Atrial cardiomyocyte-restricted cleavage of gasdermin D promotes atrial arrhythmogenesis

Abstract

Background and Aims

Enhanced inflammatory signalling causally contributes to atrial fibrillation (AF) development. Gasdermin D (GSDMD) is an important downstream effector of several inflammasome pathways. However, the role of GSDMD, particularly the cleaved N-terminal (NT)-GSDMD, in non-immune cells remains elusive. This study aimed to elucidate the function of NT-GSDMD in atrial cardiomyocytes (ACMs) and determine its contribution to atrial arrhythmogenesis.

Methods

Human atrial appendages were used to assess the protein levels and localization. A modified adeno-associated virus 9 was employed to establish ACM-restricted overexpression of NT-GSDMD in mice.

Results

The cleavage of GSDMD was enhanced in ACMs of AF patients. Atrial cardiomyocyte-restricted overexpression of NT-GSDMD in mice increased susceptibility to pacing-induced AF. The NT-GSDMD pore formation facilitated interleukin-1β secretion from ACMs, promoting macrophage infiltration, while up-regulating ‘endosomal sorting complexes required for transport’-mediated membrane-repair mechanisms, which prevented inflammatory cell death (pyroptosis) in ACMs. Up-regulated NT-GSDMD directly targeted mitochondria, increasing mitochondrial reactive oxygen species (ROS) generation, which triggered proarrhythmic calcium-release events. The NT-GSDMD-induced arrhythmogenesis was mitigated by the mitochondrial-specific antioxidant MitoTEMPO. A mutant NT-GSDMD lacking pore-formation capability failed to cause mitochondrial dysfunction or induce atrial arrhythmia. Genetic ablation of Gsdmd prevented spontaneous AF development in a mouse model.

Conclusions

These findings establish a unique pyroptosis-independent role of NT-GSDMD in ACMs and arrhythmogenesis, which involves ROS-driven mitochondrial dysfunction. Mitochondrial-targeted therapy, either by reducing ROS production or inhibition of GSDMD, prevents AF inducibility, positioning GSDMD as a novel therapeutic target for AF prevention.

Structured Graphical Abstract

Working model illustrating how cleaved N-terminal gasdermin D may create a pro-arrhythmic substrate for atrial fibrillation development. Cleaved N-terminal gasdermin D forms pores in the plasma membrane, facilitating cytokine release and increasing macrophage infiltration to the atria. Due to the activation of endosomal sorting complexes required for transport-mediated membrane repair, inflammatory cell death (pyroptosis) does not occur. The up-regulation of N-terminal gasdermin D increases pore formation in mitochondria, impairing their integrity and promoting oxygen reactive species production. These changes synergistically cause Ca²⁺-handling abnormalities, electrical remodelling (action potential duration shortening), and structural remodelling (interstitial fibrosis and conduction slowing), thereby creating a substrate for atrial fibrillation induction and maintenance. AF, atrial fibrillation; APD, action potential duration; DAD, delayed after depolarization; ERP, effective refractory period; ESCRT, endosomal sorting complexes required for transport; Kv1.5, ultra-rapid delayed rectifier K⁺ channel; NLRP3, NACHT, LRR, and PYD domains containing protein 3; NT-GSDMD, N-terminal gasdermin D; ROS, reactive oxygen species. The figure was created with BioRender.com.

Atrial fibrillation, Gasdermin D, ESCRT, Mitochondria

Translational perspective

The cleavage of gasdermin D (GSDMD), a common effector of multiple inflammasome pathways, is enhanced in atrial fibrillation (AF) patients. This study demonstrates that atrial cardiomyocyte-restricted overexpression of cleaved N-terminal (NT)-GSDMD increases AF susceptibility by promoting electrical and structural remodelling. Cardiomyocytes initiate endosomal sorting complexes required for transport-mediated membrane repair to avoid NT-GSDMD-induced pyroptosis, helping them remain functional despite heightened inflammasome signalling. The NT-GSDMD directly targets mitochondria, increasing reactive oxygen species (ROS) production and causing abnormal calcium handling in cardiomyocyte. Mitochondrial-targeted therapies, either by reducing ROS production or inhibiting GSDMD, prevent AF susceptibility in preclinical models. These findings suggest targeting GSDMD and mitochondrial ROS as potential AF prevention strategies.

Introduction

Atrial fibrillation (AF) is a very common cardiac arrhythmia that increases risk of stroke and mortality.^1,2 Enhanced inflammatory signalling causally contributes to AF initiation and persistence and is associated with AF recurrence after catheter ablation.^3–7 The ‘NACHT, LRR, and PYD domains containing protein 3’ (NLRP3) inflammasome plays a causal role in AF pathophysiology^8,9 and in conditions that promote AF (obesity/diabetes,^10,11 chronic kidney disease,¹² gut dysbiosis,¹³ sepsis,¹⁴ and heart failure with preserved ejection fraction¹⁵). Gasdermin D (GSDMD) is a key downstream effector of NLRP3 and is cleaved by the active caspase-1 (canonical pathway) or caspase-11 (non-canonical pathway).¹⁶ This cleavage produces an active 31-kDa N-terminal fragment (NT-GSDMD) that can induce lytic cell death known as pyroptosis in immune cells and a 22-kDa C-terminal fragment (CT-GSDMD) with an unknown function.^16,17 N-terminal GSDMD forms permeable pores in the plasma membrane, which are necessary for the secretion of interleukin-1 (IL-1) family cytokines.¹⁸ Although prior studies suggest that GSDMD activation could contribute to myocardial infarction and pathological cardiac hypertrophy by causing pyroptosis, the precise function of GSDMD in the heart, particularly in cardiomyocytes, remains poorly understood.^19,20 In this study, we explore the role of GSDMD, specifically the NT-GSDMD pores, in atrial cardiomyocytes (ACMs) and arrhythmogenesis.

Methods

Expanded methods and material are provided in the Supplementary data.

