Graphical abstract

   graphic file with name fx1.jpg
   [43]Open in a new tab

Highlights

     * •
       T cell remodeling in coronary sinus blood of persistent atrial
       fibrillation (PerAF)
     * •
       FGFBP2(+)TRDC(−)CD4(−) T (Ftc-T) cells were enriched in coronary
       sinus blood of PerAF
     * •
       Ftc-T cells was closely associated with AF recurrence post catheter
       ablation
     * •
       IFI27 is highly expressed in PerAF and is closely related to
       myocardial fibrosis
     __________________________________________________________________

   Biological sciences; Immunology; Transcriptomics

Introduction

   Atrial fibrillation (AF) is a prevalent cardiac arrhythmia that
   significantly impacts individuals worldwide, often resulting in reduced
   long-term survival and heightened healthcare costs.[44]^1 Although
   catheter ablation (CA) is an effective treatment for AF, it is
   accompanied by a high recurrence rate for persistent AF (PerAF) over
   the long term.[45]^2 The progression of AF and its recurrence post-CA
   may closely correlate with the persistent progression of atrial
   fibrosis.[46]^3 Although the precise biological mechanisms disrupting
   atrial electrical stability and causing persistent progression of
   atrial fibrosis remain incompletely understood, inflammation and immune
   response may contribute to this progression.[47]^4

   The role of inflammatory and immune response in the development of
   atherosclerosis and heart failure (HF) has been well established.[48]^5
   Increasing evidence also supports the role of immune response in the
   pathology of AF.[49]^6 Patients with AF were found to have elevated
   levels of inflammatory markers and neutrophil/lymphocyte ratio (NLR)
   compared with the general population.[50]^4^,[51]^7 Inflammation can
   alter atrial electrophysiology and structural substrates, thereby
   leading to the onset and persistence of AF.[52]^8 In turn, AF might
   lead to calcium overload in atrial myocytes, resulting in cell death,
   danger-associated molecular pattern (DAMP) release, and subsequent
   inflammatory response activation.[53]^9 Therefore, inflammation and
   immune response are involved in the initiation and maintenance of AF,
   whereas AF can promote inflammation, leading to an ‘AF begets AF’
   phenomenon.

   The inflammation and immune response associated with AF tend to be a
   persistent yet low-grade nature, believed to be more conducive to the
   progression of atrial fibrosis.[54]^10 The enduring inflammatory state
   of AF is closely tied to the infiltration of immune cells and
   cytokines. Previous studies have demonstrated the involvement of
   T cells in the atherosclerosis process.[55]^11 Furthermore, immune cell
   infiltration, predominantly characterized by macrophages and T cells,
   has been identified in the left atrial appendage (LAA) of PerAF
   patients, suggesting that T cell-mediated immune responses may play a
   role in the pathology of AF. However, multiple subtypes of T cells
   exist, each with distinct roles in inflammation and fibrosis, and the
   abundance of a particular subtype may have a more pronounced
   impact.[56]^12 Because the proportion of immune cells in the blood
   perfusion tissue may indirectly mirror their abundance in the tissue,
   the prevalence and functional significance of T cell subtypes in
   cardiac tissue may be indirectly represented by their proportion in the
   cardiac circulation.

   The coronary sinus (CS) serves as the gathering place for blood
   collection following cardiac circulation. In comparison to peripheral
   blood, blood derived from the CS may possess more cardiac
   information.[57]^13^,[58]^14 Consequently, the composition and
   proportion of immune cells in the cardiac circulation of AF patients
   could offer a more insightful reflection of the impact of immune cells
   on the pathological processes of AF. Despite this potential, the
   specific signatures of immune cells from the CS circulation (CSC) in AF
   patients remain unknown, and there is currently a lack of reported
   studies addressing this aspect.

   Paroxysmal supraventricular tachycardia (PSVT) is a common cardiac
   arrhythmia, typically characterized by its benign and self-limiting
   nature. The anatomical basis of PSVT lies in the congenital presence of
   one or more abnormal conduction pathways in the atrioventricular
   junction area. However, other physiological characteristics and
   biological indicators of PSVT do not significantly different those of
   the general population. Therefore, PSVT is commonly utilized as a
   control group in research related to AF.

   Herein, we opted for PSVT as our control group, subsequently pioneering
   the creation of a single-cell RNA sequencing (scRNA-seq) atlas of
   immune cells in CSC of patients with PerAF. This initiative provides a
   more profound understanding of the signatures and heterogeneity of
   immune cells in the CSC of individuals with PerAF. Notably, our dataset
   unveiled a subset of T cells, denoted as FGFBP2(+)TRDC(−) CD4(−)
   T cells T cells (Ftc-T cells), showcasing a positive linear correlation
   with AF and AF recurrence post-ablation. Furthermore, IFI27 exhibited
   significant enrichment in Ftc-T cells. Therefore, Ftc-T cells may
   participate in the pathological process of atrial fibrosis progression
   in PerAF through IFI27 and the JAK/STAT signaling pathway, presenting a
   potential target for immunological interventions in patients with
   PerAF.

Results

Single-cell transcriptome atlas of immune cells from the CS in patients with
PerAF

   We isolated coronary sinus blood mononuclear cells (CBMCs) from
   individuals with PerAF and compared them with PSVT controls.
   Subsequently, we conducted scRNA-seq using the 10xGenomics platform
   (patient characteristics are presented in [59]Table 1). After rigorous
   quality controls (QC), involving the filtration of cells with low RNA
   content or high mitochondrial RNA, we identified 11,146 cells per
   sample, with an average of 1,819 genes per cell and 5542 mean UMI
   counts per cell (refer to [60]Table S1 and [61]Figure S1). The data
   were aggregated into a t-SNE plot ([62]Figure 1A), revealing 15
   clusters of cells ([63]Figures 1B and 1C) with different proportions
   ([64]Figure 1D and details in [65]Table S2). The heatmap displayed the
   top 10 marker genes in each cluster ([66]Figure 1E). Moreover, the
   feature and violin plot of the top marker genes of clusters are
   displayed in [67]Figure S2. The t-SNE projection plot highlighted a
   higher proportion of clusters 3, 4, and 5 in the PerAF group,
   surpassing those in the PSVT group (31.21% ± 2.22% vs. 9.12% ± 1.2%,
   p < 0.001; 22.56% ± 1.22% vs. 0.86% ± 0.21%, p < 0.001, and 15.68% ±
   1.03% vs. 4.84% ± 0.99%, p < 0.001, details in [68]Table S2).

Table 1.

   Characteristics of patients undergoing single-cell RNA sequencing
   Characteristics PerAF (n = 3) PSVT (n = 3) p-value
   Age, years 54 58 53 55 ± 2.6 49 59 55 54 ± 5 0.77
   Male, n (%) Male Male Male 3 (100%) Male Male Male 3 (100%) 1
   BMI, kg/m^2 22.6 24.8 25.3 24.2 ± 1.4 23.8 26.1 24.3 24.7 ± 1.2 0.66
   Duration of AF, months 12 10 13 11.7 ± 1.5 – – – –
   LVEF, (%) 61.5 62.1 60.5 61.4 ± 0.8 62.4 61.4 62.2 62 ± 0.5 0.32
   LAD, mm 43.2 40.5 44.1 42.6 ± 1.9 33.4 32.1 34.5 33.3 ± 1.2 0.002
   Hy, n (%) 1 1 0 2 (66.3%) 1 1 1 3 (100%) 1.00
   CHD, n (%) 0 0 1 1 (33.3%) 0 0 0 0 1.00
   Current or former smoker (%) 1 0 0 1 (33.3%) 0 1 1 2 (66.7%) 1.00
   [69]Open in a new tab

   Values are mean ± SD or n (%). PerAF, persistent atrial fibrillation,
   PSVT, paroxysmal supra-ventricular tachycardia, CHD, coronary heart
   disease, Hy, hypertension, BMI, Body mass index, LAD, left atrial
   diameter, LVEF, left ventricular ejection fraction. The statistical
   method used was Fisher exact test for comparison of categorical
   variables, and a Mann-Whitney U test for comparison of continuous
   variables. Statistical tests were 2-tailed, with p < 0.05 indicating
   statistical significance.

Figure 1.

   [70]Figure 1
   [71]Open in a new tab

   T-SNE projection plot of scRNA-seq data of CBMCs from PerAF and PSVT

   (A) T-SNE projection output of all single cells of CBMCs from PerAF
   (green) and PSVT (purple) subjects, each dot corresponds to a single
   cell.

   (B) T-SNE projection plot of CBMCs, showing the identification of 15
   main clusters. Each dot corresponds to a single cell, colored according
   to the cell cluster.

   (C) T-SNE projection plot of CBMCs yielded 15 clusters for PerAF (left)
   and PSVT (right) subjects respectively. Each dot corresponds to a
   single cell, colored according to the cell cluster.

   (D) Different clusters fractions of CBMCs from PerAF (left) and PSVT
   (right) subjects, colored according to the cell cluster (N = 15).

   (E) Heatmap of top 10 marker genes of the different clusters, colored
   according to the cell cluster (N = 15). ScRNA-seq, single-cell
   ribonucleic acid sequencing; PerAF, persistent atrial fibrillation;
   PSVT, paroxysmal supra-ventricular tachycardia; CBMCs, coronary sinus
   blood mononuclear cells; T-SNE, T-distributed stochastic neighbor
   embedding.

