Increased atrial arrhythmia susceptibility induced by intense endurance exercise in mice requires TNFα

Atrial fibrillation (AF) is the most common supraventricular arrhythmia that, for unknown reasons, is linked to intense endurance exercise. Our studies reveal that 6 weeks of swimming or treadmill exercise improves heart pump function and reduces heart-rates. Exercise also increases vulnerability to AF in association with inflammation, fibrosis, increased vagal tone, slowed conduction velocity, prolonged cardiomyocyte action potentials and RyR2 phosphorylation (CamKII-dependent S2814) in the atria, without corresponding alterations in the ventricles. Microarray results suggest the involvement of the inflammatory cytokine, TNFα, in exercised-induced atrial remodelling. Accordingly, exercise induces TNFα-dependent activation of both NFκB and p38MAPK, while TNFα inhibition (with etanercept), TNFα gene ablation, or p38 inhibition, prevents atrial structural remodelling and AF vulnerability in response to exercise, without affecting the beneficial physiological changes. Our results identify TNFα as a key factor in the pathology of intense exercise-induced AF. Endurance exercise is associated with an increased risk of atrial fibrillation. Here, the authors show the adipokine TNFα is a crucial mediator of exercise-induced atrial fibrillation and irreversible atrial remodelling characterized by fibrosis and inflammation.

A trial fibrillation (AF) is the most common arrhythmia that is associated with ageing as well as hypertension, structural heart disease, hyperthyroidism and several other cardiovascular conditions 1 . Although exercise is known to provide enormous health benefits particularly by improving disease outcomes in cardiovascular patients 2 , it is now clear that high intensity endurance sports increase AF susceptibility in the absence of underlying cardiovascular disease [3][4][5][6] . The basis for the association between intense exercise and AF is poorly understood. AF is generally associated with atrial remodelling often characterized by increased parasympathetic nerve activity (PNA) 7 , atrial enlargement 8 , fibrosis 9 and inflammation 10 . A hallmark of endurance athletes is heart rate slowing due to elevated parasympathetic tone 11 and atrial hypertrophy 12 both risk factors for AF. Intense physical activity also elevates inflammatory factors such as CRP 13 and TNFa 14 , which can be activated by atrial stretch following elevated atrial pressures (B35 mm Hg) as occurs during intense exercise 15 . Indeed, although expression of TNFa was originally thought to originate primarily from macrophages 16 , other cells such as cardiac myocytes synthesize TNFa in response to myocardial stretch 17,18 and elevated cardiac TNFa induces atrial fibrosis, myocardial hypertrophy and AF 19 .
In this study we show that intense endurance exercise (involving swimming or treadmill running for 6 weeks) increases susceptibility to AF in mice and is associated with macrophage infiltration and fibrosis. This exercise-induced atrial remodelling is prevented by treatment with the TNFa inhibitor (etanercept) and is not seen in mice lacking TNFa gene.

Results
Physiological cardiac remodelling in exercised mice. Although the primary purpose of our studies was to assess the link between endurance exercise and cardiac arrhythmias, we began by characterizing our exercised mice. Six-week-old male mice were selected for swimming and treadmill running exercise at levels that exceeded 70% of the VO 2 max for the entire duration of their training protocols ( Supplementary Fig. 1A), which is similar to levels targeted by elite athlete training programmes 20 . All the mice were trained for 6 weeks (unless otherwise stated), and all the exercised mice were used in subsequent analysis (See Methods). As expected from previous studies examining endurance athletes 11 , heart rates estimated using telemetry ECG measurements declined (swim P ¼ 0.009, treadmill P ¼ 0.03, oneway ANOVA) progressively after initiating exercise ( Fig. 1 &  Supplementary Fig. 1C). The heart-rate differences (Po0.01, twoway ANOVA) resulting from exercise were eliminated by blocking the autonomic nervous system (via combined atropine and propranolol treatment, Fig. 1c). In addition, there were no differences (P ¼ 0.7, Student's t-test) in beating rates of isolated atria (which lack autonomic nerve inputs) between exercise and sedentary groups (Fig. 1d). These results support the conclusion (c) Heart rates derived from surface ECGs in anesthetized mice before and after autonomic blockade using atropine & propranolol. Administration of atropine and propranolol eliminated heart rate differences between 6-week exercised and sedentary mice. (d) To further assess the role of autonomic nerve input on exercise-induced HR reductions, atria were isolated and completely denervated. Field recordings of spontaneously beating atria were used to calculate beating rate. The n values in bars indicate number of mice used. Beating rates of denervated atria ex vivo did not differ between sedentary and 6-week exercised mice. Data presented as mean± s.e.m., # Po0.05 when compared with baseline using repeated measure one-way ANOVA with Dunnett's multiple comparison test. ## Po0.05 when compared with baseline using repeated measure two-way ANOVA with Sidak's multiple comparison test. NS, not significant.
that the exercise-induced reductions in heart rate arise primarily from differences in autonomic nerve activity, as typically seen in athletes 11 . Ventricles of exercised mice also underwent physiological remodelling, as in athletes 21 , characterized by left ventricular chamber enlargement (P ¼ 0.04, Student's t-test) and enhanced ventricular contractility at rest (P ¼ 0.002, Student's t-test) as well as following intraperitoneal injection of 1.5 mg g À 1 dobutamine (P ¼ 0.006, Student's t-test), as detailed in Table 1,  Supplementary Tables 1-3 and Supplementary Fig. 2.
