Skip to main content
2024 BMES Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.

Bioinformatics, Computational and Systems Biology

Bioinformatics, Computational and Systems Biology - Poster Session E

Poster A8 - Validating in silico evolutionary predictions for various pharmacokinetic profiles and treatment adherences using an automated continuous-culture bioreactor

Saturday, October 26, 2024
10:00 AM - 11:00 AM EST
Location: Exhibit Hall E, F & G

Introduction:  The emergence of antibiotic resistance is a rapidly growing clinical crisis, complicated by the presence of different mutations, over-prescription, and patient non-adherence. Pharmacokinetic–pharmacodynamic (PKPD) models, which describe how drug concentrations change in a patient and the drug’s effect on pathogen growth, can be employed to better understand and predict the evolution of drug resistance. These models can be used alongside optimal control theory to design dosing regimens that minimize resistant infections [1]. Ideally, this optimal dosing is robust to variation in treatment adherence, which can further promote drug resistance.

In vitro validation is a critical step in translating these clinically optimal schedules but it remains challenging to simulate in vivo pharmacokinetic profiles. Modeling approaches are also not well integrated with continuous-culture devices that can perform the precise dosing and dilution required to simulate pharmacokinetic conditions [2]. We identify the need for a robust in silico model that synergises with our low-cost automated experimental culture system, the EVolutionary biorEactor (EVE) [3]. Our design criteria for the model include simulating pharmacokinetic conditions which can be experimentally replicated, designing optimal drug dosing protocols based on those conditions, and predicting the evolution of drug resistance under altered dose schedules.

In this work, we’ve modified FEArS (Fast Evolution on Arbitrary Seascapes), a software package for modeling biological evolution with empirical and simulated data [4]. Our changes to this software package allow us to predict the evolution of drug resistance for different dosing regimens and pharmacokinetic profiles as would be implemented in the EVE.

Materials and

Methods:  Our in silico FEArS model utilizes empirical data, including cefotaxime time kill assays [5] and dose-response curves for 16 E. coli genotypes (with combinations of 4 point mutations in the β-lactamase gene). This data parametrizes the growth and death rates in our stochastic FEArS agent-based model to simulate the evolution of particular E. coli genotypes as drug selection pressure changes [4].

We adapt FEArS to run evolutionary simulations using drug concentration curves and cell dilution methods that can be experimentally replicated by an automated continuous-culture bioreactor. Specifically, we simulate a pharmacokinetic drug curve created by drug and media being pumped in and waste being pumped out of a culture unit at certain time points (“Curve Modeled with EVE Pumps,” shown in orange in Fig. 1). This approximates a continuous pharmacokinetic curve to be used for evolutionary simulations (“Approximate Pharmacokinetic Curve,” shown in blue in Fig. 1). The drug, media, and waste pump schedules used to create the pharmacokinetic curve can be precisely replicated by the EVolutionary biorEactor (EVE).

Using the FEArS simulation results, we design an EVE experimental protocol with fully adherent and various dropped dose cefotaxime regimens to study the effect of missed doses on the survival of wild-type and drug-resistant mutant genotypes. We intend to validate the in silico model and simulation results by performing experiments with the same dosing regimens and conditions in the EVE. Validation of mutant fractions will be done via targeted sequencing of bacteria in the culture units at multiple time points.

Results, Conclusions, and Discussions: First, we design an adherent cefotaxime dosing regimen that can be implemented in the EVE (Fig. 2C) and demonstrate the elimination of the wild-type and two different mutant genotypes (Fig. 2B). Our model seeds single-mutants (mutant 1 and mutant 2) at lower cell count than the wild-type, as mutants arise later and typically have a fitness cost in drug-free conditions. In the adherent condition, mutant genotypes are eliminated 25 hours earlier than the wild-type genotype, with all populations going extinct by the end of the three day schedule (Fig. 2B).

We then compare this adherent cefotaxime dosing regimen simulation result to a dropped dose regimen generated with a 0.5 probability of “forgetting” each dose (Fig. 2E). In this dropped dose regimen simulation, both mutant 1 and mutant 2 survive, out-competing the wild-type population (Fig. 2D).

To understand the range of simulation outcomes that occur with intermediate probabilities (p_forget) of forgetting doses, we run 100 simulations for p_forget in the range 0 to 0.5 (Fig. 3). As p_forget increases, the wild-type and mutant genotypes show varying rates of increased survival, suggesting that mutant persistence occurs with non-adherence to optimal regimes. Within the 100 simulation results, the mutant genotypes begin to outcompete the wild-type genotype at different time points and reach varying cell counts (Fig. 4), depending on the exact doses missed.

Our model supports clinically-relevant dosing experiments and as an example we simulate an optimized three day cefotaxime dosing schedule. We find the wild-type and mutant genotypes are eliminated at full treatment adherence but survive with varying probability at lower adherence rates.

Our model software allows PKPD dynamics and evolutionary outcomes to be simulated in ways that can be validated in vitro. Upon validation, this probabilistic model can be extended to a range of pharmacokinetic profiles, dosing intervals, and treatment adherences, which would otherwise be extremely time-intensive to simulate in vitro. Our combined EVE-FEArS workflow reinforces the utility of the precise drug and media control capabilities of the EVE and highlights how synergistic in silico and in vitro testing of optimal control can better evaluate the risks of antibiotic non-adherence.

Acknowledgements (Optional):  Thank you to John Joyce, Rowan Barker-Clarke, PhD, Eshan King, PhD, and Jacob Scott, MD, DPhil for their collaboration and mentorship on this project and to Theory Division Laboratory members for their helpful discussion and support.


References:

[1] Witzany, C., Rolff, J., Regoes, R. R., & Igler, C. (2023). The pharmacokinetic–pharmacodynamic modelling framework as a tool to predict drug resistance evolution. Microbiology, 169(7), 001368. https://doi.org/10.1099/mic.0.001368

[2] Lee, H. H., Molla, M. N., Cantor, C. R., & Collins, J. J. (2010). Bacterial charity work leads to population-wide resistance. Nature, 467(7311), 82–85. https://doi.org/10.1038/nature09354

[3] Gopalakrishnan, V., Crozier, D., Card, K. J., Chick, L. D., Krishnan, N. P., McClure, E., Pelesko, J., Williamson, D. F., Nichol, D., Mandal, S., Bonomo, R. A., & Scott, J. G. (2022). A low-cost, open-source evolutionary bioreactor and its educational use. eLife, 11, e83067. https://doi.org/10.7554/eLife.83067

[4] King, E. S., Maltas, J., Weaver, D. T., Barker-Clarke, R., Pelesko, J., Dolson, E., & Scott, J. G. (2022). Fitness seascapes are necessary for realistic modeling of the evolutionary response to drug therapy (p. 2022.06.10.495696). bioRxiv [Preprint]. https://doi.org/10.1101/2022.06.10.495696

[5] King, E. S., Stacy, A. E., & Scott, J. G. (2024). A low-footprint, fluorescence-based bacterial time-kill assay for estimating dose-dependent cell death dynamics (p. 2024.03.08.584154). bioRxiv [Preprint]. https://doi.org/10.1101/2024.03.08.584154