Fibrosis and Atrial Fibrillation: Computerized and Optical Mapping; A View into the Human Atria at Submillimeter Resolution

Abstract

Recent studies strongly suggest that the majority of atrial fibrillation (AF) patients with diagnosed or subclinical cardiac diseases have established or even pre-existing fibrotic structural remodeling, which may lead to conduction abnormalities and reentrant activity that sustain AF. As conventional treatments fail to treat AF in far too many cases, an urgent need exists to identify specific structural arrhythmogenic fibrosis patterns, which may maintain AF, in order to identify effective ablation targets for AF treatment. However, the existing challenge is to define what exact structural remodeling within the complex 3D human atrial wall is arrhythmogenic, as well as linking arrhythmogenic fibrosis to an underlying mechanism of AF maintenance in the clinical setting. This review is focused on the role of 3D fibrosis architecture in the mechanisms of AF maintenance revealed by submillimeter, high-resolution ex-vivo imaging modalities directly of human atria, as well as from in-silico 3D computational techniques that can be able to overcome in-vivo clinical limitations. The systematic integration of functional and structural imaging ex-vivo may inform the necessary integration of electrode and structural mapping in-vivo. A holistic view of AF driver mechanisms may begin to identify the defining characteristics or “fingerprints” of reentrant AF drivers, such as 3D fibrotic architecture, in order to design optimal patient-specific ablation strategies.

Introduction

Atrial fibrillation (AF) is the most prevalent cardiac arrhythmia and a significant cause of hospitalization and morbidity. Affecting more than 2.7 million Americans today, AF is projected to affect up to 16 million by 2050(1,2). Despite significant advances in this field, treatment options for AF remain limited as data regarding the mechanisms that drive and maintain AF are elusive in the human heart. The localized driver hypothesis of AF has gained momentum recently,(3-5) which suggests that paroxysmal, persistent, and permanent AF can be driven from localized sources outside the pulmonary veins and that limited ablation targeting these regions can have a successful outcome. However, targeting ablation to the functional activation patterns of AF drivers has produced mixed results, partially due to the technical limitations of clinical mapping.(6,7) Recent clinical and experimental studies have suggested that AF drivers may be perpetuated by pathologic atrial structural and electrical remodeling associated with the arrhythmia, which can be heterogeneous and patient-specific. Vast information from clinical(8,9) and experimental(10-12) studies directly suggest that fibrosis plays a key role in the progression of AF and development of AF drivers. Disease-remodeled atrial tissue may have underlying unique characteristics that make some areas prone to harboring localized AF drivers. These unique characteristics or AF driver “fingerprints” could be a combination of not only fibrosis, but other structural features, such as wall thickness variation and myofiber orientation/anisotropy,(12) as well as electrical remodeling(2), which all may play a synergistic role in reentrant AF driver formation.

The introduction of global or wide-field of view clinical mapping techniques has aided in the search for AF drivers(3-5); however, the wide-field of view has necessarily been limited in clinical use by low electrode spatial resolution as well as surface only recordings, which may miss or misinterpret some mechanisms in highly complex 3D structure of the human atria.(7,13) Accurately identifying drivers may be aided by defining heart-specific structural fingerprints to reliably locate AF drivers.(7) This review will focus on the role of 3D fibrosis architecture of human atria as an AF driver fingerprint that facilitates mechanisms of AF maintenance. This review will also delineate advances and hurdles in in-vivo and ex-vivo imaging techniques and the advantages and limitations of in-silico 3D computational techniques. We also discuss the need for consolidated structural and functional data derived from submillimeter high-resolution ex-vivo imaging of human atria that can be potentially incorporated into 3D computational models to accurately predict not only the role of fibrosis and other AF driver fingerprints but also to effectively guide targeted ablation of AF drivers.

Building the Link between AF and Fibrosis

Recent clinical studies suggest that many AF patients(1) with diagnosed or subclinical cardiac diseases (e.g. hypertension(14) or heart failure(15)) have fibrotic structural remodeling. Atrial structural remodeling in AF may have various characteristics that enhance the natural complexity of the atrial wall, such as enlargement of the atrial chamber, hypertrophy of cardiomyocytes, increased discordance between epi- and endocardial myofiber orientations and tissue anisotropy, enhanced atrial wall thickness variations, and importantly, an increase in fibrotic or connective tissue content.(1,2,16) Increased fibrosis, considered a key contributor to the functional effects of atrial remodeling(17,18), is defined as an accumulation of extracellular matrix comprised of a mesh of collagen and elastin produced by fibroblasts.