Increased vulnerability to atrial arrhythmias in exercised mice. As documented in Fig. 2  Autonomic and electrical remodelling in atria of exercised mice. Candidate mechanisms for increased AF vulnerability with exercise include abbreviated action potential durations (APD) 23 and increased APD heterogeneity 24 both of which can occur with activation of muscarinic K þ currents (I K,Ach ) following parasympathetic nerve activity (PNA) 25 . Consistent with a role for elevated PNA, exercised mice showed larger HR responses to atropine than sedentary controls (Fig. 3a), and the heart rate variability component, which is associated with PNA 11 , was elevated in exercised mice (Fig. 3b, Supplementary Fig. 9). Furthermore, blockade of the PNA with atropine caused the effective refractory period (ERP) to be longer in exercised mice compared with sedentary ( Fig. 3c), despite the lack of ERP differences in the absence of drug (Supplementary Table 4). However, although atropine prolonged ERPs, it could not entirely prevent AF in exercised mice (Fig. 3d). These effects of PNA blockade on ERP and AF in exercised mice are consistent with the longer action potential durations (APDs) in isolated exercised atria and cardiomyocytes (Fig. 4a,b), which were associated with increased L-type Ca 2 þ currents (Fig. 4c,d) without changes in voltage-gated K þ currents or I K,Ach (peak amplitude, desensitization or deactivation) measured in atrial cardiomyocytes from exercised mice ( Supplementary Fig. 10). As altered RyR2 function has been linked to AF in humans 26,27 and mice 28 , pathological cardiac remodelling 29 , and elevated oxidative stress 30 with exercise 31 , RyR2 phosphorylation and oxidation levels were measured. We found increased (Po0.05, One-way ANOVA) RyR2 phosphorylation at S2814 in the atria from mice after 6 weeks of swimming or after only four sessions of swimming ( Fig. 4e). Interestingly, the level of phosphorylation at S2814 in the ventricles was significantly decreased in mice after 6 weeks of swimming. In contrast, no differences in RyR2 oxidation, or RyR2 phosphorylation at position S2808 (PKA-dependent) or CaMKII phosphorylation at T287 were detected between the groups (Supplementary Fig. 11). In addition, no change in the expression level of RyR2 or CaMKII was seen between the groups (Supplementary Fig. 11). Elevated RyR2 phosphorylation at 2814 correlated with trends towards increased Ca 2 þ spark activity (P ¼ 0.06, Fig. 4f, Student's t-test) as well as Ca 2 þ transient amplitudes (P ¼ 0.06, Supplementary Fig. 11, Student's t-test) with no change in Ca 2 þ transient decay rates in swimming exercised atrial myocytes as compared with sedentary mice.
Structural remodelling in atria of exercised mice. The inability of PNA blockade to entirely prevent AF vulnerability in exercised mice combined with the increased AF incidence in isolated denervated exercised atria motivated us to examine other factors previously linked to AF including atrial enlargement 8 , interstitial collagen deposition (fibrosis) 32 and inflammation 10,33 . Exercised mice increased (P ¼ 0.01, Student's t-test) interstitial collagen deposition (Fig. 5a,b) in atria without differences (P ¼ 0.4, Student's t-test) in ventricles. As tissue fibrosis is invariably associated with inflammation in AF 10,33 , cardiac sections were examined with toluidine blue (Mast cells), Ly6b.2 antibodies (primarily neutrophils and recently activated inflammatory monocytes 34 ) and Mac-3 antibodies (macrophages and monocytes). Although no elevations in mast cells or Ly6b.2positive cells were observed acutely (2 days) or after 6 weeks of exercise training, Mac-3-positive macrophage/monocyte numbers were increased (P ¼ 0.005, Student's t-test) after 6 weeks of exercise (Fig. 5c,d). Increased macrophage/monocyte numbers, without neutrophil infiltration, support the conclusion that inflammation in exercised atria does not result from inflammatory cell infiltration as might occur with myocyte damage. No differences (P ¼ 0.67, Student's t-test) in staining patterns (Ly6b.2 and Mac-3 staining) were observed in ventricles between exercise (16.26±2 cells mm À 2 ) and sedentary mice (14.1 ± 4 cells mm À 2 ). We also observed increased (P ¼ 0.03, These structural changes in atria from exercised mice were associated with slowed (P ¼ 0.02, Student's t-test) conduction velocities and prolonged (P ¼ 0.003, Student's t-test) P-wave durations ( Fig. 5e-h) without changes in either connexin-40 or connexin-43 expression in swimming mice compared with sedentary mice (Supplementary Fig. 12). Interestingly, the organizational index of ECG recordings during AF episodes in exercised atria were greater, consistent with models of structural remodelling in AF 22 (Supplementary Fig. 7). Importantly, if we examined mice 6 weeks after the termination of the 6-week swim  training, the atrial AF vulnerability and fibrosis persisted (P ¼ 0.02, Student's t-test), while HR changes were reversed after detraining (to 492±7 beats/minute, P ¼ 0.02 compared with 6 weeks swim exercised, Student's t-test), establishing that the exercise-induced structural remodelling is not readily reversible ( Supplementary Fig. 13).