Characterization of CBMCs in patients with PerAF

   Referring to the transcriptomic datasets from ‘Human Primary Cell
   Atlas’,[72]^15 as illustrated in heatmap ([73]Figure 2A) and t-SNE plot
   ([74]Figure 2B and 2C), the bio-informatics cell annotation utilizing
   marker genes revealed that the majority of cells correspond to the five
   general immune cells types: T-lymphocyte cells (T cells 53.9%, cluster
   3, 5and 7), followed by monocytes (30.9%, cluster1, 8, 12 and 15) and
   natural killer cells (NK cells 13.4%, cluster 2, 4 and 6), B-lymphocyte
   cells (B cells 5.3%, cluster 9, 10, 11 and 14), and Dendritic cells (DC
   1.2%, cluster 13) ([75]Figure 2A).Furthermore, the PerAF group
   exhibited a significant increase in T cells compared with the PSVT
   group (71.2% ± 5.45%vs. 34.4% ± 3.01%, p < 0.001) ([76]Figure 2D,
   [77]Table S3). In contrast, the proportion of monocyte and NK cells was
   relatively lower in the PerAF group (details in [78]Table S3). High
   T cell abundance (as determined by flow cytometry analysis) was
   validated in the independent cohort (60.8% vs. 36.2%, p < 0.001)
   ([79]Figures 2E and 2F).

Figure 2.

   [80]Figure 2
   [81]Open in a new tab

   Identification and annotation of CBMCs between PerAF and PSVT subjects

   (A) Heatmap showing of different clusters matching traditional immune
   cell types.

   (B) T-SNE projection plot of CBMCs matched for traditional immune cell
   types. Each dot corresponds to a single cell, colored according to
   immune cell types.

   (C) T-SNE projection plot of CBMCs from PerAF (left) and PSVT (right)
   subjects matched for traditional immune cell types, respectively. Each
   dot corresponds to a single cell, colored according to immune cell
   types.

   (D) Abundance distribution of different immune cells of CBMCs of the
   PerAF (left) and PSVT (right) subjects.

   (E and F) Validation of the differential distribution of immune cells
   in the CSC of PerAF (left) and PSVT (right) subjects. (E) Flow
   cytometry analysis of differential distribution of immune cells in the
   CSC of PerAF (left) and PSVT (right) subjects in the validation cohort.
   (F) High abundance distribution of T cell were identified in CBMCs of
   PerAF patients, N = 10 vs. 10, two-tailed unpaired Student’s t tests,
   p < 0.05 indicating statistical significance.

   (G) Heatmap of top DEGs of CBMCs between PerAF and PSVT subjects.

   (H) Heatmap of top DEGs without immunoglobulin genes of CBMCs between
   PerAF and PSVT subjects.

   (I) PPI network of DEGs without immunoglobulin genes of CBMCs between
   PerAF and PSVT subjects.

   (J) KEGG classification enrichment analysis of DEGs without
   immunoglobulin genes of CBMCs between PerAF and PSVT subjects.

   (K) KEGG pathway enrichment analysis of DEGs without immunoglobulin
   genes of CBMCs between PerAF and PSVT subjects.

   (L) GO enrichment analysis of DEGs without immunoglobulin genes between
   PerAF and PSVT subjects. PerAF, persistent atrial fibrillation; PSVT,
   paroxysmal supra-ventricular tachycardia; T-SNE, T-distributed
   stochastic neighbor embedding; CSC, Coronary sinus circulation. CBMCs,
   coronary sinus blood mononuclear cells; DEGs, differentially expressed
   genes; PPI, protein–protein interaction; GO, gene ontology; KEGG, Kyoto
   Encyclopedia of Genes and Genomes.

   In comparison to the PSVT group, the PerAF group exhibited 41
   significantly up-regulated genes and 169 significantly down-regulated
   genes, as illustrated in the heatmap displaying the top 25
   differentially expressed genes (DEGs) ([82]Figure 2G). [83]Figure 2H
   further illustrates the DEGs excluding immunoglobulin genes. The
   up-regulated DEGs encompassed T cell-related marker genes
   (CD8A/CD8B/CD3D/CD3G/CD52/CD69/CCL5/LY6E), IFN-stimulated genes
   (IFI44L/IFI27/IFI6/ISG15), membrane protein genes (LIME1 and LTB), and
   genes involved in the regulation of RNAs and DNAs (NCOR2/MX1), among
   others. To understand the interactions among these genes, we
   constructed a PPI network using Metascape ([84]Figure 2I). The network
   revealed that most of the DEGs are associated with one another.
   Notably, the genes cluster: “CD8A/CD8B/CD3D/CD3G/CD52/CD69” was
   associated with IFN-stimulated genes, membrane protein genes, IL32,
   NCOR2 and MT2A either directly or indirectly. CD69, a marker gene for
   T cell activation, and CCL5, a marker gene for CD8^+ T cells, suggest
   that this specific cell subtype may represent a group of activated
   CD8^+ T cells. These cells may function by regulating the secretion of
   cytokines, influencing the expression of membrane proteins, and
   interfering with the translation and transcription of genes, among
   other mechanisms.

   As immunoglobulin genes are not part of the regular transcriptome, and
   their inclusion might impact the result of the GO analysis, we
   conducted an analysis excluding immunoglobulin genes. The DEGs without
   immunoglobulin genes were subjected to enrichment analysis in the KEGG
   and GO database. The KEGG classification enrichment highlighted a
   significant proportion of DEGs primarily influenced the “immune system,
   endocrine system, and circulatory system” in Organismal Systems, and
   their contribution to “infectious diseases, endocrine and metabolic
   diseases, and cardiovascular diseases” within Human Diseases.
   Additionally, they played roles in “signaling molecules and
   interaction, signal transduction” in Environment Information
   Processing, and were involved in “transport and catabolism, cell growth
   and death” in Cellular Processes ([85]Figure 2J). Further KEGG pathway
   enrichment analysis of the DEGs revealed highly enriched pathways,
   including rheumatoid arthritis, T cell receptor signaling pathway, Th17
   cell differentiation, Th1 and Th2 cell differentiation, NF-kappaβ
   signaling pathway, Ferroptosis, Phagosome, Lysosome, and others
   ([86]Figure 2K). Functional enrichment analysis using the GO database
   showcased major terms in the biological process (BP), such as immune
   system, neutrophil degranulation, cell surface receptor signaling
   pathway, and T cell receptor signaling pathway, among others. In terms
   of cellular component (CC), most of the genes were enrichment in plasma
   membrane, secretory granule membrane, lysosome, extracellular space,
   extracellular exosome, etc. Moreover, the significantly enriched
   molecular function (MF) term included carbohydrate binding, IgG
   binding, transmembrane signaling receptor activity, RAGE receptor
   binding, and protein binding, among others ([87]Figure 2L).

Characterization of T cells from CBMCs in patients with PerAF

   Given that T cells constituted the largest population exhibiting
   substantial transcriptional changes associated with AF ([88]Figure 3A),
   we subsequently re-clustered the entire T cell population to
   investigate the influence of T cells on AF and identify differentially
   regulated genes. The t-SNE output revealed 11 clusters representing
   different T cell subtypes ([89]Figure 3B and 3C) with varying
   proportions, notably cluster 2,exhibited a significantly higher
   proportion in T cells from PerAF patients compared to the PSVT group
   (28.8% ± 3.34% vs. 3.1% ± 1.12%, p < 0.001) ([90]Figure 3D and details
   in [91]Table S4). The heatmap displayed the top 10 marker genes of each
   cluster ([92]Figure 3E), and the feature and violin plot of the top
   markers gene of clusters are displayed in [93]Figure S3.

Figure 3.

   [94]Figure 3
   [95]Open in a new tab

   ScRNA-seq data analysis of T cells between PerAF and PSVT subjects

   (A) T-SNE projection plot of T cells of CBMCs between PerAF and PSVT
   subjects.

   (B) Dimensional reduction t-SNE projection plot of scRNA-seq data of
   T cells yielded 11 clusters.

   (C) Dimensional reduction t-SNE projection output of scRNA-seq data of
   T cells for PerAF (right) and PSVT (left) subjects yielded 11 clusters,
   respectively.

   (D) Abundance distribution of different clusters of T cells from CBMCs
   between the PerAF (left) and PSVT (right) subjects.

   (E) Heatmap of top 10 marker genes for clusters of T cells of CBMCs
   between PerAF and PSVT subjects.

   (F) Heatmap of top DEGs without immunoglobulin genes of T cells from
   CBMCs between PerAF and PSVT subjects.

   (G) KEGG classification enrichment analysis of DEGs without
   immunoglobulin genes of T cells from CBMCs between PerAF and PSVT
   subjects.

   (H) KEGG pathway enrichment analysis of DEGs without immunoglobulin
   genes of T cells from CBMCs between PerAF and PSVT subjects.

   (I) GO enrichment analysis of DEGs without immunoglobulin genes of
   T cells from CBMCs between PerAF and PSVT subjects. scRNA-seq,
   single-cell ribonucleic acid sequencing; PerAF, persistent atrial
   fibrillation; PSVT, paroxysmal supra-ventricular tachycardia; T-SNE,
   T-distributed stochastic neighbor embedding; CBMCs, coronary sinus
   blood mononuclear cells; DEGs, differentially expressed genes; GO, gene
   ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes.

   Compared with the PSVT group, we identified 37 significantly
   up-regulated and 16 significantly down-regulated genes were identified
   in T cells of the PerAF group. The top DEGs of T cells closely mirrored
   those observed in CBMCs between the two groups ([96]Figure 3F).
   Additionally, the KEGG classification analysis suggested that these
   DEGs acted on the immune system, infectious and cardiovascular diseases
   and were associated with signaling molecules and interaction, signal
   transduction and transport, and cell grown and death ([97]Figure 3G).
   KEGG pathway enrichment indicated that many genes were enriched in the
   rheumatoid arthritis, Influenza A, T cell receptor signaling pathway,
   Th1 and Th2 cell differentiation, NF-kappaβ signaling pathway,
   Ferroptosis etc.([98]Figure 3H). Additionally, the DEGs of T cells were
   associated with GO terms related to type I interferon signaling
   pathway, immune response, T cell receptor signaling pathway in terms of
   BP; extracellular region, cell surface, mitochondrion, extracellular
   exosome, plasma membrane, endosome in terms of CC; protein
   homodimerization activity, protein binding, RAN binding and DNA binding
   in terms of MF ([99]Figure 3I).

