Rearrangement of Atrial Bundle Architecture and Consequent Changes in Anisotropy of Conduction Constitute the 3-Dimensional Substrate for Atrial Fibrillation

Sham-operated goats with acutely induced AF (aAF; n=6) and goats with AF persisting for 7 months (persAF; n=5) were studied (see online-only Data Supplement for detail).^2,4,5 After electrical cardioversion, atrial effective refractory periods were determined. A square contact mapping array (256 channels; interelectrode distance, 1.5 mm; Figure IA in the online-only Data Supplement) was positioned on the LA wall, and conduction during epicardial pacing outside the mapping area near the 4 corners was recorded. AF was re-induced, and fibrillation electrograms were recorded for 30 s (sampling rate, 1 kHz; filtering bandwidth, 0.5–500 Hz). Immediately after recording, the mapping array was mechanically fixed exactly at the recording site, and tissue deformation was prevented by application of a frame (Figure 1A; see online-only Data Supplement for detailed description). The sample was stored in the Karnovsky fast-acting chemical fixative.²⁶

Original high-resolution 3D MRI data sets (acquisition details and Movie I in the online-only Data Supplement) were reconstructed to yield a stack of 2-dimensional (2D) images in a plane parallel to the epicardial recording surface (Figure 1B) and were used to determine the direction of endocardial bundles and epicardial fibers (Figure 1C and 1D). A 16×16 grid, spaced to contain 1 recording electrode each in the center of any grid square, was overlaid on these 2D reconstructions. For each electrode position, the orientation of the endocardial and epicardial bundles within that square was determined. Squares not containing tissue were excluded from analysis. Only MRI images located within <220 µm of the epicardial boundary were used for epicardial bundle identification.

Epicardial unipolar fibrillation electrogram recordings (4 s) were analyzed using a wave mapping algorithm (see online-only Data Supplement) described previously.^1,3,4 Anisotropic behavior is defined as the property of being direction dependent. To study anisotropic conduction of fibrillation waves, 2 types of anisotropy were distinguished.

First, we determined anisotropy of CV (ie, fastest versus slowest propagation direction). This was characterized by degree of aniso tropy of CV (ie, the ratio between the long axis and the short axis of the ellipse fitted through local conduction vectors) and direction of anisotropy of CV (ie, the direction of the fastest propagation). Second, we determined anisotropy of conduction likelihood (ie, the most frequent versus the least frequent propagation direction). This was characterized by the degree of anisotropy of conduction likelihood (ie, the degree of spread of propagation directions) and the direction of anisotropy of conduction likelihood (ie, the most frequently observed propagation direction). See online-only Data Supplement for a detailed description of anisotropy calculation.

Pacing (200-ms cycle length) was performed outside each of the 4 corners of the mapping array. As a result of hemodynamic instability, pacing was not possible in 1 persAF goat. For each pacing site (n=4), 4 consecutive beats were analyzed. All analyzed beats (n=16) were merged per goat, and the degree and direction of anisotropy of CV were determined by ellipse fitting. Because the pacing location dictated the conduction direction, the degree of anisotropy of conduction likelihood was not calculated.

To quantify and compare the organization of the 2 types of anisotropy, a heterogeneity index (HI) was calculated. For each electrode, the mean absolute angle difference between the direction of anisotropy of CV or anisotropy of conduction likelihood at that electrode and the direction of anisotropy of CV or anisotropy of conduction likelihood at its 8 surrounding electrodes was determined. The HI was calculated as the mean of all absolute angle differences. The same HI calculation was used to quantify the organization of epicardial and endocardial bundle orientation.

To assess the relationship between electrophysiological and structural data, directions of anisotropy of CV and anisotropy of conduction likelihood were compared with the directions of endocardial bundles and epicardial fibers, both in aAF and in persAF. For quantitative analysis, angle differences were pooled within the aAF group and within the persAF group. Only absolute angle differences between 2 directions were used for analysis so that the minimal and maximal angle differences between the 2 compared directions equal 0° and 90°, respectively. For each comparison, the mean angle difference was calculated. As a metric, the ratio [R_N=N_{small angle differences (0°–20°)}/(N_{small angle differences (0°–20°)}+N_{large angle differences (70°–90°)})] was calculated to quantify directional correlation.

Data are expressed as mean±SD. Significance of differences in means between aAF and persAF was assessed using an unpaired Student t test. As angle differences were pooled for aAF and persAF, a mixed-effects ANOVA was performed to test for significance of differences in mean angle differences between these groups. For this purpose, a linear model was used with fixed effect of aAF or persAF and with random effect of the goat the angle difference belonged to. To test for the presence of a uniform distribution of angle differences between the 2 directions (P_uniform), a Kolmogorov–Smirnov test was used. There was no adjustment for multiple comparisons. P<0.05 were considered statistically significant.

