Multi-omics data reveal the origin of cardiac myxoma

Abstract

Cardiac myxoma, the most common primary heart tumor, remains poorly understood at the molecular level. Here, we combined single-nucleus RNA sequencing, third-generation transcriptomics, and untargeted metabolomics to dissect its origin and pathology. Single-cell analyses demonstrate an endothelial origin driven by aberrant endothelial-to-mesenchymal transition (EndMT), with pseudotime and RNA-velocity tracing a continuum from endothelial-like to mesenchymal-like and metabolically active states. We identify two distinct myxoma subtypes: Subtype 1, marked by MAPK/WNT/EGFR pathway activation, and Subtype 2, characterized by ribosomal and oxidative phosphorylation signatures alongside immune-evasive programs. Third-generation data highlight extracellular matrix remodeling and endothelial signaling, while metabolomics reveal dysregulated purine, nicotinic acid, and nicotinamide metabolism. Notably, MET–PTK2 signaling emerges as a potential driver of tumor initiation and progression. These integrated findings define the cellular architecture and metabolic adaptations of cardiac myxoma and lay the foundation for future interventions.

Introduction

Cardiac myxomas are the most common primary benign heart tumors, usually located in the left atrium¹, accounting for about 50% of all primary cardiac tumors^2,3. Characterized by a mucopolysaccharide-rich extracellular matrix and diverse clinical presentations, myxomas are challenging to diagnose and manage. Patients may be asymptomatic or present with symptoms such as fever, weight loss, myalgia, dyspnea, palpitations, or syncope due to ventricular inflow obstruction or increased atrial pressure. Cerebral and peripheral embolisms are also significant complications⁴. Echocardiography is the primary diagnostic tool, and surgical resection is often recommended to prevent complications, with an excellent prognosis (80%-98% survival) following surgery^5,6,7. For patients unsuitable for surgery, developing molecular targeted therapies is crucial to effectively address their specific medical needs.

Recurrence in atrial myxoma cases is rare (1–3%)⁴, but their potential for malignancy and metastasis remains debated. Although generally considered benign, some myxomas have demonstrated metastatic behavior, spreading to the lungs, bones, soft tissues, and brain⁴. These findings underscore the importance of molecular characterization to unravel the mechanisms underlying these malignant traits and to develop targeted therapies, particularly for patients unsuitable for surgery.

A key challenge in understanding cardiac myxomas lies in determining their precise cellular origins and developmental mechanisms. Despite extensive research, these origins remain unclear, limiting progress in developing effective non-surgical treatments. Clarifying the cellular and molecular foundations of myxomas is essential for developing molecular targeted therapies that could offer alternatives to surgery. Traditional methods, such as immunohistochemistry, electron microscopy, and tissue culture, have provided inconsistent and sometimes contradictory findings regarding the cellular origins of myxomas^8,9,10,11,12. Studies have indicated that cardiac myxomas may differentiate into multiple cell types, including endothelial cells^8,9, fibroblasts¹⁰, and smooth muscle cells^8,10, as well as typical stromal cells. Glandular differentiation is rarely observed^10,12, and neurogenic differentiation has been proposed based on the presence of S100 protein and neuron-specific enolase immunoreactivity in some myxoma cells^10,12. Traditional histopathological methods have been insufficient in conclusively determining the tissue origin or cell lineage of myxomas, and electron microscopy has similarly failed to identify myogenic differentiation, further complicating their characterization¹³. Given these limitations, a comprehensive multi-omics approach, integrating single-cell RNA sequencing, transcriptomics, and metabolomics, is necessary to unravel the molecular heterogeneity of cardiac myxomas. Such advanced techniques offer the potential to definitively characterize the tumor’s cellular origins, delineate the molecular pathways involved in its progression, and identify therapeutic targets. This would significantly enhance our understanding of the tumor’s biology, paving the way for personalized therapeutic strategies and offering new directions in the management of this rare yet clinically important condition.

Our study integrates single-cell nuclear sequencing, third-generation transcriptomics, and metabolomics to provide deep insights into cardiac myxoma. This analysis elucidated the endothelial origins of these tumors and mapped their developmental processes, emphasizing the key roles of endothelial-to-mesenchymal transition (EndMT) and extracellular matrix remodeling in tumor progression. Furthermore, our research uncovered intricate cellular interactions within the tumor microenvironment and highlighted critical metabolic pathways, such as purine, niacin, and nicotinamide metabolism, involved in tumor pathogenesis. These insights are crucial for developing targeted therapies for specific tumor subpopulations. This approach offers valuable insights that may contribute to the development of personalized treatment options, potentially improving clinical management and patient outcomes for cardiac myxoma.

Results

Single-cell transcriptomic atlas of cardiac myxoma and myocardium

In this study, the patients ranged in age from 30 to 76 years, including nine females and eight males. The volume of the myxomas varied between 1.88 and 99.49 cm³. All patients exhibited varying degrees of cardiac dysfunction, with NYHA functional classifications ranging from Class II to IV, accompanied by varying levels of abnormal myocardial injury biomarkers (Table 1 and Table S1). Figure 1A displays the hematoxylin and eosin (HE) staining results of three cardiac myxoma samples. The images reveal dispersed spindle-shaped, polygonal, or stellate cells embedded within the matrix. The tumor cells are arranged either individually or in clusters, presenting a distinctive vascular-like pattern. This observation suggests that endothelial-related alterations may play a crucial role in the pathogenesis of cardiac myxomas.

Fig. 1: Single-cell transcriptomic of cardiac myxoma.

A Morphological feature of myxoma in H&E staining. B TSNE dimensionality reduction of cellular identity. C Cell type composition per sample. D Heatmap of representative marker gene expression for each cell type. E Heatmap showing average expression of signature genes for each cell type. F Differences in cell proportions between normal and myxoma samples.

Table 1 The key clinical characteristics of the patients

Single-cell/nucleus RNA sequencing (sc/snRNA-seq) was performed on eight cardiac myxoma samples from six patients, along with five matched cardiac tissue samples and two normal heart tissue samples. To investigate the spatiotemporal heterogeneity of cardiac myxomas, we collected samples from the stalk and main body of the tumor in the same patient: P2 (P2T1, stalk; P2T2, main body). After filtering out low-quality cells, the final dataset comprised 99,248 single cells, with a median of 1445 genes detected per cell. The number of cells per sample ranged from 2849 to 13,423 (Fig. 1B, C). Dimensionality reduction and batch effect correction were applied to all samples, followed by clustering using the Leiden algorithm at a resolution of 1.2, generating a distance matrix from the neighborhood graph of nearest neighbors and observations. Based on established tissue-specific marker genes, 12 major cell types were identified: adipocytes (ADIPOQ, PPARG, FABP4, LPL), B cells (FCRL5, IGKC, PTK2B), cardiomyocytes (MYH6, RYR2, TNNI3), endothelial cells (VWF, EGFL7, EMCN), fibroblasts (DCN, COL1A1, COL6A3), macrophages (CD163, CD163L, MRC1), mast cells (CPA3, KIT, MS4A2), smooth muscle cells (ACTA2, MYL6, TAGLN), T cells (CD96, CD247, BCL11B), cycling cells (EZH2, TOP2A, MKI67), and myxoma cells (PAPPA, BMPER, WWTR1, CDH11) (Fig. 1D, E, Table S2). In the normal tissue samples, cardiomyocytes and fibroblasts were the predominant cell types, whereas in myxoma samples, tumor cells were the most abundant, comprising 29.5% to 67% of all cells, with minimal cardiomyocyte presence (Fig. 1C, Table S2). Statistical analysis showed a significant increase in immune cells (B cells, T cells, macrophages) in myxoma samples compared to normal tissues (Fig. 1F), suggesting a robust immune response triggered by the tumor. Conversely, the proportion of endothelial cells and fibroblasts was markedly reduced in myxoma samples (Fig. 1F), likely reflecting the disruption of normal tissue architecture and function due to the tumor’s invasive nature. This shift in cellular composition highlights the dynamic interaction between tumor cells and the immune system, emphasizing the myxoma’s capacity to modify the local cellular environment, potentially aiding its survival and growth by modulating immune responses and compromising normal tissue integrity.