Characterization of T cells subtypes from CBMCs in patients with PerAF

   Match with the known marker genes of various T cells ([100]Table S5),
   which mainly include CD4 naive T cells, CD4 effector T cells, CD4 Treg
   cells, CD8 naive T cells, CD8 cytotoxic T cells, GdT (γδT) cells, NKT
   cells ([101]Figure 4A). In the t-SNE output, a significantly higher
   proportion of cytotoxic CD8^+ cells and exhausted CD8^+T (T-ex) cells
   were identified in PSVT (44.4% ± 5.51% vs. 17.0% ± 3.12%, p < 0.001)
   ([102]Figure 4B and 4C, [103]Table S6). The clusters of T cells were
   matched with the marker genes of general T cell subsets, indicating
   that cells in the cluster2 were mainly cytotoxic CD8^+ cells and
   exhausted CD8^+T (T-ex) cells ([104]Figure 4D).

Figure 4.

   [105]Figure 4
   [106]Open in a new tab

   T cell subtypes from CBMCs between PerAF and PSVT subjects

   (A) Heatmap of genes expressed by T cells from CBMCs between PerAF and
   PSVT subjects matching known marker genes of different T cell subsets.

   (B) T-SNE projection plot of different T cells subtypes from CBMCs
   between PerAF and PSVT subjects.

   (C) Abundance distribution of different T cell subtypes from CBMCs
   between the PerAF (left) and PSVT (right) subjects.

   (D) Heatmap of marker genes of different T cell subsets matched with
   clusters of T cells from CBMCs between PerAF and PSVT subjects.

   (E) Heatmap of top 25 DEGs of cluster #2 of T cells from CBMCs between
   PerAF and PSVT subjects.

   (F) KEGG classification enrichment analysis of DEGs without
   immunoglobulin genes of cluster#2 of T cells from CBMCs between PerAF
   and PSVT subjects.

   (G) KEGG pathway enrichment analysis of DEGs without immunoglobulin
   genes of cluster#2 of T cells from CBMCs between PerAF and PSVT
   subjects.

   (H) GO enrichment analysis of DEGs without immunoglobulin genes of
   cluster#2 of T cells from CBMCs between PerAF and PSVT subjects. PerAF,
   persistent atrial fibrillation; PSVT, paroxysmal supra-ventricular
   tachycardia; T-SNE, T-distributed stochastic neighbor embedding; CBMCs,
   coronary sinus blood mononuclear cells; DEGs, differentially expressed
   genes; GO, gene ontology; KEGG, Kyoto Encyclopedia of Genes and
   Genomes.

   In cluster 2 of T cells, 63 genes were significantly up-regulated, and
   16 genes were significantly down-regulated (Top 25 DEGs)
   ([107]Figure 4E). Intriguingly, we observed that the top DEGs in
   cluster 2 were similar to those in CBMCs and T cells ([108]Table S7).
   KEGG classification analysis indicated that these DEGs functioned in
   the immune and circulatory system, as well as in infectious and
   cardiovascular diseases. They were also associated with signaling
   molecules and interaction, signal transduction, and cell growth and
   death ([109]Figure 4F). KEGG pathway enrichment revealed that DEGs were
   primarily enriched in the T cell receptor signaling pathway, Th17 cell
   differentiation, Th1 and Th2 cell differentiation, Jak/STAT signaling
   pathway, PI3K/Akt signaling pathway, and RNA transport, among others
   ([110]Figure 4G). Furthermore, GO enrichment demonstrated that, in
   terms of biological processes (BP), most terms were associated with the
   immune response, type I interferon signaling pathway, and T cell
   receptor signaling pathway; in cellular components (CC), most terms
   were associated with the plasma membrane, extracellular exosome,
   endosome membrane, etc. Regarding molecular functions (MF), most terms
   were associated with antigen binding, immunoglobulin receptor binding,
   carbohydrate binding, RNA binding, transmembrane signaling receptor
   activity, GTP binding, and DNA-binding transcription factor activity,
   among others ([111]Figure 4H).

   Collectively, the KEGG and GO enrichment analysis of cluster 2 were
   akin to those observed in T cells. Hence, it is plausible that cluster
   2 represents functionally significant cells among immune cells, playing
   a crucial role in the pathological processes of PerAF and maintaining a
   close association with PerAF.

The redefinition and function of cluster2

   FGFBP2 was the top marker gene of cluster 2 ([112]Figure 3E).However,
   beyond cluster 2, cells expressing FGFBP2 were also scattered in
   cluster 7, 8 and 11 ([113]Figures 3E and [114]5A-a).To refine the
   representation of cluster 2, we selectively excluded FGFBP2 expressing
   cells from cluster 7, 11 ([115]Figure 5A-b) and 8 ([116]Figure 5A-d).
   This exclusion process aimed to enhance the precision of FGFBP2
   expressing cells, more accurately reflecting the characteristics of
   cluster 2. Notably, TRDC and CD4 served as marker genes for cluster 7,
   11 and 8 ([117]Figure 3E). Co-expressing TRDC with FGFBP2 and CD4 with
   FGFBP2 accounted for 95.6% and 96.3% of cluster 7, 11 and 8
   ([118]Figures 5A-c and [119]6A-e). As shown in the t-SNE output, cells
   expressing FGFBP2 but not TRDC and CD4 (FGFBP2(+)TRDC(−)CD4(−) T cells)
   ([120]Figure 5A-f) closely aligned with cluster 2 (95.3%)
   ([121]Figure 5A-g). Consequently, we apply FGFBP2(+)TRDC(−) CD4(−)
   T cells (Ftc-T cells) to represent Cluster 2.

Figure 5.

   [122]Figure 5
   [123]Open in a new tab

   The redefinition and function of cluster#2

   (A) T-SNE projection output of FGFBP2-expressing cells in T cells (a);
   T-SNE projection output of FGFBP2-expressing cells in cluster #7 and
   #11 of T cells from CBMCs (b); T-SNE projection output of cells
   co-expressing TRDC with FGFBP2 (c); T-SNE projection output of
   FGFBP2-expressing cells in cluster #8 of T cells from CBMCs (d); T-SNE
   projection output of cells co-expressing CD4 with FGFBP2 (e); T-SNE
   projection output of FGFBP2+TRDC-CD4^−cells (Ftc-T cells) of T cells
   from CBMCs (f); T-SNE projection output of cells of cluster #2 (g).

   (B) Validation of the abundance of Ftc-T cells from CBMCs of the PerAF
   subjects by flow cytometry analysis in the validation cohort, N = 10
   vs. 10.

   (C) Validation of the abundance of Ftc-T cells from CBMCs of the PSVT
   subjects by flow cytometry analysis in the validation cohort, N = 10
   vs. 10.

   (D) Validation of FTR from CBMCs of PerAF (right) and PSVT (left)
   subjects by flow cytometry analysis in the validation cohort. N = 10
   vs. 10, two-tailed unpaired Student’s t tests, p < 0.05 indicating
   statistical significance, ∗∗∗∗p < 0.001, ∗∗∗∼∗∗p < 0.01, ∗p < 0.05.

   (E) Validation of FCR from CBMCs of PerAF (right) and PSVT (left)
   subjects by flow cytometry analysis in the validation cohort. N = 10
   vs. 10, two-tailed unpaired Student’s t tests, p < 0.05 indicating
   statistical significance, ∗∗∗∗p < 0.001, ∗∗∗∼∗∗p < 0.01, ∗p < 0.05.

   (F) Light microscopy view of Ftc-T cells co-cultured with CFs.

   (G) Co-culture of Ftc-T cells with CFs can promote the proliferation of
   CFs. Representative images of EDU incorporation (green) reflecting cell
   proliferation (a) and quantification of EDU-positive cells (b). Data
   are means ± SEM (n = 3); ∗∗∗∗p < 0.001, ∗∗∗∼∗∗p < 0.01, ∗p < 0.05.
   PerAF, persistent atrial fibrillation; PSVT, paroxysmal
   supra-ventricular tachycardia; T-SNE, T-distributed stochastic neighbor
   embedding; CBMCs, coronary sinus blood mononuclear cells; FTR, Ftc-T
   cells to the total T cells ratio; FCR, Ftc-T cells to the CBMCs ratio;
   CFs, cardiac fibroblasts.

Figure 6.

   [124]Figure 6
   [125]Open in a new tab

   Effect of IFI27 on cardiac fibroblasts

   (A) Differential expression of IFI27 between PerAF and PSVT subjects.
   a: differential expression of IFI27 in CBMCs between PerAF and PSVT
   subjects; b, differential expression of IFI27 in T cells between PerAF
   and PSVT subjects; c, differential expression of IFI27 in Ftc-T cells
   between PerAF and PSVT subjects. N = 10 vs. 10, Ftc-T cells:
   FGFBP2+TRDC-CD4^− T cells, two-tailed unpaired Student’s t tests,
   p < 0.05 indicating statistical significance, ∗∗∗∗p < 0.001,
   ∗∗∗∼∗∗p < 0.01, ∗p < 0.05.

