Differentiation of hypervirulent and classical Klebsiella pneumoniae with acquired drug resistance

Abstract

Distinguishing hypervirulent (hvKp) from classical Klebsiella pneumoniae (cKp) strains is important for clinical care, surveillance, and research. Some combinations of iucA, iroB, peg-344, rmpA, and rmpA2 are most commonly used, but it is unclear what combination of genotypic or phenotypic markers (e.g., siderophore concentration, mucoviscosity) most accurately predicts the hypervirulent phenotype. Furthermore, acquisition of antimicrobial resistance may affect virulence and confound identification. Therefore, 49 K. pneumoniae strains that possessed some combinations of iucA, iroB, peg-344, rmpA, and rmpA2 and had acquired resistance were assembled and categorized as hypervirulent hvKp (hvKp) (N = 16) or cKp (N = 33) via a murine infection model. Biomarker number, siderophore production, mucoviscosity, virulence plasmid's Mash/Jaccard distances to the canonical pLVPK, and Kleborate virulence score were measured and evaluated to accurately differentiate these pathotypes. Both stepwise logistic regression and a CART model were used to determine which variable was most predictive of the strain cohorts. The biomarker count alone was the strongest predictor for both analyses. For logistic regression, the area under the curve for biomarker count was 0.962 (P = 0.004). The CART model generated the classification rule that a biomarker count = 5 would classify the strain as hvKP, resulting in a sensitivity for predicting hvKP of 94% (15/16), a specificity of 94% (31/33), and an overall accuracy of 94% (46/49). Although a count of ≥4 was 100% (16/16) sensitive for predicting hvKP, the specificity and accuracy decreased to 76% (25/33) and 84% (41/49), respectively. These findings can be used to inform the identification of hvKp.IMPORTANCEHypervirulent Klebsiella pneumoniae (hvKp) is a concerning pathogen that can cause life-threatening infections in otherwise healthy individuals. Importantly, although strains of hvKp have been acquiring antimicrobial resistance, the effect on virulence is unclear. Therefore, it is of critical importance to determine whether a given antimicrobial resistant K. pneumoniae isolate is hypervirulent. This report determined which combination of genotypic and phenotypic markers could most accurately identify hvKp strains with acquired resistance. Both logistic regression and a machine-learning prediction model demonstrated that biomarker count alone was the strongest predictor. The presence of all five of the biomarkers iucA, iroB, peg-344, rmpA, and rmpA2 was most accurate (94%); the presence of ≥4 of these biomarkers was most sensitive (100%). Accurately identifying hvKp is vital for surveillance and research, and the availability of biomarker data could alert the clinician that hvKp is a consideration, which, in turn, would assist in optimizing patient care.

Keywords: Klebsiella; biomarker; classical; diagnosis; hypervirulent.

Fig 1

Phenotypic markers. Forty-nine strains comprising the hvKp (n = 16) and cKp (n = 33) cohorts underwent assessment for LD₅₀ in the outbred CD1 murine subcutaneous challenge infection model (Panel A), in vitro quantitative siderophore production when grown in c-M9-CA, black symbols represent strains that do not possess iucA (Panel B), and in vitro quantitative mucoviscosity when grown in either LB (Panel C), or c-M9-CA-te medium (Panel D). For Panel A, see Table S1 for challenge inocula and animal number for each strain. For Panel B, a minimum of three biologic with two technical repeats was performed for each strain. For Panels C and D, a minimum of three biologic repeats was performed for each strain. Strain cohorts were compared via Mann-Whitney except siderophore production was compared by an unpaired t-test (Prism version 9.5.1). Panel A, ****P < 0.0001. Panel C, *P = 0.0113, Panel D, ****P < 0.0001.

Fig 2

Genotypic markers. Forty-nine strains comprising the hvKp (n = 16) and cKp (n = 33) cohorts underwent assessment for marker count (Panel A), 29 strains comprising the hvKp (n = 9) and cKp (n = 20) cohorts underwent assessment for Mash distance (Panel B), and for Jaccard distance (Panel C), and all 49 strains underwent assessment for Kleborate score (Panel D). Strain cohorts were compared via Mann-Whitney (Prism version 9.5.1). Panel A, **** P = <0.0001, Panel B, ***P = 0.0002, Panel C, ***P = 0.0001, Panel D, ***P = 0.001.

Fig 3

Biomarker distribution Panel A. UpSet graph for the 49 strains comprising hvKp (n = 16) and cKp (n = 33) was generated to facilitate the visualization of the presence or absence of the biomarkers iucA, iroB, peg-344, rmpA, and rmpA2. Black circles designate the presence of a given marker, whereas gray circles designate its absence. Pathotypes are color coded. The number above each bar represents the number of strains that possess that marker configuration. Panel B: Results are shown as the distribution in proportions of each biomarker count (1–5) and Kleborate virulence score (0–5) for the hvKp and cKp cohorts. The biomarkers count is the presence of some combination of iucA, iroB, peg-344, rmpA, and rmpA2. The Kleborate virulence score is calculated by the presence of the genes that encode aerobactin (three points), yersiniabactin (one point), and colibactin (one point).

Fig 4

Mash and Jaccard Distances The pLVKP-like virulence plasmid of 29 strains comprising the hvKp (n = 9) and cKp (n = 20) cohorts, for which long-read sequencing was obtained thereby enabling a closed genome/plasmid, was compared to the canonical pLVPK sequence. Scatterplots of Mash/Jaccard distance to pLVPK were plotted for each strain and colored by pathotype, number of biomarkers (some combination of iucA, iroB, peg-344, rmpA, and rmpA2), Kleborate virulence score (presence of genes that encode aerobactin (three points), yersiniabactin (one point), and colibactin [one point)], and antimicrobial resistance (AMR, presence of ESBL/carbapenemase genes).

Fig 5

Proksee alignment to pLVPK: 29 strains with a closed plasmid harboring any virulence biomarker were compared to the canonical pLVPK. Alignment of the pLVPK sequence to all closed pLVPK-like plasmid sequences from (A) hvKp (n = 9) and (B) cKp isolates (n = 20) were produced with Proksee. Virulence and landmark genes from pLVPK are labeled in each panel. The five virulence biomarkers of interest are concentrated on a roughly 30 kb region of pLVPK from iucA to iroB. Concentric circles are sorted by alignment fraction to pLVPK sequence with the outermost ring representing the isolate with the highest alignment fraction. The average alignment fraction of pLVPK-like plasmid to pLVPK is higher in hvKp isolates compared to cKp. Alignment covering the virulence biomarkers is more extensive in hvKp isolates. Most cKp isolates lack or have non-functioning iroB, peg-344, and rmpA while possessing a rmpA2 and iucA.