Human atrial samples

Human right atrial appendages were collected from patients (>18 years) undergoing elective open-heart surgery. Individual patient data were extracted from the digital health record files. Written informed consent was obtained from every patient prior to cardiac surgery. Patients were categorized based on their clinical chart diagnosis of persistent or long-standing persistent AF [chronic AF (cAF)] or the absence of an AF diagnosis [control (Ctl)]. We obtained atrial samples from 95 participants (47 Ctl and 48 cAF patients) to perform multiple sets of western blotting and immunohistochemistry analyses. Patients who were diagnosed with post-operative AF or had a history of autoimmune diseases and systemic inflammation were excluded. All experimental protocols involving human tissue specimens and health data have been approved by the local ethical review board of the University Duisburg-Essen, Germany (#12–5268-BO). Cardiomyocyte-enriched fractions were obtained from human atrial samples using bovine serum albumin (BSA)-gradient filtration.⁹

Animal studies

All studies involving mice were employed according to protocols approved by the Institutional Animal Care and Use Committee at Baylor College of Medicine and confirmed to the Guide for the Care and Use of Laboratory Animals published by the National Institutes of Health. The protocol number is AN-7259. Whole-body Gsdmd^−/− knockout mice (The Jackson Laboratory, #032663) and CREM-IbΔC-X transgenic (Crem Tg) mice were established previously.^21,22 Crem Tg mice, originally established on the FVB/N background,^21,22 were crossbred with C57Bl6/J wild-type (WT) mice for more than 10 generations, before being crossed with the Gsdmd^−/− mice, which also have a C57Bl6/J background. The offspring of the intercrossed Crem Tg and Gsdmd^−/− mice maintained the expected Mendelian ratio. Wild-type, Crem Tg, Crem:Gsdmd^−/−, and Gsdmd^−/− littermates with the same mixed genetic background were used in this study. Both male and female mice were included in the study.

Adeno-associated virus delivery

Adeno-associated virus 9 (AAV9) was used to express NT-GSDMD or mutant NT-GSDMD in an atrial myocyte-specific manner.²³ Six-week-old C57Bl6/J mice were injected with a single dose of AAV9-ANF-Flag, AAV9-ANF-GSDMD^NT, or AAV9-ANF-GSDMD^AAA virus via the retro-orbit route. For interventional studies, 2 weeks after the injection of AAV9-ANF-GSDMD^NT virus, mice were intraperitoneally injected with either saline or MitoTEMPO (0.1 mg/kg, Sigma, Cat#1334850-99-5) every other day for 2 weeks until the termination of the experiment. An equal number of male and female mice were included in the study.

Programmed intracardiac stimulation

Programmed intracardiac stimulation was performed to test AF inducibility.^8,10,12,24 Mice were subjected to the same atrial burst-pacing protocols three times. If AF lasting more than 1 s was evoked by burst pacing in at least two out of three attempts, the mouse was considered AF positive. The incidence of inducible AF was measured as the percentage of AF-positive mice divided by the total number of mice tested. Atrial fibrillation duration was capped at 10 min for any episodes lasting longer. The operator was blinded to the viral gene transfer status of mice.

Echocardiography

To characterize cardiac function and atrial diameter, echocardiography was performed under anaesthesia (1.5% isoflurane in 100%, 1.0–1.5 L/min) using VisualSonics Vevo F2. Body temperature was maintained between 36.0°C and 37.0°C. Standard short-axis M-mode images were recorded to evaluate ventricular function. Four-chamber B-mode images were recorded to measure left and right atrial size.

Optical mapping

Optical mapping studies were performed as previously described.^8,10,12 The whole heart was removed, cannulated via the aorta, and retrogradely perfused and superfused with Tyrode’s solution (2–5 mL/min). The voltage-sensitive dye di4-ANEPPS (Invitrogen, 0.1 mmol/L, 1 mL) was slowly injected into a drug port over 10-min period. Afterwards, blebbistatin (Sigma-Aldrich, 6.8 mmol/L, 0.1 mL) was administered via the drug port to eliminate motion artefacts. The emitted fluorescence V_m signal was long passed (>700 nm) and acquired via a MiCAM CMOS camera (SciMedia, USA) at the sampling rate of 1 kHz and pixel size of 100 μm/pixel. Right atrial pacing was performed with a PowerLab 26T stimulator (AD Instruments, Australia), and pseudo electrocardiogram (ECG) was recorded using a PowerLab bioamplifier (AD Instruments, Australia). The atrial effective refractory period (AERP) was assessed with S1-S2 pacing at a cycle length of 100 ms. A 5 ms pulse width and 5 V pulse amplitude were applied. Mapping results were analysed using ElectroMap, an established open-source software.²⁵ For 10 Hz pacing, activation maps were generated based on the depolarization midpoint value. Conduction velocity (CV) was calculated by multi-vector method for the entire right atrium. Action potential duration (APD) at 20%, 50%, 70%, and 90% repolarization was calculated for the right atrium. For each value, the average of 10 consecutive beats at 10 Hz pacing was calculated for each mouse.

RNA sequencing

Total RNA was extracted using a Qiagen kit, and a library was constructed. The libraries were prepared using the NEBNext Ultra RNA Library Prep Kit for Illumina and then sequenced on an Illumina paired-end sequencing platform with a 100 bp read length. The differentially expressed genes (DEGs) were identified based on a fold change of >0.5 and a P-value of <.05. Gene ontology analysis was performed to indicate biological functions of the DEGs.

Transmission electron microscopy

Mice atria were fixed by glutaraldehyde (2.5%) with phosphate buffer (pH 7.4) for 24 h. After rinse, the tissues were submitted in OsO₄. Atrial tissues were cut into 0.8 μm sections. Representative images were taken using a transmission electron microscope.

Flow cytometry

Mouse atria were subjected to enzymatic digestion with 1 mg/mL collagenase II and 60 U/mL of DNase I for 30 min at 37°C under agitation. The left atrial tissues were filtered through a 40 μm nylon filter. The cells were then washed and centrifuged at 1500 rpm for 5 min to collect single-cell suspensions. Cell suspensions were stained with anti-CD45, CD11b, CD64, F4/80, CD3, CD4, and CD8 in FACS buffer (phosphate-buffered saline with 0.5% BSA) at 4°C for 30 min. A live/death stain was used as a cell viability marker. Data were collected with LSRII (BD Biosciences) and analysed with FlowJo software.