   (B–D) Validation of the differential expression of IFI27 in Ftc-T cells
   from CBMCs between PerAF and PSVT subjects. (B) Flow cytometry analysis
   of the differential expression of IFI27 in Ftc-T cells from CBMCs
   between PerAF (red) and PSVT (blue) subjects; (C) High expression of
   IFI27 in Ftc-T cells were identified in CBMCs of PerAF (right) than
   PSVT (left) subjects; (D) Validation of the differential expression of
   IFI27 in Ftc-T cells from CBMCs between PerAF and PSVT subjects by
   qRT-PCR in the validation cohort. N = 10 vs. 10, Ftc-T cells:
   FGFBP2+TRDC-CD4^− T cells, two-tailed unpaired Student’s t tests,
   p < 0.05 indicating statistical significance, ∗∗∗∗p < 0.001,
   ∗∗∗∼∗∗p < 0.01, ∗p < 0.05.

   (E) Western blot (a, b) and qrt PCR (c) results showed that IFI27
   protein over-expression and knockdown can promote and inhibit the
   expression of col1 and a-SMA in CFs. Data are means ± SEM (n = 3);
   ∗∗∗∗p < 0.001, ∗∗∗∼∗∗p < 0.01, ∗p < 0.05.

   (F) IFI27 protein overexpression and knockdown could promote and
   inhibit the proliferation of CFs. Representative images of EDU
   incorporation (green) reflecting cell proliferation (a) and
   quantification of EDU-positive cells (b). Data are means ± SEM (n = 3);
   ∗∗∗∗p < 0.001, ∗∗∗∼∗∗p < 0.01, ∗p < 0.05.CFs: cardiac fibroblasts.

   Subsequently, we employed the Ftc-T cells to the total T cells ratio
   (FTR) and CBMCs ratio (FCR) to gauge the abundance of Ftc-T cells. In
   the validation cohort, flow cytometry analysis confirmed significantly
   higher FTR and FCR in the PerAF group compared to the PSVT group
   ([126]Figure 5B and 5C) (FTR: 27.31 ± 3.11% vs.7.24% ± 1.38%,
   p < 0.001([127]Figure 5D); FCR: 17.19 ± 2.71% vs. 2.95 ± 0.87%,
   p < 0.001, ([128]Figure 5E)). Moreover, the further cell experiments
   in vitro, we co-cultured Ftc-T cells with cardiac fibroblasts (CFs) and
   observed a substantial promotion of CF proliferation by Ftc-T cells
   ([129]Figures 5F and 5G).

IFI27can promote cardiac fibroblast proliferation and collagen expression

   Among the upregulated genes, we identified interferon alpha inducible
   protein 27 (IFI27), which was significantly enriched in CBMCs
   ([130]Figure 6A-a), T cells ([131]Figure 6A-b) and cluster 2 of T cells
   ([132]Figure 6A-c) in the CSC of patients with PerAF. The t-SNE output
   illustrating cells expressing IFI27 is presented in [133]Figure S4.
   Flow cytometry and qRT-PCR further confirmed the upregulation of
   IFI27([134]Figures 6B and 6C), with a 3.27-fold increase in mean
   fluorescence intensity (MFI) (p < 0.001) ([135]Figure 6D) in Ftc-T
   cells in the PerAF group compared to the PSVT group. This consistency
   between the scRNA-seq and the validation data reinforces the robustness
   of our findings.

   To delve deeper into the role of IFI27 in promoting fibrosis, we
   explored its effects on the proliferation and collagen expression of
   CFs. IFI27 over-expression, in contrast to control cells, resulted in
   an elevation in the expression of COL-1 and α-SMA ([136]Figure 6E) and
   promoted CFs proliferation ([137]Figure 6F). Conversely, IFI27
   knockdown yielded opposite effects ([138]Figures 6E and 6F). These
   results suggested that IFI27 exerts a profibrotic effect.

Ftc-T cells and recurrence of PerAF post-CA

   All PerAF patients from the validation cohort and the prospectively
   enrolled cohort completed a follow-up period of at least 6 months, with
   a median follow-up of 8 [7.0, 10.0] months. Among them, 22 patients
   experienced AF recurrence. Patient characteristics are summarized in
   [139]Table 2. The FTR and FCR were significantly higher in the
   recurrence group than in the sinus rhythm (SR) group ([140]Figures 7A
   and 7B). Univariate analysis suggested that FTR and FCR were
   independent risk factors for AF recurrence. Other risk factors included
   the duration of AF and left atrial diameter (LAD) ([141]Table S8).
   After controlling for potential confounders such as LAD and duration
   (Model 1), Cox regression suggested that the crude recurrence rates
   were progressively higher with increasing FTR and FCR. Moreover, hazard
   ratios (HRs) maintained a similar trend in models 2 and 3
   ([142]Table 3). Therefore, FTR and FCR might be significantly and
   positively associated with AF recurrence. Additionally, we found that
   both the dose-response trajectories of FTR and FCR were linear (P
   linear = 0.0261 and 0.0251, P nonlinear = 0.4368 and 0.4753).
   Trajectories for FTR and FCR of model 1 were therefore plotted using
   the linear form and are shown in [143]Figures 7C and 7E (FTR),
   [144]Figures 7D and 7F (FCR). The recurrence risk HR (95% CI)
   associated with FTR was 1.34 (1.08–1.65), and for FCR, it was 1.39
   (1.19–1.63) after adjustment for LA and duration of AF, suggesting that
   the trajectory was steeper for FCR than FTR.

Table 2.

   Characteristics of patients after catheter ablation
   Overall (n = 80) SR(n = 58) Recurrence(n = 22) p-value
   Age, years 50 (45, 55) 50 (45, 55) 47 (44, 55) 0.63
   Male, n (%) 52 (65.0%) 36 (62.1%) 16 (72.7%) 0.53
   BMI, kg/m^2 22.8 ± 1.8 22.9 ± 1.6 22.5 ± 2.1 0.29
   Duration of AF, months 8.0 (5.0, 11.0) 7.2 (5.2, 10.0) 11.0 (8.4, 13.2)
   <0.01
   e-GFR 99 (95,113) 102 (95,114) 98 (93,111) 0.45
   Stroke 3 1 2 0.33
   LVEF, % 58.8 (55.6, 60.9) 58.6 (55.6, 61.0) 59.1 (56.8, 60.5) 0.94
   LAD, mm 42.0 ± 3.9 41.6 ± 3.9 43.0 ± 3.8 0.15
   CHD, n (%) 23 (28.7%) 15 (25.9%) 8 (36.4%) 0.52
   Hy, n (%) 39 (48.8%) 25 (43.1%) 14 (63.6%) 0.16
   Follow-up, months 8.0 (7.0, 10.0) 8.0 (7.0, 10.0) 8.5 (7.3, 10.0) 0.36
   [145]Open in a new tab

   Values are mean ± SD, median (interquartile range, IQR) or n (%). AF,
   atrial fibrillation, CHD, coronary heart disease, Hy, hypertension,
   BMI, Body mass index, LAD, left atrial diameter, LVEF, left ventricular
   ejection fraction. The statistical method used was Fisher exact test
   for comparison of categorical variables, and a Mann-Whitney U test for
   comparison of continuous variables. Statistical tests were 2-tailed,
   with p < 0.05 indicating statistical significance.

Figure 7.

   [146]Figure 7
   [147]Open in a new tab

   Ftc-T cells and recurrence of PerAF post-CA

   (A) FTR in patients with AF recurrence (right, n = 22) and SR
   maintenance (left, n = 78) post-CA. Two-tailed unpaired Student’s t
   tests, p < 0.05 indicating statistical significance, ∗∗∗∗p < 0.001,
   ∗∗∗∼∗∗p < 0.01, ∗p < 0.05.

   (B) FCR in patients with AF recurrence (right, n = 22) and SR
   maintenance (left, n = 78) post-CA. Two-tailed unpaired Student’s t
   tests, p < 0.05 indicating statistical significance, ∗∗∗∗p < 0.001,
   ∗∗∗∼∗∗p < 0.01, ∗p < 0.05.

   (C and E) Continuous dose-response relationship between AF recurrence
   and FTR, estimated using linear Cox regression models.

   (D and F) Continuous dose-response relationship between AF recurrence
   and FCR, estimated using linear Cox regression models.

   (G) The time-dependent ROC curves for FTR and FCR for AF recurrence.

   (H) Kaplan-Meier survival curves of patients free from AF for FTR-[L]
   and FTR-[H] group following ablation. Log rank test with p < 0.05
   indicates statistical significance.

   (I) Kaplan-Meier survival curves of patients free from AF for FCR-[L]
   and FCR-[H] group following ablation. Log rank test with p < 0.05
   indicates statistical significance.

   (J) Kaplan-Meier survival curves of patients free from AF for
   FTR-[L]+FCR-[L,] FTR-[H]+FCR-[L,] FTR-[L]+FCR-[H] and FTR-[H]+FCR-[H]
   group following ablation. Log rank test with p < 0.05 indicates
   statistical significance. PerAF, persistent atrial fibrillation; Ftc-T,
   FGFBP2+TRDC-CD4^− T cells; CA, catheter ablation; FTR, Ftc-T cells to
   the total T cells ratio; FCR, Ftc-T cells to the CBMCs ratio; SR, sinus
   rhythm; ROC, receiver operating characteristic; FTR-[L], FTR<27.3%;
   FTR-[H], FTR>27.3%, FCR-[L]: FCR<17.4%, FCR-[H]: FCR>17.4%.

Table 3.

