| | Identify drug-induced T wave morphology changes by a cell-to-electrocardiogram model and validation with clinical trial dataReceived 15 May 2009 published online 24 August 2009. Abstract BackgroundIncrease of repolarization heterogeneity has been identified as a major factor for drug-induced arrhythmia event like torsade de pointes. In recent years, there have been quite a few efforts for studying T wave morphology changes, hoping to identify more sensitive proarrhythmia electrocardiogram (ECG) biomarkers than QT interval. However, the associations among ECG morphologies and the repolarization heterogeneities are still not clear. MethodA cell-to-ECG model has been built by our group to study relationship between multiple factors of ion channels on the heart tissue and ECG morphology changes measured on the torso. More specifically, we varied both transmural (from Epi to Endo myocardium layers) and apex-to-base heterogeneities by blocking rapid delayed rectifier potassium current (Ikr), slow delayed rectifier potassium current (Iks), and late sodium current (InaL) with different extents on Epi, M, and Endo myocardium. On ECG measurement part, the study was focused on some new morphology-related features including T-peak to T-end (TpTe) interval, T wave flatness, T wave symmetric, and T wave notch. Two types of transmural dispersion of repolarization (TDR) were created: global and localized heterogeneities. Vector magnitude and principal component–based composite leads were formed from multiple chest leads for robustness against large variation of individual lead due to placement and noise issues. Cross-correlation methods were used to determine the relationship of the new ECG morphology features with the heterogeneities. All the ECG morphology measurements were first analyzed with the cell-to-ECG model and then validated with previously acquired clinical trial ECG data (d-sotalol). ResultsThe results based on our cell-to-ECG model showed that the new TpTe interval of the composite signal based on V2, V3, and V4 leads has the correlation coefficients of 0.99 and 0.98 with the simulated global and localized TDR, respectively, highest among other tested ECG parameters. The combined T wave morphology score has the correlation coefficients of 0.98 and 0.92 with the simulated global and localized TDR, respectively. The validation results of d-sotalol show that new TpTe measurement has a correlation coefficient of 0.90 with plasma concentration, and the parameter's correlation with heart rate is 0.02. ConclusionsThe study provided preliminary results showing the usefulness of the cell-to-ECG model for studying relationship between multiple ion-channel factors with ECG morphology changes. The global and localized TDR generate very different T wave morphologies. The newly identified T wave morphology parameters are highly correlated with transmural dispersion and are heart rate independent. Introduction  More studies have shown that the QT interval prolongation induced by drugs is only a surrogate to torsade de pointes (TdP). There are other potential triggering factors of TdP, including early after-depolarizations (EADs) and increase of transmural dispersion of repolarization (TDR).1, 2, 3 Comparing with the other 2 factors, that is, EAD and TDR, QT is a weaker linker to TdP because some studies showed that QT interval prolongation is not highly correlated with the incidence of drug-induced TdP, but EAD and increase of TDR are.2 However, the QT interval remains to be the only electrocardiogram (ECG) marker required for the Thorough QT study for new drug test on cardiac safety, mainly because it can be visualized and edited by overreaders, although measuring the QT interval accurately and consistently remains to be a challenge. Early after-depolarization can be indirectly assessed if it triggers premature ventricular beat, but otherwise, it is difficult to determine from the surface ECG. Over the years, some studies have suggested that T-peak to T-end interval (TpTe) might be related to TDR, but TpTe has not been accepted as a valid test mainly due to 2 reasons: (1) lack of solid scientific proof to link TDR to TpTe of ECG and (2) large variability of T-peak positions from lead to lead, making it very inconsistent. Our previous study based on our cell-to-ECG model and a robust measurement of T wave parameters has shown that the TpTe based on vector magnitude (VM) of 12 leads has the highest correlation to TDR among other ECG parameters, which include Q- to T-end dispersion, Q- to T-peak dispersion, TpTe from individual lead (v3, v5), and principal-component-analysis (PCA).4 In the current study, we expanded ECG parameters into more T wave morphology–related parameters including T wave flatness, T wave symmetric, and T wave notch.5 The focus from the ECG here is