Single‐cell RNA sequencing of solid pseudopapillary neoplasms of the pancreas in children

Abstract

Solid pseudopapillary neoplasm (SPN) of the pancreas is a rare pancreatic tumor in children. Its origin remains elusive, along with its pathogenesis. Heterogeneity within SPN has not been previously described. In addition, low malignant but recurrent cases have occasionally been reported. To comprehensively unravel these profiles, single‐cell RNA sequencing was performed using surgical specimens. We identified the cell types and suggested the origin of pancreatic endocrine progenitors. The Wnt/β‐catenin pathway may be involved in tumorigenesis, while the epithelial‐to‐mesenchymal transition may be responsible for SPN recurrence. Furthermore, NOV, DCN were nominated as primary and S100A10, MGP as recurrent SPN marker genes, respectively. Our results provide insight into the pathogenesis of SPN.

Solid pseudopapillary neoplasm (SPN) of the pancreas is a rare pancreatic tumor in children with elusive pathogenesis and origins. Here, we applied single‐cell RNA sequencing to identify the underlying mechanism of SPN tumorigenesis and the possible origin. Analyses identified distinct heterogeneity within primary SPN tumor cells. The tumor microenvironment was also revealed. In addition, recurrent SPN tumor cells were found to possess epithelial‐to‐mesenchymal transition features and novel marker genes were identified.

Abbreviations

CNV

copy number variation

DEGs

differentially expressed genes

EMT

epithelial‐to‐mesenchymal transition

GO

gene ontology

GSEA

gene set enrichment analysis

H&E

hematoxylin and eosin

IHC

immunohistochemistry

PCW

post‐conception weeks

qPCR

quantitative PCR

scRNA‐seq

single‐cell RNA sequencing

SPN

solid pseudopapillary neoplasm

TNFR

tumor necrosis factor receptors

UMAP

uniform manifold approximation and projection

1. INTRODUCTION

Solid pseudopapillary neoplasm (SPN) of the pancreas is a rare tumor that accounts for ~1%–2% of the pancreatic exocrine neoplasm. ¹ Most patients with SPN are young women between the ages of 20 and 40 years. ² However, tumors make up ~8%–12.5% of pediatric pancreatic neoplasms. ³ Most of these tumors are considered low‐grade malignant, but a few show recurrence and distant metastases.

Solid pseudopapillary neoplasm is usually diagnosed by its distinctive pathogenic features. Tumor tissues consist of solid nests of tumor cells and the fake papillary structure can be observed. Immunohistochemistry (IHC) is often required to detect positive expression of α1‐antitrypsin, vimentin, β‐catenin, claudin‐5, progesterone receptor, and CD10, ⁴ but there are no specific or sensitive markers. SPN is classified as a pancreatic epithelial tumor, and almost 70% of tumors have demonstrated positive cytokeratin expression. ⁵ In addition, the origin of SPN remains elusive. Some researchers suggest that the tumor originates from the incorporation of primitive ovarian cells into the pancreatic parenchyma during the seventh week of embryogenesis. ⁶ Other scholars propose that tumors originate from pluripotent stem cells. ⁷ However, these mechanisms have not been confirmed to date. Regarding the pathogenesis of SPN, it has been proposed that it is related to mutations in the coding sequence of the somatic gene β‐catenin (CTNNB1), ⁸ which can affect the Wnt signaling pathway. ⁹ Other studies indicated the correlation between the overexpression of transcription factors SOX11 and TFE3 in the Wnt pathway and SPN. ¹⁰ , ¹¹ In the case of recurrent SPN, the available literature is even more limited.

Because both the characteristics of SPN and recurrence mechanisms remain to be clarified, along with the unclear origins of the tumor, we performed single‐cell RNA sequencing (scRNA‐seq) on SPN specimens to unravel these features at the transcriptome level.

