Analysis of fatty acid metabolism in prostate cancer and discovery of a new programmed cell death-associated luminal cell subpopulation

Abstract

Introduction

This study focuses on the role of fatty acid metabolism in prostate cancer, particularly in oncogenic luminal cells associated with programmed cell death under the influence of metabolic reprogramming.

Materials and Methods

Prostate cancer was analyzed using single-cell transcriptomics and spatial transcriptomics data. Fatty acid metabolism levels in the tumor microenvironment were quantified by multiple gene set scoring methods, and data were processed using NMF and deconvolution methods to identify different cell populations and their interactions in the tumor microenvironment.

Results

Luminal cells have significantly increased activity in fatty acid metabolism, which is associated with the aggressiveness and metastatic capability of tumors. Luminal cell subpopulations have been found to play a key role in the development of prostate cancer, especially their close association with programmed cell death.

Conclusion

This study deepens the understanding of the role of fatty acid metabolism in prostate cancer, identifies fatty acid metabolism-related luminal cell subtypes, and proposes new therapeutic targets, providing new insights into prostate cancer treatment.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-025-01982-w.

Introduction

Prostate cancer is the second most common cancer among men worldwide and one of the leading causes of cancer death among men. With the aging of the global population, the incidence and mortality rates of prostate cancer are expected to continue to rise. Although prostate cancer usually has a favorable prognosis in its early stages, once metastasis occurs, treatment becomes significantly more challenging, and patient survival rates markedly decrease. Therefore, a deeper understanding of the pathogenesis of prostate cancer, especially the biological basis for tumor metastasis and treatment resistance, is crucial for improving patient outcomes [1].

In recent years, the role of fatty acid metabolism in prostate cancer has attracted widespread attention. Fatty acids are not only major components of cell membranes but also important sources of energy storage and production. In tumors, changes in fatty acid metabolism help cancer cells adapt to increased energy demands and harsh survival conditions, promoting tumor growth and metastasis. Studies have shown that prostate cancer cells reprogram fatty acid metabolism to meet their growth and metastasis needs, including changes in fatty acid uptake, synthesis, elongation, and oxidation processes. These metabolic pathway alterations not only support the energy and biosynthetic demands of tumor cells but also promote tumor aggressiveness and metastatic capability. Therefore, fatty acid metabolism plays a key role in the development of prostate cancer and has become a focus of research [2, 3].

The present study was devoted to characterizing fatty acid metabolism in the prostate cancer tumor microenvironment and its role in tumor development. By analyzing single-cell RNA sequencing data, we aim to identify important cellular subtypes affected by reprogramming of fatty acid metabolism and explore new therapeutic targets for tumor therapy [3]. The integrated use of spatial transcriptomics data to observe the differences in fatty acid metabolism at spatial locations and to explore the interactions between cells at different metabolic levels will provide new strategies for personalized treatment of prostate cancer [4].

Materials and methods

Source of raw data

In this study, the prostate cancer single-cell sequencing data were sourced from the Gene Expression Omnibus (GEO) database, specifically the GSE176031 dataset, which includes four prostate tumor samples and four matched adjacent normal tissue samples. The spatial transcriptomics data were obtained from the official 10X Genomics website (https://www.10xgenomics.com/), encompassing sequencing data from three formalin-fixed paraffin-embedded (FFPE) samples, including one normal prostate tissue, one prostatic acinar cell carcinoma, and one human prostate adenocarcinoma sample with invasive cancer. Additionally, RNA sequencing data for prostate cancer were downloaded from the UCSC Xena platform (https://xena.ucsc.edu/), derived from the TCGA (The Cancer Genome Atlas) cohort, including sequencing information for 551 samples, along with corresponding survival data to facilitate survival analysis.

Processing of single-cell sequencing data

In this study, we conducted an in-depth analysis of single-cell RNA sequencing (scRNA-seq) data of prostate cancer using the Seurat package (version 4.3.0) in R [5]. Initially, to ensure the accuracy of our analysis, we performed strict quality control on the data, selecting cells that met the following criteria for subsequent analysis: the number of genes expressed in each cell ranged from 200 to 4000, and the proportion of mitochondrial gene expression was less than 20%. Next, we normalized the data using the NormalizeData function in the Seurat package to eliminate potential systematic biases between different samples. The normalized data were converted into Seurat objects, and 2000 highly variable genes with the most significant expression variability were identified using the FindVariableFeatures function, providing key information for subsequent principal component analysis (PCA) and cell clustering. We performed PCA dimensionality reduction on the data using the RunPCA function within the Seurat package and identified principal components that significantly impacted sample clustering through JackStraw analysis. Based on the selected principal components, cell clustering analysis was conducted using the FindNeighbors and FindClusters functions, and the clustering results were visualized using the t-SNE method, thereby intuitively revealing the similarities and differences between cells. Finally, to identify genes specifically expressed in each cell cluster, we used the Wilcoxon test method, along with the FindAllMarkers and FindMarkers functions, to complete the differential expression analysis. Cell type annotation was referenced against cell marker information provided by the CellMarker database (http://xteam.xbio.top/CellMarker/index.jsp), ensuring the accuracy of cell type identification.

We utilized the CellChat R package (version 1.6.1) for comprehensive analysis of intercellular communication patterns [6]. Beyond the CellChat package, we also employed the CellCall software package (version 1.0.7), aiming to further analyze the communication patterns of Luminal cells under different fatty acid metabolism states [7]. To delve deeper into the metabolic state differences between normal and tumor tissues, this study utilized the scMetabolism software package (version 0.2.1) for the quantitative analysis of metabolic pathway activity in single-cell data.