to identify new T wave morphology that may present stronger link to the occurrence of TDR and EADs than measuring the QT interval duration. We also formed global and localized TDR in our cell-to-ECG model. The purpose is to examine these 2 TDR's effects on the ECG morphology parameters. The cell-to-ECG model consists of a cell model and a forward model, which converts cell's action potentials to a torso potential—ECG. Several excellent cell models have been published over the years.6, 7, 8 We adopted the model created by ten Tusscher's group8, 9, 10 because it is based on human heart cells and also includes both Ikr and Iks ion channels, which can be used to simulate drug-induced action potential (AP) prolongation, hence dose-dependent QT prolongation. A late sodium model (InaL) was added to the previous model to study InaL inhibition effect.11, 12, 13 Methods The proposed cell-to-ECG model consists of 4 major portions as previously described,4 which includes cell model, propagation algorithm, finite element and boundary element (FEM-BEM) methods, and calculating body surface potential map. The cell model portion calculates up to 13 ion-channel currents and also generates transmural and base-to-apex heterogeneities. Finite element and boundary element method is used to handle forward model calculation and also tissue geometry (heart, lung, torso, etc) and conductivity. The propagation algorithm controls propagation speed on different myocardium, Purkinje fibers, and fiber orientation for anisotropic propagation. Body surface potential map is computed based on the derived conversion matrix from previous 3 portions. Cell model The cell model is based on the latest human cell model.9, 10 There are 13 ion channels built into this model, where rapid and slow delayed rectifier potassium channels (Ikr and Iks) and late sodium channel (InaL), inward rectifier potassium (Ik1), and transient outward channel (Ito) are most relevant to this study. In this latest model, a Markov process is used to describe the dynamics of Ikr. More specifically, the Ikr status is defined by 5 states as where C denotes the closed state, O the open state, and I the inactivated state. Each state is controlled by transition rates αi and βi, which are voltage- and temperature-dependent. 14 Take the open state O as an example: The Ikr current is formulated as: where gKr is the conductance of the channel, Vm is the transmembrane potential, and EK is the Nernst potential for potassium. The ion-current block factor and heterogeneity are introduced by scaling gKr. Late sodium channel The late sodium current model included was introduced by Xia et al,13 which takes 2 gates formulation: where mL is an activation gate and hL is an inactivation gate. Each of these gates is governed by Hodgkin-Huxley-type equations for gating variables and characterized by a steady-state value ( mL,∞ and hL,∞) and a time constant ( τmL and τhL) for reaching this steady-state value, both of which are functions of transmembrane potential ( Vm). GNaL is the maximum value of late sodium conductance, GNaL = 0.0065 mS/μF, and ENa is the Nernst potential for sodium. Forward model A bidomain model–based FEM-BEM coupling formulation in the cardiac electric field is developed.15 The formula to solve forward model is divided into 2 parts: inside myocardium and from the heart surface to the torso. For the inside myocardium portion, FEM method is applied to consider the anisotropy of myocardium. The equation can be described as: where κH is the bulk conductivity of the heart, φFEM the extracellular electric potential, σin the effective intracellular conductivity, Vm the transmembrane action potential, and r denotes the field point. From the heart surface to torso, BEM method is applied for higher computational efficiency, which can be described by Eq. (4): where κT is the isotropic conductivity of the torso and φBEM the potential on both heart and torso boundary. By solving Eqs. (3), (4), a transfer matrix C can be derived to calculate torso potential from cell's action potential: Twelve-lead ECGs can be extracted from φtorso.16 Definition of heart repolarization dispersion There are 3 dispersions defined in the study: (1) left ventricular (LV) versus right ventricular (interventricular [circumferential] dispersion), (2) LV base versus LV apex (basoapical [transepicardial] dispersion), and (3) across LV wall (transmural). Transmural dispersion is our focus because of its link to TdP demonstrated in the preclinical studies mentioned previously. Fig. 1 shows a cross-section of computed tomography–scanned 3-dimensional ventricle and the representative myocardium cells in Epi, Endo, and Mid myocardium layers, where Epi cell action potential duration (APD) is the shortest, and M or Endo cells have the longest APD. The layer is normalized