Atrial fibrosis is directly correlated with age of patients, which is mirrored by the increasing incidence of AF in the elderly(1,2). Structural cardiac diseases, such as hypertension(14) and heart failure(15), have been shown to have both an increase in atrial fibrosis and an increase in the occurrence of AF compared to healthier hearts. Recently, Kottkamp(9) has coined the phrase fibrotic atrial cardiomyopathy to denote the structural remodeling that accompanies or even precedes AF which is independent of other co-morbidities. Platonov et al.(19) showed by histological analysis of autopsy specimens that patients with paroxysmal or permanent AF had 2-3 times greater fibrosis than patients without AF history (Figure 1A). The DECAAF Trial,(20) a multi-center prospective trial that employed delayed-enhancement magnetic resonance imaging (DE-MRI), showed that the extent of atrial fibrosis was the most significant predictor of ablation failure. Clinical DE-MRI and postmortem histological studies suggest that the total amount of atrial fibrosis may predispose a patient to AF, but the heterogeneous 3D fibrosis architecture and the specific role that fibrosis plays in the mechanisms of AF maintenance are yet to be clarified.

Figure 1. Arrhythmogenic Fibrosis in the Progression and Maintenance of AF.

A. Masson's Trichrome staining shows that the amount of fibrosis (blue) increases with AF burden in patients. Modified from Platonov et al.(19) with permission. B. Fibrosis (red) classifications based on fibrotic tissue pattern, size, and distribution. From de Jong et al.(17) with permission. C. Anisotropic micro-reentry observed in a human pectinate muscle ex-vivo due to fibrotic tissue insulating myofibers. Modified from Spach et al.(24) with permission. AF-atrial fibrillation.

What makes Fibrosis Arrhythmogenic: Insights from Histological and Ex-Vivo Mapping Studies

Broadly, fibrosis can be sorted into four houses based on its extent, pattern, and distribution, namely compact, diffuse, patchy, and interstitial fibrosis (Figure 1B).(17) Compact fibrosis, usually due to previous infarct or ablation, is devoid of myocytes and electrically non-conductive. While it may act as a central obstacle for macro-reentry, it is not considered a significant arrhythmogenic pattern for human AF. However patchy, interstitial, and diffuse fibrosis separate and insulate myocardial bundles, which can create a structural substrate for disordered activation by slowing conduction velocity, increasing anisotropy of conduction, creating unidirectional blocks, and allowing multiple reentries and wavelets.(12,21,22) Structural remodeling leading to or due to AF can be associated with all three types of arrhythmogenic fibrosis, which could be present and spatially distributed in a highly patient specific manner. Moreover, the categories of fibrosis appear to exist on a spectrum, as it may not always be possible to clearly distinguish one type from another, so further studies are needed to assess the arrhythmogenicity of each type. However, it is clear that any form of fibrosis should affect conduction to be considered arrhythmogenic. Furthermore, while clinical DE-MRI can reveal the link between AF occurrence and the total amount of atrial fibrosis, the contribution of specific fibrotic architecture to the mechanisms that sustain AF could remain hidden within the complex 3D atrial wall mainly due to low resolution in-vivo techniques.

Most of the data on the arrhythmogenicity of specific atrial fibrotic remodeling patterns have come from animal model studies which offer the advantage of combining high resolution electrode or ex-vivo optical mapping with histological analysis of the atrial wall, which would be impossible in in-vivo human studies. Ex-vivo endocardial optical mapping, histology, and computer simulations of the isolated chronic heart failure sheep atria have suggested that patchy fibrosis in the posterior left atria could be a substrate for rotor and even lead to intramural localized reentry.(10) Verheule et al's(11) epicardial optical mapping study along with detailed histological analysis in a goat chronic AF model revealed that increased endomysial fibrosis, which could be considered interstitial fibrosis, could increase the complexity of AF. Furthermore, a small regional high density (1mm² resolution) epicardial electrode array placed in-situ along with histological analysis demonstrated that obstructive fibrosis, meaning ≥200μm thick strands encompassing both interstitial and patchy fibrosis, could initiate and maintain reentry in a goat model with 6 month tachypacing induced AF.(23) Ex-vivo animal studies also show that patchy and interstitial fibrotic strands could enhance both the natural conduction anisotropy, due to myofiber orientation and myobundle discontinuities,(10,23,24) as well as the natural transmural conduction dissociation,(11) thus increasing the ability of a discrete atrial site to support reentry or of the entire atria to support multi wavelets as possible mechanisms of AF maintenance.