Role of TNFa in exercise-induced atrial structural remodelling. To gain insight into the possible genetic and biochemical mechanisms underlying exercise-induced AF, we performed gene set enrichment analysis (GSEA) of microarray gene expression values that revealed enrichment of several signalling pathways (p38 MAPK and NFkB pathways) associated with inflammation in exercised atria ( Supplementary Fig. 14A,B, Supplementary Table 5). Although these enriched pathways are regulated by several factors 36,37 , they are nevertheless downstream of, or crosstalk with, TNFa 36,38 . These observations, combined with previous studies implicating TNFa in the promotion of AF 19,39 , suggest that TNFa plays a role in AF pathogenesis with exercise. In support of this possibility, acute bouts of exercise activated (P ¼ 0.03, Student's t-test) NFkB, the canonical downstream transcription factor of TNFa, in atria, but not in ventricles Fig. 6a,b, Student's t-test). Acute bouts of exercise also increased the phosphorylation levels of other downstream factors of TNFa signalling such as IkB (P ¼ 0.01, Student's t-test) and p38 MAPK (P ¼ 0.03, Student's t-test) compared with sedentary controls (Fig. 6c,d).
To more directly assess the involvement of TNFa in exercisemediated atrial remodelling, the pharmacological inhibitor of  Table 6) without affecting either the mild ventricular chamber dilation or HR reductions induced by exercise (Supplementary  Table 7). Moreover, APD prolongation (Po0.01, Student's t-test) was also seen with exercise in TNFa À / À atrial myocytes, with no change in the amplitude or resting membrane potentials (Supplementary Fig. 10E-H).
Previous studies have shown that increased fibrosis is linked to TNFa and p38 activation 40 , and we found that p38 phosphorylation was acutely elevated by exercise (Fig. 6d) in addition to the observed chronic differential expression of p38 pathway related genes (Supplementary Table 5). Consistent with a role for p38 activation in exercise-induced remodelling, treatment of mice with the p38 inhibitor SB203580 during the 6-week exercise period completely prevented AF (P ¼ 0.021, Fig. 7a,c, w 2 -test) as well as atrial fibrosis (Po0.01, Fig. 7b,d, oneway ANOVA). In addition, the acute exercise-induced elevation in p38 phosphorylation was significantly reduced in TNFa À / À mice (P ¼ 0.5, Supplementary Fig. 14C, Student's t-test) supporting the conclusion that exercise induces atrial-selective remodelling via TNFa-dependent p38 activation.

Discussion
Our results support the conclusion that increased PNA and structural atrial remodelling, characterized by inflammation, hypertrophy and fibrosis, increase AF vulnerability in intense exercised mice. Our studies show that the mechanosensitive inflammatory cytokine TNFa is central to the pathophysiology of AF induced by intense exercise but does not appear to interfere with the beneficial physiological ventricular remodelling as well as the electrical remodelling seen with exercise. Involvement of TNFa in exercise-induced atrial remodelling is supported by the enrichment of the p38 and NFkB pathways [36][37][38] in our microarray analysis, by increased IkB and p38 phosphorylation in atrial lysates, and by elevated NFkB activity in atria, but not ventricles, following acute bouts of exercise. Although altered TNFa-dependent signalling can occur with or without changes in TNFa levels, or TNFa shedding 41 , we were unable to determine whether atrial TNFa levels change with exercise because the specificity of commercially available antibodies could not be verified by us (using TNFa À / À heart lysates). Although our studies focused primarily on swimming-based exercise in mice, we also observed similar responses to exercise in mice undergoing tread-mill running (that is, reduced heart rates, mild ventricular hypertrophy, increased vagal tone, increased incidence of AF both in vivo and ex vivo, and increased fibrosis in atria but not ventricles). Although we did not fully explore the role of TNFa in the response of mice to tread-mill running, the similar responses in the swimming and tread-mill running mice suggested that interference with TNFa will also be effective in treadmill exercised mice that will require further studies to validate.
