Computational modeling identifies embolic stroke of undetermined source patients with potential arrhythmic substrate

  1. Savannah F Bifulco
  2. Griffin D Scott
  3. Sakher Sarairah
  4. Zeinab Birjandian
  5. Caroline H Roney
  6. Steven A Niederer
  7. Christian Mahnkopf
  8. Peter Kuhnlein
  9. Marcel Mitlacher
  10. David Tirschwell
  11. WT Longstreth
  12. Nazem Akoum
  13. Patrick M Boyle

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    In this manuscript, the authors try to answer an important clinical question about the previously observed connection between embolic stroke of undetermined source (ESUS) and atrial fibrillation (AF). Using cutting-edge personalized computational modeling of left atrium from both ESUS and AFib patients, the researchers try to understand why the fibrotic substrate found in both ESUS and AFib patients causes arrhythmia in the latter group but not the former. Their study concludes that the intrinsic capacity to sustain arrhythmias of fibrosis found in ESUS and AFib atrium are identical. The key claims of the manuscript are well supported by the data, and the modeling methodology is largely appropriate.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #2 and Reviewer #3 agreed to share their names with the authors.)

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Cardiac magnetic resonance imaging (MRI) has revealed fibrosis in embolic stroke of undetermined source (ESUS) patients comparable to levels seen in atrial fibrillation (AFib). We used computational modeling to understand the absence of arrhythmia in ESUS despite the presence of putatively pro-arrhythmic fibrosis. MRI-based atrial models were reconstructed for 45 ESUS and 45 AFib patients. The fibrotic substrate’s arrhythmogenic capacity in each patient was assessed computationally. Reentrant drivers were induced in 24/45 (53%) ESUS and 22/45 (49%) AFib models. Inducible models had more fibrosis (16.7 ± 5.45%) than non-inducible models (11.07 ± 3.61%; p<0.0001); however, inducible subsets of ESUS and AFib models had similar fibrosis levels (p=0.90), meaning that the intrinsic pro-arrhythmic substrate properties of fibrosis in ESUS and AFib are indistinguishable. This suggests that some ESUS patients have latent pre-clinical fibrotic substrate that could be a future source of arrhythmogenicity. Thus, our work prompts the hypothesis that ESUS patients with fibrotic atria are spared from AFib due to an absence of arrhythmia triggers.

Article activity feed

  1. Version published to 10.7554/elife.64213 on eLife
  2. eLife

    Author Response to Public Reviews

    Reviewer #1 (Public Review):

    [...] What is left unclear is what is unique about the fibrotic substrate in ESUS patients in comparison to AFib patients in the future.

    We thank the reviewer for these reasonable and accurate critiques. In the revised version of our manuscript, we offer a more in-depth analysis of potential cohort-scale differences in the spatial distribution of fibrosis between ESUS and AFib patients and how that might affect the overall arrhythmogenicity of fibrotic remodeling between the two populations. We further acknowledge comprehensive understanding of pathophysiological consequences of fibrosis in ESUS will require much more research in the future. Our plans include analysis of how fibrosis might affect LA hemodynamic properties and the likelihood of clot formation. Future work (both clinical and computational) will also be needed to test the hypothesis generated by the present study that ESUS patients lack the triggers needed to initiate AFib. We have added clarifying text to the Discussion section of our manuscript to acknowledge these two points (see lines 286-289, 367-368).

    Reviewer #2 (Public Review):

    [...] 1) As the authors point out, clinical studies have revealed that the fibrotic burden in ESUS patients is similar to those with aFib. The question is why then, do so few ESUS patients exhibit clinically detectable arrhythmias with long-term monitoring. The authors hypothesize and their data support the notion that while the substrate is prime for pro-arrhythmia in ESUS patients, a lack of triggering events may explain the differences between the two groups.

    We thank the reviewer for their kind comment about the level of anatomical and structural variability in our study. We concur that additional analysis of fibrosis spatial pattern properties (local fibrosis density and entropy, as calculated in our previous work) on a region-wise basis between AFib/ESUS and inducible/non-inducible models would add significant value to our work. Accordingly, we have made significant additions to the text including a completely new figure.

    1. I think the authors could go further in describing why this is surprising. Generally, severe fibrosis is thought to potentially serve as a means or mechanism for pro-arrhythmic triggers. This is because damage to cardiac tissue typically results in calcium dysregulation. When calcium overload occurs in isolated fibrotic tissue areas, or depolarization of the resting membrane potential due to localized ischemia allows for ectopic peacemaking, we might expect that the diseased/fibrotic tissue is itself the source of arrhythmia generation. I think the novel finding here is that this notion may be a simplification, and the sources of arrhythmia generation may be more complex and may need to come from outside the areas of fibrosis. I think this is a big deal.