Based on 68 fatty acid metabolism-related genes (FAM), we utilized five commonly used algorithms to score gene sets from single-cell data (AddModuleScore, ssGSEA, AUCell, UCell, and singscore). Evaluating a specific gene set with multiple scoring methods to obtain a composite score can reduce errors and biases in gene set scoring, providing more comprehensive information, robustness, and better biological interpretation [8].

To detect copy number variations (CNV) in Luminal cells within tumor samples, we utilized the infercnv software package (version 1.14.2). We used epithelial cells from adjacent normal samples as a reference, allowing for detailed CNV analysis of Luminal cells in tumor samples at high resolution. To perform enrichment analysis on cell types within prostate cancer single-cell transcriptome data, particularly focusing on Luminal cells, we employed two R packages: clusterProfiler (version 4.6.2) and fgsea (version 1.24.0). The clusterProfiler package is used to compare enrichment differences between biological themes and supports queries to various biological information databases, including GO, KEGG, and Reactome [9]. Additionally, the fgsea package, as an implementation of fast GSEA, optimizes the computation process and is capable of efficiently handling large datasets, thus allowing us to precisely assess gene set enrichment at the single-cell level [10].

In this study, we employed the unsupervised non-negative matrix factorization (NMF) method using the R package NMF (version 0.27) to analyze single-cell RNA sequencing (scRNA-seq) data [11]. Our goal was to reveal the characteristics and differences of fatty acid metabolism in Luminal cell subpopulations in more detail. Utilizing the nmf() function of the NMF package, we chose the “snmf/r” method for factorization, which is particularly well suited for dealing with sparse matrices commonly found in single-cell datasets. During the decomposition process, a suitable decomposition dimension (i.e., the number of subgroups) was chosen, which was determined by cross-validation and stability analysis. Based on the NMF decomposition results, we classified the cells into different subpopulations and further analyzed the characteristic genes of each subpopulation.

Finally, it is crucial to highlight that the ggplot2 package (version 3.4.2) serves as the core tool for visualizing our results. Based on the grammar of graphics, it offers a powerful and flexible approach to crafting complex plots, enabling us to effectively communicate our findings through sophisticated graphical representations.

Processing of spatial transcriptome sequencing data

In our study, the preliminary processing and analysis of spatial transcriptomics data were conducted using the Seurat package (version 4.3.0). Initially, the “SCTransform” function was utilized to normalize and scale UMI counts and to identify the most variably expressed features. Subsequently, we performed dimension reduction using the “RunPCA” function. The “scMetabolism” R package was also applied to spatial transcriptomics sequencing data. Assessing metabolism in spatial transcriptomics data, following dimension reduction and clustering, can reveal the metabolic characteristics of different cell clusters.

For pseudo-temporal analysis of spatial transcriptomics data, indicating the developmental and differentiation processes of cell clusters located in different spatial positions, we used the “Monocle” R package. To execute spatial transcriptomics data preprocessing, visualization, clustering, pseudo-temporal analysis, and differential expression analysis in Python, we adopted the Python-based Scanpy package [12]. Additionally, to decipher cell types and their interactions in spatial transcriptomics data, we introduced the stLearn software package developed by the University of Queensland’s Institute for Molecular Bioscience [13]. stLearn is capable of revealing cell types in tissues using gene expression, tissue morphology, and spatial location information. It can reconstruct the distribution of cell types within tissues, infer cellular evolutionary pathways, and identify tissue regions with strong intercellular interactions.

Deconvolution analysis integrated spatial transcriptomics with single-cell sequencing data

We employed deconvolution analysis to infer the proportions of different cell types from mixed cell samples. This method leverages the strengths of both single-cell sequencing and spatial transcriptomics data: the former provides cell-level gene expression information on cellular heterogeneity within tissues, while the latter offers spatial location information of cells within tissues, revealing spatial heterogeneity. Through this approach, we gained a deeper understanding of the complexity and heterogeneity of tumors at a high spatial resolution. During the deconvolution analysis, we utilized the “spacerxr” R package (version 2.2.1) for Robust Cell Type Decomposition (RCTD) analysis. To further understand the patterns of intercellular communication within the tumor microenvironment, we utilized the “mistyR” package (version 1.6.1) for intercellular interaction analysis of spatial transcriptomics data [14]. “mistyR” is a powerful R package designed specifically for analyzing spatial transcriptomics data, capable of revealing the interactions between cells within tissues using spatial location information and gene expression data.

Key FAM-related Luminal cell subgroups combined with bulk data for prognostic analysis

We explored the potential clinical prognostic value of newly identified fatty acid metabolism-related Luminal cell subgroups. To this end, we conducted an in-depth analysis combining bulk sequencing data. Using the Seurat package to process single-cell sequencing data, we first categorized the fatty acid metabolism-related Luminal cells identified from patient tumor tissues into high and low expression subgroups based on their key gene expression levels. Next, we used the FindAllMarkers function to identify marker genes for these two subgroups. After obtaining marker genes for the key cell populations, we quantitatively analyzed these genes in bulk sequencing data to construct high and low-risk groups. To assess the expression of these cell subgroups in invasive tumors, we processed bulk RNA-seq data and conducted single-sample gene set enrichment analysis (ssGSEA) using the GSVA method to quantitatively evaluate the enrichment of these marker gene sets in various tumor samples. Finally, we combined this enrichment analysis result with clinical survival data to assess the clinical prognostic differences between different risk groups through survival analysis. When performing survival analysis using the survminer (0.4.9) and survival (3.4–0) R packages, we first determined the optimal risk grouping threshold using the surv_cutpoint function, then divided the samples into high and low-risk groups based on this threshold, and constructed survival curves using the survfit function. Through this series of analyses, we not only revealed the potential role of fatty acid metabolism-related Luminal cell subgroups in tumor prognosis but also provided important biomarkers for future clinical applications.