However, studies incorporating integrated functional and structural mapping are mostly from animal models and not from human hearts,(2,25) which possess significantly different 3D atrial anatomy and function.(2) Moreover, fibrosis accumulation and 3D architecture in tachypaced animal models is experimentally induced by fast atrial rhythm, whereas fibrosis associated with AF in aged human atria have usually developed overtime secondary to preexisting co- morbidities even before AF manifestation.(1,2,8,19,26) The proportions of human atrial anatomical regions, such as the smaller appendage to free wall ratios in both left and right atria, also differs from that of commonly used sheep models.(27-29) Only a few ex-vivo human studies have been able to directly link atrial structural remodeling with conduction impairment and AF reentrant maintenance mechanisms.(24,30) Spach et al.(24) were the first to demonstrate both 1) that micro-reentry can be sustained within a human pectinate muscle if the anisotropic ratio is high enough and 2) that this micro-reentry was related to interstitial fibrosis and electrical remodeling in diseased and aged human atria (Figure 1C). However, since all previous ex-vivo studies that attempted to link atrial function with structure utilized mainly 2D histological sections, conclusions were unable to describe the 3D fibrotic architecture across the human atria.

Integration of High-Resolution Submillimeter Contrast-Enhanced MRI and Transmural Optical Mapping to Identify AF Driver Fingerprints in the Human Atria Ex-vivo

In order to overcome the limitations of clinical and animal models and to elucidate the specific role of 3D atrial structural remodeling and fibrosis patterns in AF maintenance, our group has recently developed an integrated ex-vivo approach to study the 3D structure of AF drivers in diseased human atria at submillimeter resolutions.(12,31) Explanted human right atria or intact whole atria were studied by high-resolution (330μm²)(12) simultaneous epi-endocardial and panoramic optical mapping to determine the locations and spatiotemporal characteristics of AF drivers. Subsequently, the same human atria were scanned by high-resolution (up to 80μm³) ex-vivo contrast-enhanced MRI (CE-MRI) with 9.4 Tesla scanner to study region-specific structural fingerprints of reentrant AF drivers(12).

Submillimeter resolution optical mapping that uses voltage-sensitive dyes may resolve conduction and repolarization at the cellular level and can accurately collect optical action potentials from a depth of the tissue (∼0.5-1mm deep using di-4-ANNEPS and ∼1-4mm using near-infrared dye di-4-ANDBQBS)(12,32). Thus, near-infrared optical mapping(12) allows for more detailed spatial resolution of fibrillatory conduction compared to unipolar and bipolar contact electrode mapping that records extracellular surface activation with far-field influence from surrounding tissue (Figure 2)(33). Additionally, ex-vivo human optical mapping experiments are performed at controllable physiologic conditions that ensure conduction and repolarization values(30) resemble values clinically recorded by monophasic action potentials.(34,35) Autonomic(28,31) or metabolic pharmacologic stimulation(12) also help recapitulate the autonomic nervous system and metabolic stresses that are known to play a critical role in AF.(2)

Figure 2. Complexities of the 3D Atrial Wall create Challenges for AF Substrate Detection in Patients.

A. Optical activation maps and optical action potentials (OAPs) show distinct activation patterns recorded simultaneously from the Epi (epicardium) and Endo (endocardium) in the human right atria ex-vivo. B. 2D contrast-enhanced MRI (CE-MRI) of the right atrial wall underlying catheter in panel A with corresponding bipolar electrogram (EG) signal showing the influence of activation of different myobundles on electrogram fractionation. Panels A and B modified from Hansen et al.(12) with permission. C. Bipolar electrogram recorded by two white circles in a canine infarct model with simultaneous microelectrode recordings shows electrogram fractionation with three deflections due to three distinct myocardial layers (grey) insulated by fibrosis (speckled black). Modified from Gardner et al.(22) with permission. S₁-pacing stimulus artifact.

