Author + information
- Received February 4, 2015
- Revision received April 3, 2015
- Accepted April 9, 2015
- Published online June 1, 2015.
- Nicholas Jackson, MD∗,
- Stéphane Massé, MASc∗,
- Nima Zamiri, MD∗,
- Mohammed Ali Azam, MBBS, PhD∗,
- Patrick F.H. Lai, MSc∗,
- Marjan Kusha, MEng∗,
- John Asta∗,
- Kenneth Quadros, MD∗,
- Benjamin King, MD∗,
- Peter Backx, DVM, PhD∗,
- Raymond E. Ideker, MD, PhD† and
- Kumaraswamy Nanthakumar, MD∗∗ ()
- ∗University Health Network, Toronto, Ontario, Canada
- †University of Alabama at Birmingham, Birmingham, Alabama
- ↵∗Reprint requests and correspondence:
Dr. Kumaraswamy Nanthakumar, The Hull Family Cardiac Fibrillation Management Laboratory, Division of Cardiology, University Health Network, Toronto General Hospital, 150 Gerrard Street West, GW3-526, Toronto, Ontario M5G 2C4, Canada.
Objectives This study sought to determine the characteristics of human LDVF, particularly as it contrasts with short-duration VF (SDVF), and evaluate the role of Purkinje fibers in its maintenance.
Background The electrophysiological mechanisms of long-duration ventricular fibrillation (LDVF) have not been studied in the human heart.
Methods VF was induced in 12 human Langendorff hearts, and the hearts were examined from initiation to LDVF (10 min). Endocardial, epicardial, and transmural plunge needle mapping were performed on the hearts. Simulated LDVF was studied in canine hearts to determine the potential role of Purkinje fiber automaticity.
Results The mean age at transplant was 48 ± 20 years, and the mean ejection fraction was <20%. The mean cycle length of local activation times on the endocardium was 252 ± 66 ms in SDVF and 441 ± 80 ms in LDVF (p = 0.0002). On the endocardium and the epicardium in LDVF, cycle length was 441 ± 80 ms and 590 ± 88 ms, respectively (p = 0.0002). No endocardial to epicardial activation frequency gradient was seen in SDVF. Simultaneous transmural needle activation was most common in SDVF, whereas endocardial to epicardial activation was most common in LDVF (47.7% and 38.8% of activations, respectively [p = 0.031]). Re-entry was less common in LDVF, and over time, wave break (i.e., nontransmural propagation of wave fronts) developed. Isochronal maps of the left ventricular endocardium in LDVF identified Purkinje potentials as preceding and predominating endocardial activations. In explanted canine heart preparations, rapid pacing led to spontaneous Purkinje fiber activity that was dependent on pacing rate and duration.
Conclusions LDVF in human hearts is characterized by focal endocardial activity with mid-myocardial wave break and not by re-entry. This arrhythmia is modulated by rapid activations in early VF that lead to spontaneous Purkinje fiber activity.
Survival outcomes after cardiac arrest due to ventricular fibrillation (VF) decrease exponentially as time to defibrillation increases (1,2). Nonetheless, defibrillation can be performed successfully for some patients after 10 min or more of VF, and some patients survive with relatively good clinical outcomes (1). Evaluation of the electrophysiological characteristics of long-duration ventricular fibrillation (LDVF) using animal models has shown periods of organized and synchronous endocardial activation and has implicated abnormal automaticity or triggered activity from Purkinje fibers as possible drivers of VF at this stage (3). These studies have shown endocardial to epicardial activation frequency gradients (4) and earliest activation in Purkinje fibers after LDVF defibrillation failures (5), and they have shown that chemical ablation of the Purkinje system with Lugol solution leads to earlier termination of VF and loss of the endocardial to epicardial activation frequency gradients (4).
In contrast, the characteristics of short-duration (<3 min) ventricular fibrillation (SDVF) have been studied in detail in human in vivo and ex vivo models. These studies suggest that SDVF is most frequently characterized by transmural scroll wave activation, with intramural re-entry most often localizing to regions with greater fibrosis (6,7). In the clinical setting, it is at these earlier phases of cardiac arrest (SDVF) that pharmacological therapies to improve defibrillation efficacy have been studied (8,9). If VF is maintained by different mechanisms as the rhythm progresses over time, then alternative therapeutic interventions may become important in improving patient survival.