Statistical analysis

The statistical analyses were conducted using R version 4.2.2, 64-bit, along with its support packages. The non-parametric Wilcoxon rank sum test was employed to assess the relationship between two groups for continuous variables. Spearman correlation analysis was conducted to examine correlation coefficients. A significance level of P < 0.05 was considered statistically significant for all statistical investigations.

Results

Single-cell data dimensionality reduction clustering and cell type identification

Initially, we created Seurat objects for each sample’s expression matrix. Given the significant biological differences between tumor tissues and their adjacent normal tissues, along with the complexity of intratumoral heterogeneity, we decided to merge these eight Seurat objects directly without batch effect correction [15, 16]. This decision was based on the observation of PCA dimensionality reduction plots (Fig. 1B), which supported the value of retaining heterogeneity in revealing tumor complexity. We then implemented a series of quality control measures. The distribution of data after quality control was visually presented through violin plots, showing the effectiveness of data cleaning (Fig. 1A). We used a resolution of 0.7 for clustering, ultimately identifying 16 different cell clusters. To visually present the distribution and relationships of these cell clusters, we opted for the t-distributed stochastic neighbor embedding (t-SNE) method (Fig. 1C). We clearly identified various cell types, including endothelial cells, fibroblasts, T cells, etc. (Fig. 1D). Furthermore, Fig. 1E used t-SNE visualization to display the expression patterns of selected cell markers. The effectiveness of this classification was further validated through enrichment analysis of different cell types (Supplementary Fig. 1A). To show more clearly the epithelial cell subtypes we identified, we visualized the marker genes of luminal and basal cells by bubble plots (Fig. 1G). Analyzing the proportions of these cell types in different samples, we found that the rankings of these cell types were relatively stable across samples, with Luminal cell types always having the highest proportion. However, when comparing tumor group samples with tumor-adjacent normal samples, we observed that the proportion of Luminal cells was significantly higher in tumor group samples than in tumor-adjacent normal samples (Fig. 1F), suggesting that the change in the proportion of Luminal cells in the tumor microenvironment might be related to the development of prostate cancer. Additionally, the accuracy and reliability of cell type identification were further validated by the expression heatmap of cell type-specific marker genes (Fig. 1H), ensuring the precision of our atlas construction and support for subsequent analysis. We applied the Slingshot package to estimate the approximate structure between cell types (Fig. 1I and J). Enrichment analysis results provided key insights into the specific functions of cell types (Supplementary Fig. 1A).

Fig. 1.

Single-cell transcriptomic landscape analysis of prostate cancer. A Data quality control. Violin plots depict the number of genes per cell (nFeature_RNA), total transcript counts (nCount_RNA), percentage of mitochondrial genes (percent.mt), and percentage of hemoglobin genes (percent HB) to assess sample quality. B PCA dimensionality reduction of patient samples. Principal component analysis (PCA) based on expression profiles, showing the distribution of cell populations from different patients (ANT for normal tissue, T for tumor tissue). C t-SNE clustering visualization. t-SNE dimensionality reduction technique reveals 16 distinct cell populations, each identified by a different color. D Marker genes of cell populations. Bubble plots show selected marker genes expressed in different cell populations. E Spatial expression patterns of cell marker genes. t-SNE plots display expression patterns of selected marker genes. F Identification and proportional distribution of cell types. t-SNE plots show cell types annotated based on expression profiles, with stacked bar graphs indicating the proportion of cell type distribution in different patient samples. G Marker genes of epithelial cell subtypes. Expression and percentage expression of marker genes for Luminal and Basal cells across different cell populations. H Heatmap of marker gene expression. Displays the expression levels of specific marker genes across different cell types. I and J Cell developmental trajectory. t-SNE and Pseudotime analysis reveal potential developmental paths and differentiation trajectories among cell populations

Cellular communication in the prostate cancer tumor microenvironment

In the predicted cell communication network, T cells received significantly stronger signals than other immune and stromal cells, while Luminal cells were at a lower level in both signal reception and emission (Fig. 2A–C). The strongest communication pathways emitted by fibroblasts included LAMININ, COLLAGEN, PTN, FN1, THBS, IGF, and CD99, whose activation might promote tumor growth and metastasis. The main communication pathways for T cells included MHC-1, ANNEXIN, and SEMA4 (Fig. 2D). NMF decomposition of the prostate cancer tumor microenvironment cell communication network revealed four communication patterns, with T cells and myeloid cells sharing pattern 4, possibly reflecting their synergistic action in immune response, inflammation regulation, or tumor development (Fig. 2E). Despite differences in communication intensity between the two types of epithelial cells, their communication patterns and structures were similar (Fig. 2F). The MK signaling pathway played a significant role in the communication of Luminal cells with other cell types (Fig. 2G–I). The high interaction between MDK and NCL might have significant implications for cell proliferation, migration, and interactions in the tumor microenvironment (Fig. 2J).

Fig. 2.

Analyzing cellular communication networks in the tumor microenvironment. A Cell–cell interaction network graph, where the thickness of the lines represents the frequency of interaction between different cell types. B Weighted network graph of cell–cell interactions, where the thickness of the lines reflects the strength of interactions, highlighting the main signaling pathways. C Scatter plot showing the difference in signal strength emitted (outgoing) and received (incoming) in cell communication by each cell type, with dot size representing interaction frequency. D Heatmap of the activity level of cell communication signaling pathways obtained through cellchat analysis, showing the modes of signal emission and reception. E River plot representing the outgoing communication patterns of secretory cells, revealing the dominant cell dialogues. F Hierarchical network graph showing the hierarchical relationships and influences of eight cell types in cell communication. G Chord diagram of the MK signaling pathway’s role in cell communication within the tumor microenvironment, showing its mediation between different cell types. H Heatmap of the roles of different cell types in the MK signaling pathway, revealing the distribution of senders, receivers, mediators, and influencers. I Heatmap of the interaction strength in cell communication via the MK pathway, reflecting the intensity of communication between different cell types through this pathway. J Bar graph showing the relative contribution of each ligand-receptor pair in the MK signaling pathway, highlighting the contribution of major communication pairs

