Transmission of AMR Campylobacter and Escherichia Coli During the Processing of Chicken Meat

undefined

Acknowledgements

The authors would like to thank the chicken processing sites visited in this project for their generous access to their data and support and access to their sites.

The authors would like to thank Miss Hamida Musah Alhassan, Mr Oliver Horne, Miss Sophie Bowers, Mr Bamidele Adedeji, Mr Jesse Adenola, Dr Essam Hebishy, and Dr Samson Oyeyinka for their involvement in the sample collection and sample preparation stages of this project.

We would also like to thank the UK Food Standards Agency for funding this work.

Glossary and abbreviations

Lay Summary

The emergence and spread of resistance to antimicrobials in bacteria, viruses, fungi, and parasites is of global concern. There is increasing concern that the food chain and food processing environments may significantly contribute to the transmission of antimicrobial resistant bacteria and genes (ARGs), potentially serving as hotspots for the acquisition and spread of antimicrobial resistance (AMR). However, relatively little is known about the role of the food chain in the transmission of AMR bacteria in general, including the transmission and prevalence of AMR bacteria in chicken and chicken products, before retail. The overall aim of this project was to access the impact that processing has on the presence and transfer of antimicrobial resistant Campylobacter species (spp.) and Escherichia coli and associated genes on chicken meat. Since poultry accounts for half of the meat eaten in the United Kingdom (UK) it potentially represents a significant reservoir for AMR to transfer to humans. This project therefore was intended to act as a baseline for work comparing two similar typical large scale UK chicken plants using traditional and newer approaches, as well as an incentive for further testing, to establish the importance of different processing steps in AMR transmission and identify mitigating strategies for reducing and eliminating persistence and transmission of AMR.

This project was carried out in two parts. Part 1. an initial literature review which was carried out to aid the project sampling plan/design and part 2. a field and laboratory study conducted to assess the presence and potential transmission of AMR- Campylobacter spp. and E. coli during the processing of chicken in two similar sized large scale UK chicken processing sites producing whole chickens and chicken meat for major UK retailers at different points in the year.

Key findings

Our literature review found that:

• Few published studies have investigated the transmission of AMR bacteria or ARGs during poultry processing, with existing research suggesting that AMR and ARGs in outgoing meat are primarily influenced by farm-level factors rather than in-plant procedures. There is limited published evidence on the transmission and persistence of AMR bacteria and genes occurring in processing environment.

• No published studies have looked at the impact of physical interventions (such as steam) that may be employed to reduce pathogenic contamination on chicken carcasses on the transmission of AMR. Such interventions are now being employed by some UK chicken processors and employed in both plants studied in this project.

Our field and laboratory study found that:

• Of the 376 samples collected from the poultry plants during this current project, 65.1% of samples were positive for presence of Campylobacter spp. and 95.6% were positive for presence of E. coli.

• Levels of Campylobacter spp. isolated from sample collected from the 2 UK plants ranged from 1.00±0.0 to 3.9±0.5 log₁₀ Colony Forming Units (CFU) per sample; with the highest level recorded from chicken litter and the lower levels recorded from post intervention equipment, scalding water, clean crates, and on carcasses at dispatch.

• Levels of E. coli isolated from sample collected from the 2 UK plants ranged from 2.0±0.0 to 8.3±0.2 log₁₀ CFU per sample; with the highest level recorded from chicken litter and the lower levels recorded from post intervention equipment, post intervention carcasses, scalding water, post chill, inside outside wash and clean crate processing stages.

• Campylobacter spp. and E. coli were detected at different stages throughout the production process in both plants, with higher numbers measured at the earlier stages of the process in both plants.

• At some stages of the process no Campylobacter spp. or E. coli were isolated in either plant.

• Generally, levels of Campylobacter spp. decreased to non-detectable levels in samples taken further along the processing stages. For E. coli while there was a general decrease in the level of the organisms, samples tested at the later stages of processing remained still positive for the presence of E. coli.

• The resistance of Campylobacter spp. isolates to 5 antimicrobial agents were tested and E. coli isolates were tested against 11 antimicrobial agents. Of the Campylobacter spp. isolates, we recorded the highest overall percentage resistance to tetracycline (53%) and lowest overall percentage (7%) resistance to erythromycin. For E. coli isolates, we recorded the highest overall percentage resistance to ampicillin (80%) and lowest overall percentage (13%) resistance to chloramphenicol.

• Approximately 7% of the Campylobacter jejuni isolates were resistant to three or more classes of antimicrobial (tested against examples of four different classes).

• Approximately 60% of the E. coli isolates were resistant to three or more classes of antimicrobial (tested against examples of eight different classes).

• Analysis of the genetics of cultured C. jejuni and E. coli isolates (using whole genome sequencing [WGS]) showed that both C. jejuni and E. coli isolates tested were genetically diverse across sampling points and periods. However, E. coli displayed greater variability in sequence types (STs). In total, 5 C. jejuni STs (3 recognised) and 40 E. coli STs (36 recognised) were identified.

• Four C. jejuni STs (21, 262, 5136, 6175) were identified across sites, with ST 6175 being most prevalent. All STs isolated carried resistance genes for ampicillin or tetracycline, and a gyrA gene mutation conferring resistance to ciprofloxacin and nalidixic acid.

• E. coli isolates from both sites showed high genetic diversity with 36 distinct STs identified. Prevalent STs included 10, 155, and 6448. Isolates carried multiple resistance genes and plasmids (genetic elements that can transfer genes between different organisms), with bla_TEM and tet(A) genes being common. Plasmid types varied across sampling periods, with Col156, Inc1-l(Alpha), and P0111 being frequently observed.

• An analysis of samples taken at different processing stages using shotgun metagenomic sequencing (which detects DNA from all organisms within a sample) showed a wide diversity of bacteria, ARGs, and bacteriophages (naturally occurring viruses than target bacteria, usually called phage) present at both poultry sites on the different sampling periods.

• Overall, there were similarities in the composition of the microbiomes at both sites, though there were some differences in the overall diversity (number of different bacteria, ARGs, or phage) between both sites and some seasonal differences (though mainly only at one site).

• Interestingly, the most abundant bacteria, ARGs, and phages were the same (though not always in the same order) at both sites and prevalent throughout processing. However, the overall diversity (number) of different bacteria, ARGs, and phages present at different processing stages reduced during processing at both sites.

Overall conclusions

Combining traditional culture methods with modern genomics dependent on DNA sequence analysis of bacteria and ARGs, we created a detailed map of bacterial and gene presence in the poultry processing units under investigation. Our findings using traditional methods showed that while target bacteria like Campylobacter spp. and E. coli were initially present in birds entering the plants, their presence and numbers significantly decreased along the poultry processing line especially following specific procedures such as defeathering and evisceration (internal organ removal). The prevalence of Campylobacter spp. was generally lower than E. coli overall in the samples collected. Specifically, only 65.1% of the total samples from various stages tested positive for the presence of Campylobacter spp., whilst 95.6% of total samples were positive for E. coli.

Antimicrobial resistance testing revealed varying patterns among Campylobacter spp. and E. coli isolates. Campylobacter spp. showed highest resistance to tetracycline (53%) and lowest to erythromycin (7%), while E. coli exhibited highest resistance to ampicillin (80%) and lowest to chloramphenicol (13%). Notably, 7% of Campylobacter spp. and 60% of E. coli isolates were resistant to three or more antimicrobial classes, indicating multidrug resistance (MDR), particularly in E. coli.

Genetic analysis via WGS revealed diverse C. jejuni and E. coli isolates across sampling points. Five C. jejuni STs were identified, with ST 6175 most prevalent, all carrying resistance genes. E. coli showed higher diversity with 36 STs, commonly carrying bla_TEM and tet(A) genes. Various plasmid types were observed in E. coli isolates. Genomic analysis of parallel samples confirmed reductions in bacterial and associated ARG diversity through the processing chain, suggesting that the hygienic measures implemented during the poultry processing stages can effectively reduce both bacteria and ARGs. The fate of some resistant bacteria appears less certain and for reasons unknown persist throughout the poultry processing stages. It should be noted that this study is based on only two large scale UK chicken processing plants and may not be a reflection of all chicken processing plants in the UK. Further studies are needed to confirm these results and assess the role of poultry products in spreading AMR/ARGs of concern to human health.

Executive Summary

The overall aim of this project was to access the impact that processing has on the presence and transfer of antimicrobial resistant Campylobacter spp. and E. coli and their associated genes on chicken meat. There is a growing concern that some AMR transmission to humans occurs via the food chain and food processing environments and that these could act as potential hotspots for AMR acquisition and spread. However, relatively little is known about the role of the food chain in the transmission of AMR bacteria, including the transmission and prevalence of AMR bacteria in chicken and chicken products before retail. Since poultry accounts for half of the meat eaten in the UK (BPC, 2018) it represents potentially a significant reservoir for AMR.

Key findings

Although only few studies have focused on the transmission of AMR bacteria or genes during poultry processing, existing research suggests that AMR and ARGs in incoming birds are primarily influenced by farm-level factors rather than in-plant operations leading to cross contamination. There is very limited evidence on the sequential transmission of AMR bacteria and genes during chicken slaughter or the persistence of AMR bacteria and genes in the processing environment. While some studies indicate that ARGs can persist in processing environments and contribute to cross-contamination, the evidence is scarce. Additionally, the impact of cutting operations and physical interventions such as steam on AMR or ARGs in UK plants remains largely unexplored.

This study took samples along the processing chain of two similar sized large scale UK chicken processing sites producing whole chickens and chicken meat for major UK retailers on 3 occasions during the year (March, June, and October/November).

In this study of 376 samples from most stages of poultry processing plants, overall, 65.1% were positive for Campylobacter and 95.6% for E. coli. Levels of Campylobacter spp. ranged from 1.00±0.0 to 3.9±0.5 Log₁₀ CFU/sample, while E. coli levels ranged from 2.0±0.0 to 8.3±0.2 Log₁₀ CFU/sample, with chicken litter (pre-processing stage) showing the highest levels for both bacteria. Both organisms were detected at various processing stages in both plants, with higher counts in earlier stages. Some processing stages yielded no detectable Campylobacter spp. or E. coli in either plant. Interestingly, Campylobacter spp. levels decreased from samples along the processing stages, to non-detectable levels while E. coli, although showing a general decrease, remained detectable in samples from later processing stages.

Out of testing against 5 antimicrobial agents, Campylobacter isolates showed the highest overall resistance to tetracycline (53%) and the lowest overall percentage (7%) to erythromycin. For E. coli isolates, tested against 11 antimicrobial agents, the highest resistance was observed for ampicillin (80%) and the lowest overall percentage for chloramphenicol (12%). MDR was observed in approximately 7% of Campylobacter spp. isolates and 60% of E. coli isolates.

WGS analysis of C. jejuni and E. coli isolates revealed genetic diversity across sampling points and periods, with E. coli showing greater variability in sequence types (STs). Five C. jejuni STs, including ST 6175, which was most prevalent, were identified, all carrying resistance genes for ampicillin or tetracycline, as well as a gyrA mutation conferring resistance to ciprofloxacin and nalidixic acid. E. coli isolates demonstrated high genetic diversity with 36 distinct STs identified, including STs 10, 155, and 6448, carrying multiple resistance genes and plasmids, the resistance genes bla_TEM and tet(A) being common. Plasmid types varied across sampling periods, with Col156, Inc1-l(Alpha), and P0111 frequently observed.