Atrial effective refractory periods (at cycle lengths >250 ms; Figure 2) and AF cycle length (Table 1) were significantly shorter in persAF than in aAF. Further electrophysiological parameters characterizing the AF substrate are listed in Table 1. As expected,^4,5 the number of fibrillation waves and breakthroughs per second were higher in persAF compared with aAF, whereas the CV and width of fibrillation waves were lower. Figure 3 shows the representative wave maps and electrograms of aAF and persAF. In aAF (Figure 3A), a single AF cycle consists of only a few simultaneous waves, with no or few epicardial breakthroughs occurring. However, in persAF (Figure 3C), a single AF cycle consists of multiple simultaneous waves, with the occurrence of several breakthroughs. Furthermore, electrograms in persAF show more fractionation than in aAF. These results indicate the presence of a more complex AF substrate in the persAF group.

To study whether AF-related remodeling is accompanied by changes in atrial bundle orientation, the direction of endocardial bundles was compared with the direction of epicardial fibers in aAF and persAF (Figure 4). In aAF, endocardial bundle orientation often showed large angles to epicardial bundle orientation (R_N=0.44; P_uniform<0.001; Figure 4A). However, in the persAF group, this was significantly more pronounced with an even larger fraction of large angles (R_N=0.19; P_uniform<0.001, P<0.05 versus aAF; Figure 4B), indicating that epicardial bundles were oriented more perpendicularly to endocardial bundles than in aAF (see Table 2 for mean angle differences and Results section in the online-only Data Supplement).

All angle differences between the direction of anisotropy of CV and the direction of the endocardial and epicardial bundles are plotted in Figure 5A to 5D. In aAF, the direction of anisotropy of CV corresponded well with endocardial bundle direction (R_N=0.67; P_uniform<0.001), whereas there was no correlation with the epicardial fiber direction (R_N=0.54; P_uniform=0.08). In contrast, in persAF, the direction of aniso tropy of CV correlated better with epicardial bundle direction (R_N=0.66; P_uniform<0.001) and was oriented predominantly perpendicularly to the endocardial bundle direction (R_N=0.34; P_uniform<0.01).

The same analysis was performed for the direction of anisotropy of conduction likelihood (Figure 5E–5H). For aAF, the direction of anisotropy of conduction likelihood corresponded with the direction of the endocardial bundles (R_N=0.64; P_uniform<0.001). There was no relationship with the epicardial fiber direction (R_N=0.53; P_uniform=0.20). In the persAF group, the direction of anisotropy of conduction likelihood showed no relationship with the endocardial bundle direction (R_N=0.47; P_uniform=0.11) but a good correlation with the epicardial fiber direction (R_N=0.61; P_uniform<0.01; see Table 2 for mean angle differences).

In Figure V in the online-only Data Supplement, example maps of anisotropy of CV and anisotropy of conduction likelihood are given (dark to light color indicates increasing anisotropy). The local direction of both types of conduction anisotropy is indicated by arrows (light blue). The corresponding HI and the HI of the endocardial and epicardial bundle patterns are summarized in Table II in the online-only Data Supplement. No significant differences in HI of anisotropy of CV and in HI of anisotropy of conduction likelihood were found between aAF and persAF. However, the HI of anisotropy of conduction likelihood was significantly lower than the HI of anisotropy of CV for both groups. The HI of the epicardial bundle pattern was higher than the HI of the endocardial bundle pattern for persAF, but not for aAF.

Low epicardial macroscopic anisotropy of CV during sinus rhythm and pacing has been reported in experimental and clinical studies.^{18,19,21,22,27} Also here, epicardial macroscopic anisotropy of CV in goat LA free wall was low during pacing (mean anisotropy ratio was 1.7 in nonremodeled atria and 1.9 in remodeled atria).

Houben et al²⁰ reported low anisotropy of CV in the human RA during AF (median 1.2). Here, we also report low macroscopic anisotropy of CV during AF. To our surprise, the anisotropy ratio was similar in both remodeled and nonremodeled atria (mean 1.5), indicating that an increase in anisotropy of CV as such is not required to enhance AF stability in remodeled goat atria.

