Exploring new mechanisms in cancer molecular pathways and pathogenic cell transformation: PIP4K2A as a prognostic marker and therapeutic target in cutaneous malignant melanoma Authors: Wen-fei Luo, Ding-ming Song, Tong Shen, Yi-bo He, Hong-yang Du, Ming-jue Si, Lu-Wei Fang Published: 2024 ABSTRACT BackgroundCutaneous malignant melanoma is a very aggressive and metastatic form of skin cancer, typically linked with poor outcomes. Advances in genomic analysis have underscored the crucial role of T cells in tumor immunity. Immune checkpoint inhibitors have notably transformed melanoma treatment by boosting T cell activity. Studies of gene expression have found that the phosphatidylinositol-4-phosphate kinase 2A (PIP4K2A) gene is abnormally expressed in various tumors, indicating its potential role in tumor progression. Utilizing single-cell sequencing and machine learning, researchers can now explore the complex interactions between T cells and melanoma cells at a genomic level. This study aimed to investigate the role of the PIP4K2A gene in cutaneous malignant melanoma, with a focus on its influence on T cell-mediated immune responses.MethodsSamples from cutaneous melanoma patients were analysed by single-cell transcriptome for differentially expressed genes and signalling pathways associated with cutaneous melanoma. Then, genes were identified and predictive models were built based on the transcriptomic data using machine learning models to assess whether the expression level of PIP4K2A could effectively predict the malignancy and prognosis of cutaneous melanoma. In addition, we also performed drug therapy predictive analysis and immunotherapy analysis.Finally, the critical role of PIP4K2A in cutaneous melanoma was further confirmed by immunohistochemistry.ResultsThe PIP4K2A gene exhibited a significantly elevated expression level in cutaneous malignant melanoma, showing a strong correlation with the clinical stage and patient prognosis. At the therapeutic level, high PIP4K2A expression is less responsive to immunotherapy, and this gene is a risk factor for drug therapy in cutaneous malignant melanoma. Additionally, our experimental outcomes validated this observation.ConclusionsThe PIP4K2A gene could be a crucial prognostic marker for cutaneous malignant melanoma, as it significantly affects T cell activity within the tumor microenvironment. This study offers essential insights into melanoma pathogenesis and assists in pinpointing new early diagnostic markers and therapeutic targets. Utilizing advanced genomic tools and computational techniques, the research enhances our understanding of T cell dynamics in melanoma, facilitating the development of personalized medicine and more effective immunotherapy strategies.Supplementary InformationThe online version contains supplementary material available at 10.1007/s12672-024-01555-3. Introduction Cutaneous malignant melanoma (Malignant Melanoma) is a malignant tumor originating from the melanocytes of the skin, the incidence of which has shown a significant increase in the last decades [1]. Although cutaneous malignant melanoma accounts for a small fraction of all skin cancer cases, it has the highest mortality rate among skin cancers due to its aggressive and metastatic behavior [2]. The development of cutaneous malignant melanoma is intricate, influenced by numerous genetic, environmental, and lifestyle factors [3]. Among these, exposure to ultraviolet light, particularly intense sunlight during childhood and adolescence, is the foremost environmental risk factor [4]. Other significant risk factors include family history, immunosuppression, skin type, and the number and type of moles [5]. At the molecular level, mutations in genes such as BRAF, RAS, NF-1 and WT play an important role in the development of cutaneous malignant melanoma [6]. In particular, the BRAF V600E mutation, which appears in approximately 50% of patients with cutaneous malignant melanoma, has become an important target for targeted therapy. Despite notable progress in diagnosing and treating cutaneous malignant melanoma in recent years, the prognosis remains unfavorable. This is largely due to the absence of clear symptoms in its early stages and its aggressive and metastatic behavior [1]. Consequently, gaining a deeper understanding of the disease's pathogenesis and identifying novel early diagnostic markers and therapeutic targets are crucial for enhancing patient survival and quality of life. The PIP4K2A gene, known as phosphatidylinositol-4-phosphate kinase 2A, is an isoform within the PIP4K family and is situated on chromosome 10p12.2 [7]. PIP4K2A primarily resides in the cytoplasm and plasma membrane [8]. It is abundantly expressed in peripheral blood cells [9, 10]. This gene's encoded protein is crucial in the phosphatidylinositol signaling pathway, which is vital for various biological processes like cell growth, proliferation, survival, and migration [7]. Consequently, abnormalities in the PIP4K2A gene could disrupt these processes, potentially leading to tumor formation. Research indicates that the PIP4K2A expression profile influences PtdIns5P levels under stress, impacting the cellular response to oxidative stress [11, 12]. Furthermore, PIP4K2A transcript levels are significantly correlated with leukemia susceptibility, suggesting its critical role in leukemia survival outcomes [13, 14]. Elevated expression levels of PIP4K2A have been observed in breast cancer, where its increase is linked to reduced TP53 expression [12]. A meta-analysis revealed an association between PIP4K2A and colorectal cancer risk, with significant interactions between PIP4K2A rs2230469 and environmental factors affecting colorectal cancer risk. Additionally, rs2230469 is significantly associated with PIP4K2A expression in colonic tissues [15]. Machine Learning is a branch of Artificial Intelligence, a process