false
Catalog
Young Investigator Awards Competition - Basic Scie ...
YIA Basic
YIA Basic
Back to course
[Please upgrade your browser to play this video content]
Video Transcription
We will start with our first presenter, Dr. Rogers. Dr. Rogers, you may now start your presentation and you have 10 minutes. Hello, everyone. My name's AJ Rogers. I'm a biomedical engineer by training, and I'm a cardiovascular medicine fellow at Stanford. I'm excited to say I'll be starting clinical EP fellowship next month. And I really appreciate the opportunity to present my work here. And thank you to the judges and everybody tuning in on HRS for being here and listening. I have no conflicts. This is my research support. So despite massive health consequences, risk stratification and sudden cardiac arrest has been very difficult. A reduced LV ejection fraction is an imperfect marker, although this is what we have to use clinically. This is shown by the red light on the ROC curve to the left. And it's only mildly better when we include other things like basic ECG metrics. However, a substantial opportunity does exist to save lives if we can improve risk stratification. For inherited cardiomyopathies, there is a wealth of cellular electrophysiological understanding. And these cellular mechanisms have been tracked all the way from action potentials to surface ECG all the way to clinical outcomes. However, this cell-to-bedside foundation does not yet exist for ischemic cardiomyopathy. In order to address this gap, we use machine learning. In cardiology, machine learning has provided a lot of value in biosignal interpretation. For example, ML has been used to approximate human expert classifications of complex signals. Here, a model was used to identify cardiac rhythm. And it performed about the same, at least as well as an individual cardiologist. It even made some of the same mistakes as seen by the confusion matrix here. It's also been used to discover patterns not easily interpreted by experts, such as quantifying the LV ejection fraction from 12 lead ECG. However, machine learning is a black box and has not been used to link these mechanisms with clinical outcomes in sudden death. So we asked ourselves, how could we bridge these two worlds, from cellular electrophysiology to clinical outcomes? We hypothesized that a machine learning model applied to ventricular action potentials could predict clinical outcomes at three years for ventricular arrhythmias and for death. We further hypothesized that interrogating this model and looking at the action potential partitions, we could develop phenotypes associated with specific ion current abnormalities. We recruited 53 patients. All of them had an LV ejection fraction less than 40% and coronary disease that was optimally revascularized. We excluded any patient that had a prior ventricular arrhythmia. And we took all these patients to electrophysiology study, including monophasic action potential catheter recordings. This catheter can be used in vivo in humans. And we recorded these monophasic action potential recordings. As you can see on the right, these are very closely tied and in shape to the transmembrane action potential. We did 90 seconds of pacing at a cycle length of 550 milliseconds, mostly from the RV apex, but in a minority, the LV apex. And we recorded at the site of highest regional amplitude so that we could avoid any scar or border zone. We excluded about 11 patients for having less than 50 quality MAP tracings. And the MAP signals were digitized and exported at 1 kilohertz, filtered at 0.05 to 500 hertz. I annotated all of these recordings so that single beats would be put into the models. Model inputs overall were 5,706 separate MAP recordings. For clinical outcomes, we used these as the labels. ICDs were present in 88%. And close clinical follow-up was available at our institution. So we looked at ventricular arrhythmias if they met the treatment threshold or other clinical recordings in the chart, as well as mortality by death registry and the EHR. In order to create the models, we divided all the beats from patients into training and testing in a process called cross-validation. During each run of the prospective model, 30% of the patients, like these in the red boxes, were set aside for testing. And we repeated this 10 times. And summary test characteristics were calculated for each model. Several supervised architectures were trained and evaluated. And of all of them, the support vector machines based on TSFRESH features turned out to be the best. In short, TSFRESH is a Python library that develops coefficients for different frequency and time spectrum features of the raw signal. And the purpose of using this is to decrease the dimensionality to improve supervised learning. Support vector machines is a supervised architecture that partitions the data based on rules or hyperplanes that separates the two classes during training. And then testing the patients can run through those rules and determine whether they are of the class of ventricular arrhythmias or death or survival. For patient-level analysis, we looked at the proportion of beats from each patient that predicted the outcome. If you had a high proportion of beats, it would be classified as a high risk for that outcome. These are the demographics for the patients. They were mostly in the seventh decade and predominantly male. There was no difference between the patients at baseline between those who had ventricular arrhythmias at three years and those who did not. And the same was for mortality and inducibility. Overall, we reported 5,706 MAPs. They were of good quality. And by visual inspection alone, it's difficult to differentiate the events from the non-events. At the patient-level analysis, the AUC for VT and VF was very strong, 0.901 for the ROC curve and 85.7% accuracy for predicting ventricular arrhythmia. For mortality, the numbers are seen here. On univariate and multivariate analysis, only the MAP scores predicted VT and VF, despite looking at several endpoints. For mortality, the MAP score strongly and the lack of prior beta blocker use at baseline more weakly predicted all-cause mortality. To interrogate our train model, we evaluated the mean MAP shapes from all 5,706 action potentials that were predictive and non-predictive. Monophasic action potentials predicting VT and VF had prolonged and elevated phase 2 compared to non-VT and VF. For mortality, there was an enhanced early repolarization pattern not seen in those predicting survival. These phenotypes suggest a mechanistic link between action potential shape and clinical outcomes. And importantly, they can be measured clinically and can thus be used to enable personalized risk stratification. Since the initial