2. MATERIALS AND METHODS

2.1. Patients and sample collection

Eight patients diagnosed with SPN between May 2019 and December 2020 at the Children's Hospital of Fudan University were included in this study. All samples were derived from surgical resection specimens. Five of them were processed for scRNA‐seq (assigned nos. T138, T141, T146, T219, and T387). Hematoxylin and eosin (H&E)‐stained sections and IHC findings were obtained from tumor tissues and examined by at least two pathologists. World Health Organization histological criteria were used to indicate malignant potential, including vascular infiltration, extra‐pancreatic or perineural invasion, and to identify the presence of pancreatic parenchyma disorder. ¹² T141 was a patient with recurrent SPN, who had previously accepted only radical surgery. None of the patients had a history of other cancers or other pancreatic diseases. None of the patients was treated with chemotherapy, radiation, or any other antitumor drug prior to tumor resection. We also collected four benign pancreatic biopsies (assigned P1–P4) as control. Furthermore, 10 fetal pancreas specimens were included, which were obtained from elective termination of pregnancy at different post‐conception weeks (PCW) at the Obstetrics and Gynecology Hospital of Fudan University. Two of them (assigned nos. F9 and F10) were processed for scRNA‐seq.

2.2. Single‐cell preparation and scRNA sequencing

The samples were dissected into pieces, digested, and isolated into single cells. After single‐cell partitioning, reverse transcription, and amplification, the libraries were built and processed for sequencing and analyses. The detailed protocol is given in Appendix S1.

2.3. Bioinformatics analyses

We performed Scrublet ¹³ and Seurat v3 ¹⁴ for upstream quality control. Uniform Manifold Approximation and Projection (UMAP) ¹⁵ was used for data visualization and Harmony ¹⁶ was used for data integration. Additionally, we applied inferCNV ¹⁷ , ¹⁸ , ¹⁹ and NMF ²⁰ to explore the characteristics of SPN tumor cells, and CIBERSORTx, ²¹ Jaccard similarity analysis, CytoTRACE, ²² and Monocle2 ²³ to investigate the origin of SPN tumor cells. Gene ontology (GO) and gene set enrichment analysis (GSEA) were performed using the R package ClusterProfiler. ²⁴ Python package pySCENIC ²⁵ was used for transcription factor analysis and CellPhoneDB v2.0 ²⁶ was applied for cell communication analysis. The detailed protocol is given in Appendix S1.

2.4. IHC and multicolor IHC

Formalin‐fixed and paraffin‐embedded tissue sections were deparaffinized and rehydrated. Antigen retrieval was performed using 10× Tris‐EDTA (pH 9.0) at 100°C for 10 min. Color development of IHC was performed using DAB and hematoxylin (Gene Tech, GK500710). Multicolor IHC was performed using a fluorescence IHC kit (Panovue, TSA‐RM‐275) according to the protocol of the manufacturer. IHC images were acquired with a Leica light microscope (DM750). Immunofluorescence images were captured by Leica Confocal Microscope and processed by Leica Imaging Software. The primary antibodies aere listed in Table S1.

2.5. Real‐time qPCR and western blot analysis

Total RNA and protein were extracted from specimens using TRIzol reagent and RIPA respectively. Real‐time quantitative PCR (qPCR) was performed with the primer sequences listed in Table S2. The protein was subjected to SDS‐PAGE and transferred to PVDF membranes. Detailed protocols are presented in Appendix S1. Mann–Whitney tests were used to compare the significance between the two groups, and P values <0.05 were considered significant.

3. RESULTS

3.1. Single‐cell transcriptome profiling of primary SPN

We collected four primary SPNs during surgical resection (Figure S1A). Microscopically, the appearance of all tumors presented the characteristic microscopic characteristics of SPN. The tumor was composed of a solid area, pseudopapillary area, and the transition area of these two mixed in different proportions. The pseudopapillary area formed a branched pseudopapillary structure with multiple layers organized by surface cells. Focal hemorrhage and necrosis were often observed (Figure 1A). To comprehensively characterize cellular diversity in pediatric SPN, we carried out high‐throughput scRNA‐seq on four fresh human SPN specimens using a droplet‐based 10× genomics platform (Figure 1B).