The shotgun metagenomics sequencing identified a diversity of bacteria, ARGs, and phages present across the processing lines at both sites. The results showed a wide diversity of bacteria, ARGs, and phages to be present at both sites on the different sampling periods, many of which only occurred on the individual sampling occasions. However, not surprisingly, there were similarities in the dominant composition of the bacteria, ARGs, and phages at both sites, though there were differences in the overall diversity between both sites and some seasonal differences (though many only at one site). The most abundant phyla were Proteobacteria, Firmicutes, and Actinobacteria, which accounted for the majority of detected phyla in all samples. These same phyla have been reported to be the most abundant phyla in poultry in other studies in other countries (though not always in the same order). Acinetobacter spp. were the most abundant species at both sites, particularly A. johnsonii, lwoffii, and gandensis, and were detected in samples from sampling points along the whole processing chain (from entry to dispatch) at both sites.

The results from the metagenomics revealed the occurrence of a total of 442 different ARGs which may be involved in resistance to 16 different classes of antimicrobial agents. The most abundant and commonly occurring ARGs in both plants were genes coding for tetracycline resistance. This study also detected the prevalence of phage which could play an important role in the transfer of ARGs within the microbiome or processing environment since they can be associated with the transduction of genes, including ARGs. The overall diversity of bacterial phyla, target bacteria, ARGs, and phages were clearly shown to reduce during processing at both sites. Similar reductions in the diversity of the microbiome through the poultry processing chain have been observed in limited other studies of the poultry processing environment carried out in other countries.

Conclusion and recommendation

The classical approach of traditional culture of the target pathogens combined with powerful genomic analysis allowed us to provide a detailed map of spatial and temporal distributions in the process chain of a working poultry environment. We were able to conclude that the target organisms were readily recovered from the process chain at early stages but following several identified processing stages reduced numbers of Campylobacter spp. and E. coli were recovered. Parallel sampling for genomic analysis demonstrated extensive bacterial diversity from early stages of the process which was concordant with associated ARG diversity. The loss of ARG diversity following specific interventions in the process chain suggests that ARGs along with their host bacteria may also be reduced by the interventions. However, some ARGs may persist because they are carried by particularly resilient bacteria to dispatch. We should consider whether these results can be repeated and are representative of other UK processor operations and to what extent poultry products represent a significant reservoir for AMR/ARGs. Further studies are needed to clarify if the microbiome composition and changes in diversity observed in our study are representative of other UK-wide poultry plants.

This study (in common with others) shows that shotgun metagenomic sequencing of processing environments is an appropriate approach to investigate how microbiome composition and diversity changes during processing and provide insights not provided by traditional microbiological analysis. However, such approaches also result in large data sets which are slow to acquire and must be carefully analysed and interpreted. As a result of metagenomic analysis, this study detected extensive phage presence in most samples. Phages have been associated with the transduction of genes (including ARGs) which could play a role in the transfer of ARGs within the poultry microbiome, but whether this occurs within the poultry processing environment is unknown and requires future work to establish baseline data.

It is vital that future work focuses on elucidating hygiene/contamination processes/interventions in processing chains, and which have the greatest influence on the reduction of target microbes, phages, and ARGs, thus allowing for improved targeted interventions to better manage these microbial populations to benefit environmental and public health.

1. Introduction

For the interpretation of Antimicrobial Resistance (AMR) in this study, the WHO definition will be applied (WHO, 2018): “Antimicrobial resistance is resistance of a microorganism to an antimicrobial drug that was originally effective for treatment of infections caused by it.” AMR is a complex issue driven by a variety of interconnected factors enabling microorganisms to withstand the killing or static effects of antimicrobial agents, such as antibiotics, antifungals, antivirals, disinfectants, preservatives. Microorganisms may be inherently resistant to such agents or can change and adapt to overcome the effects of such agents. Microorganisms can acquire antimicrobial resistance genes (ARGs) through mutation or by obtaining foreign DNA from other microorganisms, with the widespread use of antimicrobial agents particularly driving the selection for AMR.

According to the British Poultry Council (Maxwell, 2020), the UK poultry meat sector has achieved a 76% reduction in total antimicrobial drug use and a 97.3% reduction in the use of critically important antimicrobials (CIA) over the past seven years (2012-2019). Additionally, UK Veterinary Antibiotic Resistance and Sales Surveillance 2022 Report (UK-VARSS, 2023) reported a 57% reduction in overall antimicrobial drug sales for animals and an 81% reduction in sales of high-priority CIAs for all animals in 2022 compared to 2014. Specifically, antimicrobial usage (AMU) in broilers and turkeys decreased by 71% and 84%, respectively, with a 99% reduction in high-priority CIAs for meat poultry in 2022 compared to 2014.

However, the reduction of AMU alone may not be sufficient to control AMR because the environmental persistence and spread of AMR bacteria and ARGs is a major contributory factor (Koutsoumanis et al., 2021). As highlighted in a recent EFSA biohazards panel report (Koutsoumanis et al., 2021), apart from prudent AMU, the most important measures to mitigate AMR applicable for all the food-production sectors investigated, both at pre- and post-harvest, involve the correct implementation of effective well known general management measures (good hygiene practices, biosecurity) to prevent/reduce occurrence and transmission of pathogens and other microorganisms. Identifying activities at food processing stages that cause or prevent the spread of AMR bacteria and ARGs in the different production sectors is an important priority for intervention. Since AMR bacteria are primarily associated with gut microbiota, reducing the likelihood of introduction, spread, and persistence of faecal contamination during meat processing is a high priority.

Contamination and control pathways for AMR bacteria in the chicken food chain include: metaphylactic administration of antimicrobial drugs to poultry, biosecurity in the growing sheds; cross-contamination in growing sheds, contamination or control via feed and water, cross-contamination during thinning, cross-contamination during transport to abattoir, control through washing of transport crates, cross-contamination during abattoir operations (slaughter, bleeding, scalding, defeathering, evisceration, cutting and portioning), potential control/reduction during abattoir operations (such as hard scalding, inside-outside wash, pre-chill interventions, chilling, freezing), packaging, control during chilled and frozen storage, distribution, retail display, consumer storage etc. (Bennani et al., 2020; Hedman et al., 2020). A recent EFSA biohazards panel report (2021), highlighted the role of contaminated process water and workers, through their hands or equipment, as sources of AMR bacteria. In 2020, the Food Standards Agency published a literature review on the impact of secondary processing of meat and meat products, but while that covered secondary processing of poultry (i.e. post primary chilling) it did not cover primary processing operations. The review concluded that there was a lack of studies examining whether persistently colonised processing environments can act as a contamination source for AMR bacteria as most of the literature concerns AMR in retail meats.

Campylobacter spp. are considered as priority pathogens due to their widespread antimicrobial resistance (Halaby et al., 2013). These bacteria are a common cause of foodborne illnesses and are frequently found in poultry and other food-producing animals (Havelaar et al., 2015). The increasing prevalence of antimicrobial resistant Campylobacter strains poses a significant threat to public health and food safety (Qin et al., 2023). Due to the importance of Campylobacter spp. in public health and its rising resistance to antimicrobials, particularly fluoroquinolones, Campylobacter spp. have been recognised as one of the serious antimicrobial-resistant threats of high priority by both WHO and the CDC [US Centers for Disease Control and Prevention] (Tacconelli et al., 2018).

Since E. coli is ubiquitous in the gastrointestinal tract of warm-blooded animals, it has been extensively used to monitor AMR in food animals (including chicken). Over the past few decades, AMR has increased at a faster rate among chicken isolates of E. coli than human clinical isolates (Tadesse et al., 2012). This trend is alarming as it highlights the role of poultry production in the dissemination of AMR (Bhattarai et al., 2024), which can be transmitted to humans through the food chain. E. coli strains in poultry are increasingly resistant to multiple classes of antimicrobial, including those critical for human medicine, such as fluoroquinolones and third-generation cephalosporins (Fenollar-Penadés et al., 2024).

The rise of MDR E. coli and antimicrobial resistant Campylobacter spp. in poultry presents a complex challenge for both animal and human health. This situation highlights the critical importance of comprehensive AMR surveillance and responsible antimicrobial use in agriculture. Also, while overall antimicrobial use has decreased (UK-VARSS, 2023), the persistence and, in some cases, increase of resistant bacteria emphasises the need for continued vigilance and action across the agricultural and public health sectors.

Antimicrobial resistance (AMR) in bacterial pathogens is intensified by the frequent role of mobile genetic elements (MGEs). The transfer of ARGs within and between different species of bacteria is commonly facilitated by MGEs, although the important role of individual plasmids and bacteriophages is not entirely understood.

The detection and identification of lytic or lysogenic bacteriophages (phages) from poultry environments at scale by metagenomic sequencing is an emerging area of interest. To what extent the diversity and epidemiology of phages in poultry processing and the natural environment is important however is yet to be determined. Phages can be present in large numbers especially from contaminating faecal matter and the transfer of ARGs by genes from bacteria to bacteria with specific phage can be demonstrated at least in the laboratory. In ‘real world’ microbial ecology with poultry, multiple compounding factors make it difficult to provide conclusive evidence that transfer of ARGs by phages occurs in this environment.

Although the FSA has been monitoring the prevalence and types of AMR bacteria in retail chicken, there is strong evidence that commensal and pathogenic AMR bacteria can be found on poultry meat. However, what is not known at present, is the role that chicken processing has on the prevalence and spread of AMR bacteria. It is also unclear whether persistently colonised processing environments can act as a contamination source for AMR bacteria, thus showing the importance of environmental monitoring and identification and survey of biofilms. There is a need to evaluate the impact of processing steps used during chicken production and the processing environment on the (molecular) epidemiology and transmission of AMR in chicken meat, to identify probable pathways for transmission, and mitigate the risks of transmission.

1.1. Aims and objectives of this project

The overall aim of this project was to assess the impact that processing has on the presence and transfer of antimicrobial resistant Campylobacter spp. and E. coli and associated ARGs on chicken meat.

The project was carried out in two parts with an initial literature review carried out to aid the project sampling plan/design (see Appendix). A detailed field and laboratory study was then conducted to assess the abundance and potential transmission of antimicrobial resistant Campylobacter spp. and E. coli during the processing of chicken in two similar large scale intensive UK slaughterhouses.

The specific objectives set for the “field” and laboratory study were to determine:

• The presence and abundance of Campylobacter spp. and E. coli cultured from samples taken at different stages along the chicken processing line at different sampling periods.

• The AMR profiles in Campylobacter spp. and E. coli isolates recovered from samples taken at different stages along the chicken processing line using Antimicrobial Susceptibility Testing (AST).

• The presence and types of ARGs present in selected individual isolates of E. coli and Campylobacter spp. using WGS.

• The diversity and abundance of bacteria, ARGs, and bacteriophages (naturally present viruses that target bacteria, commonly called phages) at different stages along the chicken processing line by modern genomic technology (metagenomic sequencing).

2. Materials and Methods

The project was carried out in two parts. Part 1: an initial literature review was carried out to aid the project sampling plan/design (detailed in the Appendix) and Part 2: a field and laboratory study carried out to assess the presence and potential transmission of antimicrobial resistant Campylobacter spp. and E. coli during the processing of chicken in two similar large scale intensive UK slaughterhouses and cutting plants.

