GENOMIC CHARACTERIZATION OF ANTIMICROBIAL RESISTANCE IN Pseudomonas aeruginosa

Abstract

Antimicrobial resistance (AMR) is one of the most serious healthcare problems in the twenty first century. According to data from the World Health Organization (WHO), AMR directly caused 1.27 million deaths and indirectly contributed to 4.95 million deaths in 2019. It is projected that without significant interventions, the number of deaths may rise to approximately 10 million annually by 2050. Pseudomonas aeruginosa, belonging to the ESKAPE group (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacter spp.), has been classified as a critical priority pathogen by the WHO and U.S. Centers for Disease Control and Prevention. It is responsible for 10-16% of hospital-acquired infections and has an inherently complex resistance architecture due to the acquisition of new resistance determinants through horizontal gene transfer. While studies have used whole genome sequencing (WGS) for AMR surveillance, most remain limited to listing the resistance genes alone. It is therefore necessary to examine how these genes are organized and co-selected. A major factor contributing to the spread of resistance is the association of the resistance genes with mobile genetic elements (MGEs), which is yet to be fully understood. This dissertation focuses on the development and application of a reproducible bioinformatics framework for the integrated characterization of the resistance genes and MGEs in clinical P. aeruginosa isolates. Whole genome sequences of ten clinical isolates representing pneumonia, bloodstream infections, and urinary tract infections from geographically diverse sources were retrieved from publicly available databases. Raw paired-end reads were processed through a quality-controlled assembly pipeline using FastQC, MultiQC, fastp, and SPAdes, followed by AMR gene detection using NCBI AMRFinderPlus. MGEs belonging to four categories, including insertion sequence (IS) elements, prophage regions, integrons, and plasmid replicons, were detected using ISEScan, PHASTER, IntegronFinder, and PlasmidFinder, respectively. All downstream analysis was implemented in Python, enabling transparent and reproducible execution. Across the ten isolates, 69 unique resistance determinants including both acquired resistance genes (85%) and chromosomal point mutations (15%) were identified. Three genes aph(3’) IIb, catB7, and fosA, were universally conserved. AMR gene burden varied significantly across isolates, with the highest burden isolate carrying 32 genes across multiple resistance classes. Co-occurrence network analysis was performed using a binary gene presence-absence matrix, revealing a structured resistome comprising 20 genes connected by 91 edges. The three core genes each showed maximum degree centrality. The crpP gene, encoding a ciprofloxacin resistance enzyme, exhibited disproportionately high connectivity relative to its prevalence, vi

indicating strong co-selection pressure. Hierarchical clustering using Jaccard distance identified high, moderate, and low burden groups, indicating non-random multi-class acquisition patterns. Shannon entropy indices ranged from 1.79 to 3.47 across isolates, with higher values in high-burden isolates. These findings suggest that resistance is distributed broadly across drug classes, and this has direct relevance to the treatment limitations faced in clinical settings. Mobilome characterization revealed 297 IS elements, 15 intact prophage regions, integron associated attC sites in six isolates, and plasmid replicons confined to the highest burden isolate. A composite MGE burden score integrating all four element types showed a statistically significant positive correlation with AMR gene count (Pearson’s r = 0.762, R2 = 0.580, p = 0.010). The values show a strong association between genomic plasticity and resistance accumulation within this population. The findings of this study demonstrate that AMR in P. aeruginosa is not merely a collection of independent genetic events, but rather a structured, network-organized, and MGE-associated phenomenon. The framework utilized here generated quantitative outputs for each isolate, including network centrality scores, Shannon diversity indices, hierarchical cluster assignments, and MGE burden profiles, which are directly applicable to AMR surveillance. When applied to large sample sizes, these results may serve as a structured feature set for future machine learning-based resistance phenotype classification and outbreak risk prediction. This work contributes to the growing foundation of AI-driven infectious disease surveillance by bridging WGS-based genomic profiling with the computational framework required for large scale clinical application.