Robin Michelet [1], Ana Rita Brochado [2,3], Elisabeta Cacace [2], Athanasios Typas [2,4], Sebastian G. Wicha [5], Charlotte Kloft [1]
1 Freie Universitaet Berlin, Institute of Pharmacy, Dept. of Clinical Pharmacy & Biochemistry, Berlin, Germany, 2 European Molecular Biology Laboratory, Genome Biology Unit, Heidelberg, Germany 3 Wuerzburg University, Chair of Microbiology, Wuerzburg, Germany 4 European Molecular Biology Laboratory, Structural & Computational Biology Unit, Heidelberg, Germany 5 University of Hamburg, Institute of Pharmacy, Dept. of Clinical Pharmacy, Hamburg, Germany
Objectives:
In order to tackle the current challenge of rising antimicrobial resistance, the combination of existing compounds is one of the alternative strategies to pursue. The analysis of high-throughput combinatorial data, where compounds are tested for their antibiotic activity in vitro, can be used to find these promising combinations. These data are typically analysed using deviations from Bliss Independence or Loewe Additivity to quantify synergy or antagonism1, which do not take directionality of the interaction into account, nor do they specify victim or perpetrator drugs. For this reason, a quantitative framework for the description of drug interactions, the General Pharmacodynamic Interaction (GPDI) model2 was recently implemented in a model selection and evaluation workflow and tested for 242 drug combinations in Gram-negative bacteria.3 In the current work, we present the further development and application of this workflow and how it was applied to a dataset of 1,711 combinations of antibiotics, human-use (non-antibiotic) drugs and other compounds in 3 Gram-negative and 86 combinations in 3 Gram-positive strains. We show how the workflow allows for quantitative interpretation of interactions, going beyond classically used scores and interesting combinations for further development can be identified in a semi-automated way.
Methods:
In a first step, the earlier developed workflow was applied to a dataset of 2,883 pairwise drug combinations in six Gram-negative strains4. Briefly, linear, power and Emax-type models were fitted to single concentration-effect data, after which their parameters were fixed and all possible interaction parameters for all possible model combinations were estimated using ELS regression. Based on the preliminary analysis of the full dataset, the workflow was optimized with respect to model selection and post-processing of the results. The amount of possible model combinations of was increased from 113 to 206 to allow for all possible symmetric and asymmetric interactions. Furthermore, a post-processing step to flag (non-) significant interactions was included. For this, the precision of the parameter precision of the single concentration-effect models was used as a threshold for the relative interaction effect, quantified as the relative impact of the GPDI parameter on its respective parameter at maximum normalized perpetrator concentration: INT/(EC50,AB+ 1) > RSEmono-effect. Lastly, final models and parameter estimates were combined in a repository for further analysis. This optimized workflow was then applied to a subset of pairwise combinations of 59 compounds (1,711 combinations) for the Gram-negative strains E. coli K-12 BW25113 and O8 IAI1, S. Typhimurium LT2 and 14028s, and P. aeruginosa PAO1 and PA14 and for 86 combinations in the Gram-positive strains S.aureus Newman and DSM20231, and B. subtilis. R (v. 3.6.0) and RStudio (1.2.1335) were used for dataset preparation, model selection, estimation and evaluation, and post-processing of the results.
Results:
Using the optimized workflow, the wide variability in interactions which was detected before3 diminished significantly. Indeed, the previous large amount of asymmetric interactions collapsed often towards mono-directional interactions with a non-significant contribution of an opposite interaction (e.g. a significant synergy in one direction with a small (GPDI-term < RSEmono-effect) antagonism. For the Gram-negative dataset, the total amount of significant interactions was found to be 21.7%, of which , 37.8% of the observed interactions were mono-directional synergistic, 50.1% mono-directional antagonistic, 15% bi-directional synergistic, 3.2% bi-directional antagonistic and 9.1% asymmetric. For the Gram-positive dataset, the amount of significant interactions was similarly found to be 24.4%, of which 47.8% was mono-directional synergistic, 19.6% mono-directional antagonistic, 12.5% bi-directional synergistic, 8.8% bi-directional antagonistic and 11.3% asymmetric.
Conclusions:
A previously developed robust workflow to apply the GPDI model to high-throughput data and select the most fitting model structure per combination was optimised further and applied to combinatorial data from both Gram-negative as Gram-positive bacteria. The interactions, if present, were quantified and classified depending on their directionality. The next steps are further clustering to group interactions according to their intensity and directionality and link this to antibiotic class and bacterial strain in order to inform mechanistic hypothesis generation.
Acknowledgements
This work was supported by grants of German Federal Ministry of Education and Research [031L0098B] in context of the Joint Programming Initiative on Antimicrobial Resistance project “COMBINATORIALS”.
References:
[1] Greco WR, Bravo G, Parsons JC. The search for synergy: a critical review from a response surface perspective. Pharmacol Rev. 1995;47(2):331-385.
[2] Wicha SG, Chen C, Clewe O, Simonsson USH. A general pharmacodynamic interaction model identifies perpetrators and victims in drug interactions. Nat Commun. 2017;8(1):2129. doi:10.1038/s41467-017-01929-y
[3] Michelet R, Brochado AR, Typas A, Wicha SG, Kloft C. A workflow for application of the general pharmacodynamic interaction model on high-throughput combinatorial data in order to identify, quantify and characterise drug combinations that can overcome multi-drug-resistance. In: PAGE. Abstracts of the Annual Meeting of the Population Approach Group in Europe. ; 2019. https://www.page-meeting.org/default.asp?abstract=8866.
[4] Brochado AR, Telzerow A, Bobonis J, et al. Species-specific activity of antibacterial drug combinations. 2018. doi:10.1038/s41586-018-0278-9
Reference: PAGE () Abstr 9510 [www.page-meeting.org/?abstract=9510]
Poster: Drug/Disease Modelling - Infection