Charlotte Thomas 1, Hiie Soeorg, John Readman 1, Dagan Lonsdale 2, Joseph Standing 1
1 University College London (, United Kingdom), 2 City St. George's, University of London (, United Kingdom)
Objectives: We aimed to characterise antimicrobial activity and quantify drug interactions in Gram-negative bacteria by applying a range of approaches to checkerboard assay data across clinically relevant antibiotic combinations.
Methods: Minimum inhibitory concentration (MIC) testing was conducted across 2 E. coli (ATCC 25922, DWEC107) and 2 K. pneumoniae (DWKC01, JRKC01) strains under amikacin – meropenem and amikacin – piperacillin/tazobactam combinations, and 4 A. baumannii strains (ATCC 19606, DWAB210, DWAB211, DWAB19) across polymyxin B – fusidic acid. Checkerboard assays were undertaken, with 3 technical replicates, measuring optical density (600nm) every 10 minutes for 20 hours.
Interactions were initially quantified using the Fractional Inhibitory Concentration Index (FICI). A “merge-first” approach was undertaken to minimise experimental noise across the three technical replicates. For every unique concentration pair, the mean inhibition and standard deviation were calculated. Dose-response surface fitting was undertaken to fit a single representative dose-response surface based on these averaged inhibition values, providing a stabilised landscape for synergy assessment. The relative EC₅₀ for each monotherapy was calculated by fitting a four-parameter logistic model. Further synergy scores were calculated using a range of surface response models (Bliss Independence, Loewe Additivity, Highest Single Agent (HSA), and Zero Interaction Potency (ZIP)) and normalised for comparison. A mechanistic approach (General Pharmacodynamic Interaction Model (GPDI)) categorised the magnitude, direction and mechanism of interactions[3].
All analyses and visualisations were performed using R (v4.5.0), synergyfinder (v3.10.3)[2] and stats (v4.5.1).
Results: Using EUCAST clinical breakpoints[1], the E. coli reference strain (ATCC 25922) was susceptible to amikacin, meropenem, and piperacillin/tazobactam. The second E. coli strain (DWEC107) was resistant to amikacin and piperacillin/tazobactam but remained susceptible to meropenem. Both K. pneumoniae strains were susceptible to meropenem, but resistant to amikacin and piperacillin/tazobactam. Among A. baumannii, three strains were susceptible to polymyxin B with high fusidic acid MICs, while DWAB219 was resistant to polymyxin B, but had a substantially lower MIC for fusidic acid.
One K. pneumoniae strain (JRKC01) could not be evaluated by FICI (amikacin MIC > 22,000 mg/L). E. coli and K. pneumoniae showed additive FICI scores across both amikacin – meropenem and amikacin – piperacillin/tazobactam. 3/4 A. baumannii strains had synergistic FICI scores under polymyxin B – fusidic acid.
Despite additive FICI scores, the amikacin–meropenem combination exhibited localised concentration-
dependent synergy hotspots (scores ≥ 10/100) in both E. coli and K. pneumoniae. For E. coli ATCC 25922, synergy hotspots were consistently detected by the Bliss, HSA, and ZIP models; no synergy hotspots were observed for E. coli DWEC107. In K. pneumoniae, both strains demonstrated synergy across Bliss, HSA, and ZIP. The amikacin–piperacillin/tazobactam combination exhibited concentration-dependent synergy hotspots in both E. coli and K. pneumoniae across all strains. Polymyxin B – fusidic acid showed clear, concentration-dependent synergy across A. baumannii, with the highest scores across all combinations, consistent with FICI.
GPDI models had a good fit (R2 = 0.94-1.0, RMSE = 0.023-0.72) and EC₅₀ values were consistent with those previously calculated. Amikacin-meropenem showed weak bidirectional synergy under the ATCC 25922 E.coli strain. E.coli DWEC017 showed modest synergy in the meropenem to amikacin direction, but antagonism in the opposite direction, aligning with previous models. In K. pneumoniae strains, this combination showed consistent bidirectional synergy. The amikacin-piperacillin/tazobactam combination displayed synergistic behaviour against E. coli and K. pneumoniae strains, with amikacin generally producing a stronger reduction in the EC₅₀ of piperacillin/tazobactam.
Polymyxin B and fusidic acid also demonstrated consistent synergy in A. baumannii, with the strongest effect observed in the reduction of fusidic acid’s EC₅₀ by polymyxin B.
Conclusions: Response-surface models (Bliss/HSA/ZIP/Loewe) localised synergy hotspots that FICI alone missed, highlighting synergy hotspots across drug combinations. Loewe additivity showed synergy only in the strongest of interactions; this behaviour is expected because Loewe presumes dose additivity, an assumption less appropriate for agents with distinct primary mechanisms. The GPDI converted surfaces into mechanistic, directional parameters, quantifying potency shifts and explaining inter-strain heterogeneity. Collectively, these findings demonstrate that antibiotic interactions are highly concentration-dependent and strain-dependent, highlighting the limitations of single-point interaction metrics. Future work will involve repeating this methodology across Time-Kill and Hollow Fibre Infection Model experiments to explore results across experiment types.
References:
[1] European Committee on Antimicrobial Susceptibility Testing. Data from the EUCAST MIC distribution website. https://mic.eucast.org/search/. Accessed: February 2, 2026. 2026.
[2] Aleksandr Ianevski, Anil K Giri, and Tero Aittokallio. “SynergyFinder 3.0: an interactive analysis and consensus interpretation of multi-drug synergies across multiple samples”. en. In: Nucleic Acids Res. 50.W1 (July 2022), W739–W743.
[3] Sebastian G. Wicha et al. “A general pharmacodynamic interaction model identifies perpetrators and victims in drug interactions”. In: Nature Communications 8.1 (2017), p. 2129. doi: 10.1038/s41467-017-01929-y.
Reference: PAGE 34 (2026) Abstr 11946 [www.page-meeting.org/?abstract=11946]
Poster: Drug/Disease Modelling - Other Topics