Introduction

Neuroblastoma (NB) is a malignancy derived from sympathetic nervous system neural crest cells and represents one of the most prevalent solid tumors in childhood, comprising about 8–10% of all pediatric cancers1. Approximately 40–50% of cases are diagnosed within the first year of life, and about 90% occur before the age of ten1. NB shows marked biological heterogeneity, spanning from spontaneously regressing low-risk types to highly invasive and drug-resistant high-risk types2.

Metabolic reprogramming, serving as a crucial mechanism for tumor cells to adapt to microenvironmental changes and maintain their proliferative ability, has been designated as one of the “emerging hallmarks” of cancer. In contrast to normal cells, tumor cells display profound alterations in energy metabolism, carbon utilization, and metabolic biosynthetic pathways. Such systemic metabolic reprogramming not only facilitates rapid tumor cell proliferation but also empowers tumor cells to escape immune surveillance, resist apoptosis, and develop treatment resistance3. In NB, MYCN amplification is a well-known marker of poor prognosis and has been mechanistically linked to oncogenic metabolic rewiring, including alterations in mitochondrial function, oxidative stress response, and protein synthesis4,5. For instance, MYCN-amplified NB cells display enhanced sensitivity to the combined inhibition of mitochondrial complex I (via phenformin) and monocarboxylate transporter 1 (MCT1; via AZD3965), leading to lactate accumulation, oxidative stress, and impaired ATP production6.

As high-throughput sequencing technologies rapidly evolve, bioinformatics, a key branch of data-driven biology, has become increasingly essential in functional genomics, disease mechanism exploration, and personalized medicine. In particular, the emergence of single-cell RNA sequencing (scRNA-seq) technology in recent years has provided unprecedented resolution for deciphering cellular heterogeneity, lineage relationships, and microenvironmental dynamics7. scRNA-seq allows researchers to map the intricate metabolic landscapes inside tumors at the cellular level. For instance, different tumor cell subpopulations may exhibit significant differences in glycolysis, oxidative phosphorylation, or fatty acid oxidation. These metabolic characteristics are closely associated with tumor cell stemness, invasiveness, or therapeutic sensitivity. By integrating scRNA-seq data with metabolic pathway databases, the expression patterns and regulatory networks of key metabolic genes can be inferred, facilitating the identification of potential metabolic targets8.

In high-risk NB, bone marrow (BM) represents the most common site of distant metastasis and plays a central role in disease progression, immune suppression, and therapy resistance. Compared to primary tumor tissues, BM-residing metastatic NB cells are subject to distinct microenvironmental pressures and may acquire unique metabolic and immunological features that promote survival and dissemination2. However, the molecular characteristics of these metastatic cells remain incompletely defined. In this study, we focused on scRNA-seq data derived from BM aspirates of NB patients—with and without marrow infiltration—to systematically explore the transcriptional programs underlying metabolic reprogramming in the context of metastasis. This approach enables us to uncover mechanisms of tumor–immune interaction and microenvironmental adaptation that are often masked in bulk primary tumor analyses. Through integrative bioinformatics analyses and functional validation, we aim to identify key metabolic regulators and novel therapeutic vulnerabilities in bone marrow-infiltrating NB.

Materials and methods

Data acquisition

The full name of the GEO database (https://www.ncbi.nlm.nih.gov/geo/info/datasets.html) is Gene Expression Omnibus, a gene expression database established and maintained by the National Center for Biotechnology Information (NCBI) in the United States. Single-cell RNA-sequencing data were obtained from the GEO database (GSE216155), comprising 17 bone marrow samples from 16 NB patients. Among them, 11 samples were infiltrated by metastatic NB cells, while 5 non-infiltrated samples served as controls9. This dataset enables direct comparison between bone marrow-infiltrating NB cells and non-tumor bone marrow, providing a valuable model to investigate the transcriptional basis of NB metastasis and microenvironmental adaptation.

TARGET (https://portal.gdc.cancer.gov/projects/TARGET-NBL)10, currently the largest database of cancer genomic information, contains a wide range of data such as gene expression, miRNA expression, lncRNA expression, copy number variations, DNA methylation, and SNPs. A total of 155 processed raw expression datasets from NB patients were downloaded.

Single-cell data preprocessing and quality control

Expression data were loaded using the Seurat package11. Quality control was performed by filtering cells based on total UMI counts, number of detected genes, and the proportion of mitochondrial gene expression. Mitochondrial gene expression percentage was calculated as the proportion of mitochondrial gene counts to the total gene counts. Cells with high mitochondrial content and low RNA expression were considered apoptotic. We applied median absolute deviation (MAD)-based filtering, removing cells with values exceeding 3 MADs from the median. Doublets were removed using DoubletFinder (v2.0.4)12, applied independently to each sample to ensure clean cell populations.