Furthermore, we compared snRNA and scRNA sequencing results. The results showed no significant differences in the number of cells detected or the number of genes between the two methods (Fig. S1A, B). However, a significant reduction in macrophages was observed in scRNA-seq, particularly in the myxoma sample (P1T2) (Fig. S1D–H), indicating that snRNA-seq provides improved sensitivity for detecting specific cell populations, such as macrophages, under these sequencing conditions or sample states. Additionally, significant transcriptomic differences were detected between samples from different spatiotemporal regions of the same patient. (P2T1 [tumor stalk] vs. P2T2 [tumor body]) (Fig. S2A–C), highlighting the significant transcriptomic variability across different regions within the same cardiac myxoma, emphasizing the temporal and spatial heterogeneity characteristic of these tumors. In the stalk region, tumor cells express high levels of the endothelial marker ANGPT1, whereas in the body region, they upregulate mesenchymal markers VIM and COL3A1 (Fig. S2C, D), creating a clear endothelial-to-mesenchymal gradient. Moreover, robust expression of cell-cycle genes such as CCND3 reveals intrinsic proliferative potential (Fig. S2C, D).

The characteristics of copy number variants (CNV)

To further characterize cardiac myxoma cells, we conducted a comprehensive analysis of copy number variations (CNVs) within these tumors. Based on CNV profiles, cells were classified as either normal or tumor cells (Fig. 2A, Fig. S3). Our results revealed significant intratumoral genomic heterogeneity, uncovering subclonal structures and distinct evolutionary relationships among subclones (Fig. 2B, C, Fig. S3). Within the tumor of a single patient, all neoplastic cells exhibited a uniform, dominant CNV landscape (Fig. 2A, B, Fig. S3), providing strong evidence for clonal expansion from a single transformed precursor. A phylogenetic tree based on CNVs demonstrated marked heterogeneity between different tumors (Fig. 2D).

Fig. 2: Copy number variation (CNV) in cardiac myxoma cells.

A CNV landscape in P2T1. B CNV landscape of clonal composition in myxoma cells from P2T1. C Clonal evolution tree of myxoma cells based on CNV in P2T1. D Sample tree based on CNV. E Pathway alterations detected in more than 10 subclones, ranked by frequency of occurrence. F OncoHeat plot displaying CNV landscape across all myxoma samples.

Additionally, we observed abnormal activation of the MET_ACTIVATES_PTK2_SIGNALING and ECM_PROTEOGLYCANS pathways in most tumor subclones (Fig. 2E), suggesting that alterations in these pathways play a crucial role in the tumorigenesis and progression of myxomas. Specifically, the MET_ACTIVATES_PTK2_SIGNALING pathway is implicated in tumor initiation and progression by promoting tumor cell proliferation, migration, and invasion, while ECM_PROTEOGLYCANS may contribute to the formation of the tumor’s myxoid components.

Furthermore, CNV analysis across all tumor samples indicated that shared CNVs between different tumors are extremely rare (Fig. 2F). High-frequency CNVs included deletions on chromosome 19 (p13.3–p13.11, q13.12–q13.43, q13.2–q13.43) and amplifications on chromosome 1 (p36.33–p21.2, p33–p21.1) and chromosome 9 (p24.3–q21.33, p24.3–q21.32) (Fig. 2F). These findings suggest that these CNVs may play a critical role in the pathogenesis and progression of cardiac myxomas, underscoring the need for further genomic studies to identify potential therapeutic targets. Interestingly, we identified deletions on chromosome 17 (q22–q24.2) in two patients, a CNV linked to Carney complex-associated myxomas^14,15,16. This finding underscores the genetic complexity of cardiac myxomas and hints at possible shared pathogenic pathways among different myxoma subtypes

The characterization and cell lineages of cardiac myxoma

Next, we focused on characterizing the transcriptomes of myxoma cells defined by genes significantly overexpressed within these tumor cells. We first evaluated the expression levels of previously reported myxoma-related markers at the single-cell level. Among these, SOX5, SOX6, GATA4, and CDH11 demonstrated high specificity, while PDGFRA, CD44, TGFBR2, NFATC2, and MMP2 exhibited lower specificity (Fig. S4A). Additionally, our study revealed a series of complex gene expression changes critical to tumor development (Fig. S4A, B, Table S2). Specifically, we observed abnormal expression of phosphodiesterase (PDE) family genes, including PDE3A, PDE4B, and PDE4D, which were associated with increased tumor growth and survival. Moreover, MSI2, a key regulator of mRNA stability and translation, was identified as essential for maintaining stem cell characteristics and promoting differentiation. The activity of SOX family transcription factors (SOX4, SOX5) and hepatocyte growth factor (HGF) was found to significantly influence tumor cell proliferation and invasion. In addition, TCF7L2, a key transcription factor in the Wnt/β-catenin signaling pathway, was shown to play a pivotal role in determining cell fate, promoting growth, and guiding differentiation. These findings provide valuable insights into the transcriptional landscape of cardiac myxomas and highlight potential therapeutic targets involved in tumor progression.

Our analysis revealed that genes associated with cardiac myxoma were significantly enriched in biological processes essential for epithelial cell proliferation and migration, as well as those involved in vascular development, MAPK signaling, WNT signaling, and cell adhesion (Fig. S4C). Given the shared lineage between epithelial and endothelial cells and the cellular composition of the heart, these findings may reflect underlying endothelial biological activities. Further exploration of endothelial-related processes indicated substantial enrichment in pathways regulating endothelial cell proliferation, differentiation, and endothelial-mesenchymal transition (EndMT) (Fig. 3A). These results lend strong support to the hypothesis that cardiac myxoma originates from endothelial cells¹⁷ and suggest that endothelial dynamics play a critical role in the tumor’s pathogenesis. This provides a novel perspective on the molecular mechanisms driving cardiac myxoma, emphasizing the importance of endothelial cell behavior and signaling pathways in the development and progression of the tumor.

Fig. 3: The characterization and cell lineages of cardiac myxoma.

A Epithelial cell-related biological processes in myxoma. B Heatmap showing the expression of endothelial-to-mesenchymal transition, fibroblast, and endothelial cell-specific markers. C t-SNE plot showing the expression of endothelial and mesenchymal-related markers. D Immunohistochemical detection of endothelial cell (CD34, CD31/PECAM1), muscle cell (desmin), and fibroblast (SMA) specific marker expression. E Pseudotime analysis showing the evolution from endothelial cells to myxoma cells. F RNA velocity–inferred developmental trajectory.

Specifically, EndMT contributes to the remodeling of the extracellular matrix (ECM), enhancing collagen deposition, which is consistent with the mucoid characteristics of cardiac myxoma^18,19. Additionally, we observed abnormal activations of crucial pathways that induce EndMT, including TGF-β and Wnt/β-catenin pathways^18,20 (Fig. S4D, Table S3). Our evaluation of cardiac myxoma gene expression revealed that these tumor cells specifically express key regulators that promote the transition from endothelial cells to mesenchymal-like cells, such as WWTR, TCF7L2, PTK2, and ENG (Fig. 3B). Additionally, myxoma cells express stromal cell (fibroblast) markers including vimentin (VIM) and fibronectin 1 (FN1), as well as endothelial-related markers such as endoglin (ENG) and platelet/endothelial cell adhesion molecule 1 (PECAM1) (Fig. 3B). This distinct gene expression profile highlights the complex cellular identity of myxoma cells, reflecting their origin in endothelial-to-mesenchymal transition (EndMT) and their active role in extracellular matrix (ECM) production and remodeling. The presence of EndMT regulators supports the hypothesis that myxoma cells originate from endothelial cells undergoing this transition, while the expression of both fibroblast and endothelial markers underscores the hybrid nature of these cells, contributing to the myxoid features characteristic of cardiac myxomas. The ECM remodeling facilitated by these cells involves significant collagen deposition, aligning with the tumor’s mucoid appearance and invasive properties.