submission, I've performed additional work to identify specific ion channel abnormalities. We performed in silico action potential modeling using the O'Hara-Rudy model. And model parameter sets for the currents were permuted over a very wide range, resulting in 750,000 unique parameter sets and states of the model. For each of those states, we looked at the comparison between that and the mean action potentials on the previous slide. And we found that for ventricular tachycardia and ventricular fibrillation, L-type calcium channel currents was increased for those that had ventricular arrhythmias. For mortality, we saw an increase in the herd channel currents. In summary, machine learning applied to ventricular action potentials predicted VT and VF and death in ischemic cardiomyopathy. Interrogation of those model partitions identified phenotypes with specific ion channel abnormalities, including prolonged phase II plateau as a risk for VT and VF and earlier polarization abnormality as a risk for mortality. What's important, again, is that these phenotypes can be clinically identified and thus enable personalized risk stratification. These novel phenotypes uncovered may also provide therapeutic insights in the future. All studies do have limitations. And this one is not unique. The limitations included recordings were primarily from the RV apex. And optical imaging or unipolar electrogram recordings in the future could be used for more wide assessment. We think the RV apex was a good, stable, and reproducible place to record. We did not have access to tissue, but biopsies in certain disease states may provide in vitro assessment. And furthermore, the O'Hara-Rudy model is not specifically designed for heart failure, but our wide parameter search likely captures the heart failure state. In the future, I'd really like to extend these to more accessible recordings, such as ECGI or implanted devices. And I'd love to use this as a basis for fellow to faculty transition support. Very excited to continue this work after EP fellowship. And I'd like to thank all of my collaborators and my mentors for teaching me and ushering me through this process. Of course, Dr. Narayan, who's been my primary mentor for the last five years. He's allowed me the chance to pursue my own research interest and help me develop my research and training plan. Thank you very much. And I'll take any questions. Dr. Rogers, that was an amazing presentation. And we will now start with your first primary, first judge, who is Dr. Ruecklinger, who will be asking most of the questions, followed by a secondary reviewer, Dr. Lubitz, who will have follow-up questions. Dr. Ruecklinger, the microphone's yours. Thank you. And thank you, Dr. Rogers, for an excellent presentation. In fact, my first question for you was going to be about some future directions and how you might extrapolate this to cellular mechanisms. But it seems as if you are already on the way to doing that with your mathematical modeling. So that's great. So hopefully, I can state this clearly. I'm interested in the shapes and the similarities between your final monophasic action potential recordings that were predictive of VTVF and mortality. So if I understand correctly, you got those average monophasic action potentials by sort of taking the mean of all of the action potentials that were predictive and then saying, OK, then there's a prolonged phase 2. So I guess my question is, how consistent is that prolonged phase 2 between action potentials and between patients? So when you average thousands of them together, you see the prolonged phase 2. But have you done any sort of analysis or correlation analysis to see if each individual action potential and each individual patient shares those features or whether you have to have thousands in order to see that emerge? Yeah, I think that's a very important question. And the truth is that it's actually not as many as you would think. And I think the reason why we were able to improve the model by creating this MAP score, so for the patient level analysis, it's which proportion of those recordings actually predict the outcome. For both of the ROC curves, it's about 30%. So patients that did not ultimately have the outcome ended up with having only a few beats that ended up having a shape that was longer, whereas patients that had high risk of the outcome, it'd be more than a third. And so I think there's some sort of like a filtering method in my mind. And you have cardiac motion that's hitting the catheter. You have catheter moving around. You have the contractility of the heart beating against the pacing. And so I think it's only a fraction of those that are actually picking up the signal that is ultimately predictive. So that's why we use the predicted beats rather than just averaging all the patients with the outcome to find out which part of the signal was actually important. OK, thank you. If I understand correctly from your methods, you also took these recordings from basically viable myocardium. That was not the scar, not the border zone, because presumably those signals might be noisier. Is that correct? Well, we wanted to make sure that we had good signals that we could analyze. And if you get into the scar too deep, you don't have much of a signal to look at. And also, border zone has been shown to have kind of different electrophysiologic characteristics than tissue that's remote from the scar itself. And so we looked at other MAP recording papers in animals. And it appears that as long as you find amplitudes that are in the top quartile, top half of the amplitudes that are available, that on histologic examination is remote from scar. And so in that way, we could try to get it so that all the patient recordings lined up with each other. I think some of the patients that ended up getting excluded might have had too few recordings because there was extensive scar in that area, the apical region. That's my guess. OK. And I guess my last question would be, can you tell us a little bit more about this TSFRESH software algorithm that you used? Because it seems that that's what allowed you to extract all these different individual signal features. But some of those were really kind of arbitrary, like the imaginary parts of the Fourier transform of the signal at a certain frequency. Did you have any input in defining what those are? Or is that kind of a predefined algorithm that extracts a full set of features? So the TSFRESH package will actually give you thousands of different features. And they're both in the time and the frequency domain. And as you mentioned, they're very abstract. If it gave me something that was a simple frequency characteristic, or even better, if it would just tell me 10 features that had