2.1. Field and laboratory sampling and analysis

This section details the sample locations and collection procedures at two UK slaughterhouses and cutting plants over three distinct periods. Samples, including swabs, liquids, and neck skin, were collected from various stages of poultry processing. Using traditional microbiological methods, each sample was analysed for the presence, levels and AMR characteristics of Campylobacter spp. and E. coli isolates. WGS was then performed on selected individual isolates to evaluate the presence of ARGs. Additionally, pooled samples from each processing stage were selected and subjected to shotgun metagenomic analysis to enable the assessment of microbial diversity, bacteriophage diversity, and ARG distribution across the different processing stages.

2.1.1. Sampling location and sampling dates

Two slaughterhouses in the UK were selected as sampling locations for this study. One of the plants was located in the East Midland region and the other in the West Midland region of England, thus representing different geographical areas within the UK. Both of the sites were large scale chicken processing plants with similar processing lines but with differences in factory layout, some equipment, and distances between individual processing stages. Both supply major UK retailers with chicken and had similar throughputs and line speeds, though Site A was slightly larger than Site B in terms of throughput and ran at a faster line speed. Table 1 below shows details of the 2 sites visited for sample collection in this project. Each site was visited on 3 occasions in 2023. An initial in-field sampling occurred in March 2023 (16^th March for Site A and 14^th March 2023 for Site B; referred to as March in this report). Two further sampling visits occurred in the summer and winter months. For the summer collection, samples were collected in June 2023 (21^st June 2023 for Site A and 27^th June for Site B; referred to as June in this report) and October and November 2023 (18^th October for Site A and 21^st November for Site B; referred to October and November respectively in this report). At each site on each visit samples were collected in the morning (am) and afternoon (pm) at each different processing stages.

2.1.2. Description of chicken processing and sampling stages

At both sites, live birds were transported from the farm in crates, on arrival at the processing plants the birds were gas stunned rendering them unconscious. They were then removed from the crates by hand and suspended by their feet on a moving line, the birds were then euthanised and bled. The crates were cleaned and disinfected before being returned to the farms for use. After slaughter, birds entered a process where their feathers were removed. This began by putting the birds through a bath of hot water, which is designed to help loosen feathers, a processing stage called scalding. Feather removal (a processing stage called defeathering) was then performed by a machine called a “picker,” which includes hundreds of little rubber “fingers” that rotate around to remove the feathers. In Site A the heads were removed after defeathering, while in Site B the heads were removed prior to scalding. After the feathers were removed the feet were removed and the whole carcass inspected, and the carcasses then rehung on a separate moving line. The carcasses were then sent to an “eviscerating” (EV) line which removed the internal organs (viscera) in a series of operations. After the organs were removed, each carcass and the separated viscera were inspected (post-mortem (PM) inspection). After inspection excess neck skin was trimmed from the carcasses before the inside and outside of the carcass is washed in a machine called an “inside-outside (IO) washer”. In both of the plants the carcasses then passed through a thermal intervention process designed to reduce bacteria on the carcass prior to chilling (since different processors may use different interventions, or may use interventions in their plants, details of the exact intervention have been withheld in this report to ensure the anonymity of the plants). The same intervention was used in both of the plants. After the intervention the carcasses were chilled in continuous in-line air chillers to lower their temperature before being packed and dispatched. In Site B carcasses were also sampled at an accessible point between exit from the chiller and final dispatch, where an additional in process temperature check was made.

Due to access (for safety reasons not all processing stages were accessible) and scheduling issues (due to distances between sampling points, line speeds, and the time required to sample at different sampling points) samples could not be collected at the same intended sampling points in the plants on all of the visits. The sampling points where samples were collected at each site on each visit is shown in Table 2 and where, what type of sample, and at what time samples were collected on each visit shown in Table 3. These samples consisted of whole chicken carcass swabs (swabbing and sponging covered whole bird carcass surface area as best as possible), neck skin (excised and weight out for analysis), scalding water samples, crate swab (taken from an ~10cm x 10cm area) and equipment surface swabs (covering an ~30cm x 30cm area).

2.1.3. Sample collection and transportation

Based on discussions with the project team and the sequencing laboratory (CosmosID, an external laboratory) with which we collaborated on this project, we picked a swabbing/sponge method for sample collection, except for neck skin, which was excised at the factory and deposited in a sampling bag for analysis, and water samples (60 ml), which were directly placed in the sampling tube.

2.2. Isolation, enumeration, and confirmation of Campylobacter spp. and E. coli from samples

For Campylobacter spp., recovery and enumeration was carried out in accordance with ISO 10272-2:2017 (direct method) for the detection and enumeration of Campylobacter spp. in chicken meat (applied with a detection limit of 10 CFU per whole carcass, g, or cm³ of sample tested).

Briefly, to prepare the sample for testing, each sponge was rinsed with 10mL of buffered peptone water (Oxoid, Basingstoke, UK), and 100 μl of the resulting suspension was then subjected to threefold serial dilutions. From each dilution, 0.333ml was inoculated on modified charcoal cefoperazone deoxycholate agar (mCCDA). The plates were incubated for at least 44 hours at 41.5 ± 1°C under microaerobic conditions created with an Oxoid™ CampyGen™ sachet. The presumptive Campylobacter colonies were then further confirmed by inoculating five identical colonies on Columbia blood agar supplemented with defibrinated sheep blood.

For the enumeration, detection, and identification E. coli, 100 μl of each dilution was inoculated on Tryptone Bile X-glucuronide (TBX) agar consisting of chromogen 5-bromo-4-chloro-3-indolyl-beta-D-glucuronide (X-glucuronide). Typical E. coli colonies were counted, and their levels were enumerated with a detection limit of 100 CFU) per whole carcass, g, or cm³ of sample tested.

All presumptive bacterial isolates were further confirmed by a polymerase chain reaction (PCR) analysis. Thereafter, all confirmed bacteria colonies were tested against a broad spectrum of antimicrobial classes. All recovered and confirmed bacteria isolates were retained in storage at ambient temperature for use in further analysis (Antimicrobial Susceptibility Testing (AST) and Whole-Genome Sequencing [WGS]). Confirmed isolates were also placed in glycerol and stored at -80°C for long term storage.

2.3. Antimicrobial Susceptibility Testing (AST) of Campylobacter spp. and E. coli isolates

AST of confirmed Campylobacter spp. and E. coli isolates was carried out against a range of different antimicrobial agents (Table 4 and Table 5 respectively) to determine the resistance profile of the isolates.

Campylobacter spp. isolates were tested against ciprofloxacin, tetracycline, streptomycin, erythromycin, and nalidixic acid (Table 4). Minimal inhibitory concentration (MIC) values for ciprofloxacin (CIP), erythromycin (E), nalidixic acid (NA), streptomycin (S), and tetracycline (TE) were determined by the microdilution method. This panel was chosen in accordance with EUCAST guidelines and recommendations of the European Centre for Disease Prevention and Control (ECDC) (EUCAST, 2023).

All E. coli isolates were tested against amoxicillin + clavulanic acid, ciprofloxacin, ceftriaxone, ceftazidime, cefoxitin, trimethoprim/ sulfamethoxazole, cefotaxime, azithromycin, ampicillin, chloramphenicol, and tetracycline (Table 5) using a disk diffusion method, following guidelines from the Clinical and Laboratory Standards Institute (CLSI, 2024).

2.4. Whole-Genome Sequencing (WGS) of Campylobacter spp. and E. coli isolates

To predict genetic determinants of antibiotic resistance, WGS was conducted on 107 of the isolates that had been characterised using AST. In total 69 E. coli and 38 Campylobacter spp. isolates were selected and sent for WGS.

2.4.1. Campylobacter spp. and E. coli isolates selected along poultry processing stages chosen for further study.

To better understand the ARGs present in Campylobacter spp. and E. coli genomes, isolates that had shown the highest levels of phenotypic AMR (i.e. resistance to the largest number of antimicrobial agents) characterised using AST were selected for WGS. Details of at what stages Campylobacter spp. and E. coli isolates selected for WGS were collected from Site A and Site B are shown in Table 6 and Table 7, respectively.

2.4.2. Whole Genome Sequencing (WGS)

DNA extraction was conducted using the Promega Wizard Genomic DNA Extraction Kit (Madison, WI, USA), following the manufacturer’s protocol. Libraries for Illumina reads was prepared with the Illumina Nextera XT kit and libraries were assessed for quality with Qubit (Thermo Fisher, Waltham, MA) prior to WGS.

All the WGS was carried out by CosmosID, an external laboratory. Samples were sequenced on an Illumina NextSeq 550 platform (San Diego, CA, USA), producing paired-end reads at a maximum length of 150 bases. Raw sequence data trimming was performed for adapters and low-quality bases using bbduk and applying standard parameters (phred quality trimq = 22, and minimum length minlen = 36).

De novo assembly was performed in PATRIC v. p3-build-178 via Unicycler version 0.4.8 with minimum contig length cutoff set of 300bp. Quality assessment of assemblies was performed with QUAST version 5.0.2, SamTools version 13, and Pilon version 1.23. Closest reference genomes were identified by Mash/MinHash employing the PATRIC database. Upon submission to GenBank (Bioproject PRJNA674638), assemblies were reannotated using the NCBI Prokaryotic Genome Annotation Pipeline (PGAP) v. 6.2. The Genomic library was prepared using Illumina TruSeq Nano DNA library preparation kit supplied by Illumina Inc. Pair-end sequencing (2 x 150 base pairs) was performed using Illumina NovaSeq 6000. The quality of sequence reads was determined using FastQC Version 0.12.0. Adapter trimming, quality filtering, and per-read quality pruning was performed using fastp software. The sequence reads were merged using PEAR v0.9.6. The filtered paired-end reads were de novo assembled using SPAdes v3.15.3. Genome quality and completeness was evaluated using CheckM v1.0.18. while quality assessment of the assembled sequence was done using QUality ASsessment Tool (QUAST) v5.2.0. Staramr v0.10 was used to examine plasmids, virulence determinants, and ARGs.

2.4.3. Bioinformatic analysis of genomic data

The quality of sequence reads was determined using FastQC Version 0.12.0. Adapter trimming, quality filtering, and per-read quality pruning was performed using fastp software. The sequence reads were merged using PEAR v0.9.6. The filtered paired-end reads were de novo assembled using SPAdes v3.15.3. Genome quality and completeness was evaluated using CheckM v1.0.18. while quality assessment of the assembled sequence was done using QUality ASsessment Tool (QUAST) v5.2.0. The identity of the presumptive isolates was confirmed using the MASH algorithm, with average nucleotide identity of our genomes to reference strains ranging from 97% to 99.1%, identifying them as either C. jejuni or E. coli. Staramr v0.10 was used to examine plasmids, virulence determinants, ARGs, and the MLST profile. For E. coli, the Achtman MLST scheme was used.

2.5. Shotgun metagenomic sequencing

Shotgun metagenomic sequencing of samples collected across the processing chain on each visit to the two poultry processing sites was carried out to determine the diversity and abundance of bacteria, ARGs, and phages present in samples. In all 201 samples collected from both poultry sites in this project were sent for shotgun metagenomic sequencing.