Involvement of TNFa-dependent pathways readily explains the structural atrial remodelling with exercise since TNFa is an integral component of inflammatory responses, by recruiting circulating monocytes 42 and regulating resident tissue macrophages 43 . In addition, TNFa is a central regulator of cardiac hypertrophy 19 and fibrosis that involves altered p38 activity in fibroblasts 17,44 . In this regard, we found that pharmacological blockade of p38, as with TNFa inhibition, also prevented AF and structural remodelling without interfering with the physiological changes in heart rate and the ventricular chambers induced by exercise. The underlying mechanisms for the activation of these pathways and the differential pathological remodelling on atria versus ventricles with exercise are unclear. However, stretch has been shown to activate TNFa in cardiac 17,18 , skeletal 45 , fibroblasts 46 and smooth muscle 47 cells, and we noted that exercise in mice leads to sharp elevations in diastolic atrial pressures exceeding 30 mm Hg (Supplementary. Fig. 15), as seen in humans 15 . Although the pericardium constrains heart volumes, the atrium expansion is (B2-fold) greater than ventricular expansion under conditions when the pericardium becomes engaged 48 , and this may underlie the rise in serum TNFa 14 as well as serum atrial natriuretic peptide 49 seen in exercising humans. A role for stretch in atrial remodelling resulting from intense exercise in our mice is supported by our findings that atria have a far greater compliance than ventricles ( Supplementary Fig. 15) as well as by previous studies showing that atrial fibroblasts undergo transformation to proliferating, collagen-secreting myofibroblasts in response to stretch more easily than ventricular fibroblasts 50 . However, exercise induces a multitude of metabolic, biochemical and physiological changes in multiple cells types that could lead to stretch-independent activation of TNFa and its downstream signalling molecules like p38 (ref. 51). Clearly, additional studies will be required to address the underlying mechanisms for TNFa/p38-mediated atrial remodelling with exercise. In this regard, we are currently examining the effects of tissue-specific TNFa deletion including immune cell deletions on exercised-induced atrial remodelling in our mice. Our findings predict that interference with, or reductions in, TNFa-dependent signalling pathways could prove useful in minimizing the harmful consequences of intense exercise on atria, without interfering with either the performance of endurance athletes or the many otherwise highly beneficial effects of exercise. Nevertheless, since previous clinical trials with etanercept in advanced heart failure patients were terminated due to a lack of efficacy 52 , despite showing promise in animal studies 53 , etanercept may have limited efficacy in preventing detrimental atrial remodelling with exercise, even though etanercept is equally effective in animals and human in other diseases, such as rheumatoid arthritis 54 . It is worth noting that, although treating mice with human-based etanercept is expected to evoke an immune response, intraperitoneal injection of etanercept has been used often in animal models to treat TNFa-dependent inflammatory conditions 53,54 . The modulation of pathways downstream of TNFa such as p38 might also represent viable targets for future interventions. In this regard, p38 inhibitors have been used in heart failure models and shown to reduce cardiac apoptosis and fibrosis in the ventricles 55 . These findings are intriguing because acute bouts of exercise elevated p38 phosphorylation, and chronic p38 inhibition prevented the adverse structural remodelling in the atria associated with 6 weeks of exercise.
It should be mentioned that inhibition of the PNA reduced, but did not eliminate AF susceptibility in exercised mice. PNA involvement in exercise-induced AF is anticipated based on its link to AF in endurance athletes 56 , which presumably occurs via effects on atrial ERP (via I K,Ach ). Consistent with this possible mechanism, our exercised mice had signs of elevated PNA (that is, elevated heart rate variability and lower heart rates), similar to that seen in human athletes 11 . Indeed, following PNA blockade, atrial ERPs prolonged more in exercised mice compared with controls and these differences correlated with longer APDs in cardiomyocytes from exercised mice that were not associated with alterations in K þ currents but were linked to enhanced L-type Ca 2 þ currents (I Ca,L ), as was also reported previously in ventricular myocytes due to changes in PI3Kinase signalling with exercise 57 .
Our studies also uncovered modest increases in RyR2 phosphorylation at position S2814 (but not S2808) following swimming exercise that was accompanied by trends towards increased spontaneous Ca 2 þ spark activity in atria. Increased RyR2 phosphorylation and Ca 2 þ leak as a result of CaMKII activation has previously been reported in human AF patients 26,27 as well as mouse AF models 28 . In this regard, CaMKII can be activated by a number of inflammatory factors, including TNFa 58 , as well as a number of inter-related mechanisms including elevated Ca 2 þ and oxidative stress 59 . Interestingly, we did not observe any changes in CaMKII phosphorylation at T287 or expression levels of CaMKII with exercise, although recent studies have questioned the specificity of CaMKII phospho-specific antibodies 60 . Future studies will clearly be required to determine the roles and links between RyR2, CaMKII and TNFa in atrial remodelling induced by exercise.