    Patients with stroke were excluded from the AFib cohort because the etiology of stroke in our AFib cohort was not explicitly adjudicated to be cardioembolic, other ischemic such as atherosclerotic, or haemorrhagic and therefore would not allow us to draw reliable conclusions regarding the role of the atrial substrate in stroke in this population. A separate issue is the fact that the cell- and tissue-scale electrophysiology in models reconstructed from ESUS patients was based on the same representation as those used in AFib models. In fact, this was a deliberate design choice to ensure that our modeling results represented a “worst case scenario” for the potential impact of fibrosis in patients with ESUS. Given the fact that our aim is to determine whether there are any differences in the pro-arrhythmic capacity of fibrotic substrate in ESUS and AFib groups, we believe that this is a suitable and justifiable modeling choice – modeling fibrosis differently in the two populations would be difficult to justify due to a lack of good experimental data and would introduce more confounding factors.

    Nevertheless, we agree this is a relevant limitation of our study and we have added an acknowledgement of that fact to our revised manuscript (see lines 361-365).

    1. An acknowledged limitation of the study is the assumption of fixed conduction velocity and action potential duration/effective refractory period. Bifulco et al. base this assumption on previous studies by the group (e.g. L312), which, however, concluded that reentrant driver locations and inducibility are sensitive to changes of action potential and conduction velocity (Deng et al.). For conduction velocity, wider ranges have been reported since the publication of the supporting reference (35) in 1994, e.g. Verma et al.; Roney et al.

    The reviewer’s point is well taken. Accordingly, we have added qualifying language pertaining to RD localization analysis in our Discussion (see lines 323-326). Having said that, we do not think this issue stands to fundamentally change our top-line interpretation of the findings from simulations, as it pertains to the idea that fibrosis in ESUS might plausibly be latent proarrhythmic substrate. The point of the paper by Deng et al. was to analyze sensitivity of reentrant driver localization to altered cell- and tissue-scale electrophysiological properties, not the concept of inducibility per se. It is thus likely that if our entire study were repeated with ±10% CV or APD (both within normal physiological range for average fibrotic atrial tissue) the take-home message would be the same.

    1. The number of pacing sites is rather low for a comprehensive in silico arrhythmia inducibility test but likely a good balance of coverage and computational feasibility considering that the primary goal of this research was to check whether the two groups of models show differences when undergoing the same (but not necessarily exhaustive) protocol.

    We would argue that 15 sites in the LA alone is comparable in coverage to prior studies in biatrial models (e.g., 30 LA/RA sites in Zahid et al. [2016] Cardiovasc Res; 40 LA/RA sites in Boyle et al. [2019] Nat Biomed Eng). We would further stress that our decision to use these specific sites was based on our motivation to simulate triggered activity (i.e., rapid pacing) exclusively from sites identified as common clinically relevant trigger locations documented in AFib patients (see ref. [14] by Santangeli et al. [2016] Heart Rhythm). If we were to instead pace from randomly distributed atrial sites as in prior work, we would jeopardize our ability to draw conclusions on the potential relevance of our simulations to the real-world susceptibility of atrial fibrotic substrate in ESUS patients to ectopic beats originating from realistic locations.

    1. The discussion does a good job in putting the results into context. Two interesting observations that deserve more attention are that i) the Inducibility Score was always higher for AFib vs. ESUS (Figure 6A, no statistical test performed). However, this did not translate to a difference in silico arrhythmia burden (inducibility). ii) Reentrant drivers were about twice as likely to localize to the left pulmonary veins than the right pulmonary veins in the AFib models (Figure 6D).

    Regarding the first point (i), with corrections made to the fiber mapping process, the statement regarding uniformly higher IdS values in AFib models is no longer true. Moreover, with our revised analysis there is no significant difference in the region-wise inducibility rates (P=0.45). The reviewer’s second point (ii) still stands and is even more pronounced with a ~3x higher rate of localization to the LPV vs. RPV areas in AFib models. Notably, our new region-wise analysis of fibrosis spatial pattern (see new Fig. 6 and our response to major points 4 and 5 above) shows that LPV regions in AFib models in this cohort were much more likely to have the combination of high fibrosis density and entropy previously shown to be highly favorable to reentrant driver localization. However, we recognize that a more fulsome analysis will be required to draw truly meaningful conclusions on this subject in the context of either AFib or ESUS patients; this has been briefly noted in our Limitations section (see lines 332-335).

