Adenosine in human atrial fibrillation modulates and re-distributes atria-wide dominant frequencies and directs to reentry ablation sites with improved outcome Authors: David Calvo, Lucia Salinas, Pablo Martínez-Camblor, María Salgado, Ruth Álvarez-Velasco, Amaia Martínez-León, Minel Soroa, Daniel García-Iglesias, David Filgueiras-Rama, Juan José Gonzalez-Ferrer, Victoria Cañadas, Ricardo Salgado, Alba Cruz, Jorge Quintanilla, Andrés Redondo, Miguel Rodrigo, Julián Pérez-Villacastín, Nicasio Pérez-Castellano, Omer Berenfeld Published: 2025 ABSTRACT Understanding and therapy of persistent atrial fibrillation (AF) are suboptimal partly because the dynamics, mapping, and ablation of rotational activity that has been proposed to drive AF remain elusive. In the present work we challenge AF with Adenosine, an agonist activating the Kir3.x subfamily of inward rectifier potassium channels and known to accelerate rotors, while performing non-invasive panoramic mapping. We studied 25 persistent AF patients who underwent mapping twice at baseline (B1, B2) and again during 12 mg Adenosine (Ado) peak effect. Post-procedure, spatial distributions of dominant frequencies (DFs) and differences (ΔDFs) were analyzed. During baseline, highest 50% DF sites strongly associated with |ΔDF|≤ 0.4 Hz between B1 and B2 and most stable DFs (|ΔDF|≤ 0.1 Hz) had highest DFs. During Adenosine, rotational domains displayed cycle length shortening in agreement with DF increase. However, the DF modulation by Adenosine was heterogeneous and transformed a single to a bi-modal DF distribution. A new DFAdo Stability Index (DFASI) was thereafter calculated at each virtual node as the normalized DFAdo weighted by normalized stability (1-|ΔDF|Norm) to locate regions with the highest acceleration relative to baseline variations. Distributions of DFASI ≥ 0.8 domains and rotational domains approximate each other, but their joint probability was 0.3249 indicating not all DFASI ≥ 0.8 and rotational domains co-localized. On follow up, higher all arrhythmia freedom was observed in patients in which all DFASI ≥ 0.8 domains were ablated than in patients in which not all such domains were ablated. The study proposes a new method for identifying reentries driving AF and a physiologically-based guidance for their ablation.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-26262-z. Introduction The mechanisms underlying AF are still not fully clear and its therapy is suboptimal. The use of Adenosine, an agonist activating the Kir3.x subfamily of inward rectifier potassium channels,1 has been proposed for enhanced understanding of excitation patterns and drivers in AF. In studies with computer simulations2 and experimental optical mapping3 on AF models, increased acetylcholine concentration, that activates the same channels as Adenosine, has been demonstrated to accelerate the dominant frequency (DF) of rotational activity implicated in sustaining the arrhythmia. Subsequent optical and multi-electrode mapping studies in human atria have observed distributed AF driving reentries, with varying levels of stabilities and reduced cycle lengths in the presence of Adenosine.4 Non-invasive and bi-atrial electrocardiogram imaging (ECGi) mapping in AF patients also reveal reentrant and high-frequency domains5,6 correlated with intracardiac full-chamber mapping7 whose cycle lengths are reduced in the presence of Adenosine.8 However data on non-invasive bi-atrial mapping of reentry and atria-wide activation rate distribution during Adenosine, with concomitant ablation and outcomes in persistent AF patients, are scarce. Thus, the two-fold aim of this study is to provide a new atria-wide characterization of the impact of Adenosine on reentrant activity and local activation rate relative to baseline variations, which are typically neglected, and to demonstrate a new translational approach for a robust mapping of the Adenosine impact to potentially guide AF ablation. Methods Patients and workflow The Ethics Committee of the “Principado de Asturias” in Oviedo and “Clínico San Carlos” Hospital in Madrid approved the study and subjects gave written informed consent. All methods were performed in accordance with the relevant guidelines and regulations. Figure 1 illustrates the mapping, ablation and analyses workflow and dataflow. Patients admitted for ablation of persistent AF were instrumented with a 252-electrodes vest and underwent a CT scan. Thereafter 3-stages non-invasive atria-wide ECGi mapping (CardioInsight; Medtronic) and intracardiac navigation with regional mapping (Carto, Thermocool Smart-Thouch®, Penta-Ray® and a tetrapolar catheter; Biosense-Webster) were performed under general anesthesia.9Fig. 1Flow chart of actions and data prior, during, and following the mapping-ablation procedure. Yellow squares correspond to ECGi data acquisition and processing. Black arrows, work flow; Blue arrows, data flow; Green squares, intracardiac data acquisition, processing, and navigation; B1, basal 1 mapping; B2, basal 2 mapping; Ado, Adenosine mapping. Flow chart of actions and data prior, during, and following the mapping-ablation procedure. Yellow squares correspond to ECGi data acquisition and processing. Black arrows, work flow; Blue arrows, data flow; Green squares, intracardiac data acquisition, processing, and navigation; B1, basal 1 mapping; B2, basal 2 mapping; Ado, Adenosine mapping. Mapping during baseline and adenosine To establish baseline conditions, atria-wide virtual unipolar EGMs were acquired during 2, Basal 1 and Basal 2, non-invasive mapping stages (B1, B2; ≈ 20 min and a transeptal puncture in between) together with the RA and LA anatomy acquisition (Fig. 1). Thereafter, Penta-Ray® and tetrapolar catheters were stably positioned in the RA and LA appendages (LAA, RAA) respectively to record intracardiac contact bipolar EGMs. Following, 12 mg Adenosine was infused through the femoral vein, and a 3rd non-invasive atria-wide mapping stage (Ado mapping) was obtained together with the LAA and RAA contact bipolar EGMs at around the time of maximal atrioventricular block induced by the Adenosine. Reentry domains identification The continuous surface ECG time series were analyzed intra-procedurally on the CardioInsight mapping system to mark sequential TQ segments free of ventricular activity during the 3 mapping stages (B1, B2 and Ado). The marked sequence of TQ segments were processed by built-in algorithms to generate virtual EGM time series (1 kHz sampling rate) in each node of the virtual atria. Thereafter, the CardioInsight system used proprietary algorithms to process the virtual EGMs by means of the Hilbert transform to compute instantaneous phase in each node, to identify the presence and duration of rotational activity, and to present the local density of activity with ≥ 2 consecutive uninterrupted rotations in color-coded composite maps.5,6 Sites with high density of rotations displayed on composite maps overlapping at B1 and B2 locations (by visual inspection and majority of 3 expert observers blind to outcomes) defined reproducible rotational domains and all were considered for ablation. Due to regulatory restrictions, rotational domains identified during Ado mapping were not considered as ablation targets. Ablation procedures Circumferential pulmonary veins isolation (CPVI) was accomplished by ablation at the PVs antrum and confirmed by elimination of local bipolar potentials (temperature 43°; 20–35 Watts; Contact force 10–15 g). The B1 and B2 reproducible