As seen in our exercised mice, human athletes also show reductions in heart rate mediated by changes in autonomic nerve activity that reverse after long-term deconditioning 61 , while chamber dilatation persists. In contrast, it remains unknown whether cessation of exercise can eliminate the atrial remodelling and AF vulnerability. In this regard, we found that reversal of atrial fibrosis and enhanced vulnerability to atrial arrhythmias persisted after 6 weeks after cessation of swimming exercise (for 6 weeks), despite the reversal of HR slowing as in human. If exercised TNFa À/ À and SB203580 treated mice. The n value shown on bar graphs and present number of mice used. Data presented as mean ± s.e.m. w 2 -test *Po0.05 when exercised untreated is compared with exercised þ Etn, exercised TNFa À/ À or exercised SB203580 treated mice. **Po0.05 when exercised untreated is compared with exercised þ Etn, exercised TNFa À/ À , or exercised SB203580 treated mice using one-way ANOVA with Dunnett'smultiple comparison test. Scale bar, 100 mm.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms7018 ARTICLE these observations are applicable to humans, they highlight the critical importance of identifying and interfering with pro-fibrotic pathways activated by exercise. These results further suggest that exercise as well as other factors inciting atrial structural remodelling (such as hypertension, valve disease, heart failure) may accumulate with age, which readily explains the strong link between age and AF 62 including in endurance athletes 63 . On the other hand, it is clearly conceivable that ageing could influence the response of the atria to stresses induced by exercise and other cardiovascular conditions. Nevertheless, consistent with the potentially cumulative and irreversible nature of atrial structural remodelling, we failed to prevent the increased fibrosis in the atria of exercised mice when treatment with etanercept was started (3 weeks after initiating swimming exercise). Interestingly, many studies have established that ageing and cardiovascular disease are themselves accompanied by inflammation, with TNFa being involved in both conditions 64 . Future studies that are designed to identify individuals at risk for exercise-induced AF should develop methodologies for the longitudinal assessment of atrial remodelling in athletes. It is also imperative for future studies to identify the training and exercise conditions (intensity, duration, frequency) that most strongly predispose to pathological atrial remodelling; we are currently engaged in studies designed to address precisely these questions in our mouse models. Despite the similar AF and fibrosis between our results and exercised rats 7 , the extrapolation of rodent results to human athletes clearly requires caution. Besides having much smaller hearts, mice also have 10-fold greater HRs than humans, thus potentially making mice less susceptible to arrhythmias. Nevertheless, the ERPs in mice are also proportionally shorter, which probably explains the many previous mouse studies that reproduce human cardiac arrhythmias. Another notable difference between humans and mice, of particular relevance to our studies, is the relative differences in heart rate increases during exercise; in humans, heart rate can be increased by 3-4 times above baseline heart rates while in mice the heart rates increase by only B50%. Consequently, mice are expected to rely much more on stroke volume increases in order to elevate cardiac output during exercise compared with humans. If atrial stretch is indeed a major factor in atrial remodelling, then exercise is predicted to create a far stronger stimulus for atrial remodelling in mice compared with humans. While these species differences in response to exercise could explain the appearance of atrial hypertrophy and inflammation after only 6 weeks of intense exercise, it is also important to note that the atrial filling pressures changes during exercise are remarkably similar in mice and humans (that is, compare Supplementary Fig. 15A with Reeves et al. 15 ).
It should be mentioned that our exercise programmes were not voluntary. Nevertheless, stress was minimized in our exercised mice in several ways. Mice swam against a water current thereby avoiding the drowning stress associated with tail-weights, bubbling or mild detergents to break the surface tension of water. In addition, with the treadmill exercise, the use of electrical shocks was limited to the pre-training sessions and not used during the remainder of the exercise protocol. We believe that stress is not a major factor in the atrial remodelling in the mice for several reasons. First, the HR in our animals decline as early as 1 week after the start of the exercise protocol, whereas mice subjected to psychosocial stress show chronic HR elevations 65 . Second, though fecal corticosterone levels tended to be higher in exercised versus sedentary mice, the difference was not significant (Supplementary Fig. 1B). The interpretation of these corticosterone measurements is complicated because acute and chronic 66 intense exercise in humans independently causes elevated cortisol levels. Additional studies are needed to understand the link between stress, exercise and AF induction.

Methods
Experimental animals and exercise protocols. All mice were housed at the Division of Comparative Medicine animal facility in accordance with the standards of the Canadian Council on Animal Care. Ethical approval for all animal experiments was granted by the University of Toronto. In preliminary studies we found that CD1 strain of mice (Charles River Laboratories) instinctively swim against a current in our swimming apparatus, a behaviour that was found to be absent in other strains. In studies using mice with whole-body TNFa disruption (c57b, Taconic model #1921) or expressing an NFkB-Luc reporter (c57b, Jackson Laboratories stock #006100), a single back-cross with CD1 mice was sufficient to manifest instinctive swimming behaviour. Mice swam in water containers (30 cm diameter) with a steady water current (325 l min À 1 ) generated by a central water pump at 34°C. Swimming training began with 30-min swimming sessions (twice daily, separated by 4 h) that were increased to 90 min by adding 10 min per day. Training then continued for 6 weeks. For running exercise, mice trained on a treadmill apparatus (Harvard Apparatus) at 35 cm s À 1 with a 30°incline. Running training began with 30-min sessions (once daily) that were increased by 10 min per day to 120 min. In the first 3-4 sessions, electric stimulation (0.2 mA) was used to promote running during training; thereafter, mice were encouraged to run using a soft brush pressed against their posterior region, to reduce the potential stress of continued electric shocking. Male mice were randomly assigned to exercised or sedentary groups at 6 weeks of age. No mice were excluded from the study based on inability to train. Control animals were placed in water containers without water current for 5 min, or placed on the treadmill for 5 min at 5 cm s À 1 .
