snRNA-seq of acral melanoma reveals heterogeneity and identifies novel progression marker and cancer-associated fibroblasts subpopulation

Abstract

Acral melanoma (AM) is characterized by a high incidence of distant metastasis. Although the genomic characterization of AM has been explored, the landscape of AM in different stages is still a critical factor to be researched. We performed single-nucleus RNA sequencing analysis in three nevi, seven primary AM, four skin metastases, and three lymph node metastasis specimens. Cell subpopulations were characterized. The identified marker was verified in clinical samples, and the function was further validated by in vitro and in vivo experiments through establishing a knockdown cell line. A total of 13 types of cells were identified from 17 samples. Ten cancer cell subtypes were profiled through comprehensive analysis. Analysis identified Cancer cells_TGFBI to be the potential progressive subpopulation, and SERPINE2 was the top differentially expressed gene. SERPINE2 knockdown significantly suppressed melanoma cell growth and invasion, and downregulated the EMT pathway. Cancer-associated fibroblast (CAF) had the second largest number of cells. A new cluster, Fibroblast_TNXB, was identified and had potential correlation with melanocyte differentiation and development. The above findings demonstrated the tumor heterogeneity of AM. SERPINE2 had the potential to be a therapeutic target for melanoma, and Fibroblast_TNXB may link to melanocytic differentiation, highlighting new microenvironmental crosstalk in AM.

Introduction

Acral melanoma (AM) is a dismal subtype of melanoma developed in the skin of the glabrous acral areas, such as the palm, sole, and subungual. Although the total relative incidence of AM is rare, it accounts for ~50% of melanoma cases in East Asia¹. The development of immunotherapy and targeted therapy has greatly improved the prognosis of melanoma patients, it is still one of the deadliest skin malignant tumors. In recent years, the genomic characterization of AM has been intensively explored in several studies. However, the features of different stages of AM remain unclear.

The tumor microenvironment (TME) of melanoma comprises tumor cells, immune cells, fibroblast cells, endothelial cells, keratinocytes, etc. It is a tumor with great tumor heterogeneity. Single-cell RNA sequencing (scRNA-Seq) has the ability to identify complex admixtures of cells and reveal cellular functions at the single-cell level in unprecedented ways². However, applying scRNA-Seq technology to precious, archived human tissues has proven to be challenging, since it is difficult to dissociate intact cells from frozen biopsy tissues. Single-nucleus RNA sequencing (snRNA-Seq) techniques have overcome these technical challenges and can achieve accurate cell-type level characterization of frozen solid tissues^3,4.

Thus, in our study, we used snRNA-Seq to reveal the tumor heterogeneity among primary, skin metastases and lymph nodes metastasis of AM, and analyzed the tumor initiation and progression by comparing with nevi. We further validate the potential marker for AM progression in clinical samples and cell function assays. The novel subset of cancer-associated fibroblast (CAF) was initially validated by multiple immunofluorescence stainings. Our results demonstrated SERPINE2 as a potential gene participating in the AM progression process. CAFs have a close relationship with cancer cells, and Fibroblast_TNXB may be associated with melanocyte development.

Results

The landscape of nevi, primary and metastatic melanoma

To decipher the difference of cellular architecture among benign lesions and the TME in primary, skin metastases, and lymph nodes metastases of AM, we performed snRNA-seq of these clinical samples (Fig. 1A). We obtained 217111 cells from three nevi (n), seven primary AM (p), four skin metastasis (m_S) and three lymph nodes metastases (m_L) specimens subjected to snRNA-seq. The patients’ information has been listed in Table S1. Then, unsupervised clustering analysis was performed to define major clusters of cells with similar expression patterns. Thirteen types of cells were identified according to the expression of canonical markers and the most variable genes among the 17 specimens, including keratinocytes (gene markers: KRT14, KRT1, KRT10), sweat gland cells (gene markers: DCD, AQP5), melanocytes (gene markers: MLANA, PMEL, DCT), adipocytes (gene markers: ADIPOQ, PLIN1, FABP4), endothelial cells (ECs) (gene markers: PECAM1, VWF, ENG), fibroblasts (gene markers: DCN, LUM, COL1A1), mural cells (gene markers: ACTA2, RGS5, MYH11), B cells (gene markers: CD79A, MS4A1, IGHD), plasma cells (gene markers: JCHAIN, XBP1, MZB1), T cells (gene markers: CD3d, CD2, TRBC1), mast cells (gene markers: TPSAB1, TPSB2, CPA3), mononuclear cell (MPs) (gene markers: LYZ, C1QA, CD14), cancer cell were defined by tumor heterogeneity (Figs. 1B, C and S1A, Table S2). Pooled analysis of all samples revealed that the cancer cells have the most diverse population and that some samples formed their own distinct clusters, whereas other cell subtypes in the TME tended to cluster together (Figs. 1B and S1B). Moreover, analysis of cancer cells in each sample revealed the extreme tumor heterogeneity of melanoma (Fig. S1C). Besides cancer cells, fibroblasts had the largest population among other cell types (Table S3).

Fig. 1. The snRNA-seq analysis of the tissues of patients with nevi, primary AM, skin metastasis or three lymph nodes metastasis.

A Experimental design: the tissues were collected from the patients at The First Affiliated Hospital of Naval Medical University. B UMAP plot depicting 13 major cell types identified by snRNA-seq. C Feature plots for the canonical marker genes of major cell types. D The composing proportion of each cell type in n, p, m_S, and m_L groups. The relative contribution of each population was weighed by the number of cells and scaled to 100%. E Heatmap of the expression levels of the top 10 differentially expressed genes among 13 major cell types in all samples. UMAP uniform manifold approximation and projection.

Moreover, copy number variation (CNV) plays an important role in the pathogenesis and poor prognosis of tumor patients, and the InferCNV analysis demonstrated that the cells identified as cancer cells had patterns of amplification and deletion characteristic of malignant cells (Fig. S1D and E). The CNV score of cancer cells in each group or each cancer cell subtype was significantly higher than that of the ECs group (Fig. S1F and G). The proportions of different cell types at the four groups were compared (Fig. 1D). Melanocytes were the major cell type of nevi, and the percentage populations of cancer cells were gradually increased in p, m_S and m_L group. The proportion of fibroblasts and keratinocytes was decreased in m_S and m_L groups compared with primary tumors. Overall, the expression of the top ten marker genes among 13 major cell types was compared and displayed in a heatmap (Fig. 1E).

Distinct functional signatures of the melanoma tumor cell

To better understand the landscape of melanocytes and cancer cells in acral melanoma in finer detail, we applied a second stage of clustering. The melanocytes were further divided into four subtypes: Melanocyte_FGF14, Melanocyte_PTN, Melanocyte_SDC2, and Melanocyte_DNAJB1 (Fig. 2A). The proportions of four subtypes in the four groups were compared (Fig. 2B). The top-expressed genes in each subtype are shown in Fig. S2A. The second stage of clustering was also applied in cancer cells, and there were 10 subtypes identified (Fig. 2C). The proportion of each subtype in each group was demonstrated in Fig. 2D. Since a small number of cells in the nevi group were clustered as cancer cells, the proportion was also analyzed. The results further supported the finding that cancer cells have the most diverse population among all cell types. Most subtypes had their own unique top-expressed genes (Fig. S2B). The proportion of each subtype of cancer cells in each group was different, and some groups had their own unique subtypes, which demonstrated the tumor heterogeneity of AM.