Statistics

All data were presented as mean ± SEM. Statistical analyses were performed using GraphPad Prism. Normality was evaluated with Anderson–Darling test or Shapiro–Wilk test. Two-tailed Student’s t-tests were used to compare data between two groups maintaining normal distributions. Mann–Whitney tests were used for data where normality could not be assumed. Fisher's exact test was used to analyse categorical data. Nested t-tests and RStudio were used for data of biological replicates with repeated measurements. In MitoTracker and co-staining experiments, where each sample could contribute multiple data points, data were delineated based on (i) the individual sample ID and (ii) the ID of HL-1 cells within each sample. Sample sizes were denoted as n/N, where n represents HL-1 cells and N represents samples. Multilevel mixed-effect models were applied to compare groups. The random effect incorporated in the model was the intercept, accounting for non-independent measurements in multiple HL-1 cells from individual samples. Multilevel models were implemented in RStudio using lme4, and P-values were derived employing the Kenward–Roger approximation. A P-value of <.05 was considered statistically significant. One-way analysis of variance (ANOVA) was used for multiple comparisons of normally distributed data, with P-values corrected for multiple testing using Bonferroni adjustment. For non-normally distributed data, the Kruskal–Wallis test with Dunn’s correction was applied in place of ANOVA. A P-value of <.025 was considered statistically significant.

Results

Increased levels of N-terminal gasdermin D in atria create a proarrhythmic substrate

Western blotting using right atrial appendage and ACMs from sinus rhythm Ctl and long-standing persistent/cAF patients revealed significantly up-regulated protein levels of NT-GSDMD in whole atria and ACMs of cAF patients (Figure 1A and B; Supplementary data online, Tables S1 and S2). To elucidate the role of NT-GSDMD in ACMs, we employed the previously established AAV9 vector to overexpress NT-GSDMD (AAV9-ANF-GD^NT, 1 × 10¹² GC/mouse) in C57Bl6/J mice in an ACM-specific manner (aGD^NT mice; Figure 1C).^23,24 Control mice received the AAV9 virus expressing Flag (AAV9-ANF-Flag, 1 × 10¹² GC/mouse, aFlag mice). Four weeks after virus injection, western blot and qRT-PCR confirmed the ACM-selective increase of NT-GSDMD expression (Figure 1D; Supplementary data online, Figure S1A and B). To assess vulnerability to AF, programmed intracardiac stimulation was performed. While ECG parameters were comparable between aFlag and aGD^NT mice (see Supplementary data online, Table S3), upon rapid atrial pacing, the incidence of pacing-induced AF was significantly higher (75% vs. 20%; Figure 1E) and the duration of inducible AF episodes was longer in aGD^NT compared with aFlag mice (Figure 1E), indicating the evolution of an AF-maintaining substrate. Echocardiography, Doppler, and histological analyses demonstrated that the ventricular function and structure, as well as left and right atrial sizes, were comparable between aFlag and aGD^NT mice (see Supplementary data online, Figure S1C–F and Table S4).

Figure 1

Atrial cardiomyocyte-specific overexpression of N-terminal gasdermin D enhances atrial fibrillation susceptibility. (A) Western blots of the full-length and N-terminal gasdermin D protein in atrial tissue of control and chronic atrial fibrillation patients. n = 11 per group. (B) Western blots of N-terminal gasdermin D protein in atrial cardiomyocytes from control and chronic atrial fibrillation patients. n = 8 per group. (C) Schematic representation of the development of aFlag and aGD^NT mouse models using adeno-associated virus 9 viruses. (D) Western blots and quantification of N-terminal gasdermin D protein levels in the atria of aFlag and aGD^NT mice. n = 5 per group. (E) Left: Representative simultaneous recordings of surface electrocardiogram (ECG) and intracardiac electrograms in mice and the incidence of pacing-induced atrial fibrillation. Right: Duration of inducible atrial fibrillation. (F) Quantification of atrial effective refractory period at 10 Hz pacing. n = 8 per group. (G) Left: representative activation maps. Right: Quantification of conduction velocity at 10Hz pacing in aFlag and aGD^NT mice. n = 8 per group. (H) Immunostaining and quantification of Cx43, cell membrane wheat germ agglutinin (WGA), and nuclei 4′,6-diamidino-2-phenylindole (DAPI) in mouse atria. Scale bars, 50 or 25 μm. (I) Picrosirius red staining of whole hearts in aFlag and aGD^NT mice and quantification of fibrotic area (pink colour) in left and right atria. Scale bars, 1 mm or 100 μm. n = 5 per group. The bar graph data are as mean ± SEM with individual values. P-values are determined with two-tailed Student’s t-test in A, B, D, G, H, and I, Fisher’s exact test in E (left), and Mann–Whitney test in E (right). AAV9, adeno-associated virus 9; ACM, atrial cardiomyocyte; AF, atrial fibrillation; cAF, chronic atrial fibrillation; Ctl, control; CV, conduction velocity; FL-GSDMD, full-length gasdermin D; NT-GSDMD, N-terminal gasdermin D; RA, right atrium

Optical mapping studies on perfused hearts revealed that both AERP and APD at 70% and 90% of repolarization were abbreviated in aGD^NT mice (Figure 1F; Supplementary data online, Figure S2A). These findings were consistent with the up-regulated protein levels of ultra-rapid delayed-rectified K⁺ channels (Kv1.5) and voltage-gated Na⁺-channels (Nav1.5), while L-type Ca²⁺ channel (Cav1.2) proteins remained unaltered in the atria (see Supplementary data online, Figure S2B). Conduction velocity was slowed, and the CV-regulating gap junction protein connexin (Cx)43 was decreased in the atria of aGD^NT mice (Figure 1G; Supplementary data online, Figure S2D), while the Cx40 levels remained unchanged. Ventricular Kv1.5, Nav1.5, Cav1.2, Cx43, and Cx40 protein levels were comparable in aGD^NT and aFlag mice (see Supplementary data online, Figure S3A–C). Immunofluorescence analysis further revealed an increased lateralization of Cx43 in the atria of aGD^NT mice (Figure 1H), along with a more condensed distribution of Nav1.5 at the cardiomyocyte surface (see Supplementary data online, Figure S2C). Finally, Picrosirius red staining and western blot analyses detected increased levels of collagen deposition (fibrosis) and pro-fibrotic proteins in the atria of aGD^NT mouse (Figure 1I; Supplementary data online, Figure S2E). Thus, the ACM-selective up-regulation of NT-GSDMD, as seen in cAF patients, promotes the formation of a pro-arrhythmic AF-accommodating substrate, characterized by electrical (ERP/APD abbreviation, ion-channel dysfunction) and structural (fibrosis and connexin) remodelling.