of fitting predictive models to data or identifying groupings of information in data [16]. Its goal is to develop and apply algorithms that allow computers to learn from data and make predictions or decisions [17]. In recent years, machine learning has been increasingly used in bioinformatics analysis. For example, machine learning can be used for classification and clustering of gene expression data, prediction of gene function, prediction of protein structure, diagnosis and prognosis of diseases [17, 18]. The advantages of machine learning are that it can handle large-scale, high-dimensional data, it can detect complex patterns in data, and it can make predictions and inferences. The combination of bioinformatics analysis and machine learning has now become a popular research area. Many research teams are developing new algorithms and tools to improve the accuracy and efficiency of their analyses. Also, many studies are applying these methods to solve specific biological problems, such as functional annotation of genes, molecular mechanisms of diseases, and drug discovery and optimization [18]. In our study, we combined single-cell transcriptome and machine learning to screen key genes for cutaneous melanoma. In addition, we verified the expression of PIP4K2A in cutaneous melanoma by immunohistochemistry. Materials and methods Data acquisition The SKCM scRNA-seq data was acquired from the GSE72056 dataset, which was downloaded from the Gene Expression Omnibus (GEO) database (http://www.ncbi.nih.gov/geo) [19]. This dataset is associated with the article "Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq" [20]. In our study, we utilized the "seurat" package in R (version 4.3.0) to conduct a dimensionality reduction cluster analysis on our single-cell data. We identified 17 clusters with a resolution of 1 (RES = 1) and assessed the expression of feature genes within these clusters. The transcriptomic data used in this study were obtained from GSE15605 [21] and GSE46517 [22] in the GEO database. Sixteen normal skin tissue samples and 46 primary cutaneous melanoma tissue samples were selected from the SKCM dataset of GSE15605. Metastatic cutaneous melanoma samples were excluded from this study. In dataset GSE46517, 7 normal skin tissue samples and 31 primary cutaneous melanoma tissue samples were collected. All normal skin tissue samples were defined as the normal group and primary skin melanoma tissue samples were defined as the cancer group.Gene expression profiling arrays for GSE15605 and GSE46517 were based on the GPL570 platform ([HG-U133_Plus_2] Affymetrix Human Genome U133 Plus 2.0 Array) and GPL96 platform ([HG-U133A] Affymetrix Human Genome U133A Array), respectively. Raw gene expression data from primary cutaneous melanoma were processed using R software (version 4.2.2) and Bioconductor Packages, and files of the transcriptome were normalised by trans-IDing with the Platform File Probe to TPM format files.Seurat objects with UMI counts of < 1000 or mitochondrial origin UMI counts of more than 20% of the cells and cells with more than 20% UMI count of erythrocyte origin were considered as low quality cells and were removed. Following quality control, the UMI count matrix underwent log-normalization. In this step, the top 2000 variable genes were selected for further analysis. To reduce the dimensionality of the scRNA-Seq dataset, principal component analysis (PCA) was applied to the integrated data matrix. The Elbowplot function in Seurat was used to select the top 5 principal components for downstream analysis. Major cell clusters were identified with the FindClusters function in Seurat, using a resolution of 1 (res = 1), and visualized with 2D tSNE or UMAP plots. Conventional labeling from previous studies was used to categorize each cell into known biological cell types. Initially, cells were grouped into seven major cell types. Each major cell type was then divided into subsets and further clustered into subclusters to examine cell type heterogeneity. Preferentially or differentially expressed genes in tumor and normal cells were identified using the Seurat FindAllMarkers function. Gene set enrichment analysis To explore the regulatory mechanisms of differential genes in the GSE15605 and GSE46517 datasets, we conducted gene set enrichment analysis (GSEA) utilizing the "clusterProfiler" R package [23]. Nine hub genes were identified through Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) to highlight their biological and functional characteristics. The clusterProfiler (version 4.6.2) and DOSE (version 3.24.2) packages were used to analyze these hub genes for GO and KEGG, retaining results with P-values under 0.05, and giving more significance to higher GeneRatio values. Machine learning Using the R package "mlr3verse", a machine learning diagnostic model was built using dataset GSE15605 and matching clinical data, and a validation model was built using GSE46517 and matching clinical data. Multiple machine learning modelling was used, and finally the ranger model, which had the best benefit, was used to predict the important role of the nine pivotal genes in disease diagnosis. And then the lasso regression model was used to screen out the 3 core genes that accounted for the most important of these 9 pivotal genes. Survival analysis Survival analyses were performed using the website GENT2 (http://gent2.appex.kr/gent2) for meta-survival analysis of the key gene PIP4K2A, and further validated in the BEST website (https://rookieutopia.com/app_direct/BEST) for PIP4K2A The differential survival impact of this gene on tumour TMN staging and progression. Clinical data in both online sites were obtained from the GEO database. Immunoinfiltration analysis Using the datasets GSE22153, GSE133713, GSE98394, GSE65904, TCGA_SKCM, GSE46517, GSE54467, GSE53118, GSE99898, GSE22154, GSE78220, GSE100797, in