Optimizing quasi-metagenomic sequencing for strain-level detection and genomic surveillance of foodborne pathogens in complex food matrices

Abstract

Foodborne disease surveillance relies heavily on whole-genome sequencing (WGS) of bacterial isolates to support outbreak detection and source attribution. However, culture-based workflows introduce delays that can limit timely public health response, particularly for perishable foods and complex matrices. This thesis evaluates quasi-metagenomics, defined as sequencing enriched food matrices without secondary isolate recovery, as a culture-independent approach for strain-level pathogen detection and characterization. The objective was to determine whether quasi-metagenomic sequencing can provide strain-resolved genomic information suitable for foodborne outbreak detection. This thesis therefore investigates the laboratory and analytical requirements needed to achieve outbreak-level genomic resolution directly from enriched food samples. Chapter 3 establishes the bioinformatic foundation by benchmarking taxonomic classifiers, reference databases, and analytical parameters. Results demonstrated that a curated foodborne pathogen (FBP)-specific reference database used with KrakenUniq provided the most accurate and consistent taxonomic classification while reducing computational burden. Chapter 4 applies this optimized workflow to short-read quasi-metagenomic datasets generated from enriched food matrices artificially contaminated with Listeria monocytogenes, verotoxigenic Escherichia coli (VTEC), and Shigella spp. Selective enrichment had a greater influence on pathogen recovery than eukaryotic DNA depletion. When sufficient pathogen DNA was recovered, quasi-metagenomic sequencing enabled accurate pathogen detection, serotype prediction, genome assembly, and whole-genome multilocus sequence typing (wgMLST)-based outbreak characterization with genomic distance ranges comparable to isolate-derived WGS. To support standardized interpretation of quasi-metagenomics, a pathogen-presence likelihood metric was developed that integrates sequencing depth-normalized unique k-mer counts (unique KPMR; k-mers per million reads) with robust z-scores to distinguish true pathogen signal from background noise. Chapter 5 extends this workflow to Oxford Nanopore Technologies (ONT) long-read sequencing. Despite higher raw read error rates, long-read quasi-metagenomic sequencing supported accurate taxonomic classification, genome assembly, and outbreak characterization. Single nucleotide polymorphism (SNP)- and wgMLST-based analyses produced genomic distance ranges comparable to those obtained using short-read quasi-metagenomics and isolate-derived WGS. Overall, this research establishes an end-to-end quasi-metagenomic workflow for strain-level pathogen detection and outbreak characterization directly from enriched food samples. These findings provide a strong methodological foundation for the continued development of culture-independent genomic approaches to support foodborne disease surveillance and outbreak response.