Fig. 2. The subtypes of melanocytes and cancer cells, and the identification of four common expression programs of cancer cells.

A UMAP plot of the melanocytes landscape colored by cluster. B The composing proportion of melanocyte subtypes in n, p, m_S, and m_L groups. C UMAP plot of the cancer cells landscape colored by cluster. D The composing proportion of cancer cell subtypes in n, p, m_S, and m_L groups. E Hallmark enrichment analysis was performed to compare the differences in pathways in each subtype of cancer cells. The Z score was used to normalize each column. F A pySCENIC analysis was conducted to find the core regulons identifiable in cancer cell subtypes. G Fourteen modules derived from the cancer cell gene expression profile. H Four common expression programs across cancer cells are aggregated into clusters. I The biological functions and relevant signaling pathways of six aggregated clusters were evaluated based on hallmark gene sets. J Jaccard similarity was used to identify the correlation of 10 cancer cell subtypes and 14 gene modules. K Gene set scoring analysis of the EMT score of each cancer cell subtype.

The potential biological functions and relevant signaling pathways of ten cancer cell subtypes were evaluated by gene set variation analysis (GSVA) based on hallmark gene sets (Fig. 2E). Specifically, Cancer cells_TGFBI showed enrichment in almost all signatures; Cancer cells_LHFPL3, MIR4300HG and CTSL showed enrichment of G2M checkpoint; Cancer cells_PIZEO2 showed high enrichment of angiogenesis; Cancer cells_PKHD1 showed enrichment of epithelial-mesenchymal transition (EMT); Cancer cells_BRD9 showed enrichment of interferon gamma response; Cancer cells_CEMIP showed enrichment of TNFA signaling via NFKB. The gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment of Cancer cells_TGFBI revealed up-regulation of cell adhesion, cell differentiation, and PI3K-Akt signaling (Fig. S2C and D). Transcription factors (TFs) enrichment heatmap accomplished by pySCENIC analysis demonstrated the top-enriched regulons of the ten subtypes (Fig. 2F). The top regulons of Cancer cells_TGFBI included PRRX1, MEIS2, MAFF, and NFIA, all of which were associated with the progression of different cancers (Fig. S2E).

We then use hotspot analysis to group the genes into modules and identify functional gene modules that illustrate heterogeneity within cancer cell subpopulations. This resulted in 14 transcriptional modules, and the module scores are presented in Fig. 2G. Next, the 14 modules were aggregated into multiple recurrent expression programs. Six common expression programs with different cell statuses and biological functions were identified (Fig. 2H). According to the hallmark analysis of these expression programs, four signatures were identified: EMT, UV-response-DN, interferon alpha (IFN-α) response, PI3K-AKT-mTOR signaling (Figs. 2H, I and S2F). Then, the Jaccard similarity coefficient was used to estimate the similarity between gene modules and cancer cell subtypes, and Cancer cells_TGFBI can correspond to Modules 9–11 (Fig. 2J). Moreover, gene set scoring analysis demonstrated Cancer cells_TGFBI had relatively high enrichment in EMT, stemness, and hypoxia (Figs. 2K and S2G, H).

The top expressed gene SERPINE2 in Cancer cells_TGFBI plays a key role in tumor progression

Since the above analysis indicated Cancer cells_TGFBI to be the key subtypes in tumor progression, we then revealed the differentially expressed genes (DEGs) in this group. As shown in Table S4, SERPINE2 is the top one DEG in Cancer cells_TGFBI. The expression of SERPINE2 was examined in nevi, primary tumor and metastatic melanoma samples by immunohistochemical staining. Results demonstrated the increased expression of SERPINE2 with tumor progression (Fig. 3A and B). Moreover, the mRNA expression was also higher in primary tumor samples than in normal skin, and expression in metastatic samples was much higher than in primary samples (Fig. 3C). We compared the SERPINE2 expression in melanoma cell lines and the primary melanocytes. It was found that A375 expressed the highest SERPINE2, compared with the other melanoma cell lines (Fig. 3D).

Fig. 3. The biological function of SERPINE2 in melanoma.

A The expression of SERPINE2 in nevi, primary tumor and metastatic tumor. B The SERPINE2 positive cells in (A). C The mRNA level in normal skin, primary tumor tissues and metastatic tumor. D The protein expression of SERPINE2 in primary melanocytes and melanoma cell lines (A375, A2058, HMY-1, SK-MEL-1, SK-MEL-5, and SK-MEL-28). E The construction of SERPINE2 knockdown (shSERPINE2) A375 cell line. F The proliferation of shNC and shSERPINE2 A375 cells. Representative images of colony formation ability of shNC and shSERPINE2 A375 cells (G) and the statistical analysis (H). Representative images of cell invasion ability of shNC and shSERPINE2 A375 cells (I) and the statistical analysis (J). K The WB results show EMT markers' expressions of shNC and shSERPINE2 A375 cells. L Nude mice subcutaneously implanted with shNC or shSERPINE2 A375 cells, the tumors were circled with a dotted line. M The photo of tumors dissected from mice in (L). N Tumor volume in (M). O The tumor progression curve over 10 days. The immunohistochemical staining of SERPINE2 (P) and its quantitative analysis (Q). The immunohistochemical staining of Ki67 (R) and its quantitative analysis (S). The immunofluorescent staining of TUNEL (T) and its quantitative analysis (U) (* indicates P < 0.05; ** indicates P < 0.01; *** indicates P < 0.001).

The SERPINE2-KD cell line (shSERPINE2) was established using the A375 and SK-MEL-28 cell lines (Fig. 3E and S3A). The effect of SERPINE2 expression on melanoma cell proliferation and invasion was examined via colony formation and CCK-8 cell proliferation assay, and Transwell cell invasion assay. As shown in Figs. 3F–H and S3B–D, the cell proliferation and colony formation ability were significantly decreased in the shSERPINE2 group, and the invasion ability was also impaired after SERPINE2 knockdown (Figs. 3I, J and S3E, F). Moreover, western blot results showed that the expression of EMT activation markers was decreased in shSERPINE2 cells (Figs. 3K and S3G), indicating the potential role of SERPINE2 in the metastasis program.

We then investigated the in vivo tumor growth of shNC and shSERPINE2 cells by subcutaneously injecting cells in nude mice. Results demonstrated that the tumor growth was significantly inhibited in the shSERPINE2 group 10 days after implantation (Fig. 3L and M), and the tumor volume (Fig. 3N and O) and tumor weight (Fig. S3H) in the shSERPINE2 group were smaller by statistical analysis. The immunohistochemical staining of SERPINE2 (Fig. 3P and Q) and Ki67 (Fig. 3R and S) further demonstrates the successful construction of the shSERPINE2 melanoma animal model and the decreased proliferation in vivo. Moreover, the immunofluorescent staining of TUNEL indicated the increased cell apoptosis in the shSERPINE2 group (Fig. 3T and U).