Samples were processed as follows: The QIAGEN DNeasy PowerFood Pro Kit was used to extract genomic DNA from samples according to the manufacturer’s instructions. The extracted DNA was tested for quality and quantity using a Qubit 4 fluorometer and Thermofisher Scientific’s QubitTM dsDNA HS Assay Kit. DNA was then transported to CosmosID (a commercial external laboratory) for library preparation and sequencing. Briefly, the Nextera XT DNA Library Preparation Kit (Illumina) was used to produce genomic DNA libraries, which were then indexed using the IDT Unique Dual adapters with a total DNA input of 1ng. The genomic DNA was fragmented with a proportionate quantity of Illumina Nextera XT fragmentation enzyme prior to amplification. Following that, adapters were added to each sample, followed by 12 cycles of PCR amplification to create libraries. The DNA libraries were purified with Beckman Coulter Ampure magnetic beads and eluted in QIAGEN EB buffer. The Qubit 4 fluorometer and QubitTM dsDNA HS Assay Kit were used to confirm the quality and quantity of DNA libraries. Afterwards, the libraries were sequenced at 2x150bp on an Illumina NovaSeq 6000 platform.

2.5.1. Bioinformatics analysis of metagenomic data

Bioinformatic analysis was carried out on the CosmosID interface (CosmosID Metagenomics Cloud, CosmosID Inc.,). To quote the CosmosID method document, “The system utilizes a high-performance data-mining k-mer algorithm that rapidly disambiguates millions of short sequences reads into the discrete genomes engendering the particular sequences. The pipeline has two separable comparators: the first consists of a pre-computation phase for reference databases and the second is a per-sample computation. The input to the pre-computation phase are databases of reference genomes, virulence markers and antimicrobial resistance markers that are continuously curated by CosmosID scientists. The output of the pre-computational phase is a phylogeny tree of microbes, together with sets of variable length k-mer fingerprints (biomarkers) uniquely associated with distinct branches and leaves of the tree. The second per-sample computational phase searches the hundreds of millions of short sequence reads, or alternatively contigs from draft de novo assemblies, against the fingerprint sets. This query enables the sensitive yet highly precise detection and taxonomic classification of microbial NGS reads. The resulting statistics are analysed to return the fine-grain taxonomic and relative abundance estimates for the microbial NGS datasets. To exclude false positive identifications the results are filtered using a filtering threshold derived based on internal statistical scores that are determined by analysing a large number of diverse metagenomes. The same approach is applied to enable the sensitive and accurate detection of genetic markers for virulence and for resistance to antibiotics.”

3. Results

All results are presented in this section except for the systematic review. A detailed summary of the literature can be found in the Appendix of this report.

3.1. Field and laboratory sampling and analysis

This section presents the findings from our comprehensive field and laboratory investigations involving sample collection from 2 poultry processing environments and laboratory analysis using traditional and molecular biology techniques.

Site A and Site B were two chicken processing facilities with some notable similarities and differences. Both sites operate two processing lines and source their chickens from Red Tractor Assurance (RTA) farms in the UK. However, Site A has a higher processing capacity, handling up to 400,000 birds per day from Monday to Thursday and 200,000 birds on Friday and Saturday, with faster line speeds of about 217 birds per minute. In contrast, Site B processes 320,000 to 370,000 birds daily across two shifts, with slower line speeds of 175 birds per minute. A total of 17 farms supplied all the chickens processed in Site A during the collection period and a total of 16 farms supplied Site B during the collection period. The number of farms supplying each site varied slightly across different visits, ranging from 5 to 7 for both sites. Site A’s supplying farms were generally closer, with average distances ranging from 18 to 23 miles, while Site B’s farms were farther away, with average distances between 50 and 82 miles. Both sites showed some variation in their supplying farms across different visits, in addition in Site B there were variations in the farms that supply each processing line on each collection day. For Site A October collection all farms that supplied chicken varied completely from farms that supplied March and June samples.

For the laboratory sampling and analysis, we report on:

1. the prevalence and distribution of Campylobacter spp. and E. coli throughout various stages of poultry processing;

2. antimicrobial susceptibility profiles of isolates, including MIC data;

3. the resistance patterns observed in isolates across different processing stages;

4. WGS analysis of individual bacterial isolates, focusing on the identification of ARGs and plasmids,

   and finally,

5. shotgun metagenomic analysis to assess genetic diversity among isolates and detect ARGs within the broader microbial community.

3.1.1. Occurrence of Campylobacter spp. and E. coli along the processing chain

In this study a total 376 samples were collected along the processing line in 2 poultry plants on 3 visits to each. In Table 8 and 9, “No”: Indicates that the specific bacterium was not isolated from the sample; “Yes”: Indicates that the specific bacterium was successfully isolated from the sample and “Ns” (Not Sampled): Indicates that no sample was collected or analysed for the specific bacterium at this point or stage.

Of these 376 samples, 65.1% were positive for Campylobacter spp. (Table 8). Generally, Campylobacter spp. were detected at fewer processing stages in both plants as the carcasses progressed along the processing chain, although Campylobacter spp. were still detected in samples at late processing stages.

E. coli were ubiquitous in samples collected along the processing chain in both sites on all visits, with E. coli being detected in 95.6% of samples (Table 9).

3.1.2. Levels of Campylobacter spp. measured along the processing chain.

Numbers of Campylobacter spp. ranged from 1.00±0.0 to 3.9±0.5 Log₁₀ CFU per sample (Figure 1 to Figure 4). The highest numbers were measured in chicken litter. Samples collected at either site at some processing stages yielded no detectable Campylobacter spp.. Generally, Campylobacter spp. numbers decreased to non-detectable levels as sampling progressed along the processing chain, with fewer Campylobacter spp. detected in samples from later processing stages.

3.1.3. Levels of E. coli measured along the processing chain

Mean E. coli numbers ranged from 2.0±0.0 to 8.3±0.2 Log₁₀ CFU per sample (Figure 5 to Figure 8). As with Campylobacter spp., the highest number of E. coli were detected in chicken litter. Although E. coli numbers showed a general decrease along the processing chain, unlike Campylobacter spp., E. coli remained detectable in samples taken at later processing stages.

3.1.4. Antibiotic Susceptibility Testing (AST) of Campylobacter spp. and E. coli isolates

3.1.5. Campylobacter spp.

All isolates were confirmed to be C. jejuni. The resistance of isolates to each antimicrobial agent tested were variable according to site and collection month. A total of 137 Campylobacter spp. isolates were selected from the two sites and from different processing stage where possible. The summary interpretation of MIC for ciprofloxacin, erythromycin, nalidixic acid, streptomycin and tetracycline and individual MIC results for Campylobacter spp. isolates collected from different sites and different collection periods are shown in Notably, there are substantial differences in percentage resistance rates between sites and time periods for the same antimicrobial agent in some instances.

Table 10 and Figure 9. Also, overall antimicrobial resistance profile of all of the Campylobacter spp. isolates to the 5 antimicrobial agents tested are presented in Table 11.

From the Campylobacter spp. isolates collected from each site during each collection period, AST data reveal variations in percentage resistance among the different antimicrobials and sampling site. The highest percentage of resistance was recorded for tetracycline (72%) in isolates from Site B in November. Resistance to this antimicrobial agent demonstrated the highest variability, with only 32% of isolates from Site A in October being resistant to this agent. Isolates generally showed a lower resistance to streptomycin and nalidixic acid. No isolates from Site A collected in October showed any resistance to erythromycin or nalidixic acid in ( Notably, there are substantial differences in percentage resistance rates between sites and time periods for the same antimicrobial agent in some instances.

Table 10 and Figure 9). Ciprofloxacin resistance showed considerable variability, with resistance rates ranging from 9% (in isolates from Site A in October) to 60% (in isolates from Site B in June).

Overall, a higher percentage of resistance was recorded for tetracycline in all of the isolates. Of the total 137 Campylobacter spp. isolates examined, 53% (72/137) were resistant to tetracycline and 7% (9/13%) were resistant to erythromycin, the lowest resistance recorded overall (Table 11). Notably, there are substantial differences in percentage resistance rates between sites and time periods for the same antimicrobial agent in some instances.

MDR (defined as reduced susceptibility to at least three antimicrobial classes) was found in 7% of all of the Campylobacter spp. isolates examined (Table 12). No Campylobacter spp. isolate was resistant to all 5 antimicrobial agents and in total 38 (28%) were fully sensitive to all antimicrobial agents tested.

The following tables (Due to the lower frequency of Campylobacter spp. isolation and the lack of representative isolates for some processing stages, it is difficult to draw definitive conclusions from the results. However, the pattern of percentage resistance varied between sites and collection periods. As processing progressed, the percentage of resistance did not consistently change but fluctuated across different stages where Campylobacter spp. were isolated. Additionally, isolates from post-intervention stages did not consistently demonstrate any clear change in resistance. Though it should be noted that fewer Campylobacter spp. were isolated from later processing stages than processing stages earlier in the process at all sites, and no Campylobacter spp. isolates were isolated from samples taken at late stage processing stages on some visits, such as in Site A on both the June and October visits where no isolates post-neck skin cutting were detected.

Table 13 – Table 16) show the percentage of resistant and susceptible Campylobacter spp. isolates collected from Site A and Site B in June and October or November 2023 against the 5 antimicrobial agents tested. Each table represents the percentage resistant Campylobacter spp. isolates collected at different site and different collection period; Site A, June Collection (Due to the lower frequency of Campylobacter spp. isolation and the lack of representative isolates for some processing stages, it is difficult to draw definitive conclusions from the results. However, the pattern of percentage resistance varied between sites and collection periods. As processing progressed, the percentage of resistance did not consistently change but fluctuated across different stages where Campylobacter spp. were isolated. Additionally, isolates from post-intervention stages did not consistently demonstrate any clear change in resistance. Though it should be noted that fewer Campylobacter spp. were isolated from later processing stages than processing stages earlier in the process at all sites, and no Campylobacter spp. isolates were isolated from samples taken at late stage processing stages on some visits, such as in Site A on both the June and October visits where no isolates post-neck skin cutting were detected.

Table 13), Site A, October Collection (Table 14), Site B, June Collection (Table 15), Site B, November Collection (Table 16).

Due to the lower frequency of Campylobacter spp. isolation and the lack of representative isolates for some processing stages, it is difficult to draw definitive conclusions from the results. However, the pattern of percentage resistance varied between sites and collection periods. As processing progressed, the percentage of resistance did not consistently change but fluctuated across different stages where Campylobacter spp. were isolated. Additionally, isolates from post-intervention stages did not consistently demonstrate any clear change in resistance. Though it should be noted that fewer Campylobacter spp. were isolated from later processing stages than processing stages earlier in the process at all sites, and no Campylobacter spp. isolates were isolated from samples taken at late stage processing stages on some visits, such as in Site A on both the June and October visits where no isolates post-neck skin cutting were detected.

3.1.6. Escherichia coli

E. coli isolates were tested against 11 antimicrobial agents from 8 classes. These were: amoxicillin + clavulanic acid (a beta-lactam), ampicillin (a penicillin), azithromycin (a macrolide), cefotaxime (a cephalosporin), cefoxitin (a cephalosporin), ceftazidime (a cephalosporin), ceftriaxone (a cephalosporin), chloramphenicol (a phenicol), ciprofloxacin (a quinolone), trimethoprim/sulfamethoxazole (a cotrimoxazole), and tetracycline (a tetracycline).

A total of 469 E. coli isolates selected from the two sites and from different processing stages were tested for antimicrobial resistance. The antimicrobial susceptibility profile for E. coli isolates collected from individual sites and collection period are presented in Figure 10 and Figure 11. Also, overall antimicrobial susceptibility profiles for the 469 E. coli isolates are presented in Table 17: Overall resistant and susceptible E. coli isolates.