In this study, we examined myocardial activation patterns in SDVF and LDVF in myopathic Langendorff-perfused human hearts, with particular attention to transmural activation gradients and the role of Purkinje fibers in LDVF. We hypothesized that focal activity from the endocardium aids in maintaining LDVF in its later stages and in creating endocardial to epicardial activation frequency gradients.
Initially we sought to characterize the transmural activation patterns of VF in 12 cardiomyopathic Langendorff human hearts from onset to LDVF (10 min). Particular attention was given to the role of Purkinje fibers and the presence of re-entry or focal activity as VF progressed over time. A dog model was used to test the effect of rapid activations (simulated VF) on Purkinje fibers because the Purkinje system in dogs is most similar to that in humans compared to other mammals and allows for the isolation and mapping of individual Purkinje fibers.
Human LDVF study
The study protocol was approved by the University Health Network Human Research Ethics Board and complies with the Declaration of Helsinki. Twelve patients with cardiomyopathy requiring heart transplant consented to use of their explanted hearts for the study. Global transmural plunge needle mapping was performed in 11 of the 12 hearts, endocardial activation was mapped with a balloon array (unipolar and bipolar recording) in 5 hearts, and epicardial activation mapping with a sock array was performed in 6 hearts. Because of limitations with signal processing and the feasibility of array positioning, all mapping techniques were not used concurrently on each heart.
Human Langendorff hearts were perfused with modified Tyrode solution via the coronary arteries, and VF was induced by burst pacing from the right ventricular apex. With VF induction, perfusion was halted and pseudosurface electrocardiography was monitored to confirm irregular activity and heart rate >220 beats/min (6). Details of the human Langendorff methodology and VF mapping arrays are included in the Online Appendix.
During VF mapping, the local activation times (LATs) on unipolar recordings were taken as the maximum negative dV/dt (change is voltage/change in time) at each electrode, provided it was at least –0.5 mV/ms (10). Early VF recordings were taken at 3 s after onset (6), and LDVF recordings were taken at up to 10 min (3,4,10). Purkinje fiber activations were initially identified on the left ventricular (LV) septum with the bipolar endocardial balloon during basal pacing. Capture of the His-Purkinje system allowed identification of high-frequency potentials (1 to 2 ms duration) (11) preceding local ventricular activation (Figure 1A). The corresponding electrodes were later examined for Purkinje potentials during VF on the bipolar and unipolar needles and endocardial balloon arrays. Local activation time for bipolar recordings was taken as the peak of the positive deflection (12).
Transmural VF mapping
Transmural needle activations were examined for simultaneous, endocardial to epicardial, epicardial to endocardial, and nonuniform multidirectional patterns (6). A <10 ms difference among LATs of all 4 electrodes along a needle defined simultaneous activation (6). To meet the criteria for uniform transmural activation, at least 3 of 4 electrograms had to be in the appropriate sequence, with the fourth not more than 20 ms out of sequence to allow for some heterogeneity in conduction and slanting or curved wave front propagation. Nonuniform multidirectional patterns were defined as chaotic patterns that did not fit the sequences previously described. When analyzing the needle data during LDVF, wave break (i.e., failure of a wave front to propagate transmurally) was seen (13). Activation patterns with wave break were identified when 2 or 3 local needle activations were seen, but propagation of activation to the remaining needle pole(s) was not seen. Examples of how transmural and nontransmural wave front propagations were classified are shown in Figures 2B and 2C.
Cycle length determination and re-entry detection
To assess for re-entry and to compare activation sequences, a cycle length (CL) for VF was defined (6). The median of the number of LATs on each needle was used to define the number of “beats” and defines the context in which the term “beat” is used hereafter in this paper. The number of beats divided by the duration of the analyzed VF segment determined the CL (6) (CL and median activation times for beats 11 and 12 in SDVF are shown in Online Figure 1). Re-entry was evaluated in 2 orthogonal planes (both parallel and perpendicular to the epicardium and endocardium). Each needle electrode in turn was examined for intramural re-entry involving that electrode plus the 9 adjacent electrodes around it. To meet the criteria for re-entry, one full rotation was required with progression of local activation on at least 75% of the 9 electrodes, spanning ≥85% of the CL of the beat (6). Re-entry was assessed in this way at all 4 layers of the myocardium (endocardial to epicardial), in larger groups of needles (with 16 and 25 needles), and then orthogonally along the length of the needle as well.