into a range between 0 and 1. The right ventricular–septum surface is taken as epicardium layer. In this study, to define TDR under the influence of drug-induced QT prolongation, Ikr block factor and InaL are used to control Epi, M, and Endo myocardium layers separately. After 3 anchor points across the ventricle wall are defined, a smooth second-order curve is used to interpolate the Ikr block factor to all other layers as shown in Fig. 2. In the simulation, the Ikr block ratios were increased for all 3 cell layers but with different incremental factors. The end results of this simulation were the increase of the TDR as well as APD. Global versus localized TDR The global TDR is formed by applying block factors displayed in Fig. 1 to whole heart from apex to base, whereas the local TDR is created by applying the transmural dispersion as shown in Fig. 2 to only defined local zone. In the study, an anterior zone was selected. The localized TDR was generated by only applying gradient of Ikr current block as shown in Fig. 2 across the anterior wall. Measurements of ECG parameters Standard 12-lead ECG is selected from model-generated body surface potential map. Some ECG parameters were defined in our previous study, including the VM of all 12 leads.4 Also, the following new ECG morphology parameters were defined: 1.The anterior VM ECG is calculated by the root mean square formula:and the other VM is based on all 12 leads. The TpTe of VM is calculated from the VM signal by detecting the peak in the T wave region and the end of VM signal. 2.Morphology measuresThree T wave morphology features, that is, asymmetry, notch, and flatness, are defined in the previous study,5 as shown in Fig. 3. Here is a simplified description for each morphology parameter:Asymmetry:The difference in slopes of the ascending and descending parts of the T wave was considered a measure of asymmetry. The slopes at each point of the descending part of the T wave were mirrored to be compared with the slopes at corresponding points of the ascending segment. Asymmetry was defined as the average squared difference between the slope segments. Notch:A curvature signal that measures how fast a curve is changing direction at a given point was obtained from the first and second derivatives of T waves. This signal indicates the presence of notches because deflections in curvature correspond to deviations from the normal smooth progress of a T wave. The magnitude of a notch was measured on a unit amplitude T wave and assigned to 1 of 3 categories: no deflection = 0, moderate notch (perceptible bulge) = 0.5, and pronounced notch (distinct protuberance above the apex) = 1.0. Flatness:T wave amplitudes were normalized to yield a unit area under the T wave curve. From this signal, flatness was calculated using instantaneous amplitudes, the weighted mean, and the variance, analogous to the way kurtosis is calculated from a sample distribution. A morphology combination score (MCS) was developed to combine T wave asymmetry, flatness, and notching where  Cardiac safety clinical trial data (d-sotalol) Electrocardiogram data from the d,l-sotalol group was recorded with 12-lead digital Holter at a sample rate of 180 Hz, described in Sarapa et al.17 Each 10-second ECG segment extracted from Holter was resampled to 500 Hz. From each subject, 22.5-hour recordings were obtained per day. At approximately 30-minute intervals, 12-lead ECGs of 10-seccond duration were derived from Holter. To put HR-RR hysteresis effect in consideration, a representative 10-second segment was extracted from Holter, only if it was preceded by 1 minute of stable heart rate. All representative median beats were derived from such stable and noise-free 10-second Holter segments. Experiment design and statistical analysis Two types of transmural dispersion were tested by setting global and localized TDR as described above. For each setting, the gradient of Ikr blocks was created to 3 myocardium layers, as shown in Fig. 2. For these 2 tests of dispersion, 8 incremental cases of AP and ECG were generated. The TDR for each case was calculated as the difference between maximum APD of Endo and M layers and the minimum APD on the Epi layer of the ventricle. The correlation coefficients between TDR and the ECG parameters defined previously were calculated for each type of TDR. The study also compared Q- to T-peak (QTpeak) with TpTe changes when TDR was increased. Results Compare Q- to T-peak changes to TpTe changes Fig. 4 shows the results of comparing QTpeak and TpTe changes when TDR increase. On the left are QTpeak and TpTe changes for global TDR cases; on the right are changes for localized TDR cases. It is shown that both QTpeak and TpTe are increased for global TDR cases, whereas only TpTe is increased for localized TDR cases. This phenomenon