FIGURE 1.

scRNA‐seq reveals the composition of SPN. (A) H&E staining of four primary SPNs. (B) General workflow of droplet‐based single‐cell RNA sequencing. (C) UMAP visualization of merged SPN profiles colored by cell types. (D) Dot plot displaying marker genes for each cell type. (E) Heatmap showing the inferred copy number variation (CNV) profiles in SPN tumor cells and comparison with endothelial cells. (F) UMAP visualization of the average score of the chromosome 4 duplication and chromosome 6 loss calculated by inferred CNV

After raw data processing, filtering low‐quality cells and doublets, we obtained 28,817 cells for further analysis (Figure S1B,C). We then used classic known marker genes to define cell types, including two types of classical dendritic cells (IRF8, XCR1, and CD1C, CLEC10A), endothelial cells (VWF, PLVAP), fibroblasts (COL1A1, COL3A1), macrophages (CD163, CD68), two groups of monocytes (CD14, FCGR3A), mast cells (TPSAB1, TPSB2), B cells (CD79A, CD79B), and T and NK cells (CD3E, NKG7; Figures 1C,D and S1D). We used markers commonly specified for SPN IHC and marker genes extracted from previous reports ¹⁰ , ¹¹ , ²⁷ , ²⁸ , ²⁹ to identify tumor cells in our data. SOX11, TFE3, LEF1, HAND2, MME, AR, and PGR are all effective marker genes for identifying tumor cells, while CTNNB1 and VIM are less specific (Figure S1E). In addition, we applied large‐scale inferred copy number variation (CNV) analysis to distinguish malignant from nonmalignant cells. We selected endothelial cells as the “normal” reference and found clusters having a similar pattern, which showed losses of chromosome 6 and duplicates on chromosome 4 (Figure 1E,F). Consistent with the marker genes, these clusters with aberrant karyotypes were convincingly identified as tumor cells.

3.2. Intraheterogeneity of primary SPN tumor cells

We extracted the subset of tumor cells and performed unsupervised clustering to elucidate the characteristics within the SPN (Figure 2A). To further explore how expression states varied among the different subtypes within the same tumor, we used NMF ²⁰ to identify the coherent sets of genes that were significantly expressed by different subsets. Four meta‐programs were generated (A to D), including three specific ones (A, B, D) and one nonspecific (D) (Figure 2B). These were functionally annotated using GO terms (Figure 2C). Meta‐program A regulated stem cell proliferation (HNRNPU, NUMB, N4BP2L2) and was associated with cellular junctions (DSP, TJP1). Meta‐program B was associated with cell adhesion (ACTG1, ACTB, LGALS1). Meta‐program C contained genes relative to the lumen of the granule (SRP14, FTH1, EEF1A1). Meta‐program D was characterized by stress response genes (FOS, JUN, JUNB). Focusing on all tumor cells, the programs were expressed across different clusters, which implied intratumoral heterogeneity (Figures 2D and S2A). To further investigate the relative differentiation state of these tumor cells, CytoTRACE analysis was utilized. The differentiation state of the tumor cells varied (Figures 2E and S2B). Cells with higher expression in meta‐program B were less differentiated (Figure 2F). Genes associated with differentiation and predicted by CytoTRACE were enriched in pathways possibly correlated with SPN tumorigenesis, including oxygen metabolism, MYC targets, and DNA repair (Figures 2G and S2C).

FIGURE 2.

Intraheterogeneity within SPN tumor cells. (A) UMAP visualization of SPN tumor cells colored by clusters. (B) Heatmap depicting pairwise correlations of 40 modules extracted by NMF analysis derived from four tumors. Clustering of four coherent expression meta‐programs across tumors (meta‐programs A to D). (C) Enriched GO terms of the top 50 genes of each meta‐program. (D) Feature plot of gene sets of four meta‐programs in all tumor cells. (E) UMAP visualization of the CytoTRACE score of all tumor cells. (F) Heatmap displaying the Pearson's correlation coefficients calculated between the scores of meta‐programs and CytoTRACE. (G) GSEA analysis based on genes correlated with CytoTRACE scores in the Hallmark gene set