   Associations of AF recurrence with FTR and FCR
   Models FTR p-value FCR p-value FTR-[H] vs. FTR-[L] p-value FCR-[H] vs.
   FCR-[L] p-value
   Model 1 1.34 (1.08–1.65) 0.007 1.39 (1.19–1.63) <0.0001 2.84
   (1.03–7.86) 0.04 10.42 (2.98–36.43) <0.0001
   Model 2 1.33 (1.07–1.65) 0.011 1.31 (1.13–1.52) <0.0001 2.72
   (0.99–7.50) 0.052 8.45 (2.49–28.64) <0.0001
   Model 3 1.30 (1.05–1.62) 0.018 1.28 (1.11–1.49) <0.0001 2.47
   (0.90–6.78) 0.0785 8.19 (2.40–27.89) <0.0001
   [148]Open in a new tab

   Continuous variables (per 0.01). Model 1 = LA + Duration adjusted;
   Model 2 = age+sex adjusted; Model 3 = sex+BMI adjusted; LA, left atrial
   diameter, Duration, duration of AF; BMI, body mass index; AF, atrial
   fibrillation; HR, hazard ratio; CI, confidential interval; FTR, Ftc-T
   cell to total T cells ratio; FCR, Ftc-T cell to CBMCs ratio. FTR-L,
   Ftc-T cell to total T cells ratio<27.3%; FTR-H, Ftc -T cells to total
   T cells ratio>27.3; FCR-L, Ftc-T cells to CBMCs ratio<17.4%; FCR-H,
   Ftc-T cells to CBMCs ratio>17.4%. Data are expressed as HR (95% CI).
   Statistical tests were 2-tailed, with p < 0.05 indicating statistical
   significance.

   The time-dependent ROC curve illustrated the predictive capacity of
   continuous values for FTR and FCR in AF recurrence, yielding AUROC
   values of 0.711 and 0.799, respectively. When comparing the AUROC of
   FTR with that of FCR, there was an increase of 0.088 for FCR,
   suggesting its potential superiority in discriminating patients at high
   risk of recurrence compared to FTR ([149]Figure 7G). Optimal cut-off
   values for FTR and FCR, determined using the maximum Youden index, were
   27.3% and 17.4%, respectively. These thresholds demonstrated sufficient
   sensitivity (0.875 and 0.938) and specificity (0.547 and 0.625) for
   predicting the risk of AF recurrence. Based on these cut-off values,
   individuals were categorized into the following groups: FTR-[H]
   (FTR≥27.3%), FTR-[L] (FTR<27.3%), FCR-[H] (FCR≥17.4%), and FCR-[L]
   (FCR<17.4%). Cox proportional hazard regression and survival analysis
   (Kaplan-Meier analysis) revealed that individuals with FTR-[H] and
   FCR-[H] had a significantly higher risk of recurrence, with hazard
   ratios (HRs) of 2.84 (1.03–7.86) and 10.42 (2.98–36.43) in Model 1
   ([150]Table 3; [151]Figures 7H and 7I). Even after adjusting for other
   factors in Models 2 and 3, the risk of AF recurrence in the FTR-[H] and
   FCR-[H] groups remained significantly higher than that in the FTR-[L]
   and FCR-[L] groups.

   A joint analysis of FTR and FCR with AF recurrence risk was further
   performed. The characteristics of patients in different groups are
   displayed in [152]Table S9. Results from the joint analysis are shown
   in [153]Table 4. In all models, AF recurrence risk was higher for
   patients with FTR-[H]+FCR-[H] than for patients with only FTR-[H] or
   with only FCR-[H], which was indicative of an additional interaction.
   The survival curve also suggested that the AF recurrence in the
   FTR-[H]+FCR-[H] group was significantly higher than that in the other
   groups, followed by the FTR-[L]+FCR-[H] group ([154]Figure 7J).

Table 4.

   Joint associations of AF recurrence with FTR and FCR
   Models FTR-[L]+FCR-[L] FTR-[H]+FCT-[L] p-value FTR-[L]+FCR-[H] p-value
   FTR-[H]+FCR-[H] p-value
   Model 1 1 (ref) 0.84 (0.07–9.55) 0.890 7.25 (1.12–46.92) 0.038 10.32
   (2.27–47.07) 0.003
   Model 2 1 (ref) 0.67 (0.06–7.51) 0.749 4.40 (0.73–26.46) 0.105 8.27
   (1.88–36.24) 0.005
   Model 3 1 (ref) 0.61 (0.05–6.77) 0.686 3.98 (0.65–24.33) 0.135 7.69
   (1.74–33.96) 0.007
   [155]Open in a new tab

   Continuous variables (per 0.01). Model 1 = LA + Duration adjusted;
   Model 2 = age + sex adjusted; Model 3 = sex + BMI adjusted; LA, left
   atrial diameter, Duration, duration of AF; BMI, body mass index; AF,
   atrial fibrillation; HR, hazard ratio; CI, confidential interval; FTR,
   Ftc-T cell to total T cells ratio; FCR, Ftc-T cell to CBMCs ratio.
   FTR-L, Ftc-T cell to total T cells ratio<27.3%; FTR-H, Ftc -T cells to
   total T cells ratio>27.3; FCR-L, Ftc-T cells to CBMCs ratio<17.4%;
   FCR-H, Ftc-T cells to CBMCs ratio>17.4%. Data are expressed as HR (95%
   CI). Statistical tests were 2-tailed, with p < 0.05 indicating
   statistical significance.

Discussion

   To the best of our knowledge, this study represents the inaugural
   exploration of immune cell signatures within the CSC of PerAF patients
   by scRNA-seq. The primary findings reveal profound transcriptional
   alterations in T cell profiles within the CSC of PerAF patients,
   unveiling a T cell subset termed Ftc-T cells, predominantly comprised
   cytotoxic CD8^+ T cells and T-ex cells. Notably, Ftc-T cells were
   identified as potent promoters of CFs proliferation. Additionally, the
   study identifies FTR and FCR as significantly associated with AF
   recurrence, displaying a positive linear dose-response relationship.
   The top DEGs between PerAF and PSVT patients in CBMCs, T cells, and
   Ftc-T cells exhibited similarities. KEGG and GO enrichment analysis
   implicated these DEGs in immune diseases, inflammatory diseases,
   signaling pathways, cell growth, death, etc. Moreover, among the top
   DEGs in Ftc-T cells, IFI27 was identified as capable of promoting CFs
   proliferation and collagen expression. Consequently, IFI27may be a
   potentially crucial factor in the progression of atrial fibrosis
   induced by Ftc-T cells, providing new insights into the underlying
   mechanisms through which Ftc-T cells may impact PerAF.

   CA is an pivotal treatment for PerAF, however, its long-term efficacy
   is compromised, revealing individual heterogeneity likely arising from
   the continuous and personalized progression of atrial fibrosis.[156]^16
   The inflammatory response triggered by the ongoing infiltration of
   immune cells plays a pivotal role in the progression of atrial
   fibrosis.[157]^12 CS, serving as the collection site for blood
   post-cardiac circulation, contains cardiac-specific information that
   may be richer than that found in peripheral blood.[158]^14
   Investigating the characteristics of immune cells in the CS may thus
   offer better insights into the impact of immune cells on inflammation
   and atrial fibrosis in PerAF.

   A potential relationship between T cell immunity and the development of
   AF has been proposed.[159]^17^,[160]^18 Elevated CD4^+ T cells have
   been documented in the peripheral blood of AF patients. Moreover,
   circulating T helper 17 (Th17) cells have been linked to post-operative
   AF (POAF), with a higher Th17/Treg ratio positively correlated with AF
   risk.[161]^19^,[162]^20 While CD8^+ T cells have a pivotal cytotoxic
   role in immune diseases and viral infections, but little is known about
   their role in AF progression. Recently, CD8^+CD28^null T cells were
   proved to be an independent predictor for POAF.[163]^21 In the present
   study, utilizing scRNA-seq, we unveiled substantial alterations in
   immune cells within the CS of PerAF patients, particularly a
   significant increase in T cells .Subsequent re-clustering of T cells
   led to the identification of a T cell subset termed Ftc-T cells,
   capable of promoting CF proliferation. Ftc-T cells were predominantly
   composed of CD8^+ T cells and T-ex cells, simplifying the potential
   involvement of CD8^+ T cells in PerAF progression. Furthermore, we
   reported for the first time that T-ex cells were significantly elevated
   in CSC of patients with PerAF. T-ex cells, representing CD8^+ T cells
   in a dysregulated state of “exhaustion” due to prolonged exposure to
   sustained immune stimulation, have been recognized for their immune
   significance in chronic viral infections and tumors.[164]^22 The
   notable increase in T-ex cells in the CS of PerAF patients suggests
   their potential participation in the chronic and persistent
   inflammation associated with PerAF, even though the detailed mechanism
   remains unclear.

   Followup data suggested that FTR and FCR were positively correlated
   with AF recurrence post-CA. On average, each additional 1% of FTR was
   associated with a 34% increase in multivariable-adjusted risk for AF
   recurrence. Dose-dependent increased risk of 39% was also observed for
   each 1% increase in FCR. When FTR and FCR were jointly classified,
   PerAF patients with the FTR-[H] and FCR-[H] had the greatest AF
   recurrence risk. Even after adjusting for different confounding
   factors, the results also indicated that FTR and FCR could effectively
   predict recurrence post-CA. Therefore, Ftc-T cells may be the important
   immune cell subset that promotes the progression of PerAF.

