IDENTIFYING AMR GENE RESERVOIRS AND BACTERIAL HOST-AMR GENE ASSOCIATIONS TO IDENTIFY BACTERIAL HOST RANGE OF AMR GENES IN SWINE PRODUCTION SYSTEMS

<b>Project Methods</b><br> Research Objective 1- Determine the effects of antimicrobial use on the ecology of microbial community and microbial resistance gene reservoirs in the porcine gut and wastewater (to evaluate downstream effects) at different production stages using longitudinal sampling. Specifically, we will evaluate the changes in the microbial community composition and microbial resistance gene reservoirs in the gastrointestinal tract of antimicrobial treated (55 mg/Kg carbadox and 250 mg/Kg copper sulfate in (AB1) and chlortetracycline/tiamulin hydrogen formate (AB2) and untreated animals (CONTROL) at different stages of the production cycle (day 28, 56, 91, and 126 covering weaning, nursery, growing and finishing phase of the production cycle) and in wastewater by sequencing the virome, plasmidome, microbial community and associated microbial resistance genes. The study will be carried out at the pork production facility at U.S. Meat Animal Research Center (USMARC) using 12 pens with 8 head per pen for each rep resulting in a total of 192 pigs with 64 pigs per treatment.Research Objective 2 – Develop neural network models to predict the bacterial host range of raw plasmid metagenomic sequences in complex bacterial communities for rapid risk assessment of emerging multi-drug resistant strains using Nanopore sequencing.We will train a language model with the fastai deep learning library using dereplicated plasmid genomes (RefSeq) split into 5,000 base pair segments with a 100 base pair overlap. A subword tokenization algorithm will be used to tokenize the genomic segments. The validation set for the language model will be genomic segments from 10% of the dereplicated genomes, rather than 10% of all genomic segments, to avoid data leakage between the training and validation datasets. We will train a four-layer quasi-recurrent neural network (QRNN) based on the concepts of masked language modeling (15% of tokens masked). Transfer learning involves further training (i.e., fine-tuning) a model to be more specific to a downstream task, which improves model performance. Nanopore sequences have unique error-profiles which can be simulated with NanoSim. We will fine-tune the pretrained plasmid language model using simulated Nanopore metagenomic sequences generated from dereplicated plasmid reference genomes. The training approach will be identical to pretraining, yielding a plasmid language model fine-tuned to the characteristics of Nanopore sequencing data. The fine-tuned plasmid language model will be used to develop a classifier to assign Nanopore metagenomic sequences to previously described and novel PTUs derived from the plasmid enriched metagenomic data in Objective 1 or an 'Other' class (bacteria, viruses, etc.). Nanopore plasmid metagenomic sequences for training the classifier will be simulated with NanoSim from a dereplicated set of RefSeq genomes (bacterial, viral, and plasmids). The classifier will be evaluated based on the accuracy of the test data, the accuracy on simulated and real metagenomic datasets, and the accuracy of newly sequenced plasmids not included in the training data. The plasmid enriched metagenomes from Objective 1 will be used to evaluate the classifier by comparing the assigned bacterial taxa obtained from methylation assisted binning and the plasmid language model classifier. The classifier accuracy will be compared to baseline tools, minimap2, a metagenomic mapping tool for long-reads, and Kraken. The developed model will be applied to the plasmid enriched metagenomic data generated in Objective 1 to better understand the host-range of plasmid-borne AMR genes throughout different production stages. Nanopore metagenomic reads will be fed into the model and sequences classified to PTUs and will be screened for AMR genes with DeepARG. Using the PTU network, we can assess the if the potential mobility of plasmid-borne AMR genes into foodborne pathogens is higher at different stages of production. For instance, the model will allow us to assess the potential host-range of plasmids with AMR genes early in production that may transfer to Escherichia coli and verify if those transfer events occurred via isolate screening included in Objective 1.Research Objective 3 – Develop a novel science-based management and nutritional strategy to reduce metaphylactic antibiotic use by utilizing a novel essential oil as an alternative to conventional antibiotics, evaluating the changes in the microbiome due to utilizing the essential oil as a feed additive and evaluating its applicability as a nutritional strategy to reduce the use of antibiotics. Currently, antibiotics are widely used to reduce pathogen colonization associated with weaning stress and this objective will help develop dietary intervention strategies using alternatives to antibiotics to reduce antibiotics in the swine industry.Utilizing the pork production facility at USMARC, the feeding experiments will be performed over the whole production cycle from farrowing to finish. Two replicates of 16 sows per farrowing group will be utilized that will contain n=4 litters/replicate. Litters will be standardized to a difference of 1 pig or less (10 or 11, 11 or 12, etc.), depending on number of pigs per sow, and cross-fostering will be performed in the first week of life. Pigs from the 32 litters (8 total litters per treatment) will be fed one of 4 treatments (1) Control, 2) Control + essential oil (EO, 2.5% inclusion rate), 3) Control + carbadox/copper sulfate (AB1, 55 mg/kg and 250mg/kg diet, respectively), or 4) Control + chlortetracycline/tiamulin hydrogen formate (AB2, 55 mg/kg and 1.65 g/kg, respectively) as creep feed during lactation starting at 7 d prior to weaning and throughout the nursery phase (28d; 35 d total). All pigs will be fed common diets in the growing and finishing phases of production. The pigs will have free access to each diet and diets will be formulated to meet or exceed NRC nutrient recommendations (1). At weaning, treatments will be moved into the nursery together and be group-housed and blocked by litter. Body weight will be measured gravimetrically starting on days 0, 10, 28, 56, 91 and 126 of life. Weekly pen feed disappearance will be measured in the nursery (d28 to 56). Individual weights and feed intakes will be measured in the USMARC Feed efficiency Barn (described in Obj. 1). At days 0, 28, 56, 91 and 126 a fecal grab sample will be collected from each animal. Subsamples of fecal grabs will be pooled by pen and be snap frozen and stored at -80°C for evaluating microbial community compositional changes.Outreach Activity #1: Deliver relevant outcomes generated by achievement of the project's research objectives through the iAMResponsibleTM Project and the eXtension Livestock and Poultry Environmental Learning Community (LPELC).Outreach Activity #2: Deliver project outcomes through the existing iAMResponsibleTM Project-led university graduate-level course, AMR from a One Health Perspective.