the BEST website. GSE190113, and the CIBERSORT, CIBERSORT_ABS, EPIC, ESTIMATE, MCPcounter, Quantiseq, TIMER, and xCell algorithms were used to comprehensively evaluate the immune infiltration in the high PIP4K2A expression group of cutaneous malignant melanoma. Predictive analysis of drug therapy Based on the drug prediction analysis cohort (1.GDSC_V1 2.GDSC_V2 3.CTRP 4.PRISM) from the BEST website (https://rookieutopia.com/app_direct/BEST), the datasets GSE100797, GSE78220, GSE53118 were used, GSE54467, GSE99898, GSE190113, TCGA_SKCM, GSE65904, GSE22154, GSE98394, GSE22153, GSE46517, GSE133713 predicted information about possible target therapeutic drugs for cutaneous malignant melanoma as well as the analysis of immune modulatory inhibitors. In addition to targeted therapy predictions, studies have emphasized the importance of optimizing surgical and therapeutic strategies to improve patient outcomes. For example, one study demonstrated that tranexamic acid effectively reduces blood loss and transfusion requirements during surgical procedures, highlighting its role in perioperative management. Similarly, another study compared robot-assisted and laparoscopic surgical techniques, showing the impact of advanced approaches on perioperative and oncologic outcomes. These findings underscore the importance of tailoring surgical strategies to individual patient needs to enhance cancer treatment effectiveness [24, 25]. Immunotherapy prediction Targeted therapy analysis of genes was performed based on the BEST website. The Homet cohort, Cho cohort, Kim cohort, and Nathanson cohort were used to predict the immune response of high expression of the PIP4K2A gene for PD-1 immune-targeted therapy, PD-L1 immune-targeted therapy, and CTLA-4 immune-targeted therapy [26]. Immunohistochemistry In this study, immunohistochemical analysis with PIP4K2A polyclonal antibody was conducted on four cutaneous melanoma samples (No. 523797, 501653, 519937, 519909) and four normal skin samples (No. C1, C2, C3, C4). Initially, fresh tissues were fixed in 10% neutral buffered formalin for 24 h. They were then dehydrated with gradient alcohol, cleared with xylene, and embedded in paraffin with sections cut to 4 μm thickness. Sections were deparaffinized using xylene, rehydrated with gradient alcohol, placed in citrate buffer (pH 6.0), and subjected to microwave heating for 15 min for antigen retrieval. Endogenous peroxidase activity was inhibited with 3% hydrogen peroxide, and nonspecific binding was blocked by incubating sections in PBS buffer containing 10% normal goat serum for 30 min. To differentiate between DAB chromogen staining and the natural brown color of melanin in melanoma tissues, a melanin bleaching step was performed. Sections were treated with 3% hydrogen peroxide for 20 min before antibody incubation to remove endogenous melanin pigments, which helped ensure that the brown staining observed was primarily due to PIP4K2A expression rather than melanin. After melanin bleaching, the sections were incubated overnight at 4 °C with PIP4K2A polyclonal antibody (diluted 1:200), washed with PBS, and treated with an HRP-labeled secondary antibody for 1 h at room temperature. Following DAB color development, hematoxylin was used for contrast staining. Sections were dehydrated using gradient alcohol, cleared with xylene, and sealed with neutral gum. PIP4K2A expression was observed and evaluated under a light microscope, with representative areas selected for imaging. The images were quantitatively analyzed using image analysis software to compare PIP4K2A expression between cutaneous melanoma and normal skin tissues. Future studies may consider alternative chromogens, such as AEC or Fast Red, to further differentiate specific staining from melanin, if necessary. Results The flow chart of this study is shown in Fig. 1.Fig. 1Flowchart of multi-step screening strategy for bioinformatics data Flowchart of multi-step screening strategy for bioinformatics data Single-cell transcriptome altas of SKCM In this study, we analyzed 4645 cells from the GSE72056 dataset, comprising melanoma, immune, and stromal cells from 19 patients [19]. The t-SNE method was used for clustering visualization, resulting in 17 distinct clusters (Figure S1). Feature expression was identified for each cluster (Figure S2A, Figure S2B), leading to the identification of seven key clusters (Fig. 2A-figure supplement 2B): B cells with high CD79A expression, NK cells with elevated NKG7 expression, melanoma cells marked by high MLANA and SOX10 expression, T cells with high CD3D expression, macrophages with elevated MARCO expression, endothelial cells with high VWF expression, and fibroblasts with elevated PDRGA expression (Fig. 2C, Figure S2C). The proportions of these clusters are shown in Fig. 2D, where T cells, melanoma cells, and NK cells are the most prevalent, while macrophages, endothelial cells, and fibroblasts are less common. Using the FindAllMarkers function, we identified the differential gene expression within each cluster.Fig. 2Single-cell sequencing analysis. A 7 cell types identified based on marker gene expression. B Classification of cells. A total of 7 cell types can be found, such as B cells, NK cells, Melanoma cells and T cell. C Violin plots displaying expression of marker genes for the 7 cell types across cells. D Proportional representation of different cell types. E GSVA investigation of different cell types. F Network diagram of the frequency of ligand-receptor interactions between different cell types Single-cell sequencing analysis. A 7 cell types identified based on marker gene expression. B Classification of cells. A total of 7 cell types can be found, such as B cells, NK cells, Melanoma cells and T cell. C Violin plots displaying expression of marker genes for the 7 cell types across cells. D Proportional representation of different cell