   ScRNA-seq analysis of DEGs of CBMCs, T cells, and Ftc-T cells subset
   from CSCs of PerAF patients showed significantly elevated IFI27
   expression. Emerging data suggest that IFI27 is a mitochondrial
   protein, which can activate the Jak/STAT signaling pathway,[165]^23
   which not only can promote release of pro-inflammatory and pro-fibrotic
   factors and then facilitates the differentiation of fibroblasts into
   myofibroblasts,[166]^24 but can also interact with AKT and MAPK/ERK
   pathways to promote tumor cells proliferation, growth, and
   migration.[167]^25 In addition, IFI27 was also found to be elevated in
   liver fibrosis and idiopathic pulmonary fibrosis tissues, suggesting
   that IFI27 is closely related to tissue fibrosis.[168]^26^,[169]^27 Our
   study confirmed a substantial presence of Ftc-T cells highly expressing
   IFI27 in the CSC of PerAF patients. Moreover, over-expression of IFI27
   significantly promoted the proliferation and collagen expression of
   CFs, while silencing IFI27 inhibits CFs proliferation and collagen
   expression. Ftc-T cells continuously infiltrates the atrial tissue with
   the blood circulation, resulting in the enrichment of IFI27 in the
   atrial tissue. This enrichment promotes fibroblasts transformation and
   proliferation through the release of pro-fibrotic factors, ultimately
   leading to the progression of atrial fibrosis. In addition, KEGG
   enrichment analysis of DGEs suggested the involvement of multiple genes
   in the JAK/STAT signaling pathway. Therefore,the JAK/STAT signaling
   pathway may have an important role in the promotion of atrial fibrosis
   by Ftc-T cells and IFI27. Ftc-T cells may participate in the
   pathological process of atrial fibrosis progression in PerAF through
   IFI27 and, simultaneously, serve as potential target for immunological
   interventions in patients with PerAF.

   In summary, scRNA-seq revealed the immune cells atlas of CSC in
   patients with PerAF, highlighting a high proportion of T cells and
   Ftc-T cell subset in the CSC of patients with PerAF.Ftc-T cells, in
   addition to promoting the proliferation of CFs, but high FTR and FCR
   are closely related to the recurrence of AF post-CA of PerAF.
   Furthermore, IFI27 emerged as a potentially crucial factor in the
   progression of AF and may serve as a potential target for AF
   intervention.

Limitations of the study

   Despite its numerous strengths, this study should be interpreted in
   light of some limitations. First, scRNA-seq technology has inherent
   constraints, such as a limited ability to infer cell-cell interactions
   and capture spatial information, necessitating supplementation with
   other technologies. Second, while this study represents the largest
   scRNA-seq investigation of immune cells in CSC of patients with PerAF
   to our knowledge, the number of subjects studied is limited. We cannot
   exclude the possibility that comorbidities, differences in risk
   factors, or therapies might have influenced the observed changes in
   gene expression patterns. Thirdly, although we identified a
   high-abundance T cell subset (Ftc-T cells) in the CSC of patients with
   PerAF, subsequent follow-up data demonstrated that this cell subset is
   closely related to the prognosis of PerAF, and suggested that IFI27 may
   be a functional substance. However, this study has not been further
   verified through vivo experiments. Finally, the mechanism by which
   Ftc-T cells interact with IFI27 in target cells remains unclear and
   necessitates further study.

STAR★Methods

Key resources table

   REAGENT or RESOURCE SOURCE IDENTIFIER
   Antibodies
     __________________________________________________________________

   Rabbit monoclonal anti-IFI27 Abcam ab171919
   Rabbit monoclonal anti-a-SMA Abcam ab244177
   Rabbit monoclonal anti-Col-1 Abcam ab270946
   Rabbit monoclonal anti-GAPDH Abcam ab9485
   Horseradish peroxidase (HRP) -conjugated secondary antibody Abcam
   ab205718
     __________________________________________________________________

   Biological samples
     __________________________________________________________________

   Coronary sinus blood in patients with persistent atrial fibrillation
   Zhongshan Hospital, Fudan university N/A
   Coronary sinus blood in patients with paroxysmal supraventricular
   tachycardia Zhongshan Hospital, Fudan university N/A
     __________________________________________________________________

   Chemicals, peptides, and recombinant proteins
     __________________________________________________________________

   10% male human AB serum Sigma-Aldrich H4522
   IL-2 Sigma-Aldrich I4161
   5′- ethynyl-2′-deoxyuridine Sigma-Aldrich BCK-EDU488
   High-Capacity cDNA Reverse Transcription Kit Thermo Fisher 15596–018
   Nuclear and Cytoplasmic Protein Extraction Kit Beyotime Biotechnology
   NDC-RBB-XFA61F-1
   BCA Protein Quantitative Detection Kit Sigma-Aldrich BCA1
   PVDF membranes Millipore 03010040001
   Endo-Free Plasmid Maxi Kit Omega D6926
   Lipofectamine 3000 reagent Lipofectamine L3000-008
     __________________________________________________________________

   Critical commercial assays
     __________________________________________________________________

   GEXSCOPETM Single-Cell RNA Library Kit Singleron Biotechnologies N/A
   Chromium Next GEM Single Cell 3′ Kit v3.1 10X Genomics PN-1000269
     __________________________________________________________________

   Deposited data
     __________________________________________________________________

   Single-cell RNA-seq raw and analyzed data This paper; Sequence Read
   Archive (SRA) for Human (SRA-Human) PRJNA1108417
     __________________________________________________________________

   Experimental models: Cell lines
     __________________________________________________________________

   Primary rat cardiac fibroblasts Procell N/A
     __________________________________________________________________

   Oligonucleotides
     __________________________________________________________________

   siRNA targeting sequence: siIFI27-1:CTGCAGAGAAGAGAACCAT This paper N/A
   siRNA targeting sequence: siIFI27-2:TCTGGCTCTGCCGTAGTTT This paper N/A
   GAPDH Primer, Forward: GCACGACTTCTTCAAGTCCGCCATGCC This paper N/A
   GAPDHPrimer Reverse: AATCCGTTGACTCCGACCTTC This paper N/A
   Col-1 Primer, Forward: TCTAGACATGTTCAGCTTTGTGGAC This paper N/A
   Col-1 Primer Reverse: TCTGTACGCAGGTGGATTGGTG This paper N/A
   a-SMA Primer, Forward: GTCCCAGACATCAGGGAGTAA This paper N/A
   a-SMAPrimer Reverse: TCGGATAGCGTCAGGA This paper N/A
   IFI27 Primer, Forward: TGCTCTCACCTCATCAGCAGT This paper N/A
   IFI27Primer Reverse: CACAACTCCTCCAATCACAACT This paper N/A
   Recombinant DNA
   pLenti-CMV-MCS- SBP-3Flag-tRFP-F2A-Neo GenePharma N/A
   pLKO.1-TRCcopGFP-2A-PURO GenePharma N/A
     __________________________________________________________________

   Software and algorithms
     __________________________________________________________________

   ImageJ version 10 1.5i Open source [170]https://imagej.nih.gov/ij/
   FlowJo version 10 Open source [171]https://www.flowjo.com/
   Metascape Open source [172]http://metascape.org/gp/index.html
   R version 4.0.2 Open source [173]https://www.r-project.org/
   10x Genomics Cell Ranger Open source [174]https://www.10xgenomics.com/
   [175]Open in a new tab

Resource availability

Lead contact

   Further information and requests for resources should be directed to
   and will be fulfilled by the Lead Contact, wenqing zhu
   (20111210013@fudan.edu.cn).

Materials availability

   This study did not generate new materials.

Data and code availability

     * •
       Single-cell RNA-seq data have been deposited at SRA and are
       publicly available as of the date of publication. Accession numbers
       are listed in the [176]key resources table. This paper analyzes
       existing, publicly available data. These accession numbers for the
       datasets (PRJNA1108417) are listed in the [177]key resources table.
     * •
       All original code has been deposited at Code Ocean and is publicly
       available as of the date of publication. The capsule number is
       listed in the [178]key resources table.
     * •
       Any additional information required to reanalyze the data reported
       in this paper is available from the [179]lead contact upon request.

Experimental model and study participant details

   Human PATICIPANTS: a total 6 subjects for scRNA-seq were divided over
   two cohorts (PerAF-1,PerAF-2, PerAF-3; PSVT-1, PSVT-2,PSVT-3) to
   isolate immune cells from coronary sinus blood (CSB) for scRNA-seq,
   detailed patient information could be found in [180]Table S10.
   Subsequently, 10 patients were enrolled in each cohort, and flow
   cytometric analysis was used to validate the subtype cells, detailed
   patient information could be found in [181]Table S11. Next, we
   prospectively enrolled 80 patients with PerAF who underwent CA for
   subsequent follow-up. CSB samples of all participants were available
   for further flow cytometry analysis, detailed patient information could
   be found in [182]Table S12. Informed consents were obtained from all
   subjects and the experiments with human subjects were approved by the
   Ethics Committees of Zhongshan Hospital.

   CELL LINES: primary rat cardiac fibroblasts were cultured in DMEM-F12
   at temperature: 37°C with 5% CO2.

Method details

Participants and coronary sinus blood collection

   A total of 6 subjects for scRNA-seq were divided over two cohorts for a
   total of 6 participants (PerAF: n = 3, PSVT: n = 3) to isolate immune
   cells from coronary sinus blood (CSB) for scRNA-seq. Subsequently, 10
   patients were enrolled in each cohort, and flow cytometric analysis was
   used to validate the subtype cells identified by scRNA-seq. Next, we
   prospectively enrolled 80 patients with PerAF who underwent CA for
   subsequent follow-up. CSB samples of all participants were available
   for further flow cytometry analysis. All clinical investigations were
   conducted in accordance with the principles in the Declaration of
   Helsinki. The protocol of the study was approved by the Ethics
   Committees of Zhongshan Hospital. Prior to any study procedure, all
   participants provided written informed consent.