Mice treated with etanercept (Enbrel) were injected subcutaneously daily at 2.5 mg kg À 1 for the entire 6-week exercise training. Mice treated with SB203580 (AbMole Bioscience) were injected i.p. at 4 mg kg À 1 . All experimental assessments were conducted 24 h after the final exercise session. The identity of the groups (exercised versus sedentary) was concealed to the investigators. Sample sizes were . The n value shown on bar graphs and present number of mice used. Data presented as mean ± s.e.m. **Po0.05 when exercised untreated is compared with exercised þ Etn or exercised TNFa À/ À using one-way ANOVA with Dunnett'smultiple comparison test. Scale bar, 100 mm.
determined using power calculations at 95% confidence interval and 80% power using comparable data from previously published results 67 .
VO 2 consumption measurements. Oxygen consumption was determined using Oxymax oxygen monitoring (Columbus instruments, Columbus, OH). For swimming, mice were placed in a sealed container containing water pumps to generate current (34°C). Maximal O 2 consumption was determined using the swimming apparatus with two simultaneous pumps generating current on opposite ends. We have found that mice in this set-up consume the greatest amount of O 2 (11260 ± 563 ml kg À 1 h À 1 ), which was much greater than values that could be obtained using exhaustive treadmill exercise with speeds of up to 50 cm s À 1 and 30°incline (9384±643 ml kg À 1 h À 1 ).
Echocardiography. Mice were anaesthetized using 1.5% isoflurane oxygen mixture. Mice were placed on a heating pad and body temperature was maintained between 36.9 and 37.3°C for the duration of the measurements. Transthoracic M-mode echocardiographic examination was conducted using a Vevo 7700 system (VisualSonics) equipped with ultrasonic linear transducer scanning heads operating at 30 MHz. The left ventricular long axis view was used for measurement of chamber size and wall diameter. Data analysis was performed using the Visual-Sonics data analysis suite.
Electrocardiography. Surface ECG measurements were conducted on mice anaesthetized using a 1.5% isoflurane oxygen mixture with sub-dermal platinum electrodes placed in lead II arrangement. Body temperature was maintained at 36.9 to 37.3°C. To record heart rates from unanaesthetized, unrestrained animals, radiofrequency emitting ECG units (Data Sciences International) were implanted in the intraperitoneal cavity and the electrodes placed in lead II arrangement subdermally. After 1 week of recovery, mice were exercised according to the standard regimen and heart rate (HR) was recorded weekly over a 48-h period. Data were analysed using Ponemah Physiology Platform software. To assess pharmacological inhibition of autonomic nerve activity, HRs of anaesthetized mice were monitored before and after intraperitoneal injection of 2 mg kg À 1 BW of atropine sulphate and 10 mg kg À 1 BW of propranolol hydrochloride (Sigma-Aldrich) to block the parasympathetic and sympathetic branches of the autonomic nervous system.
Heart-rate variability analysis. Minimum anaesthesia deepness was strictly monitored and maintained at equivalent levels between sedentary and exercised mice using the toe pinch-pedal reflex. Surface ECG recordings were marked for R-R intervals using P3 Plus (Data Sciences International), and the raw text files of the inter-beat-intervals (IBI) were input for analysis using Kubios HRV software (University of Eastern Finland). Frequency bands of 0-0.1 Hz and 0.1-5 Hz for low and high frequency components, respectively, were determined in a series of preliminary experiments ( Supplementary Fig. 9) using atropine (2 mg kg À 1 ) and propranolol (10 mg kg À 1 ). An interpolation rate of 20 Hz, with 256 point window was used to generate frequency domain plots. The absolute power from the frequency domain plot was used for analysis. A minimum of 5 min of data were used for the analysis.

Assessment of electrical properties and arrhythmia vulnerability of hearts.