The differentiation states of cancer cell subtypes

We then projected melanocytes and cancer cells onto the two-dimensional state-space defined by Monocle2 for pseudotime analysis, to infer lineage trajectory for cell development (Fig. 4A). A total of three differentiation states were captured, of which state 1 was the earliest stage and was then separated into states 2 and 3 (Fig. 4B). Our result showed that the aforementioned ten tumor cell subtypes and 4 melanocyte subtypes were in different developmental states. Melanocytes were mainly at the root of the phylogenetic tree, which was consistent with the benign status of those cells (Fig. 4C). The GO, KEGG, and hallmark enrichment were also conducted for genes that changed with developmental time. The hallmark analysis demonstrated high enrichment of almost every signature, while state 2 had enrichment in UV_RESPONSE_DN, angiogenesis and hedgehog signaling, and state 3 in UV_RESPONSE_UP, MYC_TARGETS_V1, and oxidative phosphorylation (Fig. 4D).

Fig. 4. Differential gene expression profiles along melanoma progression.

A Pseudotime analysis of cancer cells inferred by Monocle2. Each point corresponds to a single cell. B Indication of three states along the trajectory. C The developmental states along the pseudotime trajectory of cancer cells and melanocytes. D Hallmark gene sets analysis was applied to the 3 states. E The differentially expressed genes along the pseudo-time are clustered into 14 clusters and three patterns. The representative pathways of each pattern were shown. F RNA Velocity analysis illustrated the direction of cell differentiation and developmental lineages of cancer cells and melanocytes. G GeneSwitches analysis of pathway activity changes in the transition from state 1 to state 3 in pseudotime analysis.

Next, we analyzed the single-cell transcriptomes along the trajectory and identified genes with significant expression changes, which could be clustered into three expression patterns (Fig. 4E). Cell fate 1 and fate 2 represent states 2 and 3, respectively. Pattern 1 included genes that showed increased expression levels along the trajectory of states 1 to 3. Pathway enrichment analysis revealed that these genes were associated with cell-cycle signaling and IFN-γ response. The genes in pattern 2 were upregulated along the trajectory of states 1 to 2, and these genes participated in cell adhesion, EMT, and various signaling pathways that were involved in tumor progression. The genes in pattern 3 were activated at the early stage of development, and were enriched in cell development, translation, cell differentiation, response to ROS, etc.

The pseudotime analysis was also conducted in cancer cells alone (Fig. S4A and B). Cancer cells in the primary group were mainly at state 1 and 2, m_S group were at state 1 and 3, and m_L group were at state 2 and 3 (Fig. S4C). Results in Supplementary Fig. 4D demonstrated that those 10 cancer cell subtypes were in different developmental states. The Cancer cells_TGFBI, PIZEO2, and HMCN1 were mainly located in the roots of the phylogenetic trees, which means those subtypes may be primitive tumor cells. While Cancer cells_BRD9, ENTHD1, LHFPL3, and MIR4300HG were in the late stages of development and well-differentiated.

To further analyze the time-resolved phenomena of tumor cell evolution, the RNA velocity method was employed to investigate developmental lineages and cellular dynamics. Melanocyte_PTN was likely upstream of Melanocyte_SDC2 and FGF14. Notably, Melanocyte_DNAJB1 presented developing trends toward cancer cells, and Cancer cells_PKHD1 and TGFBI had trends toward Cancer cells_PIEZO2, LHFPL3, CTSL, BRD9, and CEMIP (Fig. 4F and S4E).

GeneSwitches analysis identified multiple biological processes that were activated along the pseudo-timeline. Along the timeline from state 1 to state 2 (P1) in Fig. 4B, many proteins in the ribosomal protein family, including RPL27A, RPL32, RPL37A, etc., were downregulated at the early stage. And genes involved in growth delay and reduction of metastatic potential, GPNMB, were also downregulated (Fig. S4F), indicating the progression of AM. Cancer cells_TGFBI were in this timeline. Moreover, in the timeline from state 1 to state 3 (P2), melanin-related genes, PMEL, PRAME, and OCA2, were upregulated at the late stage (Fig. S4G and H). We then performed pathway enrichment analysis on key genes in different pathways. Ribosome-related pathways were downregulated in both paths at the early stages (Fig. S4I and J), while pathways like kinase activity, microtubule-related, and nuclear-related were upregulated at late stages in P2 (Fig. 4G).

Identification of CAFs in the TME

CAFs are crucial components in TME, and they play a crucial role in orchestrating TME reorganization, involved in processes such as angiogenesis, extracellular matrix remodeling, tumor progression, and immune evasion. Among all four groups, a total of 20126 fibroblasts were identified, and were further divided into seven subtypes (Fig. 5A). The clusters in each sample were also analyzed (Fig. 5B). It can be seen from the proportion of CAFs in each group showed in Fig. 5C, the proportion of Fibroblast_MMP2 gradually decreased when tumor progressed from nevi to skin metastasis, while Fibroblast_POSTN increased along the process. The top expressed genes in each subtype were listed in tracksplot (Fig. 5D). The GSVA based on hallmark gene sets was used to identify potential biological functions and relevant signaling pathways. Specifically, Fibroblast_MMP2 was highly enriched in almost every signature; Fibroblast_TNXB was enriched in PEROXISOME, MITOTIC_SPINDLE, and HEME_METABOLISM; Fibroblast_APCDD1 was enriched in IFN-α and IFN-γ response; Fibroblast_POSTN was enriched in EMT, angiogenesis, E2F_TARGETS, G2M_CHECKPOINT, and various signaling pathways (Fig. 5E). Moreover, the GO analysis of the top biological functions was also conducted (Fig. 5F and S5A). Notably, Fibroblast_POSTN was associated with extracellular matrix organization and collagen metabolic process; Fibroblast_MMP2 was associated with regulation of the immune system and lymphocyte differentiation; Fibroblast_TNXB was associated with cell junction assembly, developmental pigmentation, and Ras protein signaling transduction.

Fig. 5. Fibroblast heterogeneity in melanoma.

A UMAP plot depicting 7 subtypes of fibroblasts identified by snRNA-seq. B Subtypes of fibroblast were displayed as samples. C The composing proportion of each subtype in n, p, m_S, and m_L groups. The relative contribution of each population was weighed by the number of cells and scaled to 100%. D The tracksplot of the top-expressed genes among 7 subtypes in fibroblasts. E Hallmark gene sets analysis was applied to the 7 fibroblast subtypes. F The GO analysis of the top biological functions of Fibroblast_POSTN, Fibroblast_MMP2, and Fibroblast_TNXB. G The dot plot of 7 fibroblast subtypes with the corresponding selected marker genes for each CAF type showing the expression level of those markers. H The pySCENIC analysis was used to determine essential regulons of the 7 subtypes of CAFs.

It has been reported that CAFs can be divided into inflammatory CAFs (iCAFs), matrix CAFs (mCAFs), myofibroblastic CAFs (myCAFs), secretory-papillary CAFs (spCAFs), and secretory-reticular CAFs (srCAFs)^5–7. spCAFs and srCAFs are generally located in the papillary dermis and reticular dermis, respectively, and express corresponding markers. The signatures of iCAFs include inflammatory response, cell chemotaxis, immune trafficking, and complement regulation. mCAFs can participate in processes including matrix remodeling and production, inflammatory factors production, facilitating cell adhesion and migration. myCAFs are associated with smooth muscle contraction, focal adhesion, ECM organization, and collagen formation^6,8. Makers of the above CAFs were used to classify the seven subtypes of our results (Figs. 5G and S5B). Fibroblast_MMP2 can be classified as iCAFs, Fibroblast_POSTN as mCAFs, Fibroblast_APCDD1 as spCAFs, Fibroblast_IGF1 and CREB5 as srCAFs, and Fibroblast_SLC22A3 as myCAFs. The feature plot of top-expressed genes in each subtype is shown in Fig. S6. This classification can also be verified by the hallmark and GO enrichment results.