Studies on fatty acid metabolism in prostate cancer

By comparing the metabolic intensity of different cell types, we found that Luminal cells exhibited the highest metabolic activity among all cell types tested, particularly showing significant enhancement in three key metabolic pathways: fatty acid elongation, fatty acid degradation, and fatty acid synthesis (Fig. 3A and B). In response to the significant upregulation of fatty acid metabolism in prostate cancer, we further conducted gene set scoring analysis using fatty acid metabolism-related genes, employing five different scoring methods, including AUCell, UCell, singscore, ssgsea, and addmodulescore. Through the composite scores of these methods, we assigned a final fatty acid metabolism score to each cell to ensure the robustness of our scoring system and the reliability of metabolic intensity assessment (Supplementary Fig. 2A). The scores of each cell type through the five scoring methods as well as the final scores are shown through bubble plots (Fig. 3C). Consistent with the results from the scMetabolism package, Luminal cells scored significantly higher in fatty acid metabolism than other cell types (Fig. 3D). We also calculated the average scores for cells from different groups and compared them using Wilcox statistical analysis. The results showed that the scores of Luminal cells in the tumor group were significantly higher than those of Luminal cells in the tumor-adjacent normal group, with a statistically significant difference (p-value < 0.01), further emphasizing the potential link between metabolic characteristic changes and tumor growth, invasion, metastasis, and drug resistance (Fig. 3E).

Fig. 3.

Analysis of cell metabolic levels. A Bubble plot distribution of comprehensive metabolic activity across cell types, analyzed using the scMetabolism package, showing metabolic pathway activity differences in tumor tissue. B Box plots highlight the activity levels of three fatty acid metabolism pathways (synthesis, oxidation, and degradation) across different cell types. C Bubble plot of fatty acid metabolism gene set scores in different cell types, with scores based on the expression levels of specific gene sets. D t-SNE plot showing the distribution of metabolic scores for cells, with color depth representing score magnitude, revealing metabolic heterogeneity among different cell types. E Violin plots compare metabolic scores of cell types in tumor tissue and adjacent normal tissue, showing changes in metabolic states within the tumor microenvironment

Analysis of Luminal epithelial cell properties in the tumor microenvironment

After identifying the significant upregulation of fatty acid metabolism in Luminal cells, we further analyzed the metabolic activity of Luminal cells in each sample (Fig. 4A). In analyzing copy number variations (CNVs) in Luminal cells within the tumor group, we utilized the infercnv package to identify and compare the CNV status of Luminal cells in the tumor group versus the tumor-adjacent normal group. The analysis indicated relatively mild CNV variations in Luminal cells of the tumor group [17] (Fig. 4B).

Fig. 4.

Multidimensional analysis of tumor Luminal cells. A Heatmap of metabolic activity in Luminal cells from different sample sources, analyzed using the scMetabolism package, showing expression patterns across various metabolic pathways. B Copy number variation (CNV) analysis chart of Luminal cells, generated using the inferCNV package, reflecting CNV distribution in Luminal cells from different samples. C Bubble plot showing differences in communication strength between high and low fatty acid metabolism Luminal cells and other cell types, analyzed using the CellCall package, with a focus on different signaling pathways. D Chord diagram of cell communication networks involving Luminal cells, drawn using the CellChat package, revealing the main communication paths between these cells and other cell types. E Chord diagram of signals received by Luminal cells in the cell communication network, also analyzed using the CellChat package, highlighting the diversity of information they receive in the network. F–G Bubble plots show the activity level of major ligand-receptor pairs in communication between Luminal cells and other cell types, obtained through communication probability analysis with the CellChat package. H–K Gene set enrichment analysis (GSEA) charts for Luminal cells, performed using the fgsea package, displaying pathways related to prostate cancer, ribosome, fatty acid metabolism, and antigen processing and presentation, with positive and negative enrichment scores reflecting the activity differences of these pathways in Luminal cells

Exploring the cell communication patterns of Luminal cells under the influence of fatty acid metabolism, we divided Luminal cells into high and low fatty acid metabolism groups based on the median score of fatty acid metabolism and conducted a more comprehensive cellcall cell communication analysis (Fig. 4C). In parts D and E of Fig. 4, we showed the intensity of signals sent and received by Luminal cells to other cell types. Notably, Luminal cells engaged in significant intercellular communication with myeloid cells and endothelial cells through the APP-CD74 signaling pathway. Furthermore, communication with endothelial cells revealed the pathway’s potential role in tumor angiogenesis, crucial for tumor growth and metastasis, as it provides the tumor with necessary nutrients and oxygen supply (Fig. 4F). Further analysis showed extremely significant intercellular communication between Luminal cells and T cells through the HLA-E-CD8A, HLA-C-CD8A, HLA-B-CD8A, and HLA-A-CD8A pathways. These pathways involve interactions between major histocompatibility complex (HLA) molecules and CD8 + T cells, crucial for activating CD8 + T cells to recognize and kill tumor cells. Findings in the GSEA enrichment analysis, especially the top enrichment of Luminal cells in prostate cancer and fatty acid metabolism, and the bottom enrichment in ribosome, antigen processing, and presentation, significantly highlight the critical role of Luminal cells in the development of prostate cancer (Fig. 4H–K).

