Benchmarking Illumina and Oxford Nanopore Technologies (ONT) sequencing platforms for whole genome sequencing of bacterial genomes and use in clinical microbiology

Background: In microbial diagnostics, whole-genome sequencing (WGS) is used to address key questions such as species identification, presence of antimicrobial resistance genes (ARGs), virulence genes, and outbreak detection. The choice of sequencing technology is crucial to ensure high-quality data, cost-effectiveness, and efficient reporting times. We aimed to compare Illumina (short-read) and ONT (long-read) sequencing methods for WGS on different bacterial species for base accuracy and reliable taxonomic and ARG identification. Materials and methods: We used clinical isolates of ESKAPE pathogens (n = 12) and ATCC strains (n = 8) of varying %G + C. Illumina sequencing was performed on MiSeq (PE150) and ONT sequencing using GridION with R9.4.1 and R10.4.1 flowcells. Base-calling was performed using Guppy, Dorado, and Rerio software. We performed de novo assembly with Unicycler for Illumina and Flye for ONT, and two types of hybrid assemblies, Unicycler and Polypolish. We annotated genomes with Bakta and assessed the quality (QUAST, GTDB-Tk). We identified ARGs (AMRFinderPlus) and plasmids (MOB-suite). We mapped reads and called SNPs using Minimap2, Pilon, vcftools, and Snippy (Illumina). Core genome MLST analysis was conducted with Ridom Seqsphere+. Results: We observed that Illumina sequencing provided consistently high-quality reads (median Q-score 35), whereas for ONT R10.4.1, SUP model showed higher median quality (median Q-score 15.3) compared to R9.4.1 (median Q-score 13.9, SUP model). We observed that Illumina-based assemblies generated fewer genes annotated as disrupted; for ONT assemblies, the base-caller affects assembly annotation accuracy, with High accuracy (HAC) and Super accuracy (SUP) base-calling modes perform better than FAST model. ONT assemblies resolved rRNA operons better than Illumina assemblies. Sequencing errors were determined by SNP calling, and varied widely by species, with ONT often generating more sequencing errors compared to Illumina. Hybrid assemblies combine accuracy and completeness effectively. Taxonomic identification and ARG detection were reliable across all methods. Conclusion: Combining Illumina and ONT technologies yielded optimal bacterial genome sequencing results, leveraging the high accuracy of short reads and improved contiguity of ONT long reads. The HAC and SUP ONT models with Dorado notably enhance genome assembly annotation and resolution of complex regions, although species-specific issues, likely due to repeat regions and base modifications, remain challenging even in SUP model with Dorado. Hybrid approaches currently offer the most comprehensive and accurate genome assemblies for clinical microbiology. For reliable cgMLST even using the most recent ONT methods, resolution must be assessed on a species-by-species basis.