Via pySCENIC analysis, we determined essential regulons of the 7 subtypes of CAFs. Immune regulation genes like ANXA1, CEBPD, KLF6 were enriched in Fibroblast_MMP2; the genes STAT2, BACH2, and NFIL3 associated with immune and SIX4, TWIST1 associated with cancer progression were enriched in Fibroblast_POSTN; muscle differentiation and regeneration related genes MEOX2 and PAX7 were highly enriched in Fibroblast_SLC22A3 (Figs. 5H and S5C). The collagen expression pattern has also been linked to fibroblast functions. The results indicated relatively lower global collagen expression levels of Fibroblast_SLC22A3 and TNXB. Cartilage development-associated collagen COL11A1⁹ was highly expressed in Fibroblast_POSTN, suggesting its mesenchymal component. COL13A1 and COL23A1, two known markers of papillary fibroblasts^10,11, and epidermal-dermal junction collagen COL18A1^12,13are relatively highly expressed in Fibroblast_APCDD1 (Fig. S5D). The above gene expression also corresponds to the classification of each subtype.

The differentiation states of fibroblast subtypes

The differentiation trajectory of the fibroblast was visualized by monocle 2. The chronological order of the trajectory was illustrated by the gradation of color (Fig. 6A). The distribution of the seven subtypes in the trajectory was presented in Fig. 6B and C. Fibroblast_MMP2 and SLC22A3 were mainly located at the late stage of the trajectory, which implied relatively mature differentiation. Fibroblast_CREB5, IGF1, and TNXB were dispersive, suggesting multiple differentiation states. Most cells of the subtypes Fibroblast_POSTN and SLC22A3 were presented at different locations of the trajectory, indicating completely different differentiation states. Consistent with the distribution of Fibroblast_MMP2, the top five expressed genes in Fibroblast_MMP2, which were CLU, CFD, GSN, LUM, and CXCL14, were also extremely concentrated at the late stage (Fig. 6D, F). While the top five expressed genes in Fibroblast_POSTON, which were POSTN, ADAM12, FN1, KIF26B, and COL11A1, were also consistent with the status of this subtype (Fig. 6E, F). Moreover, the top five expressed genes in other subtypes were in correspondence with their location in the trajectory (Fig. S7A).

Fig. 6. Differential gene expression profiles along CAFs.

A Pseudotime analysis of CAFs inferred by Monocle2. B Each CAF subtype was projected along the trajectory. C The developmental states along the pseudotime trajectory of CAFs. The expression of the most variable genes involved in Fibroblast_MMP2 (D) and Fibroblast_POSTN (E) state transition. F The top expressed genes in Fibroblast_MMP2 and Fibroblast_POSTN were projected along the trajectory. G Changes in pathway activities of CAF subtypes along the trajectory of pseudotime. GeneSwitches analysis of pathway activity changes in the transition in path 1 (H) and path 2 (I).

The expression patterns of CAFs along the trajectory were analyzed by GO, KEGG, and hallmark enrichment. The top 100 genes can be divided into two groups, which were those expressed at the early stage (group 1) and those expressed at the late stage (group 2) (Fig. S7B). Genes in group 1 had high enrichment in extracellular matrix organization, cartilage development, and other fibroblastic features. Genes in group 2 preserved fibroblastic features and developed other cancer-related signatures, including EMT, hypoxia, angiogenesis, response to oxidative stress, and proteoglycans in cancer (Fig. S7B).

To dissect the expression dynamics along the pseudotime trajectory, we conducted trajectory analysis of the regulation of extracellular matrix constituent secretion, wound healing, angiogenesis, and collagen biosynthetic process. It was found that those biological processes were upregulated along the CAFs transition trajectory (Fig. 6G). The pseudotime trajectory of CAFs can be divided into two main paths, which were path 1 (included states 1, 2, 6, 7) and path 2 (included states 1, 2, 3, 4) (Fig. S7C).

To investigate key genes involved in the developmental timeline of CAFs, GeneSwitches analysis was conducted. Along the timeline of path 1, represented by Fibroblast_POSTN, classified as mCAFs, cell proliferation-associated TFs (KLF4, FOSB, and JUND) were increased, cancer-related genes (S100A6, CD63, melanoma-associated; MT-CYB, MT-ND2, bladder cancer-associated; ACKR3, glioma-associated) were upregulated, and genes participating in collagen fibril assembly, cell adhesion, like MYADM, BGN, LUM, CCDC80, MFAP5, and MFAP4, were increased (Fig. S7D). Along the timeline of path 2, represented by Fibroblast_MMP2, classified as iCAFs, immune-related genes (CFD, CD81, SERPING1), fibroblastic features-related genes (ISLR, MFAP4) were increased (Fig. S7E and F).

GeneSwitches analysis identified differences in multiple biological processes that were downregulated or activated along these two paths. We performed GO and hallmark pathway enrichment analysis on key genes in different pathways. The pathway analysis results demonstrated that EMT, extracellular matrix, and cell substrate junction were increased with these two timelines, while extracellular structure organization was upregulated in path 1, of which Fibroblast_POSTN was one of the major components (Fig. 6H and I). Moreover, in the process of path 2, genes related to skeletal system development, collagen trimer, biological adhesion, and ECM receptor interaction were suspended (Fig. S7G).

The specific expression profiles of fibroblast subtypes

RNA velocity analysis further revealed the complex developmental trajectories within heterogeneous fibroblast populations in melanoma TME. As demonstrated in Fig. 7A, the arrowheads of Fibroblast_POSTN, Fibroblast_APCDD1, Fibroblast_IGF1, and Fibroblast_CREB5 are oriented in the same direction, indicating that these clusters have a stable state and appear to differentiate from the Fibroblast_TNXB. To further identify the gene expression pattern of each fibroblast subtype, we analyzed the common and unique up-regulated genes of subtypes (Fig. 7B) and their GO and KEGG enrichment. As shown in Fig. S7A and B, Fibroblast_MMP2 had enrichment in terms of extracellular matrix organization, T cell activation, leukocyte migration, and antigen processing and presentation, which further verified its classification as iCAFs (Fig. S8A). Fibroblast_POSTN had enrichment in collagen fibril organization, collagen metabolic process, and ECM–receptor interaction, which verified its classification as a mCAFs (Fig. S8B). A total of 133 genes were specifically up-regulated in Fibroblast_TNXB subtype (Fig. 7B), and GO and KEGG enrichment were also conducted regarding those genes (Tables S5 and S6). Terms including cell junction assembly, developmental pigmentation, and melanogenesis were enriched, indicating the correlation of Fibroblast_TNXB with melanocyte development (Fig. 7C).

Fig. 7. Fibroblast_TNXB may be the original subtype of CAFs and is associated with melanocyte development.