3.3. Putative origin of malignant cells

The origin of SPN is unclear, and many hypotheses have been proposed. However, there is no definitive conclusion or concrete supporting evidence. Some theories indicate the origin being the pancreatic stem cell. To confirm that, we performed scRNA‐seq on two fetal pancreas specimens (10.5 and 14 PCW, respectively), as our reference. After upstream processing, we obtained 8991 high‐quality cells, among which 425 exocrine and endocrine cells were defined by canonical markers (Figures 3A,B and S3A–C). We then applied the computational method CIBERSORTx to estimate the abundances of member cell types in mixed tumor populations. After building the signature matrix file of fetal pancreas data using CIBERSORTx, mixed tumor data were deconvolved by calculating the similarity with the input signature gene expression profile, and then the presumed cell fractions were enumerated (Figures 3C and S3D,E). There is a sizable proportion of endocrine cells, which implies an endocrine origin. Additionally, the Jaccard similarity analysis showed that the endocrine cell type identified in these two samples was highly transcriptionally similar to the meta‐programs in the tumor subsets (Figure 3D). Furthermore, two types of endocrine cells were identified by mature pancreatic beta cell marker genes (INS, MAFA, IAPP) and endocrine lineage genes (PAX4, NEUROD1, CHGB) (Figure 3E,F). Both the meta‐programs and the differentially expressed genes (DEGs) of each tumor cluster were more similar to the developing endocrine subtype (Figures 3G and S3F). Due to the small number of endocrine cells in our dataset, we collect the DEGs of all types of endocrine cells from another dataset ³⁰ (Table S3) and used them to score SPN cells. Consistently, markers of endocrine progenitor populations scored higher than those of mature populations (Figure 3H). We chose several DEGs which were highly expressed in endocrine progenitor cells (FEV, NEUROG3 (also as NGN3), SPP1, MDK, CHGB, STMN1) and then validated the expression in fetal pancreases and SPN specimens using IHC and qPCR (Figures 3I, 4A,B, and S4A). We found that the expression of these genes in SPN was at nearly the same or a higher level compared to fetal pancreas samples. On the contrary, SPN had a much lower expression of mature endocrine cell marker genes (Figure S4B). SPN tumor cells were more alike to immature endocrine cells than mature ones. Except for being found in the fetal pancreas, we also discovered this type of endocrine progenitor cells exists in adult and children's pancreas by using multicolor IHC (Figure S4C,D). Meanwhile, we subset the endocrine cells from the normal adult pancreas dataset ³¹ and found a small cluster of endocrine progenitors (Figure S4E). These indicated the existence of endocrine progenitor cells through childhood and adulthood. Next, we combined the SPN cells and endocrine cells from fetal and adult datasets (Figure S4F). Then we applied Monocle2 to analyze the pseudotime trajectory among immature endocrine cells, mature endocrine cells, and tumor cells. Most SPN cells were located in the later stage of the development trajectory (Figures 4C,D and S4F,G). In summary, SPN tumor cells may originate from pancreatic endocrine progenitor cells.

FIGURE 3.

Putative origin of malignant cells. (A) UMAP plot showing the cell atlas of two fetal pancreases. (B) Heatmap showing the top 20 DEGs of all cell types. Marker genes are listed on the right. (C) Bar plot displaying the proportion calculated by CIBERSORTx. (D) Jaccard similarities of four meta‐programs with signatures of three developing cell types. (E) UMAP visualization of two types of endocrine cells in the fetal pancreas specimens. (F) Dot plot showing marker genes of two developing endocrine cell types. (G) Jaccard similarities of four meta‐programs with the signatures of two developing endocrine cell types. (H) Boxplots showing scores of gene sets of seven reference cell types. The immature cell types scored significantly higher than the mature cell types respectively. ****P < 0.0001. (I) IHC staining of SPN and control samples

FIGURE 4.