We then assessed whether the elevation of NT-GSDMD in ACMs drives atrial NLRP3-inflammasome activation. While protein levels of NLRP3, pro-IL-1β, and caspase-11 were unchanged, the protein levels of ASC, pro-caspase-1, cleaved caspase-1 (p20), and mature IL-1β were all up-regulated in the atria (see Supplementary data online, Figure S2F–H), but not in the ventricles (see Supplementary data online, Figure S3D–F) of aGD^NT mice. The levels of the IL-1β and IL-6 were higher in sera of aGD^NT mice (Figure 2A), suggesting potential activation of immune cells. Indeed, flow cytometry and immunofluorescence confirmed that CD11b⁺ myeloid cells and F4/80⁺ macrophages were expanded in the atria of aGD^NT mice compared with aFlag mice (Figure 2B–D). Given the recent report that recruited macrophages facilitate fibrosis as a substrate for AF development,²⁶ our results suggest that the up-regulation of NT-GSDMD in ACMs might be a mechanism for macrophage recruitment to the atria, which accelerates the evolution of AF-promoting atrial remodelling.

Figure 2

Activation of atrial inflammation in aGD^NT mice. (A) Quantification of serum interleukin-1 beta and interleukin-6 levels in aFlag and aGD^NT mice. n = 9 per group. (B) Representative flow cytometry and quantification of CD3⁺ T cells and CD11b⁺ myeloid cells in the atria of aFlag and aGD^NT mice. n = 8 per group. (C) Representative flow cytometry and quantification of F4/80⁺ macrophages in the atria of aFlag and aGD^NT mice. n = 8 per group. (D) Representative immunostaining and quantification of infiltrated macrophages (F4/80), Flag and nuclei DAPI in the atria of aFlag and aGD^NT mice. Scale bars, 50 or 25 μm. All bar graph data are presented as mean ± SEM with individual values. P-values were determined with unpaired two-tailed Student’s t-test in A–D. IL-1β, interleukin-1 beta; IL-6, interleukin-6

Membrane pore repair is enhanced in aGD^NT mice

It is known that NT-GSDMD forms membrane pores that facilitate cytokine release and lead to pyroptosis.¹⁶ To determine whether the NT-GSDMD-induced pyroptosis occurs, we measured the levels of propidium iodide (PI) intake and lactate dehydrogenase (LDH).²⁷ Although the number of PI-positive nuclei was increased in the atria of aGD^NT mice (Figure 3A), LDH levels—the common marker of pyroptosis—were comparable between aFlag and aGD^NT mice, indicating the absence of pyroptosis (see Supplementary data online, Figure S4A). Western blotting and immunofluorescence demonstrated increased levels of sarcolemmal NT-GSDMD in ACMs and the plasma membrane fraction of atrial lysates in the aGD^NT group (Figure 3B and C). These results suggest that, although NT-GSDMD protein forms membrane pores in ACMs, leading to increased PI intake, it does not cause pyroptosis. To determine the mechanism protecting ACMs from pyroptosis, we performed bulk RNA sequencing on atrial tissue from aFlag and aGD^NT mice. We identified 95 DEGs between the groups (see Supplementary data online, Table S5). Gene ontology analysis based on the DEGs indicated that the up-regulated gene sets in aGD^NT atria were related to the activity of transmembrane transporters (see Supplementary data online, Figure S4B and C). Therefore, we determined whether the ‘endosomal sorting complexes required for transport (ESCRT)’ machinery, responsible for membrane repair,^28–30 is up-regulated in ACMs of aGD^NT mice. We found that the protein levels of ESCRT components CHMP3, CHMP4B, and VPS4A were elevated in the atrial tissues of aGD^NT mice (Figure 3D). Moreover, in the atria of aGD^NT mice, CHMP3 was enriched at the plasmalemma of ACMs (Figure 3E). Similarly, in the atria of cAF patients, the protein levels of ESCRT components were significantly up-regulated and preferentially localized at the plasmalemma of ACMs (Figure 3F and G; Supplementary data online, Tables S6 and S7). These findings suggest that ACMs can counteract the increase of NT-GSDMD by up-regulating membrane-repair mechanisms, which prevents ACM pyroptosis.

Figure 3

Enhanced endosomal sorting complexes required for transport membrane repair in aGD^NT mice. (A) Left: Representative staining of propidium iodide, Flag, and nuclei DAPI. Right: Quantification of propidium iodide-positive nuclei in mouse atria. n = 4 per group. Scale bars, 50 or 25 μm. (B) Western blots and quantification of cytosol and membrane-bound full-length gasdermin D and N-terminal gasdermin D proteins in mouse atria. n = 4 per group. (C) Left: Representative staining of cell membrane (Cavolien3), Flag, and nuclei (4′,6-diamidino-2-phenylindole) in isolated atrial cardiomyocytes. Right: Pearson analysis of co-localization. n = 5–6 per group. (D) Western blots and quantification of CHMP3, CHMP4B, and VPS4A proteins in mouse atria. n = 4 or 5 per group. (E) Representative immunofluorescence staining and quantification of CHMP3 and cell membrane WGA co-localization in mouse atria. Scale bars, 50 or 25 μm. n = 5 per group. (F) Western blots and quantification of CHMP3, CHMP4B, and VPS4A proteins in the right atria of control and chronic AF patients. n = 15 or 16 per group. (G) Representative immunofluorescence staining and quantification of co-localization of cell membrane (wheat germ agglutinin) with CHMP3 or gasdermin D in human atria. Scale bars, 100 or 25 μm. n = 9 or 14 per group. All bar graph data are presented as mean ± SEM with individual values. P-values were determined with unpaired two-tailed Student’s t-test in A–G. cAF, chronic atrial fibrillation; Ctl, control; Cyto, cytosol; FL-GSDMD, full-length gasdermin D; Mem, membrane; NT-GSDMD, N-terminal gasdermin D

