Quantifying fluoroquinolone resistance-inducing point mutations from metagenomics data

Abstract

Bacterial resistance to fluoroquinolones primary arises from chromosomal point mutations in the gyrA and parC genes. Currently, bioinformatics tools designed for measuring resistance from metagenomics data are not able to detect resistance arising from point mutations. To address this gap, two methods for quantifying fluoroquinolone resistance-inducing SNPs from metagenomic reads were created.GyrAPointCounter is a software which can analyze the presence and abundance of point mutations present on gyrA that cause resistance and infer the relative proportion of reads carrying potential resistance causing sequences vs wildtype sequences. Additionally, we developed a classification tree machine learning model, which was trained using the physicochemical proprieties of amino acids. Both methods form their decision rules by using the available sequences and the mutational patterns of fluoroquinolone-resistant GyrAs present in the CARD database. The two methods were validated using an Escherichia coli (E. coli) WGS dataset (n = 201) and an external dataset comprised of gyrA sequences (n = 40) belonging to species distinct from those present in the training data. The results of the analyses show that both methods display excellent concordance with the phenotypic data for E. coli sequences. The classification performance for novel species for GyrAPointCounter and the supervised learning model showed a True Positive Rate (TPR) of 0.87 and 1 and True Negative Rate (TNR) of 0.75 and 0.68 respectively. An in-house shotgun time series metagenomomics datatset containing Illumina short-reads from farm animals treated with enrofloxacin was submitted for the analysis with our two methods. The enrofloxacin-treated samples displayed higher average resistance levels compared to the control groups for both methods. Conclusions: This research introduced the first version of GyrAPointCounter, which is a promising tool for monitoring the resistance levels from metagenomics data. However, stricter validation is needed before confidently evaluating the tool’s performance. For the current version, we propose using the tool as a relative quantification method, rather than absolute.