rotational domains on any PV antrum were targeted by the CPVI. For reproducible rotational domains located outside of the PV antrum, linear ablation or patches were applied by covering the distribution of singularity points (and connecting to unexcitable boundaries when available). All ablations were annotated on the anatomical map created by the intracardiac Carto mapping system. After ablation, external cardioversion was performed if the patient remained in AF. Post-procedure reentry and frequency analysis Following the mapping and ablation procedures, virtual EGMs and the atrial geometry were exported for post-procedural analyses of the same TQ segments used to generate intra-procedurally the rotational density maps (Matlab R2019a; Mathworks, Natick, MA). The cycle length, number of consecutive uninterrupted rotations, duration of the rotations, number of rotations per time unit, and the mapping time were calculated for the same TQ intervals analyzed intra-procedurally during the 3 mapping stages. Custom software was developed to compute DF maps for the virtual EGMs during the same TQ intervals. Each TQ segment (> 0.5 s long) of the exported virtual EGMs was pre-processed by a non-biased 1–20 Hz band-pass filter to minimize baseline wander and high-frequency harmonics. Thereafter the TQ segments were processed to generate DF for each mapping point as follows (See also Fig. S1 in Supplemental Material). Initially a short fast Fourier transformation (SFFT) was implemented on each TQ segment after Hanning windowing and zero padding to a uniform length of 8192 points for all segments.10 The SFFT signal in the frequency domain was then used to generate a power spectrum for each TQ segment. Thereafter the spectra of all the TQ segments in the sequence for each mapping stage were averaged according to the Welch method of weighting each spectra by length of the original TQ segments.11 The DF for each node was defined as the frequency with the highest power in the Welch averaged power spectrum with frequency steps of 0.112 Hz and was color-coded on the CardioInsight atria-wide 3D shell (with nearest-point interpolations between nodes) to create DF maps for the for the B1, B2 and Ado mapping stages. Contact bipolar EGMs recorded at the LAA and RAA during Adenosine mapping were also exported post-procedurally for additional frequency analysis (see Fig. S2 in Supplemental Material). Virtual EGMs were further processed as follows to analyze atria-wide alterations of the DFs during Adenosine (DFAdo) and the significance of the alteration relative to baseline alterations (DFB1, DFB2). For baseline alterations and stability quantification, the differences of the DFs of each virtual EGM between the 2 basal mapping stages (ΔDF = DFB2-DFB1) were calculated point-by-point first, and thereafter the absolute value of the differences, |ΔDF|, was obtained. Given the variability in fibrillatory dynamics and DFs across the patients,12,13 maps of DFAdo and |ΔDF| for further analysis were normalized for each individual patient to establish a unform scaling between minimal and maximal values common to all patients. Accordingly, the DFAdo map was normalized between 0 and 1, corresponding to the patient minimal and maximal values, respectively, for each patient. In addition, a common point-by-point stability weigh factor for each patient was generated as a normalized stability factor maps presenting the values of 1-|ΔDF|Norm at each point, whereby |ΔDF|Norm was calculated by subtracting from the |ΔDF| the minimal |ΔDF| value in that map, and then dividing by the difference between the maximal and minimal |ΔDF|s in that map. The resulting normalized stability factor maps ranged between 0 and 1 for the least and most basal stable local DFs, respectively. Finaly, to quantify the significance of the alterations in DF during Adenosine effect (DFAdo) we calculated a new parameter termed DFAdo Stability Index (DFASI), that was generated by multiplying point-by-point the normalized DFAdo map by the normalized stability factor 1-|ΔDF|Norm map for that same patient. The resulting DFASI maps thus correspond to the normalized value of the DFAdo weighted by the DF stability at baseline whereby points get the highest values, closest to 1, at locations that were the fastest during Adenosine and the most stable at baseline. Follow-up At time of enrolment, 6 follow-up visits were scheduled for the following 2-year period. Holter monitoring (24 h) and resting ECG were conducted at each visit. Antiarrhythmics were suspended after 3 months of blanking period if stable sinus rhythm was maintained. All patients were followed for a minimum of 1 year period. Statistical analysis Continuous variables are reported as mean ± standard deviation (SD). We report 95% confidence interval (CI95%) for most relevant quantities. Categorical variables are reported as number and percentage. The Chi-square test and Student’s t-test were used for comparisons between groups of those variables. Paired t-test was applied when appropriate. ANOVA methods were used for comparisons between multiple categories. For multiple repeated paired measurements, a General Linear Model was used when appropriate. A proportional hazard Cox regression model was used to test the influence of patients and clinical variables on ablation outcomes during follow-up. Hazard ratio with 95% CI is reported. All comparisons accounted for uneven data replications in patients and resampling methods have been used to make inferences when dealing with dependent data structures. Specifically, the general Bootstrap Algorithm14 (gBA) was used to contrast the intra- and inter-cluster similarity or difference between groups. Similarly, a standard naïve bootstrap method was used to calculate the CI95%. Analyses were performed by customized routines for the R environment (www.r-project.org), Origin 8.1 (OriginLab Corp, Northampton, MA) and Matlab R2019a (Mathworks, Natick, MA). Statistical significance was established at p < 0.05. Code availability The software developed in the present work is available upon reasonable request to the corresponding author. Results Patients and reentry mapping We studied 25 consecutive patients with persistent AF duration of > 6 months (mean ± SD: 15.3 ± 8.4 months; see Tables 2 and 3 for clinical characteristics). All patients underwent 3-steps B1, B2 and Adenosine mappings prior to ablation. In Fig. 2A we demonstrate ECGi maps of rotational density during those 3 mapping stages in a representative patient. The maps show domains of high density (red areas) with varying stability across the stages. Table 1 summarizes the characteristics of the non-invasive mapping and the rotational patterns it identified. During the B1 and B2 mapping of sequential TQ intervals we identified 81 reproducible rotational domains (3.2 ± 1.3 per patient) distributed over the RA (n = 44) and the LA (n = 37). Seventy one of the 81 domains (87.7%) also showed ≥ 2 consecutive rotations during Adenosine mapping. Overall, in 17 patients the domains were distributed over both the RA and LA, in 4 exclusively over the RA, and in 4 exclusively over the LA. The time percentage with ≥ 2 consecutive rotations presence considered for reproducible rotational domains was 63.5% (CI95% 51.8–75.3%) for B1 mapping, 66.3% (52.5–80.2%) for B2 mapping and 68.2% (53.9–82.5%) for Adenosine mapping (p = 0.622). Figure 2B summarizes the total number, mean duration and cycle length of all rotations identified during B1, B2 and Adenosine mapping. Among