to do with phase 0, 10 features that had to do with phase 1, then it would be very easy to interpret, rather than doing the mean of all the predictive beats. So in order to pick those features that ultimately defined what was predictive, what you do is you compute all of the different features for all of the different beats across the entire data set. And then you run a logistic regression that shows which one of those correlate with the outcome most strongly. And then you can limit the number, because you don't want to use thousands of endpoints. Otherwise, you're not really reducing the dimensionality, which was the goal of using the TSFRESH. And in that way, you can train on 30, 40 features that ultimately can be partitioned a little bit more efficiently. Now in future, what I would really like to do is get away from using TSFRESH and just have the raw data go right in. But I think the problem is that using neural networks, whatever architecture you pick based on the raw data, since that input data is so highly dimensional, you're actually going to have to need a much larger number. So I think we have a large swath of data and probably the biggest available for this type of monophasic action potential type recordings. But if we go to something like UCGI or maybe unipolar recordings from an RV tip of an ICD, something like that, if I can get that on a much larger range of patients, then I could probably use the more data-hungry methods that don't require that preprocessing step. Thank you. I'll turn it over to my co-judge. Dr. Rogers, congratulations. That was a great presentation and very innovative work. And you also successfully anticipated many of my questions for you. So I'm going to have to think a little bit harder here. But I will follow up just on the last point that was being discussed there. And I was particularly interested in the methods, how you ended up using an SVM and this TS-Fresh package. And it looked like in early slides in your presentation that you had tested out neural networks as well. I wasn't sure how you compared the neural networks to the SVMs and exactly what you used to determine superiority and performance. And as a follow-up to that question, I'd like to just ask you to explain to us what the optimal experiment would look like from start to finish, if you had your way. Sure. So I'll try to attack the first part. And then I may ask you to remind me of the last part. So the first part is, how do you compare across different models to determine which one's the best? And so I think that there's a couple of different things that you have to keep in mind. So if you have different data sets and you're comparing models across different data sets, you have to really pay attention to, what is the prevalence of the different diseases in your population to determine which metrics you're going to use? So not all AUCs are created equal. An ROC curve is very different from a precision recall curve, especially if you're dealing with imbalanced data sets. So comparing from one paper to another, you may have to be very careful about those metrics. In this case, what we did is, at the very beginning, we took our data set and we split it into this training and testing cohort. And we took all of those different assignments and we froze them. And we used the same ones and put that into all the different models that we try. So that way, the incidence, the splits, and the randomization, all of that was equal across all the different models. And then we just look at a beat by beat basis. So when you do the cross validation, you have the predictive results from all of the different beats, and you can just calculate accuracy. You could calculate an F1 score. Any of those should be comparable across those because you're using the same test splits. So I think this is a robust method. I think if I had all the data in the world, if the data labeling wasn't so labor intensive, and if we weren't limited by measuring time in the lab and risk to patients and all of that that we have to be concerned about, I'd like to greatly increase the numbers, in which case, the ideal would be we train on a data set at one institution, and then we go to another institution on different machinery, on different set of patients, and then we test the data on that group of patients. I think this would just, we'd have to use a great number of patients and a great deal of time in the lab to do it. So I think it would be possible to do that type of thing if we use one of these MAP surrogates that I discussed earlier. Thanks very much. I think that qualifies as two answers to two questions embedded into one. Congratulations on the work. Perfect. Thank you so much. Thank you, Dr. Rogers. That was an excellent presentation, good question and answer session. We will now move to the second finalist, Dr. Wimbo. Dr. Wimbo, please take your time loading up your slides and we'll come back on. Thank you for the opportunity to share this research. I'm going to be speaking about increased neurotransmission in synthetic neurons derived from long QT syndrome patient-induced glucopotent stem cells. This work focuses on the long QT syndrome, an ion channelopathy associated with sympathetically triggered cardiac events. In long QT type one, most notably, increased beta adrenergic signaling is an established trigger of life-threatening cardiac events. Long QT1 is, as you know, caused by a loss of function variance in the KCNQ1 gene. While KCNQ1 is expressed in the heart, it's also expressed in neuronal networks and tissues. KCNQ1 dysfunction has been identified in epilepsy and sudden death in epilepsy, both associated with neuronal hyperacceptability. Major preventive therapeutic options for long QT1 are neuromodulators, including beta blockers and cardiac synthetic generation. However, the human sympathetic neuronal long QT1 phenotype remains unknown. So we hypothesize that sympathetic neuronal function might be altered in long QT1 and develop long QT1-controlled sympathetic neurons, human-induced puripotent stem cells, to study their function in mono- and co-culture with IKS-derived polymyosinase. So the first patient that we included has a compound heterozygous genotype, meaning a severe KCNQ1 function loss, a QTC of 557 milliseconds, early symptoms to abuse, and multiple triggered cardiac events, including aborted cardiac arrests. For this presentation, however, we've also included data from a second patient with a novel class V KCNQ1 variant, a family history of several sudden deaths, a QTC of 530 milliseconds, and exercise-triggered syncope. So the included genotypes are compound heterozygous and heterozygous KCNQ1, and wild-type KCNQ1 as control. This is an overview of our methods. We reprogrammed white blood cells