Cell trajectory of SPN. (A) Identifying pancreatic endocrine progenitor marker genes in fetal pancreas and SPN by multicolor IHC. (B) Bar plots showing relative expression of the indicated genes between SPN and control group from qPCR analysis. *P < 0.05, **P < 0.01, ***P < 0.001; ns, not significant. (C) Distribution of immature and mature endocrine cells, and SPN tumor cells in the pseudotime trajectory. (D) Bar plots showing the proportion of each cell type in different states re‐clustered by Monocle2

3.4. Identifying specific tumor markers for primary SPN

To uncover more specific marker genes for SPN, we downloaded normal pancreas single‐cell data from GSE155698 ³¹ and compared it with our SPN tumor cells (Figure S5A–D). We subset normal endocrine cells as a reference and performed a DEG analysis (Figure 5A). The marker genes selected from significant DEGs met the following criteria: (1) >2.5 average log2 fold‐change; (2) >60% tumor cells expressed; (3) >50% delta percentage of cells ([(percentage of cluster1) − (percentage of cluster2)]/(percentage of cluster1) × 100%). After filtering and ordering the genes by fold change, NOV (also known as CCN3), AC104126.1 (a long noncoding RNA), DCN, IFITM3, and DKK3 emerged as the top five DEGs (Figure 5B). We then examined these marker genes in the bulk SPN dataset from GSE43795, ³² all of which showed higher expression in tumors, except for AC104126.1, which was not probed due to sequencing limitations (Figure 5C). Additionally, we verified that SPN had higher expression of NOV and DCN than control samples by IHC, qPCR, and western blot analysis (Figures 5D–F and S5E). We further performed GSEA analysis on all genes upregulated in tumor cells that were detected by DESeq2 methods to assess the enrichment of hallmark pathways (Figure 5G,H). SPN tumor cells possessed a more proliferative state, as indicated by the enrichment of MYC and E2F targets pathways. It should be noted that the Wnt/β‐catenin signaling pathway was also enriched, which has constantly been reported in the literature as involved in tumorigenesis.

FIGURE 5.

Identifying characteristics of primary SPN. (A) Volcano plot showing the top five DEGs between SPN (red) and normal pancreatic endocrine cells (blue). Gray dots represent insignificant genes. (B) Violin plots displaying the expression of the selected genes between two groups. (C) Boxplots showing the expression of the selected genes between tumors and normal samples in bulk SPN data. (D) qPCR analysis showing relative expression of DCN and NOV in SPN and control groups. (E) IHC staining of SPN and control samples. (F) Western blotting analysis of DCN and NOV in SPN and control groups. (G) GSEA of the Hallmark gene set based on DEGs of SPN tumor cells. (H) GSEA enrichment plot of the expression signatures of Hallmark MYC Targets V1 and Hallmark Wnt/β‐catenin signaling in SPN tumor cells. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

3.5. Tumor microenvironment in primary SPN

The tumor microenvironment plays a key role in tumor initiation and progression. Dynamic and intricate interactions between tumor cells and stromal cells, immune cells, and blood vessels may be important in tumor survival, invasion, and metastasis. ³³ After extracting noncancer cells, T and NK cells were present in considerable proportions in the microenvironment (Figure 6A). CellphoneDB (V2.0) was performed to explore cell–cell interactions among different cell types. The most interaction pairs with tumor cells were endothelial cells and macrophages for immune cells (Figure 6B). NOV, a noncanonical NOTCH1 ligand, ³⁴ was overexpressed in tumor cells and interacted with endothelial cells through the NOTCH1 receptor. In addition, the VEGF receptor 1 (also known as FLT1) in endothelial cells was also upregulated and interacted with tumor cells VEGFRA and VEGFRB. For immune cells, SPN tumor cells revealed the involvement of pleiotropic cytokine macrophage migration inhibitory factor (MIF) to interact with macrophages expressing the CD74 receptor. Furthermore, the analysis identified GRN (granulin precursor, also called PGRN), a direct ligand of tumor necrosis factor receptors (TNFR), was also overexpressed in macrophages and had strong interactions with TNFR in tumor cells (Figure 6C–E).