Based on our high resolution integrated ex-vivo studies, we propose that human AF is maintained by a limited number of spatially stable but temporally competing, intramural microanatomic reentries within discreet patient-specific atrial regions with specific structural fingerprints, such as fibrotic insulation, wall thickness variation, and endo-epicardial myofiber misalignment (Figure 3). We use the term “microanatomic” reentry to distinguish the mechanism from macroanatomic reentry that circulates around large anatomic structures, such as the tricuspid valve and venous ostia. The microanatomic tracks shown in Figures 3B and 4A are composed of several adjacent fibrotically-insulated myobundles (tunnels) 0.5-2mm thick that are partially uncoupled along their lateral borders. These tracks are defined by not only the unexcitable, physical discontinuity between these muscle bundles but also by the bundles themselves. The 3D size of microanatomic reentrant tracks observed in our integrated optical mapping and 3D CE-MRI study(12), and more recently with 3D micro-CT(28,31), were on average ∼15×6 mm, with a 3 mm depth, but were defined by microanatomic substrates (i.e. intramural fibrotic strands ∼200μm thick) that may only be visible at submillimeter resolution. Importantly, the microanatomic reentrant AF drivers were confirmed by targeted radiofrequency ablation that eventually terminated and prevented AF while ablation outside the driver tracks had no effect on arrhythmia.(12) In our recent review,(7) we described how the same intramural microanatomic reentrant AF driver could be visualized differently at the atrial epi- vs. endocardial surfaces as either 1) complete reentry 2) partial reentry 3) stable breakthrough/ focal activity or 4) unstable breakthroughs due to the heart-specific complexity of the 3D atrial wall (Figure 3A). Furthermore, at low millimeter resolution or after phase signal processing,(13) 3D transmural microanatomic reentry may be indistinguishable from rotor type activity.(36)

Figure 3. Human Intramural Reentrant AF Drivers localized to Fibrotically Insulated Myobundle Regions.

A. Simultaneous panoramic and dual-sided optical mapping of a reentrant AF driver in the human right atria ex-vivo. B. Left: Panoramic optical activation map of the human right atria shows structurally defined lines of conduction block in the AF driver region. Right: Histologically validated contrast-enhanced MRI (CE-MRI) shows that fibrotically insulated myobundles create a track for reentry. Modified from Hansen et al.(12) with permission. Abbreviations as in Figure 2; AF- atrial fibrillation; CAD-coronary artery disease; HTN-hypertension; Inf-inferior; OAPs-optical action potentials; Sup-superior; TA- tricuspid annulus.

Figure 4. A limited Number of Spatially Stable, Temporally Competing Reentrant AF Drivers Sustain AF in the Human Atria.

A. Optical activation maps and contrast-enhanced MRI (CE-MRI) show that the superior AF driver occurred at a region of slow conduction during pacing defined by fibrotically insulated discontinuous myobundles. B. Three schematic activation patterns and optical action potentials (OAPs) recorded during AF driven by two competing reentrant AF drivers (purple ovals). Modified from Li et al.(31) with permission. Abbreviations as in Figure 3; IVC/SVC-inferior/superior vena cava; LAA/RAA-left/right atrial appendage.

A continuum may exist with a variety of mechanistic explanations of localized AF drivers having varying degrees of experimental support. After originally proposed by the Soviet Union scientists Krinskii (1966)(37), Pertsov et al (1984)(38), and others and in parallel in the US by Winfree(39) and others in the 1960s-1980s, rotors (i.e. reverberators) and 3D scroll waves were applied to the field of AF drivers at the turn of the 21^st century.(40,41) Rotors are functional reentries with an excitable but unexcited core that are based on the phenomena originally seen in excitable chemical systems, such as the Belousov-Zhabotinsky reaction. Rotors and 3D scroll waves have been proposed to be induced and maintained by many of the same substrates discussed in this review, including fibrosis/structural remodeling and electrical remodeling. However, our high-resolution transmural optical mapping and 3D contrast-enhanced MRI of the human heart identified activation conducting unidirectionally through microanatomic reentry tracks(12), which we have termed microanatomic reentries rather than rotors, since they lack a functional rotor core in their structural tracks/tunnels.(7) The microanatomic reentrant AF drivers observed in our ex-vivo studies would be visualized with similar AF activation and organizational patterns(36) as the reentrant drivers seen in current clinical mapping studies.(3-5) As clinical surface mapping is limited to few millimeters resolution and thus fails to distinguish between microanatomic reentries, rotors, or focal drivers of AF, we refer to the results of these clinical studies with the generic term reentrant drivers.