types. E GSVA investigation of different cell types. F Network diagram of the frequency of ligand-receptor interactions between different cell types Enrichment analysis of single cells in scRNA-seq In order to explore the function of differential genes in each clump, we performed GSVA enrichment analysis on seven clumps of single cells. Among them, the differential genes of B cells clusters were mainly enriched in the functions of OXIDATIVE-PHOSPHORYLATION, REACTIVE-OXYGEN-SPECIES-PATHWAY, etc. The up-regulated differential genes in Endothelial cells clusters were mainly enriched in ANGIOGENESIS, EPITHELIAL-MESENCHYMAL-TRANSITION, COAGULATION, TGF- BETA-SIGNALING, and other functions. In contrast, the down-regulated differential genes of Endothelial cells clusters were mainly enriched in the ALLOGRAFT-REJECTION function.The differential genes of Fibroblasts cells clusters were mainly enriched in the MYC-TARGETS -V1, OXIDATIVE-PHOSPHORYLATION and EPITHELIAL-MESENCHYMAL-TRANSITION functions.Macrophages Cluster differential genes were mainly enriched in INTERFERON-GAMMA-RESPONSE, TNFA-SIGNALING-VIA-NFKB INFLAMMATORY-RESPONSE and IL6-JAK-STAT3-SIGNALING.The up-regulated differential genes in NK cells clusters were mainly enriched in ALLOGRAFT -REJECTION and INTERFERON-GAMMA-RESPONSE. The down-regulated differential genes of NK cells clusters were mainly enriched in functions such as COAGULATION and OXIDATIVE-PHOSPHORYLATION. The up-regulated differential genes of Melanoma cells clumps were mainly enriched in functions such as MYC-TARGETS-V1 and OXIDATIVE-PHOSPHORYLATION. Conversely, the down-regulated differential genes of Melanoma cells clusters were mainly enriched in functions such as ALLOGRAFT-REJECTION. up-regulated differential genes of T cells clusters were mainly enriched in functions such as ALLOGRAFT-REJECTION. Conversely, the down-regulated differential genes of T cells clusters were mainly enriched for functions such as MYC-TARGETS-V1 and OXIDATIVE-PHOSPHORYLATION (Fig. 2E). From these Cluster enrichment results we found that there was a striking consistency in the pathways to which the up-regulated differential genes in the Melanoma cells clusters were enriched with the down-regulated differential genes in the T cell clusters. From this phenomenon, we hypothesised that the development of cutaneous malignant melanoma may be more related to T cells, and so we focused on the differential genes in the T cell clusters down the pathway. These pathways, particularly MYC-TARGETS-V1 and OXIDATIVE PHOSPHORYLATION, were chosen as key upregulated functions due to their established roles in cancer biology. The MYC pathway is a critical regulator of cell proliferation, metabolism, and survival, and its dysregulation is commonly associated with tumorigenesis in many cancers, including melanoma. MYC activation drives uncontrolled cell division, contributing to melanoma aggressiveness by promoting the transcription of genes that support cell growth and division. In parallel, OXIDATIVE PHOSPHORYLATION is crucial for energy production in rapidly proliferating tumor cells. In melanoma, metabolic reprogramming enables cancer cells to meet the increased demands for energy (ATP) and biosynthetic precursors, supporting their high proliferative rate. The upregulation of oxidative phosphorylation indicates that melanoma cells rely heavily on this metabolic process to maintain growth, especially under the often hypoxic conditions of the tumor microenvironment. Together, these pathways highlight key mechanisms that sustain melanoma malignancy and offer potential targets for therapeutic intervention. CellChat analysis In this study, we used "CellChat" simulation to analyse the cell–cell communication network generated by the interaction between different cells (Fig. 2F). We can see that the interaction between NK cells and macrophages is strong, Melanoma cells have strong interaction with NK cells, Fibroblasts cells, macrophages and T cells, and T cells have strong interaction with NK cells (Figure S3). Figure S4 shows that T cells were interacting strongly with Melanoma cells through the CD99-CD99 pathway. We surmised that T cells do have some relevant roles with Melanoma cells based on this result, so we extracted the T cell-related differential genes for subsequent analysis (Fig. 3A).Fig. 3Gene expression characterization of SKCM-infiltrating tregs and exhausted CD8 + T cells A t-SNE plot showing T lymphocyte subtypes. B Volcano plot showing differentially expressed genes in SKCM -infiltrating Tregs. C GSEA analysis results in GSE15605 D The same volcano plot as in B in the GSE46517 dataset. E GSEA analysis results in GSE46517 F The Venn graph showing the overlap of the two data sets with T cell genes Gene expression characterization of SKCM-infiltrating tregs and exhausted CD8 + T cells A t-SNE plot showing T lymphocyte subtypes. B Volcano plot showing differentially expressed genes in SKCM -infiltrating Tregs. C GSEA analysis results in GSE15605 D The same volcano plot as in B in the GSE46517 dataset. E GSEA analysis results in GSE46517 F The Venn graph showing the overlap of the two data sets with T cell genes Differentially expressed genes(DEGs) Flowchart of the overall data screening strategy (Fig. 1). Two GEO datasets GSE15605 and GSE46517 were obtained for analysis. We analysed the SKCM and normal skin tissues from these two datasets for differences with the screening conditions of logFC > 0.5 and p < 0.05, and finally obtained the differential genes (Fig. 3B, D). We performed GSEA enrichment analysis on the acquired differential genes to obtain the diseases or functions regulated by the differential genes corresponding to the two datasets.Diseases or functions enriched in the GSE15605 dataset were Allograft rejection, Antigen processing and presentation, Arginine biosynthesis, Drug metabolism—cytochrome P450, Fatty