Based on the expression of specific endothelial cell markers, we characterized myxoma cells as negative for von Willebrand factor (VWF) and kinase insert domain receptor (KDR or VEGFR2), yet positive for endoglin (ENG) and platelet/endothelial cell adhesion molecule 1 (PECAM1 or CD31) (Fig. 3C). This distinctive marker profile underscores the endothelial origin and phenotypic attributes of myxoma cells. The absence of VWF and KDR indicates a reduction in typical endothelial functions, while the expression of ENG suggests activation of endothelial-mesenchymal transition (EndMT) by TGF-β. The persistence of PECAM1 indicates the retention of certain endothelial characteristics. To further investigate the cellular origin of cardiac myxomas, we conducted immunohistochemical analyses using markers specific to various cell types, including neuroendocrine (S100, NSE200, SOX9), fibroblast (SMA), myocyte (desmin), and endothelial (CD34, CD31) cells (Fig. 3D and Fig. S5A). Our findings revealed consistent expression of the endothelial markers CD34 and CD31 across all myxoma cells, whereas markers for other cell types were either absent or only sporadically positive. CD34 is a key marker of endothelial cells, commonly expressed on vascular endothelial and hematopoietic progenitor cells, and helps distinguish endothelial cells from mesenchymal stem cells. Combined with PECAM1 (CD31) and ENG, it confirms endothelial origin. PDGFRA, though typically linked to mesenchymal cells, can appear in endothelial cells during endothelial-mesenchymal transition (EndMT) in tumors. The co-expression of CD34, PECAM1, ENG, and PDGFRA forms a strong molecular signature of endothelial origin, highlighting endothelial cell plasticity and their role in cardiac myxoma development.

To further delineate the characteristics of myxoma cells, we conducted a comparative transcriptomic analysis between myxoma cells and endothelial/fibroblast cells in cardiac myxoma samples. Our findings revealed that tumor-associated signature molecules were significantly overexpressed in myxoma cells relative to endothelial and fibroblast cells, underscoring the unique tumorigenic properties of myxoma cells (Fig. S5B). Concurrently, markers specific to endothelial and stromal (fibroblast) cells were significantly downregulated, indicating a loss of their typical cellular characteristics (Fig. S5C). This genomic profile suggests that while myxoma cells derive from endothelial cells undergoing endothelial-mesenchymal transition (EndMT), they acquire distinctive tumorigenic traits that differentiate them from their non-tumorigenic counterparts. Furthermore, myxoma cells exhibited significant upregulation of genes involved in fibroblast differentiation processes and endothelial-related processes when compared to endothelial cells (Fig. S5D). This indicates that myxoma cells retain some endothelial and fibroblast characteristics. Conversely, when compared to fibroblasts, myxoma cells predominantly showed enrichment in endothelial-related biological processes (Fig. S5E). This dual enrichment highlights the hybrid nature of myxoma cells, which amalgamate endothelial and mesenchymal properties. Such a hybrid nature facilitates their role in the production and remodeling of the extracellular matrix (ECM), contributing to the characteristic myxoid features of cardiac myxomas.

To further elucidate the relationship between endothelial and myxoma cells, endothelial cells were stratified into three distinct subtypes (Fig. S6A), each of which consistently maintained high expression of canonical endothelial markers—VWF, PECAM1, KDR, and ENG (Fig. S6C). Cluster Endo1 was predominantly derived from normal samples and exhibited a transcriptional signature indicative of enhanced biosynthetic and metabolic activity. It showed high expression of canonical endothelial markers ID1, ID3 and RNASE1 (Fig. S6B), together with upregulation of ribosomal proteins (RPL and RPS families), the eukaryotic initiation factor EIF1, and mitochondrial electron-transport components (COX family and ATP synthase ATP5 series) (Table S4). These features suggest that Endo1 represents a metabolically “privileged” endothelial subtype tasked with sustaining vascular barrier function and cellular homeostasis through elevated protein synthesis and energy production. Cluster Endo2 was predominantly derived from myxoid-like samples (Fig. S6D) and co-expressed key regulators of endothelial-to-mesenchymal transition, including ZEB2, SMAD3, SPARC, and ENG, alongside the myxoma-associated transcription factor MECOM (Fig. S6B, Table S4). This combined expression profile implies that Endo2 occupies an intermediate state between a quiescent endothelial phenotype and a mesenchymal or myxoid identity. Cluster Endo3, the smallest population, displayed a lymphatic endothelial program with pronounced expression of PROX1, NRP2, and the chemokine CCL21 (Fig. S6B, Table S4).

Next, we reconstructed developmental trajectories for endothelial and myxoma cells using pseudotime analysis, which suggested a direct lineage transition from endothelial cells into myxoma clusters (Fig. 3E, Fig. S7). To explore this further, we applied scVelo to infer RNA velocity across endothelial cells, myxoma cells, and fibroblasts. On the dimensionality reduction plot (Fig. 3F), fibroblasts (blue points) formed a discrete branch whose velocity vectors remained largely confined within that cluster, with no apparent extension toward the myxoma clusters (red points). In contrast, endothelial cells (cyan points) exhibited predominantly unidirectional streamlines projecting into the myxoma clusters, supporting a developmental trajectory from endothelium to tumor. Together with our CNV and marker-expression data, these results further indicate that cardiac myxoma originates from an endothelial lineage rather than from fibroblast transdifferentiation. Furthermore, we validated our key findings in an independent cardiac myxoma dataset, which yielded concordant results (Figs. S8 and S9).

Finally, pathway enrichment analysis uncovered a stage-dependent phenotypic shift. Early-stage myxoma cells (P2T1, tumor pedicle) were enriched for vascular endothelial–related pathways, including vasoconstriction and circulatory system processes (Fig. S2E, F). By contrast, late-stage samples (P2T2, tumor body) displayed a more pronounced mesenchymal phenotype, with markedly elevated VIM and COL3A1 expression (Fig. S2C). Collectively, our data are consistent with a model in which a single transformed endothelial cell gives rise to cardiac myxoma and gradually acquires mesenchymal-like features during tumor progression.

The heterogeneity and dynamic development of Myxoma cells

To uncover the distinct cellular states within myxoma tumors and identify the key transition events and signaling pathways that drive their evolution from origin to more advanced stages, we extracted tumor cells to construct pseudo-temporal developmental trajectories. We classified myxoma cells into three distinct states based on their developmental trajectories (Fig. 4A, Fig. S10). State1 cells retain marked endothelial characteristics, as indicated by the significant expression of endothelial markers such as PECAM1 and ENG, in addition to mesenchymal markers like COL1A2, VIM, and DCN (Fig. 4B). This state is distinguished by the activation of the TGFB and MET pathways, along with matrix remodeling activities, including collagen synthesis and degradation, signaling the early stages of cellular transition and myxoma formation (Fig. 4E). As the cells progress to state2 and state3, there is a significant reduction in both endothelial (PECAM1, ENG) and mesenchymal (COL1A2, VIM, and DCN) markers (Fig. 4C, D), signifying a progressive loss of endothelial characteristics with the development of the myxoma. State2 is marked by a shift toward enhanced material transport and metabolic activities (Fig. 4F), suggesting an adaptation phase where cells adjust their metabolism and transport mechanisms to support tumor growth and survival. Conversely, state3 cells enter a relatively quiescent phase, showing suppressed biological activities despite the activation of some tumor-related pathways (Fig. 4G).

Fig. 4: Dynamic progression of cardiac myxoma cells.

A Pseudotime analysis showing three distinct states of myxoma cells. B–D Volcano plot showing state-specific differentially expressed genes in different states of myxoma cells. E–G Enrichment analysis of state-specific biological processes in different states of myxoma cells.