Our integrated functional and structural studies(12,28,31) reveal that reentrant AF driver can be facilitated by structural fingerprints characterized by the combination of increased interstitial and patchy fibrosis forming intramural fibrotic strands (∼200μm thick), higher endo-epicardial myofiber misalignment, and variation in atrial thickness. Importantly, patchy or interstitial fibrosis was found concentrated in the perimysium within driver regions that insulated endocardial myobundles from one another. Insulating myobundles with fibrosis increased the natural discontinuity between myobundles and created discreet tunnels or microanatomic tracks for reentry (Figure 3B). It is currently unknown if arrhythmogenic fibrotic strands alone can create microanatomic reentrant AF drivers in the human atria, or if the natural 3D myofiber orientation and wall thickness variation of the atria are also necessary and are thus identifiable AF driver fingerprints as well (Central Illustration).

Furthermore, structural fingerprints could be supported by functional fingerprints, such as heterogeneously short refractoriness and slow conduction velocity due to disease-related electrical remodeling, which would create conditions favorable to support the wavelengths of reentry needed for a microanatomic reentrant AF driver (Figure 4). Following clinical protocols of AF induction(3), AF in intact ex-vivo human atria could be provoked and maintained by autonomic stimulation with clinically used agonists including adenosine(31) or isoproterenol.(28) While AF complexity and frequency increased during heterogonous action potential shortening by adenosine for example, the same distinct region supported a reentrant driver at both baseline and adenosine.(31) At the same time, the path length of any microanatomic reentry track would be governed by the functional parameters that dictate wavelengths of reentry. Thus, action potential shortening during autonomic stimulation may allow for a “short cut” along a previously refractory myobundle to become active, which shortens the path length and thus increases the frequency. In this way, autonomic stimulations and/or electrical remodeling would affect which specific myobundles would compose the microanatomic reentry track in the driver region.(7,31)

Furthermore, when two or more reentrant AF drivers were identified, each driver may not be active during every cycle (temporally unstable) (Figure 3B), but AF was always maintained by at least one driver, and reentry repeatedly returned to and remained at the same location without drifting (spatially stable) due to the favorable structural and functional fingerprints that could support microanatomic reentry at the location (Figure 3C). The temporal instability of AF drivers, as well as limitations in extracellular electrogram recordings and signal processing, may underlie the spatial instability of AF drivers seen clinically.(3,4) However, these clinical mapping studies revealed that even targeting the location of highest driver probability achieved promising long-term success, which presumably represent spatially stable drivers with localized fibrotic structural substrates.

Are Reentrant AF Driver Fingerprints within the 3D Atrial Wall Hidden from clinical In-vivo Electrode Mapping and DE-MRI Techniques?

Despite advances in imaging techniques, the hunt is still on for clinically measurable characteristics that accurately define AF driver fingerprints and can guide targeted ablation. While extracellular electrode recordings are currently the most widely used method for mapping atrial activation patterns and identifying AF drivers, interpretations and activation analysis of these recordings may be complicated by contributions from multiple layers in the 3D atrial wall and far-field influences.(33) The human atrial wall, although often simplified to a 2D surface when compared to the thick musculature of the ventricles, has an innate complex 3D anatomy. Detailed histological studies of human atria(12,29) revealed the atrial wall varies in thickness from about 1 to 7 mm due to multiple layers of naturally discontinuous myobundles. Moreover, our simultaneous dual-sided optical mapping of the human right atria showed differences in activation patterns between the epicardial and endocardial surfaces during 2Hz pacing (Figure 2A), and integration with structural analysis by CE-MRI showed how atrial wall thickness variation, myofiber orientation misalignment, and patchy fibrosis could lead to activation discordance and fractionation on a bipolar electrogram (Figure 2B).(12)

Complex Fractionated Atrial Electrograms and Intramural Fibrosis

It was believed that complex fractionated atrial electrograms (CFAEs) could represent sources of AF and thus were considered as a useful guide for ablation treatment.(42) Gardner et al's 1985 study(22) of a chronic canine infarct model elegantly showed by direct microelectrode intracellular action potential recordings that intramural fibrosis can create distinct electrically isolated/discontinuous layers of myofibers that each contribute to deflections seen on a fractionated extracellular electrogram recorded with an inter-electrode distance of only 1 mm (Figure 2C). Consequently, fractionated electrograms have proven non-specific and ineffective as ablation targets in AF treatment,(43) yet specific morphologies of electrogram recordings could in theory be eligible as an AF driver fingerprint with appropriate validation.(33)