acid biosynthesis, Glycosphingolipid biosynthesis—ganglio series, Graft-versus-host disease, Terpenoid backbone biosynthesis (Fig. 3C). Diseases or functions enriched in the GSE46517 dataset were Focal adhesion, Histidine metabolism, Nicotine addiction, Ribosome, Staphylococcus aureus infection, Steroid biosynthesis, Tuberculosis, Valine, leucine and isoleucine degradation (Fig. 3E). It is predicted that the development of cutaneous malignant melanoma may be associated with changes in the above related functions. Screening of candidate genes associated with SKCM Taking the intersection of the above modules with the obtained differential genes, the 9 genes with the highest correlation with cancer were obtained as pivotal genes (Fig. 3F). Functional enrichment analysis To uncover the molecular biological processes underlying common disease genes and the pathogenesis of cutaneous malignant melanoma, we conducted functional enrichment analyses on differentially expressed genes (DEGs) using GO, KEGG, and Hallmark pathways. Among the enriched pathways, oxidative phosphorylation and metabolic reprogramming were significantly highlighted, indicating a central role in melanoma progression. This aligns with findings in prostate cancer, where mitochondrial dysfunction was identified as a critical factor influencing molecular subtypes and patient prognosis [27]. These results suggest that mitochondrial dysfunction and associated metabolic changes are conserved mechanisms across different cancer types, contributing to tumor aggressiveness and treatment resistance. GO enrichment was performed with the clusterProfiler and DOSE packages in R (Fig. 4A). The analysis indicated that, in biological processes, these genes were predominantly involved in CD4-positive, alpha–beta T cell activation, and carbohydrate transmembrane transport. For cellular components, enrichment was primarily observed in tertiary granules and secretory granule membranes. In terms of molecular function, genes were mainly enriched in carbohydrate-related activities. KEGG pathway analysis revealed that common genes were primarily associated with pathways such as ubiquitin-mediated proteolysis, cell cycle, DNA replication, mismatch repair, and T cell receptor signaling (Fig. 4B). Hallmark gene set enrichment analysis showed concentration in pathways linked to Myc targets v1, protein secretion, G2/M checkpoint, mitotic spindle, E2F targets, unfolded protein response, PI3K/AKT/mTOR signaling, oxidative phosphorylation, heme metabolism, MTORC1 signaling, and DNA repair (Fig. 4C). The results from GO, KEGG, and Hallmark analyses were consistent with existing reports on cutaneous malignant melanoma, indicating their reliability.Fig. 4Enrichment analysis and machine learning. A Based on GO enrichment analysis, these genes are mainly involved in CD4-positive, alpha–beta T cell activation and carbohydrate transmembrane transport in biological processes, and in terms of cellular components are mainly focused on tertiary granule and secretory granule membranes, while in terms of molecular functions, they mainly exhibit carbohydrate transmembrane transport activity. B KEGG-enriched analyzed genes were mainly focused on pathways related to Ubiquitin mediated proteolysis, Cell cycle, and DNA replication. C Enrichment analysis of the Hallmark gene set showed that common genes were mainly concentrated in pathways related to Myc targets v1, Protein secretion, G2/M checkpoint, etc. D Machine learning model comparison using 7 different machine learning algorithms. E Receiver operating characteristics (ROC) curves for multiple machine learning models. F The “logreg” machine learning algorithm model exhibits the highest AUC of 0.995 Enrichment analysis and machine learning. A Based on GO enrichment analysis, these genes are mainly involved in CD4-positive, alpha–beta T cell activation and carbohydrate transmembrane transport in biological processes, and in terms of cellular components are mainly focused on tertiary granule and secretory granule membranes, while in terms of molecular functions, they mainly exhibit carbohydrate transmembrane transport activity. B KEGG-enriched analyzed genes were mainly focused on pathways related to Ubiquitin mediated proteolysis, Cell cycle, and DNA replication. C Enrichment analysis of the Hallmark gene set showed that common genes were mainly concentrated in pathways related to Myc targets v1, Protein secretion, G2/M checkpoint, etc. D Machine learning model comparison using 7 different machine learning algorithms. E Receiver operating characteristics (ROC) curves for multiple machine learning models. F The “logreg” machine learning algorithm model exhibits the highest AUC of 0.995 Machine learning To elucidate the progression of cutaneous malignant melanoma, we utilized the identified hub genes to construct predictive models for further examination. The GSE15605 dataset served as the training cohort, while the GSE46517 dataset was used to evaluate the models' predictive capabilities. We applied seven distinct machine learning algorithms, optimizing each model's parameters through five repetitions of ten-fold cross-validation (Fig. 4D). The models' performance was assessed by examining their area under the curve (AUC) values in the validation cohort (Fig. 4E). Through these stringent evaluations, the "logreg" machine learning algorithm emerged as the top performer, achieving the highest AUC of 0.995 (Fig. 4F). Additionally, we developed a linear regression model to assess the pathogenicity of the nine hub genes for cutaneous malignant melanoma. Using a lasso regression algorithm, we identified three key genes—PIP4K2A, SLC2A3, and RORA—that were strongly linked to disease pathogenicity (Fig. 5A, B). Among these, PIP4K2A had the highest weight in the lasso regression model, warranting a more detailed subsequent analysis.Fig. 5Identification of