The proportion of resistance to the 11 antimicrobial agents in E. coli isolates showed varied between sites and between collection period (Figure 10 and Figure 11). The highest level of occurrence of resistance was shown to ampicillin in E. coli isolates from Site A in June (95% resistant) and from Site B in November (86% resistant) (Figure 10). The lowest occurrence of resistance in isolates were to chloramphenicol, ciprofloxacin, and ceftriaxone. The lowest occurrence of resistance was recorded to chloramphenicol in isolates from both sites and all collection periods, except in isolates from Site B in June where the lowest occurrence of resistance found was to azithromycin (2%) and chloramphenicol (10%).

For other antimicrobial drugs and comparing the 2 sites and collection periods (Figure 10 and Figure 11), percentage occurrence of resistance to amoxicillin/clavulanic acid in E. coli isolates at Site B decreased from March (33%) to June (25%), then dropped further in November (5%). Site A showed a different pattern with a higher percentage occurrence of resistance to amoxicillin/clavulanic acid isolates in June and October (both 49%), but lower occurrence of resistance in isolates in March (18%).

The pattern of resistance to azithromycin in E. coli isolates differed between sites. Isolates from Site B showed a lower percentage occurrence of resistance in June (2%) but a higher percentage occurrence of resistance in March (53%). While isolates from Site A showed a consistently higher occurrence of resistance, with a peak in June (45%) and a lower occurrence of resistance in March (37%) and October (21%).

Resistance to ceftazidime in isolates from Site A and Site B were similar for the 3 collection periods. The highest percentage occurrence of resistance was recorded in isolates collected in June at both Site A (55%) and Site B (60%), with lower occurrence in March (24% and 28%) and November (18% and 15%) in isolates from Sites A and B, respectively.

Percentage resistance to ceftriaxone in isolates from Site A ranged between 25% and 71%, with the lowest percentage of resistance in isolates collected in November and the highest in isolates collected in June. Site B showed a different pattern with no isolates collected in November being resistant to ceftriaxone, while 34% and 46% of isolates collected in March and June, respectively, were resistant.

Cefotaxime resistance patterns differed between sites. Isolates from Site B showed the higher percentage resistance in June (50%) with lower percentage resistance in March (36%) and November (1%). The highest percentage resistance (62%) was found in isolates collected from Site A in June. Compared to Site B, Site A still showed a higher resistance in isolates collected in October (26%) compared to that in isolates from Site B collected in November (1%).

More isolates collected from Site A were resistant to cefoxitin compared to isolates from Site B across all time periods. At Site A, resistance peaked in isolates collected in June (55%) and was lowest in isolates from March (12%). Isolates from Site B showed a gradual increase in resistance from March (25%) to June (21%), then a sharp decline in November (1%).

Tetracycline resistance showed a decreasing trend in isolates from Site B from March (64%) to November (39%). In contrast, Site A exhibited fluctuations, with the highest resistance in isolates collected in June (68%) and the lowest in March (50%). Overall, isolates from Site A tended to be more resistant than isolates from Site B, particularly in June and October or November.

The percentage of isolates that were resistant to trimethoprim/sulfamethoxazole collected at Site B decreased from March (72%) to November (25%). Isolates from Site A showed more extreme fluctuations, with a high percentage of resistant isolated collected in June (82%) but lower percentages of resistant isolates in March (46%) and October (42%). The June period stands out with high resistance in isolates collected at both sites.

Overall, antimicrobial susceptibility testing identified that 80% (376/469) and 55% (256/469) of E. coli isolates examined were resistant to ampicillin and tetracycline, respectively. The highest overall resistance recorded in this study (Table 17). Lowest resistance was recorded for chloramphenicol, 13% (62/469) (Table 17).

MDR (defined as reduced susceptibility to at least three antimicrobial classes) was found in 60% of E. coli isolates examined (Table 18), 3% (14/469) of the isolates were resistant to all 11 agents and a total of 25/469 (5%) were fully sensitive to all antimicrobials tested.

The following set of back-to-back histograms (Figure 12-Figure 22) show the percentage of resistant and susceptible E. coli isolates collected from Site A and Site B in March, June, and October or November 2023. These isolates were tested against 11 different antibiotics: amoxicillin + clavulanic acid (Figure 12), ampicillin (Figure 13), azithromycin (Figure 14), cefotaxime (Figure 15), cefoxitin (Figure 16), ceftazidime (Figure 17), ceftriaxone (Figure 18), chloramphenicol (Figure 19), ciprofloxacin (Figure 20), tetracycline (Figure 21), and trimethoprim/sulfamethoxazole (Figure 22). Each histogram pair represents the percentage resistant and susceptible E. coli isolates at various stages of the poultry processing chain, from crate litter to dispatch.

Overall, the pattern resistance in isolates varied between sites and collection periods. Isolates from different sites showed reduced resistance to specific antimicrobial agents, including cefotaxime (Figure 15), cefoxitin (Figure 16), ceftriaxone (Figure 18), chloramphenicol (Figure 19) and ciprofloxacin (Figure 20), particularly in isolates collected in November at Site B and in June at Site A for ceftazidime (Figure 17). As processing progressed, the percentage of resistant isolates fluctuated across different stages, independent of the processing sequence. Additionally, post-intervention stages did not consistently demonstrate any change in the percentage of resistant isolates.

3.1.7. Whole Genome Sequencing (WGS) of Campylobacter and E. coli isolates

To predict genetic determinants of AMR, WGS was conducted on 107 isolates, 38 Campylobacter spp. isolates and 69 E. coli isolates. The isolates represented those that showed resistance to the highest number of antimicrobial agents using AST at different sampling points across the processing chain. The quality of the assembled bacterial genomes was verified using software called QUAST. The genome length of the Campylobacter spp. isolates ranged from 1.7 to 1.8 million base pairs with 28 to 70 contigs, while the E. coli isolates ranged from 4.6 to 5.7 million base pairs with 6 to 847 contigs. The identity of the isolates was confirmed using the MASH algorithm, with average nucleotide identity of our genomes to reference strains ranging from 97% to 99.1%, identifying them as C. jejuni and E. coli.

Overall, there was greater variability in sequence type (ST) in E. coli isolates characterized from each sampling point at both sites and throughout the sampling period compared to C. jejuni isolates. A total of five and 40 different C. jejuni and E. coli STs were identified with 3 and 36 assigned to recognised STs, respectively. The ST profiles of one C. jejuni strain and three E. coli strains were unclassified at the time of drafting this report. However, these sequences will be submitted to PubMLST (for C. jejuni) and Enterobase (for E. coli) for sequence type allocation. Additionally, the sequencing output for one E. coli isolate was insufficient for ST or resistance genotyping.

The phenotypic susceptibility profiles, AMR-related genes, plasmids, point mutations detected and ST outputs for all sequenced C. jejuni isolates are presented in Table 19 to Table 22.

A total of four distinct C. jejuni STs were identified in both sample sites throughout the sampling period. These STs include ST 21, ST 262, ST 5136, and ST 6175. However, ST 6175 was the most prevalent being detected at various sampling points in both sample sites regardless of the sampling period. Each of the sequence types exhibited at least one of the following: the bla_OXA-61 gene (associated with ampicillin resistance) or the tetracycline resistance determinants tet(O) or tet(O/32/O). In addition, the ResFinder algorithm predicted that all the STs had the point mutation ACA -> ATA (T -> I) on the house keeping gene gyrA (T86I). A mutation predicted to confer resistance to ciprofloxacin and nalidixic acid.

Overall, the genotypic resistance determinant predicted by the ResFinder algorithm for ciprofloxacin, nalidixic acid and tetracycline matched the observed phenotypes. However, reduced susceptibility to erythromycin and streptomycin was observed in several isolates subject to AST but no corresponding resistance determinant gene was predicted for this observation.

The phenotypic susceptibility profiles, AMR-related genes, plasmids, point mutations detected and MLST outputs for all sequenced E. coli isolates are presented in Table 23 to Table 28.

Across all samples obtained from both sites at all timepoints, the following E. coli STs were observed: 10, 48, 57, 68, 69, 93, 101, 111, 117, 155, 162, 215, 224, 517, 522, 641, 770, 877, 1011, 1049, 1084, 1140, 1196, 1201, 1485, 1564, 1611, 1665, 1721, 1841, 2491, 3249, 6422, 6448, 9120, and 12034.

Four E. coli STs: ST 10, ST 155, ST 641, and ST 1611 were characterised from Site A in March. These STs were associated with 20 resistance-related genes, with bla_TEM-1C being the most prevalent. Additionally, 12 plasmids were detected, with Col156 being the dominant type (as shown in Table 23).

In June, seven STs were detected at Site A: ST 10, ST 68, ST 69, ST 117, ST 224, ST 1084 and ST 6448. These STs carried 30 resistance-related genes, with bla_TEM-1B as the prevailing gene. Also,17 plasmids were identified, with Inc1-l(Alpha) being the most common (as shown in Table 25).

In October there were nine STs at Site A (ST 10, ST 48, ST 101, ST 155, ST 1196, ST 1201, ST 1841, ST 2491, ST 6448) with 22 resistance-related genes, where tet(A) was the most prevalent and 9 plasmids, with P0111 being the most prevalent were found (Table 27).

In Site B, during March sampling period, 12 E. coli STs (ST 69, ST 155, ST 215, ST 517, ST 1140, ST 1485, ST 1564, ST 1665, ST 1721, ST 6422, and ST 12034) were detected, along with 17 resistance-related genes, with bla_TEM-1B being the most prevalent, and 16 plasmids, with IncFIB being the most prevalent (Table 24).

In June eight STs were identified at Site B: ST 10, ST 57, ST 93, ST 522, ST 1011, ST 1049, ST 1140, and an ST 17270. These STs carried 23 resistance-related genes, with sul(2) being the most abundant. Seventeen plasmids were observed, with P0111 as the prevailing type (as shown in Table 26).

In November 11 STs (ST 10, ST 69, ST 111, ST 162, ST 877, ST 1841, ST 3249, ST 6422, ST 6448, ST 9120 and an unidentified sequence type) were found at Site B, along with 22 resistance-related genes, with tet(A) being the most prevalent, and 9 plasmids, with P0111 being the most prevalent (Table 28).

In general, the genotype predictions for resistance to beta-lactams, cotrimoxazole, quinolones, and tetracyclines were consistent with the phenotypic profile, however, there were some inconsistencies with other drug classes. Specifically, the ResFinder tool failed to detect resistance genes in isolates that displayed phenotypic resistance to cephalosporins. A similar pattern was observed for macrolides in the majority of the sequence types. Conversely, resistance genes were detected in several isolates that were not phenotypically resistant.

3.1.8. Shotgun Metagenomic Sequencing

The terminology used to describe bacteria, ARG classes, ARGs, and phage in this report is that of the Sequence and analysis provider CosmosID, the external laboratory that carried out the shotgun metagenomic analysis. In the terminology of Cosmos an ARG “Branch” is a designation for a higher-level node which represents shared homology among a group of specific gene sequences. This indicates that the sequences generated for that samples were not able to resolve to a specific ARG allele.