Mice were anaesthetized using 1.5% isoflurane and oxygen mixture, and a 2.0 French octapolar recording/stimulation electrophysiology catheter (CI'BER Mouse, Numed) was inserted into the right jugular vein and advanced into the right ventricle. The his bundle signal was used for consistent positioning of the catheter (see Supplementary Fig. 4). Appropriate leads were used to deliver programmed stimulation to either the right atria or right ventricle. All stimulations were delivered at a voltage magnitude of 1.5 times capture threshold, with 1 ms pulse duration. Refractory periods were determined by delivering seven pulses at 20 ms below the R-R interval followed by extrastimulation. The S2 coupling interval was initially delivered at below capture (B15 ms) and increased by 5 ms increments until capture. The coupling interval was reduced by 1-2 ms again until loss of capture. Our protocols for arrhythmia were based on previously published protocols using burst pacing with intervals less than atrial effective refractory in both exercise and sedentary mice (Supplementary Table 4). For arrhythmia induction, 27 pulses (of 2 ms duration) at 40 ms intervals were applied to the right atrium or ventricle and reduced to 20 ms intervals by 2 ms decrements, repeated three times. If arrhythmia events were not induced, 20 trains (applied every 1.5 s) of 20 pulses (2 ms duration) with an interpulse interval of 20 ms duration were used. Sustained arrhythmias were defined as reproducible episodes of rapid, chaotic and continuous atrial or ventricular activity, lasting longer than 10 s (examples in Supplementary Fig. 6).
Optical mapping and intracellular action potential recording. Heparinized mice were anaesthetized using isoflurane and killed via cervical dislocation. The thorax was opened by midsternal incision, and the heart was excised into warm (35°C) Tyrode's solution (in mmol/l): 140 NaCl, 5. For optical mapping measurements, isolated atrial preparations were stained with 10 mM voltage-sensitive dye Di-4-ANEPPS (Sigma-Aldrich) for 10 min and then superfused continuously with carbogenized Krebs solution (35°C). The flow rate, volume and temperature of the solution was kept constant throughout and inbetween the experiments. The pH of the bath solution in the perfusion vessels was monitored to ensure that the pH was between 7.35 and 7.4. To calculate conduction velocity, atria were paced at 90 ms interval from the left atrial appendage, and images were captured at 1408 frames s À 1 using a high speed camera (Photometrics, AZ, USA). Activation maps were generated to calculate conduction velocity. To induce atrial arrhythmias, the same stimulation protocols that were used in vivo using intracardiac catheter were applied. Sustained arrhythmias were defined as a reproducible episode of rapid, chaotic atrial activation lasting longer than 10 s. Images were acquired using ImagePro Plus software (Mediacy) and analysed using ImageJ or Scroll (custom software) 68 , which was modified to perform phase analyses.
For intracellular action potential recordings, pipettes were pulled from borosilicate glass using a Flaming/Brown pipette puller (Sutter Instrument Company). Pipettes were filled with 3 M KCl solution and the resistance was between 30 and 50 MO. Intracellular recordings of atrial myocytes (in tissue, not isolated) were acquired by placing the microelectrodes into the left atrial appendage. ClampFit software (Axon) was used for analysis.
Isolation of atrial myocytes and I Ca measurement. Left atrial myocytes were obtained from 6-week exercised and age-matched sedentary (CD-1, male 14 weeks old) mice. Mice were anaesthetized with 2.5% isoflurane, and hearts were removed rapidly and retrograde perfused with Ca 2 þ -free Tyrode's solution (mM): 137 NaCl, 5.4 KCl, 1.0 MgCl 2 , 10 D-Glucose, 10 HEPES, pH 7.4 at 37°C through the aorta for 10 min. After perfusing the heart with collagenase (1.0 mg ml À 1 , Worthington) and elastase (0.2 mg ml À 1 , Worthington) for 8-10 min, the left atrial appendage was dissected and stored in Kraftbruhe (KB) buffer (mM): 120 potassium glutamate, 20 KCl, 20 HEPES, 1.0 MgCl 2 , 10 D-glucose, 0.5 K-EGTA, 0.1% bovine serum albumin, pH 7.4. Single myocytes were obtained by gently triturating the left atrial appendage using a polished glass pipette with a 5 mm opening for 6-10 min. Single atrial myocytes were stored in KB buffer at 4°C until use. For recording Ca 2 þ currents (I Ca ), single left atrial myocytes were superfused with bath solution containing (mM): 140 NaCl, 4 CsCl, 1 MgCl 2 , 1.5 CaCl 2 , 10 HEPES, 10 D-glucose (pH 7.3) for 10 min and only Ca 2 þ tolerant cells were used for recording. I Ca was recorded using whole-cell patch clamp (Axopatch 200B and Clampex 8 software, Axon Instrument, CA, USA) at room temperature. The pipette resistance ranged between 4 and 6 MO when filled with internal solution, containing (in mM): 135 CsCl, 6 NaCl, 1 MgCl 2 , 3 MgATP, 10 HEPES and 10 EGTA (pH to 7.2 with CsOH). I Ca was recorded under voltage-clamp mode with 80% compensation of cell capacitance and series resistance. After membrane rupture, cell capacitance was measured by integrating capacitance transients initiated by 5 mV steps from holding potential of À 50 mV and was used to normalize the magnitude of recorded currents. For recording I Ca , cells were held at À 75 mV and stepped to À 40 mV for 50 ms to inactivate voltage-gated Na þ currents before stepping to test potentials from þ 50 mV to À 40 mV decremented by 10 mV. For recordings acetylcholine-activated K þ currents (I k,Ach ), single left atrial myocytes were superfused with bath solution containing (mM): 140 NaCl, 4 KCl, 1 MgCl 2 , 1.2 CaCl 2 , 10 HEPES, 10 D-glucose (pH 7.3) for 10 min and only Ca 2 þ tolerant cells were used for recording. To block K ATP 10 mM Glybenclamide (Sigma) was added to the bath solution. Two different concentrations of carbachol (CCh, Sigma) were delivered to cells through 0.1 mm internal-diameter tubing sitting on top of the cells. The pipette resistance ranged between 4 and 6 MO when filled with internal solution, containing (in mM): 120 potassium aspartate, 20 KCl, 1 MgCl 2 , 3 MgATP, 0.1Na 2 GTP, 10 HEPES and 10 EGTA (pH 7.2). Cells were held at À 70 mV and stepped to þ 30 mV for 120 s to ensure inactivation of all voltagegated currents before addition of CCh for additional 120 s. I K,Ach current density was measured by subtracting baseline current before addition of CCh at þ 30 mV from maximal current induced by CCh addition normalized to cell size.
