Disease-associated genotypes of the commensal skin bacterium Staphylococcus epidermidis

Guillaume Méric, Leonardos Mageiros, Johan Pensar, Maisem Laabei, Koji Yahara, Ben Pascoe, Nattinee Kittiwan, Phacharaporn Tadee, Virginia Post, Sarah Lamble, Rory Bowden, James E. Bray, Mario Morgenstern, Keith A. Jolley, Martin Maiden, Edward J. Feil, Xavier Didelot, Maria Miragaia, Hermı́nia de Lencastre, T. Fintan Moriarty, Holger Rohde, Ruth C. Massey, Dietrich Mack, Jukka Corander, Samuel K. SheppardNature Communications 2018 · DOI

Abstract

Some of the most common infectious diseases are caused by bacteria that naturally colonise humans asymptomatically. Combating these opportunistic pathogens requires an understanding of the traits that differentiate infecting strains from harmless relatives. Staphylococcus epidermidis is carried asymptomatically on the skin and mucous membranes of virtually all humans but is a major cause of nosocomial infection associated with invasive procedures. Here we address the underlying evolutionary mechanisms of opportunistic pathogenicity by combining pangenome-wide association studies and laboratory microbiology to compare S. epidermidis from bloodstream and wound infections and asymptomatic carriage. We identify 61 genes containing infection-associated genetic elements (k-mers) that correlate with in vitro variation in known pathogenicity traits (biofilm formation, cell toxicity, interleukin-8 production, methicillin resistance). Horizontal gene transfer spreads these elements, allowing divergent clones to cause infection. Finally, Random Forest model prediction of disease status (carriage vs. infection) identifies pathogenicity elements in 415 S. epidermidis isolates with 80% accuracy, demonstrating the potential for identifying risk genotypes pre-operatively.

Introduction

Staphylococcus epidermidis presents a biological paradox: while it is a ubiquitous and harmless commensal on human skin, it is also a leading cause of hospital-acquired (nosocomial) infections, particularly those involving surgical implants and catheters. For clinicians, the primary challenge lies in distinguishing between harmless "commensal" strains and dangerous "pathogenic" invaders. Because S. epidermidis is so common, it is often dismissed as a sample contaminant in clinical settings. This research aims to identify the specific genetic signatures that drive pathogenicity and determine if a strain's risk level can be predicted before surgical intervention.

Methods and Materials

The researchers employed a multi-stage computational and experimental pipeline to analyze 415 S. epidermidis isolates, comprising 274 infection-associated samples and 141 asymptomatic carriage samples.

Experimental and computational workflow for identifying infection-associated genetic elements.

Key methodological steps included:

  1. K-mer based GWAS: An alignment-free approach analyzing short DNA sequences (k-mers) was used to capture variation in the "accessory genome"—genes present in some strains but not others—rather than just single-nucleotide polymorphisms (SNPs).
  2. Replicate GWAS: To filter out noise from population structure, the analysis was run on independent datasets.
  3. Phenotype Correlation: Isolates were tested in vitro for biofilm formation, cell toxicity, induction of the immune marker IL-8, and methicillin resistance.
  4. Machine Learning: A Random Forest classifier was trained to predict disease status (carriage vs. infection) based on the presence of specific genetic elements.

Results and Insights

The study identified 61 genes containing infection-associated elements that correlate with pathogenicity traits. A significant finding is the "Divided Genome" model: pathogenicity in S. epidermidis is not restricted to a single lineage but is spread horizontally across the species via horizontal gene transfer (HGT). This suggests the species maintains a pool of "pathogenicity islands" that can move between different genetic backgrounds.

Distribution of isolates, phylogeny, and genomic localization of infection-associated elements.

The mecA gene, which provides methicillin resistance, emerged as the strongest predictor of infection. Its presence strongly correlated with other virulence traits, suggesting an evolutionary link between antibiotic resistance and pathogenicity. The Random Forest model achieved 80% accuracy in predicting whether an isolate was from a carriage or infection source. Notably, using just the top four phenotype-linked predictors yielded nearly the same accuracy as using the entire pangenome.

Discussion and Future Work

The identification of 61 high-risk genes and the success of the predictive model highlight the potential for rapid, pre-operative diagnostic tools. However, a fundamental limitation exists: some "carriage" isolates may be pathogenic strains that simply haven't had the opportunity to cause infection, placing a ceiling on predictive accuracy. Future work will involve larger, geographically diverse datasets to refine the risk score and investigate the molecular mechanisms of the identified hypothetical proteins. The ultimate goal is to flag high-risk patients carrying dangerous genotypes before they undergo surgery.

Summary updated

Abstract

Some of the most common infectious diseases are caused by bacteria that naturally colonise humans asymptomatically. Combating these opportunistic pathogens requires an understanding of the traits that differentiate infecting strains from harmless relatives. Staphylococcus epidermidis is carried asymptomatically on the skin and mucous membranes of virtually all humans but is a major cause of nosocomial infection associated with invasive procedures. Here we address the underlying evolutionary mechanisms of opportunistic pathogenicity by combining pangenome-wide association studies and laboratory microbiology to compare S. epidermidis from bloodstream and wound infections and asymptomatic carriage. We identify 61 genes containing infection-associated genetic elements (k-mers) that correlate with in vitro variation in known pathogenicity traits (biofilm formation, cell toxicity, interleukin-8 production, methicillin resistance). Horizontal gene transfer spreads these elements, allowing divergent clones to cause infection. Finally, Random Forest model prediction of disease status (carriage vs. infection) identifies pathogenicity elements in 415 S. epidermidis isolates with 80% accuracy, demonstrating the potential for identifying risk genotypes pre-operatively.
BibTeX
@inproceedings{meric2018disease,
  author = {Guillaume Méric and Leonardos Mageiros and Johan Pensar and Maisem Laabei and Koji Yahara and Ben Pascoe and Nattinee Kittiwan and Phacharaporn Tadee and Virginia Post and Sarah Lamble and Rory Bowden and James E. Bray and Mario Morgenstern and Keith A. Jolley and Martin Maiden and Edward J. Feil and Xavier Didelot and Maria Miragaia and Hermı́nia de Lencastre and T. Fintan Moriarty and Holger Rohde and Ruth C. Massey and Dietrich Mack and Jukka Corander and Samuel K. Sheppard},
  title = {{Disease-associated genotypes of the commensal skin bacterium Staphylococcus epidermidis}},
  booktitle = {Nature Communications},
  year = {2018},
  doi = {10.1038/s41467-018-07368-7},
}
Figure