these 3 measurements, only the mean reentry cycle length varied between the mapping stages (158.2 ± 20, 161.8 ± 20 and 156.1 ± 19 ms for B1, B2 and Adenosine, respectively. p < 0.001). For the subset of longest lasting reentries of > 3 rotations, a respective increase in the number of consecutive rotations (3.51 ± 0.6, 3.62 ± 0.8 and 3.96 ± 1.3, p < 0.001) and duration (555.4 ± 115, 571.9 ± 124 and 623.5 ± 201 ms, p = 0.001) was observed between B1, B2 and Adenosine mappings (see Table 1).Fig. 2Rotational domains in ECGi mappings. (A) ECGi atrial maps showing density of rotations (composite maps; CardioInsight;) in the Basal 1 (top), Basal 2 (middle) and Adenosine infusion (bottom) mappings in a representative patient. Numbers on maps are number of reentries and colors indicate total number of rotations, regardless of belonging to the same reentry. High number of rotations were consistently observed in the lateral-high RA in this patient (red). Maps are from case N° 7. (B) Cumulative mean ± std of total number of reentries (left), average reentry duration (middle) and average reentry cycle length (right) during Basal 1, Basal 2 and Adenosine (Ado) mapping in each patient (N = 25). Reentry cycle length shows a variation across mapping stages (See also Table 1).Table 1Mapping and reentry characteristics in 25 patients.Mapping characteristicsBasal 1Basal 2Adenosinep valueNumber of TQ intervals (N)17 ± 3.215.2 ± 5.811.7 ± 60.017Duration of TQ intervals (ms; [CI95%])819.4 [799.8–838.2]805 [785.8–825.9]1150 [1016.8–1289.6]< 0.0001AF time processed (sec)13.96 ± 3.113.26 ± 3.514 ± 7.30.920Ratio of TQ to RR intervals0.64 ± 0.040.63 ± 0.060.71 ± 0.08< 0.0001Total time of mapping (min)6.7 ± 6.44.1 ± 2.70.31 ± 0.18< 0.0001Number of epicardial virtual nodes (N)2102 ± 3572102 ± 3572102 ± 357NARotational domains (All reentries lasting ≥ 1 rotation)Number of reentries (N)37.5 ± 16.835.7 ± 16.938.6 ± 23.50.277Number of consecutive rotations2.22 ± 0.72.24 ± 0.82.25 ± 0.90.756Cycle length (ms)158.2 ± 20161.8 ± 20156.1 ± 19< 0.001Duration of reentries (ms)351 ± 122360 ± 123350 ± 1440.211Reentries during mapping (N/sec)2.63 ± 12.7 ± 12.8 ± 1.10.677Total rotations during mapping (# of rotations/sec)5.9 ± 2.16 ± 2.36.4 ± 2.70.663Rotational domains (Reentries lasting ≥ 2 rotations)Number of reentries (N)21.2 ± 8.921.4 ± 1122.7 ± 13.30.845Number of consecutive rotations2.65 ± 0.62.65 ± 0.72.68 ± 0.90.850Cycle length (ms)156 ± 18158.6 ± 17153.5 ± 18< 0.001Duration of reentries (ms)416 ± 121421 ± 123413 ± 1570.669Reentries during mapping (N/sec)1.5 ± 0.61.6 ± 0.61.7 ± 0.70.557Total rotations during mapping (# of rotations/sec)4.1 ± 1.64.2 ± 1.84.5 ± 20.641Rotational domains (Reentries lasting ≥ 3 rotations)Number of reentries (N)6.5 ± 3.65.9 ± 4.95.7 ± 3.60.408Number of consecutive rotations3.51 ± 0.63.62 ± 0.83.96 ± 1.3< 0.001Cycle length (ms)157.6 ± 16157.8 ± 15157.9 ± 150.982Duration of reentries (ms)555.4 ± 115571.9 ± 124623.5 ± 2010.001Reentries during mapping (N/sec)0.5 ± 0.20.4 ± 0.30.4 ± 0.30.893Total rotations during mapping (# of rotations/sec)1.6 ± 0.81.5 ± 1.21.7 ± 1.30.815Significant values are in [bold]. Rotational domains in ECGi mappings. (A) ECGi atrial maps showing density of rotations (composite maps; CardioInsight;) in the Basal 1 (top), Basal 2 (middle) and Adenosine infusion (bottom) mappings in a representative patient. Numbers on maps are number of reentries and colors indicate total number of rotations, regardless of belonging to the same reentry. High number of rotations were consistently observed in the lateral-high RA in this patient (red). Maps are from case N° 7. (B) Cumulative mean ± std of total number of reentries (left), average reentry duration (middle) and average reentry cycle length (right) during Basal 1, Basal 2 and Adenosine (Ado) mapping in each patient (N = 25). Reentry cycle length shows a variation across mapping stages (See also Table 1). Table 1Mapping and reentry characteristics in 25 patients.Mapping characteristicsBasal 1Basal 2Adenosinep valueNumber of TQ intervals (N)17 ± 3.215.2 ± 5.811.7 ± 60.017Duration of TQ intervals (ms; [CI95%])819.4 [799.8–838.2]805 [785.8–825.9]1150 [1016.8–1289.6]< 0.0001AF time processed (sec)13.96 ± 3.113.26 ± 3.514 ± 7.30.920Ratio of TQ to RR intervals0.64 ± 0.040.63 ± 0.060.71 ± 0.08< 0.0001Total time of mapping (min)6.7 ± 6.44.1 ± 2.70.31 ± 0.18< 0.0001Number of epicardial virtual nodes (N)2102 ± 3572102 ± 3572102 ± 357NARotational domains (All reentries lasting ≥ 1 rotation)Number of reentries (N)37.5 ± 16.835.7 ± 16.938.6 ± 23.50.277Number of consecutive rotations2.22 ± 0.72.24 ± 0.82.25 ± 0.90.756Cycle length (ms)158.2 ± 20161.8 ± 20156.1 ± 19< 0.001Duration of reentries (ms)351 ± 122360 ± 123350 ± 1440.211Reentries during mapping (N/sec)2.63 ± 12.7 ± 12.8 ± 1.10.677Total rotations during mapping (# of rotations/sec)5.9 ± 2.16 ± 2.36.4 ± 2.70.663Rotational domains (Reentries lasting ≥ 2 rotations)Number of reentries (N)21.2 ± 8.921.4 ± 1122.7 ± 13.30.845Number of consecutive rotations2.65 ± 0.62.65 ± 0.72.68 ± 0.90.850Cycle length (ms)156 ± 18158.6 ± 17153.5 ± 18< 0.001Duration of reentries (ms)416 ± 121421 ± 123413 ± 1570.669Reentries during mapping (N/sec)1.5 ± 0.61.6 ± 0.61.7 ± 0.70.557Total rotations during mapping (# of rotations/sec)4.1 ± 1.64.2 ± 1.84.5 ± 20.641Rotational domains (Reentries lasting ≥ 3 rotations)Number of reentries (N)6.5 ± 3.65.9 ± 4.95.7 ± 3.60.408Number of consecutive rotations3.51 ± 0.63.62 ± 0.83.96 ± 1.3< 0.001Cycle length (ms)157.6 ± 16157.8 ± 15157.9 ± 150.982Duration of reentries (ms)555.4 ± 115571.9 ± 124623.5 ± 2010.001Reentries during mapping (N/sec)0.5 ± 0.20.4 ± 0.30.4 ± 0.30.893Total rotations during mapping (# of rotations/sec)1.6 ± 0.81.5 ± 1.21.7 ± 1.30.815Significant values are in [bold]. Mapping and reentry characteristics in 25 patients. Significant values are in [bold]. Atria-wide DFs mapping during baseline and adenosine In Fig. 3 we demonstrate the method of creating DF maps from virtual EGMs obtained during B1, B2 and Ado mapping stages based on DF maps for the individual TQ segments free of ventricular activity and with which rotational patterns were determined (Fig. 2). Panel A shows DF maps for 4 representative TQ segments in temporal order from early (top) to late (bottom). The DF map of each TQ segment was computed based on power spectra calculated with SFFT algorithm. Those individual DF maps demonstrate that during Ado mapping there were dynamic DF domains (color-coded orange-red) with values higher than during B1 and B2 mapping. Panel B presents averaged DF maps for the same patient as in Panel A (see Methods; all DF maps analyzed below are averaged as in this panel). The averaged DF map during Ado shows also orange-red domains with DF values higher than the maps during B1 and B2 mapping stages, demonstrating that Ado DF mapping during maximal RR interval prolongation captures the dynamic and heterogenous nature of the DF acceleration across the atria during Ado infusion.Fig. 3DF maps of TQ segments and their Welch’ average power spectra during Basal 1. Basal 2 and Ado mapping in a representative patient. (A) DF maps of representative sequential TQ segments in temporal order from early (top) to late (bottom) during Basal 1, Basal 2 and Ado mappings. DF is calculated for each TQ segment using SFFT. Highest DF domains (orange-red) are visible in DF maps during Ado mapping. (B) Cumulative DF maps of all TQ segments during Basal 1 (n = 15), Basal 2 (n = 14) and Ado (n = 8) based on SFFT-Welch’s