NMF analytical study of fatty acid metabolism in Luminal epithelial cells

We applied the NMF method to analyze the expression patterns of genes related to fatty acid metabolism in single-cell data. Through the NMF algorithm, we decomposed Luminal cells into 11 distinct cell clusters based on expression patterns (Fig. 5A). Subsequently, we constructed pseudotemporal trajectories of these cells using a pseudotime analysis method, these genes were categorized into 6 groups based on their expression timing, as shown in the heatmap (Fig. 5B). The cellular developmental trajectory showed multiple branches, dividing the entire trajectory into 9 states, including 4 branching points (Fig. 5C and D). Differential expression analysis of the cell clusters obtained from NMF analysis identified genes that were highly expressed in each cluster compared to all others, with statistical significance at a p-value of less than 0.05. When the most differentially expressed gene in a cell cluster was related to fatty acid metabolism and the differential expression logFC value was greater than 1, we defined this cluster as a Luminal cell subpopulation specific to fatty acid metabolism-related genes. Accordingly, we identified 6 Luminal cell subpopulations significantly related to fatty acid metabolism, including SCD + LE, ECHS1 + LE, ACSL1 + LE, ACLY + LE, ECH1 + LE, ACADVL + LE (Fig. 5E). In addition, luminal cells contain groups in which no fatty acid metabolism-related genes were identified or such genes were not significantly differentially expressed, as well as subpopulations of cells associated with programmed cell death. After identifying the six newly defined Luminal cell subpopulations, this study further explored the intercellular communication patterns between these subpopulations and with Basal cells. Notably, the communication activity between the SCD + LE subpopulation and Basal cells was the most significant, indicating that this subpopulation plays a key role in information transfer between Luminal and Basal cell populations. This communication process primarily occurs through the MK, PTN, and VISFATIN pathways for signal sending and the ncWNT pathway for signal reception, highlighting the importance of these pathways in intercellular communication (Fig. 5F, G, and H). Moreover, the ECHS1 + LE subpopulation also occupies an important position in the cell communication network, with its communication with other cell types also being significant (Fig. 5I). From a metabolic perspective, the SCD + LE and ECHS1 + LE subpopulations have significantly higher metabolic activity than other Luminal subpopulations, underscoring their potential key roles in regulating energy supply and metabolic remodeling in the tumor microenvironment (Fig. 5J).

Fig. 5.

Non-negative matrix factorization (NMF) analysis of Luminal cells. A NMF analysis of fatty acid metabolism-related gene expression in Luminal cells, resulting in the decomposition of tumor group Luminal cells into 11 different cell subpopulations. B Heatmap showing the changes in expression of fatty acid metabolism-related genes in Luminal cells over pseudo-time, revealing patterns of gene expression evolution over time. C Pseudo-temporal trajectory plot from single-cell trajectory analysis, with different colors representing different developmental stages of cells over time. D Pseudo-temporal trajectory plot based on NMF results, showing the evolutionary paths of different cell subpopulations. E UMAP plot annotated with Luminal cell types following NMF, identifying six subtypes of Luminal cells closely related to fatty acid metabolism. F–G Chord diagrams reveal the interactions between cell subtypes identified by NMF analysis, showing the number (F) and weight/strength (G) of interactions between cell subtypes. H Heatmap of the activity levels of pathways in different Luminal cell subtypes, showing the differences in activity of signaling pathways in cell communication. I Heatmap of the roles of cell types in the MK signaling pathway, revealing the distribution of senders, receivers, mediators, and influencers. J Bubble plot analysis of the activity levels of various metabolic pathways in different subpopulations of Luminal cells, with bubble size and color depth reflecting relative metabolic activity levels

Metabolic characterization in spatial transcriptome data

After processing the spatial transcriptome sequencing data of three prostate samples with the Seurat package, we compared the count values of normal, acinar cell carcinoma, and adenocarcinoma with infiltrating cancer prostate samples (Supplementary Figs. 3A, C, and E). After dimensionality reduction clustering of the spatial transcriptome data, we continued with tSNE for dimensionality reduction visualization (Supplementary Figs. 3B, D, and F). By comparing the changes in color depth on HE-stained slides, we were able to verify the accuracy of the clustering effect. Furthermore, by using the scMetabolism package for metabolic analysis of the spatial transcriptome data, focusing on glycolysis, oxidative phosphorylation, purine metabolism, pyrimidine metabolism, and three fatty acid metabolism pathways, we mapped the metabolic intensity to HE-stained slides (Fig. 6A, C, and E). At the same time, we also demonstrated the specific metabolic levels of each cell cluster in the three samples, further highlighting the differences in metabolic characteristics between different types of prostate cancer (Fig. 6B, D, and F).Integrating spatial transcriptomics with single-cell data for deconvolution and analysis of cell interactions.

Fig. 6.

Spatial transcriptomics analysis revealing changes in metabolic activity. A Spatial transcriptomics data from normal prostate tissue sections, analyzed using the Seurat package for dimensionality reduction clustering, showing the activity heatmap of different metabolic pathways. B Bubble plot analyzed using the scMetabolism package, displaying the metabolic activity level of different cell clusters in normal prostate tissue, highlighting the performance of each cluster in various metabolic pathways. C Spatial transcriptomics data from prostate adenocarcinoma tissue sections, with multiple metabolic pathway activity heatmaps post-Seurat package clustering, reflecting the metabolic characteristics of transformed tissue. D Bubble plot of metabolic activity levels of cell clusters in prostate adenocarcinoma tissue, revealing metabolic heterogeneity in the tumor microenvironment. E Spatial transcriptomics data from prostate adenocarcinoma with invasive carcinoma tissue sections, showing the activity heatmap of cell clusters and related metabolic pathways analyzed by the Seurat package. F Bubble plot of metabolic activity levels of cell clusters in prostate adenocarcinoma with invasive carcinoma tissue sections, analyzed using the scMetabolism package, highlighting the differences in metabolic states during tumor progression

Due to the limitations of spatial transcriptomics technology, current spatial transcriptome data have not yet achieved the single-cell resolution of single-cell sequencing data (Fig. 7A and G). On the basis of deconvolution analysis of two tumor samples, we further applied the MISTy (Multiview Intercellular SpaTial modeling framework) framework for spatial transcriptome cell interaction analysis. Our analysis results, presented in bar graphs, show the contribution of three different views to cell interactions, finding that intraview and paraview15 contribute the most in the two tumor samples (Fig. 7B and J). This reveals the importance of intracellular regulation and paracrine regulation in tumor samples. Further heatmap and network graph analyses reveal the specific patterns of these two views in tumor samples, highlighting the significant interaction relationships between Luminal cells and other cell types (such as smooth muscle cells and fibroblasts), consistent with the findings of cell communication networks in single-cell data (Fig. 7C, D, E, F, H, I, K, and L).