Epicardial anisotropy of CV varies between regions in the atria, partly because of differences in underlying histoanatomy. For example, epicardial activation in the LA free wall is relatively uniform with anisotropy ratios of 1.4 to 1.5.²⁷ Most of the LA endocardium consists of a smooth wall composed of overlapping layers of differently aligned myocardial fibers, with major changes in fiber orientation from the epicardial to the endocardial surface.^23,24 In contrast, epicardial propagation in the posterior LA exhibits marked CV anisotropy, ranging from 1.8 to 3.2.^1,4,22,25 The posterior LA wall is dominated by the septopulmonary bundle, where the muscle strands tend to follow a predominant direction.^23,24,28 Overall, regional variability in fiber orientation, together with tight electrical coupling between the fibers, leads to a lower degree of anisotropy when conduction is observed on a macroscopic scale.

Interestingly, there is a pronounced discrepancy between these macroscopic findings and reports of high anisotropy at the microscopic level,^13,16 which ranges from 4.5 to 9.8.¹³ Spach et al¹³ were the first to demonstrate highly anisotropic propagation in human atrial bundles. Not only the development of microscopic collagenous septa¹³ but also the redistribution of gap junctions toward cell ends¹⁶ lead to reduced lateral electrical coupling between myocytes, leading to tortuous conduction pathways in the transverse direction while the longitudinal conduction may be unaffected.^11–16 Furthermore, fibrosis-induced tissue discontinuities can cause electrical source-to-sink mismatch along these pathways, contributing to rate-dependent local slowing of conduction or conduction block.¹⁴ Finally, AF can be associated with fibroblast proliferation and differentiation into myofibroblasts.²⁹ Heterocellular electrotonic coupling of myofibroblasts and myocytes has been shown in native atrial tissue³⁰ and is thought to increase heterogeneity in excitability, refractoriness, and electrical load,³¹ potentially inducing heterogeneous slowing of conduction.³² These mechanisms may also contribute to microscopic zigzag conduction, allowing re-entry to occur in regions as small as 1 to 2 mm²,^13,16 largely facilitating re-entrant arrhythmias.^12,15–17

With the transition of aAF to persAF, the orientation differences between epicardial and endocardial bundles in goats become larger. This alteration in atrial bundle architecture might be related to progressive atrial dilatation, occurring during prolonged AF in humans³³ and goats (as much as ≈12% dilatation during the first 5 days).³⁴ Chronic atrial dilatation and stretch in fibrillating atria are caused by an elevation in LA pressure and increased wall stress^6,33 and loss of atrial contractility, resulting in enhanced atrial compliance.³⁴ As shown in Figure 4C and 4D, atrial dilatation itself may lead to rotation of epicardial fibers with respect to endocardial bundles, leading to a more perpendicular arrangement. This change would be most likely to occur in the thin epicardial layer, rather than in the endocardial bundles that are relatively thick and firmly anchored to the macroanatomy.

Atrial bundle rearrangement has consequences for the observed behavior of fibrillation waves. The alignment of the direction of anisotropy of CV with the endocardial and epicardial bundle directions showed marked differences in aAF and persAF. In nonremodeled atria, the direction of the fastest conduction was mainly dominated by the direction of the endocardial bundles (Figure 5A), in agreement with previous studies demonstrating that lines of block occur parallel to pectinate muscles in sheep¹⁸ and rabbit.²⁷ This suggests that the overall activation pattern is mainly determined by the endocardial bundle network and that this network is electrically well connected to the epicardial layer.

However, after 7 months of AF, the direction of anisotropy of CV is primarily determined by the epicardial fiber orientation (Figure 5D). Earlier studies from our group demonstrated that perpetuation of AF in goat is associated with increased heterogeneity in connexin 40 expression, an increase in intermyocyte distance (endomysial fibrosis) and myocyte hypertrophy.^4,5,27,35,36 Increased interstitial fibrosis underlies electrical uncoupling of side-to-side connections between neighboring muscle bundles,¹³ potentially leading to electrical dissociation within the epicardial layer^1,3,4 but more importantly between the epicardial layer and the endocardial bundle network.⁵ This is in keeping with a recent report by Verheule et al³⁵ showing that AF-induced endomysial fibrosis is much more pronounced in the outer millimeter of the atrium than in the deeper layers. In this study, a computer simulation of endo-epicardial dissociation and conduction showed that fibrosis exclusively occurring in the epicardial layer was sufficient to increase electrical dissociation and the complexity of AF.³⁵

Although the role of ectopic activity originating in the pulmonary veins is well established as a mechanism of paroxysmal AF,³⁷ persistent AF mechanisms are still under debate. Both experimental and human studies suggest rotor and ectopic activity to drive persistent AF.^7–10 However, detailed high-density direct contact mapping studies in goat and humans identified multiple wavelets, increasing longitudinal dissociation and transmural conduction (epicardial breakthrough) as dominant AF mechanisms.^1–5 In these studies, stable rotors were rare and not sustained, and breakthroughs were largely because of transmural conduction rather than ectopic activity.^1,2 The present study was not designed to clarify the mechanism of AF. Rather, our data provide insight into how bundle anatomy affects fibrillatory conduction in the atria, independent of the mechanism driving AF.