    1. The study succeeded in answering the question it posed in the sense that no marked difference was found between the ESUS and AFib models. This leads to the question what the stroke-inducing mechanism is in the ESUS patients. A hypothesis for future work could be that the fibrotic infiltrations in the ESUS patients reduce the hemodynamic efficacy of the left atrium and render clot formation (e.g. in the atrial appendage) more likely in this way.

    The reviewer’s comment is duly noted and entirely consistent with our plans for future work. In fact, we recently published a white paper (Boyle et al. [2021] Heart) outlining a vision to combine electrophysiological models of the left atrium with biomechanics and hemodynamics simulation to comprehensively understand how fibrosis might influence clot formation. Our revised Discussion emphasizes this exciting trajectory for future work (see lines 370-372).

    1. The negative finding in this study (no difference between groups) does not naturally allow us to draw clinical implications for diagnosis or stratification. Additional ways to put the hypothesis proposed by the authors (fewer arrhythmogenic triggers in the ESUS patients) to test could be to consider readouts/surrogate measures of the autonomic nervous system.

    We have noted in our Discussion (see lines 286-289) that future work could test the hypothesis arising from this project via electrocardiographic monitoring in ESUS patients with different levels of fibrosis. Concerning the idea of using direct readouts of autonomic tone, we chose to leave this out since we are unaware of any clinically available systems. The usefulness of surrogate measurements (e.g., heart rate variability) in this context also remains unclear.

    Reviewer #3 (Public Review):

    [...] 1) As the authors point out, clinical studies have revealed that the fibrotic burden in ESUS patients is similar to those with aFib. The question is why then, do so few ESUS patients exhibit clinically detectable arrhythmias with long-term monitoring. The authors hypothesize and their data support the notion that while the substrate is prime for pro-arrhythmia in ESUS patients, a lack of triggering events may explain the differences between the two groups.

    We thank the reviewer for these kind remarks. It is encouraging to have our results interpreted so elegantly and accurately. We are excited to test this new hypothesis (and others prompted by the peer review process for this manuscript) in future studies.

    1. I think the authors could go further in describing why this is surprising. Generally, severe fibrosis is thought to potentially serve as a means or mechanism for pro-arrhythmic triggers. This is because damage to cardiac tissue typically results in calcium dysregulation. When calcium overload occurs in isolated fibrotic tissue areas, or depolarization of the resting membrane potential due to localized ischemia allows for ectopic peacemaking, we might expect that the diseased/fibrotic tissue is itself the source of arrhythmia generation. I think the novel finding here is that this notion may be a simplification, and the sources of arrhythmia generation may be more complex and may need to come from outside the areas of fibrosis. I think this is a big deal.

    This is an excellent point and we strongly concur that the “trigger-centric” interpretation of the pathophysiological consequences of fibrotic remodeling should be reconsidered. To further reinforce this fact, we ran additional simulations to rule out the possibility that there might be exaggerated resting membrane potential depolarization in AFib but not in ESUS, which might provide an alternative explanation for the clinical manifestation of arrhythmia in the former but not the latter. Our new results support the point raised by the reviewer and, in our opinion, increase the overall impact of the work.

  3. Version published to 10.1101/2020.09.03.20184051v3 on medRxiv
  4. eLife

    Reviewer #3 (Public Review):

    This is a well written and elegant study from a collaboration of groups carrying out models based on high resolution imaging. I think the study also serves as a prime example for where modeling and simulation bring added value in the sense that the insights revealed in the study would not likely be gained through other methods.

    1. As the authors point out, clinical studies have revealed that the fibrotic burden in ESUS patients is similar to those with aFib. The question is why then, do so few ESUS patients exhibit clinically detectable arrhythmias with long-term monitoring. The authors hypothesize and their data support the notion that while the substrate is prime for pro-arrhythmia in ESUS patients, a lack of triggering events may explain the differences between the two groups.

    2. I think the authors could go further in describing why this is surprising. Generally, severe fibrosis is thought to potentially serve as a means or mechanism for pro-arrhythmic triggers. This is because damage to cardiac tissue typically results in calcium dysregulation. When calcium overload occurs in isolated fibrotic tissue areas, or depolarization of the resting membrane potential due to localized ischemia allows for ectopic peacemaking, we might expect that the diseased/fibrotic tissue is itself the source of arrhythmia generation. I think the novel finding here is that this notion may be a simplification, and the sources of arrhythmia generation may be more complex and may need to come from outside the areas of fibrosis. I think this is a big deal.

  5. eLife

    Reviewer #2 (Public Review):

    Bifulco et al. performed a large-scale in silico study to test whether the spatial fibrosis distribution measured via LGE-MRI in 45 patient with embolic stroke of undetermined source (ESUS) as compared to the distribution in 45 atrial fibrillation (AFib) patients without stroke leads to differences in reentrant arrhythmia inducibility of dynamics.