Pore-forming defective N-terminal gasdermin D prevents arrhythmogenicity

Several basic residues of NT-GSDMD, including Lys146, Arg152, and Arg154, are pivotal for the assembly of NT-GSDMD into membrane pores.^31,32 To directly ascertain the role of NT-GSDMD pore formation in AF development, we mutated the Lys146, Arg152, and Arg154 residues of NT-GSDMD to alanine (K146A/R152A/R154A, AAA) and injected C57Bl6/J mice with the ACM-specific, pore-forming defective mutant AAV9-ANF-GD^AAA virus (1 × 10¹² GC/mouse, aGD^AAA mice) (Figure 4A). Four weeks post-injection, western blots confirmed comparable transduction efficiency between aGD^NT and aGD^AAA mice (Figure 4B). Echocardiography and Doppler analyses showed comparable cardiac structure and function in mutant aGD^AAA and aGD^NT mice (see Supplementary data online, Figure S5A–D and Table S8). However, the incidence and duration of pacing-induced AF were significantly decreased in aGD^AAA mice compared with aGD^NT mice (Figure 4C; Supplementary data online, Table S9). Subsequently, we assessed the underlying mechanism of AF protection in aGD^AAA mice. Optical mapping showed that the shortening of AERP and APD was reversed in aGD^AAA mice vs. aGD^NT mice (Figure 4D; Supplementary data online, Figure S5E), along with a normalization of Nav1.5 and Kv1.5 protein levels (see Supplementary data online, Figure S5G). Additionally, CV and Cx43 protein levels were increased, the lateralization of Cx43 in ACMs was reduced, and the preferential plasmalemmal localization of Nav1.5 channel was ameliorated in aGD^AAA mice compared with aGD^NT mice (Figure 4E and F; Supplementary data online, Figure S5F and H). Fibrosis and levels of pro-fibrotic proteins were reduced in the atria of aGD^AAA compared with aGD^NT mice (Figure 4G; Supplementary data online, Figure S5I). These changes were accompanied by a reduction in the protein levels of active p20 and mature IL-1β in the atria and a decrease in the levels IL-1β and IL-6 cytokines in the sera of aGD^AAA mice (see Supplementary data online, Figure S5J, K, and M). Consistently, flow cytometry and immunofluorescence revealed a reduced atrial infiltration of macrophages in aGD^AAA vs. aGD^NT mice (see Supplementary data online, Figure S5L; Figure 4H). To verify whether the AAA mutation decreased NT-GSDMD pore formation, we tested the levels of PI intake and the sarcolemmal NT-GSDMD and found that plasma pore formation was lower in the atria of aGD^AAA mice (see Supplementary data online, Figure S6A and B). Notably, the levels of ESCRT proteins (CHMP3, CHMP4B, and VPS4A) and membrane-bound CHMP3 were also reduced in the aGD^AAA group compared with the aGD^NT group (see Supplementary data online, Figure S6C and D). Lactate dehydrogenase levels were comparable between aGD^NT and aGD^AAA mice (see Supplementary data online, Figure S6E). Overall, these results not only suggest that pore formation is essential for NT-GSDMD-induced arrhythmogenesis but also reaffirm the earlier observation that enhanced membrane repair is a compensatory response to the increased pore formation associated with NT-GSDMD up-regulation in ACMs.

Figure 4

Pore-forming deficient aGD^AAA mice do not show increased susceptibility to atrial fibrillation. (A) Schematic representation of developing aGD^AAA mice using adeno-associated virus 9 viruses. (B) Western blots and quantification of Flag and N-terminal gasdermin D proteins in mouse atria. n = 4 or 5 per group. (C) Left: Representative simultaneous recordings of surface ECG and intracardiac electrograms in mice and atrial fibrillation incidence. Right: Duration of inducible atrial fibrillation. (D) Quantification of atrial effective refractory period at 10 Hz pacing. n = 8 per group. (E) Left: Representative activation maps. Right: Quantification of CV at 10 Hz pacing in aGD^NT and aGD^AAA mice. n = 8 or 10 per group. (F) Representative immunofluorescence staining and quantification of Cx43, cell membrane WGA, and nuclei DAPI in mouse atria. Scale bars, 50 or 25 μm. (G) Representative Picrosirius red staining of whole hearts in aGD^NT and aGD^AAA mice and quantification of fibrotic area in the left and right atria. Scale bars, 1 mm. n = 5 per group. (H) Representative immunofluorescence staining and quantification of infiltrated macrophages (F4/80), Flag, and nuclei DAPI in the left atria from aGD^NT and aGD^AAA mice. Scale bars, 50 μm. n = 4 per group. All bar graph data are presented as mean ± SEM with individual values. P-values were determined with unpaired two-tailed Student’s t-test in B and D–H, Fisher’s exact test in C (left), and Mann–Whitney test in C (right). AERP, atrial effective refractory period; AF, atrial fibrillation; AAV9, adeno-associated virus 9; NT-GSDMD, N-terminal gasdermin D

Up-regulated N-terminal gasdermin D causes mitochondrial dysfunction

Since pore formation driven by NT-GSDMD is not restricted to the plasmalemma, we hypothesized that NT-GSDMD might localize to the mitochondria of cardiomyocytes,^33–35 particularly given that several pathways related to the activities of mitochondrial electron transport chain were enriched in the RNA-sequencing data set of aGD^NT mice (see Supplementary data online, Figure S4C). Western blotting and immunofluorescence showed that NT-GSDMD protein was enriched in mitochondrial fraction and more strongly co-localized with the mitochondrial marker cytochrome c oxidase subunit IV in the atria of cAF compared with Ctl patients (Figure 5A and B; Supplementary data online, Tables S10 and S11). Consistently, NT-GSDMD protein was more abundant in the mitochondrial fraction of atrial lysates and showed a higher degree of co-localization with the mitochondrial marker ATP synthase F1 subunit alpha (ATP5A1) in the atria of aGD^NT vs. aFlag mice (Figure 5C and D). Conversely, the AAA mutation in NT-GSDMD reversed the mitochondrial enrichment of NT-GSDMD protein and its co-localization with ATP5A1 in aGD^AAA mice compared with aGD^NT mice (Figure 5C and D). Transmission electron microscopy revealed an increased number of swollen or injured mitochondria in the atrial tissue of aGD^NT mice compared with either aFlag or aGD^AAA mice (Figure 5E). We also noted a moderately increased cytosolic double-stranded DNA (dsDNA) level, whereas protein levels of mitochondrial Complexes I, II, and IV were lower in the atria of aGD^NT compared with aFlag mice (see Supplementary data online, Figure S7A;  Figure 5F). All these changes were attenuated in aGD^AAA mice. Additionally, to assess the direct effects of NT-GSDMD overexpression on mitochondrial function, we transfected HEK293T cells with GFP-tagged Flag, GSDMD^NT, or GSDMD^NT−AAA, respectively (see Supplementary data online, Figure S7B). Seahorse assay revealed that the oxygen consumption rate was lower in HEK293T cells transfected with GSDMD^NT compared with the Flag-transfected HEK293T cells (see Supplementary data online, Figure S7C). These data suggest that the increased mitochondrial localization of NT-GSDMD can promote mitochondrial dysfunction, which may, in turn, contribute to its arrhythmogenic effects.

