Good morning, everyone. Thank you, Mr. Chairman, for your invitation. I've been invited to talk about understanding of stem cell derived cardiomyocyte through computational modeling. It is a potentially quite wide topic. I choose to focus on single cell electrical activity and human-induced pluripotent stem cells derived cardiomyocytes. I will call them IPs cardiomyocytes in the following. Otherwise, all the time will be spent saying this name. And as this says, as in vitro models. So I won't address many other things like mechanical aspects and application to regenerative medicine and so on. So the aims are quite simple of my presentation to illustrate different ways in which computational modeling can be used to gain some understanding on these cells and to give you a few relevant examples of this. Just a quick slide of introduction, IPs cardiomyocytes, maybe you know better than me, can be derived by reprogramming of adult cells and differentiations into the cardiac phenotype and they show morphological and electrophysiological cardiac phenotype, but still differences with respect to adult cardiomyocytes in terms of morphology, of dimension, but also of electrical activity. And the most striking difference is that usually they show spontaneous electrical activity and this is one of the effects of a different mixture, a different balance of underlying current. The most important is thought to be the lack or great reduction in IK1 current. The computational model we have developed and has been used in the results I'm going to show is the patchy model of IPs cardiomyocytes, action potential. We developed it starting from 2013 and with some updates. The last version has been published last year in Frontiers of Physiology and this is the schematic representation of the structures included in the model, the cells and the sarcoplasmic reticulum and all the currents included. Including a current in the model obviously means have a set of equations describing the kinetics of the currents and when all the equations are solved numerically, what we obtained is the time course of all the variables included in the model starting from the membrane potential in the top panel and all the currents, for example. And we based the model on the quite extensive characterization of IPs cardiomyocytes that the general lab published in 2011 on American Journal of Physiology. They divided IPs cardiomyocytes in ventricular-like, atrial-like, nodal-like and so we also developed a version for ventricular-like and atrial-like IPs. But in the following I will always refer to ventricular-like IPs cardiomyocytes. Let me also very quickly recall what other kind of understanding we can gain from modeling. I think there are three kinds of this understanding. The first is the quantitative testing of hypotheses generated by the experimental data. And the second, the analysis of non-observable quantities like the simultaneous time course of the currents and the action potential. The most important one is the formulation of new hypotheses. Models are used to make predictions. Predictions should be seen, in my view, as hypotheses generated by the model. When we developed the model, the first question we tried to address was trying to understand the mechanism underlying this difference, the main difference in the electrical activity that is the spontaneous beating. And to do this, we used an approach that we can use in simulation. I call it virtual transplantation. That is changing, taking the formulation of one component of a model, let's say a current from a model, and insert it in another one. We used the ORROD model of adult ventricular cardiomyocytes and the PACI model. And one at a time, one current of the adult cardiomyocyte was inserted, transplanted virtually in the IPs cardiomyocyte model. And using this approach, we found that three combined mechanisms sustain spontaneous beating. The first one is the most expected instability of resting potential due to IK1. In this panel, you can see in the thin continuous line the IPs cardiomyocyte section potential. The dashed line is the adult one. And when we transplanted IK1 from the adult to the IPs, the spontaneous beating was stopped. We will be back on this panel in a few seconds. The second mechanism is increased excitability. When we transplanted the sodium current from the adult to the IPs cardiomyocytes, again, we found no beating. Here we see the beats synchronized with the adult because this is stimulated. Otherwise, no spontaneous activity. And the third one is slow diastolic depolarization due to the funny current, again, contributing to instability of resting potential. Going back to IK1, another observation was the adult IK1 is not sufficient per se to obtain adult ventricular action potential. Indeed, the thick line here is without spontaneous beating, but it's not resting potential. It's depolarized, something like a depolarization failure. We had to increase by 50% the amount of IK1 with respect to the ORRD model to obtain this dashed thick line resting stable potential. This is relevant to the dynamic clamp technique in which a virtual IK1 is injected in real cardiomyocytes to make them more adult, to make more stable. Indeed, in our hands, in a real dynamic clamp, we need to inject definitely more IK1 with respect to the formulation in the ORRD model to obtain stable resting. Later, we addressed the comparison in response to a current block between IP's cardiomyocyte model and the adult one in order to understand if findings from IP's in terms of drug effects can be translated to adult cardiomyocytes. Again, here are some examples. On the right, the ORRD model. On the left, the PACI model. And for L-type calcium current, IKR, IK1, the effect of progressive block. You can see that the kind of effect is similar. So, shortening for L-type calcium current, prolongation, and so on. But the amount, the sensitivity, somehow, sometimes is different, especially for the L-type calcium current. More sensible are the IP's cardiomyocytes. To a lesser extent, IKR and IK1, again, more sensitivity. This was also