averaged power spectra. Maps are from case N° 24. DF maps of TQ segments and their Welch’ average power spectra during Basal 1. Basal 2 and Ado mapping in a representative patient. (A) DF maps of representative sequential TQ segments in temporal order from early (top) to late (bottom) during Basal 1, Basal 2 and Ado mappings. DF is calculated for each TQ segment using SFFT. Highest DF domains (orange-red) are visible in DF maps during Ado mapping. (B) Cumulative DF maps of all TQ segments during Basal 1 (n = 15), Basal 2 (n = 14) and Ado (n = 8) based on SFFT-Welch’s averaged power spectra. Maps are from case N° 24. Baseline stability of DFs Figure 4 shows in Panel A additional representative DF maps obtained during B1 (DFB1, left) and B2 (DFB2, right) mapping stages with DF values ranging between about 5 (yellow) and 6.5 (red) Hz and the absolute difference |ΔDF| map in Panel B. The |ΔDF| values in this example vary between zero (red) and up to about 1 Hz (blue), with an overall absence of within-patient correlation in the whole cohort of patients analyzed (Pearson 0.36 ± 0.2; p = 0.29 ± 0.13). The histogram in Panel C displays the distribution of the |ΔDF| for the whole patient’s cohort. For the majority of the virtual EGMs, the DFs vary by ≤ 0.4 Hz and a general decay in number of virtual EGMs was observed for less stable behavior. Areas with highest DF reproducibility of |ΔDF|≤ 0.1 and |ΔDF|≤ 0.2 Hz represented 10.6 ± 6.5% and 32.6 ± 16.5% of the number of virtual EGMs, respectively.Fig. 4Baseline stability of DF measurements from virtual EGMs. (A) A posterior view of an ECGi-based DF bi-atrial map and sample virtual EGMs obtained during Basal 1 (DFB1, left) and Basal 2 (DFB2, right) mapping in a representative patient. Maps show stability at 6.1 Hz at the right atrium appendage and a reduction from 6.1 in DFB1 to 5.4 Hz in DFB2 near the pulmonary veins. (B) A map of the DF differences (ΔDF) between DFB1 and DFB2 maps in panels A and B. Note: values in the color bar are downward from zero (red) to 1 (blue). (C) Cumulative histogram of absolute DF differences (|ΔDF|) between Basal 1 and Basal 2 across all nodes and maps (N = 25). (D) 2D histograms of normalized DFB1 (left) and DFB2 (right) versus |ΔDF| (N = 25). DF values were normalized between 0 and 1 based on each patient’s minimal and maximal DF values. Maps in panels A and B are from case N° 21. Baseline stability of DF measurements from virtual EGMs. (A) A posterior view of an ECGi-based DF bi-atrial map and sample virtual EGMs obtained during Basal 1 (DFB1, left) and Basal 2 (DFB2, right) mapping in a representative patient. Maps show stability at 6.1 Hz at the right atrium appendage and a reduction from 6.1 in DFB1 to 5.4 Hz in DFB2 near the pulmonary veins. (B) A map of the DF differences (ΔDF) between DFB1 and DFB2 maps in panels A and B. Note: values in the color bar are downward from zero (red) to 1 (blue). (C) Cumulative histogram of absolute DF differences (|ΔDF|) between Basal 1 and Basal 2 across all nodes and maps (N = 25). (D) 2D histograms of normalized DFB1 (left) and DFB2 (right) versus |ΔDF| (N = 25). DF values were normalized between 0 and 1 based on each patient’s minimal and maximal DF values. Maps in panels A and B are from case N° 21. Interestingly, the mean DF of areas with highest DF reproducibility of |ΔDF|≤ 0.1 Hz was significantly higher than the mean DF at less reproducible sites (6.19 Hz [CI95% 6.18–6.2 Hz] vs. 5.96 Hz [CI95% 5.95–5.97 Hz] respectively, p < 0.001). The 2-dimensional histograms in Panel D of Fig. 4 further explore the relationship between DFs and their stability. It was found that the highest 50% DFs are strongly associated with variation by ≤ 0.4 Hz (χ2 with Yates correction test: p < 0.0001) and constitute 48% and 52% of all DFs during B1 and B2 mappings, respectively. Effect of adenosine infusion on DFs and their distribution Figure 5 shows the effect of Adenosine infusion on DFs across both atria surfaces in another representative patient. In Panel A, DFs during B1 mapping were mostly between 6 and 7 Hz (green-yellow tones). Panel B shows that during Adenosine, on the other hand, two regions of the atria notably accelerated to DFs of > 7 Hz (red) and increased DF gradients formed with surrounding areas. The histograms in Panel C show that the increased spatial DF gradient in the maps from baseline (Panel A) to Adenosine (Panel B) is accompanied by a modulation of DFs from a single (left) to a double (right) mode distribution, with a visible nadir at 7.14 Hz that splits the histogram into low and high band modes when Adenosine was infused.Fig. 5Adenosine effect on DFs in a representative patient. (A) Anterior (top) and posterior (bottom) views of DF maps during Basal 1 mapping with sample virtual EGMs signal segments and corresponding DFs. (B) DF maps with same view as in A during Adenosine mapping showing areas with increased DFs (red areas) relative to Basal 1 mapping. (C) Histograms of DFs during Basal 1 mapping shown in A (DFB1, left) and during Adenosine mapping shown in B (DFAdo, right). The distribution during Adenosine is showing a marked band split relative to the distribution during Baseline 1. Data are from case N° 20. Adenosine effect on DFs in a representative patient. (A) Anterior (top) and posterior (bottom) views of DF maps during Basal 1 mapping with sample virtual EGMs signal segments and corresponding DFs. (B) DF maps with same view as in A during Adenosine mapping showing areas with increased DFs (red areas) relative to Basal 1 mapping. (C) Histograms of DFs during Basal 1 mapping shown in A (DFB1, left) and during Adenosine mapping shown in B (DFAdo, right). The distribution during Adenosine is showing a marked band split relative to the distribution during Baseline 1. Data are from case N° 20. The partition of the DF distribution into slow and fast frequency modes was observed in 21 out of 25 (84%) patients and is further characterized in Fig. 6. At baseline (Panel A), histograms of DFB1 and DFB2 for the 21 patients are not statistically different and best fitting with single Gaussian distribution functions yields full width half maximum (FWHM) values of 1.73 and 1.79 Hz, and Adjusted R2 values of 0.8738 and 0.8153, respectively. Following Adenosine, Panel B shows that a single Gaussian distribution best fit of the DFAdo histogram (left graph, p < 0.001 vs. DFB1 and DFB2) has a wider FWHM of 2.12 Hz and lower Adjusted R2 value of 0.7322 compared with baseline, suggesting a re-distribution of DFs and reduced model accuracy relative to the distribution models in Panel A. We surmised that the re-distribution of DFAdo values around a band split, as seen in Fig. 5C, could be better modeled with a dual instead of a single mode distribution. Thus, a new histogram of DFAdo was created (right graph in Fig. 6B) in which each DF value was shifted by a visually determined (blinded to results) split frequency value (the nadir between the low and high frequency bands in the DFAdo histogram) specific for each patient (values of split frequency were 6.2 ± 0.8 Hz). The best fit of the split distribution with a bi-modal model resulted in an increased Adjusted R2 value of 0.8864 (p < 0.001 vs. non-split DFAdo histogram). In addition, the 2D histogram in Panel C shows that most of the virtual EGMs at low and high DFAdo bands maintained |ΔDF|≤ 0.4 Hz at baseline. Panels B and C together support the portrayal of the Adenosine effect as re-distributing the DFs from a single-mode into a bi-modal distribution.Fig. 