Figure 5

Up-regulated N-terminal gasdermin D localizes at mitochondria and causes mitochondrial dysfunction. (A) Western blots of full-length gasdermin D and N-terminal gasdermin D proteins in cytosol and mitochondrial fractions of atria from control and chronic atrial fibrillation patients. n = 7 per group. (B) Representative immunofluorescence staining and quantification of cytochrome c oxidase subunit IV, gasdermin D, cell membrane WGA, and nuclei DAPI in the atria of control and chronic atrial fibrillation patients. Scale bars, 100 or 25 μm. n = 9 per group. (C) Western blots and quantification of N-terminal gasdermin D protein in cytosol and mitochondrial fractions of mouse atria. n = 4 per group. (D) Representative immunofluorescence staining and quantification of co-localization of N-terminal gasdermin D (Flag) and mitochondrial marker (ATP5A1) in mouse atria. n = 4 per group. (E) Representative transmission electron microscopy images and quantification of injured mitochondria in mouse atria. n = 3 per group. (F) Western blots and quantification of mitochondrial complex protein in mouse atria. n = 4 per group. All bar graph data are presented as mean ± SEM with individual values. P-values were determined with unpaired two-tailed Student’s t-test in A and B, one-way analysis of variance multiple with Bonferroni correction for multiple comparisons in C–F. cAF, chronic atrial fibrillation; Ctl, control; CV, conduction velocity; FL-GSDMD, full-length gasdermin D; NT-GSDMD, N-terminal gasdermin D

Dysfunctional mitochondria increase the production of reactive oxygen species (ROS), which can cause pro-arrhythmic Ca²⁺-handling abnormalities in cardiomyocytes.^36–38 Using the mitochondrial ROS (mtROS)-specific indicator, mitoSOX, we detected increased levels of mtROS in ACMs of aGD^NT mice compared with either aFlag or aGD^AAA mice (Figure 6A). Similarly, both mitoSOX signal and the levels of mitochondrial fission were increased in the lipopolysaccharide-treated vs. saline (Ctl)-treated HL-1 cells, which were prevented by Gsdmd knockout in HL-1 cells (see Supplementary data online, Figure S7E–G). Ca²⁺ imaging in ACMs revealed that mitochondrial dysfunction in aGD^NT mice was associated with a greater incidence of spontaneous Ca²⁺ waves compared with aFlag or aGD^AAA mice (Figure 6B).

Figure 6

MitoTEMPO (TEMPO) reduced atrial fibrillation susceptibility in aGD^NT mice. (A) Representative mtROS (MitoSOX) and nuclei DAPI staining in isolated atrial cardiomyocytes and quantification of MitoSOX signal in atrial cardiomyocytes from aFlag, aGD^NT, and aGD^AAA mice. n = 24 or 25 cells per group. (B) Representative traces of Ca²⁺ transients induced by 1 Hz pacing, followed by baseline recording and the application of 10 mmol/L caffeine. The incidence of spontaneous Ca²⁺ waves (indicated by red arrows) in atrial cardiomyocytes of aFlag, aGD^NT, and aGD^AAA mice. (C) Timeline of TEMPO injection in aGD^NT mice. (D) Representative simultaneous recordings of surface ECG and intracardiac electrograms in aGD^NT mice treated with saline or TEMPO, and the incidence of pacing-induced atrial fibrillation. (E) Duration of inducible atrial fibrillation. All bar graph data are presented as mean ± SEM with individual values. P-values were determined with one-way analysis of variance with Bonferroni correction for multiple comparisons in A, Fisher’s exact test in B and D, and two-tailed Mann–Whitney test in E. AAV9, adeno-associated virus 9; SCaWs, spontaneous Ca²⁺ waves

MitoTEMPO therapy attenuates N-terminal gasdermin D-induced mitochondrial dysfunction and atrial arrhythmogenesis

To assess whether the increased mtROS levels contribute to atrial arrhythmogenesis in aGD^NT mice, we randomized the aGD^NT mice to receive treatment with the mitochondrial-specific antioxidant MitoTEMPO (TEMPO)^39,40 or saline for 2 weeks (Figure 6C). While TEMPO did not affect the protein levels of NT-GSDMD or the cardiac function and structure in aGD^NT mice (see Supplementary data online, Figure S8A–C and Table S12), AF inducibility and duration were reduced in the TEMPO-treated aGD^NT mice compared with saline-treated aGD^NT mice (Figure 6D and E; Supplementary data online, Table S13). We also observed a moderate decrease in cytosolic dsDNA levels, whereas the protein levels of mitochondrial Complexes I, II, and IV remained unchanged in TEMPO-treated aGD^NT mice (see Supplementary data online, Figure S8D and E). These data suggest that the formation of NT-GSDMD membrane pores in mitochondria increases mtROS production, potentially causing Ca²-handling abnormalities that may lead to ectopic (triggered) activity and AF promotion.

Genetic ablation of gasdermin D ameliorates the development of spontaneous atrial fibrillation