into iPS cells using non-integrating centivirus vectors. iPS cells were differentiated into synthetic neurons and cardiomyocytes, matured, and then studied in mono-culture and co-culture using various methods. Mature synthetic neurons showed similar morphology across genotypes, including large somas and extensive neurites. Cells were positive for tyrosine hydroxylase, the enzyme needed to make noradrenaline, rifrin expressed in peripheral neurons, and nicotinic acetylcholine receptors expressed in post-cardionic sympathetic neurons that project to the heart. As you can see in the left panel, 86% of nucleated cells were Th-positive, and there was a significant co-localization between Th and nicotinic receptors in the soma and varicosities of the sympathetic neurons. More importantly, all the derived sympathetic neurons were able to fire action potentials, either spontaneously as seen in A and B, or after provocation with hypotassium seen in C, or by current injection as seen in D. All derived synthetic neurons also presented with synaptic current activity as seen in E, meaning they form functional connections and communicate with each other. Both wild-type and locative-1 sympathetic neurons were largely comparable with regards to resting membrane potential, around minus 60 millivolts, and had comparable membrane resistances. So with that established, the talk will now focus on how the locative-1 sympathetic neurons differed functionally from wild-type neurons. Firstly, age-matched locative-1 sympathetic neuronal cultures, plated at the same density as wild-type cultures, released more than double the amount of noradrenaline. As has been reported in other disease states, increased noradrenaline release regulates further presynaptic transmitter release via positive feedback pathway. And increased noradrenaline increases calcium influx and susceptibility to triggered arrhythmia. Secondly, when exposed to increasing levels of current injection, locative-1 sympathetic neurons consistently responded with a higher action potential frequency. And thirdly, locative-1 sympathetic neurons presented with a larger total inward current density, so current corrected for capacitance, between minus 30 to plus 20 millivolts. Moreover, the amplitudes of synaptic currents were significantly larger in locative-1 neurons. And you can see clearly on the right-hand side how the histogram for the locative-1 synaptic currents is shifted to. These are examples of spontaneous action potentials from all three genotype groups. Normally, following an action potential, a neuron will repolarize fully, and then hyperpolarize, resulting in that dip that you see after the spike, which allows for inactivation of the neuron. Wild-type neurons, after hyperpolarization, or AHP, were seen in 93% of the cells. In the long QT cells, however, lack of AHP was seen in 57% of cells, typically followed by excitatory post-synaptic potentials, or EPSPs, or triggered action potentials. While typically caused by synaptic activity, EPSPs can result from a decrease in outgoing positive charge, in this case, potentially, from a decrease in outgoing potassium via casein-Q1 channels. So with decreased repolarization, impaired AHP, and thereby impaired inactivation, this could provide a potential mechanistic link between casein-Q1 genotype and our findings of hyperexcitability in locative-1 synaptic neurons. So up until now, the presented data has related to sympathetic neurons in monoculture. From here on, I will focus on neocardiac cultures. So we co-cultured sympathetic neurons in cardiomyocytes up to 14 days before performing experiments. During co-culture, we saw successive up-growth of neurites reaching into cardiomyocyte clusters, so resulting in close proximity between neurites and cardiomyocytes as seen in these images. To test whether the neurons in the cardiomyocytes are functionally connected, we exposed cultures to nicotine to induce noradrenaline release from the sympathetic neurons and measured the cardiomyocytes' beating rate response. In co-culture, a significant beating rate increase was seen, but not in monocultured cardiomyocytes, supporting that the beating rate increase in co-cultures is indeed due to sympathetic activation. Similarly, in this simultaneous recording with cardiomyocyte action potentials above and sympathetic neuronal action potentials below, an increase in both neuronal firing frequency and cardiomyocyte beating rate is seen during nicotine exposure. Here, we measured the action's potential duration during nicotinic activation, normalized to baseline. Notably, the long-cut T1 cardiomyocytes responded with marked prolongation of the action potential duration. And in the inset, we can see an example of what that APD prolongation looks like. Here again, note the successive prolongation of the action potential duration during nicotine wash-in, leading to plateau phase prolongation and calcium channel reactivation, similar to what you might see in long-cut T1 experiments using sympathetic memetic floats. So in summary, we generated electrophysiologically functional synthetic neurons from human iPS cells, modeling compound heterozygous and heterozygous long-cut T type I. Long-cut T1 sympathetic neurons presented with an hyperactivity or hyperexcitability phenotype with increased noradrenaline release, increased action potential firing frequency, increased synaptic current amplitudes, and decreased repolarization or AHPs. In neurocardiac co-culture, the sympathetic neurons modulate cardiomyocyte beating rates and action potential duration. And activating these synthetic neurons recapitulates the long-cut T1 arrhythmia phenotype in the cardiomyocytes. So in conclusion, we have identified a novel neuronal long-cut T1 phenotype that could contribute to algorithmic mechanisms underlying sympathetic trigger ability. And I'd love to thank all my collaborators and co-authors that worked with me on this study, and I'd be happy to take any questions. Thank you. Dr. Wimbo, that was an excellent presentation. You will now have questions from your primary reviewer, Dr. Tereschenko, followed by Dr. Replinger. Dr. Tereschenko, the microphone is yours. Thank you so much. This is really a beautiful presentation and very important critical novel contribution and important advancement in the field. So a few questions. How generalizable are findings from your point of view? Because you are taking cells from one particular person, would you think that you can generalize it to the whole disease, long-cut T1 at least? Or how important it is donor gender, ethnicity, and specific technique which you may use in laboratory such as manual pipetting, cell passage, or any other technical points which you can apply in your experimental