Induction of focal activity from Purkinje fibers in LDVF from rapid myocardial activation in early VF
To explore the possibility that focal activity from Purkinje fibers in late VF could be promoted by rapid activation during early VF, we used a separate preparation with which Purkinje and myocardium can be visually separated and recorded. Explanted canine hearts were dissected into islands of ventricular myocardium joined by strands of Purkinje fibers and placed in a tissue bath (Figure 1B). A glass microelectrode was used to impale muscle and Purkinje fibers to record action potentials (Figure 1B, inset). Bipolar electrodes were placed onto ventricular myocardium and Purkinje fibers to record local electrical activity. Rapid pacing of the myocardium at 6 Hz for 5 min was used to simulate LDVF (this rate approximated the dominant frequency of canine LDVF found by Newton et al. ) in the presence of ischemia (no perfusion) and 0.2 ml of isoproterenol. After this, burst pacing of the ventricular myocardium at progressively faster rates (1 to 5 Hz CLs) and for progressively longer durations (5 to 20 s) was performed to look for subsequent spontaneous Purkinje fiber activation. Because Purkinje strands were separated from adjacent myocardium, only true Purkinje activations could be detected at the corresponding bipolar electrodes.
Analyses were performed with SAS version 9.1 (SAS Institute, Cary, North Carolina), and results are expressed as mean ± SD where stated. Comparison of activation patterns in SDVF and LDVF in Figure 2 and comparison of re-entry incidence in Table 1 were performed with a Wilcoxon signed rank test because of repeated measures. Comparison of endocardial and epicardial CLs in SDVF and LDVF was performed by 2-way repeated measures analysis of variance with Bonferroni correction for multiple comparisons. A p value of <0.05 was considered statistically significant.
The baseline characteristics of the 12 patients whose hearts were used in this study are summarized in Online Table 1. Patients consisted of 5 women and 7 men with a mean age of 48.4 ± 19.6 years. The predominant disease was dilated cardiomyopathy with ejection fraction <20%, and the mean LV internal dimensions in diastole and systole were 61.9 ± 2.3 mm and 55.3 ± 2.2 mm, respectively.
Transmural activation sequences during human SDVF and LDVF
Figure 2A shows the relative frequency of different transmural activation sequences in SDVF and LDVF from plunge needle data. The most frequent activation sequence seen (of those defined in Figure 2) in SDVF was simultaneous activation of all 4 needle electrodes (47.7% of activations), whereas the most frequent activation sequence seen in LDVF was endocardial to epicardial activation (38.8%; p = 0.031). Including endocardial to epicardial activation with wave break, a total of 54.2% of all needle activations appeared to originate at the endocardium in LDVF compared with 14.5% (p = 0.031) in SDVF. Wave break overall was a much more common phenomenon in LDVF than in SDVF (27.9% vs. 3.3%; p = 0.031). These differences in transmural activation between SDVF and LDVF are further shown in the spatiotemporal activation plot in Online Figure 1.
Differential activation rates during human LDVF
The cycle lengths of SDVF and LDVF at the endocardium and epicardium from endocardial balloon and epicardial sock arrays are shown in Table 2. Activation frequency gradients can be seen between LDVF at the endocardium and LDVF at the epicardium (mean CL 441 ± 80 ms vs. 590 ± 88 ms, p = 0.0002) and between SDVF and LDVF at both the endocardium and epicardium. These findings are consistent with the longer mean CLs seen in LDVF than in SDVF and with the prevalence of endocardial to epicardial activation with wave break seen on plunge needle mapping (Figure 2A).
Re-entry during human LDVF
The incidence of re-entry in SDVF and LDVF at each of the 4 myocardial layers during a 3-s period of VF is shown in Table 1. Overall, a greater number of wave fronts met the criteria for re-entry in SDVF than in LDVF (15.8% vs. 4.2%, p = 0.035). Re-entry along the length of the needle transmurally was seen only 2% of the time in SDVF and never in LDVF. A greater number of “beats” during 3 s of SDVF led to a greater number of total wave fronts assessed for re-entry in SDVF than in LDVF (171 vs. 72 beats).
