Here’s an interesting little result that is thanks to Kylie. She found it whilst doing the work for her paper on prediction of drug compound effects at the tissue scale using data on multiple ion channel block. But we didn’t manage to squeeze it into that paper, so thought I’d put it up here.
Lots of people have done simulations of drug effects on cardiac tissue cells, we wrote a review on this a couple of years ago.
All you need to know here is that the heart cells experience (and generate) a wave of electrical activity that causes them to contract and pump blood. The wave seen by any particular cell is called its action potential. Blocking certain ion channels can lead to the heart cells being electrically active (‘depolarized’) for a longer time than normal – giving a longer Action Potential Duration (APD); and blocking others can lead to cells being active for less time – having shorter APD. (Of course, you can get anything in between as blocking lots of channels at once could do either*).
Prolonging the APD is linked to prolongation of what’s called the QT interval on the electrocardiogram (ECG) – see Fig 1. When induced by taking pharmaceutical drugs, both effects are often due to a particular potassium current being blocked (the current is called IKr, it flows through an ion channel mostly made of proteins encoded by the hERG gene).
In Kylie’s paper, the idea of the study was to use ion channel screening data to predict changes to the QT interval that would be observed in experiments on rabbit heart tissue.
If you want to simulate ECG changes then you have a few options:
- Simulate a single cell (0D) then just differentiate the action potential (you don’t see many people do this, but I can’t see why not as a first approximation).
- Simulate a monodomain tissue and use a ‘pseudo-ECG’ calculation – details in this paper.
- Simulate a bidomain tissue and compare extracellular potentials between locations directly (this is why you have a number of electrodes stuck on you for a real ECG).
Anyway, we did the middle one with a 1D ‘strand’ of tissue, to see if it would get closer to observed ECG changes in experiment than just looking at action potential duration changes (APD90). You can see the correlation between both approaches in Fig 2.
A change in APD90 is highly correlated with a change in QT in the simulations (as you’d probably expect). What we noticed, that I haven’t seen much comment on before, is that the relationship is not 1:1 (red line), instead it is steeper at about 1:1.34 (green line of best fit), so you get more change in QT in the simulation than you do in APD.
I’m not quite sure why we see this yet, whether different methods for measuring APD and QT would make much difference to this, or whether it is some effect of the boundaries in a tissue of finite length (obviously real tissues are finite so it might be a good boundary effect!). But if the relationship shown in Figure 2 holds more widely (for different models, in different tissues, for bidomain simulations etc.) it suggests a simple rescaling may allow more accurate QT predictions to be made from single cell APD predictions.
*Pharmaceutical drugs have a bit of a habit of blocking ion channels, sometimes that is how they work, but often it is an unintended side effect. For a lot of potential drugs block of cardiac ion channels is an issue. Both IKr block and QT prolongation are associated with a drug-induced increase in the risk of pro-arrhythmia: the heart going into an unusual rhythm which can be fatal. Obviously that’s an unacceptable side effect for something like a hay fever drug, and so pharmaceutical companies and regulators have to spend a lot of time and money checking this isn’t likely to happen, or weighing up the risk/benefit for e.g. a cancer treatment with a small risk of pro-arrhythmia.