The subtypes of cardiac myxoma

In clinical settings, cardiac myxomas are categorized into two main gross anatomical types: a solid and smooth mass type, and a soft and friable gelatinous type, each associated with distinct clinical presentations²¹. Patients with solid myxomas typically exhibit symptoms of congestive heart failure, whereas those with gelatinous myxomas are more susceptible to cerebral and peripheral embolism²¹. To further dissect the molecular heterogeneity of cardiac myxoma, we extracted myxoma cells and integrated them with previously published single-cell myxoma profiles for joint reclustering, resolving two predominant cell subtypes (Fig. 5A). In the external GSE265921 cohort, both subtypes were detected in every tumor, with Subtype 2 showing relative enrichment, whereas in our primary dataset Subtype 2 predominated only in sample P1T2 and Subtype 1 in the remaining tumors (Fig. S11B, C). These variable proportions likely reflect sample-specific microenvironmental influences or genuine biological diversity rather than technical or batch-effect artifacts. Importantly, the consistent presence and differential dominance of Subtype 1 or Subtype 2 across tumors underscore their identity as two biologically meaningful cardiac myxoma subtypes.

Fig. 5: Identification of myxoma subtypes.

A t-SNE plot showing the subtypes of cardiac myxoma. B Dot plot showing the differentially expressed genes in different myxoma subtypes. C Dotplot showing KEGG signaling pathways associated with sub1. D Dotplot showing KEGG signaling pathways associated with sub2. E t-SNE embedding showing sub1-specific expression markers. F t-SNE embedding showing sub2-Specific expression markers. G Heatmap of top 10 differentially expressed genes between myxoma subtypes 1 and 2.

Subtype 1 was defined by strong upregulation of oncogenic drivers—PVT1, CCND3, MLLT10, SETBP1, and BCAS3 (Fig. 5B, E, G, Table S5)—alongside enrichment of cancer-related pathways (e.g., prostate and endometrial carcinoma) and pronounced activation of the MAPK, WNT, FOXO, EGFR, and ERBB signaling cascades (Fig. 5C), consistent with a protumorigenic phenotype. Subtype 2 exhibited a hyperproliferative, stress-resistant transcriptional program characterized by overexpression of ribosomal proteins (RPL18A, RPS10, RPS5, RPL39) and translation elongation factors (EEF1B2, EEF1G, EEF2), together with elevated mitochondrial respiratory chain subunits (NDUFA4, NDUFB8, COX4I1, COX5B, ATP5F1E) to reinforce oxidative phosphorylation (Figs. 5B, D, F, G, Table S,5). This subtype also activated interferon-stimulated genes (IFITM2/3, IFI27), upregulated antigen-presentation and complement-regulatory molecules (HLA-C, CD59), and enhanced CD63-mediated exosome secretion and cell–matrix interaction signaling, implying that Subtype 2 couples metabolic reprogramming with multilayered immune-regulatory strategies to drive tumor progression and evade immune surveillance. Furthermore, we overlaid subtype annotations onto the pseudotime trajectory (Fig. S11A). Both Subtype 1 and Subtype 2 cells span the full differentiation continuum rather than occupying discrete trajectory segments, demonstrating that these represent two coexisting tumor phenotypes within a unified developmental landscape.

The tumor microenvironment of cardiac myxoma

To gain a comprehensive understanding of the immune microenvironment in cardiac myxoma, we analyzed primary immune cells, particularly T cells and macrophages, in both myxoma and normal tissue samples. The analysis revealed that the predominant macrophage subtypes in myxoma were M2 macrophages (CD163, MRC1, IL2RA) and monocytes (CSF3R, FCN1, CD14), with a significant increase in their absolute numbers within myxoma tissues (Fig. 6A–E), suggesting that myxoma promotes M2 polarization to support tumorigenesis. The T cell subtypes in myxoma included CD4+ (CCR7, SELL, LEF1), CD8+ (CD8A, CD44, ITGAL), cytotoxic T cells (CCL5, CD69, GZMA), NK cells (NCR1, KLRD1, KLRC1), and Treg cells (NRP1), all of which were significantly elevated in myxoma (Fig. 6F–J), indicating a complex immune landscape potentially contributing to tumor progression.

Fig. 6: Immune microenvironment landscape of cardiac myxoma.

A t-SNE plot showing macrophage subtypes. B t-SNE plot showing the distribution of macrophages organized by sample type. C The number of macrophage subtypes in normal and myxoma samples. D Percentage distribution of macrophage subtypes in each sample. E Expression of specific markers for different macrophage subtypes. F t-SNE plot showing T cell subtypes. G t-SNE plot showing the distribution of T cells organized by sample type. H The number of T cell subtypes in normal and myxoma samples. I Percentage distribution of T cell subtypes in each sample. J Expression of specific markers for different T cell subtypes.

The cell–cell interaction between myxoma and tumor microenvironment cells

To elucidate the intricate interactions among different cell types within the tumor microenvironment, we conducted an extensive analysis of intercellular communication. Results showed significant interactions between mucinous tumor cells and other cellular components, including macrophages, endothelial cells, fibroblasts, cardiomyocytes, T cells, and B cells. Our findings uncover a complex molecular network that plays a critical role in the activation of signaling pathways, including endothelial cell migration, TGF-β, Notch, G-protein-coupled signaling, and inflammatory chemokine signaling (Fig. 7A, B). Further research delved into the interactions between mucinous tumor cells and endothelial and stromal cells. We identified a substantial network of signals associated with vascular endothelial activity, centered around VEGFA and involving the VEGFA-KDR/NRP1/FLT1/NRP2 axis (Fig. 7B). These signals play a critical role in both physiological and pathological processes within the tumor microenvironment, including angiogenesis, cell migration, and proliferation. Additionally, our findings suggest that amyloid precursor protein (APP) may contribute to the abnormal accumulation of amyloid-β (Aβ) through its interactions with CD74, SORL1, TNFRSF21 (DR6), and PLXNA4 (Fig. 7B, C). This mechanism not only plays a role in the pathological progression of Alzheimer’s disease but may also be linked to the formation of mucinous components in mucinous tumors.

Fig. 7: The cell–cell interaction between myxoma and tumor microenvironment cells.

. (A) Heatmap showing the frequency of cell-cell communication events. (B) Cell communication network showing key ligand-receptor interactions. (C) Ligand-receptor interactions between myxoma cells and stromal cells. (D) Ligand-receptor interactions between myxoma cells and immune cells.

We also discovered that interactions between ADGRG6 and COL4A1, COL4A2, and LAMA2 (Fig. 7B, C) primarily regulate the structure and function of the basement membrane. Abnormal interactions in this network may lead to excessive deposition of type IV collagen and laminin, subsequently affecting tissue structural stability. Moreover, JAG1, through its interactions with NOTCH3 and NOTCH4 (Fig. 7B, C), activates the Notch signaling pathway, a process critical in endothelial-to-mesenchymal transition (EndMT). We also observed that TGFB1 and TGFB2, by interacting with TGFBR3 (Fig. 7B, C), regulate extracellular matrix deposition and tissue remodeling, potentially promoting cell invasion and metastasis in tumors. This enhanced intercellular communication facilitates complex interactions among endothelial cells, fibroblasts, immune cells, and mucinous tumor cells. Notably, we detected the activation of PECAM1-CD38 signaling between circulating cells and mucinous tumor cells (Fig. 7C), which is known to contribute to the transformation of vascular endothelial cells into myofibroblast-like cells.

Within the patient’s immune microenvironment, we observed frequent communication between mucinous tumor cells, macrophages, and T cells. This interaction is characterized by the activation of chemokine signaling pathways, such as the CXCL12-CXCR4 axis (Fig. 7B, D), which plays a significant role in tumor cell migration and metastasis. Additionally, the regulatory roles of DPP4 and CXCL14 (Fig. 7B, D) may influence the aggressiveness and metastatic potential of the tumor within the tumor microenvironment. These complex signaling networks are crucial in determining the behavior and pathological characteristics of mucinous tumor cells, highlighting their importance as potential therapeutic targets.