Single-cell data dimensionality reduction and clustering

We applied the global normalization method LogNormalize, scaling each cell’s total expression to 10,000 by multiplying a coefficient s0, followed by log-transformation for normalization. CellCycleScoring was used to compute cell cycle scores. FindVariableFeatures was applied to detect highly variable genes. The ScaleData function was used to regress out gene expression variability caused by mitochondrial gene percentage, ribosomal gene percentage, and cell cycle phase. Principal component analysis (PCA) was applied via RunPCA for linear dimensionality reduction, and “Harmony” R package13 was used to eliminate batch effects. Finally, RunUMAP was applied to perform nonlinear dimensionality reduction based on the Uniform Manifold Approximation and Projection (UMAP) algorithm.

Cell type annotation

Cell type annotation was performed mainly based on CellMarker14 and PanglaoDB15 databases and relevant literature, with assistance from automated annotation via the SingleR package16. We employed the FindAllMarkers function to screen for marker genes in each cluster. The screening parameters were: only.pos = TRUE, min.pct = 0.25, and logfc.threshold = 0.25. That is, only positively expressed marker genes were selected, with expression in at least 25% of the cells and a differential expression threshold of LogFC > 0.25.

Ligand-receptor interaction analysis

CellChat17 is a tool that enables quantitative inference and analysis of intercellular communication networks from single-cell data. It utilizes network analysis and pattern recognition methods to predict the major signaling inputs and outputs of cells, and how these cells and signals coordinate to perform functions. In this study, the normalized single-cell transcriptomic data were used as input, along with the cell subtypes identified from single-cell analysis, to investigate cell-cell interactions. The strength (weights) and frequency (count) of interactions between cells were used to quantify the closeness of intercellular communication. This allowed us to evaluate the activity and influence of each cell type within the tumor microenvironment.

Transcriptional regulatory analysis

SCENIC (single-cell regulatory network inference and clustering)18 is a computational approach that reconstructs gene regulatory networks and identifies cellular states from single-cell transcriptomic data using co-expression and motif enrichment analysis. Initially, GEne Network Inference with Ensemble of trees (GENIE3)19 is employed to identify sets of genes that are co-expressed with specific transcription factors (TFs). Each co-expression module undergoes motif enrichment analysis, where only motifs with significant enrichment are retained. Motifs are annotated with their corresponding TFs based on reference databases, with the annotations classified into high- or low-confidence levels. TFs identified through direct database annotation or homology-based inference are considered high-confidence, whereas those annotated via motif sequence similarity are considered low-confidence. The retained motifs are used to score genes within the co-expression modules, identifying high-scoring genes (assumed to be near the transcription start sites (TSS) of these motifs). Genes with low motif scores are removed from the co-expression modules, and the remaining gene sets are defined as regulons. Each regulon comprises a TF and its directly regulated target genes; SCENIC then scores the activity of each regulon across individual cells. Based on gene expression, a higher score indicates stronger activation of the gene set, and the resulting activity matrix helps in distinguishing cell types and functional states.

Random survival forest analysis

We applied the randomForestSRC R package to carry out feature selection. The random survival forest algorithm was also used to rank the importance of prognostic genes, with nrep = 500 representing 500 iterations in the Monte Carlo simulation. Genes with relative importance values exceeding 0.4 were identified as the final marker genes.

Immune infiltration analysis based on metabolic gene expression

To investigate the potential interaction between tumor metabolic activity and the immune microenvironment, we systematically analyzed the relationship between the expression of prognostic metabolic genes and the abundance of immune cell populations in the bone marrow. For this purpose, five metabolism-associated genes (MRPL21, NHP2, RPL13, RPL18A, and RPL27A) identified through single-cell and bulk transcriptomic integration were used as stratification markers. For each gene, samples from the TARGET-NB bulk RNA-seq cohort were divided into high and low expression groups based on median expression levels. The CIBERSORT algorithm20 was then used to estimate the relative proportions of 22 immune cell types per sample. Group-wise comparisons of immune infiltration were conducted using Wilcoxon rank-sum tests. In addition, Spearman correlation analyses were performed to examine continuous relationships between gene expression and immune cell abundance.

This integrative design allowed us to assess whether high expression of key metabolic genes was associated with immunosuppressive remodeling of the tumor microenvironment, potentially contributing to NB progression.

Gene set enrichment analysis (GSEA)

Based on the expression levels of key genes, samples were divided into high and low expression groups, and GSEA21 was used to further analyze pathway differences between the two groups. The background gene set used for pathway annotation was version 7.0 of the gene set collections downloaded from the MSigDB database22, which was used for differential pathway analysis between groups. Significantly enriched gene sets (adjusted p-value < 0.05) were ranked according to their normalized enrichment scores (NES). GSEA analysis is commonly used to explore disease subtypes in conjunction with biological significance.