study? So how do you think is generalizable your findings? Thank you for that question. It's really important. One of the things that we've been working with with the original model, which is the compound heterotagous model, is to make sure we repeat differentiation protocols multiple times. So for example, this data set is from five different differentiation rounds, and we see the same phenotype in our neurons every single time we start from the beginning from the active cells and to differentiate them. So I think that's very important. But then also, like you said, we started out with a very severe case in Q1 loss of function patients. And it was very heartening to see that in the heterotagous model or the heterotagous patients, we see a similar phenotype with hyperactivity and hyperexcitability. So at least between compound heterotagous and heterotagous, the results look the same. For the compound heterotagous patient, that's a young adult male, and the heterotagous patient is a young adult female. So far, the numbers are too few to be able to look at gender, of course, and that's something that would be quite important to focus on as we go ahead. Same with ethnicity, because we see that a lot with our LQT cohort in New Zealand. We have quite wide, well, differences in ethnicities. So about one third are Maori or Polynesian heritage, and that could also be an important thing to look at. But of course, right now, we only have a limited number of patients. Thank you. And follow-up question. How do you plan to translate your finding into clinical practice? How do you think it change clinical practice in the future? So as I said in the beginning, so most of the therapies that we use for LQT are neuromodulators. And of course, if neurons are actually actively driving this phenotype, we'll need to look more mechanistically on how to target the neuronal phenotype, which could then alter the way that we treat our patients. So in that sense, I think this will be quite important or potentially important. And what about other LQT, for example, LQT1 and LQT2 clearly has important involvement of the sympathetic nervous system, and type one is provoked by exercise, type two by emotional stress. Did you target specifically LQT1 or what are your thoughts on other types of LQT? So far we've focused on LQT type one because of the clear connection with a synthetically triggered events. But I agree that LQT2 also would be quite useful to look at and the proportion of epileptic cases in LQT2 is actually higher than in LQT1. So I do think that that would be quite important to look at and see whether the neuronal phenotype is there as well. Okay, thank you so much. Congratulations, it's very good. I don't have questions. All right, thank you. Congratulations, Dr. Wimbo, that's an exciting study. So in my mind, the major outlying question is what is the mechanism of the hyper excitability in the neurons? And so have you looked to see, do they have KCNQ1 channels in culture and how would you test that? And do you think that that's the underlying mechanism that it relates back to that channel or do you think that there's another potentially more complex mechanism driving that? That's an excellent question. One of the ways that I've tried to understand this data is when you look at the repolarization deficits that we see, that could be directly linked to KCNQ1. But as you said, we have to go back and actually prove that KCNQ1 is expressed in these cells in culture. So that would be a clear next step to do. But based on the way that these neurons are not inactivated properly and they tend to keep firing and they have this high firing rate, that could relate back to KCNQ1. But on the other hand, since we see this increased neuroadrenaline release, it's hard to tell whether that is a secondary effect because of the higher activity or whether that actually drives it. And there is also, if you looked at the IV code, this increased inward currents looking like there could be an increased calcium current activity as well. And to tease out which one actually is the primary cause is quite difficult and will be very exciting to work with, I think. And I guess my other question would be, and maybe I should know the answer to this, but I don't. In LQT1 patients, we know that it's autonomically mediated or sympathetically triggered arrhythmias. Has the experiment been done where anyone has looked at the, pulled out the stellates of a LQT patient and recorded to see if they are hyper excitable in these patients? Or is it possible that this is an artifact of the IPS system somehow? Yeah, excellent question. So that is definitely one thing that we want to do if we can get some human tissue to actually test this in LQT patients undergoing cardiac denervation, which would be ideal. To my knowledge, there has been no studies in human sympathetic neurons on a cellular level. There has been some studies in specimens from operations where they've seen signs of inflammation, for example. But none in living functional human cells. Okay, thank you. Excellent work. Excellent presentation, Dr. Willembo. I have one question in closing. You've used this paradigm in LQT1. If you were to use a different cell type applicable to more clinical, bigger population, is there a different cell type you would use to think about it? Because there are many other arrhythmias that are very much sympathetically driven post-MI and other sort of environments. Have you done some experiments like that? Or would that be a target towards that? So far, we've only looked at synthetic neurons in normal and LQT1. But I completely agree. I think the beauty sort of of this co-cultural system is you could use any type of genotype that you were interested in and be able to look at how those neurons communicate with cardiomyocytes. And it doesn't have to be not QT1 or CPBT, which is also, of course, the bites would be very interesting to look at. But it could also be other types of disease. Thank you very much. That was an excellent presentation and a good discussion. Thank you. I'd like to introduce Jonathan Huang, our third finalist of the basic science competition. Jonathan Huang, you may now load your slides. And you have 10 minutes for your presentation. Thank you for having me here today to talk about my study, looking at the autonomic and electrophysiological effects of thoracic epidural anesthesia on the chronically infarcted porcine heart. So my name is Jonathan Huang, and I'm a PhD candidate in the lab of Dr. Marmar Bussegi at the UCLA Cardiac Arrhythmia Center. I have no conflicts of interest to declare. So with VT recurrence, despite catheter ablation occurring in up to 20% of patients with ischemic cardiomyopathy, up to 50% of patients with non-ischemic