Integrative Third-Generation Transcriptomics and Untargeted Metabolomics Reveal Unique Signatures of Cardiac Myxoma

We employed third-generation integrated transcriptomic and untargeted metabolomic sequencing to further delineate the molecular landscape of cardiac myxoma, thereby unveiling its unique molecular characteristics. In alignment with single-cell sequencing data, our differential gene expression analysis revealed significant overexpression of CDH11, TIMP1/3, PECAM1, MMP2, and PED1A in myxomas (Fig. 8A, Figure S12A). Conversely, VWF and DCN were significantly downregulated compared to myocardial tissue (Fig. 8A). Myocardial tissue showed prominent overexpression of genes related to cardiac function such as MYL7, MYL4, MYL5, MB, TNNI3, and TPM1 (Figure S12A), which were expressed at very low levels in myxomas. Enrichment analysis revealed activation of NF-κB and ERK pathways, in addition to immune-related biological processes (Figure S12B). Notably, a biological process related to endothelial cell proliferation was also observed (Fig. 8B), which corroborates our single-cell data and supports the hypothesis that myxomas may originate from endothelial cells. In addition, we observed the frequent occurrence of the EFEMP1:CCDC85A fusion gene in myxomas compared to normal heart tissues (Figure S12C). The EGF-containing fibrin-like extracellular matrix protein 1 (EFEMP1) is integral to fibrogenesis and crucial for the development and repair of connective tissues. On the other hand, Coiled-Coil Domain-Containing 85A (CCDC85A) is implicated in cell cycle regulation and signaling processes. The fusion of EFEMP1 with CCDC85A may lead to aberrant cellular behaviors, including disrupted cell signaling, compromised cell adhesion, and abnormal tissue development. These alterations are potentially critical in the pathogenesis of cancerous or fibrotic diseases.

Fig. 8: Characteristics of myxoma based on third-generation transcriptomics and Metabolomics.

. (A) Volcano plot showing differentially expressed genes between myxoma and normal samples. (B) Biological function enrichment analysis of myxoma-specific expressed genes. (C) Biological function enrichment analysis of differentially expressed metabolites. (D) Venn diagram showing significantly differentially expressed purine metabolism-related genes in myxoma single-cell data. (E-F) t-SNE embedding of cell-type distribution and purine metabolism scores. (G-H) t-SNE embedding of key enzymes NNMT and NAMPT in niacinamide metabolism.

Next, we investigated the distinct metabolic signatures of myxomas, revealing that the metabolic characteristics of myxoma tissues are significantly different from those of normal cardiac tissues, indicative of unique biological processes. Our comprehensive analysis identified 1,268 high-quality metabolites; among these, 210 were significantly upregulated, while 132 were significantly downregulated in myxoma tissues (Figure S12D). Enrichment analysis revealed these metabolites predominantly participate in several key metabolic pathways: purine metabolism, arginine and proline metabolism, nicotinate and nicotinamide metabolism, and histidine metabolism (Fig. 8C).

Purine metabolism, critical for nucleic acid synthesis, is central to DNA and RNA production. Enhanced purine metabolism can lead to increased cell proliferation and resistance to apoptotic signals. Conversely, disruptions in arginine metabolism can impair immune cell functions, potentially enabling tumors to evade immune surveillance. Proline, crucial for collagen synthesis, also plays a role in cellular stress responses. The elevated metabolism of nicotinate and nicotinamide, precursors of NAD + , aids in meeting increased energy demands and combating oxidative stress. Additionally, the histidine metabolic pathway generates bioactive molecules such as nitric oxide (NO), which may facilitate tumor proliferation, invasiveness, and immune evasion by stimulating angiogenesis and modifying immune responses through the upregulation of enzymes like nitric oxide synthase (NOS). These biological processes contribute to the development of myxoma. In addition, we evaluated metabolic pathways at the single cell transcriptomic level, focusing on myxoma cells. Our analysis revealed pronounced expression of genes associated with purine metabolism (Fig. 8D–F) and nicotinate and nicotinamide metabolism (Fig. 8G-H). Specifically, abnormalities in purine metabolism were predominantly linked to the upregulation of phosphodiesterases (PDEs), which are crucial for purine nucleotide signal transduction. This upregulation primarily affects the degradation of cyclic adenosine monophosphate (cAMP) and cyclic guanosine monophosphate (cGMP). Moreover, the dysregulation of nicotinate and nicotinamide metabolism was primarily due to significant upregulation of NNMT (nicotinamide N-methyltransferase) and NAMPT (nicotinamide phosphoribosyltransferase). NNMT plays a direct role in the methylation of nicotinamide, impacting the recycling and metabolism of NAD+ (nicotinamide adenine dinucleotide). NAMPT is a key enzyme that facilitates the conversion of nicotinamide mononucleotide (NaMN) to NAD+. The dysregulated expression of these genes is critical for maintaining cellular NAD+ levels and influences various biological processes, including energy metabolism, DNA repair, and cell signaling transduction.

Discussion

Our study is the first to comprehensively analyze cardiac myxomas using multi-omics integration, including single-nucleus RNA sequencing (snRNA-seq), third-generation transcriptomics, and metabolomics. This approach has provided valuable insights into the transcriptional landscape, cellular heterogeneity, and dynamic developmental processes of cardiac myxomas, shedding light on aspects of their biology that were previously obscured by less sensitive techniques, and offering a precious dataset for understanding this rare disease.

Research into cardiac myxomas has faced significant challenges, particularly in unraveling their origins and developmental mechanisms. Despite extensive investigations, the exact origins of these tumors remain elusive^8,9,10,11,12. There is ongoing debate regarding whether cardiac myxomas originate from endothelial cells, mesenchymal stem cells, or potentially other cell types, highlighting the complexity and multifactorial nature of their developmental origins. Understanding the cellular and molecular basis of cardiac myxomas is crucial for developing effective targeted therapies.

Accumulating evidence increasingly supports the hypothesis that cardiac myxomas arise from embryonic endocardial endothelial cells. Morphological analyses reveal that cardiac myxomas exhibit a strong resemblance to endothelial cells, particularly in their vascular-like structures^17,22,23. Immunohistochemical studies, including the expression of markers such as CD34, vimentin, and Notch1, further support this view^17,22,23. Molecular analyses indicate that myxoma cells produce endothelial-specific factors like endothelin-1, while lacking mesenchymal-associated factors^17,22,23, challenging the mesenchymal stem cell hypothesis. The presence of vessel-like formations and the consistent expression of endothelial markers in myxomas suggest an origin linked to embryonic endothelial cells.

Further supporting this hypothesis, myxoma cells also express markers associated with endothelial-to-mesenchymal transition (EMT), such as Sox9 and alpha-smooth muscle actin (α-SMA)^24,25,26, suggesting that these cells may derive from embryonic endothelial cells that underwent EMT but were arrested in a transitional state. This interpretation is consistent with the vascular-like morphology observed in myxomas and the expression of EMT-related markers^24,25,26, reinforcing the notion that the tumor originates from developmental remnants of endothelial cells that failed to fully differentiate. While mesenchymal stem cells are multipotent and capable of differentiating into various cell types, including endothelial cells, this differentiation requires specific microenvironmental cues that are not present in cardiac myxomas²⁴. The consistent endothelial-like phenotype across different myxoma cases suggests a direct endothelial lineage rather than secondary differentiation from mesenchymal stem cells. Detailed morphological analyses have demonstrated structural similarities between myxomas and endothelial tissues²⁵, further supporting the endothelial origin hypothesis and weakening the mesenchymal hypothesis.

Some studies have reported the expression of cardiomyocyte-specific transcription factors like Nkx2.5 and GATA-4 in myxoma cells^13,24. However, these factors are more indicative of early embryonic endocardial differentiation than mesenchymal stem cell activity. This aligns with the hypothesis that myxoma cells represent endothelial cells that underwent partial differentiation during early cardiac development, rather than differentiation from mesenchymal stem cells.

Taken together, the body of evidence strongly favors the hypothesis that cardiac myxomas originate from embryonic endothelial cells that have retained their differentiation potential and exhibit characteristics of arrested endothelial-to-mesenchymal transition^13,27. This theory provides a coherent explanation for the tumor’s endothelial-like features, consistent expression of endothelial markers, and vascular-like morphology²⁴. In contrast, the mesenchymal stem cell origin hypothesis is weakened by the lack of mesenchymal differentiation markers in myxoma cells and the inability to account for the tumor’s predominantly endothelial characteristics. Therefore, the endothelial origin hypothesis is a more plausible explanation for the pathogenesis of cardiac myxoma.