Low Voltage Electroanatomic Mapping as a Surrogate for Fibrotic Tissue

Another application of extracellular electrode recordings to define AF sources is electroanatomic voltage mapping that evaluates atrial fibrotic tissue in-vivo, which are defined by the surrogate <0.5mV bipolar signal.(44) Box isolation of fibrotic /low voltage areas (BIFA) from the rest of the atria has shown some success in treating paroxysmal and non-paroxysmal AF.(45) However, any intra-cardiac electrode would record an averaged signal spanning an area on the order of millimeters,(46) and would be unable to distinguish if the fibrosis, defined by a surrogate of low voltage, is interstitial and arrhythmogenic or merely a thick connective tissue layer on the endocardial surface with minor effect on conduction through sub-epicardial myocardium (Figure 2B).(33) The unknown cause of fractionated electrograms, the unpredictability of driver pattern, and the temporal instability of AF drivers all add to the challenges to accurately define AF drivers by surface electrogram mapping alone.

Delayed Contrast Enhancement MRI to define Atrial Fibrotic Regions

It is important to consider that a definitive answer on intramural AF driver mapping may not come from electrogram recordings alone, but instead may require integration with data from 3D structural imaging. DE-MRI may be able to place in-vivo functional, electrode-based maps in the context of 3D human atrial anatomy. However, current in-vivo techniques for quantifying atrial fibrosis may be unable to accurately distinguish 3D patterns of arrhythmogenic atrial structural remodeling. Clinical cardiac DE-MRI has only been able to achieve ∼ 1mm³ resolution even with state-of-the-art 3 Tesla systems.(20,47) DE-MRI uses the diffusion properties of the contrast agent, such as gadolinium, to increase the signal intensity of fibrosis, usually defined as any voxel above a set threshold.(8) As discussed above, obstructive patchy/interstitial arrhythmogenic fibrotic strands could be as small as 200μm across, which would inevitably have its high signal averaged in an DE-MRI voxel with surrounding myocytes that may fail to pass the threshold set to define fibrotic tissue. Moreover, the entire thickness of the atrial wall may be seen by only one or two pixels with the common clinical DE-MRI resolution of 1.25×1.25×2.5mm. An ongoing multicenter clinical trial DECAAF II (NCT02529319) would be the first study where ablation would be guided by DE-MRI defined areas of fibrosis.

Validation of Clinical Techniques

Importantly, current clinical techniques, even those with some promising initial results, still require validation. While neither technique can be considered the gold standard, more studies have shown good correlations(48-50) rather than poor correlations(51) between DE-MRI and voltage mapping. Unvalidated low resolution techniques could be one of the main reasons for contradictions in the field between structural and functional substrates or why some clinical studies do not show a correlation between DE-MRI(52) or low voltage(53) fibrosis and FIRM drivers while others show good correlation between DE-MRI and body surface mapping ECGI studies.(54) The best way to overcome these contradictions would be to validate clinical techniques with high submillimeter resolution ex-vivo techniques (Figures 2A&B).(7) For example, in our recent studies(30,47) we used the relatively large fibrotic content of the human sinoatrial node (SAN) (Figure 5A) as a test to translate high-resolution ex-vivo CE-MRI, validated by optical mapping and histology (Figure 5B), to low resolution in-vivo DE-MRI. Fibrosis of the human SAN found ex-vivo was used to inform in-vivo DE-MRI that was able to correctly identify the human SAN fibrotic structure for the first time in healthy volunteers (Figure 5C). The correct identification of the SAN could be used as one metric for validating fibrosis of AF drivers detected by in-vivo DE-MRI (Figure 5C). Additionally, the development of patient-specific 3D computer modeling based on integrated structural and functional mapping could also help to determine possible AF driver fingerprints and predict successful targeted ablation.(54)

Figure 5. Translation of High-Resolution Ex-Vivo Data for Clinical Application.

A. Fibrotic content in the human sinoatrial node (SAN). From Shiraishi et al.(66); used with permission. B. Validation of high-resolution ex-vivo CE-MRI fibrosis detection with strong correlation to fibrosis defined by gold standard histological analysis. C. Left: In-vivo DE-MRI correctly shows a high percent of fibrotic tissue in the human SAN. Right: Fibrosis found in the left atrium could represent a structural substrate for a reentrant AF driver and guide future targeted ablation. Panels B and C modified from Csepe et al.(47) with permission. Abbreviations as in Figure 4; CT-crista terminalis; IAS-interatrial septum, LPV/RPV- left/right pulmonary veins.