cardiomyopathy, it has become increasingly appreciated that the genesis of ventricular arrhythmias is truly a combination of the anatomic substrate and autonomic imbalances. So where conventional arrhythmia management has failed to act, of which beta-blockers are a cornerstone of treatment, thoracic epidural anesthesia may provide valuable anti-arrhythmic benefits. In a 2005 case report, a 75-year-old man presented with numerous shocks that were refractory to conventional arrhythmia management. His ICD was eventually depleted and he required external defibrillation. Shocks continue despite general anesthesia and increased doses of both amiodarone and Esmolol. However, upon institution of thoracic epidural anesthesia, he showed complete freedom from VT without compromising hemodynamic function. The prognostic power of bare reflex sensitivity, a measure of parasympathetic function, in predicting cardiac mortality has been comprehensively studied for decades. Ischemia-induced losses of bare reflex sensitivity were first shown by Takeshita et al. in 1980 and further shown by Schwartz et al. in 1988 to further predict sudden cardiac death. Since then, bare reflex sensitivity has been used extensively as a powerful clinical predictor of cardiac mortality in patients with myocardial infarction. Current studies looking at the effects of TEA, however, on bare reflex sensitivity report wildly conflicting results. However, these results may be contingent on the resting autonomic tone as well as the animal and disease model employed. Moreover, experimental studies to date involve the administration of TEA for acute conditions in normal hearts and really failed to take into account the neural remodeling present in chronic myocardial infarction, while clinical studies lack sufficient mechanistic power. Thus, the goal of this study was to assess the effects of TEA, particularly on parasympathetic function, in the setting of chronic myocardial infarction. Before induction of myocardial infarction, you can appreciate here that there's a strong coronary perfusion, but after the percutaneous injection of microspheres into the LED, there's a marked loss of flow through the distal portion of the LED. Four to six weeks later, significant scarring of the left and right ventricles noted. Dense scar adjacent to viable myocardium represents heterogeneous domains of myocardial, electrical, and neural function. Grossly, we can also begin to identify potential regions of electrical border zone characterized by islands of viable myocardium within scarred regions. Hopefully, from looking at just the myocardium here, you can already begin to appreciate the modeled appearance that really characterizes this heterogeneous substrate. So, to complement surface and intracardial electrograms, we can also advance 56-electrode SOC over the ventricles to obtain local unipolar electrograms. Then, using a closely spaced bipolar catheter, we can measure electrical potentials from the epicardium and validate the functional electrical region over which each of these electrodes sits. So, with the high thoracic spinal cord providing sympathetic innervation to the heart, the epidural catheter, seen here, was placed at the junction of the C7 and T1 spinal segments using a standard loss of resistance approach. The animal was then placed in the supine position, and ECG, as well as left and right ventricular hemodynamics, were recorded by intracardiac pressure conductance catheters before and after TEA. There was a small decrease in heart rate consistent with previous reports, and while all LV hemodynamic parameters saw a modest reduction, this may be founded by potential lumbar sympathectomy with our approach. However, there were no effects on RV hemodynamics, which has really been the primary concern due to the preload dependence of the RV, the dysfunction of which may ultimately precipitate hemodynamic collapse. So, looking at some basic electrophysiological parameters before and after induction of TEA, using intracardiac EP catheters, we saw that the age interval was prolonged while the HV interval was unchanged, and moreover, both the atrial effective refractory period and ventricular effective period were increased. However, for more insight into the antiarrhythmic effects of TEA, we had to turn to higher resolution electrophysiological techniques. So, using that 56 electrode socket mentioned earlier, we measured local unipolar electrograms from the ventricular epicardium. After thoracic epidural anesthesia, we saw a prolongation in activation recovery intervals, a surrogate for local action potential duration, as indicated here by the shift from cooler to warmer colors. Globally, this represented an average prolongation of 15 milliseconds, an effect which may underlie the cardioprotective benefits of TEA, as prolongation of APD is a major antiarrhythmic mechanism. Notably, these effects were not restricted to any particular region of these diseased hearts, having similar effects in viable and scar tissue, and importantly, the border zone. Next, to evaluate the effects of preganglion-like blockade on postganglionic and vagal cardiomotor function, we stimulated the cell-like ganglion and cervical vagi. The effects of bilateral cell-like stimulation on action potential duration were unaffected by thoracic epidural anesthesia, but this is perhaps unsurprising as we directly drove synthetic efferents at the level of the cell-like, effectively circumventing the blockade. Similarly, both right and left vagal nerve stimulation were still effective at prolonging ventricular APD despite thoracic epidural. To now assess afferent and parasympathetic efferent function, we employed VAER reflex sensitivity as a measure of vagal function and predictor of sudden death. By administering phenylephrine, we were able to evoke an acute rise in arterial blood pressure and a reflex-induced increase of IR interval. At baseline, these animals had poor VAER reflex sensitivity, not dissimilar to a patient with a large myocardial infarction. Surprisingly, epidural blockade to block spinal afferents and synthetic efferents improved VAER reflex sensitivity in these infarcted animals, perhaps indicating an augmentation of parasympathetic function. However, while the prolongation of ERP and APD, as well as the enhancement of vagal function, are all great, it's still rather speculative. What does all this actually amount to in terms of cardioprotection? Thus, the next step was to induce ventricular tachycardia by ventricular extracellular spacing. So in this animal, after entraining the heart to steady cycle length of 450 milliseconds, the introduction of just two extracellular stimuli caused this animal to degenerate into ventricular tachycardia. However, the