FIGURE 6.

Tumor microenvironment in SPN. (A) The proportion of different cell types in the SPN tumor microenvironment. (B) Heatmap displaying the number of potential ligand‐receptor pairs between cell types. (C) Dot plots showing the top five highly expressed ligand‐receptor interactions of tumor cells and endothelial cells, tumor cells, and macrophages. (D) Feature plots showing selected ligand‐receptor pairs of tumor cells and endothelial cells. (E) Feature plots showing selected ligand‐receptor pairs of tumor cells and macrophages

3.6. Uncovering transcriptomic characteristics of recurrent SPN

Although SPN rarely presents metastasis and recurrence, it does occur and it may be difficult to predict its malignant potential based on radiological or histopathological features. ³⁵ We collected T141 as a recurrent sample. The histopathological assessment of the tumor was consistent with SPN (Figures 7A and S1A). We applied scRNA‐seq on this sample to explore its differences and potential mechanism. The tumor cells were subset and a combined analysis with primary cells was performed as advanced research. First, the Euclidean similarity between primary and recurrent SPN was assessed, which showed significant differences (Figure 7B). Using the same method as before, S100A10, MGP, BST2, LUM, and TPM2 were identified as the five main marker genes (Figure 7C,D). We applied IHC staining of S100A10 and MGP in primary and recurrent SPN (Figure 7E). GSEA analysis on DEGs also revealed enrichment of hallmark pathways, including MYC and E2F targets. Moreover, because more mesenchymal genes were upregulated and most of the screened marker genes were correlated with the mesenchyme, epithelial‐to‐mesenchymal transition (EMT) may also involve the mechanism of the recurrence (Figures 7F,G and S6A,B). Meanwhile, we verified that the expression of recurrent tumor mesenchymal markers was higher than that of primary tumors by IHC (Figures 7H and S6C). Taken together, these findings indicated the involvement of the EMT process in recurrent SPN.

FIGURE 7.

Characteristics of recurrent SPN. (A) H&E staining of recurrent SPN (T141). (B) Euclidean distance between primary and recurrent SPN tumor cells. (C) Volcano plot showing the top five DEGs between recurrent (red) and primary (blue) SPN. Gray dots represent insignificant genes. (D) Violin plots displaying the expression of the selected genes between two groups. ****P < 0.0001. (E) IHC staining of primary and recurrent SPN samples. (F) GSEA in the Hallmark gene set based on DEGs of recurrent and primary SPN tumor cells. (G) GSEA enrichment plot of expression signatures of Hallmark MYC Targets V1 and Hallmark EMT in recurrent SPN tumor cells. (H) IHC staining of primary and recurrent SPN samples. (I) Rank for regulons in recurrent SPN tumor cells based on the regulon specificity score (RSS)

Next, we explored the differences by SCENIC regulon analysis. We selected the regulons with the highest regulon specificity score for either the recurrent or primary tumor. Our analysis identified NR2F2 and ZNF717 as specific regulons associated with recurrent SPN (Figures 7I and S6D,E). For primary SPN, the most specific regulons included LEF1 and TCF7 (Figure S6F,G).

4. DISCUSSION

Solid pseudopapillary neoplasm of the pancreas is a rare tumor whose origin is still unclear. Most tumors are low‐grade malignant, but a few cases presenting recurrences and distant metastasis have been reported. There is limited evidence available describing tumor characteristics at the transcriptome level, and the mechanism of disease recurrence remains uncertain. In this study, we performed scRNA‐seq to investigate these issues at single‐cell resolution. We constructed the first cell transcriptomic atlas of SPN.

After identifying cell types using recognized markers, we verified the results by inferCNV to distinguish tumor cells. We observed an elevated Ki‐67 index in the recurrent sample. The Ki‐67 index has been suggested as an indicator of malignant potential and poor outcome of SPN, and also as an index of recurrence. ²⁷ Primary tumors showed a low Ki‐67 index (≤5%) that indicated a slow growth of the tumors. These suggested the indolence of primary SPN and the malignancy of the recurrent tumor.