It should be noted that metagenomic sequencing detects DNA from both dead and alive organisms and further testing is required to establish whether any of the organisms, genes, and/or phage detected are viable and contributing to the microbial ecology or risk (Feye et al., 2020). In addition, this data is not quantitative, relative abundance describes the contribution/proportion of a given taxon/feature to the total microbial community detected within a sample (CosmosID, 2023; Feye et al., 2020).

The results presented are “filtered” sequence data supplied by CosmosID. The Cosmos filtering process is a proprietary stage of data processing, where a framework is applied that has been developed using hundreds of samples of known composition, in order to balance sensitivity and specificity. This framework is then used to predict which assignments are highly likely to be correct (Filtered) vs those which are possibly present but may be inaccurate or spurious (Total).

3.1.8.1. Bacteria

In total, 14 and 17 phyla were detected in Sites A and B, respectively. Proteobacteria, Firmicutes, Actinobacteria, and Bacteroidetes accounted for the majority of detected phyla in all samples. A total of 1407 different bacterial species (including classifiable and unclassifiable species [i.e., where the genus but not the specific species could be identified]) were detected using shotgun metagenomic analysis in all of the 201 samples collected from the poultry sites in this project. There was a greater diversity of bacterial species (including classifiable and unclassifiable species) detected over the sampling period at Site B compared to Site A, 1220 vs 940, respectively. Of these species, 439 and 510 occurred in Sites A and B on all sampling periods, respectively, with 334 different species being detected in samples taken from both sites on all occasions.

The relative abundance of the dominant bacterial species (including classifiable and unclassifiable species) detected in samples collected on each of the visits to Site A are presented in Figure 23, Figure 24, and Figure 25, and from Site B in Figure 26, Figure 27, and Figure 28, respectively. The overall most abundant classifiable bacterial species detected at Site A were: Bifidobacterium pullorum, Acinetobacter johnsonii, and Acinetobacter lwoffii (in order). E. coli was the seventh most abundant classifiable bacterial species overall. A. johnsonii, A. lwoffii and Acinetobacter gandensis (in order) were the overall most abundant classifiable bacterial species at Site B. E. coli was the eighth most abundant classifiable bacterial species overall. The most abundant species at Site A, B. pullorum, was the sixth most abundant at Site B.

A summary of the number of different species detected on the different sampling periods is shown in Table 29. Although further sampling is required to statistically establish seasonal trends, there did appear to be a seasonal trend in both sites, with a lower diversity (number) of different bacterial species being present in samples taken in October/November, than in March. With the highest diversity measured in samples in both sites collected in June. While a large range of different species of bacteria were detected, many of these bacteria species were only detected in samples collected at only one sampling point along the processing chain.

Generally, the diversity of bacterial species detected in the samples collected from the sites appeared to reduce through the processing chain in each site on all of the sampling occasions (Figure 29 and Table 30). A greater number of different bacterial species were detected in samples taken before evisceration than in samples taken after evisceration in both sites on all occasions. The number of bacterial species detected at both sites in samples taken after the thermal intervention process stage (which is specifically designed to reduce microbial contamination) used at both sites was particularly low but began dropping after defeathering. However, samples taken in June in Site B did not show as big a drop in bacterial diversity during processing (Figure 29Figure ) and there were more bacterial species present in samples taken after the pre-chilling thermal intervention process stage in samples collected in June than detected in samples collected from this Site (Site B) in March or November. It is not clear why this was the case and further sampling is required to establish whether there is a seasonal pattern or whether this was a one-off occurrence due to something that specifically occurred during processing in this plant on the day in June that samples were collected.

Only 51 of the 1407 bacteria species (including classifiable and unclassifiable species) were detected at more than 10 sampling points along the processing chain at each site in each month (March, June, and October or November 2023) and were also present in samples taken at post-intervention stages in the sites on at least one occasion (such cells are highlighted in Table 31 below). Of these species A. johnsonii, B. pullorum, and E. coli were the most commonly occurring classifiable bacterial species in samples taken at both sites on all occasions.

3.1.8.2. Antimicrobial Resistance Genes (ARGs)

Shotgun metagenomic analysis revealed the occurrence of a total of 442 different ARGs which may be involved in resistance to 16 different classes of antimicrobial (hereafter referred to as ARG classes) in all the 201 samples collected along the processing chain in the poultry sites sampled. The detected classes were: Tetracyclines, Aminoglycoside, Macrolide-Lincosamine-Streptogramin B, Class D beta-lactamases, Sulfonamides-Trimethoprim, Polymixin-Colistin, Sulfa drugs, Class C beta-lactamases, Class A beta-lactamases, Phenicol, Quinolones and fluoroquinolones, Glycopeptides, Fosfomycin, 5-Nitroimidazole, Class B beta-lactamases, and Rifampicin.

The relative abundance of the ARG classes detected in samples collected on each of the visits at Site A are presented in Figure 30, Figure 31, and Figure 32; and at Site B in Figure 33, Figure 34, and Figure 35 respectively. The overall most abundant ARG classes detected at Site A were: Tetracyclines, Macrolide-Lincosamine-Streptogramin B, Aminoglycosides, and Class D beta-lactamases (in order). These were also the most abundant classes detected at Site B, but differed in order, with Class D beta-lactamases being the second most abundant and Macrolide-Lincosamine-Streptogramin B being the fourth.

A summary of the number of different ARG classes detected on the different sampling periods is shown in Table 32. As with the bacterial species, generally the diversity of ARG classes detected in the samples collected from the sites appeared to reduce through the processing chain in each site on all of the sampling occasions, particularly post evisceration (Figure 36 and Table 33). The exception was Site B in June, where a high diversity of different ARG classes were still detected at post evisceration sampling points.

Overall, genes in classes for Tetracyclines, Aminoglycoside, Macrolide-Lincosamine-Streptogramin B, Class D beta-lactamases, Sulfonamides-Trimethoprim, Polimixin-Colistin, Sulfa drugs, Class C beta-lactamases, Class A beta-lactamases, and Phenicol were the most detected in the samples taken throughout processing at both of the sites on all visits. Genes in these classes were detected in samples taken at more than 10 different sampling points along the processing chain in both site A and B on each of sampling periods (Table 34). In contrast, genes in classes for Fosfomycin, 5-Nitroimidazole, Class B beta-lactamases, and Rifampicin were detected in very few samples taken from these sites. For example, Rifampicin class genes were only detected in one sample (viscera sampled at post-mortem inspection) from Site B in November.

The relative abundance of the ARGs detected in samples collected on visits in March, June, and October to Site A are presented in Figure 37, Figure 38, and Figure 39 respectively; and in visits in March, June, and November to Site B in Figure 40, Figure 41, and Figure 42. The overall most abundant ARGs detected at both Site A and B were: Tetracycline tet(39) 1 KT346360, Tetracycline 41 2477 Branch, and Macrolide lnu(A) 1 M14039 (in order).

Of the 442 different ARGs detected in total, there was a slightly greater diversity of ARGs detected over the sampling period at Site B compared to Site A, 354 vs 343, respectively. Of these ARGs, 140 occurred in Site A on all visits and 144 occurred in Site B on all visits, with 103 different ARGs being detected in samples taken from both sites on all visits (Table 35).

A summary of the number of different ARGs detected at the different sampling periods is shown in Figure 39. Although further sampling is required to statistically establish seasonal trends, as with bacterial diversity there did appear to be a seasonal trend in both sites, with a lower diversity (number) of different ARGs being present in samples in both site A and B taken in October/November and March in comparison to those in June 2023.

As with the bacterial species, generally the diversity of ARGs detected in the samples collected from the sites appeared to reduce through the processing chain in each site on all of the sampling occasions, particularly post evisceration (Figure 43 and Table 36). The exception was Site B in June, where a high diversity of different ARGs were still detected at post evisceration sampling points.

While a total of 442 different ARGs were detected in all of the samples, only 14 different ARGs were detected at more than 10 sampling points along the processing chain at each site on each month and were also present in samples taken at post-intervention stages in the sites on at least one occasion (Table 37). The most commonly occurring ARGs were for Tetracycline.

3.1.8.3. Bacteriophages (phages)

Shotgun metagenomic analysis revealed the occurrence of a total of 758 different phages in all the 201 samples collected along the processing chain in the poultry sites sampled in this project. The number of different phages detected over the sampling period at both sites was similar, 644 and 648 at Sites A and B, respectably, but differed in the type of phages present. 291 different types of phage occurred in Site A on all occasions and 326 different types of phage occurred in Site B on all occasions, with 250 different phages being detected in samples taken from both sites on all occasions.

The relative abundance of the phages detected in samples collected at different processing stages at Site A on visits in March, June, and October are presented in Figure 44, Figure 45, and Figure 46, respectively; and on visits in March, June, and November at Site B in Figure 47, Figure 48, and Figure 49, respectively. The overall most abundant different phages detected at Site A were: Escherichia phage P1, Escherichia phage P694, and Erwinia phage FE44. Whereas the overall most abundant different phages detected at Site B were: Escherichia phage P1, Acinetobacter phage B1251, and Escherichia phage P694.

A summary of the number of different phages detected at the different sampling periods is shown in Table 38.

As with the bacteria and ARGs, generally the diversity of phages detected in the samples collected from the sites appeared to reduce through the processing chain in each site on all of the sampling occasions, particularly post evisceration (Figure 50 and Table 39). The exception was Site B in June, where a high diversity of different phages were still detected at post evisceration sampling points.

While a total of 758 different phages were detected in all of the samples, only 54 different phages were detected at more than 10 sampling points along the processing chain at each site on each month and were also present in samples taken at post-intervention stages in the sites on at least one occasion (Table 40). Of these 13 different phages were detected at more than 10 sampling points, including post intervention, at each site on all occasions. Of these phages, Escherichia phage P1, Salmonella phage SJ46, and Enterobacteria phage mEp460 were detected at the most sampling points. Escherichia phage P1 and Salmonella phage SJ46 were detected in approximately 80% or more of all the samples taken across the processing chain at both sites on all occasions.

Escherichia phage were ubiquitous at both sites, but only one Campylobacter phage, CPt10, was detected. The Campylobacter phage was detected at 2 sampling points in samples taken in November at Site B. Both were feather samples taken at the post-stun (pm) and post-defeathering (pm) stages. The Campylobacter phage CPt10 is known to lyse with C. jejuni and C. coli and has been isolated from UK poultry processing environments (Salama et al., 1989; Timms et al., 2010). Both C. jejuni and C. coli were also detected in these samples.

4. Discussion

4.1. Literature review

Overall, the literature review (see Appendix) found that few studies have focussed on the transmission of AMR bacteria or genes during poultry processing. Studies that have looked at the relationship between AMR on incoming birds and outgoing meat appear to show that AMR and ARGs are related to incoming birds so related to farm drivers rather than contamination occurring from equipment within plant. There is little evidence in the literature on the sequential transmission of AMR (bacteria or genes) during the chicken slaughter process or persistence or transmission of antimicrobial resistant bacteria or ARGs in the processing environment. Few studies have considered the persistence of ARGs in the processing environment and whether they could be a vector or contribute to transmission. But there is some limited evidence of ARGs persisting in processing environments (on equipment and in scald water) that may be a source of cross-contamination during processing. There is also some evidence on role of flock-to-flock contamination during processing on AMR. While many studies (not reviewed here) have determined AMR prevalence in retail poultry, few studies have considered cutting operations and what role they may have on transmission.