same animal, after administration of epidural lidocaine, was no longer inducible with even three extracellular stimuli. Overall, thoracic epidural anesthesia was effective at suppressing ventricular arrhythmias in six of these nine animals, thus proving a significant reduction in the inducibility of ventricular arrhythmias. Yet another potential mechanism by which TEA may be cardioprotective is by homogenizing the activation wavefront. Before TEA, there are numerous isoelectric potentials in the SCAR border zone and areas of early activation directly adjacent to areas of late activation, representing possible circuits for reentry. TEA decreased heterogeneity in activation while mitigating areas of late activation as reflected by a reduction in isopotentials in SCAR and border zone regions, potentially suppressing ventricular tachycardia. So in conclusion, we thus report five key findings in the study in the chronically infarcted porcelain heart. Thoracic epidural anesthesia prolongs the AH interval as well as atrial and ventricular ERP, prolongs global and regional APD, preserves postganglionic sympathetic and parasympathetic efferent function, and enhances vagal afferent efferent function. And so together, TEA suppresses ventricular arrhythmogenesis, perhaps through the combination of electrophysiological and autonomic effects seen here. Overall, myocardial infarction induces structural and neural remodeling, generating the anatomic and functional substrate for ventricular arrhythmias. By blocking both afferent and sympathetic efferent neurotransmission through the spinal cord, thoracic epidural anesthesia may be antiarrhythmic. For the first time in the study, it has been shown that the electrophysiological and autonomic effects of TEA persist even in the setting of structural heart disease. TEA does not compromise hemodynamic function, as was shown previously, and possibly due to normalization of autonomic function rather than the depression of a normally functioning one. Moreover, TEA does not interrupt intrathoracic reflex loops between afferents and sympathetic efferents in the stellate ganglia and the intrinsic cardiac nervous system, potentially allowing these local cardiocardiac circuits to continue to accommodate changes in myocardial demand. Similarly, TEA does not directly affect vagal afferent nor efferent neurotransmission, but by inhibiting spinal afferent signaling, it may, by inhibiting spinal afferent signaling, which may in turn depress vagal function, TEA may indirectly enhance vagal function and exert antiarrhythmic effects. Thus, TEA may actually act on both arms of the autonomic nervous system and reduce the overall arrhythmogenicity of chronically infarcted hearts. And with that, I'd like to conclude my presentation by thanking my co-authors for their help in making the study possible, my mentor, Dr. Russegi, for her support and guidance, and lastly, a big thank you to the Young Investigator Award Committee for their invitation to speak today and their attention. Thank you. Jonathan, that was an excellent presentation. Your judges are Dr. Lubitz and Dr. Tereschenko. Dr. Lubitz is the primary judge who will ask most of the questions, followed by Dr. Tereschenko. Dr. Lubitz, you have the microphone. Well, thanks very much, Jonathan. That was a terrific presentation. Congratulations on an immense amount of work. In real world practice, patients post myocardial infarction who are having arrhythmias are often on beta blockers, potentially antiarrhythmics and other medications that may modify the way the sympathetic nervous system is influencing susceptibility to arrhythmia as well as the parasympathetic nervous system. How do you think thoracic epidural anesthesia would play a role in a setting in which patients are being exposed to such medications? I mean, referencing back to that first 2005 case report, it really, I think, needs to act at the intersection of or act on these patients that are refractory to other therapies. Of course, beta blockade and other antiarrhythmic agents will remain to be the cornerstone of treatment, but TEA should be used as either treatment for these refractory patients or a bridge to a longer term solution like heart transplant. You can take that a little bit further. Where exactly in the clinical pathway do you see thoracic epidural anesthesia potentially playing a clinical role in light of emphasis on other methods such as cervical sympathectomy, for example? There was a recent study done maybe a few years ago looking at the clinical outcomes comparing cardiac synthetic denervation and thoracic epidural. Actually, it showed that thoracic epidural had a greater effect in suppressing VT storm in a number of patients than did cardiac synthetic denervation. It may play a role there, but also I think one of the important findings of the study, which I didn't emphasize too much at the end, is that it didn't terribly impact the ventricular mechanical function. In these patients who are receiving other treatments such as beta blockade and they have the concern for loss of ejection fraction and worsening cardiac function, thoracic epidural may be a nice middle ground where it's not a surgical option and it's much it's able to be much more easily in these susceptible patient populations. Thank you. Can you tell me if you think that any additional work is necessary to demonstrate the safety and efficacy and safety really of thoracic epidural anesthesia? If so, what that experiment might look like? I think one of the main challenges with this is really to see how I think that it does have the strength of being able to suppress VT storm when people have no other options, but also in the long term how well thoracic epidural anesthesia is tolerated. While we were presenting the chronic MI state, the application of the thoracic epidural itself was only acute, and so how well patients will respond both consciously and again, so one of the limitations of the study was that this was done on sedated animals. So how this is tolerated consciously, as well as in long-term treatment needs to be evaluated before this can be used in humans, I think. Thanks very much. That's all the questions that I have. Thank you, Dr. Lubitz. Okay, so then I continue. Congratulations again. It is very interesting work. I have several questions on dispersion of activation and recovery. Several slides, and one slide was illustrated change of activation recovery interval, but on my eye, and again, you did not comment on that, but usually there was similar dispersion of refractoriness, which you did not measure. So my question is, did you measure dispersion, not only duration of activation recovery interval, but distribution of it, and did you measure it, and if you did, what