where A f , A i and A s are the amplitudes of the fast, intermediate and slow kinetic components decaying with time constants of t f , t i and t s respectively. C is the amplitude of the steady-state, non-inactivating component. All data analysis was performed using ClampFit software (Axon).
Histology. Hearts were first perfused with saline containing 1% KCl followed by 4% paraformaldehyde in 0.1 M phosphate buffer. Hearts were embedded in paraffin and 5-mm thin sections were stained with Picrosirius red for visualization of collagen. Confocal laser scanning microscopy was used to image Picrosirius red stained sections (491 nm laser, 750 nm band pass emission), and imageJ software 69 was used for quantification. Perivascular and endocardial fibrosis were not included in the analysis for the quantification of % fibrosis. For quantitative measurement of macrophage infiltration, antibodies against mouse Mac-3 (1:200, BD Pharminogen, Cat. # 553322) were used with the streptavidin-biotin diaminobenzidine chromogen detection method (Vector Laboratories). Mac-3-positive cells were counted in three different sections (100 mm apart) of the left atrial appendage or left ventricle per replicate and standardized to the total tissue area of each slice. Images were acquired using Metamorph software (Molecular Devices), and data analysis was performed using ImageJ software. Mast cells were stained for using Toludine Blue at the appropriate PH. Mast cells were identified by metachromatic granules, counted and normalized to total tissue area. For quantitative measurement of connexin staining, heart sections were stained against connexin 40 (Abcam ab16585, 1:100 dilution) and 43 (Abcam ab11370, 1:100 dilution). Alexa Fluor 488 conjugated secondary antibody (Life technologies, 1:500 dilution) was used for connexin detection. Threshold method was used to determine number of pixels corresponding to connexin 40/43 staining and presented as value normalized to area of atria.
Ca 2 þ spark activity. Isolated left atrial myocytes were loaded with 1 mM fluo-3 Am at room temperature for 15 min followed by 15 min of desertification by continuously superfusing the cells with Tyrod's buffer containing 1.2 mM Ca 2 þ . After 15 min of desertification at room temperature, Ca 2 þ tolerant cells were field stimulated at 1 Hz for 30 s at room temperature and Ca 2 þ transients were recorded using a Yokagawa spinning disk confocal (491 nm excitation, 510 nm emission) with a high speed camera (Cascade 128 þ ). For measurement of spark activity, the field stimulation was stopped and images were acquired after 2 s and analysed using Sparkmaster plugin for Image J (background fl. U: 10, criteria: 3.8, no of intervals: 1) 70 .
NFjB activity assays. Transgenic 6-to 8-week old male mice expressing firefly luciferase driven by two copies of the NFkB regulatory element (Jackson Laboratories Stock #006100) were swim exercised for 2 days, two sessions per day for 90 min separated by 4 h. Two hours after the final session atria and ventricles (dissected and quickly rinsed in cold PBS) were flash frozen in liquid nitrogen. Luciferase activity was assessed in tissue homogenates according to the manufacturer's specification (Promega), and the values were standardized to total protein concentration (Bio-Rad).
Statistical analysis. Data are presented as mean ± s.e.m.. Statistical significance was determined using unpaired or paired student's t-test (two-tailed) as appropriate, a repeated measure one-way ANOVA with Sidak's multiple comparison test, or a repeated measure two-way ANOVA with Sidak's multiple comparison test. Welch's t-test was used when the variance differed between groups (assessed via F-test). The Mann-Whitney U-test was used to compare AF durations as the data were not normally distributed according to a D'Agostino& Pearson omnibus normality test. For comparison of arrhythmic events between different treatment groups, a 2 Â 2 contingency table using w 2 -test without yates correction was used. P values o0.05 were considered statistically significant.