6Cumulative analysis of the DFs in 21 patients during B1, B2 and Ado mappings. (A) Histogram of DFs during B1 (DFB1, left) and B2 (DFB1, right) mappings with superimposed single-mode Gaussian best fit (green) and fitting parameters. (B) Left: Histogram of DFs during Ado (DFAdo) mapping with superimposed single-mode Gaussian best fit (green) showing broadening of distribution relative to A and reduced Adjusted R2. Right: Histogram of same data in left graph transformed to have DFAdo = 0 at the specific band splitting DF of each patient (see Fig. 5C). A superimposed bi-modal Gaussian best fit (green) with low (red arrow) and high (blue arrow) bands shows a greater Adjusted R2 value than in left graph. (C) 2D histogram of DFAdo transformed to have 0 value at the band split frequency for each patient separately and |ΔDF| for Adenosine mapping (N = 21). Peak distributions at low and high bands (yellow bins) are visible at |ΔDF|≤ 0.4 Hz. (D) Temporal behavior of DFs in all ECGi nodes belonging to low (red) and high (blue) frequency DFAdo bands. Nodes during B1 and B2 mappings were assigned to the low and high frequency bands based on the DFAdo values at the nodes during Ado mapping. Cumulative analysis of the DFs in 21 patients during B1, B2 and Ado mappings. (A) Histogram of DFs during B1 (DFB1, left) and B2 (DFB1, right) mappings with superimposed single-mode Gaussian best fit (green) and fitting parameters. (B) Left: Histogram of DFs during Ado (DFAdo) mapping with superimposed single-mode Gaussian best fit (green) showing broadening of distribution relative to A and reduced Adjusted R2. Right: Histogram of same data in left graph transformed to have DFAdo = 0 at the specific band splitting DF of each patient (see Fig. 5C). A superimposed bi-modal Gaussian best fit (green) with low (red arrow) and high (blue arrow) bands shows a greater Adjusted R2 value than in left graph. (C) 2D histogram of DFAdo transformed to have 0 value at the band split frequency for each patient separately and |ΔDF| for Adenosine mapping (N = 21). Peak distributions at low and high bands (yellow bins) are visible at |ΔDF|≤ 0.4 Hz. (D) Temporal behavior of DFs in all ECGi nodes belonging to low (red) and high (blue) frequency DFAdo bands. Nodes during B1 and B2 mappings were assigned to the low and high frequency bands based on the DFAdo values at the nodes during Ado mapping. In Panel D of Fig. 6 we analyze the Adenosine DF bi-modal partition effect by grouping nodes on the CardioInsight atrial surface based on whether they belong to low (red) or high (blue) DFAdo band following Adenosine infusion. Comparing the DFs of the low and high frequency groups during B1 and B2 mappings reveals slight deceleration in the frequencies of the indistinguishable low and high groups (B1: 6.12 ± 0.15 vs. 6.16 ± 0.17 Hz, p = 0.8; B2: 5.92 ± 0.16 vs. 5.98 ± 0.16 Hz, p = 0.708). However, following Adenosine, the low (red) and high (blue) frequency groups separate, mostly by acceleration of the high frequency designated group (Adenosine: 5.82 ± 0.17 vs. 6.45 ± 0.21 Hz, p = 0.009). The analysis demonstrates that the DF acceleration induced by Adenosine is significant despite DF variations at baseline. High DFs stability and rotational domains distribution We proceed to analyze in Fig. 7 the spatial relationship between DF response to Adenosine, DFAdo, the DF variations at baseline, |ΔDF|, and the locations of reproducible rotational domains. Panel A displays the normalized DFAdo map during the Adenosine infusion and Panel B displays the baseline stability map quantified as 1-|ΔDF|Norm (most stable equals 1 and least equals zero) in a representative patient. In this example there were sites with high DFAdo levels in the posterior LA and lateral RA, but only the one in the RA co-localized with the region of maximal baseline stability.Fig. 7Atrial distributions of high DFAdo, stable basal DFs and rotational domains. (A) A normalized DFAdo map from a representative patient. (B) A map of the stability in DFs during the 2 basal DF mappings in the same patient as in A. High stability regions are in red and low stability in blue. (C) Map of the DFAdo map in panel A weighted by the basal Stability map in panel B results in a normalized DFAdo-Stability index (DFASI) map. Red arrow points to highest DFASI value domain in the lateral RA. (D) Rotational domains maps during basal mapping of the same patient. Red arrows point to a stable reentry domain with maximal number of rotations correlating with the high DFASI domain in panel C. Maps in A–D are from case N° 18. (E) Distribution of stable CardioInsight rotational domain (green) and DFASI ≥ 0.8 (red) across the atria of all patients. Numbers in parenthesis indicate the number of those occurring in the same patient. DFASI, DFAdo stability index; ILA, inferior LA; SLA, superior LA; LAA, LA appendage; IRA, inferior RA; SRA, superior RA; RAA, RA appendage. Atrial distributions of high DFAdo, stable basal DFs and rotational domains. (A) A normalized DFAdo map from a representative patient. (B) A map of the stability in DFs during the 2 basal DF mappings in the same patient as in A. High stability regions are in red and low stability in blue. (C) Map of the DFAdo map in panel A weighted by the basal Stability map in panel B results in a normalized DFAdo-Stability index (DFASI) map. Red arrow points to highest DFASI value domain in the lateral RA. (D) Rotational domains maps during basal mapping of the same patient. Red arrows point to a stable reentry domain with maximal number of rotations correlating with the high DFASI domain in panel C. Maps in A–D are from case N° 18. (E) Distribution of stable CardioInsight rotational domain (green) and DFASI ≥ 0.8 (red) across the atria of all patients. Numbers in parenthesis indicate the number of those occurring in the same patient. DFASI, DFAdo stability index; ILA, inferior LA; SLA, superior LA; LAA, LA appendage; IRA, inferior RA; SRA, superior RA; RAA, RA appendage. Panel C shows the map with the new DFASI generated by the product of the maps in Panels A and B and is presenting in red the location of the maximally accelerated DFAdo and the most baseline stable DF at the lateral RA region (red arrow). Panel D shows the CardioInsight rotational domains maps for this patient and demonstrates that the stable high rotational activity during B1 and B2 mappings at the lateral RA (arrows) colocalizes with the region of DFASI ≥ 0.8 in Panel C. We found that overall, the number of virtual atrial nodes with DFASI ≥ 0.8 represented 4.8 ± 4.6% of the total number of nodes, and selected 0.8 as a cut-off value for designating highest DFASI domains across all patients. In Panel E of Fig. 7, the diagram of the atria is marked with the numbers of regional cumulative reproducible rotational domains identified during the 2 basal mapping stages (green, n = 81), and the highest DFASI domains (DFASI ≥ 0.8; red, n = 95) as annotated by a majority decision by 3 observers inspecting visually the maps. Both the number of reproducible rotational domains and the highest DFASI domains per any region distributed equally normal across the atrial regions (p = 0.4480). Nevertheless, differences in the local values were observed. In the 17 patients with bi-atrial reproducible rotational domain, the highest DFASI domains were observed in both atria in 88.2% of the cases. In 4 patients with reproducible rotational domains exclusively in the RA, bi-atrial highest DFASI domains appeared in 2 cases, confined to the RA in 1 