Purkinje activity during human LDVF
Whereas frequent Purkinje-like potentials could be identified in all hearts during LDVF, Purkinje potentials were only clearly identified at baseline in hearts 3 and 4 (from Online Table 1), so these hearts were primarily used in constructing Figures 3 and 4⇓. Examples of discrete Purkinje potentials preceding local ventricular activation are shown on bipolar endocardial mapping in LDVF (arrows) in Figure 3B. Four septal splines of the bipolar endocardial array are shown.
Isochronal maps of the corresponding endocardial activation are shown in Figure 3A with earliest Purkinje activation represented by a star. Rapid endocardial activation can be seen to spread out across the septum via the Purkinje network, with latest activation consistently at the basal and lateral endocardium. On beats 5 and 6, separate wave fronts originating in ventricular myocardium can be seen around splines 2 and 3 of the array and appear to contribute to even more rapid global endocardial activation. Endocardial activation on these beats occurs within 100 ms.
Figure 4A shows unipolar endocardial and epicardial activation every 2 min from induction to 10 min of VF. Sharp and discrete Purkinje potentials can be seen more frequently and with increasing regularity on the septal endocardium as VF progresses. A significantly greater activation rate on the endocardium with low-frequency, longer CL signals on the epicardium shows that the endocardium drives LDVF in myopathic human hearts with spontaneous endocardial Purkinje activity.
Isochronal maps of endocardial needle activation at 10 min of VF are shown in Figure 4B. These maps correspond to the isochronal maps in Figure 3A; however, the true septum was not mapped by the transmural plunge needles and is not present on the left-hand side of the maps. Purkinje potentials adjacent to the septum and more laterally can be seen again to correspond with the points of earliest endocardial activation on beats 3, 4, 5, 6, and 9. Multifocal activations are also seen on beats 2, 3, 6, and 8 with variable wave front propagation.
Focal activity from Purkinje fibers in LDVF is induced by rapid activations during early VF
To determine the mechanism for spontaneous focal Purkinje fiber activations in LDVF, Purkinje fibers were isolated from canine myocardium. During simulation of VF by rapidly pacing canine ventricular myocardium, 1:1 capture of myocardium and adjacent Purkinje tissue can be seen initially (Figure 5A). After 3 min of pacing, 1:1 capture of local Purkinje tissue continues with variable and significantly less-frequent capture of ventricular myocardium (Figure 5B). No ventricular myocardial capture was seen beyond 4 min.
After a rapid burst pacing protocol for 5 s, a single Purkinje extrasystole can be seen that conducts to adjacent myocardium. When this burst pacing protocol is continued for 20 s, multiple Purkinje extrasystoles are induced that then conduct to adjacent ventricular myocardium (Figures 5B and 5C). A rate-dependent aspect to this phenomenon is also shown in Figure 5D, when more frequent spontaneous Purkinje extrasystoles are seen as the CL of burst pacing decreases.
This study demonstrates the following findings in isolated human hearts. An endocardial to epicardial gradient in activation rate develops during LDVF in humans. The dominant pattern of activation is not re-entry or scroll wave activation but predominately activation from the endocardium with increasing breakdown of wave front propagation toward the epicardium. Focal Purkinje potentials precede local ventricular activation on the endocardium. Our experiments are consistent with the hypothesis that this focal activity is modulated by rapid endocardial activations during SDVF. Together these findings suggest that during the development of LDVF in humans, a progressive change occurs from a rhythm driven by re-entry (6) to one driven by focal activations from Purkinje fibers on the endocardium. Although this study was performed in cardiomyopathic human hearts, the findings are consistent with those from noncardiomyopathic animal heart preparations in previous studies (3–5). This is a mechanistic evaluation of a rhythm that is responsible for the majority of sudden cardiac deaths (15) and may have implications for different treatment strategies for VF in its later stages.
Transmural activation sequences during human SDVF and LDVF
The incidence of simultaneous transmural needle activations in SDVF in this study was similar to that found by Nair et al. (6) (48.7% (6) vs. 47.7%). The incidence of re-entry in SDVF was also similar (14.3% (6) vs. 15.8%) and is consistent with the transmural scroll waves or “mother rotors” that were seen migrating though the myocardium in SDVF and giving rise to multiple smaller chaotic wave fronts (6). In LDVF, however, wave break or nontransmural propagation of wave fronts occurred 27.9% of the time and the re-entry incidence was only 4.2%, suggesting that transmural scroll wave activation is not a predominant feature in human LDVF. In Online Figure 1, LDVF frequently displays a different number of beats with different median activation times across adjacent plunge needles, which is also inconsistent with regional organization from migrating transmural scroll waves.
