Clinical pilot of bacterial transcriptional profiling as a combined genotypic and phenotypic antimicrobial susceptibility test

Abstract

Antimicrobial resistance is a growing health threat, but standard methods for determining antibiotic susceptibility are slow and can delay optimal treatment, which is especially consequential in severe infections such as bacteremia. Novel approaches for rapid susceptibility profiling have emerged that characterize either bacterial response to antibiotics (phenotype) or detect specific resistance genes (genotype). Genotypic and Phenotypic AST through RNA detection (GoPhAST-R) is a novel assay, performed directly on positive blood cultures, that integrates rapid transcriptional response profiling with the detection of key resistance gene transcripts, thereby providing simultaneous data on both phenotype and genotype. Here, we performed the first clinical pilot of GoPhAST-R on 42 positive blood cultures: 26 growing Escherichia coli, 15 growing Klebsiella pneumoniae, and 1 with both. An aliquot of each positive blood culture was exposed to nine different antibiotics, lysed, and underwent rapid transcriptional profiling on the NanoString platform; results were analyzed using an in-house susceptibility classification algorithm. GoPhAST-R achieved 95% overall agreement with standard antimicrobial susceptibility testing methods, with the highest agreement for beta-lactams (98%) and the lowest for fluoroquinolones (88%). Epidemic resistance genes including the extended spectrum beta-lactamase bla_CTX-M-15 and the carbapenemase bla_KPC were also detected within the population. This study demonstrates the clinical feasibility of using transcriptional response profiling for rapid resistance determination, although further validation with larger and more diverse bacterial populations will be essential in future work. GoPhAST-R represents a promising new approach for rapid and comprehensive antibiotic susceptibility testing in clinical settings.IMPORTANCEExposure to antibiotics causes differential transcriptional signatures in susceptible vs resistant bacteria. These differences can be leveraged to rapidly predict resistance profiles of Escherichia coli and Klebsiella pneumoniae in clinically positive blood cultures.

Keywords: antimicrobial resistance; bacteremia; diagnostics; transcriptional regulation.

Fig 1

Assay workflow and summary of transcriptional profiles in the study population. Transcriptional response to antibiotic exposure differs by susceptibility classification. (A) Schematic workflow for an individual sample. (B) Clinical isolates show predictable transcriptional response to antibiotic treatment based on susceptibility profile. The absolute value of log₂-fold change of all transcript counts of genes targeted by the probeset upon antibiotic exposure averaged across isolates, subset by antibiotic, and grouped according to clinical resistance profile (susceptible, intermediate, and resistant) and species (E. coli and K. pneumoniae). Antibiotics are abbreviated as per Table 1 and grouped by class. The color of each bar represents the average log₂-fold change in transcript counts, while the height of each bar corresponds to the number of samples with that susceptibility profile for each antibiotic.

Fig 2

SPD calculation from transcriptional response and distribution. The transcriptional profile is distilled into a measure of treatment response with SPD, a single-value summary of the transcriptional response across the genes of interest (see Materials and Methods). (A and B) Heatmaps of the log₂-fold change in the transcriptional response of each (A) E. coli or (B) K. pneumoniae isolate after exposure to ceftriaxone. Each row corresponds to a different gene of interest, while each column is a different blood culture sample. KPC and CTX-M-15 genotypes are shown above the heatmap, while sample numbers, SPD values, and MICs (µg/mL) are shown below. (C) Box-and-whisker plot of SPD values for each species and antibiotic combination, grouped by clinical susceptibility profile. Antibiotic abbreviations are listed in Table 1.

Fig 3

Predictions of drug susceptibility by SVM modeling and genotype. SVM thresholds for each drug class and bacterial species are shown by the dotted line. SVM is a model to find the optimal separation between two classes of data (see Materials and Methods). Each point corresponds to an SPD value (y-axis) and is shaped according to gene content and colored according to whether the SVM algorithm made the correct resistance assignment. Samples are grouped according to clinical susceptibility classification (x-axis).