Plants outside of Europe often use pre-chill chemical intervention steps not permitted or used in the UK or Europe. No studies appear to have looked at physical interventions, such as steam, on AMR or ARGs that have been implemented and are being used in UK plants, including the plants sampled in this study. In addition, plants outside of Europe also often use immersion (water) chilling (usually in chlorinated water) rather than air chilling which is used in UK plants. Finally, most studies outside of Europe have focussed on Salmonella spp. contamination rather than Campylobacter spp. or commensal gut bacteria that may harbour ARGs.

These insights were used to inform the sampling plan used in this project to ensure that some of these knowledge gaps were covered within this project.

4.2. Field and laboratory sampling and analysis

4.2.1. Sampling sites and poultry supply

The farms supplying chickens to each site varied across the collection periods, in addition this variation was also observed between processing lines at Site B, potentially contributing to differences in AMR profiles observed in this study. The farms that supplied Site A in October were different from those that supplied chickens in March and June, possibly explaining seasonal variations in bacterial loads and AMR patterns. At Site B, while 5-7 farms supplied each processing line during each collection period, one farm consistently supplied chickens across all three collection periods. This consistent supplier might have acted as a stable source of specific bacterial strains or AMR profiles, providing a baseline for comparison with other fluctuating sources. The diversity in farm sources likely contributed to the overall variability in microbial populations and their resistance characteristics observed throughout the study. Other studies have also reported that farm location and management practices significantly influence the prevalence and AMR patterns of poultry bacteria (Mudenda et al., 2023; Sibanda et al., 2018; Taylor et al., 2016). This highlights the importance of considering farm-specific factors when developing strategies to mitigate antimicrobial resistance in poultry production systems.

4.2.2. Presence and levels of Campylobacter and E. coli during chicken processing

The prevalence of Campylobacter spp. and E. coli was varied with ~65% samples positive for Campylobacter spp. and ~96% samples testing positive for E. coli. The analysis of Campylobacter spp. contamination in poultry processing reveals substantial variations across different processing stages, between sites, and across seasons. Crates with litter consistently showed the highest contamination levels, ranging from 2.1 to 4.6 Log₁₀ CFU, compared to clean crates (0.7 to 4.0 Log₁₀ CFU) and dirty crates (0.7 to 3.6 Log₁₀ CFU). Viscera handling and neck skin trim also exhibited consistently higher contamination, with viscera contamination reaching up to 4.0 Log₁₀ CFU at post-evisceration stages. Site B generally demonstrated higher counts than Site A across most processing stages. For instance, in carcass processing, Site B showed contamination levels up to 3.8 Log₁₀ CFU at post-mortem inspection in June, compared to Site A’s maximum of 3.9 Log₁₀ CFU in the same period. Notably, June datasets showed higher contamination levels compared to October or November. This seasonal effect was particularly evident in carcass contamination, where Site A’s post-stun contamination reached 2.9-4.0 Log₁₀ CFU in June but only 0.7-2.6 Log₁₀CFU in October. Similarly, Site B’s post-chill contamination ranged from 0.7 to 3.3 Log₁₀CFU in June, dropping to 0.7-1.7 Log₁₀CFU in November. The prevalence and level of E. coli in this study varied processing stages, sites, and seasons. Crates with litter consistently showed the highest contamination levels similar to the Campylobacter spp. results, with counts ranging from 4.7 to 8.4 Log CFU, compared to clean crates (1.7 to 6.1 Log CFU) and dirty crates (4.1 to 8.4 Log CFU). Viscera handling also demonstrated high contamination levels, reaching up to 8.4 Log CFU at post-evisceration stages. Also, Site B generally showed higher counts than Site A across most processing stages, particularly in June. For instance, in carcass processing, Site B showed contamination levels up to 8.3 Log CFU at post-mortem inspection in June, compared to Site A’s maximum of 8.3 Log CFU in the same period. Notably, June data showed higher contamination levels compared to October or November, indicating a strong seasonal effect. This was particularly evident in carcass contamination, where Site A’s post-stun contamination reached 8.0-8.3 Log CFU in June but only 4.6-4.9 Log CFU in October.

Furthermore, certain interventions showed varying effectiveness between sites and time periods. For example, post-intervention carcass contamination at Site B ranged from 3.0 to 3.3 Log₁₀ CFU in June but consistently measured 0.7 Log₁₀ CFU in November, suggesting potential improvements in intervention efficacy or changes in initial contamination levels. The inside-outside wash stage showed high variability, with Site B ranging from 1.9 to 3.3 Log₁₀ CFU in June and 0.7 to 2.9 Log₁₀ CFU in November, while Site A ranged from 1.3 to 3.5 Log₁₀ CFU in June and consistently measured 0.7 Log₁₀ CFU in October. Scalding water contamination varied from 1.9 to 3.4 Log₁₀ CFU at Site B and 0.7 to 2.8 Log₁₀ CFU at Site A, highlighting the potential for cross-contamination at this early processing stage. These findings emphasise the need for tailored, site-specific approaches to effectively reduce Campylobacter spp. contamination. For E. coli, the effectiveness of interventions varied between sites and time periods. Post-intervention carcass contamination at Site B ranged from 4.0 to 5.2 Log CFU in June but decreased slightly to 4.1 to 4.5 Log CFU in November. Also, the inside-outside wash stage showed high variability, with Site B ranging from 2.8 to 4.1 Log CFU in November, while Site A ranged from 4.3 to 5.0 Log CFU in October. Scalding water contamination varied from 4.1 to 4.5 Log CFU at Site B and 3.0 to 8.3 Log CFU at Site A. The variability in intervention effectiveness suggests that regular monitoring and adjustment of process controls may be necessary to maintain consistent low levels of contamination throughout the year and across different processing facilities.

The finding of our study aligns with existing literature, which notes that while processing stages like scalding and chilling typically reduce pathogen levels for Campylobacter spp. and E. coli, steps like defeathering can negate these reductions, highlighting the importance of validating critical control points within each facility to optimise pathogen reduction strategies (Duffy et al., 2014; Hutchison et al., 2022; Pacholewicz, Swart, et al., 2015). In a study conducted by Duffy et al. (2014), it was observed that scalding and chilling stages typically reduce both Campylobacter spp.and E. coli levels, with reported reductions of 1.8-2.9 log₁₀ CFU/carcass for Campylobacter spp. and 1.3-2.5 log₁₀ CFU/carcass for E. coli. In addition, other similar studies have also observed significant reductions in Campylobacter spp. levels after scalding and chilling steps in both broiler and turkey slaughterhouses (Beterams et al., 2024; Hutchison et al., 2022; Pacholewicz, Swart, et al., 2015). E. coli concentrations in some studies generally followed similar patterns to Campylobacter spp., suggesting its potential as an indicator organism (Duffy et al., 2014; Pacholewicz, Swart, et al., 2015). The consistent differences between sites and seasons suggest that processors should consider both location-specific factors (such as equipment design, processing methods, or local environmental conditions) and seasonal variations when developing and implementing food safety measures.

4.2.3. Phenotypic profiles and genotype predictions of Campylobacter spp. and E. coli isolates

Overall, this study found higher resistance to tetracycline [53%] and ciprofloxacin [26%] in Campylobacter spp. Isolates compared to streptomycin [15%], nalidixic acid [12%] and erythromycin [7%]. This findings are generally in agreement with the UK-VARSS report (UK-VARSS, 2022) that reported very high (66%) resistance to tetracycline and 59% resistance to the ciprofloxacin in C. jejuni in broilers. UK-VARSS, 2022 also reported a lower resistance to erythromycin (2.8%).

The phenotypic susceptibility profile of E. coli varied across different processing stages along the production chain, with no clear correlation between the processing stages and the predicted genetic determinants of antimicrobial resistance. Our results also indicate that at least 46.5% of E. coli showed the lowest levels of susceptibility to ampicillin, tetracycline, ceftazidime, and trimethoprim/sulfamethoxazole.

The majority of AMR surveys on the occurrence of C. jejuni and E. coli in poultry have focused on specific stages across the value chain with many looking at retail poultry. Similar to our observations, Fearnley and Rodgers (2023) who surveyed AMR in E. coli (n=35) on raw fresh chicken meat in UK retail stores in 2022, found that 75% to 100% of the E. coli isolates showed reduced susceptibility to ampicillin, 75% to 90% were less susceptible to tetracycline, and all isolates were less susceptible to cefotaxime and ceftazidime. Fearnley and Rodgers (2023) tested trimethoprim and sulfamethoxazole separately, reporting reduced susceptibility in 61% to 67% of isolates for trimethoprim and 75% to 88% for sulfamethoxazole. It is noteworthy to mention that the study by Fearnley and Rodgers (2023) also considered chicken meat imported from outside the UK. Also, Willis et al. (2022) reported 47.2% of a total of 17 E. coli isolated from battered poultry products on retail sale in the United Kingdom were less susceptible to tetracycline, and all isolates were less susceptible to ampicillin, cefotaxime and ceftazidime.

We observed reduced susceptibility among 11.8 - 32.1% of C. jejuni to streptomycin, ciprofloxacin, nalidixic acid, erythromycin, and tetracycline. In both sampling sites, the phenotypic susceptibility pattern remained consistent between the C. jejuni isolates regardless of the stage of processing from which they were recovered. Fearnley and Rodgers (2023) reported that ciprofloxacin and tetracycline had the most common reduced susceptibility in C. jejuni isolates from raw fresh chicken meat, although they did not test for streptomycin.

Generally, the genotype predictions for AMR to beta-lactams, cotrimoxazole, quinolones, and tetracyclines were consistent with the phenotypic profiles. However, there were some inconsistencies with other drug classes. Specifically, the ResFinder tool failed to detect resistance genes in isolates that displayed phenotypic resistance to cephalosporins, and a similar pattern was observed for macrolides in most sequence types. Conversely, ARGs were detected in several isolates that were not phenotypically resistant. This shows the value of additional work to enhance the prediction of phenotypic AMR from genotype, for example as occurring for Campylobacter spp. and other organisms in the PATH-SAFE Programme.

4.2.4. Genomic diversity of Campylobacter spp. and E. coli

All Campylobacter spp. isolates were identified as C. jejuni, with no other Campylobacter spp. being identified by WGS. The presence of C. jejuni and E. coli STs identified in this study, whether they contained ARGs or were free of them, have been documented over the years in other poultry related studies, except for the novel ST. We cannot definitively state that the C. jejuni and E. coli sequence types observed in this study are directly associated with the specific processing stages they were recovered from. This is because only one of six isolates per stage was selected for genomic sequencing based on the phenotypic resistance profile.

For C. jejuni, STs 21, 262, 5136 and 6175 were detected. ST21, ST262 and ST6175 are all members of the ST21 clonal complex (CC) of closely related sequence types. The ST21-CC is the most common CC in the pubMLST database for the UK (updated 13/08/2024, accessed 15/08/2024), representing 19.8% of accessions to the database. It is the second most common ST isolated from poultry in the UK, representing 11.3% of accessions. ST5136 is part of the ST464 CC, the eighth most common UK CC in pubMLST, both across all sample types and from chicken specifically (3.2% and 3.8% of accessions respectively). Therefore, all identified CCs are consistent with CCs previously identified in UK poultry. ST 6175 was the most prevalent ST, being detected at various sampling points in both sample sites regardless of the sampling period. According to Mughini-Gras et al. (2021) ST 6175 is a poorly documented ST in the literature, but they reported it to be highly prevalent in poultry in an analysis they carried out of C. jejuni/coli isolates from the Netherlands in 2017–2019 from a variety of sources. They also found it to be predominant in isolates from humans. Our study also identified ST 21 as being present, which has been reported to be particularly prevalent in poultry (Mughini-Gras et al., 2021).