Given that NT-GSDMD increases susceptibility to inducible AF, we tested whether inhibition of GSDMD could prevent the development of spontaneous AF using a well-established mouse model—Crem Tg (Crem) mice, which mimic the phenotype of cAF patients.^21,22,41 Crem mice also exhibit enhanced inflammasome activation and mitochondrial dysfunction.^13,41–43 We crossbred Crem mice with Gsdmd^−/− mice. Western blots showed that the protein levels of GSDMD were higher in atria of Crem mice than in either WT or Crem:Gsdmd^−/− mice (Figure 7A). The levels of sarcolemmal NT-GSDMD and PI intake were higher in the atria of Crem mice compared with WT and were reduced significantly in the atria of Crem:Gsdmd^−/− mice (see Supplementary data online, Figure S9A and B). The protein levels of CHMP3, CHMP4B, and VPS4A were also higher in Crem mice compared with WT and were normalized in Crem:Gsdmd^−/− mice (see Supplementary data online, Figure S9C). Western blotting and immunofluorescence revealed increased protein levels of mitochondrial NT-GSDMD and a stronger co-localization with the mitochondrial marker ATP5A1 in Crem mice compared with WT mice; these were attenuated in Crem:Gsdmd^−/− mice (Figure 7B; Supplementary data online, Figure S9D). The protein levels of mitochondrial Complexes I and IV were lower in Crem mice vs. WT mice, while other complex proteins remained unchanged (see Supplementary data online, Figure S9F). Cytosolic dsDNA levels were also elevated in Crem mice (see Supplementary data online, Figure S9E). In contrast, the protein levels of mitochondrial Complexes I and IV and cytosolic dsDNA were normalized in the atria of Crem:Gsdmd^−/− mice (see Supplementary data online, Figure S9E and F). Previous work established that Crem mice develop spontaneous and long-lasting persistent AF after 5 months of age. To assess the contribution of GSDMD to the development of spontaneous AF, we recorded surface ECG in age-matched 6-month-old mice. The incidence of spontaneous AF was significantly higher, and the duration of spontaneous AF episodes was longer in Crem mice than in WT and Crem:Gsdmd^−/− mice (Figure 7C and D; Supplementary data online, Table S14). Although echocardiography showed that left ventricular ejection fraction was mildly reduced in Crem mice, the diameters of left ventricles were comparable among four groups of mice at 6 months of age (see Supplementary data online, Table S15). Consistent with previous reports,^22,41 the left atrial area was enlarged in Crem compared with WT mice, which was normalized in Crem:Gsdmd^−/− mice (see Supplementary data online, Figure S10A and B). Moreover, the protein levels of caspase-1/p20, ASC, and IL-1β were reduced in the atria of Crem:Gsdmd^−/− vs. Crem mice (see Supplementary data online, Figure S10C). Immunofluorescence demonstrated fewer macrophages infiltrating the atria of Crem:Gsdmd^−/− mice (see Supplementary data online, Figure S10D). Fibrosis and levels of pro-fibrotic proteins were normalized in the atria of Crem:Gsdmd^−/− compared with Crem mice (see Supplementary data online, Figure S10E and F). These findings suggest that GSDMD inhibition prevents the development of spontaneous AF by ameliorating mitochondrial dysfunction and the development of an arrhythmogenic AF substrate.

Figure 7

Genetic ablation of Gsdmd prevents spontaneous atrial fibrillation by improving mitochondrial integrity and function in mice. (A) Western blots and quantification of gasdermin D protein in the atria of Crem transgenic (Crem) mice with and without the Gsdmd knockout. n = 5 per group. (B) Representative immunofluorescence staining images and the Pearson score for co-localization of gasdermin D and mitochondria (ATP5A1) in mouse atria. Scale bars, 50 or 25 μm. n = 3 per group. (C) Left: Representative recordings of surface ECG in wild-type, Crem, Gsdmd^−/−, and Crem:Gsdmd^−/− mice. Right: Incidence of spontaneous atrial fibrillation in four groups of mice. (D) Quantification of spontaneous atrial fibrillation episodes. All bar graph data are presented as mean ± SEM with individual values. P-values were determined with one-way analysis of variance with Bonferroni correction for multiple comparisons in A and B, two-tailed Fisher’s exact test in C, and two-tailed Mann–Whitney test in D. FL-GSDMD, full-length gasdermin D; NT-GSDMD, N-terminal gasdermin D; WT, wild type

Discussion

The causal role of NLRP3-inflammasome pathway in AF development has been established in recent years, and NLRP3/IL-1β pathway has been demonstrated to mediate between various risk factors, such as obesity,¹⁰ chronic kidney disease,¹² aging,¹³ clonal haematopoiesis,⁴⁴ and AF pathogenesis. Although the involvement of downstream effectors of the inflammasome, such as IL-1β, in AF has been established, the role of GSDMD and particularly the cleaved NT-GSDMD in AF pathogenesis has not been reported before. Complementing the animal experiments with human atrial samples, our study demonstrates a direct role of NT-GSDMD pore formation in ACM dysfunction and atrial arrhythmogenesis. Utilizing the ACM-selective overexpression of NT-GSDMD via AAV9 virus, we show that while NT-GSDMD pores can localize to the plasmalemma of ACMs, facilitating cytokine release, the majority of ACMs do not undergo pyroptosis, a phenomenon common in immune cells or under ischaemic injury.^19,45,46 Interestingly, our data also uncover that ACMs can engage the ESCRT-mediated membrane repair mechanism to counteract the up-regulation of NT-GSDMD at the sarcolemma. This previously unrecognized mechanism could explain why the majority of ACMs in diseased atria maintain their contractile and electrophysiological properties and remain viable even in the presence of persistently increased ‘NLRP3 -GSDMD’ inflammatory signalling.^8,9

Gasdermin D is a key downstream effector of several inflammasome pathways and can be activated via both canonical and non-canonical pathways. Emerging evidence indicates that NT-GSDMD is essential for pore formation and cytokine secretion in living immune cells.^47–49 Our findings also suggest the existence of an NT-GSDMD-driven feedforward loop involving the NLRP3 inflammasome pathway that increases IL-1β release from ACMs. Consistent with a recent study revealing the involvement of recruited macrophages in atrial enlargement and atrial fibrosis,²⁶ our results show that ACM-restricted overexpression of NT-GSDMD can increase cytokine release. These cytokines can attract macrophage infiltration and activate atrial fibroblasts, subsequently augmenting collagen deposition, potentially causing heterogeneous and slow atrial conduction. The increased IL-6 may also directly affect the expression of Cx43 in the atria of aGD^NT mice, as suggested by previous reports showing that Cx43 levels negatively correlate with circulating IL-6 levels in patients at risk for atrioventricular block or AF and that direct application of IL-6 on HL-1 cells suppresses the Cx43 expression.^50,51 Overall, atrial fibrosis and Cx43 remodelling may slow atrial conduction thereby promoting the evolution of a proarrhythmic substrate for AF development.

