# P-1796. A dual-mechanism model incorporating efflux pumps and resistance genes to predict multi-drug resistance in Klebsiella pneumoniae

## TL;DR

This study improves prediction of drug resistance in Klebsiella pneumoniae by combining analysis of resistance genes and efflux pump regulators using genomic data.

## Contribution

A novel dual-mechanism model integrating efflux pump regulation and resistance genes to better predict multi-drug resistance in K. pneumoniae.

## Key findings

- A model combining ARGs and efflux pumps achieved 89% accuracy in predicting gentamicin resistance.
- Incorporating efflux pump regulatory sequences significantly improved AMR prediction sensitivity.
- The study identified genomic signatures associated with elevated efflux pump activity in K. pneumoniae.

## Abstract

Antimicrobial resistance (AMR) is a global threat projected to become a leading cause of death by 2050. Persistent selective pressure from antimicrobial overuse has driven the adaptation of resistant strains. While many AMR pathways are mediated by the presence of AMR genes (ARGs), efflux pumps (EPs) represent an often overlooked contributor to multi-drug resistance (MDR). Enterobacterales such as Klebsiella pneumoniae contain chromosomal EPs and are identified as a critical threat on the WHO Bacterial Priority Pathogens List because of growing rates of MDR. Our goal is to leverage EP regulatory sequences with known ARG markers to better predict AMR from urinary tract infection (UTI) clinical isolates using next-generation sequencing.

We collected approximately 3000 genomic sequences with known phenotypes from the Bacterial and Viral Bioinformatics Resource Center (BV-BRC), identifying common ARGs with AMRFinder-plus. EPs and global regulators were characterized by aligning to a custom set of EP operons and global regulatory elements derived from type strains. Variant enrichment analysis was conducted to identify mutations in EP regulatory sequences that are significantly associated with resistant phenotypes. A machine learning model was developed and trained using the ARGs and enriched EP variants, then cross-validated using 369 clinical UTI isolates from our in-house strain collection.

We examined aminoglycoside resistance in K. pneumoniae, typically mediated by ARGs such as aminoglycoside-modifying enzymes and 16S rRNA methyltransferases. A model including only ARGs predicted gentamicin resistance with 73% accuracy. Adding major efflux pumps (AcrAB-TolC, AcrD, OqxAB) and global regulators (SoxS, MarA, BaeSR) improved accuracy to 89%, suggesting a synergistic role of ARGs and efflux systems in resistance.

Incorporating EPs and regulatory sequences into AMR predictive models substantially increased sensitivity beyond ARGs alone. We identified signatures which might contribute to elevated EP capacity, and might be important genomic factors in resistance. We are continuing to adapt similar methods for additional resistance types found in K. pneumoniae, including fluoroquinolones and extended spectrum beta lactamases (ESBLs).

Gabor Fidler, PhD, Biotia Inc: Employee Mara Couto-Rodriguez, MS, Biotia: Employee Heather L. Wells, MPH, PhD Candidate, Biotia: Employee Sol Rey, BS, Biotia: Employee Tiara Rivera, B.S., Biotia: Employee John C. Papciak, BS, Biotia: Employee Ford Combs, PhD, Biotia: Employee Caitlin Otto, PhD, D(ABMM), Biotia: Employee Lorenzo Uccellini, PhD, Biotia: Employee Christopher E. Mason, PhD, Biotia: Board Member Niamh B. O'Hara, PhD, Biotia: Employee Dorottya Nagy-Szakal, MD PhD, Biotia: Employee David C. Danko, Ph.D., Biotia: Employee