can be further explained on the morphology changes shown in Fig. 5, where we are comparing ECG morphology changes on global and localized TDR cases with Ikr block by 70% on Endo and M myocardial layers and 40% on Epi cardial layer. In Fig. 5, row 1 is AP profiles of whole ventricle: the earliest and the latest AP endings; row 2 is the AP profile of anterior area: the earliest Epi ending to the latest M and Endo cardial endings; row 3 is the VM of anterior zone; row 4 is the VM of 12 leads; row 5 is ECG of V3. Fig. 5A displays baseline ECG without Ikr block. Both global and localized Ikr blocks result in APD prolongation and triangular shape on phases 3 to 4 of AP. For global TDR case displayed in Fig. 5B, both QTpeak and TpTe are prolonged, and T wave notch appears before the major T peak, whereas for localized TDR cases in Fig. 5C, TpTe change much more than QTpeak change, and T wave notch appears after the major T peak as indicated by the arrow sign. The major T peak is most correlated with the earliest AP ending on the whole heart. In the global TDR cases, that AP ending is prolonged with the TDR increases; thus, QTpeak is also prolonged. On the other hand, in the localized TDR cases, the earliest AP ending on the whole heart is not changed because only local zone AP is affected by the TDR change; thus, QTpeak is not changed significantly. Correlation of transmural dispersion with the T wave morphology measures In this experiment, TDR was created by applying different Ikr block to Endo, M, and Epi cells, as shown in Fig. 2, where Ikr was blocked from 20% to 90% for Endo cardial cells 10% to 70% for M cells, and 5% to 30% for Epi cardial cells. The extracellular potassium concentration [K+]° was also altered from 5.4 to 3.9 mmol/L gradually. Fig. 6 shows the correlation of TDR and TpTe of VM_ant. The TpTe of VM_ant is correlated with both global and localized TDR cases well (r2 = 0.99 and 0.97, respectively), although it shows even higher correlation with global TDR than localized TDR changes. Fig. 7 shows the correlation of TDR and the T wave MCS. The cross-correlation coefficients for the morphology score and global localized TDR cases are 0.98 and 0.92, respectively. Late sodium channel's effects on ECG morphology Fig. 8 shows how TDR is changed by increasing late sodium current (InaL) on different layers. The left side of the figure shows AP profile and ECG morphology change when InaL increases 50% on Endo and M cells and 20% on Epi cells. The right side shows the QTpeak and TpTe change and the correlation of TDR versus TpTe (r2 = 0.99). Validation results of T wave morphology measurements with d-sotalol clinical trial data Fig. 9 shows the measurements of TpTe of VM in d-sotalol clinical trail cases with 24-hour tracking.17 There are high-dose (320 mg) and low-dose (160 mg) cases. The upper traces are mean compound concentration curves of 24 hours, and the lower traces are TpTe curves. The correlation coefficients between the concentration and TpTe are 0.9 and 0.91 for low and high doses, respectively. Fig. 10 shows the measurements of the T wave morphology of the PCA vector for the same d-sotalol clinical trail data set. The upper traces are combination score (MCS) curves on the left and QTcF curve on the right, and the lower traces are results of flatness, asymmetry, and notches of T wave. New T wave morphologies follow the concentration curve shown in Fig. 9 very well. Discussions From this study, it is shown that the cell-to-ECG model, as an analytical tool, can help us to understand meaning of different ECG morphologies related to cell and tissue level abnormalities. Transmural dispersion of repolarization has been shown in other biological laboratories to contribute to drug-induced severe arrhythmias like TdP. Our new study showed that TpTe correlates with global and localized TDR very well, although more significant TpTe changes were found associated with localized TDR increases. At the meantime, the new T wave morphology parameters correlate with global TDR much better than localized TDR changes. The simulation study also showed that the global and the localized TDR increase generated very different T wave morphologies. T-peak to T-end interval of VM signal is also proven to be a useful measure of TDR, where localized TDR generated longer TpTe changes than global TDR. The ratio between TpTe and Q-to-T peak could be a potential indicator for differentiating global versus local TDR changes. At the meantime, the more general morphology descriptors such as symmetry, flatness, and T wave notch are also good indicators of the global TDR changes. Because our previous study has shown that QT interval dispersion and QTpeak interval dispersion are not correlated with TDR changes,4 we can see that morphology features tested in this study can be better ECG