Fig. 7.

Deconvolution and cell interaction analysis based on spatial transcriptomics data. A Spatial distribution probability of cell types from normal prostate tissue section data analyzed using the RCTD deconvolution method. B Bar graph using the Mistyr package, assessing the contribution of different views (view) to cell interactions during the interaction process, showing the relative importance of different views in cell interactions. C–D Heatmap and network graph of cell interactions within intraview, revealing the strength and patterns of interactions within the same cell type. E–F Heatmap and network graph of cell interactions within paraview15, showing the strength and communication network of cross-cell type interactions. G RCTD deconvolution analysis results of prostate adenocarcinoma with invasive carcinoma tissue sections, displaying the probability and spatial distribution of different cell types. J Bar graph in prostate adenocarcinoma with invasive carcinoma tissue, assessing the contribution of different views to cell interactions, evaluating the relative contributions of each view. H–I Heatmap and network graph of intraview cell interactions in prostate adenocarcinoma with invasive carcinoma tissue, showing the interaction relationships among the same type of cells in the tumor environment. K–L Heatmap and network graph of cell interactions within paraview15 in the same tissue, revealing the strength and network structure of interactions across different cell types

Developmental trajectories revealed by spatial transcriptome data

Spatial transcriptomics data provide transcriptional information of cells at precise locations within tissues. We utilized the stLearn package to conduct in-depth analysis of spatial transcriptomics data to explore the developmental process of tumors, including issues of invasion and metastasis. By integrating data quality control and dimension reduction with NumPy, and clustering using stLearn’s Louvain method, we identified different cell clusters in prostate adenocarcinoma samples (Fig. 8A and B). We focused on cell clusters with higher malignant potential, using the Diffusion Pseudotime (DPT) algorithm to reconstruct developmental trajectories, combined with spatial coordinate information, to reveal the stepwise invasion and metastasis process of tumor cells in a pseudo-temporal sequence (Fig. 8C and E). Divergent bar charts from trajectory analysis revealed gene expression changes based on trajectory differences, showing genes upregulated and downregulated throughout the tumor development process from start to finish (Fig. 8D and F). Prognostic study of FAM-related prognosis in prostate cancer combining NMF subgroups with clinical data.

Fig. 8.

Spatial developmental trajectory analysis of prostate cancer tissue. A Clustering map obtained by classifying prostate adenocarcinoma tissue section sequencing data using the louvain method in the stLearn package, showing the spatial distribution of different cell groups. B Image of prostate adenocarcinoma with invasive carcinoma tissue sections post-louvain clustering, showing the spatial heterogeneity of cell groups in the tumor microenvironment. C Spatial developmental trajectory map of cells in high fatty acid metabolism areas of adenocarcinoma tissue sections, drawn using the stLearn package, reflecting the impact of fatty acid metabolism activity on cell state changes. D Divergent bar graph of differentially expressed genes related to the developmental trajectory in adenocarcinoma tissue, statistically analyzed using numpy, revealing key regulatory genes associated with developmental trajectories. E Spatial developmental trajectory map of cells in high fatty acid metabolism areas of adenocarcinoma with invasive carcinoma tissue sections, also analyzed using the stLearn package. F Divergent bar graph of differentially expressed genes related to the developmental trajectory in adenocarcinoma with invasive carcinoma tissue, showing trends and correlations in gene expression changes

Given the significance of the new Luminal cell subtype related to fatty acid metabolism identified by NMF for tumors, to understand whether these six key fatty acid metabolism-related genes are associated with the clinical prognosis of prostate cancer, we conducted an in-depth prognostic analysis using the TCGA dataset and clinical information. Initially, we extracted positive cells of the six key fatty acid metabolism genes in Luminal cells, and through differential analysis between positive and negative cells, we identified marker genes of key gene positive cells. Then, in the TCGA data, by quantifying the expression of each marker gene, patients were divided into high and low groups for survival analysis. The survival analysis results showed that patients with high expression of ECHS1 and ECH1 cell marker genes have a poorer prognosis, showing statistical significance (Fig. 9A and D). Additionally, we observed the expression of these two prognostically valuable fatty acid metabolism-related genes in the spatial transcriptome data of two tumor samples (Fig. 9B, C, E, and F). After determining the prognostic value of these two Luminal cell subtypes, we directly explored the significance of these two fatty acid metabolism-related genes in prostate cancer. We performed ssGSEA scoring on 11 prostate cancer RNA-seq datasets, including sequencing data and its prognosis information from TCGA, GEO, and ICGC. The results of Cox regression analysis indicated that in most datasets, high expression of these two genes is closely related to poor prognosis of patients, i.e., HR values greater than 1, consistent with the analysis results of Kaplan–Meier curves (Fig. 9G). Further GO enrichment analysis results showed that highly expressed genes are significantly enriched in biological processes such as cytoplasm, membrane-bound organelles, and protein binding. These enrichment results reveal that prostate cancer cells may support their rapid proliferation and survival by regulating fatty acid metabolism, including enhancing fatty acid synthesis and oxidation, providing raw materials and energy sources for cell membrane biosynthesis, thereby reflecting the molecular mechanism of fatty acid metabolism reprogramming related to the development of prostate cancer (Fig. 9H).