Recent Advances in Computer Modeling of AF and Fibrosis

The Evolution of Computer Models from 2D to 3D Patient Specific Atrial Structure

Computer modeling's unique ability to simulate different scenarios by incorporating multiple factors/conditions, which is not possible clinically or experimentally, has proven to be a very useful interrogation tool to determine the role of fibrosis in AF.(27,55) Multiple computer models have been developed with varying accuracies to describe atrial structure. Early work of 2D human atrial models based on posterior left atrial slices with structural remodeling demonstrated that heterogeneous spatial distribution of fibrosis governs AF dynamics.(10) To address the role of discontinuous transmural conduction due to fibrotic remodeling, a bi-layer geometry model with varying connectivity between the two layers, was developed by the Schotten(11) and Vigmond groups.(56,57) The former modeling study demonstrated that the progressive loss of endo-epicardial coupling to represent fibrosis stabilized AF.(11) The recent study by the Vigmond group(57) indicated that rotor cores, identified using in-vivo ECGI, are not localized over regions of highest fibrosis (Figure 6A). The conceptual bi-layer model provides a simplified and efficient computational framework to examine the effects of epi-endocardial dissociation(56); however, it does not model the actual 3D structural heterogeneity, e.g., fibrosis architecture and intramural complexities, which could be important parameters of specific fingerprints of AF drivers.

Figure 6. Heart Specific Computer Models can pick apart the Individual Fingerprints of AF Drivers.

A. Bi-layer atrial computer model incorporated with late gadolinium enhanced MRI (LGE-MRI) patient-derived fibrosis distribution at 1.25mm³ resolution compared to clinically acquired ECGI (CardioInsight) AF driver (PS-phase singularity) density from the same patient. From Roney et al.(57); used with permission. B. Left: Functional and structural characteristics incorporated into the ex-vivo based computer model. Middle: 3D reconstruction of ex-vivo CE-MRI. Right: Simulation of AF in high-resolution computer model reproduces reentrant AF driver (white arrow) seen by ex-vivo optical mapping. Modified from Zhao et al.(65) with permission. Abbreviations as in Figure 5.

More translational models should be developed by incorporating patient-specific 3D atrial geometry and fibrosis distributions based on in-vivo DE-MRIs or ex-vivo CE-MRIs.(58-60) One of these studies conducted by the Trayanova group using DE-MRI based 3D models demonstrated that reentrant AF drivers are perpetuated at fibrotics boundary zones.(60) These studies clearly delineate the crucial role of patient-specific fibrosis architecture in AF initiation and maintenance; however, there is still no consensus in the field as to how atrial fibrosis should be modelled.

Computer Modeling of Atrial Fibrosis and its Limitations

In-vivo DE-MRI can provide heart specific distribution of fibrosis that can be incorporated into atrial image-based computer models. There are currently two major categories of methodologies for modeling atrial fibrosis that treat a fibrotic voxel as either with altered passive conduction or non-conductive elements.(57)

The first method of modeling fibrosis incorporates the electronic loading of increased fibroblast remodeling that slows conduction. In computer models, fibroblast-myocyte coupling has been represented using a fibrosis kinetic ionic model coupled to neighboring normal atrial myocytes.(59-61) In one of these studies conducted by the Trayanova group using the sophisticated biophysics-based cellular model for fibroblasts and patient-specific fibrotic distribution based on in-vivo DE-MRI suggested that the distribution of atrial fibrosis dictates where reentrant AF drivers are perpetuated.(59,60) These studies also showed that simulated reentrant activities were exacerbated with increased fibroblast-myocyte coupling and absent when fibrosis was made non-conductive. Biophysics-based approaches that take into account fibroblast-myocyte coupling can reveal the role of this coupling in AF mechanism in-silico but the existence of fibroblast-myocyte coupling in human atria is still debated.(62)

The second method is simply the addition of non-conducting tissue(63,64) based on either in-vivo DE-MRI or high resolution histological data(10,55). An image-based computer model by the Smaill group has demonstrated that structural discontinuities due to patchy fibrosis can amplify discordant conduction alternans and provide a rate-dependent substrate for reentry.(55) On the other hand, models developed by the Panfilov group concluded that arrhythmia induction is more pronounced with the increase of both the spatial size and the degree of heterogeneity of fibrosis.(64) Interestingly, they discovered that AF induction associated with fibrosis was usually driven by a reentrant AF driver but devolves into multiple wavelets when the simulated atrial tissue size is increased.(64)

How to best model atrial fibrosis is still open for debate, and the models mentioned above, by definition, are but simplifications of reality.(55) Of note, the recent study by Roney et al. has delivered a very important reminder by showing how different methodologies for modeling fibrosis had distinct impacts on AF dynamics.(57) Thus, future computer models should incorporate the fewest assumptions, and wherever possible, more integrated information should be obtained directly and reliably from the same human heart in order to closely predict the relationship between fibrosis and AF.