Our single-cell data analysis provides compelling evidence for the endothelial origin of cardiac myxomas, demonstrated by the enrichment of endothelial-related biological processes and validated through immunohistochemical analysis. Additionally, analysis of myxoma samples from different developmental stages in the same patient provides direct evidence of a phenotypic transition from endothelial to mesenchymal characteristics, from early (tumor pedicle) to late (tumor body) stages. Such phenotypic plasticity is likely crucial for tumor progression, consistent with previous studies reporting both endothelial-like and mesenchymal-like states in cardiac myxomas²⁸.

Our research emphasizes the role of EndMT as a key driver of phenotypic transition in cardiac myxomas, offering deeper insights into tumor progression and identifying new opportunities for therapeutic intervention. EndMT is a fundamental biological process in which endothelial cells acquire mesenchymal and invasive properties, contributing to embryonic development, wound healing, and the pathogenesis of various diseases, including fibrosis and cancer^27,29,30. In cardiac myxomas, EndMT involves a significant reorganization of the cytoskeleton, leading to the loss of junctional integrity and endothelial markers while acquiring mesenchymal traits such as enhanced migratory and invasive abilities²⁹. This transition helps explain the mesenchymal features observed in myxomas^28,31.

Our findings indicate that endothelial cells in cardiac myxomas undergo a transformation marked by the loss of endothelial markers such as VWF and KDR and the acquisition of mesenchymal markers like ENG and CD31. The expression of mesenchymal-associated genes such as MECOM and WWTR1 further supports this phenotypic shift, which promotes extracellular matrix deposition, a hallmark of myxoma structure. Interestingly, ENG and CD31, typically associated with endothelial lineage, remain expressed even as cells acquire mesenchymal characteristics, suggesting a partial or incomplete transition. This partial EndMT state may endow tumor cells with unique properties, including enhanced migration, resistance to apoptosis, and altered intercellular interactions.

Sustained activation of signaling pathways such as TGF-β, Notch, and Wnt/β-catenin further supports the role of EndMT in myxoma progression. TGF-β is well known for its role in inducing EndMT, promoting cell proliferation, differentiation, and extracellular matrix production, which are all observed in myxoma cells^32,33,34. The Notch signaling pathway also contributes to tumor progression by influencing cell fate and enhancing cell survival and proliferation^35,36, while the Wnt/β-catenin pathway regulates intercellular communication and maintains stem cell-like characteristics, impacting tumor growth and metastasis^37,38. These signaling pathways collectively drive the phenotypic plasticity observed in myxomas and offer potential targets for therapeutic intervention.

Our results echo findings from other cancer studies, such as those by Zhang et al. in breast cancer³², suggesting that EndMT is a common mechanism across various tumor types. Targeting EndMT in cardiac myxomas could involve inhibiting key pathways such as TGF-β, Notch, and Wnt, which have been explored in oncology and fibrotic disease therapies. For instance, TGF-β inhibitors could reduce the mesenchymal traits of myxoma cells, potentially limiting their invasiveness and impact on cardiac tissue. Additionally, modulation of Notch and Wnt signaling presents further opportunities to disrupt EndMT progression.

Pseudotime analysis has revealed the dynamic evolution of cellular states in cardiac myxomas, transitioning from predominantly endothelial-mesenchymal characteristics to those with higher metabolic activity. This transition outlines the progression pathways of myxomas, offering insights into the functional heterogeneity of myxoma cells at various developmental stages and enhancing our understanding of tumor progression. By elucidating these processes, our study provides essential insights into the complex biology of cardiac myxomas and highlights multiple potential points for therapeutic intervention.

Notably, in each tumor sample, we observed a consistent, dominant CNV signature across all myxoma cells, supporting clonal expansion from a single transformed precursor. This finding implies that endothelial cells harboring specific DNA events (e.g., copy-number alterations) undergo endothelial-to-mesenchymal transition (EnMT) to generate the bulk mesenchymal-like myxoma population. Other mesenchymal cell types are exceedingly rare, whereas a small fibroblast-annotated subset shares comparable CNV scores with the endothelial- and tumor-derived cells, indicating a limited EnMT-derived fibroblast transition within the tumor. Furthermore, our integrated single-cell and third-generation transcriptomic analyses revealed robust activation of the MET–PTK2 signaling axis in the majority of tumor subclones, accompanied by significant alterations in upstream and downstream effectors. In particular, we detected a marked upregulation of hepatocyte growth factor (HGF), a critical ligand that engages MET to initiate oncogenic signaling cascades, thereby driving myxoma onset and progression²⁷.

Furthermore, MET activation exhibited substantial cross-talk with other signaling pathways, including EGFR (epidermal growth factor receptor) and VEGFR (vascular endothelial growth factor receptor). This cross-activation, detected in both single-cell and third-generation transcriptomic analyses, adds to signaling complexity and contributes to resistance against therapies targeting single receptors. Once activated, MET phosphorylates and activates PTK2 (focal adhesion kinase), a non-receptor tyrosine kinase crucial for focal adhesion dynamics, which are essential for cell adhesion, migration, and survival—all key processes in myxoma development and progression²⁹.

PTK2 activation leads to cytoskeletal reorganization, promoting focal adhesion turnover and actin filament remodeling, ultimately enhancing cell motility and enabling tumor invasion into surrounding tissues. Our data also showed activation of the downstream PI3K/AKT pathway, indicating potential interactions between PTK2 and these downstream effectors that promote cell survival and proliferation, thereby supporting tumor growth and inhibiting apoptosis.

We also detected alterations in another key downstream effector, the MAPK/ERK signaling pathway, which plays a role in cell proliferation and tumorigenesis. Through this pathway, MET activation promotes cancer cell growth and differentiation, facilitating tumor progression. Additionally, PTK2 activates Rho GTPases, such as Rac1 and RhoA, which regulate cell shape and motility, further enhancing cancer cell migration potential. Collectively, these findings suggest that MET activation of the PTK2 pathway plays a central role in cardiac myxoma development by driving its initiation and progression through multiple downstream signaling changes.

The MET–PTK2 signaling axis significantly enhances cell motility, enabling tumor cells to invade the extracellular matrix (ECM). This axis is also associated with epithelial-to-mesenchymal transition (EMT), which aligns with the phenotypic changes observed in myxomas, where endothelial cells acquire mesenchymal characteristics, thereby promoting migratory and invasive properties. Activation of the PI3K/AKT and MAPK/ERK pathways further supports cancer cell survival and proliferation, even under adverse conditions, providing a growth advantage.

Moreover, MET activation influences angiogenesis by upregulating pro-angiogenic factors, consistent with the high expression of vascular-related molecules detected in myxomas, thereby enhancing the supply of nutrients and oxygen to the tumor and further supporting its growth and invasive potential. In conclusion, the activation of the MET–PTK2 signaling pathway is central to the initiation and progression of cardiac myxomas. It drives phenotypic plasticity, promotes EMT, enhances motility, and supports survival and angiogenesis. These findings highlight the MET–PTK2 axis as a promising target for therapeutic intervention to combat cardiac myxoma progression. Future strategies targeting MET, PTK2, and their downstream effectors may offer promising avenues for limiting tumor growth, particularly in inoperable cardiac myxomas, and improving patient outcomes.

Cardiac myxomas have two main gross anatomical types: a solid and smooth mass type, and a soft and friable gelatinous type. Shimono et al. discovered that the clinical presentation is related to the tumor’s gross appearance²¹. Patients with solid myxomas are more prone to symptoms of congestive heart failure, while those with gelatinous myxomas are more likely to experience cerebral and peripheral embolism²¹. Solid myxomas are vascular-rich tumors and are prone to intratumoral hemorrhage due to repeated contact with the mitral annulus²¹. Here, we identified two molecular subtypes of myxomas based on their distinct molecular characteristics, offering possibilities for developing tailored therapeutic strategies. For instance, the activation of Wnt, MAPK, and Ras pathways in subtype 1 suggests potential targets for inhibiting tumor growth and aggressiveness. The immunological features of subtype 2 highlight the potential utility of immunomodulatory treatments. These findings provide a theoretical basis for personalized medicine driven by distinct biological mechanisms.