A RNA Velocity analysis illustrated the direction of cell differentiation and developmental lineages of CAFs. B The Venn diagram demonstrated the co-expressed or unique up-regulated genes in 7 subtypes of CAFs. C The GO (ⅰ) and KEGG (ⅱ) analysis of the 133 specifically up-regulated genes in Fibroblast_TNXB. D The GO (ⅰ) and KEGG (ⅱ) analysis of up-regulated pathways in m_S groups compared with the p group of Fibroblast_TNXB. E Scatter plot showed the RSSs in Fibroblast_TNXB and each group. The top 5 regulons are highlighted. F and G The immunofluorescent staining of melanoma samples. CD3 was the marker for T cells, gp100 was the marker for melanoma cells, vimentin was the marker for fibroblasts, TNXB was the marker for Fibroblast_TNXB, MITF and TYR were the markers for melanocytes, and the nucleus was stained with DAPI.

We also analyzed the gene expression patterns of Fibroblast_TNXB in 4 groups. Compared with the n group, malignancy-associated pathways, including PI3K-Akt signaling pathway, cGMP-PKG signaling pathway, and cAMP signaling pathways were enriched in p group (Fig. S8C). Compared with p group, pigmentation associated terms were up-regulated in m_S group, including developmental pigmentation, melanosome membrane, pigment granule membrane. Moreover, KEGG analyses demonstrated enrichment in ErbB signaling pathway, which has been reported in melanocyte maturation and regeneration^14,15, and PI3K-Akt signaling pathway, which has been reported to be involved in melanocyte’s survival, migration, and proliferation^16–18 (Fig. 7D). The pySCENIC analysis were applied to determine the essential regulons of Fibroblast_TNXB and those in 4 groups (Fig. 7E). Genes related with cell differentiation had increased expression in Fibroblast_TNXB, including TCF7, which participated in T cells, natural killer cells, and innate lymphoid cells development, ZNF467, which enhances adipocyte differentiation and suppresses osteoblast differentiation in bone marrow, ETV7, which plays an important role in a variety of cellular processes throughout development and differentiation, PAX3, which involved in neural development and myogenesis, and LEF1, which may be related with hair cell differentiation and follicle morphogenesis. The above transcription factors also indicated the initial status of Fibroblast_TNXB in all fibroblast clusters. The immunofluorescent staining of primary melanoma samples demonstrated the close relationship of fibroblast and immune cells (Fig. 7F), and co-location of Fibroblast_TNXB and melanocytes in melanoma (Fig. 7G).

Validation of our results in the published scRNA-Seq databases

We verified our analysis results in the previously published scRNA-Seq databases. In the GSE189889 database¹⁹, both the metastasis group and the primary group underwent clustering analysis, obtaining 7 cell types (Fig. 8A and S9A). The clinical composition was presented in Table S7. Density plots showing the distribution of the TGFBI-associated malignancy signature score in cancer cells, stratified by sample type (Primary vs. Metastasis). The signature score was calculated using the Seurat AddModuleScore function based on the top 20 marker genes of the TGFBI⁺ subpopulation. The shift to the right in the metastasis group indicates an enrichment of cells with a high-malignancy state (Fig. 8B). Furthermore, a stacked bar plot illustrates the compositional changes within the fibroblast lineage. We observed a substantial increase in the proportion of Fibroblast_POSTN subpopulation in the metastasis group compared with primary tumors (Fig. 8C), confirming the remodeling of the TME towards a pro-metastatic state during progression. Additionally, we found the expression of SERPINE2, the core marker of the CancerCells_TGFBI subpopulation, was significantly upregulated in metastasis samples compared to primary samples (P < 0.0001, Wilcoxon test, Fig. 8D), which is consistent with our results.

Fig. 8. Dataset verification of our results.

A UMAP plot depicting 7 types of cells in metastasis and primary samples of dataset GSE189889. B The malignant scores of CancerCells_TGFBI were higher in the metastasis group. C The proportion of Fibroblast_POSTN was increased in the metastasis group. D The expression of SERPINE2 was higher in the metastasis group. E The identification of fibroblasts in AM (ⅰ) and CM (ⅱ) samples of dataset GSE215121. F The subtypes of fibroblasts were identified in AM (ⅰ) and CM (ⅱ) samples. The TNXB (G) and POSTN (H) signatures of fibroblasts in AM (ⅰ) and CM (ⅱ) samples. I The signature of makers for cancer cells in AM (ⅰ) and CM (ⅱ) samples. J The TGFBI score of melanoma cells in AM (ⅰ) and CM (ⅱ) samples.

Furthermore, we also conducted verification in the GSE215121 database¹. The clinical composition was presented in Table S8. Firstly, we conducted clustering analysis on the AM and CM samples. The AM group yielded 11 cell types, while the CM group resulted in 12 cell types (Fig. S9B). Based on the marker COL1A1, fibroblast clusters were screened out. Among them, cluster 6 in the AM group and cluster 8 in the CM group were fibroblasts (Fig. 8E). We performed a second stage of clustering on the fibroblasts of the two groups (Fig. 8F). Combined with the top 20 markers of Fibroblast_TNXB and Fibroblast_POSTN in our analysis, we then verified them in this dataset, and results demonstrated clusters with the same independent feature (Figs. 8G, H and S9C, D). By cross-referencing with CM data, the results revealed that the Fibroblast_TNXB signature was highly enriched and defined a dominant subpopulation in AM. In contrast, in CM, this signature was expressed at low levels across the vast majority of clusters and failed to form a dominant subpopulation comparable to that in AM. Furthermore, we selected cancer cell clusters based on common melanoma markers (MLANA, MITF, SOX10, PMEL, TYR) and then conducted a re-clustering analysis (Fig. 8I), and the subclusters of cancer cells were also analyzed (Fig. S9E). The top 20 markers of Cancer cells_TGFBI that we previously analyzed were also identified as independent clusters in AM (Figs. 8J and S9F).

Through this dual-dataset validation, we have not only confirmed that CancerCells_TGFBI (along with SERPINE2) is a key subpopulation driving AM progression (based on GSE189889) but also established that Fibroblast_TNXB represents an acral-specific microenvironmental feature (based on GSE215121).

Discussion

Cutaneous melanoma plays an important role in oncological research since it is one of the most aggressive skin cancers with high potential for metastasis. It is a malignancy of melanocytes with high heterogeneity, and it has become a significant public health concern due to the rising incidence globally²⁰. The advancements and application of genomic sequencing have identified key genetic mutations and pathways involved in melanoma pathogenesis^21,22. Compared with cutaneous melanoma, AM is very rare in Caucasian populations and is more common in Asian populations. It predominantly affects extremities, including the palms of the hands, soles of the feet, and nail beds²³. Studies about AM are relatively fewer than those about cutaneous melanoma, and there is a lack of analysis about intratumoural heterogeneity and TME of AM. Our study applied snRNA-Seq in AM with different stages and deciphered the AM heterogeneity and the underlying transcriptional programs that regulate melanoma cell diversity and behavior.

