Importance of modelling hERG binding in predicting drug-induced action potential prolongations for drug safety assessment

Figures

• 

  FIGURE 1. (A) The six-state model by Li et al. (2016) (in black) and its extension to model drug block and trapping by Li et al. (2017) (black and blue states). (B) Fractional block predictions for 10 paces of Milnes’ protocol (top) for the SD model with a trapped drug, dofetilide (middle) and a non-trapped drug, verapamil (lower). The SD model’s response (blue) is overlaid on the experimental data (orange) from Li et al. (2017). (C) Steady state response (after 1,000 paces) for the SD model during a single step of Milnes’ protocol, for a drug-free (left), trapped (middle), and non-trapped configuration (right). The lower row shows the state occupancy: the fraction of channels in any one state at a given time. (D) Like panel C, but using a predetermined elongated AP signal (1 s) instead of a rectangular pulse (25 s). For comparison, the grey lines in each column of panels C and D show the data from the other columns. Note that the labelling of the IC and C states, adapted from the original model, does not always correspond to their respective physical states.

• 

  FIGURE 2. A schematic of the process of fitting the CS model to the SD model. First, simulations are run with the SD model and Milnes’ protocol at various drug concentrations. The peak of the hERG currents are extracted and normalised, and a Hill curve is fitted to the normalised data points. This Hill curve is then used to scale the ionic conductance of the CS model. Finally, the simulated APs of both models are compared.

• 

  FIGURE 3. The AP-CS model can provide a reasonable approximation of the AP-SD model for an example drug T. (A) The matching of the peak hERG currents simulated by both drug block models under Milnes’ protocol (top row) at various drug concentrations of an example drug T. (B) The AP (first row) and its hERG current (second row) for the AP-SD model and the AP-CS model at steady state. (C) The APD90 values of the AP-SD model and the AP-CS model under the effect of example drug T. APs that show EAD-like behaviour are indicated with an asterisk. (D) The qNet values of the AP-SD model and the AP-CS model under the effect of example drug T. qNet values of APs with EAD-like behaviours are not shown.

• 

  FIGURE 4. The AP-CS model is less successful at replicating the model behaviour of the AP-SD model for an example drug N. (A) The comparison of the I Kr simulated by both the SD model and the CS model after calibration of the hERG ionic conductance for an example drug N. The top row shows Milnes’ protocol used to stimulate the hERG channel models. (B) The AP (first row) and its hERG current (second row) for the AP-SD model and the AP-CS model at steady state. (C) The APD90s of the AP-SD model and the AP-CS model under the effect of an example drug N. APs that show EAD-like behaviour are indicated with an asterisk. (D) The qNet values of the AP-SD model and the AP-CS model under the effect of example drug N. qNet values of APs with EAD-like behaviours are not shown.

• 

  FIGURE 5. The accumulation of drug compounds at the channel due to being trapped is well-observed in the transient phase. Example drug T shows a longer transient phase than an example drug N in the SD model. The first six pulses of AP (top row) and its hERG current (bottom row) for the AP-SD model after the addition of (A) an example drug T and (B) an example drug N. The vertical black dashed lines indicate the addition of drug into the system after 1,000 pulses of pacing. The first six pulses of the AP (top row) and the hERG current (bottom row) for the AP-CS model after the addition of (C) example drug T and (D) example drug N, of which its hERG ionic conductance is scaled to model the drug effect. (E) The APD90 values of both the AP-SD model and the AP-CS model are compared at 100 and 200 nM of example drug T. (F) The APD90 values of both the models are compared at 1,000 nM and 104 nM of example drug N.

• 

  FIGURE 6. The Hill curves for an example drug T and an example drug N are protocol dependent, thus affecting the resultant APD prolongation. (A) The four protocols used to calibrate the hERG ionic conductance from the SD model. Note the differences in the time axes for visualisation purposes. The Hill curves of the peak I Kr simulated from the SD model under the protocols given in panel A for (B) an example drug T and (C) an example drug N. (D) The APD90s of the AP-CS model under the effect of an example drug N with its hERG conductance scaled based on the Hill curves shown in panel C.

• 

  FIGURE 7. The AP-SD model is most likely to return higher APD90s than the AP-CS model. (A) The APD90 differences for combinations of V half−trap, K max and K u parameters. The color of the markers indicate the signed RMSD of each virtual drug in the parameter space. (B) The grey circles are parameter value combination where the signed RMSD is between −30 and 30 ms. The triangles are the synthetic drugs taken from Li et al. (2017), color coded with their signed RMSD value. These triangles are projected to the K max—K u plane as red circles for better visualisation. Some data points are missing due to numerical issues with these parameter combinations.