Developing Integrated Computer Models based on Ex-Vivo Submillimeter High-Resolution Functional and Structural Studies of the Human Atria

Computer models could be substantially improved by substituting generic functional and structural parameters from existing literature with submillimeter functional and structural data collected directly from the same human heart, to create a truly patient/heart specific model.(58,65) However, the in-vivo low resolution of DE-MRI may mean that voxels at or just above the fibrotic threshold represent an unknown mix of myocytes, fibroblasts, and connective tissue, making the most important assumptions of the computer models less robust. Thus, integration of ex-vivo functional and structural studies may provide the most accurate submillimeter high-resolution data on the human heart in order to improve current human atrial models.

In our preliminary study(65), we created a 3D heart-specific atrial anatomy model by integrating 3D CE-MRI structural and panoramic optical mapping functional data to test AF induction, maintenance, and ablation strategies (Figure 6B). The 3D heart-specific atrial anatomy model utilizes an image-based approach, which not only includes an accurate 3D representation of fibrosis architecture, but also includes accurate atrial geometry, wall thickness variations, and myofiber orientation, in addition to functional data including regional action potential duration, conduction velocity, and rate-dependent behaviors. Further development of highly integrated heart-specific computer models may help define the most arrhythmogenic regions and improve our understanding of human AF mechanisms. For example, if the model could accurately reproduce functionally mapped AF patterns, then altering functional and structural features one by one can quantify and accurately isolate the individual contributions of arrhythmogenic factors as identifiable fingerprints of AF drivers. Based on our preliminary results(12), once AF driver tracks are identified by driver fingerprints, we propose substrate modifying ablation of reentrant tracks (SMART) to include initial lesions that disrupt the structurally defined and functionally confirmed microanatomic tracks of reentrant AF drivers with subsequent continuous lesions traced toward the nearest atrial anatomical border to prevent reentrant arrhythmia reinduction (Central Illustration).

Conclusion and Future Direction

To define future diagnostic and therapeutic targets, integrated electrical and structural high-resolution studies are required to address, in a reproducible and robust manner, the patient specific patterns of fibrosis and atrial tissue architecture that facilitate clinical AF. Based on recent ex-vivo experimental results and multiple clinical studies, we hypothesize that one of the primary mechanism driving AF in diseased human hearts is a limited number of localized intramural reentry within patient-specific 3D microanatomic tracks composed of fibrotically-insulated myobundles.

We suggest that both the broader use of high-resolution integrated ex-vivo mapping of more human atria with and without AF history together with 3D computational approaches are needed to distinguish the functional and structural fingerprints of these reentrant AF drivers at submillimeter resolutions. This information could then be utilized to improve interpretations of clinical imaging and electrode mapping techniques and overcome their millimeter resolution limitations. This systematically integrated approach would eventually shift ablation paradigms from extensive ablation of poorly estimated targets to a new mechanism-based, minimally damaging, targeted treatment of fibrotically-insulated microanatomic tracks harboring reentrant AF drivers.

Central Illustration.

Arrhythmogenic Fibrotic remodeling creates Microanatomic Reentrant AF Drivers in the Human Atria. Fibrotic remodeling in AF patients, specifically an increase in arrhythmogenic patchy and interstitial fibrosis, can insulate myofibers creating structural tracks favorable for microanatomic reentrant AF drivers. Once AF driver tracks are identified by driver fingerprints, we propose substrate modifying ablation of reentrant tracks (SMART) to include initial lesions that disrupt the structurally defined and functionally confirmed microanatomic tracks of reentrant AF drivers with subsequent continuous lesions traced toward the nearest atrial anatomical border to prevent reentrant arrhythmia reinduction. AF- atrial fibrillation; Endo- endocardium; Epi- epicardium.

Abbreviations

AF

atrial fibrillation

BIFA

box isolation of fibrotic areas

CE-MRI

contrast-enhanced magnetic resonance imaging

CFAE

complex fractionated atrial electrogram

DE-MRI

delayed-enhancement magnetic resonance imaging

SAN-

sinoatrial node