Serpin family E member 2 (SERPINE2) has anti-serine protease activity against thrombin, urokinase, and plasminogen²⁴, and has been reported to be a critical factor in vasculature, nervous system, and the behavior of cancer cells²⁵. SERPINE2 has been identified to play an important role in metastasis of renal cell carcinoma²⁶, liver cancer²⁷ and esophageal squamous cell carcinoma²⁸, and promote angiogenesis and lymphangiogenesis in oral squamous cell carcinoma²⁹. High SERPINE2 expression correlates with poor prognosis in lung adenocarcinoma³⁰ and urothelial carcinoma³¹. Moreover, SERPINE2 has been identified to be involved in the promotion of melanoma metastasis through the glycogen synthase kinase 3β³². These results were based on experiments in melanoma cell lines. However, in our study, we identified Cancer cells_TGFBI may be related to the progression of melanoma, and SERPINE2 was the top-expressed gene. We have verified in clinical samples that the expression of SERPINE2 in the metastatic lesions is higher than that in the primary lesions. It is proven to promote the proliferation and invasion of melanoma cells and induce EMT by decreasing the expression of the epithelial marker E-cadherin and increasing the level of N-cadherin, Snai1, and MMP9 expression to invade the ECM. We have also verified in animal models that SERPINE2 can promote tumor growth.

The heterogeneity of melanoma is influenced by the plasticity of melanoma cells and their dynamic adaptations to the diverse stromal and immune contexture of the TME³³. scRNA-seq and snRNA-seq have revealed the plasticity and heterogeneity of melanoma and can differentially adapt to microenvironmental cues³⁴. CAFs are one of the most prominent and abundant cell populations and play a crucial role in remodeling TME, participating in processes including tumor promotion, angiogenesis, ECM remodeling, and immune evasion^35,36. Studies about conjunctival melanoma have shown that higher abundance of CAFs in TME is correlated with increased angiogenesis and VEGFR expression in distal metastatic tumors³⁷.

CAFs can be divided into different subpopulations in various cancer types, including vascular CAFs (vCAFs), matrix CAFs (mCAFs), cycling CAFs, interferon-response CAFs (ifnCAFs), inflammatory CAFs (iCAFs), tumor-like CAFs (tCAFs), antigen-presenting CAFs (apCAFs), reticular-like CAFs (rCAFs), etc. mCAFs are related to cartilage and activated myofibroblasts development, matrix remodeling, and migration. iCAFs showed high expression of cytokines and chemokines and are associated with pro-inflammatory activities, and a study revealed their significant correlation with the response to immunotherapy in melanoma³⁸. vCAFs demonstrated a link to angiogenesis. tCAFs have a gene expression signature resemble that of tumor cells and is associated with chemoresistance. ifnCAFs showed a strong upregulation of IFN-α and IFN-γ responses. apCAFs have the top differentially expressed marker genes in antigen presentation. rCAFs express markers of reticular fibroblasts in lymphoid tissues that facilitate the homing of naïve T cells. Although each subtype has a dominant biological feature or function, they all have subgroups with different expression patterns and other biological functions⁷.

The relationship between CAFs and tumor progression is complicated. It has been reported that CAFs not only secrete cytokines into the TME but also produce a stiff ECM that surrounds the tumor³⁹. Galbo Jr et al. ⁴⁰ found that myCAF and iCAF have significant enrichment in metastatic melanoma patients who experienced progressive disease while receiving anti-PD-1, which indicated they may have an immunosuppressive role in the TME. Other researchers identified that CAF receptor Endo180 (Mrc2) is predominantly expressed on myCAFs, and high MRC2 mRNA levels in melanoma were associated with a poor response to αPD-1 therapy⁴¹. The work of Verginadis et al. ⁴² highlights the importance of ATF4 in regulating the functionality and activation of CAFs. The downregulation of ATF4 led to the decreased function of CAFs and subsequently caused pronounced growth delay of melanoma. Forsthuber et al. ⁴³ reported that the number of iCAFs increases significantly in aggressive tumors (e.g., late-stage melanoma), which indicates that a high abundance of iCAFs correlates with malignant progression. Moreover, fibroblasts synthesize most cytokines and chemokines and are key players in immune cell recruitment and activation. Furthermore, they found mCAFs ensheath tumor nests and synthesize large amounts of ECM proteins, which suggests they are involved in T cell exclusion.

In our study, we identified iCAFs, mCAFs, spCAFs, srCAFs, and myCAFs according to their expression patterns. And they all demonstrated typical biological functions as previously reported. Typically, Fibroblast_POSTN was identified as mCAFs, and the proportion gradually increased with melanoma progression, indicating its association with tumor metastasis. Moreover, iCAFs, Fibroblast_MMP2, were decreased along tumor progression, which may be due to the decreased immune cells in melanoma tissue. However, Fibroblast_TNXB cannot be classified into the current subpopulations. Analysis demonstrated that it might be associated with melanocytes' survival, migration, proliferation, and differentiation. The multi-immunofluorescent staining also demonstrated the co-location of Fibroblast_TNXB and melanocytes in melanoma samples. Further studies should be focused on the experimental verification of the relationship between Fibroblast_TNXB and melanocyte.

A limitation of the present study is the exclusive use of conventional cutaneous melanoma cell lines (A375 and SK-MEL-28), as acral melanoma cell lines were not easily accessible to us in China. Future validation in acral melanoma models is warranted. We are currently pursuing several approaches to address this issue, including establishing primary acral melanoma cell cultures from surgical specimens and exploring potential collaborations with other Chinese research groups that have access to acral melanoma models. Additionally, in subsequent experiments, the experimental parameters will be optimized and the observation period for tumor‑bearing mice will be extended, so as to generate more reliable in vivo data.

In summary, our results described the landscape of malignant, stromal, and CAFs for benign lesions, primary tumors, skin metastases, and lymph nodes metastases of AM. Although further research regarding the relation of SERPINE2 and melanoma progression is needed, the biological function found in this study may help develop new treatment strategies for AM. Fibroblast_TNXB was found to be associated with melanocyte development, and its function needs further research. Our identification of the potential progression marker SERPINE2 and a new CAF subpopulation may facilitate understanding of melanoma progression and TME.

Methods

Nuclei isolation sorting from nevi and melanoma

Nevi and melanoma tissues were harvested and washed in pre-cooled PBSE (PBS buffer containing 2 mM EGTA). Nuclei isolation was carried out using GEXSCOPE® Nucleus Separation Solution (Singleron Biotechnologies, Nanjing, China), as referred to in the manufacturer’s product manual. Isolated nuclei were resuspended in PBSE to 106 nuclei per 400 μL, filtered through a 40 μm cell strainer, and counted with Trypan blue. Nuclei enriched in PBSE were stained with DAPI (1:1000) (Thermo Fisher Scientific, USA). Nuclei were defined as DAPI-positive singlets.

Single-nucleus RNA-sequencing library preparation

The concentration of single-nucleus suspension was adjusted to 3–4 × 10⁵ nuclei/mL in PBS. Single-nucleus suspension was then loaded onto a microfluidic chip (GEXSCOPE® Single Nucleus RNA-seq Kit, Singleron Biotechnologies) and snRNA-seq libraries were constructed according to the manufacturer’s instructions (Singleron Biotechnologies). The resulting snRNA-seq libraries were sequenced on an Illumina Novaseq 6000 instrument with 150 bp paired-end reads.