In LDVF, the greater presence of endocardial to epicardial needle activation patterns (54.2% of all activation patterns) and the more rapid CL on the endocardium compared with the epicardium (441 ± 80 ms vs. 590 ± 88 ms) suggests that this phase of the arrhythmia is primarily driven by the endocardium, as has previously been reported in dogs but not pigs (14,16). The presence of wave break in 27.9% of all wave fronts was also a unique finding to LDVF in this study. Endocardial to epicardial activation with wave break (seen 15.4% of the time) predominantly accounts for the endocardial to epicardial activation rate gradient seen in LDVF (Table 2). This development of an endocardial to epicardial activation rate gradient likely relates to greater ischemia of epicardial cardiomyocytes, away from the oxygenated LV blood pool (16). Despite the relatively regular endocardial activity in LDVF shown in Figure 2C, the phenomenon of wave break at variable myocardial levels is likely to contribute to the disorganized appearance of VF on the surface electrocardiography (17).
Re-entry is uncommon during human LDVF
Given the endocardial to epicardial activation rate gradients in LDVF, it is possible that intramural re-entry within the surviving endocardium and subendocardium is responsible for maintaining fibrillation. The criteria for re-entry in this study, however, were only met in 3 instances in LDVF, all on the endocardium or subendocardium (Table 1). These findings are consistent with those of other studies that have found less re-entry and increasingly frequent focal endocardial activations as VF progresses over time (5,10,11,18). We did not look for re-entry in a diagonal line, however, and needle spacing may have failed to identify small re-entry circuits in this study.
Purkinje activity determines endocardial activity during human LDVF
During human LDVF, Purkinje potentials could be seen to arise focally (occasionally from more than one Purkinje site) and then activate the endocardium rapidly via the Purkinje system (Figure 3A). At times endocardial foci away from the Purkinje network were also seen (Figure 3A, beats 5 and 6), which also appear to contribute to rapid LV endocardial activation. Purkinje fibers have been shown to be more resistant to ischemia than ventricular myocardium and to receive oxygen by diffusion from the blood pool (19,20). In combination with this phenomenon, Purkinje fiber activation has been shown to precede myocardial activation after LDVF defibrillation failures in canines, and chemical ablation of the Purkinje system has been shown to lead to earlier spontaneous termination of LDVF (4). We did not perform Purkinje fiber ablation with Lugol solution in this study because the endocardial necrosis is not specific to Purkinje fibers, and we found that nonperfused human hearts were only capable of sustaining one complete LDVF protocol.
Upon unipolar mapping, sharp Purkinje potentials become more prominent on the endocardium as VF progresses, whereas on the epicardium, the local CL slows and lower frequency signals are seen (Figure 4). Newton et al. (14) showed that in both canines and pigs, the regions with dominant frequency in LDVF are those where the Purkinje fibers distribute (epicardially in pigs and endocardially in canines). Purkinje potentials may be more difficult to see in SDVF (Figure 4, VF onset) because they are overdrive suppressed by more rapid re-entrant wave fronts or, conversely, persistent bombardment by these rapid wave fronts may lead to abnormal automaticity in Purkinje fibers as seen in canine Purkinje fibers in this study and previously in a sheep model (21).
Focal activity from Purkinje fibers in LDVF may be induced by rapid activations during early VF
By simulating rapid activation during early VF with rapid pacing, focal Purkinje fiber activity occurred that increased in frequency with increases in both pacing duration and pacing rate (Figures 5C and 5D). In addition, we were able to show a greater resistance to ischemia and a greater capacity for continued 1:1 capture of Purkinje fibers compared with ventricular myocardium (Figures 5A and 5B). It has been shown previously in animal myocardial infarct models that after infarction Purkinje fibers display spontaneous automaticity, enhanced responses to adrenergic interventions, and a tendency to triggered activity (22–24). These mechanisms may also underlie the spontaneous Purkinje fiber activity seen in this canine model and in human LDVF in this study. Repeating this LDVF protocol with continued perfusion in future studies may help clarify the precise mechanism of Purkinje automaticity.