Intratumoral heterogeneity represents a major challenge in oncology. Due to these genetic and phenotypic differences, tumor cells can exhibit differences in proliferation, metastasis, and drug resistance. We applied different approaches to characterize intratumoral heterogeneity. Different meta‐programs identified distinct cellular functions and were expressed by different cell subpopulations. Meta‐program B was associated with cell adhesion, which included genes ACTG1 and ACTB. They were reported to be of great significance in the regulation of cytoskeleton in cancer development and correlated with the poor prognosis of some tumors. ³⁶ , ³⁷ Meanwhile, cells with a higher score of meta‐program B were presumed to be less differentiated, which may take responsibility for the development of SPN.

We next explored the origin of SPN since it is currently elusive. Some studies have provided evidence supporting an endocrine origin ³⁸ , ³⁹ or the involvement of exocrine cells, ⁴⁰ , ⁴¹ and others support primitive pancreatic cells. ⁴² , ⁴³ Thus, we used fetal pancreas scRNA‐seq data as a reference for involving developmental cell types. In mice, all pancreatic epithelial lineages derive from multipotent progenitor cells (Pdx1+). During development, the expression of endocrine progenitor cells transitions from Spp1+/Ngn3+, Ngn3+, Ngn3+/Fev+ to Fev+. Then, Fev+ progenitor cells differentiate into hormone‐producing cells. ³⁰ Through similarity analyses and experimental validation of endocrine marker genes, the results showed great evidence of the transcriptomic similarities between SPN tumor cells and pancreatic endocrine progenitor cells, instead of any other cell types in the pancreas. Pseudotime analysis revealed that most immature cells were in the early stage and SPN cells were in the later stage of the simulated endocrine development trajectory. Meanwhile, most SPN tumor cells unsupervisedly clustered with a small number of immature cells downstream of the trajectory, which indicated pancreatic endocrine progenitor cells as the possible origin of SPN cells and the similarities between these two cell types. We found endocrine progenitor cells exist in children's and adult pancreases, suggesting SPN can occur in both children and adults.

Through the evaluation of DEGs between SPN and normal pancreatic endocrine cells and further validation in the bulk SPN data, we identified several marker genes, among which NOV showed good diagnostic potential. NOV encoded a secretion protein that could be detected in human serum, which indicated the possibility of diagnosing SPN by serum examination in the future. Further experiments should be pursued. Nonetheless, top markers such as NOV, DCN, and DKK3 have been reported to act as a powerful suppressor of tumor growth, ⁴⁴ , ⁴⁵ , ⁴⁶ which may be related to the promising prognosis of SPN. The Wnt/β‐catenin signaling pathway, which has been described in previous studies, ⁸ , ⁴⁷ , ⁴⁸ was enriched in our analysis and may contribute to tumorigenesis. As the specific regulons of primary SPN, LEF1 and TCF7 had a central regulatory role in the Wnt/β‐catenin signaling pathway by inducing or preventing aberrant protein expression. ⁴⁹ Overexpression of LEF1/TCF7 may abnormally affect the Wnt/β‐catenin pathway, upregulate downstream genes (such as cyclin D1, c‐Myc, and c‐Jun), and result in the development of SPN, which was similar to the pathogenesis observed in other tumors. ⁵⁰ , ⁵¹

Regarding the tumor microenvironment, strong connections bound tumor cells and endothelial cells, most of which took an effect on neovascularization ⁵² through NOTCH1 receptor in Notch signaling ⁵³ or FLT1 in VEGF pathway, ⁵⁴ and may participate in tumor progression. Based on interactions between macrophages and tumor cells, we suggested that macrophages have an immunosuppressive state. Activation of CD74‐MIF could prompt macrophage conversion to the immunosuppressive state, which thus could activate the pro‐tumorigenic potential. ⁵⁵ , ⁵⁶ Meanwhile, GRN‐TNFR pairs could inhibit TNFα signaling ⁵⁷ and help tumor cells escape antitumorigenic immune responses. ⁵⁸