biomarkers for detecting repolarization abnormality, more specifically, for drug-induced repolarization changes. Studies by other researchers have shown different drug-induced ECG morphology changes.18, 19, 20 This study further shows different ECG patterns when TDR is induced by different ion currents, such as Ikr and InaL. Late sodium current becomes more interesting lately because some InaL inhibition drugs like renolazine showed that inhibition of late sodium current can reduce TDR even QT interval is prolonged by partial Ikr block effect of the same drug.21 Our study shows that increase of InaL can indeed contribute to both QT interval prolongation and morphology changes. However, the morphology changes caused by InaL are very different from the changes caused by Ikr block. The most interesting morphology changes in this study are shown in Fig. 6, where notches are formed when TDR is increased more dramatically. Also, a U wave pattern is formed in the localized TDR change as shown in Fig. 6B). The similar T notches are also observed often in high-dose d-sotalol ECGs. This similarity can help us to assume that high-dose d-sotalol causes significant global TDR increase. This morphology pattern can also lead us to ask these 2 questions again: (1) How U wave in ECG is defined? (2) Is U wave part of repolarization process? For the first question, it is still not very conclusive, but we do need to be very careful on differentiating T wave and U wave. For the second question, the study clearly indicates that whole later part of T wave belongs to the repolarization process and should be counted as part of QT when we measure QT interval. The lower amplitude of the later part of the T wave is mainly due to triangular wave pattern on action potentials caused by Ikr and [K+]° changes. We also learned from this example that T wave notches can be formed either before or after major T wave peaks. Another intriguing question is what T peak represents for. Previous research based on animal wedge preparation has indicated that T peak is corresponding to the early action potential ending of the epicardial layer.22 A truly elegant study as it is, however, shows that the wedge study is limited by a small region of heart tissues. Our model, based on the whole human heart, could provide a more complete view to study the relationship of T peak with the activities of action potentials on the heart. Based on our preliminary study, we can see that T peak is corresponding to the early ending on a whole heart region, not necessary of the local epicardial action potential ending, Fig. 6B shows that T peak follows closely with global AP ending on row 1, instead of local ending on row 2. The correlation coefficient between global early AP ending and T peak (r2 = 0.98), a very high correlation. The clinical data in this study clearly showed that TpTe is highly correlated with the plasma Ikr concentration (Fig. 9); hence, the drug effects of QT prolongation and dispersion increase. The new T wave morphologies also showed similar correlation. The clinical results are consistent with the simulation of the whole heart model. However, we also observed that in the actual clinical ECG results, there is more fluctuation on the measurements of TpTe and the T wave morphology than the modeled cases, mainly due to various noises in the ECG data. It also reminds us the importance to have robust ECG measurements for the new features. This is the major reason why the VM of multilead ECG was used in this study instead of single-lead ECG. We would also indicate that there are several limitations of the current simulation work: (1) the limited heart and torso geometries tested in this study. We have used 2 different heart and torso geometries scanned from computed tomography. Due to large variations to the heart and the torso sizes and their relative coordination, ECG morphologies can also have large variations. Therefore, we should be careful to generalize the cases in this study; (2) the propagation model used in the simulation was referenced from published special cases; and (3) the cell models of different ion channels, especially human cell models, are still evolving. Acknowledgment The authors thank Claus Grass and Mads Anderson of Aalborg University for providing valuable help on the T wave morphology analysis. References 1. 1Antzelevitch C, et al. The M cell: its contribution to the ECG and to normal and abnormal electrical function of the heart. J Cadiovasc Electriphysiol. 1999;10:1124. 2. 2Belardinelli L, Antzelevitch C, Vos M. Assessing predictors of drug-induced torsade de pointers. Trends Pharmacol Sci. 2003;24:619. MEDLINE |
CrossRef
3. 3Antzelevitch C, et al. Electrophysiological effects of Ranolazine, a novel antianginal agent with antiarrhythmic properties. Circulation. 2004;110:904.
