Eva Wehrhahn1, Chenyan Zhao1, Pikkei Wistrand-Yuen2, Pernilla Lagerbäck2, Thomas Tängdén2, Elisabet Nielsen1
1Department of Pharmacy, Uppsala University, 2Department of Medical Sciences, Uppsala University
Introduction/Objectives: Carbapenem-resistant Gram-negative bacteria such as Enterobacterales and Pseudomonas aeruginosa are considered critical or high priority pathogens by the updated WHO bacterial priority pathogens list [1]. Understanding additional facets of antibiotic effects may aid in fighting these pathogens. Carbapenems have been shown to induce morphological changes in bacteria [2], but the relationship between morphology changes and drug concentration is not well known. Bacterial filaments are of particular concern as they may increase the risk of resistant progeny or cause rapid regrowth by dividing at a higher rate than normal cells [3,4]. Time-lapse imaging of bacteria can provide rich and rapid bacterial count data, while traditional time-kill experiments that require counting of colony forming units (CFUs) are laborious, relatively slow, and do not provide information on the individual bacterial level. Although direct imaging of bacteria can be complicated by morphological changes, such as filamentation and cell fading, a method that accounts for these changes could provide unique, in-depth information on the effect of antibiotics. This study aimed to develop an image-based methodology that uses time-lapse microscopy data of carbapenem-exposed Gram-negative bacteria to inform pharmacokinetic-pharmacodynamic (PKPD) models which describe bacterial morphology and count changes. Methods: Individual bacteria were identified and measured in time-lapse microscopy (oCelloScope) (5) images of E. coli exposed to ertapenem and P. aeruginosa exposed to meropenem at concentrations ranging from 0-16 x the minimum inhibitory concentration (MIC) for up to 24 hours using a CellProfiler [6] image analysis pipeline. A total of 3396 E. coli and 2695 P. aeruginosa images were evaluated. Bacterial counts from the image analysis pipeline were compared to manual counts from the same images and CFU counts from the same experiments. Identified bacteria were classified into morphological categories using a machine learning model and bacterial counts of each morphological class were obtained for each timepoint. Pharmacokinetic-pharmacodynamic (PKPD) models were built in NONMEM [7] to describe the relationship between drug concentration and bacterial morphology and count. The PKPD models were evaluated using Visual Predictive Checks. Results: An image analysis pipeline was developed, where the inter- and intra-observer agreement for manual total bacterial counts per image (84-87%) was similar to the agreement with the automated pipeline (82-86%). The identified morphology classes were healthy, filament, bulge, and faded. The morphology classifiers obtained 87% accuracy with a minimum F1 score of 0.81 per class. The faded class was excluded from PKPD modelling due to inclusion of background objects. A compartmental PKPD model was developed, consisting of healthy (susceptible and resistant), filament (active and resting), and bulge (dying and resting) bacteria. The presence of resistant bacteria for P. aeruginosa was modelled using a mixture model. The two datasets had a shared structural model but differed in rate parameterization, which included either fixed rate constants, or concentration-dependent rate constants, according to sigmoidal Emax, and linear functions. Conclusions: The image-analysis pipeline was able to identify bacteria from time-lapse images, with bacterial morphology identified using a machine learning model. The generated PKPD models were able to describe the bacterial morphology changes and death in relation to antibiotic concentration. This framework for evaluating antibiotic effect on bacteria has the potential to efficiently provide more detailed information regarding the concentration-effect relationship for both bacterial killing as well as potentially undesirable morphological changes such as filamentation.
[1] WHO Bacterial Priority Pathogens List 2024: Bacterial Pathogens of Public Health Importance, to Guide Research, Development, and Strategies to Prevent and Control Antimicrobial Resistance. 1st ed. Geneva: World Health Organization; 2024. 1 p. [2] Cushnie TPT, O’Driscoll NH, Lamb AJ. Morphological and ultrastructural changes in bacterial cells as an indicator of antibacterial mechanism of action. Cell Mol Life Sci CMLS. 2016 Dec;73(23):4471–92. [3] Bos J, Zhang Q, Vyawahare S, Rogers E, Rosenberg SM, Austin RH. Emergence of antibiotic resistance from multinucleated bacterial filaments. Proc Natl Acad Sci U S A. 2015 Jan 6;112(1):178–83. [4] Cayron J, Dedieu-Berne A, Lesterlin C. Bacterial filaments recover by successive and accelerated asymmetric divisions that allow rapid post-stress cell proliferation. Mol Microbiol. 2023 Feb;119(2):237–51. [5] oCelloScope – Automated Microscopy – Live Cell Imaging – BioSense Solution [Internet]. BioSense Solutions. [cited 2025 March 07]. Available from: https://biosensesolutions.dk/overview/ [6] Stirling DR, Swain-Bowden MJ, Lucas AM, Carpenter AE, Cimini BA, Goodman A. CellProfiler 4: improvements in speed, utility and usability. BMC Bioinformatics. 2021 Sep;22(1):433. [7] Bauer RJ. NONMEM Tutorial Part I: Description of Commands and Options, with Simple Examples of Population Analysis. CPT Pharmacomet Syst Pharmacol. 2019 May;
Reference: PAGE 33 (2025) Abstr 11745 [www.page-meeting.org/?abstract=11745]
Poster: Drug/Disease Modelling - Infection