The reported recurrence rate after SPN resection is approximately 4.4%, ⁵⁹ while the underlying mechanisms and predictors of recurrence are unknown. We compared recurrent SPN with primary tumors. Hallmark pathways such as MYC targets and EMT were enriched. MYC is known as a proto‐oncogene and has been implicated in many human tumorigeneses. Together with its target genes, they collectively enforce the characteristics of cancer cells, such as abnormal DNA replication and transcription, uncontrolled cellular proliferation, and vigorous metabolism. More importantly, EMT is also an underlying mechanism of carcinoma progression and metastasis, ⁶⁰ as cell polarity changes and gains the ability to break down the extracellular matrix for invading surrounding tissue. ⁶¹ In our case, recurrent SPN expressed an elevated level of mesenchyme‐associated genes. Marker genes like S100A10 and MGP also coded secreted protein that can be detected in serum, which may be used to predict the risk of SPN recurrence in the future. Meanwhile, abnormal expression of S100A10 or MGP has also been found in various types of tumors, including breast cancer, glioblastomas, and colorectal cancer, where the relationship between upregulation of S100A10 or MGP and a poor prognosis was identified. ⁶² , ⁶³ S100A10 was considered a key regulator of plasminogen activation during EMT and functioned in cancer progression in complex with ANXA2. ⁶⁴ , ⁶⁵ The induction of EMT to promote proliferation and migration by MGP was previously demonstrated in triple‐negative breast cancer. ⁶⁶ Suppressing MGP in colorectal cancer inhibited tumor proliferation and reversed oxaliplatin resistance. ⁶⁷ As specific regulons in recurrent SPN, both NR2F2 and ZNF717 were well‐characterized regulators for tumor cell proliferation. ⁶⁸ , ⁶⁹ In addition, NR2F2 has been reported to promote the progression of EMT and can help maintain a malignant tumor state by improving tumor renewal and restricting differentiation. ⁶⁸

In summary, our study disclosed the SPN cellular atlas and provided strong evidence on the possible tumor origin of pancreatic endocrine progenitor cells. NOV and DCN might be considered specific marker genes for identifying SPN tumor cells. Our analysis highlighted the intra‐tumoral heterogeneity in SPN. Furthermore, the Wnt/β‐catenin pathway was verified to associate with primary tumorigenesis, while the MYC and EMT pathways may play a role in the recurrence of SPN. S100A10 and MGP were selected as recurrence marker genes. Further investigation is needed to validate our conclusion.

ACKNOWLEDGEMENTS

We gratefully thank Dr. Qiangda Chen from Zhongshan Hospital Fudan University and Dr. Linan Xu from Children's Hospital of Fudan University for their friendly help in the experiments. And we wish to thank all the patients for their involvement in this study.

FUNDING INFORMATION

This study was supported by the Cyrus Tang Foundation and the National Natural Science Foundation of China [No. 82072782, No. 82172852].

CONFLICT OF INTEREST STATEMENT

The author declares no conflict of interest.

ETHICS STATEMENT

Approval of the research protocol by an Institutional Reviewer Board: Ethics Committee of the Children's Hospital of Fudan University (protocol no. 2020–415) and the Obstetrics and Gynecology Hospital of Fudan University (protocol no. 2018–069).

Informed Consent: Informed consent was obtained from the patient or guardians.

Registry and the Registration No. of the study/trial: N/A.

Animal Studies: N/A.

Supporting information

Appendix S1

Meng L, Zhan Y, Wei M, et al. Single‐cell RNA sequencing of solid pseudopapillary neoplasms of the pancreas in children. Cancer Sci. 2023;114:1986‐2000. doi: 10.1111/cas.15744

DATA AVAILABILITY STATEMENT

Raw sequencing data of SPN and fetal pancreas specimens are available in the Genome Sequence Archive database under accession numbers HRA002834 and PRJCA005331. Single‐cell RNA sequencing data for normal pancreases were obtained from GSE155698. ³¹ Bulk SPN samples were derived from GSE43795. ³²