key genes strongly associated with disease pathogenicity and survival analysis. A Binomial deviation of overall survival (OS) for LASSO coefficient curve. B LASSO coefficient profiles of genes. C Patients in the PIP4K2A high-expression group in cutaneous malignant melanoma have a somewhat shorter survival time. D Meta-survival analysis results of PIP4K2A expression. E Higher expression of PIP4K2A in elderly patients > 65 years of age. F Higher expression of PIP4K2A in patients with clinical manifestations of ulceration. G, H PIP4K2A is more highly expressed in cutaneous malignant melanoma tissue samples. I Higher PIP4K2A expression at higher T stage. J The expression of PIP4K2A showed differences in the performance of individual staging Identification of key genes strongly associated with disease pathogenicity and survival analysis. A Binomial deviation of overall survival (OS) for LASSO coefficient curve. B LASSO coefficient profiles of genes. C Patients in the PIP4K2A high-expression group in cutaneous malignant melanoma have a somewhat shorter survival time. D Meta-survival analysis results of PIP4K2A expression. E Higher expression of PIP4K2A in elderly patients > 65 years of age. F Higher expression of PIP4K2A in patients with clinical manifestations of ulceration. G, H PIP4K2A is more highly expressed in cutaneous malignant melanoma tissue samples. I Higher PIP4K2A expression at higher T stage. J The expression of PIP4K2A showed differences in the performance of individual staging Survival analysis Based on the transcriptomic data of GSE190113 with matched clinical data, the Kaplan–Meier survival curves indicated that the survival time of patients in the group with high expression of PIP4K2A in cutaneous malignant melanoma was a little shorter (Fig. 5C). In the results of Meta-survival analysis, the HR was 1.21 for the Fixed effect model and 1.26 for the Random effects model, further testifying that PIP4K2A is a risky prognostic factor in cutaneous malignant melanoma (Fig. 5D). Clinical data according to the GSE22153 dataset showed that PIP4K2A also exhibited a significant role in cutaneous malignant melanoma in the elderly, with higher expression of PIP4K2A in elderly patients greater than 65 years of age compared to those under 65 years of age (Fig. 5E). According to clinical data from the GSE98394 dataset, PIP4K2A expression was higher in patients with cutaneous malignant melanoma with clinical manifestations of ulceration (Fig. 5F). On the tissue side, both the GSE98394 and GSE46517 datasets showed higher expression of PIP4K2A in cutaneous malignant melanoma tissue samples (Fig. 5G, H). According to the clinical data of GSE98394, the higher the T-stage the higher the expression of PIP4K2A in cutaneous malignant melanoma was correspondingly higher (Fig. 5I), demonstrating an enhanced proliferative capacity of cutaneous malignant melanoma with high PIP4K2A expression. In terms of cancer stage, clinical data from the GSE98394 datasets showed differences in the expression of PIP4K2A across stages (Fig. 5J). Immunoinfiltration analysis Utilizing preprocessing from the BEST online platform, we analyzed datasets GSE22153, GSE133713, GSE98394, GSE65904, TCGA_SKCM, GSE46517, GSE54467, GSE53118, GSE99898, GSE22154, GSE78220, GSE100797, and GSE190113. Immune infiltration in cutaneous malignant melanoma, particularly in the high PIP4K2A expression group, was evaluated using the CIBERSORT, CIBERSORT_ABS, EPIC, ESTIMATE, MCPcounter, Quantiseq, TIMER, and xCell algorithms (Fig. 6). In the CIBERSORT_ABS analysis, significant infiltration was observed in follicular helper T (Tfh) cells, naive B cells, activated memory CD4 + T cells, activated dendritic cells, and resting memory CD4 + T cells. The Quantiseq algorithm highlighted notable infiltration of B cells and dendritic cells. According to the CIBERSORT results, naive B cells, Tfh cells, activated memory CD4 + T cells, M1 macrophages, resting memory CD4 + T cells, and activated dendritic cells showed pronounced infiltration. The xCell algorithm revealed high levels of B cells, naive B cells, mast cells, naive CD4 + T cells, memory B cells, class-switched memory B cells, memory CD4 + T cells, CD4 + T cells, plasmacytoid dendritic cells (pDC), central memory CD8 + T cells (CD8 + TCM), CD8 + T cells, activated dendritic cells (aDC), conventional dendritic cells (cDC), NK cells, effector memory CD8 + T cells (CD8 + TEM), Tregs, granulocyte-monocyte progenitors (GMP), gamma delta T cells (Tgd), macrophages, M1 macrophages, and melanocytes. MCPcounter analysis showed significant infiltration of T cells, B lineage cells, myeloid dendritic cells, CD8 + T cells, monocytic lineage cells, cytotoxic lymphocytes, and NK cells. TIMER analysis indicated higher levels of infiltration in B cells, CD8 + T cells, dendritic cells, CD4 + T cells, neutrophils, and macrophages.Fig. 6Immunoinfiltration analysis of PIP4K2A expression Immunoinfiltration analysis of PIP4K2A expression Predictive analysis of drug therapy In the drug therapy prediction analysis, in the PRISM database, we found that high PIP4K2A expressing cutaneous malignant melanomas were resistant to carbimazole, methimazole, barnidipine, talniflumate, cyanocobalamin, sulisobenzone, dexniguldipine, brucine, proxyphylline, AGI-6780, lomefloxacin, hydroflumethiazide, salazodine, bicalutamide, bemesetron, uric -acid, otenzepad, fenobam, ufenamate and pirodavir, among a range of other drugs, produced resistance. In the CTRP database analysis we found that in the high PIP4K2A expression group for erismodegib, BRD-K64610608, Platin, BRD-K16147474, semagacestat, AGK-2, BRD-K09587429, SB-431542, austocystin D, etomoxir, spautin-1, erlotinib, canertinib, MK- 0752, BRD-K17060750, CI-976, CAY10576, hyperforin, procarbazine and silmitasertib, and a host of other drugs were resistant. In the GDSC1 database analysis we found that high