CrossRef
4. 4Xue J, et al. Study of repolarization heterogeneity and electrocardiographic morphology with a modeling approach. J Electrocardiol. 2008;41:581. Abstract | Full Text |
Full-Text PDF (572 KB)
|
CrossRef
5. 5Struijk JJ, Kanters JK, Andersen MP, et al. Classification of the long QT syndrome based on discriminant analysis of T-wave morphology. Med Biol Eng Comput. 2006;44:543. MEDLINE |
CrossRef
6. 6Luo C, Rudy Y. A dynamic model of the cardiac ventricular action potential. I. Simulations of ionic currents and concentration changes. Circulation. 1994;74:1071. MEDLINE 7. 7Priebe L, et al. Simulation study of cellular electric properties in heart failure. Circ Res. 1998;82:1206. MEDLINE 8. 8ten Tusscher KHWJ, et al. A model for human ventricular tissue. Am J Physiol Heart Circ Physiol. 2004;286:H1573. MEDLINE |
CrossRef
9. 9ten Tusscher KHWJ, Panfilov AV. Alternans and spiral breakup in a human ventricular tissue model. J Physiol Heart Circ Physiol. 2006;291:H1088. 10. 10Fink M, Nobleb D, et al. Contributions of HERG K+ current to repolarization of the human ventricular action potential. Prog Biophys Mol Biol. 2008;96:357.
CrossRef
11. 11Maltsev VA. Ultraslow inactivating sodium current in human ventricular myocytes. Circulation. 1998;98:2545. MEDLINE 12. 12Maltsev VA, et al. Studies in failing human myocardium. J Mol Cell Cardiol. 2001;33:923. Abstract |
Full-Text PDF (149 KB)
|
CrossRef
13. 13Xia L, Zhang Y, Zhang HG. Simulation of Brugada syndrome using cellular and three-dimensional whole-heart modeling approach. Physiol Meas. 2006;271125. 14. 14Hille B., Ionic channels of excitable membranes. Sunderland (Mass):Sinauer; 2001. 15. 15Fischer G, et al. A bidomain model based BEM-FEM coupling formulation for anisotropic cardiac tissue. Ann Biomed Eng. 2000;28:1229. MEDLINE |
CrossRef
16. 16Malmivuo J, Plonsey R. Bioelectromagnetism: principles and applications of bioelectric and biomagnetic fields. Oxford University Press; 1995;. 17. 17Sarapa N, Morganroth J, Couderc JP. Electrocardiographic identification of drug-induced QT prolongation: assessment by different recording and measurement methods. Ann Noninvasive Electrocardiol. 2004;9:48.
CrossRef
18. 18Vaglio M, Couderc JP, McNitt S, Xia X, Moss AJ, Zareba W. A quantitative assessment of T-wave morphology in LQT1, LQT2, and healthy individuals based on Holter recording technology. Heart Rhythm. 2008;5:11. Abstract | Full Text |
Full-Text PDF (301 KB)
|
CrossRef
19. 19Kanters JK, Fanoe S, Larsen LA, Bloch Thomsen PE, Toft E, Christiansen M. T wave morphology analysis distinguishes between KvLQT1 and HERG mutations in long QT syndrome. Heart Rhythm. 2004;1:285. Abstract | Full Text |
Full-Text PDF (266 KB)
|
CrossRef
20. 20Padrini R, Butrous G, Statters D, Camm AJ, Malik M. Morphological algebraic models of the TU-wave patterns/in idiopathic long QT syndrome. Int J Cardiol. 2001;77:151. Abstract | Full Text |
Full-Text PDF (483 KB)
|
CrossRef
21. 21Lin W, et al. Antiarrhythmic effects of ranolazine in a guinea pig in vitro model of long-QT syndrome. JPET. 2004;310:599. 22. 22Yan GX, Antzelevitch C. Cellular basis for the normal T wave and the electrocardiographic manifestations of the long-QT syndrome. Circulation. 1998;98:1928. MEDLINE a Diagnostic Cardiology, GE Healthcare, Wauwatosa, WI, USA b GE Global Research, Shanghai, China  Corresponding author. Diagnostic Cardiology, GE Healthcare, 9900 Innovation Drive, Wauwatosa, WI 53226, USA.
PII: S0022-0736(09)00279-9 doi:10.1016/j.jelectrocard.2009.07.010 © 2009 Elsevier Inc. All rights reserved. | |
|
FULL TEXT
