Fig. 1

Overview of the study methods. a Episodes with at least two blood isolates at least 3 days’ apart, and episodes with at least one isolate from a nasal swab were selected from a combined cohort of S. aureus bacteraemia. b DNA was extracted from one single colony. Reads from whole genome sequencing were mapped to the reference genome S. aureus TW20. Unrelated same-patient isolates (based on clustering on the phylogenetic tree, MLST, and SNP distance) were excluded from further analysis. c–f Episode-specific phenotypic and genomic analysis. Phenotypic tests included oxacillin and vancomycin MIC, measured by E-test, overnight growth curves in HI broth, and cell toxicity assays (c). Variants calling for SNPs and short indels was performed by mapping on the closest available complete genome and the de novo assembly of the index isolate, respectively. Variants were filtered based on read depth (≥ 10) and fraction of reference alleles (> 0.5) in the index isolate reads and confirmed by manual inspection of the alignments (d). To identify regions of genome loss that were unique within episode isolates, we scanned the read alignment to the complete genome for intervals with at least 400 bp read coverage loss (e). Screening for structural variants was performed by detecting split reads (along the alignment to the complete genome) that were unique within episode isolates. Structural variants were annotated and confirmed by blasting split intervals on the assembly graph of the episode isolates