    1. This study comprises a high number of simulations and is one of the computational electrophysiology studies that covers the most anatomical and structural variability on the atrial level. In their comprehensive analysis, Bifulco et al. answered their question and found no pronounced differences in arrhythmia inducibility and dynamics between ESUS and AFib models. It would be interesting to learn how the spatial fibrosis distributions compare in terms of the previously suggested features density and entropy (Zahid et al.). This might also influence the statements in L170/L207.

    2. The authors chose to exclude patients with stroke from the AFib group, the reasons for this choice are not entirely clear. The same holds for the fact that the ESUS models included AFib-induced electrophysiological remodeling even though these patients have not been diagnosed with AFib (by definition).

    3. An acknowledged limitation of the study is the assumption of fixed conduction velocity and action potential duration/effective refractory period. Bifulco et al. base this assumption on previous studies by the group (e.g. L312), which, however, concluded that reentrant driver locations and inducibility are sensitive to changes of action potential and conduction velocity (Deng et al.). For conduction velocity, wider ranges have been reported since the publication of the supporting reference (35) in 1994, e.g. Verma et al.; Roney et al.

    4. The number of pacing sites is rather low for a comprehensive in silico arrhythmia inducibility test but likely a good balance of coverage and computational feasibility considering that the primary goal of this research was to check whether the two groups of models show differences when undergoing the same (but not necessarily exhaustive) protocol.

    5. The discussion does a good job in putting the results into context. Two interesting observations that deserve more attention are that i) the Inducibility Score was always higher for AFib vs. ESUS (Figure 6A, no statistical test performed). However, this did not translate to a difference in silico arrhythmia burden (inducibility). ii) Reentrant drivers were about twice as likely to localize to the left pulmonary veins than the right pulmonary veins in the AFib models (Figure 6D).

    6. The study succeeded in answering the question it posed in the sense that no marked difference was found between the ESUS and AFib models. This leads to the question what the stroke-inducing mechanism is in the ESUS patients. A hypothesis for future work could be that the fibrotic infiltrations in the ESUS patients reduce the hemodynamic efficacy of the left atrium and render clot formation (e.g. in the atrial appendage) more likely in this way.

    7. The negative finding in this study (no difference between groups) does not naturally allow us to draw clinical implications for diagnosis or stratification. Additional ways to put the hypothesis proposed by the authors (fewer arrhythmogenic triggers in the ESUS patients) to test could be to consider readouts/surrogate measures of the autonomic nervous system.

  6. eLife

    Reviewer #1 (Public Review):

    Previous research showed a close link between sub-clinical AFib (Atrial Fibrillation) and ESUS (Embolic Stroke of Undetermined Source). As such, current established clinical care for ESUS patients is long-term monitoring for evidence of AFib and anticoagulant treatment for an individual with high risk for AFib. Nevertheless, questions are still unanswered about who the individuals with high-risk for ESUS are and how to properly identify this population.

    This research tries to identify the fibrotic properties of ESUS patients and its pro-arrhythmic potential using computational modeling of patient's left atria reconstructed from cardiac LGE-MRI (Late-Gadolinium Enhanced Magnetic Resonance Imaging). Ultimately, their results of the comparison between left atria of ESUS and AFib patients revealed that the fibrotic substrate that could induce arrhythmia in ESUS and AFib patients are indistinguishable, raising more questions that would need to be addressed in further studies.

    This study uses a sophisticated personalized computational modeling approach that has been validated in previously published papers. This study is also well designed, clearly written, with robust data and proper statistical analysis.

    What is left unclear is what is unique about the fibrotic substrate in ESUS patients in comparison to AFib patients in the future.

  7. eLife

    Evaluation Summary:

    In this manuscript, the authors try to answer an important clinical question about the previously observed connection between embolic stroke of undetermined source (ESUS) and atrial fibrillation (AF). Using cutting-edge personalized computational modeling of left atrium from both ESUS and AFib patients, the researchers try to understand why the fibrotic substrate found in both ESUS and AFib patients causes arrhythmia in the latter group but not the former. Their study concludes that the intrinsic capacity to sustain arrhythmias of fibrosis found in ESUS and AFib atrium are identical. The key claims of the manuscript are well supported by the data, and the modeling methodology is largely appropriate.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #2 and Reviewer #3 agreed to share their names with the authors.)

  8. Version published to 10.1101/2020.09.03.20184051v2 on medRxiv
  9. Version published to 10.1101/2020.09.03.20184051v1 on medRxiv