Interferometric Antibiotic Susceptibility Testing

Antimicrobial resistance is one of the greatest threats to modern medicine. Globally, it is responsible for the loss of millions of lives and billions of dollars every year, and this problem has continued to worsen. Added selective pressure from increased antibiotic use in hospitals, in agriculture, and at home are generally blamed for the rapidly increasing rate of evolution. Combine this with our inability or unwillingness to develop novel, non-derivative antibiotics for more than three decades, and we can see the result. We must find sustainable ways to address the complex problem that antimicrobial resistance presents. One way to do this is to find ways to use our current, and any new, antibiotics more effectively. This can be done through faster, more detailed antibiotic susceptibility testing. In this thesis we propose a novel method of antibiotic susceptibility testing using white light interferometry which provides single nanometer out-of-plane resolution that we use to measure bacterial colonies grown on agar pads. First, we examined the current clinical standard susceptibility phenotypes, resistant and susceptible, using nearly 300 isolate-antibiotic pairs including 7 clinically relevant antibiotics and nearly 135 distiinct clinical isolates from the Enterobacter, Escherichia, Klebsiella, and Pseudomonas genera. After incubating at 37C for just four hours on Mueller Hinton agar mixed with the clinical breakpoint concentration of the antibiotic, we measured their topographies using coherence scanning interferometry. From these topographies, we extracted 11 biophysically relevant and relatively simple features. We trained machine learning classifiers on these features using the leave one out cross validation approach and found that we can accurately determine 98.3% of the susceptibility phenotypes with only a 1.3% very major error rate, 0.4% major error rate, and a 1% test inconclusivity rate -- all within FDA requirements. Next, in order to address a contributor to a more than 10% antibiotic treatment failure rate, even when tested for susceptibility, we developed a method to detect monoclonal heteroresistant isolates. Heteroresistance is the presence of a distinct, repeatable, and often unstable subpopulation of cells within a larger population that shows significantly stronger phenotypic resistance to an antibiotic. It is common, but often overlooked in the lab and the clinic due to its instability and the labor intense methods of testing for it. Here, we show that the detection efficiency of this phenotype is highly dependent on the antibiotic and the mode of action of the resistance to that antibiotic. Using machine learning on biophysically relevant features from their topographies, we found that for 71 Enterobacteriaceae isolates (31 were heteroresistant) and a Beta-lactam antibiotic combination, piperacillin-tazobactam, to which the resistance mechanism is generally the production of Beta-lactamase enzymes, we accurately detected generally low rates of this phenotype 68% of the time when compared to susceptible isolates, with an overall accuracy of 80%. We then tested 10 isolates with the aminoglycoside antibiotic, gentamycin, to which the main mechanisms of resistance are ribosomal mutation or drug efflux. With 1 resistant, 4 heteroresistant, and 5 susceptible isolates tested in triplicate, we were able to accurately determine 100\% of the susceptibility phenotypes just by measuring the height of one region of the topography. In sum, we have successfully developed a method for antibiotic susceptibility testing using bacterial colony topographies using white-light interferometry. This method is rapid and sensitive to nontraditional, though highly relevant, phenotypes. More work must be done to test more, different isolates and relevant antibiotics as well as expand the capabilities of this technology for further microbiological diagnostic needs. The fight against antimicrobial resistance is multi-faceted, but novel susceptibility testing methods, like the one proposed here, will be key to implementing a sustainable solution.

Date

2024-07-01

Resource Type

Text

Resource Subtype

Dissertation

Full item page