Primary analysis of raw read data

Raw reads were processed to generate gene expression profiles using CeleScope v1.14.0 (Singleron Biotechnologies) with default parameters. Briefly, Barcodes and UMIs were extracted from R1 reads and corrected. Adapter sequences and poly A tails were trimmed from R2 reads, and the trimmed R2 reads were aligned against the GRCh38 (hg38) transcriptome using STAR (v2.6.1b). Uniquely mapped reads were then assigned to genes with Feature Counts (v2.0.1). Successfully Assigned Reads with the same cell barcode, UMI, and gene were grouped together to generate the gene expression matrix for further analysis.

Quality control, dimension-reduction, and clustering

Scanpy v1.8.1 was used for quality control, dimensionality reduction, and clustering under Python 3.7. For each sample dataset, we filtered the expression matrix by the following criteria: (1) cells with gene count less than 200 or with top 2% gene count were excluded; (2) cells with top 2% UMI count were excluded; (3) cells with mitochondrial content å 5% were excluded; (4) genes expressed in <5 cells were excluded. After filtering, 217111 cells were retained for the downstream analyses, with on average 989 genes and 1844 UMIs per cell. The raw count matrix was normalized by total counts per cell and logarithmically transformed into normalized data matrix. The top 2000 variable genes were selected by setting flavor = ‘seurat’. Principle component analysis (PCA) was performed on the scaled variable gene matrix, and the top 20 principal components were used for clustering and dimensional reduction. Cells were separated into 28 clusters by using the Louvain algorithm and setting the resolution parameter at 1.2. Cell clusters were visualized by using Uniform Manifold Approximation and Projection (UMAP)⁴⁴.

Differentially expressed genes (DEGs) analysis

To identify DEGs, we used the scanpy.tl.rank_genes_groups function based on the Wilcoxon rank sum test with default parameters, and selected the genes expressed in more than 10% of the cells in either of the compared groups of cells and with an average log(Fold Change) value greater than 1 as DEGs. Adjusted p-value was calculated by Benjamini–Hochberg correction, and the value 0.05 was used as the criterion to evaluate the statistical significance.

Cell differentiation potential evaluation: CytoTRACE

CytoTRACE v0.3.3⁴⁵ (a computational method that predicts the differentiation state of cells from single-cell RNA-sequencing data using gene counts and expression) was used to predict the differentiation potential of cell subpopulations.

Jaccard similarity analysis

In order to estimate the similarity between cancer cell subtypes, the Jaccard similarity coefficient⁴⁶ was measured between genes from modules derived from hotspot and the top 50 cell type marker genes.

UCell gene set scoring

Gene set scoring was performed using the R package UCell v 2.2.0⁴⁷. UCell scores are based on the Mann–Whitney U statistic by ranking query genes in order of their expression levels in individual cells. Because UCell is a rank-based scoring method, it is suitable to be used in large datasets containing multiple samples and batches.

scRNA-seq-based copy number analysis (CNA) detection

The InferCNV package⁴⁸ was used to detect the CNAs in cancer cells. ECs were used as baselines to estimate the CNAs of malignant cells. Genes expressed in more than 20 cells were sorted based on their loci on each chromosome. The relative expression values were centered at 1, using 1.5 standard deviations from the residual-normalized expression values as the floor and ceiling. A slide window size of 101 genes was used to smooth the relative expression on each chromosome, to remove the effect of gene-specific expression. The inferred CNAs on each short or long arm or full length of the chromosomes were visualized with heatmaps generated by the R pheatmap function. The clonal relationships of malignant cell clusters in each sample were determined by the accumulation of CNAs. UPhyloplot2 software was used to draw clonality trees of malignant cell clusters. The CNA events were manually labeled on the branches. The CNV score of each cell was calculated as the quadratic sum of the CNA region.

For each cell, the CNV score is defined as the average of the absolute values of the inferred CNVs, obtained by subtracting 1 from the CNV values in the matrix output by inferCNV.

Cell-type recognition with Cell-ID

Cell-ID is a multivariate approach that extracts gene signatures for each individual cell and performs cell identity recognition using hypergeometric tests (HGT)⁴⁹. Dimensionality reduction was performed on a normalized gene expression matrix through multiple correspondence analysis, where both cells and genes were projected into the same low-dimensional space. Then, a gene ranking was calculated for each cell to obtain the most featured gene sets of that cell. HGT was performed on these gene sets, reference from SynEcoSys^TM database, which contains all cell-type’s featured genes in the specific organ/tissue⁵⁰. The identity of each cell was determined as the cell type with the minimal HGT p-value. For cluster annotation, the frequency of each cell-type was calculated in each cluster, and cell-type with the highest frequency was chosen as the cluster’s identity. The cell type identification of each cluster was determined according to the expression of canonical markers from the reference database SynEcoSys^TM (Singleron Biotechnology). SynEcoSys^TM contains collections of canonical cell type markers for single-cell seq data, from CellMakerDB, PanglaoDB, and recently published literature. The canonical markers and their corresponding cell types were listed in Table S1.

Subtyping of major cell types and filtering cell doublets

To obtain a high-resolution map of Cancer Cells/Monocytes/Melanocytes/MPs/T and NK/Mural Cells/Keratinocytes/Macrophages/ECs, cells from the specific cluster were extracted and reclustered for more detailed analysis following the same procedures described above⁵⁰ and by setting the clustering resolution as 0.3/1/0.5/1.2/1.2 /1/0.3/0.5/1.2.

Cell doublets were estimated based on the expression pattern of canonical cell markers. Any clusters enriched with multiple cell-type-specific markers were excluded from downstream analysis.

Pathway enrichment analysis

To investigate the potential functions of different clusters, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis were used with the “clusterProfiler” R package v 4.0.0⁵¹. Pathways with p_adj value < 0.05 were considered significantly enriched. Selected significant pathways were plotted as bar plots. GSEA was also performed⁵². For GSVA pathway enrichment analysis, the average gene expression of each cell type was used as input data⁵³. Gene Ontology gene sets, including molecular function (MF), biological process (BP), and cellular component (CC) categories, were used as reference.

Pseudotime trajectory analysis: monocle2

Cell differentiation trajectory of monocyte subtypes was reconstructed with the Monocle2 v 2.22.0⁵⁴. For constructing the trajectory, top 2000 highly variable genes were selected by Seurat (v3.1.2) FindVairableFeatures, and dimension-reduction was performed by DDRTree. The trajectory was visualized by plot_cell_trajectory function in Monocle2.

RNA velocity

For RNA velocity, a BAM file containing cancer cells and fibroblasts and reference genome GRCh38 (hg38) was used in the analysis with velocyto (v0.2.3) and scVelo (v0.17.17) in Python with default parameters. The result was projected to the UMAP plot from Seurat clustering analysis for visualization consistency^55,56.

Functional gene module analysis (Hotspot)

Hotspot was used to identify functional gene modules that illustrate heterogeneity within cancer cells' subpopulations⁵⁷. Briefly, we used the ‘danb’ model and selected the top 500 genes with the highest autocorrelation z-score for module identification. Modules were then identified using the create_modules function, with min_gene_threshold = 15 and fdr_threshold = 0.05. Module scores were calculated by using calculate_module_scores function.