The inclusion and exclusion criteria for patients

   The inclusion criteria were as follow: (a) patients with symptomatic
   persistent AF or PSVT who received radiofrequency ablation (RF) for the
   first time at the department of cardiology at the zhongshan hospital,
   Fudan University; (b) age: 40< age <60 years old; (c) All patients
   signed an informed consent form according to the standards of the
   Institution. The exclusion criteria were as follow: (a) patients
   complicated with valvular heart disease, heart failure (HF), diabetes,
   chronic renal failure (CRF), tumor, hematological diseases, autoimmune
   disease, etc; (b) transthoracic echocardiography of left atrial
   diameter (LAD) > 50 mm.

Coronary sinus blood sampling

   Coronary sinus blood (CSB) sampling: participants were in supine
   position, the right femoral vein was punctured and a guide wire was
   inserted. Under the guidance of X-ray, the sheath was placed along the
   guide wire to the inferior vena cava ostium, and then the mapping
   electrode was inserted through the sheath to the distal end of the CS.
   The sheath is then inserted into the CS along the mapping electrode,
   and CS blood sample is finally drawn from the lateral hole of the
   sheath. The 6 mL of whole blood sample was collected in sodium
   citrate–containing cell preparation tubes (CPT Vacutainer; Becton
   Dickinson, Germany) for scRNA-Seq experiment.

CBMCs collection of CSB before single-cell RNA-seq

   For cell populations that can be used for single-cell sequencing, we
   preprocess whole blood samples. 5 mL of blood was collected in sodium
   citrate–containing cell preparation tubes (CPT Vacutainer; Becton
   Dickinson, Germany). The samples were filtered through a 40 um cell
   sieve (Becton Dickinson, Germany) and transferred to a 15 mL centrifuge
   tube. These tubes, containing a Ficoll plug, were centrifuged at 1500 g
   for 20 min at 18°C, the supernatant was discarded, and 5-10 mL of
   erythrocyte lysate (Miltenyi, 130-094-183) was added to the pelleted
   cells. Let stand for 10 min at room temperature, then centrifuge at
   500 g for 10 min at 18°C, discard the supernatant and repeat the above
   steps 1–2 times. The precipitated cells were resuspended and washed
   with 5 mL of DMEM medium, centrifuged at 300 g for 30 min at 18°C, the
   supernatant was discarded, and then resuspended with an appropriate
   amount of medium. The cell suspension concentration and cell viability
   were counted by a Luna cell counter.

Single-cell RNA-seq library preparation

   In brief, the initial step consisted in performing an emulsion where
   individual cells were isolated into droplets together with gel beads
   coated with unique primers bearing 10x cell barcodes, UMI (unique
   molecular identifiers) and poly (dT) sequences. Reverse transcription
   reactions were engaged to generate barcoded full-length cDNA followed
   by the disruption of emulsions using the recovery agent and cDNA clean
   up with DynaBeads MyOne Silane Beads (Thermo Fisher Scientific,
   Germany). Bulk cDNA was amplified using a Biometra Thermocycler T
   Professional Basic Gradient with 96-Well Sample Block (98°C for 3 min;
   cycled 14×: 98°C for 5 s, 67°C for 20 s, and 72°C for 1 min; 72°C for
   1 min; held at 4°C). Amplified cDNA product was cleaned with the SPRI
   select Reagent Kit (Beckman Coulter, USA). Indexed sequencing libraries
   were constructed using the reagents from the Chromium Single Cell 30 v2
   Reagent Kit, as follows: fragmentation, end repair and A-tailing; size
   selection with SPRIselect; adaptor ligation; post-ligation cleanup with
   SPRIselect; sample index PCR and cleanup with SPRI select beads.
   Library quantification and quality assessment were performed using
   Bioanalyzer Agilent 2100 using a High Sensitivity DNA chip (Agilent
   Genomics, USA). Indexed libraries were equimolarly pooled and sequenced
   on two Illumina HiSeq4000 using paired-end 26 × 98 bp as sequencing
   mode by GenomeScan (Leiden, Netherlands).

Single cell RNA library construction and sequencing

   Single-cell suspensions with 13105 cells/mL in concentration in PBS
   (HyClone) were prepared. The P3 library was constructed according to
   Singleron GEXSCOPETM protocol by GEXSCOPETM Single-Cell RNA Library Kit
   (Singleron Biotechnologies). Libraries of other samples were
   constructed by Chromium Next GEM Single Cell 30 Kit v3.1 (10X
   Genomics). Individual libraries were diluted to 4 nM and pooled for
   sequencing. Pools were sequenced on Illumina HiSeq XTen with 150 bp
   paired-end reads to obtain a sequencing depth of approximately 6.5K
   reads/cell and a 70–80% saturation level.

Single-cell RNA-seq data analyses

   Single-cell expression data were processed using the Cell Ranger
   software pipeline (version 3.1.0) provided by 10×Genomics to perform
   quality control (unique molecular identifier count/gene, UMI/gene;
   mitochondrial genes), sample de-multiplexing, barcode processing, and
   single-cell gene counting. After scRNA-seq and aggregating all sample
   data from Cell Ranger, 77946 cells were yielded from sequencing of
   2,251,874,418 PE-reads for further analysis. Sequencing libraries
   showed valid barcodes of 96.5%, sequencing saturation of 85% per
   sample, with an average of 93.6% of mapping reads to the genome. Total
   genes per sample averaged 22,879, with 1,764 genes per cell and 6,036
   mean UMI counts per cell. Sequencing reads were aligned to the human
   reference genome GRCh38 using the Cell Ranger suite with default
   parameters. Principal component analysis (PCA) was performed to reduce
   the dimensionality, and the principal components were used to generate
   clusters by the graph-based algorithm. Cells were visualized using a
   2-dimensional t-distributed stochastic neighbor embedding (t-SNE)
   algorithm. Cell types were identified according to the expression of
   canonical marker genes for each cluster. Single-cell type was
   identified by Single R package with the reference transcriptomic
   datasets ‘Human Primary Cell Atlas’. FindMarkers” function of Seurat
   (version 3.1.1) was used for differential gene expression analysis
   between PerAF and PSVT groups using the Wilcox test. Only those with
   adjusted p value < 0.05 and |log2foldchange| > 0.58 were identified as
   DEGs. A false discovery rate (FDR) correction was jointly applied for
   all genes in the dataset.

Single-cell RNA-seq data analyses

Quality control

   We processed the unique molecular identifier (UMI) count matrix using
   the R package Seurat (version 3.1.1). To remove low quality cells and
   likely multiplet captures, which is a major concern in
   microdroplet-based experiments, we applied a criteria to filter out
   cells with UMI/gene numbers out of the limit of mean value +/− 2-fold
   of standard deviations assuming a Guassian distribution of each cells'
   UMI/gene numbers. Following visual inspection of the distribution of
   cells by the fraction of mitochondrial genes expressed, we further
   discarded low-quality cells where >10% of the counts belonged to
   mitochondrial genes. After applying these QC criteria, single cells
   were included in downstream analyses. Library size normalization was
   performed with Normalize Data function in Seurat[1] to obtain the
   normalized count. Specifically, the global-scaling normalization method
   “LogNormalize” normalized the gene expression measurements for each
   cell by the total expression, multiplied by a scaling factor (10,000 by
   default), and the results were logtransformed.

Dimensionality reduction and clustering

   The most variable genes were selected using FindVariableGenes function
   (mean. function = ExpMean, dispersion.function = LogVMR) in Seurat.
   Principal component analysis (PCA) was performed to reduce the
   dimensionality with RunPCA function in Seurat. Graph-based clustering
   was performed to cluster cells according to their gene expression
   profile using the FindClusters function in Seurat. Cells were
   visualized using a 2-dimensional t-distributed stochastic neighbor
   embedding (t-SNE) algorithm with the RunTSNE function in Seurat.

Markers gene and cells type identification

   We used the FindAllMarkers function (test.use = bimod) in Seurat to
   identify marker genes of each cluster. For a given cluster,
   FindAllMarkers identified positive markers compared with all other
   cells. Then, we used the R package SingleR, a novel computational
   method for unbiased cell type recognition of scRNA-seq, with the
   reference transcriptomic datasets ‘Human Primary Cell Atlas’ to infer
   the cell of origin of each of the single cells independently and
   identify cell types. Differentially expressed genes (DEGs) were
   identified using the FindMarkers Function (test.use = MAST) in Seurat.
   p value <0.05 and |log2foldchange| > 0.58 was set as the threshold for
   significantly differential expression.

Functional and pathway enrichment analysis

   Gene ontology (GO) and the Kyoto Encyclopedia of Genes and Genomes
   (KEGG) pathway enrichment analysis of DEGs were performed by Metascape
   ([183]http://metascape.org/gp/index.html) (version 3.5). Results were
   visualized with the ggplot2 R package
   ([184]https://ggplot2.tidyverse.org/) (version 4.1.3). The
   significantly enriched GO terms and KEGG pathways met the criterion of
   corrected p < 0.05 ([185]supplementary information, [186]STAR methods).

Protein–protein interaction network analysis

   DEGs were imported into Metascape. Protein-protein interaction network
   analysis (PPI) was performed using Metascape following the protocol.

Flow cytometry and cell sorting

   Candidate cell subsets were preliminarily determined according to the
   results of single-cell sequencing. Then, corresponding antibodies were
   selected according to the specific proteins expressed by the cell
   subsets, and the blood samples of participants subsequently enrolled
   were collected for subsequent analysis by FACS analysis using a
   FACSAria II cell sorter (BD Biosciences). Data were analyzed with
   FlowJo software (Tree Star).

Cardiac fibroblasts

   Primary rat cardiac fibroblasts (CFs) were obtained from Procell
   (Wuhan, China). Cells were cultured in DMEM-F12 supplemented with 10%
   fetal bovine serum (FBS). The cells were incubated at 37°C in a
   humidified atmosphere containing 5% CO[2].

