina reads generated in this study are deposited in the NCBI SRA database: BioProject PRJNA1309778. All scripts for bioinformatic and statistical analysis are deposited at: https://github.com/laurahuber/Poultry_Farm_Resistome_HuberLab.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.