PIP4K2A expressing cutaneous malignant melanoma was resistant to AST-1306_381, JNJ38877605_284, Tanespimycin_1026, AZD8931_1416, Dyrk1b_0191_1407, Pilaralisib_372, Trametinib_1372, Selumetinib_1498, Refametinib_1526, Refametinib_1014 and Cetuximab_1114, amongst a range of other drugs, have produced resistance. In the GDSC2 database we found that cutaneous malignant melanoma in the high PIP4K2A expression group was resistant to VX-11e_2096, SCH772984_1564, Trametinib_1372, Selumetinib_1736, PD0325901_1060, Sapitinib_1549, ERK_6604_1714 and ERK_2440_1713, and a range of other drugs produced resistance. These data assess that high PIP4K2A expression is indeed a risk factor for cutaneous malignant melanoma drug therapy. (Figure S5). Immunotherapy prediction We plotted AUC curves based on the Homet cohort, Cho cohort, Kim cohort, and Nathanson cohort, and verified that the high-expression PIP4K2A group had a poor response to Anti-PD-1, Anti-PD-L1 and Anti-CTLA-4 immunotherapy were all poorly responsive (Fig. 7A–D).Fig. 7Immunotherapy predictive analytics and immunohistochemistry. A–D Prediction of immunotherapy sensitivity in different cohorts. (E)The expression of PIP4K2A was significantly higher in cutaneous melanoma tissues than in normal skin tissues Immunotherapy predictive analytics and immunohistochemistry. A–D Prediction of immunotherapy sensitivity in different cohorts. (E)The expression of PIP4K2A was significantly higher in cutaneous melanoma tissues than in normal skin tissues Expression of PIP4K2A in cutaneous melanoma To explore the significance of PIP4K2A, immunohistochemical experiments were performed using four normal skin tissues as a control group and four cutaneous melanoma tissues as an experimental group to assess the expression of PIP4K2A in both tissue types (Fig. 7E). The immunohistochemical staining results showed that the expression of PIP4K2A was significantly higher in cutaneous melanoma tissues compared to normal skin tissues. In cutaneous melanoma tissues, PIP4K2A was strongly positively stained and widely distributed with high staining intensity, predominantly localized in the tumor cells. These tumor cells were characterized by irregular shapes and dense, disorganized cellular structures, indicating that PIP4K2A plays a potential role in promoting melanoma cell proliferation and tumor progression. In contrast, in normal skin tissues, PIP4K2A expression was lower, with lighter staining intensity primarily observed in the basal layer of the epidermis. This layer contains stem-like keratinocytes that are responsible for skin regeneration and repair, suggesting that PIP4K2A in normal skin is involved in regulating basic cellular growth and maintenance functions. No significant PIP4K2A expression was observed in other cell types such as immune or stromal cells in the normal skin samples. These results are consistent with our single-cell transcriptome and machine learning analyses, further confirming that PIP4K2A is an important malignant prognostic marker for cutaneous melanoma. The differential expression between melanoma and normal skin highlights PIP4K2A's role in the aggressiveness of melanoma, while its restricted expression in normal skin suggests a more regulated involvement in normal tissue homeostasis. Discussion Cutaneous malignant melanoma, a highly aggressive skin tumor, has seen a notable rise in incidence recently. Despite its low occurrence among skin cancer types, it is the leading cause of skin cancer-related deaths [28]. The epidemiology of cutaneous melanomas is becoming increasingly concerning due to global aging and greater exposure to ultraviolet light [29]. Additionally, some studies suggest a link between dietary habits and skin melanoma [30]. These tumors are notoriously malignant and often resist conventional treatments like surgery, radiotherapy, and chemotherapy, highlighting the urgent need for novel therapeutic strategies targeting specific molecular pathways. Phosphatidylinositol-5-phosphate 4-kinase (PIP4K) and its homologue PIP4K2A have garnered significant interest as potential molecular targets. PIP4K enzymes play a crucial role in phosphatidylinositol signaling by phosphorylating PtdIns5P to produce PtdIns4,5P2, a key molecule in various signaling pathways regulating cell proliferation, survival, and migration [31, 32]. Disruptions in PIP4K can affect these pathways and potentially lead to tumor formation. PIP4K2A activity regulates PI5P and PI(4,5)P2 levels, influencing multiple intracellular signaling pathways [33]. Studies indicate that PIP4K2A is essential for maintaining phosphatidylinositol-like molecules' homeostasis. Its activity is regulated by factors such as phosphorylation, protein interactions, and intracellular localization. PIP4K2A interacts with proteins like the tumor suppressor TP53, affecting cell cycle and apoptosis [34]. It is also implicated in pathways responding to DNA damage and stress [35]. Overexpression of PIP4K2A is linked to tumor aggressiveness and poor prognosis, influencing tumor cell migration and invasion by regulating PI(4,5)P2 levels. As a critical player in cell signaling networks, PIP4K2A is a promising cancer therapy target, with research focusing on small molecule inhibitors to control PI5P and PI(4,5)P2 levels and inhibit tumor growth and spread [36]. In our study, we employed multidimensional approaches, including single-cell transcriptomics, gene enrichment analysis, machine learning predictive modeling, and experimental validation, to investigate PIP4K2A expression in cutaneous melanoma and its impact on disease prognosis. Our findings revealed that PIP4K2A expression significantly influences melanoma development, progression, and prognosis. We analyzed 4645 cells from the GSE72056 dataset using single-cell analysis, identifying high proportions of T cells, melanoma cells, and NK cells. GSVA enrichment