A statistical genomics framework to trace bacterial genomic predictors of clinical outcomes in Staphylococcus aureus bacteraemia

Abstract

Outcomes for patients with severe bacterial infections are determined by the interplay between host, pathogen, and treatments. Most notably, patient age and antibiotic resistance contributes significantly to poor outcomes. While human genomics studies have provided insights into the host genetic factors impacting outcomes of Staphylococcus aureus infections, comparatively little is known about S. aureus genotypes and disease severity. Building on the idea that bacterial pathoadaptation is a key driver of clinical outcomes, we develop a new genome-wide association study (GWAS) framework to identify adaptive bacterial mutations associated with clinical treatment failure and mortality in three large and independent S. aureus bacteraemia cohorts, comprising 1358 episodes. We discovered S. aureus loci with previously undescribed convergent mutations linked to both poorer infection outcomes and reduced susceptibility to vancomycin. Our research highlights the potential of vancomycin-selected mutations and vancomycin MIC as key explanatory variables to predict SAB severity. The contribution of bacterial variation was much lower for clinical outcomes (heritability < 5%), however, GWAS allowed us to identify additional, MIC-independent candidate pathogenesis loci. Using supervised machine-learning, we were able to quantify the predictive potential of these adaptive S. aureus signatures, along with host determinants of bacteraemia outcomes. The statistical genomics framework we have developed is a powerful means to capture adaptive mutations and find bacterial factors that influence and predict severe infections. Our findings underscore the importance of systematically collected, rich clinical and microbiological data to understand bacterial mechanisms promoting treatment failure.

A high-throughput cytotoxicity screening platform revealsagr-independent mutations in bacteraemia-associatedStaphylococcus aureusthat promote intracellular persistence

“…Major MDS axes correlated with the most prevalent STs, for example ST239 was mainly defined by MDS1 (negative correlation) and MDS2 (positive correlation) (Figure 4-Supplementary Figure 3A). We then tested the association between the first 10 MDS axes (90% of the genetic variance explained) and the PI uptake phenotype in Pyseer (Earle et al, 2016;Giulieri, Guérillot, Holmes, et al, 2022;Lees et al, 2018). In agreement with the initial observations based on the phylogeny and cytotoxicity heatmap (Figure 4A), we observed significant cytotoxicity-lineage associations represented by MDS3 and MDS4 (Figure 4D, Figure 4-Supplementary Figure 2).…”

“…While intracellular cytotoxicity was strongly associated with some S. aureus lineages, this analysis showed that lineage alone does not completely explain the phenotype, as indicated by the significant overlap between the three cytotoxicity clusters across MDS3 and MDS4 (Figure 4D). This pattern is consistent with other adaptive phenotypes (Earle et al, 2016;Giulieri, Guérillot, Holmes, et al, 2022;Su et al, 2021) and suggests that locus effects from specific micro-evolutionary events modulate cytotoxicity, supporting the use of GWAS and convergent evolution approaches to identify these mutations.…”

“…We next used GWAS to identify genetic correlates of strain-level cytotoxicity, expressed as mean PI uptake AUC. The fraction of cytotoxicity variation explained by genetic variation (heritability: h 2 ) was 49%, a figure lower than the ones obtained for other phenotypes such as vancomycin resistance (Giulieri, Guérillot, Holmes, et al, 2022). A lower heritability could be resulting from the InToxSa assay variability or caused by differences in gene expression levels or due to epigenetic changes.…”

“…Current statistical genomics strategies in human genetic support combining allele-counting methods (GWAS), for the detection of common variants, with comparative genomics approaches to identify rare variants (Singh et al, 2022;Trubetskoy et al, 2022). In microbial genomics, this strategy is best achieved by combining microbial GWAS and convergent evolution studies (Chen & Shapiro, 2021;Giulieri, Guérillot, Holmes, et al, 2022;Guérillot et al, 2018;Saund & Snitkin, 2020). Whilst our GWAS approach only identified agrA as significantly associated with low cytotoxicity (Figure 4E), our evolutionary convergence analysis on genetic pairs among our 387 bacteraemia isolates identified mutations in several S. aureus genes that led to reduced cytotoxicity (Figure 5).…”

“…Despite the relatively small sample size for this kind of analysis, the gene-burden GWAS detected the agrA locus with a high significance, but it did not have sufficient power to detect mutations other than the agr genes. We further sought to identify rare mutations that might alter the intracellular cytotoxicity using comparative genomics approaches, a complementary strategy to microbial GWAS (Chen & Shapiro, 2021;Giulieri, Guérillot, Holmes, et al, 2022;Saund & Snitkin, 2020). We used evolutionary convergence analysis to identify additional loci associated with intracellular cytotoxicity among the 387 S. aureus isolates.…”