The extensive interactions among cells within the myxoma microenvironment reveal its complexity. Frequent interactions between tumor cells and endothelial, fibroblastic, and immune cells demonstrate the collaborative roles of various cell types in tumor development. Notably, ADGRG6 is involved in tissue development and maintenance³⁹ and may be associated with the extensive mucin components found in myxomas. WNT5B activates the non-canonical Wnt signaling pathway, influencing cell polarity and migration, and plays a crucial role in endothelial-to-mesenchymal transition^40,41. FGFR2 regulates cell proliferation, differentiation, and migration during development and tissue repair^42,43, while CCR5 is involved in immune responses and inflammation⁴⁴. Additionally, interactions involving endothelial cell-related receptor-ligand pairs such as VEGFC–KDR, TGFB2–TGFBR3, and PECAM1–CD38 have been detected, where TGFB2–TGFBR3 mediates cell growth, differentiation, and collagen synthesis^45,46, and PECAM1–CD38 promotes endothelial to fibroblastic transition^47,48,49,50.

Third-generation transcriptomic sequencing results have validated and extended the findings of single-cell RNA sequencing at the molecular and pathway levels, with significant activation of collagen and vascular-related processes indicating their critical roles in myxoma progression. Overall, these results enhance our understanding of the molecular mechanisms of myxomas and provide a solid foundation for future research.

Metabolomic analysis has revealed significant metabolic differences between cardiac myxomas and normal tissues, particularly in pathways like purine metabolism and niacin and nicotinamide metabolism. Enhanced purine metabolism may promote tumor cell proliferation and anti-apoptotic capabilities^51,52, while alterations in niacin and nicotinamide metabolism suggest adaptability in energy supply and oxidative stress in myxoma cells^53,54,55. Abnormalities in these metabolic pathways not only reveal unique strategies for growth and survival of myxoma cells but also offer new avenues for potential therapeutic targets. At the single-cell level, the significant upregulation of metabolic genes such as those from the PDE family, NNMT, and NAMPT further highlights the importance of these pathways in myxomas.

In conclusion, this study offers a comprehensive exploration of the transcriptional and metabolic landscape of cardiac myxomas, providing valuable insights that pave the way for future clinical research and therapeutic innovations. Our findings highlight the critical role of the EndMT process, metabolic pathways, and MET-PTK2 signaling axis in the pathogenesis of cardiac myxomas, suggesting that these elements could serve as key targets for intervention. Future investigations should aim to validate the clinical utility of these biomarkers and pathways, with particular emphasis on developing novel prevention and treatment strategies for cardiac myxomas. Moreover, expanding this line of research to examine the involvement of these pathways in other tumor types may reveal broader therapeutic opportunities, potentially leading to new directions in cancer treatment.

Methods

Samples collection

Due to the rarity of cardiac myxoma, we were only able to collect six qualified myxoma specimens and seven normal endocardial control tissues for single-cell/single-nucleus RNA sequencing. Additionally, 20 samples (10 myxomas and 10 normals) were subjected to third-generation transcriptome sequencing, and a further 20 samples (10 myxomas and 10 normals) were analyzed by non-targeted metabolomics. All specimens were obtained from Sichuan Provincial People’s Hospital, with pathological confirmation of cardiac myxoma and informed consent from all patients. This study was approved by the Medical Ethics Committee of Sichuan Provincial People’s Hospital. All ethical regulations relevant to human research participants were followed. To further validate and extend our observations, we downloaded the publicly available single-cell RNA-seq dataset GSE265921 from the Gene Expression Omnibus⁵⁶, which comprises five independent cardiac myxoma samples.

Single-cell/nucleus sequencing

For single-cell and single-nucleus RNA sequencing, a total of 15 samples (8 cardiac myxomas and 7 normal endocardial tissues) were processed. Tissue samples were first mechanically dissociated to create a single-cell suspension, which was then filtered through a 40-μm cell strainer. For single-nucleus sequencing, the filtered cells were subjected to a lysis buffer to isolate nuclei, which were then stained with DAPI and sorted using fluorescence-activated cell sorting (FACS). For single-cell sequencing, intact cells were used directly.

Both single cells and nuclei were loaded onto a microfluidic chip for RNA sequencing using the 10× Genomics Chromium platform. In this step, cells/nuclei were encapsulated with barcoded gel beads in emulsion (GEMs), where reverse transcription occurred to generate barcoded cDNA. The cDNA was amplified, and sequencing libraries were constructed by adding sequencing adapters and indices.

The resulting libraries were quantified, quality-checked, and sequenced on an Illumina platform, ensuring comprehensive transcriptomic profiling of cardiac myxoma and normal endocardial tissue at the single-cell and single-nucleus levels.

Preprocessing of single-cell/nucleus sequencing data

The raw single-cell RNA sequencing data generated from the 10x Genomics Chromium platform were processed using the Cell Ranger software (version 7.1). First, the cellranger mkfastq command was used to demultiplex the raw base call (BCL) files into FASTQ files, corresponding to each sample. These FASTQ files were then processed using the cellranger count pipeline, which performs alignment, filtering, barcode counting, and UMI counting.

The reference genome for human (GRCh38) was used for alignment. During this step, Cell Ranger aligned the reads to the reference genome using the STAR aligner, filtered out low-quality reads and cells, and counted unique molecular identifiers (UMIs) associated with each gene. The output of this process included a gene-barcode matrix, which indicates the expression level of each gene in each cell, as well as quality control metrics such as the number of genes detected per cell, total UMI counts, and percentage of mitochondrial reads. All Cell Ranger parameters were used at their default/recommended settings. The processed single-cell and single-nucleus RNA sequencing data were analyzed using the Scanpy57 package (version 1.10.0) in Python (Version 3.9.12). Initially, we calculated the proportions of mitochondrial, ribosomal, and hemoglobin genes for each cell. Cells with more than 20% mitochondrial or ribosomal gene expression were filtered out, as these cells are likely in an apoptotic state. Similarly, cells with more than 20% hemoglobin gene expression were identified as erythrocytes and excluded. Additionally, cells with expression profiles deviating by more than 5 median absolute deviations from the median were identified as aberrant and excluded from the analysis.

To correct for ambient RNA contamination, we utilized SoupX⁵⁷, which estimates the environmental mRNA expression profile from empty droplets using the unfiltered Cell Ranger matrix and calculates the contamination fraction specific to each cell. The expression matrix was then corrected based on this environmental mRNA profile and the estimated contamination. Additionally, genes not detected in at least 20 cells were filtered out, as they were deemed non-informative. We employed scDblFinder⁵⁸ to address the issue of doublets by artificially generating doublets from randomly selected droplet pairs, averaging their gene expression profiles. The doublet score for each droplet was defined as the fraction of artificial doublets within its k-nearest neighbors in the principal component space, allowing us to identify and exclude doublet cells from further analysis. All scDblFinder parameters were set according to the authors’ guidelines.

Normalization, integration, dimensionality reduction, and clustering of scRNA/snRNA-seq data

Cells that passed quality control were selected for further analysis. Background-corrected raw UMI count matrices were processed in Scanpy⁵⁹ (version 1.10.0) by first normalizing each cell’s counts to a total library size of 10,000 (sc.pp.normalize_total(adata, target_sum=1e4)), followed by log-transformation (sc.pp.log1p(adata)) and archiving of the log-normalized values in adata.layers[“logcounts”]. We then identified the top 2000 highly variable genes using the Seurat v3 method while controlling for batch effects (sc.pp.highly_variable_genes(adata, n_top_genes=2000, flavor = “seurat_v3”, batch_key = “batch”)) and retained only those genes for downstream analysis (adata = adata[:, adata.var[“highly_variable”]]). The filtered expression matrix was centered and scaled to unit variance with extreme values capped at 10 (sc.pp.scale(adata, max_value = 10)), and principal component analysis was performed to compute the first 30 components using the ARPACK solver (sc.tl.pca(adata, n_comps=30, svd_solver = “arpack”)). To remove batch effects, we applied Harmony⁶⁰ integration on the PCA embedding (sc.external.pp.harmony_integrate(adata, key = “batch”)), constructed a k-nearest-neighbor graph in the Harmony-corrected PC space (30 PCs, k = 30, Euclidean metric; sc.pp.neighbors(adata, use_rep = “X_pca_harmony”, n_pcs=30, n_neighbors=30, metric = “euclidean”)), and generated a two-dimensional representation via t-distributed stochastic neighbor embedding on the Harmony PCs (sc.tl.tsne(adata, use_rep = “X_pca_harmony”)) for all downstream visualization and clustering. The Wilcoxon test was employed to identify differentially expressed genes between clusters. Based on the CellMarker database⁶¹, the SCSA algorithm⁶² was employed for initial cell type annotation using cluster-specific markers. Final cell type determination was made by referencing specific markers reported in relevant literature. Subsequently, the annotated normal and cardiac myxoma samples were merged using the Harmony method, followed by dimensionality reduction and differential analysis conducted according to the same procedure.