studied in a more comprehensive sensitivity analysis. But these results have been reproduced in an even more comprehensive analysis by Eric Sobey in his paper last year. And in this panel, you can see, again, changing L-type calcium current, the effect on the two models. And in the IP's cardiomyocyte, the sensitivity is definitely larger. Whereas for IKR, the difference is less. But if differences are there between the two kinds of cells, how can we predict what will happen in adult cardiomyocytes, starting from IP's cardiomyocytes? And in the same paper, Eric Sobey proposed, in my view, a very elegant approach that is based on the population of models approach, make a regression and a prediction on what will happen on adult models. Trying to be brief, the population of models approach in this approach, many models, many virtual cells are generated by randomizing some parameters. So the structure of the model is the same, but a few parameters are randomized in a physiological range. So many cells, virtual cells, are obtained. In each cell, through simulation, the action potential and the calcium transient are generated. And some biomarkers, like action potential duration at different levels, MDP, and so on, are measured. And output in a line for these cells in this matrix. The same is done for the adult model, so a population of adult cardiomyocytes. Then a regression matrix is estimated in order to make these two terms as equal as possible. Once the matrix is estimated, it can be used to predict, just by product, what happens in the adult myocytes, starting from results in IP's. And quite surprisingly to me, at least, it works. Surprisingly, because the systems are highly non-linear, but this is, at the end, a simple linear regression. Anyway, you can see some examples here. The effect of prolongation due to IKR block on the adult action potential, the prediction is shown by the circles, and the simulation of the block in the adult is shown by the curve. And this has been applied to different selective and non-selective ion channel blocking drugs, with good results. And the main message is that results that are much better than assuming IP's cardiomyocytes as a predictor, per se, of what happens in adult. You can see here the green markers are the predictions made with IP's, and the dark one after regression application. The last application I want to show you is based, again, on population models, and we used it to analyze the variability of the phenotypes, and also of the responses to drugs in mutant IP's cardiomyocytes, in IP's used as disease model. The approach is quite similar, starting from one model, a population is generated. In each cell of the population, the mutation is introduced through the formulation of the current affected by the mutation, and then drug effect can be tested in both populations. We applied this in two papers, one to addressing the LongQT3, and one more recent in LongQT1 and 2. We'll see only a few examples here. In LongQT3, the prolongation due to increase in late sodium current was recapitulated by the model, and then a population, each trace here is the action potential of one virtual cell is generated, but only those whose action potential lies within a range of physiological variability are kept in the population, and the mutant population is generated, the purple one here. As occurs in real life, the mutant population has a very variable amount of effect, some very long prolongations of action potential, but in some cases even no prolongation at all. So we tried to investigate how this, what the mechanism, why some of these cells are symptomatic and other symptomatic. By dividing in two, I'm sorry, I'll go to the conclusions, we found which currents are much compensating the effect of the mutation in the asymptomatic one. Okay, then we also tested some drugs, but going to the conclusions, I think that several computation approaches can help understanding IPs, cardiomyocytes, I have shown you, and I have also to say that, last but not least, simulations are inexpensive, so we have to be careful to not over-interpret the results of simulations, but given all this, I think that simulations run in parallel with adult NIPs, cardiomyocytes models, should always integrate experiments. For example, in the SIPA framework, for those of you who are familiar with this safety assessment framework. Okay, thank you. Well, thank you, and we do have time for questions, so I'm going to start with one. So as you know, IPs, cells, there's different stages of differentiation and maturation. So in your model, how do you represent and how do you reflect those changes in your system? Yes, in this model, we didn't address the development of different stages. We did a previous work, it was the first one on stem cells derived cardiomyocytes, it was on embryonic stem cells derived, and in that case, we had two data sets, quite uncommon, but at quite different stages, and we developed two different models. So at the end, it depends on the data you have. On the other side, you can also think to this population of models approach as a way to extend the range of cells you are representing. So maybe also at different stages, with some caution. So, as a follow-up to this question, how do you account for the electrophysiology role in the question of transients and the relationship with the NCS in your model? First of all, the sub-intracellular structure is simplified with respect to adult one, so we don't have diadic space, we don't have sub-membrane space compartments, and this is one aspect. The other one, we have identified part of the model based on a measurement on calcium transients. So the idea is that also the kinetic, at least, of the calcium transient is modulated by all these aspects of the calcium handling on one side of the structure on the other, and so what we can do at present is to do this. Well, thank you very much, and we're going to proceed with the next talk. Thank you.

computational modeling

stem cell-derived cardiomyocytes

human-induced pluripotent stem cells

electrical activity

spontaneous beating