The observation in this study of a changing mechanism sustaining human VF over time suggests that conventional cardiac arrest drugs such as amiodarone (8) or lidocaine (9) may not be the optimal choice in VF of longer durations. Instead, medications that decrease triggered Purkinje fiber activity (by stabilizing ryanodine receptor calcium release , for example) may be more effective adjuvants when VF is resistant to defibrillation or constantly reinitiates. In the current era, VF is often treated early by implantable cardioverter-defibrillators; however, in patients whose first presentation is out-of-hospital cardiac arrest or when implantable cardioverter-defibrillators are not readily available for financial reasons, patients may experience 10 min or more of VF before defibrillation. Provided cardiopulmonary resuscitation is performed, these patients can survive with good clinical outcomes (1) and may benefit from newer adjunctive therapies for LDVF.
The use of explanted human hearts has inherent limitations such as a lack of autonomic innervation of the myocardium. However, there are no ethical means of studying the mechanisms of nonperfused human LDVF in in vivo hearts. The human hearts studied are myopathic, because normal hearts from deceased donors are used for transplantation at our institution. Myopathic hearts, however, are the most relevant substrate to study given that VF is far more likely to occur in this setting than in structurally normal hearts.
Purkinje fiber identification in the intact, cardiomyopathic human heart is challenging (particularly with left bundle branch block). So the human hearts could remain intact, a canine model was used because the Purkinje fiber distribution in dogs is most similar to that of humans compared with other mammals, and the Purkinje fibers can be identified readily on the canine endocardium. In this model, simulated VF (with rapid pacing) was used, which may also lead to different Purkinje fiber effects than true VF in the human heart.
Human LDVF is characterized by an endocardial to epicardial activation frequency gradient created by focal endocardial activations with mid-myocardial wave break. Re-entry is an uncommon mechanism in human LDVF; instead, focal endocardial activations originate most commonly from Purkinje fibers. Rapid activations during early VF may mediate focal activity in LDVF and facilitate its maintenance.
COMPETENCY IN MEDICAL KNOWLEDGE: VF is the rhythm most frequently responsible for sudden cardiac death. In humans, VF changes over time from a rhythm characterized by re-entry and transmural scroll waves to one dominated by focal endocardial Purkinje fiber activations with mid-myocardial wave break. It is not clear whether the optimal adjunctive strategies for treating VF should also change as the rhythm progresses in time.
TRANSLATIONAL OUTLOOK: Additional research is needed to further improve outcomes for patients who experience cardiac arrest as a result of VF. This research may include the investigation of medications to reduce Purkinje fiber–triggered activity (such as ryanodine receptor stabilizing medications) to improve the efficacy of defibrillation and prevent refibrillation in LDVF.
This work was supported by the Canadian Institute of Health Research (grant number MOP 77687). The authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- cycle length
- local activation time
- long-duration ventricular fibrillation
- left ventricular
- short-duration ventricular fibrillation
- Received February 4, 2015.
- Revision received April 3, 2015.
- Accepted April 9, 2015.
- American College of Cardiology Foundation
- Valenzuela T.D.,
- Roe D.J.,
- Cretin S.,
- et al.
- Dosdall D.J.,
- Tabereaux P.B.,
- Kim J.J.,
- et al.
- Nair K.,
- Umapathy K.,
- Farid T.,
- et al.
- Masse S.,
- Downar E.,
- Chauhan V.,
- et al.
- Li L.,
- Jin Q.,
- Dosdall D.J.,
- et al.
- Tabereaux P.B.,
- Walcott G.P.,
- Rogers J.M.,
- et al.
- Newton J.C.,
- Smith W.M.,
- Ideker R.E.
- Zipes D.P.,
- Wellens H.J.
- Li L.,
- Zheng X.,
- Dosdall D.J.,
- et al.
- Huang J.,
- Rogers J.M.,
- Killingsworth C.R.,
- et al.
- Friedman P.L.,
- Stewart J.R.,
- Fenoglio J.J. Jr..,
- et al.
- Wald R.W.,
- Waxman M.B.
- El-Sherif N.,
- Gough W.B.,
- Zeiler R.H.,
- et al.
- Kimura S.,
- Bassett A.L.,
- Kohya T.,
- et al.
- Shannon T.R.,
- Lew W.Y.