Fig. 9.

Association analysis of fatty acid metabolism-related Luminal cell subtypes with clinical prognosis in patients with prostate cancer. A Kaplan–Meier survival curve analysis showing the survival probability difference between prostate cancer patients with high and low expression of ECHS1-related Luminal cell subtypes. B–C Spatial expression heatmaps of the ECHS1 gene in prostate adenocarcinoma and adenocarcinoma with invasive carcinoma tissue sections, showing the expression levels of ECHS1 in different areas. D Another set of Kaplan–Meier survival curves, showing the survival probability comparison of prostate cancer patients expressing different levels of ECH1-related Luminal cell subtypes. E–F Spatial expression heatmaps of the ECH1 gene in prostate adenocarcinoma and adenocarcinoma with invasive carcinoma tissue sequencing data. G Forest plot of Cox regression analysis results of ECHS1 and ECH1 in multiple prostate cancer bulk sequencing datasets, assessing the correlation of these two genes with overall survival (OS), disease-free survival (DFS), recurrence-free survival (RFS), disease-specific survival (DSS), and progression-free survival (PFS) in prostate cancer patients. H Bar chart from enrichment analysis of overexpressed ECHS1 and ECH1, revealing biological processes, cellular components, and molecular functions related to their overexpression, showing their roles and impacts on cell metabolism

Discussion

Fatty acids play a crucial role in the human body, not only as fundamental components for building cell membranes but also as sources for energy supply and the synthesis of other key biomolecules [18]. In the field of tumor biology, metabolic reprogramming is a significant characteristic that distinguishes cancer cells from normal cells, enabling them to adjust energy production and biosynthetic pathways to support their rapid proliferation and growth. In tumor development, fatty acid metabolism, like other key metabolic processes such as glycolysis, amino acid metabolism, and nucleotide metabolism, plays an important role. A deep understanding of these metabolic processes and their roles in tumor development is crucial for developing innovative cancer treatment strategies. This not only reveals how cancer cells adapt their metabolic demands to support their uncontrolled growth and survival but also provides opportunities for discovering new therapeutic targets[19]. Especially in the study of prostate cancer, it has been found that tumor cells gain an energy generation advantage by enhancing the absorption and metabolism of fatty acids, thereby boosting their β-oxidation. This metabolic reprogramming not only meets the high energy demands of cancer cells but may also alter the composition and function of cell membranes, promoting the survival and spread of tumor cells. Fatty acid metabolism not only provides essential energy for tumor cells but also promotes the aggressiveness and migration capability of tumor cells by changing the fluidity of cell membranes and affecting signaling pathways. Moreover, changes in fatty acid metabolism may also impact the tumor microenvironment, such as by promoting angiogenesis, providing new pathways for tumor metastasis [20]. Therefore, fatty acid metabolism plays a key role in the development and metastasis of prostate cancer, offering possibilities for researching and developing new treatments targeting prostate cancer.

In prostate cancer tumor tissues, luminal cells exhibit significantly higher levels of fatty acid metabolism compared to other cell types, and the fatty acid metabolism score of luminal cells in tumor tissues is higher than that in normal tissues. This phenomenon indicates the crucial role of fatty acid metabolism in the development and progression of prostate cancer. To support rapid growth and division, tumor cells require substantial energy and biosynthetic materials. The increase in fatty acid metabolism may fulfill these demands, including energy production, cell membrane synthesis, and signaling molecule generation [3]. Fatty acids serve not only as an energy source but may also promote tumor growth and metastasis by affecting intercellular communication within the tumor microenvironment and the function of immune cells. In rapidly growing and dividing tumor cells, fatty acid metabolism provides essential energy and biosynthetic molecules necessary for cell membrane construction, supporting tumor cell survival and spread. Fatty acid metabolism can regulate the function of immune cells by producing specific metabolites, such as affecting the activity of T cells and macrophages, potentially leading to immune evasion and enabling tumor cells to avoid detection and elimination by the immune system. Metabolites from fatty acid metabolism act as signaling molecules, participating in interactions between tumor cells and other cell types within the TME (such as fibroblasts, immune cells, and endothelial cells), promoting the establishment and maintenance of the TME, which is crucial for tumor growth and metastasis. The increase in fatty acid metabolism in luminal cells may be associated with the worsening of prostate cancer and poorer prognosis. Changes in fatty acid metabolism may enhance the invasiveness and metastatic capability of tumors, impacting patient survival rates. The specificity enhancement of fatty acid metabolism pathways offers new therapeutic targets. Inhibiting key enzymes in fatty acid synthesis or oxidation could block the tumor cells’ ability to obtain energy and biosynthetic materials, thereby inhibiting tumor growth and metastasis.

By applying Non-negative Matrix Factorization (NMF) analysis, we succeeded in identifying a new subpopulation of tubulointerstitial cells significantly associated with fatty acid metabolism and found its association with programmed cell death [21]. This discovery underscores the prominent role of fatty acid metabolism within specific cellular subgroups in prostate cancer and its close connection with tumor progression and patient prognosis. Using the TCGA dataset and corresponding clinical information, a profound prognostic analysis of these fatty acid metabolism-related genes was conducted. By categorizing patients into high and low groups based on the expression levels of these gene markers, it was found that high expressions of ECHS1 and ECH1 are significantly associated with poor prognosis in prostate cancer patients. ECHS1 and ECH1 play a central role in the fatty acid β-oxidation process in prostate cancer cells, a process that not only provides the necessary energy source for tumor cells but also relates to the capacity for cell proliferation, survival, and malignant transformation. ECHS1 plays a key role in the fatty acid β-oxidation process, involving the metabolism of short-chain and medium-chain fatty acids. By catalyzing the hydration step of enoyl-CoA, it assists cells in utilizing fatty acids as an energy source, which is crucial for cell proliferation and survival. In tumor cells, where energy demands are increased, the role of ECHS1 may become particularly important, enabling cells to efficiently utilize fatty acids for energy metabolism to support rapid proliferation. Therefore, high expression of ECHS1 may reflect an increased dependency of tumor cells on energy metabolism, which is associated with poor prognosis in prostate cancer patients [22]. Similarly to ECHS1, ECH1 also participates in fatty acid β-oxidation, but it primarily acts on long-chain fatty acids. This indicates that ECH1 plays a significant role in maintaining cellular lipid balance and energy metabolism. In prostate cancer, high expression of ECH1 may facilitate tumor cells’ utilization of long-chain fatty acids, providing an additional energy source for proliferation and survival. This metabolic flexibility may enable tumor cells to adapt more effectively to various microenvironment conditions, thereby enhancing their invasiveness and malignancy.