To date, studies relating AF propagation to the underlying 3D atrial anatomy are scarce. Dyssynchrony of electrical activation between the epicardial and the endocardial layer was first demonstrated by Schuessler et al.³⁸ Interestingly, the observed electrical endo-epicardial differences were larger in the thick trabeculated part of the RA. Eckstein et al⁵ confirmed these findings and reported that endo-epicardial dissociation increases with AF-induced remodeling. Yamazaki et al⁸ performed simultaneous epi/endocardial optical mapping and identified atrial scroll waves, transmural rotors that form and meander around regions of sharp transition in myocardial thickness, suggesting that wall thickness variability is an important factor for AF stabilization. Also in humans, fibrillatory conduction is thought to be influenced by the underlying anatomy. Allessie et al¹ reported interwave conduction block to be predominantly oriented parallel to the large pectinate bundles. Narayan et al⁷ suggested sustained rotors and repetitive focal beats as drivers of AF; however, the relationship with underlying atrial anatomy has not yet been studied. Our results may help to better understand the relationship between fibrillatory conduction and 3D atrial anatomy. In nonremodeled atria with extensive coupling between all layers, the large endocardial bundles prevail over thin epicardial fibers as pathways for the fastest propagation of fibrillation waves. In remodeled atria, in contrast, loss of endo-epicardial coupling caused by structural remodeling allows epicardial fibrillation waves to primarily propagate along the epicardial muscle bundles. Because during the process of AF the latter become oriented more perpendicularly to the endocardial muscle bundles, epicardial fibrillation waves also preferentially propagate perpendicularly to the endocardial muscle bundles. Because endocardial fibrillation waves conduct along the large muscle bundles, the change in the atrial bundle architecture further enhances endo-epicardial dissociation and thereby the 3D character of the histoanatomic AF substrate.

Anisotropy of conduction likelihood provides an additional measure of conduction anisotropy. In our study, the degree of anisotropy of conduction likelihood was strongly reduced in goats with persAF compared with goats with aAF. This observation is in agreement with the increase in complexity of AF propagation patterns over time as reported in this study and others.^1,2,4,5 As more and narrower waves simultaneously meander over the epicardial surface, preferential pathways are less likely to coexist.

Intuitively, it may seem obvious that the direction of the fastest conduction would also be the most frequently encountered conduction direction. However, neither in aAF nor in persAF a strong correlation between the direction of the fastest and the most likely conduction was found. Potentially this is because of the low HI of anisotropy of conduction likelihood. As such, no strong correlation between the direction of anisotropy of conduction likelihood at a certain electrode and the surrounding electrodes exists. In contrast, the HI of anisotropy of CV is 2× to 3× higher than the HI of anisotropy of conduction likelihood and, more importantly, comparable with the HI of epicardial fiber orientation. This supports the hypothesis that conduction likelihood depends more on the macroscopic atrial anatomy, whereas CV is determined by microscopic fiber orientation. In agreement with this, we found the correlation between the direction of fastest conduction with the underlying bundle direction to be larger than the match between the direction of bundles and most likely conduction.

Caution is warranted when extrapolating results on the relationship between conduction anisotropy and atrial anatomy from goat to humans. For example, human RA shows a strongly parallel orientation between endocardial bundles, whereas in goat RA endocardial trabeculae follow highly variable directions. However, we think that our general conclusions still hold. These include the observation that endocardial bundle anatomy and epicardial fiber orientation are important determinants of both the velocity and the likelihood of conduction of fibrillation waves and that the extent and the directionality of conduction anisotropy may change with progressive structural remodeling of the atria and electrical dissociation.

Although this study focuses on the relationship between atrial bundle orientation and anisotropy of conduction, the general conclusions are relevant for clinical practice, in particular the mechanisms contributing to domestication of persAF.

We propose that interindividual differences in AF substrate, as observed in humans,1 are partly caused by differences in tissue architecture and that in an individual patient, AF characteristics are also determined by the unique underlying atrial anatomy.

It is obvious that the strong influence of the underlying atrial anatomy on fibrillatory conduction demonstrated in this study is relevant for the development of computer models of AF and vice versa (as predictions from structure to function benefit immensely from quantitative modeling). Our data show that not only electrophysiological characteristics but also a detailed anatomy of endocardial and epicardial muscle bundles should be implemented in realistic computer models for AF, in particular if they are to be individualized.