Improved Genomic Prediction of Staphylococcus epidermidis Isolation Sources with a Novel Polygenic Score.

Staphylococcus epidermidis infections can be challenging to diagnose due to the species frequent contamination of clinical specimens and indolent course of infection. Nevertheless, S. epidermidis is the major cause of late-onset sepsis among premature infants and of intravascular infection in all age groups. Prior work has shown that bacterial virulence factors, antimicrobial resistances, and strains have up to 80% in-sample accuracy to distinguish hospital from community sources, but are unable to distinguish true bacteremia from blood culture contamination. Here, a phylogeny-informed genome-wide association study of 88 isolates was used to estimate effect sizes of particular genomic variants for isolation sources. A "polygenic score" was calculated for each isolate as the summed effect sizes of its repertoire of genomic variants. Predictive models of isolation sources based on polygenic scores were tested with in-samples and out-samples from prior studies of different patient populations. Polygenic scores from accessory genes (AGs) distinguished hospital from community sources with the highest accuracy to date, up to 98% for in-samples and 65% to 91% for various out-samples, whereas scores from single nucleotide polymorphisms (SNPs) had lower accuracy. Scores from AGs and SNPs achieved the highest in-sample accuracy to date, up to 76%, in distinguishing infection from contaminant sources within a hospital. Model training and testing data sets with more similar population structures resulted in more accurate predictions. This study reports the first use of a polygenic score for predicting a complex bacterial phenotype and shows the potential of this approach for enhancing S. epidermidis diagnosis.