Improved diagnostic prediction of the pathogenicity of bloodstream isolates of Staphylococcus epidermidis

With an estimated 440,000 active cases occurring each year, medical device associated infections pose a significant burden on the US healthcare system, costing about $9.8 billion in 2013. Staphylococcus epidermidis is the most common cause of these device-associated infections, which typically involve isolates that are multi-drug resistant and possess multiple virulence factors. S. epidermidis is also frequently a benign contaminant of otherwise sterile blood cultures. Therefore, tests that distinguish pathogenic from non-pathogenic isolates would improve the accuracy of diagnosis and prevent overuse/misuse of antibiotics. Attempts to use multi-locus sequence typing (MLST) with machine learning for this purpose had poor accuracy (~73%). In this study we sought to improve the diagnostic accuracy of predicting pathogenicity by focusing on phenotypic markers (i.e., antibiotic resistance, growth fitness in human plasma, and biofilm forming capacity) and the presence of specific virulence genes (i.e., mecA, ses1, and sdrF). Commensal isolates from healthy individuals (n = 23), blood culture contaminants (n = 21), and pathogenic isolates considered true bacteremia (n = 54) were used. Multiple machine learning approaches were applied to characterize strains as pathogenic vs non-pathogenic. The combination of phenotypic markers and virulence genes improved the diagnostic accuracy to 82.4% (sensitivity: 84.9% and specificity: 80.9%). Oxacillin resistance was the most important variable followed by growth rate in plasma. This work shows promise for the addition of phenotypic testing in clinical diagnostic applications.


53
Rapid multiplex PCR on suspected pathogenic isolates reduced unnecessary antibiotic use by identifying 54 likely contaminating species, however it offers no way to distinguish pathogenic from commensal S. 55 epidermidis isolates(14). The development of a superior discriminatory tool to distinguish pathogenic 56 from non-pathogenic blood culture isolates of S. epidermidis would help prevent overuse and/or misuse 57 of antibiotics, which contributes to the ongoing threat of antibiotic resistance. We begin by reviewing 58 potential mechanisms for evaluating the pathogenicity of S. epidermidis.

59
S. epidermidis isolated from infections have been suggested to be a molecular subset of those 60 found on the skin surface (i.e., they continue to carry out original functions of the non-infectious 61 'lifestyle') (15)(16)(17). This implies that, rather than passive infection, there may be certain lineages or 62 specific virulence factors associated with the emergence of pathogens from a background of harmless 63 ancestors. Specifically, conferred pathogenicity is manifested in remaining competitive through 64 selection, namely antibiotic treatment, host microenvironment, or immune clearance(18). Virulence genes, including antibiotic resistance, biofilm formation capacity and other metabolic advantages, are 66 commonly passed on pathogenicity island plasmids exogenously or through horizontal gene transfer.

67
Genotype testing has been used to help identify isolates that carry specific virulence genes(19), and thus 68 have potential to express virulence. Genotyping, however, does not take into consideration the gross   TSBg supplemented with 10% human plasma. Single colonies were grown overnight in TSBg at 37°C then 146 diluted to OD 600 of 0.5-0.6 and a 1:100 dilution was used to achieve initial inoculum of 5x10 5 cells per 147 well. Growth curves for each isolate in each medium were obtained in at least triplicate. Curves were fit 148 to a Gompertz function to determine the maximum growth rate (µ), lag phase (λ), and maximum OD 600 149 (A) (see Supplemental Figure S1 and Supplemental Table S1). 167 For growth fitness we were interested in differential growth parameters in the presence and 168 absence of plasma as a surrogate for growth fitness in blood. These differential parameters were

175
For virulence genes, the frequency with which the gene is present within isolate types was 176 determined and hypothesis testing using the χ 2 test with Bonferroni correction for multiple comparisons 177 was performed like that for the antibiotic resistance.

178
Given that we evaluated methicillin resistance from both a genotypic and phenotypic 179 perspective, we chose to compare these methods by generating and comparing confusion tables for 180 each isolate group.

182
Predictive modeling: We initially performed an analysis of similarity (ANOSIM) which demonstrated that 183 dissimilarity between isolates types was higher than that within each type but quite similar to that of the 184 commensal and contaminates types (Supplemental Figure S2). The pathogen type however had 185 significantly less dissimilarity indicating the possibility that this isolate type is a more homogenous 186 subgroup. Therefore, we chose to proceed under the assumption that isolates characterized as 187 commensal or contaminant should be considered to be from one population and the pathogens from 188 another. Therefore, all predictive modeling henceforth will be for the binary pathogen vs non-pathogen Biofilm formation: Biofilm formation is a commonly described virulence factor for S. epidermidis. There 218 was high variability in biofilm forming capacity within isolate types. While the pathogen type had a 219 greater median crystal violet staining, the difference was not statistically significant (Figure 2). Growth fitness: We hypothesized that pathogenic isolates grow better in human plasma when 225 compared to non-pathogenic isolates. This is based on the idea that transition to the new environment 226 of the bloodstream favors pathogenic strains over commensals. In most cases, a delay in lag time (λ)

227
was seen in all isolates grown in plasma, suggesting overall inhibition. However, commensal isolate 228 growth seemed to be most negatively affected by the presence of plasma. This was manifested in 229 increased lag time (λ) and decreased growth rate (µ) for the commensal isolates relative to the 230 contaminants and pathogens (Figure 3). Pathogenic isolates tended to be less inhibited in plasma  As noted earlier, there is the potential for discordance between genotypic and phenotypic traits. with which a particular variable appears in each validation of the five models is indicative of its 278 predictive power and is shown in Figure 5. Oxacillin, vancomycin, and erythromycin phenotypic 279 resistance and the growth rate (µ) ratio were the most frequent features in the models. Oxacillin 280 resistance was present in all but one model. interpretations, followed by a discussion of the performance of the predictive models, and conclude 292 with a discussion of the implications, limitations, and future directions for this work.

293
Pathogenic S. epidermidis isolates originate from patients in a healthcare setting where 294 antibiotic treatment leads to potential selection of antibiotic resistance cassettes and pathogenicity 295 islands(3). This phenomenon is highlighted in this study as pathogens and contaminants from 296 hospitalized patients had consistently higher frequency of antibiotic resistance than commensal isolates 297 from healthy individuals. In particular, pathogens were significantly more likely to be resistant to 298 ciprofloxacin and oxacillin. In addition, commensal isolates had both reduced growth rates and 299 increased time to exponential growth in the presence of blood plasma compared to contaminants and 300 pathogens, as has been demonstrated previously(44). This could result from the fact that the 301 commensal isolates were obtained from thumb prints of healthy patients rather than blood cultures of 302 patients with suspected bloodstream infection and were never exposed to the blood microenvironment.

303
There was significant variability in biofilm forming capacity within isolates types and no detectable 304 difference between groups. All types of S. epidermidis isolates were able to build biofilms as is likely 305 required for colonization of the skin(16). Pathogens, however, may have more overlapping biofilm 306 formation mechanisms as part of the selection for pathogenicity islands.

307
While several of the individual tests performed demonstrated differential trends between 308 groups, no one factor, genotypic or phenotypic, distinguished isolates with perfect accuracy. This is not overuse and misuse of antibiotics as antibiotic resistance has emerged as one of the major urgent 376 threats to public health(57). Antibiotic treatment has potential adverse outcomes with adverse drug 377 reactions and hypersensitivity reactions accounting for more than 3% of hospital admissions(58), which 378 generates a significant burden on the health care system through secondary effects(59).