Quantitative real-time PCR (qRT-PCR)

Total RNA was extracted and reverse transcribed into cDNA. Gene transcripts were quantified by SYBR Green Real-Time PCR Master Mix (Thermo Fisher Scientific, USA) and normalized to the β-actin expression. The sequences of primers were as follows: SERPINE2: 5’-TGGTGATGAGATACGGCGTAA-3’, 5’-GTTAGCCACTGTCA CAATGTCTT-3’; β-actin: 5’-CATCCTCACCCTGAAGTACCC-3’, 5’-AGCCTGGATAGC AACGTACATG-3’.

Lentivirus transfection

The plasmid of shRNA SERPINE2 (shSERPINE2) was produced by BioeGene (China), and the sequence used is 5’-CAGTGTATTTCAAGGGTCT-3’. The lentiviral vector was constructed as reported. The stable SERPINE2-KD A375 cell lines and SK-MEL-28 cell lines were constructed using sh-SERPINE2 lentivirus with the assistance of Lipofectamine 3000 reagent (Thermo Fisher Scientific, USA).

Cell function assays

For the CCK-8 assay, A375 NC or SERPINE2-KD cells, SK-MEL-28 NC or SERPINE2-KD cells were seeded into 96-well plates and incubated at 37 °C, 5% CO₂ for 4 days. The CCK8 kit (Sigma-Aldrich, USA) was used to detect the proliferation of cells at days 1–4.

For the colony formation assay, cells were seeded into six-well plates at a density of 800 cells per well and were incubated at 37 °C in 5% CO₂ for 12 days to allow colony formation. The colonies were then fixed with 4% paraformaldehyde, stained with 0.5% crystal violet, and colonies containing ≥50 cells were counted under a microscope.

For invasion experiments, transwell chambers (8 µm, Corning, USA) were used with Matrigel (BD Biosciences, USA). A375 NC or SERPINE2-KD cells, SK-MEL-28 NC or SERPINE2-KD cells were seeded into the upper chamber and cultured with medium without FBS, and medium with FBS was added into the lower chamber. The cells were incubated at 37 °C, 5% CO₂ for 24 h, and cells at the lower side of the chamber were fixed with 4% Paraformaldehyde Fix Solution (Sangon Biotech, Shanghai), stained with crystal violet (Sangon Biotech).

Multiplexed immunofluorescence

After deparaffinization and rehydration, antigen retrieval was performed in citrate sodium antigen retrieval buffer (pH 8.0) (YEASEN, 36319ES60) using a microwave oven. Tissue sections were first heated using medium-high power for 8 min, then stopped for 8 min, and then using medium-low power for 7 min. Wash tissue section 3 times with PBS buffer for 5 minutes each time, and then blocked in goat serum at room temperature for 1 h. Tissue sections were then incubated overnight with primary antibodies (CD3, Abcam, ab135372; gp100, Abcam, ab137078; Vimentin, Abcam, ab20346; TNXB, Abcam, ab183346; MITF, Abcam, ab303530; TYR, Abcam, ab170905) overnight at 4 °C. After three washes with PBS for 5 min each time, tissue sections were incubated with secondary antibodies for 1 h at room temperature in the dark. After an incubation with DAPI (Servicebio, GB1012) and anti-fluorescence attenuation buffer (Servicebio, China; GB1401), images of stained tissue sections were captured using a fluorescence microscope (Zeiss LSM980).

Western blot analysis

Total protein of shNC and shSERPINE2 A375 cells was extracted using SDS-PAGE and then transferred to the PVDF membrane (Thermo, USA). The membrane was incubated with the primary antibodies: SERPINE2 (ab134905, Abcam), β-actin (#4967, CST), Tubulin (ab215037, Abcam), E-caderin (ab314063, Abcam), N-caderin (ab76011, Abcam), Snail (#3879, CST), MMP3 (#14351, CST), and MMP9 (#3852, CST). The membrane was incubated with a 1:2000 diluted horseradish peroxidase conjugated goat anti-rabbit (#7074, CST).

Animal model of melanoma

All animal experiments were approved by the ethical review board of The First Affiliated Hospital of the Naval Medical University (Shanghai, China). The 6-week-old male nude mice were housed in a specific pathogen-free (SPF) facility under a 12-h light/dark cycle at a constant temperature (22–24 °C) and humidity (50–60%). Autoclaved water and irradiated food were provided ad libitum. All mice were allowed to acclimatize for one week prior to the start of the experiment.

For tumor cell inoculation, shSERPINE2 or shNC A375 cells in the logarithmic growth phase were harvested and resuspended in sterile phosphate-buffered saline (PBS) at a concentration of 5 × 10⁶ cells/mL. Each group included 8 mice. Mice were anesthetized by inhalation of 2% isoflurane in oxygen to minimize suffering and to ensure immobility during the procedure. A total of 200 μL of the cell suspension (containing 1 × 10⁶ cells) was injected subcutaneously into each mouse using a 27-gauge needle. After injection, mice were monitored until full recovery from anesthesia. Tumor growth was monitored every 2 days by measuring the diameters of tumors.

All mice were humanely euthanized by cervical dislocation under deep anesthesia induced by isoflurane after 10 days of tumor implantation. Tumor tissues were then carefully excised and fixed in 10% neutral buffered formalin for immunofluorescence staining of TUNEL and immunohistochemical staining of shSERPINE2 and Ki67.

Gene signature scoring and visualization

To evaluate the malignancy status of cancer cells, a specific “TGFBI-associated signature” was defined based on the top 20 significantly upregulated genes in the TGFBI+ subpopulation (e.g., TGFBI, SLC38A1, SERPINE2). A composite signature score was calculated for every single cell using the AddModuleScore function in Seurat, which subtracts the aggregated expression of control feature sets from the target feature set to control for technical noise. The distribution of these scores across primary and metastatic groups was visualized using kernel density estimation plots (generated via ggplot2 and ggridges packages in R) to assess the enrichment of high-grade malignant cells in metastatic tissues.

Statistical analyses

GraphPad Prism 10.0 was used to analyze the data from at least three independent experiments. Student's t-test was employed to compare the difference between two groups, and differences among >two groups were subjected to one-way ANOVA. P < 0.05 was set as statistical significance.

Ethical approval

This research was performed in accordance with the Declaration of Helsinki. All procedures regarding sample collection and animal experiments were approved by the ethical review board of The First Affiliated Hospital of the Naval Medical University (CHEC2025-022). Informed consent to participate in the study has been obtained from all participants

Author contributions

Conception and design: Ji Zhu and Chunyu Xu; Development of methodology: Minliang Wu, Kexin Chen, and Yuchong Wang; Acquisition of data: Minliang Wu, Kexin Chen, Yuchong Wang, and Jianguo Xu; Analysis and interpretation of data: Minliang Wu, Kexin Chen, Ang Li, and Chaoying Jin; Writing, review, and/or revision of the manuscript: Minliang Wu, Kexin Chen, Yuchong Wang, Ji Zhu, and Chunyu Xue.

Data availability

The sequencing data reported in this paper had been uploaded in GSA for human database (accession number: HRA016925), and OMIX database under GSA (OMIX ID: OMIX015260). These data have been deposited in the Genome Sequence Archive under project PRJCA058317.

Code availability

The complete analysis code has been uploaded to the GitHub website (https://github.com/SingleronBio/celelenscloud).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41698-026-01618-4.