For the first time, we demonstrate that up-regulated NT-GSDMD can target the mitochondria of ACMs in both human and mouse samples. The formation of NT-GSDMD membrane pores, likely at the outer membrane, can compromise mitochondrial stability.^52–54 This is supported by evidence showing deteriorated mitochondrial morphology, reduced mitochondrial respiratory chain complexes, and an increased level of dsDNA, likely due to mtDNA leakage into the cytoplasm, in aGD^NT mice. Consequently, it is not surprising that such mitochondrial distress enhances ROS production in ACMs of aGD^NT mice, leading to aberrant pro-arrhythmic Ca²⁺-release events in ACMs and AF promotion.⁵⁵ These events can be prevented by selectively targeting mitochondria-derived ROS with MitoTEMPO in aGD^NT mice. We have previously shown that constitutive activation of NLRP3 in a cardiomyocyte-specific knockin mouse model can promote aberrant Ca²⁺-release events via Type 2 ryanodine receptor.⁸ The findings from the current study provide an explanation for the augmented Ca²⁺ release from the sarcoplasmic reticulum following ‘NLRP3-GSDMD’ activation. Utilizing the spontaneous AF mouse model, which exhibits inflammasome activation and mitochondrial dysfunction,^22,43 we further establish that targeting GSDMD can improve mitochondrial function, thereby attenuating several arrhythmogenic features associated with mitochondrial dysfunction.

Our study has limitations. While we did not observe differences between the right atria and left atria of mice in terms of both atrial dimensions (see Supplementary data online, Figures S1D and S5B) and protein expression patterns (see Supplementary data online, Figure S11), for accessibility reasons, we could assess the GSDMD and ESCRT pathway in human right atrial samples. It remains to be determined whether the GSDMD and ESCRT pathways are similarly remodelled in the left atria of AF patients. Due to the small size of the obtained human right atrial appendage tissue, multiple experiments cannot be performed systematically on the same specimen. To account for this variability, we have provided the patient characteristics corresponding to each figure panel involving human samples in separate supplementary tables to allow the readers to see the specific clinical context of each analysis. Similarly, due to the very small mouse atria and technical limitations, the number of observations between the experimental groups varied, and we could not isolate enough mitochondria from ACMs of aGD^NT mice to perform classic mitochondrial functional assays. The Seahorse assay on primarily isolated ACMs also presents challenges, and as a result, we only performed this assay on a heterologous system. Direct evidence confirming that NT-GSDMD pores are located at the outer membrane of mitochondria has yet to be established. Super-resolution microscopy imaging may be employed in future studies to assess the precise mitochondrial location of NT-GSDMD. While the levels of circulating IL-1β and IL-6 cytokines were elevated in aGD^NT mice, C-reactive protein levels and white blood cell counts were comparable between sinus rhythm Ctl and cAF patients. This suggests that the enhanced cleavage of GSDMD in AF patients may be independent of systemic inflammation. However, since the levels of other circulating proinflammatory cytokines, such as IL-1β and IL-6, and the composition of immune cells within the atria are not available, we cannot rule out the possibility that differential systemic inflammation may exist in these patients, potentially biasing the results. Clearly, the relationship between GSDMD cleavage and the regional inflammatory status in AF patients warrants further in-depth investigation. Since emerging evidence points to an association between chronic immune diseases and risk for AF,⁵⁶ future studies should determine the precise role of NT-GSDMD in autoimmune disease-related atrial arrhythmia. Given the extensive number of statistical tests in this study, the risk of Type I errors is elevated. Some results show borderline statistical significance and should, therefore, be interpreted with caution.

Conclusion

Our study reveals a novel role of NT-GSDMD in the pathogenesis of AF through multifaceted mechanisms, as illustrated in the Structured Graphical Abstract. Our findings support a unique pyroptosis-independent role of NT-GSDMD in ACMs and arrhythmogenesis, involving mitochondrial dysfunction. Mitochondrial-targeted therapy, either by reducing ROS production or inhibiting GSDMD, prevents AF inducibility. This positions GSDMD targeting as a novel therapeutic option for AF.

Acknowledgements

The authors thank the team at Essen-Huttrop Cardiac Surgery for providing the human atrial appendages and the Imaging Center Essen Service Core at University Duisburg-Essen, Germany for their assistance. Special thanks are extended to Henrike Frey, Simone Olesch, Dennis Hofmann, Bettina Mausa, Ramona Löcker, Annette Kötting-Dorsch, and Melanie Kaufmann from the Institute of Pharmacology, University Duisburg-Essen, for their excellent technical support. The authors thank Prof. Eric Metzen at the Institute of Physiology, University Duisburg-Essen and Dr. Xander H.T. Wehrens at Baylor College of Medicine, USA for providing the pMD2.G plasmid and the AAV-ANF construct, respectively. The authors also thank M. ‘Sayeed’ Sayeeduddin, Shahida Salar, Zahida Sayeeduddin, and Joel M. Sederstrom at Baylor College of Medicine for their expert assistance in histology and flow cytometry. The authors also thank Dr. Frank U. Müller at University of Münster, Germany for providing the Crem transgenic mice.

Supplementary data

Supplementary data are available at European Heart Journal online.

Declarations

Disclosure of Interest

All authors declare no conflicts of interest for this contribution.

Data Availability

All data are available in the main text or the Supplementary data.

Funding

This study is supported by grants from the National Institutes of Health (NIH) (R01HL164838, R01HL136389, and R01HL163277 to N.L. and D.D., R01HL131517, R01HL089598, and R01HL165704 to D.D., and R01HL127717, R01HL169511, and R01HL171574 to J.F.M.), the American Heart Association (936111 to N.L. and 23POST1013888 to Y.Y.), the European Union (large-scale integrative project MAESTRIA, No. 965286 to D.D.), and the Deutsche Forschungsgemeinschaft (Research Training Group 2989, project 517043330 to D.D.). This project was assisted by core facilities at Baylor College of Medicine, including the Mouse Phenotyping Core (NIH UM1HG006348, R01DK114356, and S10OD023380), the Pathology and Histology Core (HTAP, NCI P30CA125123), the Genomic and RNA Profiling Core (NIH 1S10OD023469), and the Cytometry and Cell Sorting Core (CPRIT-RP180672, P30CA125123, and S10RR024574).

Ethical Approval

• Human atrial samples

To assess human cardiac samples, tissue specimens from right atrial appendages were collected from patients (>18 years) undergoing elective open-heart surgery. Individual patient data were extracted from the digital health record files. Written informed consent was obtained from every patient prior to cardiac surgery. All experimental protocols involving human tissue specimens and health data have been approved by the local ethical review board of the University Duisburg-Essen, Germany (#12–5268-BO).

• Animal studies

All involving mice were employed according to protocols approved by the Institutional Animal Care and Use Committee at Baylor College of Medicine and confirmed to the Guide for the Care and Use of Laboratory Animals published by National Institutes of Health. The protocol number is AN-7259.

Pre-registered Clinical Trial Number

None supplied.

Author notes

Pascal Martsch and Xiaohui Chen contributed equally to the study.