case and confined to the LA in 1 case. In the 4 patients with exclusive LA reproducible rotational domains, bi-atrial highest DFASI domains appeared in 3 cases and confined to the LA in 1 case. In general, 85.3% of the highest DFASI domains displayed rotational activities, and 50 reproducible rotational (62%) and 50 highest DFASI (53%) domains occurred simultaneously in the same location (black markings), resulting in a joint probability of 0.3249 (Fisher’s exact test for numbers per anatomical region: p = 0.2852) indicating that not all highest DFASI and rotational domains co-localized. Ablation and follow-up relationship with high DFASI locations CPVI was achieved in all the patients following mapping. Additional ablation targeted all reproducible rotational domains identified during the B1 and B2 mapping stages (Fig. 1). The CPVI coincided with encircling 13 reproducible rotational domains. For extrapulmonary rotational domains, wide lines were performed for 38 domains (23 in RA vs. 15 in LA), while ablation patches were performed in other 26 domains (18 in RA vs. 8 in LA). The longest axis of the ablation lesions was 2.4 ± 1.2 cm (range 0.5–7.3 cm). Overall, 77 reproducible rotational domains out of 81 were ablated. In 4 reproducible rotational domains ablation was not performed due to location deep in the left or right appendages and risk of perforation, or safety concerns of phrenic nerve palsy (See Table 2).Table 2Clinical and procedure characteristics for all patients. AF: atrial fibrillation; LA: left atrium; LVEF: left ventricle ejection fraction, TAPSE: tricuspid annular plane systolic excursion.Patients and procedure characteristicsNo of patients (N)25Time in AF (months)15.3 ± 8.4Range (6.4–40.9)Prior AF ablation (%)8Age (years)56.4 ± 9.6Male (%)72Body mass index (kg/m2)29.6 ± 5.3Creatinine Clearance (ml/min)105 ± 27.5CHADS-VASC scale1.04 ± 0.8HADS-BLED scale0.2 ± 0.5EHRA I/IIa/IIb/III/IV (%)0/12/20/64/4Hypertension (%)36Diabetes mellitus (%)8Heart Failure (%)28Sleep apnea (%)24Prior stroke (%)0LA diameter (cm)3.9 ± 0.5LVEF (%)55 ± 12.3TAPSE (mm)20.7 ± 5.4Reproducible rotational domains per patient (N)3.2 ± 1.5DFSAI ≥ 0.8 domains per patient (N)4.3 ± 1.3Total procedure time (min; not including CT scan)230 ± 29.7Fluoroscopy time (min)6.9 ± 2.1Total ablation time (min)52.4 ± 9Extrapulmonary rotational domains ablation time (min)19.6 ± 7.3 Table 2Clinical and procedure characteristics for all patients. AF: atrial fibrillation; LA: left atrium; LVEF: left ventricle ejection fraction, TAPSE: tricuspid annular plane systolic excursion.Patients and procedure characteristicsNo of patients (N)25Time in AF (months)15.3 ± 8.4Range (6.4–40.9)Prior AF ablation (%)8Age (years)56.4 ± 9.6Male (%)72Body mass index (kg/m2)29.6 ± 5.3Creatinine Clearance (ml/min)105 ± 27.5CHADS-VASC scale1.04 ± 0.8HADS-BLED scale0.2 ± 0.5EHRA I/IIa/IIb/III/IV (%)0/12/20/64/4Hypertension (%)36Diabetes mellitus (%)8Heart Failure (%)28Sleep apnea (%)24Prior stroke (%)0LA diameter (cm)3.9 ± 0.5LVEF (%)55 ± 12.3TAPSE (mm)20.7 ± 5.4Reproducible rotational domains per patient (N)3.2 ± 1.5DFSAI ≥ 0.8 domains per patient (N)4.3 ± 1.3Total procedure time (min; not including CT scan)230 ± 29.7Fluoroscopy time (min)6.9 ± 2.1Total ablation time (min)52.4 ± 9Extrapulmonary rotational domains ablation time (min)19.6 ± 7.3 Clinical and procedure characteristics for all patients. AF: atrial fibrillation; LA: left atrium; LVEF: left ventricle ejection fraction, TAPSE: tricuspid annular plane systolic excursion. 15.3 ± 8.4 Range (6.4–40.9) One patient converted during ablation to sinus rhythm and 2 to atrial tachycardia. The rest of the patients remained in AF and external cardioversion was performed. Three patients required incremental doses of diuretics due to worsening heart failure soon after ablation. Despite prolonged hospitalization, they recovered with no additional incidences. No other significant complications were recorded, nor acutely nor during follow-up for 1.3 ± 0.6 years (range 1–2.2 years). In 7 patients, persistent AF was documented during the blanking period and external cardioversion was performed, but in 1 patient the procedure and additional antiarrhythmic drugs failed to maintain sinus rhythm. To characterize the possible role of Adenosine challenge with high DFASI localization as guidance for targets in persistent AF ablation we retrospectively identified the patients in which all domains with DFASI ≥ 0.8 were ablated (either by CPVI or by rotational domains targeting). The identification was based on whether all DFASI ≥ 0.8 domains on the DFASI map overlapped with the anatomical locations of ablations as marked on the intracardiac Carto anatomical map of the same patient and determined by majority of 3 expert observers blind to the ablation outcomes. Figure 8A demonstrates maps from a representative patient in which a highest DFASI domain near the left inferior PV (left, black arrow) colocalized with the ablation markers (right, red arrow). Panel B demonstrates another patient with highest DFASI domains visible at the low posterior LA (left, black arrow) and the high RA. The LA highest DFASI domain is seen on the Carto map to not been ablated (right, blue arrow). Supplemental Fig. S3 displays additional representative examples of DFASI maps with varying locations of highest DFASI domains and ablations. Overall, 63 of 95 DFASI ≥ 0.8 domains (66%) were ablated. In Panel C, stratified cumulative survival curves for patients in which all DFASI ≥ 0.8 domains were ablated (red) vs. patients in which not all such domains were ablated (blue) show improved survival from all arrhythmia recurrences in the former group (p = 0.037). Considering the patients’ clinical characteristics shown in Table 3, including antiarrhythmic drugs and ablation times, demonstrated that ablation of all DFASI ≥ 0.8 domains corresponded to higher all arrhythmia freedom on follow-up (Hazzard ratio 3.8; CI95% 1.1–13.8). Further emphasizing the beneficial outcome of ablating all DFASI ≥ 0.8 domains, patients displaying reproducible rotational domains distributed over both atria (N = 17) and patients with reproducible rotational domains exclusively in the RA (N = 4) or LA (N = 4) did not show different survival rate when all DFASI ≥ 0.8 domains were ablated (Cox regression analysis; p = 0.816).Fig. 8Ablation at High DFASI domains and freedom of arrhythmias. (A) A representative DFASI map (left) with high value domain (black arrow) at an area which was ablated (red arrow in right panel). Maps are from case N° 20. (B) A representative DFASI map with high value DFASI domain (black arrow) which was not ablated (blue arrow in right panel). Maps are from case N° 25. (C) Survival probability for all arrhythmia freedom for all patients. Traces are for patients in which ablation included all domains with DFASI ≥ 0.8 (red, (+)) and those in which ablation did not include all such domains (blue, (−)). Number of patients at risk in the 2 groups are indicated below graph.Table 3Clinical and procedure characteristics grouped by patients in which all domains with DFASI ≥ 0.8 were ablated (either by CPVI or by rotational domains targeting) vs. those patients in which not all DFASI ≥ 0.8 were ablated. AF: atrial fibrillation; LA: left atrium; LVEF: left ventricle ejection fraction, TAPSE: tricuspid annular plane systolic excursion. AADs: antiarrhythmic drugs.Patients and procedure characteristicsAll DFASI ≥ 0.8 ablated N = 18Not all DFASI ≥ 0.8 ablated N = 7p valueTime in AF (months)16.4 ± 9.212.3 ± 5.30.276Prior AF ablation (%)11.100.358Age (years)57.7 ± 8.853.3 ± 11.90.324Male (%)83.342.90.043Body mass index (kg/m2)29.3 ± 4.330.7 ± 7.50.564Creatinine clearance (ml/min)103 ± 25111 ± 330.527CHADS-VASC scale1.17 ± 0.90.71 ± 0.80.235HADS-BLED scale0.22 ± 0.50.14 ± 0.40.730Hypertension (%)38.928.60.629Diabetes mellitus (%)11.100.358Heart Failure (%)42.922.20.302Sleep apnea (%)22.228.60.739Prior stroke (%)00NALA diameter (cm)3.9 ± 0.54 ± 0.50.999LVEF (%)53 ± 1359 ± 70.243TAPSE (mm)19.1 ± 0.524.7 ± 0.50.080Reproducible rotational domains per patient (N)3.6 ± 1.52.4 ± 1.10.081DFASI ≥ 0.8 domains per patient (N)4.5 ± 1.23.6 ± 1.30.161Total procedure time (min; not including CT scan)226 ± 33242 ± 120.225Fluoroscopy time (min)6.1 ± 1.78.6 ± 1.80.015Total ablation time (min)52.7 ± 951.6 ± 100.829Extrapulmonary rotational domains ablation time (min)21 ± 7.714 ± 2.20.108On AADs during follow-up (%)27.814.30.637 Ablation at High DFASI domains and freedom of arrhythmias. (A) A representative DFASI map (left) with high value domain (black arrow) at an area which was ablated (red arrow in right panel). Maps are from case N° 20. (B) A representative DFASI map with high value DFASI domain (black arrow) which was not ablated (blue arrow in right panel). Maps are from case N° 25. (C) Survival probability for all arrhythmia freedom for all patients. Traces are for patients in which ablation included all domains with DFASI ≥ 0.8 (red, (+)) and those in which ablation did not include all such domains (blue, (−)). Number of patients at risk in the 2 groups are indicated below graph. Table 3Clinical and procedure characteristics grouped by patients in which all domains with DFASI ≥ 0.8 were ablated (either by CPVI or by rotational domains targeting) vs. those patients in which not all DFASI ≥ 0.8 were ablated. AF: atrial fibrillation; LA: left atrium; LVEF: left ventricle ejection fraction, TAPSE: tricuspid annular plane systolic excursion. AADs: antiarrhythmic drugs.Patients and procedure characteristicsAll DFASI ≥ 0.8 ablated N = 18Not all DFASI ≥ 0.8 ablated N = 7p valueTime in AF (months)16.4 ± 9.212.3 ± 5.30.276Prior AF ablation (%)11.100.358Age (years)57.7 ± 8.853.3 ± 11.90.324Male (%)83.342.90.043Body mass index (kg/m2)29.3 ± 4.330.7 ± 7.50.564Creatinine clearance (ml/min)103 ± 25111 ± 330.527CHADS-VASC scale1.17 ± 0.90.71 ± 0.80.235HADS-BLED scale0.22 ± 0.50.14 ± 0.40.730Hypertension (%)38.928.60.629Diabetes mellitus (%)11.100.358Heart Failure (%)42.922.20.302Sleep apnea (%)22.228.60.739Prior stroke (%)00NALA diameter (cm)3.9 ± 0.54 ± 0.50.999LVEF (%)53 ± 1359 ± 70.243TAPSE (mm)19.1 ± 0.524.7 ± 0.50.080Reproducible rotational domains per patient (N)3.6 ± 1.52.4 ± 1.10.081DFASI ≥ 0.8 domains per patient (N)4.5 ± 1.23.6 ± 1.30.161Total procedure time (min; not including CT scan)226 ± 33242 ± 120.225Fluoroscopy time (min)6.1 ± 1.78.6 ± 1.80.015Total ablation time (min)52.7 ± 951.6 ± 100.829Extrapulmonary rotational domains ablation time (min)21 ± 7.714 ± 2.20.108On AADs during follow-up (%)27.814.30.637 Clinical and procedure characteristics grouped by patients in which all domains with DFASI ≥ 0.8 were ablated (either by CPVI or by rotational domains targeting) vs. those patients in which not all DFASI ≥ 0.8 were ablated. AF: atrial fibrillation; LA: left atrium; LVEF: left ventricle ejection fraction, TAPSE: tricuspid annular plane systolic excursion. AADs: antiarrhythmic drugs. Discussion The main findings in this study are that during non-invasive bi-atrial mapping of persistent AF: (i) The DF modulation by Adenosine during persistent AF is heterogeneous and leads to a spatial re-distribution of DFs. (ii) The locations with the highest DF acceleration relative to baseline variations approximately overlap the locations of reproducible rotational domains. And (iii), ablation of all accelerating rotational domains is associated with better follow-up outcomes, suggesting a possible new ablation guidance approach. The effects of adenosine on reentrant activity in AF Mapping reentries in the presence of Adenosine has been performed to elucidate on excitation patterns driving AF. Hansen et al. studied Adenosine effects on AF in a relevant ex vivo model of explanted atria of patients with a history of AF as well as in vivo in patients and computer simulations.4 Their study found that Adenosine induced a decrease in reentry cycle lengths, as in the present study, and suggested that shortening of the action potential refractoriness by Adenosine infusion may turn unstable reentry during AF into a more organized pattern of activation around a common path in fibrotic hubs that is better visualized. Their optical mapping observations in experiments were reproduced in a clinical setting of persistent AF. Using contact basket catheters, the investigators were able to improve stable reentry localization and ablation with AF termination in multiple patients. Consistent with the observations of this and other studies of an increase in activation rate and reentries stability, our study has found also that Adenosine induced a significant general reduced cycle length of reentries as well as increased number of rotations and duration prolongation in reentries with > 3 rotations (Table 1). Interestingly, Hansen et al. also describe situations in which a different reentries response could be observed when Adenosine challenged a reentrant driver that is stable at baseline. In those situations, rotors which tend to colocalize with regions of conduction block may destabilize as blockades may be eliminated by the increased excitability secondary to transmembrane potential hyperpolarization induced by the Adenosine presence, as proposed by computer simulations15,16 and employed clinically to test for dormant conduction paths.17 The destabilization of reentries may lead to their slowing or termination, and may explain our new observation that the change in activation rate following Adenosine is heterogeneous; while some virtual nodes and atrial regions accelerate, others do not and therefore DF gradients become more apparent (see Fig. 6D). The distinct fate of the reentries and DFs under the Adenosine challenge may indicate the possibly different role of each of the detected reentries in maintaining the fibrillation. The proposed distinction between driving and passive reentries stems from optical mapping in cholinergic sheep heart model demonstrating that only a subset of all reentries appearing in series during AF onset are contributing to its sustenance.18 Further research is needed to characterize the mechanisms that distinguish the driving from the passive reentries. In a recent publication using the similar non-invasive mapping system as in our study, Dhillon et al. explored the relationship of Adenosine infusion with potential AF drivers in the form of both rotations and focal discharges.8 They also found that Adenosine reduced the cycle lengths of reentries, but it did not increase the number of rotational activities and rather increased the number of focal discharges in the RA. Our observations on Adenosine effect on cycle lengths and density of rotors as described in Table 1 are consistent with those of Dhillon et al. however, focal discharges were not studied in our ECGi mapping because their origin is not clear; while repeated focal discharges may originate at fast local ectopic excitations expected to be diminished by Adenosine,1 they may also be the manifestation of intramural reentries.19,20 Reentrant activity leading to surface