Given the critical roles of ECHS1 and ECH1 in fatty acid β-oxidation and their significant association with poor prognosis in prostate cancer, inhibitors targeting these enzymes could directly slow down tumor cell energy metabolism, limiting their growth and survival capabilities. By reducing the efficiency of fatty acid metabolism, the energy supply to tumor cells can be weakened, thereby inhibiting tumor development [23]. Besides directly targeting ECHS1 and ECH1, disrupting the entire fatty acid metabolism pathway is also a potential therapeutic strategy. This could be achieved by targeting upstream or downstream metabolic pathways, such as inhibiting key enzymes in the fatty acid synthesis pathway or promoting the oxidation of fatty acids, to adjust the metabolic balance within tumor cells [24]. Considering that tumor cells may evade the impact of a single therapeutic strategy through various pathways, combining inhibitors targeting ECHS1 and ECH1 with other treatment methods, such as chemotherapy, radiotherapy, or other metabolic pathway inhibitors, might enhance treatment efficacy. This combined strategy could attack multiple survival mechanisms of tumor cells simultaneously, reducing the development of resistance [25]. Therapeutic strategies targeting ECHS1 and ECH1 genes in prostate cancer not only offer new treatment targets but also provide important insights into the energy metabolism mechanisms of tumor cells. As the role of fatty acid metabolism in tumor development is further unveiled, therapies targeting this metabolic pathway could represent a significant advancement in the treatment of prostate cancer. Future research should focus on developing and validating therapeutic drugs targeting these metabolic pathways, as well as assessing the potential of these treatment strategies in clinical applications.

Limitations

Despite revealing the important roles of ECHS1 and ECH1 in prostate cancer, our study has some limitations. First, the limited sample size and data heterogeneity may affect the generalization and reliability of the results. Future studies should include larger sample sizes and integrate data from different sources for comprehensive analysis. Second, we mainly relied on single-cell RNA sequencing and spatial transcriptomics data, which, although they provide high-resolution information, also have limitations in sequencing depth and coverage. Future studies should incorporate other high-throughput technologies, such as mass spectrometry and multi-omics data, to fully elucidate the functions of fatty acid metabolism-related genes in prostate cancer. Third, our study focused on the effects of ECHS1 and ECH1 on tumor cell function and tumor prognosis, with less emphasis on function in other cell types, especially immune cells. Future studies should expand the scope of the study to fully understand the effects of fatty acid metabolism on the tumor microenvironment in prostate cancer. In addition, although our analysis was relatively adequate, in vivo experimental data are still lacking. Future studies should validate the specific roles of ECHS1 and ECH1 in prostate cancer using animal models. Finally, we explored the relationship between ECHS1 and ECH1 and immunotherapeutic responses, but the specific mechanisms remain unclear. Future studies should deeply analyze how ECHS1 and ECH1 regulate the immune microenvironment and their applications in immunotherapy.

Conclusion

This study delves into the role of fatty acid metabolism in prostate cancer, identifying new therapeutic targets. The research demonstrates that fatty acid metabolism is significantly upregulated in Luminal cells, highlighting their key role in tumor aggressiveness and metastatic potential. Utilizing advanced single-cell and spatial transcriptomics, this study unveils the cellular dynamics and metabolic reprogramming within the tumor microenvironment, providing insights for personalized treatment strategies. Specifically, the identification of Luminal cell subpopulations with distinct metabolic characteristics emphasizes the heterogeneity of prostate cancer, paving the way for targeted therapy. This conclusion integrates the complex interplay of cell communication, metabolic pathways, and tumor microenvironment interactions, underscoring the potential of metabolic reprogramming as a therapeutic avenue.

Author contributions

YZ and ZJ conceptualized the study. ZJ was responsible for the data analysis. ZJ also took on the visualization tasks. The manuscript was written by YZ and ZJ. Furthermore, YZ was involved in reviewing and editing the manuscript.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data availability statement

The datasets analyzed for this study can be found in the Gene Expression Omnibus (GEO): single-cell sequencing data from dataset GSE176031 at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc = GSE176031, spatial transcriptomics data from 10X Genomics website (https://www.10xgenomics.com/), and bulk sequencing data in the Xena database for TCGA at https://xenabrowser.net/datapages/?cohort = GDC%20TCGA%20Prostate%20Cancer%20(PRAD)&removeHub = https%3A%2F%2Fxena.treehouse.gi.ucsc.edu%3A443. Additionally, all analysis codes, gene sets, and raw data involved in this study have been uploaded to jianguoyun website, accessible via https://www.jianguoyun.com/p/DfetgrgQ85jYCxiK09kFIAA.

Code availability

The code used for the analysis has been uploaded to jianguoyun website, accessible via https://www.jianguoyun.com/p/DfetgrgQ85jYCxiK09kFIAA.

Declarations

Conflict of interest

The authors declare no competing interests.