Enrichment analysis

The clusterProfiler package⁶³ in R (Version: 4.3.1) was used to perform enrichment analysis of biological functions and KEGG pathways on a group of genes. A corrected p-value < 0.05 was considered statistically significant. For significantly enriched biological functions, the rrvgo package was employed to calculate the similarity between different terms, which helped to simplify the GO enrichment results.

CNV analysis

The R package SCEVAN⁶⁴ was utilized to infer CNVs in tumor samples based on scRNA/snRNA sequencing data. The pipelineCNA function was employed to infer the benign or malignant nature of cells within a single tumor sample and to assess the heterogeneity within the tumor, including the subclone structure. For comparing multiple tumor samples, the multiSampleComparisonClonalCN function was used to analyze clonal CNVs across different samples. For each cell, we calculated the CNVscore by averaging the inferred CNV values across all genomic segments and then squaring this mean. Squaring ensures that both copy-number gains and losses contribute positively and that larger aberrations carry greater weight.

Pseudotime analysis

To perform pseudotime analysis, we utilized the Monocle2⁶⁵ package in R. Initially, we constructed a CellDataSet object using our gene expression matrix, sample information, and gene annotations. The data were then normalized using the estimateSizeFactors and estimateDispersions functions, followed by the identification of highly variable genes for ordering. We employed the reduceDimension function with the DDRTree method to perform dimensionality reduction, which enabled the visualization of cells in a lower-dimensional space. Cells were then ordered along a pseudotime trajectory using the orderCells function, which organizes the cells based on their progression through biological processes.

For trajectory visualization, we used the plot_cell_trajectory function, coloring cells by pseudotime, state, or gene expression levels. To identify genes with significant changes along the pseudotime, we conducted a differential gene expression analysis using the differentialGeneTest function, focusing on genes with a q-value < 0.01. The dynamic expression of these genes across the pseudotime trajectory was further explored using plot_genes_in_pseudotime. All Monocle2 functions were executed with their default parameters as recommended by the package authors.

Next, we inferred RNA velocity. First, we ran the velocyto run command on the Cell Ranger–aligned BAM files—providing the matching barcodes file, repeat-mask GTF, and reference gene annotations—to generate spliced and unspliced count matrices. The resulting loom files were loaded into Python using Scanpy (v1.10.0) and scVelo (v0.2.4). After inspecting spliced/unspliced ratios for quality control, we used the function scv.pp.filter_and_normalize to normalize and filter the data (parameters: min_shared_counts = 20, n_top_genes = 2000). We then used the function sc.pp.neighbors to construct a neighborhood graph on the Harmony-corrected embedding (use_rep = “X_harmony”, n_neighbors = 30) and the function scv.pp.moments to compute second-order moments (n_pcs = 30, n_neighbors = 30). Finally, we estimated velocities by applying the function scv.tl.velocity (mode = “stochastic”), built the velocity graph with the function scv.tl.velocity_graph, and visualized streamlines over the UMAP embedding using the function scv.pl.velocity_embedding_stream. All functions were executed with their recommended default settings to ensure reproducibility.

Immunohistochemistry

Formalin-fixed, paraffin-embedded cardiac myxoma sections (4 µm) were deparaffinized in xylene and rehydrated through graded ethanol. Antigen retrieval was performed in a pressure cooker using 0.01 M sodium citrate buffer (pH 6.0) for 15 min. Endogenous peroxidase activity was quenched with 3% hydrogen peroxide for 10 min, and nonspecific binding was blocked with 5% normal goat serum for 30 min at room temperature. Sections were incubated overnight at 4 °C with primary antibodies against eight markers—PECAM1 (CD31), CD34, Desmin, α-SMA, SOX9, S100, NES200, and PGP9.5—each at dilutions of 1:100–1:500. After washing, slides were incubated with biotinylated secondary antibodies and the VECTASTAIN ABC reagent, and antigen–antibody complexes were visualized using 3,3′-diaminobenzidine (DAB). Finally, sections were counterstained with hematoxylin, dehydrated, and mounted. All antibody dilutions, incubation times, and detection steps were performed according to the manufacturers’ instructions.

Cellphone

We used CellPhoneDB⁶⁶ (version 5.0) to statistically infer potential cell-cell interactions. The analysis was conducted by running the cellphonedb method statistical_analysis command, which computes interactions between cell types based on the expression of ligand–receptor pairs. Permutation tests were performed to assess the significance of these interactions, with p-values adjusted for multiple testing.

Third-generation transcriptome sequencing

For third-generation transcriptome sequencing, RNA was extracted from the samples following a standard protocol to ensure high integrity for subsequent analysis. The extracted RNA was then subjected to long-read sequencing using the PacBio platform⁶⁷, which generates full-length cDNA reads, providing a comprehensive view of transcript structures, including isoforms, alternative splicing events, and potential gene fusions.

After sequencing, raw reads were processed using the PacBio SMRT Analysis software⁶⁷ to remove low-quality reads and adapter sequences. The high-quality reads were then aligned to the reference genome using the minimap2 aligner⁶⁸, which is specifically optimized for long-read data. Transcript quantification was performed using StringTie⁶⁹, resulting in an expression matrix of full-length transcripts.

The expression data were normalized using the TPM (Transcripts Per Million) method to account for differences in sequencing depth and gene length, ensuring comparability across samples. Differential gene expression analysis was performed using the limma package⁷⁰, focusing on identifying genes with significant expression changes between different conditions. Genes with an adjusted p-value < 0.05 were considered statistically significant and differentially expressed. For fusion gene analysis, the aligned reads were further analyzed using STAR-Fusion⁷¹, which detects chimeric reads indicative of potential gene fusions.

Non-targeted metabolomics sequencing

For non-targeted metabolomics sequencing, metabolites were extracted from the samples using a standardized protocol involving solvent extraction, typically with a mixture of methanol and water. The extracted metabolites were subsequently analyzed using high-resolution liquid chromatography–mass spectrometry (LC–MS)⁷² to obtain a comprehensive metabolite profile.

After data acquisition, raw mass spectrometry data were preprocessed using software tools such as XCMS⁷³ or MZmine⁷⁴ for peak detection, retention time alignment, and normalization. This preprocessing included noise reduction and correction for batch effects. The resulting metabolite features were annotated by matching mass-to-charge (m/z) ratios and retention times against established metabolite databases (Human Metabolome Database (HMDB))⁷⁵.

Normalization of the metabolite data was performed to account for variations in sample preparation and instrument performance, typically using total ion current (TIC) normalization. The normalized data were then log-transformed to stabilize variance across samples.

Differential metabolite analysis was conducted using statistical methods such as limma⁷⁰, depending on the experimental design, to identify metabolites with significant differences in abundance between groups. Metabolites with an adjusted p-value < 0.05 were considered statistically significant. Enrichment analysis of the differential metabolites was performed using the MetaboAnalyst platform⁷⁶ to identify relevant metabolic pathways and biological functions. All analysis parameters were used according to the software authors’ recommended settings.

Statistics and reproducibility

Data from single‑nucleus RNA sequencing, third‑generation transcriptomics, and targeted sequencing were processed using established analytical workflows, with statistical methods detailed in the Methods section. Reproducibility was validated using an independent external cohort and multiple independent datasets.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.