Plasmid construction and transfection

   Expression plasmids for IFI27 ([187]NM_001130080.3) were purchased from
   GenePharma (Suzhou, China). The IFI27 sequence was cloned into
   pLenti-CMV-MCS- SBP-3Flag-tRFP-F2A-Neo and pcDNA3.1 (+)-HA vectors. The
   target sequences used for shRNA or siRNA gene-silencing plasmids were
   as follows: siIFI27-1, CTGCA GAGAAGAGAACCAT; siIFI27-2,
   TCTGGCTCTGCCGTAGTTT. The sequences were inserted into
   pLKO.1-TRCcopGFP-2A-PURO (GenePharma, Suzhou, China). The amplified
   vectors were transformed into competent Escherichia coli DH5α cells
   (TSINGKE, Beijing, China) and confirmed by sequencing. The vectors were
   isolated sequentially using an Endo-Free Plasmid Maxi Kit (Omega,
   Norcross, GA, USA). Transient transfection was performed using
   Lipofectamine 3000 reagent (Invitrogen, Carlsbad, CA, USA) according to
   the manufacturer’s instructions.

   For stable cells, 2 × 10^5 CFs were seeded on 6-well plates. When the
   cells had grown to 70–80% confluence, we replaced with fresh DMEM-F12
   (Invitrogen, Carlsbad, CA, USA) medium without FBS and used the IFI27
   or shIFI27 virus to infect the cells. After incubating for 72 h, the
   stable cells were obtained by puromycin screening.

Western blot assay

   Cells were lysed in radio-immunoprecipitation assay (RIPA) buffer
   (Thermo Fisher, MA, USA) containing protease inhibitor
   phenylmethanesulfonyl fluoride (PMSF) (Thermo Fisher, MA, USA). Nuclear
   and cytoplasmic proteins were separated using a Nuclear and Cytoplasmic
   Protein Extraction Kit (Beyotime Biotechnology, Shanghai, China)
   according to the manufacturer’s protocol. The protein concentration was
   quantitatively analyzed using a BCA Protein Quantitative Detection Kit
   (Sigma-Aldrich, Darmstadt, Germany). Protein samples were separated on
   SDS-PAGE gel and transferred to PVDF membranes (Millipore, Billerica,
   MA, USA). The PVDF membranes were blocked with 5% skimmed milk at room
   temperature for 1 h and incubated overnight with the primary antibody
   at 4°C. They were then incubated with secondary antibodies at room
   temperature for 1 h and PVDF membranes were detected using a Meilunbio
   Pico Chemiluminescent Substrate (Meilunbio, Dalian, China). The
   membranes were then observed using a ChemiDoc Touch Imaging System
   (Bio-Rad, Hercules, CA, USA). Primary antibodies were as follows:
   anti-IFI27 (1:1000; Abcam, Boston,USA), anti-a-SMA (1:1000; Abcam,
   Boston,USA), anti-Col-1 (1:1000; Abcam, Boston,USA), and anti-GAPDH
   (1:10,000; Abcam, Boston,USA). GAPDH was used as the loading control.
   Horseradish peroxidase (HRP)-conjugated secondary antibody (1:3000;
   Cell Signaling Technology) was incubated for 1 h at room temperature.
   Super Signal West Femto Substrate Trial Kit (Thermo Fisher, MA, USA)
   was used for chemiluminescence detection. The immunoreactive bands were
   analyzed in triplicate with ImageJ software (National Institutes of
   Health, Bethesda, MD, USA).

Quantitative RT-PCR

   An TRIzol (Thermo Fisher, MA, USA) was used to extract total RNA
   following the manufacturer’s protocol. Reverse transcription PCR and
   qRT-PCR were performed using High-Capacity cDNA Reverse Transcription
   Kit (Thermo Fisher, MA, USA) and SYBR Green Master Mix Kit (Thermo
   Fisher, MA, USA) using the ABI ViiA7 Real-Time PCR system (Thermo
   Lifetech, Carlsbad, CA, USA) under the following conditions: 95°C for
   30 s and then 50 cycles of 95°C for 5 s and 60°C for 30 s.

Co-culture of FCs and Ftc-T cells

   Culture media for Ftc-T cells were composed of RPMI 1640 (Gibco),
   supplemented with 2 mM Ultraglutamine I, 1:100 penicillin/streptomycin
   and 10% male human AB serum (Sigma-Aldrich) (“T cell medium”). One day
   before co-culture, Ftc-T cells were thawed in pre-warmed (37°C) T cell
   medium (human serum was replaced with FBS during thawing) and incubated
   for 15 min with 25 U/mL benzonase (Merck). After washing, cells were
   re-suspended at 2×10^6 cells/mL in T cell medium supplemented with
   150 U/mL IL-2 and cultured overnight at 37°C. Prior to co-culture, CFs
   from earlier passages 3′ were seeded into a 6-well and cultured in
   DMEM-F12, supplemented with 10% FBS in cell culture incubator (37°C,
   CO2). When CFs grew to 60–70% confluence, we co-cultured them with
   Ftc-T cells.

Detection of cells proliferation

   Cell proliferation was examined by 5′- ethynyl-2′-deoxyuridine (Edu,
   10 μM). Cells were transferred into 96-well plates and cultured at 37°C
   overnight. Edu was incubated for 4 h before cells were fixed and
   permeabilized. Click-iT Edu Alexa Flour 488 imaging kit (Life
   Technologies, Eugene, OR, USA) was used according to the manufacturer’s
   instructions. Images were captured at ten randomized fields by using
   light microscopy (Nikon Eclipse Ti-S; Nikon, Tokyo, Japan).

Follow-up

   After discharge, the patients were scheduled for follow-up visits with
   baseline electrocardiogram (ECG), 24 h Holter monitoring and
   transthoracic echocardiography at 3, 6, 12, 18, and 24 months.
   Additional telephonic interviews were conducted regularly. In the case
   of symptoms suggestive of recurrent arrhythmia, additional visits were
   recommended. The first 3 months post ablation were designated as a
   ‘blanking period’. A recurrence of arrhythmia during this time was not
   considered as an efficacy endpoint failure. After blanking period,
   recurrence of arrhythmia is defined as recurrence of AF or atrial
   tachycardia (≥30 s in duration on ECG monitoring). Until 8/2022, all
   patients were followed up for at least 6 months.

Quantification and statistical analysis

Statistical analyses

   All analyses were conducted using R (version 4.1.3, R Foundation for
   Statistical Computing, Vienna, Austria) or SPSS (version 29.0, IBM,
   Armonk, NY, USA). Categorical variables were presented as frequency,
   and continuous variables were described as mean ± SD or median
   (interquartile range, IQR). Continuous variables were compared using
   Student’s t test (normal distribution) or Mann–Whitney U (nonnormal
   distribution) tests. Proportions for categorical variables were
   compared using the Pearson chi-square test. The Fisher exact test was
   used when the data were limited. Statistical tests were 2-tailed, with
   p < 0.05 indicating statistical significance (∗∗∗∗p < 0.001,
   ∗∗∗∼∗∗p < 0.01, ∗p < 0.05).

   Hazard ratios (HRs) and 95% CIs for AF recurrence were estimated with
   Cox proportional hazards regression. Three Cox models were examined for
   outcome. Model 1 was adjusted for LA and duration of AF, model 2 was
   adjusted for potential confounders: age and sex, and Model 3 was
   adjusted for sex and BMI. Tests of linear trend were computed with Cox
   models that treated FTR and FCR as continuous variables. Proportional
   hazards assumptions were assessed with tests based on Schoenfeld
   residuals, and no variables violated the assumption.

   The dose-response relation of AF recurrence with the continuous
   variables FTR and FCR were examined by means of 2 steps. First, we
   tested the dose-response trajectory for nonlinearity by repeating model
   1 after including restricted cubic spline functions of FTR and FCR
   using the Regression Modeling Strategies (rms) package in R (R
   Foundation for Statistical Computing; Vienna, Austria). To test whether
   the shapes of the dose-response trajectories were sensitive to the
   number of knots used, we ran models with 3, 4, 5 knots, respectively,
   and R2 was used to assess power of model. Plots of the dose-response
   trajectories were reviewed for outcome for each model fit, and χ2 tests
   for nonlinearity were performed. After determining the most appropriate
   functional form of the dose-response trajectories, we plotted them for
   each outcome. Dose-response trajectories were plotted for model 1 to
   visualize the influence of adjustment. The model with 5 knots was
   identified with meaningfully different of trajectories and largest
   value of R2, so χ2 tests were performed for models with 5 knots to
   maximize statistical power.

   The time-dependent receiver operating characteristic (ROC) curves were
   analyzed using the survival ROC package in the R platform to determine
   cut-off values with the largest Youden index for FTR and FCR. A
   C-statistic for the Cox model was applied to show the performance of
   FTR and FCR (area under of receiver-operating characteristic curve
   [AUROC]) in predicting the risk of AF recurrence in the cohort. The
   selection of the optimal FTR and FCR cut-off point was based on the
   highest Youden index. Results from all multivariable analyses are
   reported as HR with 95% CIs or coefficients. All statistical parameters
   are indicated in the figure legends or table footnote.

   To explore the association of jointly classified FTR and FCR with AF
   recurrence, we categorized our sample into high and low FTR and high
   and low FCR by cut off value of time-depedent ROC, to create the
   following 4 mutually exclusive groups: low FTR, low FCR (reference
   group); low FTR, high FCR; high FTR, low FCR; and high FTR, high FCR.
   The HR for each jointly classified exposure group associated with AF
   recurrence was then estimated using the model 1. For Kaplan-Meier plot
   and patients at risk, a log rank test was used to compare the AF
   recurrence-free survival among different groups.

Acknowledgments