breakthroughs in human explanted hearts was indeed recently suggested whereby mapped stable RA endocardial micro-reentry followed myocardial strands of the pectinate bundles and may have transmitted electrical waves discontinuously to the mapped epicardium.21 The discontinuous propagation is consistent with breakdown of propagation at high frequency of activations which was observed at a similar anatomical structure in the sheep.22 The frequency mapping guidance of AF ablation There is substantial experimental and clinical evidence in support of considering the activation rate of the myocardium as an important parameter to study AF mechanisms. For the inverse-solution AF mapping approach as utilized in this study, frequency domain analysis and localization of the highest DF was proposed to enhance robustness in localization of reentrant drivers.23 In Fig. 5, the increase in maximal DF following Adenosine infusion (DFAdo) was associated with increased DF gradients consistent with previous atrial optical mapping experiments with a pacing stimuli at an increasing frequency.22 Optical mapping in a sheep acute AF model has shown that increasing the concentration of acetylcholine, that acts on the same inward rectifier potassium channel as adenosine, in the perfusate leads to increased DFs in both atria, but also increased the DF gradient between the two atria.3 We previously observed in patients that there is a link between the spatial distribution of rotational domains and the direction of the frequency gradients, with the atria activated at the highest frequency being the preferential location of the rotational domains.9 Moreover, circumferential PVI with ablation near the expected location of the main inter-atrial conduction fascicle decreased the frequency in the slowest atrium while the fastest atrium remained unaffected, supporting the idea that the rotational domains at the fastest atrium played a role in maintaining the observed gradient.9 The frequency gradients have been explored also in animal models, were disruption of electrical communication between the left and right atria during AF led to similar phenomena: In a cholinergic sheep model of AF, Mansour et al. elegantly demonstrated that ablation of Bachmann’s bundle and the infero-posterior pathway connecting left and right atria caused a decrease in the activation frequency in the slower atrium, while the frequency of the faster atria remained unaltered.24 Despite a sound mechanistic rationale, clinical attempts did not demonstrate significant improvements in ablation outcomes guided by DF mapping.25 One possibility for such outcome is a temporal instability in the frequency content during AF26 which would reflect altered activation patterns between different time windows of the mapping and the ablation. We previously demonstrated that using short signal segments when analyzing local activation frequency may result in instantaneous variability in DFs that can yield un-interpretable results.9 On the other hand, Li et al.27 quantified recurrences of high DF regions in patients with persistent AF by calculating correlations between DF maps at different time windows in noncontact LA panoramic mapping and showed that different patterns of DF consistently reappear. In the present study we employed the SFFT-Welch’s frequency analysis method on TQ segments to guarantee that no ventricular signals contaminate the quantification of the stability of the DF at baseline (Fig. 4). It was found that in most of the nodes those baseline DFs vary by < 0.4 Hz between B1 and B2 mappings (Fig. 4D). This variation is less than the FWHM values of 1.73 and 1.79 Hz found in DFB1 and DFB2 maps histograms, respectively (Fig. 6A), suggesting similar DF distributions over the two mapping stages, however the local variations in DFs in each individual node, as demonstrated in Fig. 4D can affect assessment of the DF modulation by Adenosine. To rank the increase of DFAdo relative to varying baseline values we therefor weigh the normalized DFAdo with the normalized baseline stability (opposite of variability) and created the DFASI maps whereby the nodes with the highest DFAdo and that the most stable DFB1 and DFB2 (DFASI ≥ 0.8) colocalized in a significant number with reproducible rotational domains at baseline (Fig. 7). The fact that the colocalization between highest DFASI domains and reproducible baseline rotational domains was only partial may be the result of the opposing effect of Adenosine on rotors stability4 with the split in the DFAdo distribution into a high and low frequency bands (Fig. 6B). A follow-up probability analysis for freedom of any arrhythmia shows (Fig. 8) that in patients in which ablation targeted all domains determined post-procedurally as having DFASI ≥ 0.8 the outcomes were better than in patients in which not all such domains were ablated. Notably the DFASI formalism we employ here is weighing equally the contributions of the DFAdo and the stability factor, however improved outcomes may be accomplished with unequal weights that are yet to be determined. Notwithstanding, it is suggested that the panoramic evaluation of the distribution of the DF increase by Adenosine relative to the baseline DF variation can provide a method for testing which reentries drive AF and a guidance to ablation of AF drivers. Limitations There are several limitations to consider in our study. First, the non-invasive CardioInsight mapping does not include the atrial septum as does the intracardiac Carto mapping system and localizations of domains and ablation sites in maps by the two systems were based on visual assessments as conventional in the field.5,6,28,29 The visual localization in different mapping system may have reduced the correlation between the DFs in non-invasive virtual and intracardiac contact bipolar EGMs relative to other in-silico studies that demonstrated accurate frequency and phase-based markers for non-invasive AF mapping.23,30 Our non-invasive results are also consistent with those of optical mapping experiments and clinical contact electrogram recordings, including a correlation between non-invasively mapped activation frequencies and reentrant drivers7,31 and their correlation with ablation outcomes.32 Second, the CardioInsight system processes only TQ segments which are free of ventricular activity, but may limit analysis of long duration reentries. Third, we did not perform a sensitivity analysis for the highest DFASI threshold of 0.8 for optimal outcome. Instead, we selected the threshold value such that the number of virtual atrial nodes displaying DFASI ≥ 0.8 represented 4.8 ± 4.6% of the total number of nodes (less than 10% of the atrial nodes for most patients) yielding conventional significance and specificity in matching between highest DFASI domains and ablation regions. And finally, several rotational domains identified on the mapping were not ablated for safety concerns. Conclusions A new non-invasive bi-atrial mapping analysis of DFs in AF during Adenosine infusion shows modulation and bi-modal re-distribution of the DFs, and can localize the regions with the highest acceleration of activation rate relative to rate at baseline. Ablation targeting all such regions resulted in better ablation outcomes on follow-up. Thus, the bi-atrial mapping during baseline and Adenosine with analysis of local activation stability and rate could potentially assist in physiologically-based guidance of AF ablation. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1