Improving Rose Bengal Photodynamic Antimicrobial Therapy efficacy by validating predictive model

Abstract

Purpose : Antimicrobial resistant and atypical infectious keratitis has been effectively treated with experimental Rose Bengal Photodynamic Antimicrobial Therapy (RB-PDAT). RB-PDAT creates antimicrobial singlet oxygen (¹O₂) by exciting Rose Bengal (RB) photosensitizer with green light. A proof-of-concept chemistry kinetics model was previously developed (Peterson et al., 2021) to predict ¹O₂ distribution within the cornea during RB-PDAT for a given set of input parameters (light dose, RB concentration, and application time). Although developed using experimental data, the model had not yet been validated using direct ¹O₂ dose measurement. Validation and fitting of this model will enable optimization of ¹O₂ dose by optimizing RB-PDAT treatment parameters.

Methods : 4 groups of 3 donor eyes were treated with different RB concentrations (0.054, 0.12, 0.58, 1.2 mM). Donor eyes were treated with 20% dextran for 24 hours to de-swell the cornea and then treated with RB for 30 min. A ¹O₂ dosage measurement system previously developed for RB-PDAT (Peterson et al., 2021) was used to measure ¹O₂ generated during green light stimulation (at 10⁷ mW/cm²) before and after RB treatment. Measured ¹O₂ signal was normalized and compared with normalized ¹O₂ level predicted by chemical kinetics model for the corresponding RB concentrations at 6 and 12 mW/cm² at 525 nm (irradiances typical for RB-PDAT used to compare because solver is unstable beyond 30 mW/cm²). All values normalized to respective ¹O₂ value at 1.2 mM. The model was solved numerically in MATLAB.

Results : Predicted normalized ¹O₂ values were within a standard deviation of measured ¹O₂ signal at every RB concentration except at RB concentration of 0, where a non-zero ¹O₂ signal was observed. Comparing normalized predicted ¹O₂ signal values with mean normalized measured ¹O₂ dosimeter signal showed relative error ranging from 9.5% to 80.9% and 10.6% to 62.6% at 6 and 12 mW/cm² respectively. Highest error is seen at 0.054 mM for all conditions when excluding 0 mM RB (100% error as model assumes no ¹O₂ production).

Conclusions : We show a proof-of-concept predictive model for ¹O₂ dose generated during RB-PDAT, capable of testing a range of experimental parameters which can be used for optimizing RB-PDAT clinical efficacy. While limited due to variability of experimental data, early results demonstrate reasonable correlation between measured and predicted ¹O₂ levels.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.