Assessing drug safety by identifying the axis of arrhythmia in cardiomyocyte electrophysiology

Reviewer #1 (Public Review):

Summary:
Heitmann et al introduce a novel method for predicting the potential of drug candidates to cause Torsades de Pointes using simulations. Despite the fact that a multitude of such methods have been proposed in the past decade, this approach manages to provide novelty in a way that is potentially paradigm-shifting. The figures are beautiful and manage to convey difficult concepts intuitively.

Strengths:
(1) Novel combination of detailed mechanistic simulations with rigorous statistical modeling

(2) A method for predicting drug safety that can be used during drug development

(3) A clear explication of difficult concepts.

Weaknesses:
(1) In this reviewer's opinion, the most important scientific issue that can be addressed is the fact that when a drug blocks multiple channels, it is not only the IC50 but also the Hill coefficient that can differ. By the same token, two drugs that block the same channel may have identical IC50s but different Hill coefficients. This is important to consider since concentration-dependence is an important part of the results presented here. If the Hill coefficients were to be significantly different, the concentration-dependent curves shown in Figure 6 could look very different.

(2) The curved lines shown in Figure 6 can initially be difficult to comprehend, especially when all the previous presentations emphasized linearity. But a further issue is obscured in these plots, which is the fact that they show a two-dimensional projection of a 4-dimensional space. Some of the drugs might hit the channels that are not shown (INaL & IKs), whereas others will not. It is unclear, and unaddressed in the manuscript, how differences in the "hidden channels" will influence the shapes of these curves. An example, or at least some verbal description, could be very helpful.

Reviewer #2 (Public Review):

Summary:
In the paper from Hartman, Vandenberg, and Hill entitled "assessing drug safety, by identifying the access of arrhythmia and cardio, myocytes, electro physiology", the authors, define a new metric, the axis of arrhythmia" that essentially describes the parameter space of ion channel conductance combinations, where early after depolarization can be observed.

Strengths:
There is an elegance to the way the authors have communicated the scoring system. The method is potentially useful because of its simplicity, accessibility, and ease of use. I do think it adds to the field for this reason - a number of existing methods are overly complex and unwieldy and not necessarily better than the simple parameter regime scan presented here.

Weaknesses:
The method described in the manuscript suffers from a number of weaknesses that plague current screening methods. Included in these are the data quality and selection used to inform the drug-blocking profile. It's well known that drug measurements vary widely, depending on the measurement conditions.

There doesn't seem to be any consideration of pacing frequency, which is an important consideration for arrhythmia triggers, resulting from repolarization abnormalities, but also depolarization abnormalities. Extremely high doses of drugs are used to assess the population risk. But does the method yield important information when realistic drug concentrations are used? In the discussion, the comparison to conventional approaches suggests that the presented method isn't necessarily better than conventional methods.

In conclusion, I have struggled to grasp the exceptional novelty of the new metric as presented, especially when considering that the badly needed future state must include a component of precision medicine.

Author Response

We thank the reviewers for their thoughtful suggestions, which we will address in the revised manuscript.

Briefly, we purposely fixed the Hill coefficients to h=1 on the grounds that one drug molecule binding to the channel is sufficient to block the channel and there is no strong evidence for co-operative binding in the literature. Doing so also helped to constrain the degrees of freedom in the face of noisy observations in the public datasets. As noted by Reviewer 2, the quality of the drug measurements varies widely across laboratories and this is particularly noticeable in estimates of Hill coefficients which are therefore less reliable.

The dose-dependent curves of multi-channel block (Figure 6) are plotted for all four dimensions in the Supplementary Dataset. We omitted GKs and GNaL from Figure 6 in an attempt at brevity since they do not add much to the story.

It is true that pacing frequency was not considered in this study.

The drugs were assessed across a range of doses (1x to 30x) but dosage only had a minimal impact on accuracy (88.1% to 90.8%) as shown in Figure 8A.

Finally, we emphasize that the metric’s novelty lies in deriving a simple linear model from biophysical principles of ion-channel blockade rather than blind statistical model fitting.