did you find? Thank you, Dr. Tereshek, and that's an excellent question. So I did actually try looking at the dispersion, and I didn't see any significant changes either before or after TEA, and so possibly one of the reasons might that be is that, whereas normally like the left and right sympathetic chains have sort of complementary actions on the ventricular myocardium, I think, again, one of the benefits of thoracic epidural anesthesia is that it really acts bilaterally, and so it's likely acting on both the left and right, both ventral root ganglia and dorsal root ganglia, and so it acts much more homogenously than with a unilateral denervation, and so in this study, we did not see any changes in dispersion actually. And then follow-up question, you mentioned you noticed a reduction or change in dispersion of activation, and I honestly don't understand that. I cannot reconcile those two together, and again, when you showed those pictures to my eyes, they looked very similar before and after, so any comments on that? So if you don't see any dispersion of recovery, but you see dispersion of activation, how would it be if your substrate is essentially the same? You have the same, so scar did not change. You did not manipulate on scar portion of that. How would you explain change of activation without change in dispersion in recovery? Sure, I think that's another excellent point that you're making, and I agree. I think it's difficult to assess the dispersion in this case, because even looking at micro-domains of the dispersion, we really, it's such a small circuit for in that case that we were looking at, where you see, unfortunately I can't bring up the slide, where you really see that period of early activation right next to late activation potentially providing that re-entrant circuit, and then following the institution of thoracic epidural anesthesia, it was eliminated. I agree it's a rather small-looking effect, but it could be a very clinically important one nonetheless. And last question. There was a large human study which was published in circulation in 2016 about effect of the type of anesthesia which you are studying on right ventricular dysfunction, and it documented pretty well that negative effect. How would you compare animal experimental results with the clinical human study? Do you think that your experiment is sufficient to disprove previous human study findings? So, I'm not exactly sure which study that you're referring to, but I think that definitely, again, the Wink paper, yes. So, I think that one of the, again, a major limitation of our study is that this is on sedated animals under anesthesia, and so there's already a large depression of the autonomic nervous system in this case, and so any differences that we're sort of looking at may be sort of suppressed. So, one may be underestimating a difference, but also when we're affecting the basal autonomic tone, it can really shift the balance in a way that's unrepresentative of a conscious person. I'm sure the study was also likely done in anesthetized patients, but the modality that we're looking at may be completely different because I don't think that that Wink paper took into account any chronic remodeling that happens, such as after myocardial infarction, and so there's already no pre-existing imbalances in the autonomic nervous system, and so when they're depressing or affecting one system, it's throwing it out of balance, whereas our system, I like to think that we're looking at a hyperactive sympathetic nervous system and a dysfunctional parasympathetic, and so we're more restoring that balance, and like the changes that are going on within both the stelic anglia and the intrinsic cardiac nervous system have, I feel, have already adapted to these changes in RV function, and so they're more able to accommodate the changes from higher inputs. Okay, thank you. I don't have more questions. Thank you, Jonathan, for an excellent presentation and great discussion. This concludes the 2020 Young Investigators Basic Science Competition. I have some closing remarks. I'd like to thank all the finalists. The Heart Rhythm Society in due course will post the outcome of this competition. However, I'd like to remind you that in our eyes, you're all winners as you've gone through three rounds of selection to reach this point as finalists with very high scores. You should be very proud of your work, and we are indeed proud of your work, too. The pandemic has robbed you of a stage, but hopefully these recordings will do some justice. I would also like to thank the judges for all the hard work they have done to have this competition in spite of the pandemic. Thank you very much, everyone, for participating.
Video Summary
The 2020 Young Investigators Basic Science Competition featured three finalists presenting their innovative research. Dr. Rogers presented his work on using machine learning to improve risk stratification for sudden cardiac arrest. By applying machine learning algorithms to ventricular action potentials, he was able to predict clinical outcomes such as ventricular arrhythmias and death. He also identified specific ion channel abnormalities associated with these outcomes. Dr. Wimbo's presentation focused on the effects of thoracic epidural anesthesia on autonomic and electrophysiological function in the setting of chronic myocardial infarction. His study found that thoracic epidural anesthesia had antiarrhythmic effects, prolonging action potential duration and improving vagal function. Finally, Jonathan Huang presented his research on the effects of thoracic epidural anesthesia on the chronically infarcted porcine heart. His study demonstrated that thoracic epidural anesthesia prolonged atrial and ventricular effective refractory periods, preserved sympathetic and parasympathetic function, and enhanced vagal afferent efferent function. It also suppressed ventricular arrhythmias and reduced heterogeneity in activation. Overall, these presentations showcased the innovative research being conducted in the field of cardiac electrophysiology and highlighted potential new approaches to risk stratification and antiarrhythmic therapies.
Keywords
2020 Young Investigators Basic Science Competition
machine learning
risk stratification
sudden cardiac arrest
ventricular action potentials
ventricular arrhythmias
ion channel abnormalities
thoracic epidural anesthesia
autonomic function
electrophysiological function
Heart Rhythm Society
1325 G Street NW, Suite 500
Washington, DC 20005
P: 202-464-3400 F: 202-464-3401
E: questions@heartrhythm365.org
© Heart Rhythm Society
Privacy Policy
|
Cookie Declaration
|
Linking Policy
|
Patient Education Disclaimer
|
State Nonprofit Disclosures
|
FAQ
×
Please select your language
1
English