Considerably more E. coli STs were isolated than C. jejuni STs, with many STs only found in this study at a maximum of one timepoint at one site (29 out of 36 identified STs). No STs were observed at all six time points, although ST10 was observed at five out of six time points. No clear site or seasonal patterns in ST distribution were seen, except that ST 1841 was observed at both sites sampled at the October/November timepoint, but not at any other time.

Among the E. coli strains, the prevalence of ST10, one of the most persistent ST regardless of sampling site and sampling period, has been observed particularly among broiler chickens in various countries e.g., China, Ghana, Poland, Romania, UK. This ST is known to be a common sequence type found in poultry, often associated with ESBL production (Lim et al., 2015; Liu et al., 2022; Oteo et al., 2009; Randall et al., 2011). Recent studies conducted in UK, have also reported ST 93, 117, 155, 224, 1011, 6446 among others in poultry products (Fearnley & Rodgers, 2023). These similarities could suggest that these STs are persistent in the poultry setting in UK.

4.2.5. Shotgun Metagenomic Sequencing

Shotgun metagenomic sequencing was performed on 201 samples collected across the processing chain of two poultry processing sites in March, June, and October/November 2023. Overall, this showed a great diversity of bacterial species, ARGs, and phages to be present in both plants and that the diversity in numbers of different bacterial species, ARGs, and phages generally decreased along the processing chain indicating that contamination was progressively reduced on the chicken carcasses as they were processed. Reductions in the diversity of the microbiome through the poultry processing chain have been observed in other studies (Handley et al., 2018; Marmion et al., 2023; Oakley et al., 2013; Rothrock et al., 2016). The association of ARG diversity with taxonomic diversity, suggests that ARGs may be reduced with hygiene/contamination reduction. Further studies are required to confirm whether similar patterns occur in other UK poultry processing plants.

4.2.5.1. Bacterial diversity

The shotgun metagenomic sequencing showed a great diversity of bacterial species in the 201 samples collected along the processing chain in the two poultry sites sampled. The most abundant phyla were Proteobacteria, Firmicutes, and Actinobacteria, which accounted for the majority of detected phyla in all samples. These same phyla have been reported to be the most abundant phyla in poultry in other studies (though not always in the same order) in the US (Handley et al., 2018; Rothrock et al., 2016; Wages et al., 2019), Italy (De Cesare et al., 2018), Australia (S. H. Chen et al., 2020), and Ireland (Marmion et al., 2023).

A total of 1407 different bacterial species (including classifiable and classifiable species) were detected, but only 334 of these species were detected in samples taken from both plants on all occasions. Acinetobacter spp. were the most abundant species at both sites, particularly A. johnsonii, lwoffii, and gandensis, and were detected in samples from sampling points along the whole process chain (from entry to dispatch) in both sites. It is not known whether any of these Acinetobacter spp. were antimicrobial resistant. Infections caused by antimicrobial resistant Acinetobacter spp. are a known clinical risk (Hamouda et al., 2008). Foods of animal origin have been postulated as a vector for human infection with antimicrobial resistant strains of Acinetobacter (Carvalheira et al., 2017), but studies are limited and there is some evidence that human and animal isolates may differ (Hamouda et al., 2008). Acinetobacter spp. have been identified as relatively abundant in poultry in studies in Portugal (Carvalheira et al., 2017), Italy (De Cesare et al., 2018), the US (Hinton et al., 2004; Li et al., 2020; Oakley et al., 2013).

To our knowledge, few studies appear to have reported on the prevalence of B. pullorum in poultry, it has been reported to be prevalent in the cecum of laying hens surveyed in Ireland (Corrigan et al., 2023).

Other studies of poultry processing have reported Pseudomonas spp. to be prevalent through the processing chain (Handley et al., 2018; Hinton et al., 2004; Oakley et al., 2013; Wages et al., 2019). While present, Pseudomonas spp. were not that abundant or prevalent in samples from the two sites we studied. US studies by Handley et al. (2018) and Wages et al. (2019) also identified Chryseobacterium spp. as a potential indicator organism in poultry processing as they isolated them from 3 processing plants at multiple sampling locations. Our study did not detect this species as being very abundant or prevalent in the two UK sites we studied.

4.2.5.2. ARG diversity

The most abundant and commonly occurring class of ARGs in both plant A and B were tetracycline resistant genes. Since tetracycline is widely used in animal production this is not surprising. The overall most abundant ARGs detected from samples at both sites were: Tetracycline GENE tet(39) 1 KT346360, Tetracycline 41 2477 Branch, and Macrolide GENE lnu(A) 1 M14039.

Tet(39) is a tetracycline efflux pump found in Gram-negative bacteria, including Brevundimonas, Stenotrophomonas, Enterobacter, Alcaligenes, Acinetobacter, and Providencia spp. (The Comprehensive Antibiotic Resistance Database). Since Acinetobacter spp. were prevalent in the samples it is likely that they are a source for this gene, particularly A. johnsonii which has been reported to host this gene (The Comprehensive Antibiotic Resistance Database), but further studies are needed to confirm this.

Tetracycline 41 2477 Branch is a higher-level node/branch that represents individual AMR gene allele sequences or leaves. Tetracycline 41 2477 Branch contains Tetracycline GENE tet(W) 6 FN396364 and Tetracyclines DOXYCYCLINE MINOCYCLINE HCL GENE tetW, thus is an AMR allele homologous to the tet(W) gene (Personal communication from Cosmos). The gene encodes a ribosomal protection protein and known to provide tetracycline resistance to a wide range of anaerobic intestinal bacteria [The Comprehensive Antibiotic Resistance Database; Ammor et al. (2008)]. This gene was both abundant at both sites and detected in samples from many different sampling points. Further studies are required to confirm the host and any role it may have in foodborne AMR.

The lnu(A) gene (The Comprehensive Antibiotic Resistance Database), coding for macrolide and lincomycin resistance, has been described in several gram-negative and positive bacteria species of poultry origin, such as Clostridium spp., Campylobacter spp., Salmonella spp., and lactobacilli (Cauwerts et al., 2006; Farooq et al., 2022). Lactobacilli (particularly L. gallinarum and L. johnsonii) were abundant and found in samples taken along the process chain in both plants in our study. Further studies are required to confirm the sources of the genes detected in this study and to confirm whether similar patterns occur in other UK poultry processing plants.

4.2.5.3. Bacteriophage (phage) diversity

Phages are ubiquitous in the environment (Hassan et al., 2021), and a high diversity of different phages were detected in the samples. While many studies have looked at phages as a natural control for food pathogens in poultry, there is little information in the literature on the occurrence and role of naturally occurring phages during poultry processing and on the phages detected in this project in a poultry context. There is some evidence that some phages may play a role in the transfer of genes (including ARGs) between bacteria (Hassan et al., 2021), though as highlighted by Colavecchio et al. (2017), there is considerable debate on their importance in natural environments.

The most abundant and commonly occurring phage in both plants, P1, was a transducing phage. This phage is known to be both a reservoir and vector for horizontal gene transfer (HGT) that can infect E. coli and to transduce genes to bacteria other than E. coli (Riquelme et al., 2019). It has been identified as a potential vehicle for the spread and maintenance of virulence genes and ARGs in both clinical and livestock production settings (Venturini et al., 2019; M. Wang et al., 2022).

Escherichia phage P694 (the second and third most abundant phage in samples from Sites A and B, respectively) has been isolated from duck faeces in China (M. Chen et al., 2016) and reported to lyse avian pathogenic E. coli strain DE069. We have found no other literature identifying other hosts. The phage has been reported to be relatively sensitive to heat, not surviving a treatment of 65°C for 30 min. The scalding treatments used at Sites A and B are unlikely to deactivate it as the temperatures and times used to scald are too low and short, but the thermal interventions used at both plants may affect survival. However, this phage was detected in post intervention samples at both plants on a number of visits. It was detected at Site A post-intervention in October and at Site B post-intervention in June and November.

We have found no relevant data on Erwinia phage FE44. Acinetobacter phage B1251, the second most abundant phage at Site B, has been reported to lyse Acinetobacter baumannii (Jeon et al., 2012). A. baumannii was detected in samples from both sites on all sampling periods. It was detected in samples from post-intervention processing stages at Site A on all sampling months and was the seventeenth most abundant classifiable bacterial species detected at this site. A. baumannii causes infections in hospitals, and antimicrobial resistant A. baumannii has been reported, but data is limited regarding the role of A. baumannii in foods.

The detection of Salmonella phage SJ46 in majority of samples supports speculation in the literature that SJ46 is related to P1 and SJ46 may have a broader host range and also target E. coli (Gabashvili et al., 2020; Strange et al., 2021). There is also evidence that SJ46 may serve as a vector for the transfer of bla_CTX-M ARGs that code for B-Lactamase resistance (Gabashvili et al., 2020; Hassan et al., 2021). Enterobacteria phage mEp460, also detected in a majority of samples, infects E. coli, and has been reported to have a possible role in the transfer of virulence genes to other bacteria in the chicken gut microbiome (Oladeinde et al., 2019). Only one Campylobacter phage was detected (in November at Site B). This phage, CPt10, is known to lyse C. jejuni and C. coli and has been isolated previously from UK poultry processing environments (Salama et al., 1989; Timms et al., 2010).

5. Conclusion and recommendations

The approach of traditional culture combined with genomic analysis allowed us to provide a detailed map of potential contamination events in the process chain of a working poultry environment. We were able to conclude that the target organisms were readily recovered from the process chain at early stages but following several hygienic contamination reduction practices in general the ability to recover Campylobacter spp. and E. coli was reduced. Moreover, parallel sampling for genomic analysis showed early-stage extensive bacterial diversity which was concordant with associated ARG diversity and their reduction through the processing chain suggests that ARGs along with their host bacteria may also be reduced by hygienic contamination reduction practices, rather than ARGs persisting in particular resilient bacteria throughout.

This study (in common with others) shows that shotgun metagenomic sequencing is a suitable approach to investigate how the microbiome composition and diversity changes during processing and provide insights not provided by traditional microbiological analysis. However, such approaches also result in large data sets which must be analysed and interpreted.

The association of ARG diversity with taxonomic diversity observed in the shotgun metagenomics results suggests that ARGs may be reduced with hygiene/contamination reduction, rather than ARGs persisting in particular, resilient taxa.

This study detected the prevalence of phages associated with the transduction of genes (including ARGs) which could in theory play a role in the transfer of ARGs within the poultry microbiome. Whether such transfer takes place, and whether it is significant, not only in the poultry microbiome but also in other food animal microbiomes is unknown. Further work to investigate and quantify the risk associated with the possible phage-mediated transfer of such ARGs may be worthwhile.

It should be stressed that these findings are only based on samples from two poultry processing sites. Further studies are needed to clarify if the microbiome composition and changes in diversity observed in the two UK poultry plants investigated in this study are representative of other UK-wide poultry plants. It is vital that future work focuses on elucidating the processing variables that have the greatest influence on these microbiomes, thus allowing for eventual targeted interventions to better manage these microbial populations to benefit environmental and public health.