analysis of seven major cell clusters revealed functional roles for differential genes within each cluster. Results showed that pathways enriched for up-regulated genes in melanoma cell clusters corresponded to down-regulated genes in T cell clusters, suggesting a potential link between melanoma development and T cells. We further confirmed activation of pathways related to MYC target genes and oxidative phosphorylation in melanoma cells, indicating their role in tumor proliferation and metabolism. Cell–cell communication analysis indicated strong interactions between T cells and melanoma cells via the CD99-CD99 pathway, shedding light on immune cell function in the tumor microenvironment. Similar findings have been reported in studies on bladder cancer and kidney renal clear cell carcinoma, where endothelial-related molecular subtypes and their interactions with immune cells were shown to play critical roles in shaping the tumor microenvironment. These studies highlight the importance of endothelial and immune cell interplay in influencing tumor progression and therapy responses across different cancer types [37–41]. Differential analysis of SKCM and normal skin tissues from GSE15605 and GSE46517 datasets identified differentially expressed genes intersecting with T cell genes, resulting in nine hub genes highly correlated with cancer. Machine learning models identified PIP4K2A, SLC2A3, and RORA as key genes, with PIP4K2A showing the strongest association with melanoma behavior. SLC2A3, a key member of the glucose transporter protein family, has been shown in recent studies to correlate with lower survival rates in various tumors. In head and neck squamous cell carcinoma (HNSC), elevated SLC2A3 expression promotes cell proliferation and migration through the NF-κB and epithelial-mesenchymal transition (EMT) signaling pathways. Furthermore, SLC2A3 has a negative correlation with immune cells, suggesting a role in HNSC immune response [42]. Three independent SNPs (RORA rs782917, RORA rs17204952, and DNMT1 rs7253062) were identified as predictors of cutaneous melanoma-specific survival (CMSS) using a GWAS dataset of 191 related genes from MD Anderson Cancer Center, and these findings were validated in the Harvard GWAS dataset. Genetic variants in RORA and DNMT1 may serve as potential biomarkers for CMSS [43]. Furthermore, the microRNA/isomiR network has been shown to play a pivotal role in melanoma pathogenesis, influencing tumor growth and immune responses [44]. Among them, the strong association between PIP4K2A and the malignant behaviour of melanoma. Kaplan–Meier survival curves and meta-analysis confirmed that high PIP4K2A expression correlates with shorter patient survival, emphasizing its potential as a therapeutic target. High PIP4K2A expression was particularly noted in older patients and those with ulcerative manifestations, indicating a link to tumor aggressiveness and poor outcomes. Similar findings have been observed in pan-cancer analyses, which highlighted the role of oncogenic drivers across multiple cancer types. Additionally, studies demonstrated how targeting specific molecular markers could enhance treatment efficacy. These findings underscore the importance of identifying and targeting key molecular drivers to improve therapeutic outcomes in aggressive cancers [45, 46]. Additionally, mutations in BRAF, KIT, and NRAS are frequently observed in melanoma, particularly in specific subtypes such as acral melanoma, and play a crucial role in tumor progression and resistance to therapies [47]. Tumors with high PIP4K2A expression showed resistance to several therapeutic agents, including immunotherapies like Anti-PD-1, Anti-PD-L1, and Anti-CTLA-4, suggesting that PIP4K2A may affect treatment efficacy through immune escape mechanisms. Experimental validation through immunohistochemistry demonstrated significantly higher PIP4K2A expression in melanoma tissues compared to normal skin, correlating closely with clinical stage and prognosis. These findings align with our single-cell transcriptome and machine learning results, solidifying PIP4K2A as an important prognostic marker for cutaneous melanoma. However, this study has limitations, such as a relatively small sample size despite extensive data sources, and potential bias from single-dataset results. Future research should include larger independent cohorts for validation. While our analyses linked PIP4K2A to disease prognosis, its specific molecular mechanisms require further experimental exploration. Additionally, therapeutic predictions were based on existing drug databases, necessitating clinical trials to assess real-world applications. Additionally, the immunohistochemistry study in this research was limited to 4 cutaneous melanoma samples and 4 normal skin samples due to restricted access to clinical samples and financial constraints. While the findings are consistent with the bioinformatic and machine learning analyses, we acknowledge that a larger cohort of samples is required to further validate the prognostic significance of PIP4K2A. Future studies should aim to expand the sample size to strengthen the statistical power of immunohistochemistry results. In conclusion, PIP4K2A plays a crucial role in cutaneous melanoma pathogenesis, with its expression level closely tied to disease malignancy and treatment response, making it a promising therapeutic target. These insights provide a foundation for developing novel melanoma treatments targeting PIP4K2A. Future studies should further investigate PIP4K2A's mechanisms in melanoma and evaluate the clinical potential of targeting this molecule. Supplementary Information Supplementary Material 1.Supplementary Material 2.Supplementary Material 3.Supplementary Material 4.Supplementary Material 5. Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Supplementary Material 4. Supplementary Material 5.