Gene set variation analysis (GSVA)

GSVA23 is a non-parametric and unsupervised method used to assess the enrichment of gene sets in transcriptomic data. By scoring predefined gene sets, GSVA converts gene-level expression changes into pathway-level signals, enabling the assessment of biological functions in samples. In our analysis, gene sets were downloaded from the Molecular Signatures Database (MSigDB)24. The GSVA algorithm was used to calculate enrichment scores for each gene set, evaluating potential biological function changes across different samples.

Cell culture and transfection

Human NB cell lines (SH-SY5Y, SK-N-BE2, SK-N-SH) and normal human astrocyte cell line (SVGp12) were purchased from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). Cells were periodically tested for mycoplasma contamination (C0297, Beyotime, Shanghai, China).

In this study, two human NB cell lines, SK-N-BE2 and SH-SY5Y, were selected for in vitro functional assays. SK-N-BE2 was derived from a bone marrow metastasis of a Stage IV NB patient and is characterized by MYCN amplification. SH-SY5Y is a subclone of the SK-N-SH cell line, also derived from a bone marrow metastatic lesion, but lacks MYCN amplification. Due to their distinct differentiation states and metabolic profiles, these cell lines are widely used as complementary models to investigate the generalizability and specificity of candidate gene functions in NB research.

Lipofectamine™ 2000 Transfection Reagent (11668019, Thermo, USA) was mixed with small interfering RNA (siRNA) to transfect NB cells.

The sequences of the si-MRPL21 were as follows:

MRPL21 si-1: (sense: 5’- CUGAUCUUAAUUGGAAAUGAATT − 3’,

antisense: 5’- UUCAUUUCCAAUUAAGAUCAGTT − 3’);

MRPL21 si-2: (sense: 5’- CCGGAUAAACAGCAUUGAGAUTT − 3’,

antisense: 5’- AUCUCAAUGCUGUUUAUCCGGTT − 3’).

Real-time quantitative PCR (RT-qPCR)

The RNA from the cell lines was extracted using TRIzol reagent (Invitrogen), and the RevertAid First-Strand cDNA Synthesis Kit (Thermo Fisher Scientific, Inc.) was used to synthesise cDNA. RT-qPCR analysis was performed using SYBR Green (Takara). Primer sequences utilized in the RT-qPCR are detailed in Supplementary File 1. The mean and standard error for each point were calculated for each sample in three separate reactions. The relative levels of mRNA transcripts were normalized to the control GAPDH.

Western blotting

Total protein lysates were subjected to separation by SDS-PAGE and subsequently transferred onto polyvinylidene difluoride (PVDF) membranes. The membranes were blocked and incubated overnight at 4 °C with primary antibodies specific for MRPL21 (16978-1-AP; Proteintech) and GAPDH (#5174; Cell Signaling Technology). After primary antibody incubation, the membranes were washed three times with Tris-buffered saline containing 0.1% Tween-20 (TBST), followed by incubation with horseradish peroxidase (HRP)-conjugated secondary antibodies (#7074; Cell Signaling Technology) at room temperature. After a second series of TBST washes, the immunoreactive bands were detected using an enhanced chemiluminescence (ECL) substrate according to the manufacturer’s instructions.

Cell counting kit-8 (CCK-8) assay

Transfected SH-SY5Y and SH-N-BE2 cells were seeded into 96-well plates at a density of 3 × 10⁴ cells per well and incubated for 0, 24, 48, and 72 h. At each time point, 10 µL of CCK-8 solution (BS350B, Biosharp, Hefei, China) was added to each well, followed by incubation at 37 °C for 1 h. The absorbance was measured at 450 nm using a microplate reader, and cell viability was quantified based on the optical density values.

Wound healing assay

SH-SY5Y and SH-N-BE2 cells transfected with siRNA were seeded into 6-well plates (2 × 10⁶ cells/well) and cultured to full confluence. A straight scratch was made using a sterile 10 µL pipette tip, and detached cells were removed by PBS wash. After incubation for 24 h, images were captured at 0, 24, and 48 h using a 40× inverted microscope. Wound closure was quantified using ImageJ software (version 1.53t, National Institutes of Health), and migration ability was calculated as the percentage reduction in scratch area over time.

Transwell invasion assay

Cell invasion was evaluated using Transwell chambers pre-coated with Matrigel matrix. Transfected SH-SY5Y and SK-N-BE2 cells (2 × 10⁶ cells/well) were seeded into the upper chambers in 100 µL serum-free medium, while the lower chambers were filled with medium supplemented with 10% fetal bovine serum (FBS) as a chemoattractant. After 48 h of incubation, the inserts were removed, washed twice with PBS, fixed in 4% paraformaldehyde for 15 min, and stained with 0.1% crystal violet for 10 min. Non-invaded cells were removed from the upper membrane surface, and invaded cells on the lower surface were imaged under a light microscope at 100× magnification. The number of invaded cells was quantified using ImageJ software by counting stained cells in at least five randomly selected fields per insert.

Mitochondrial oxygen consumption rate (OCR) assay

Cellular OCR was measured using the Seahorse XFe96 Extracellular Flux Analyzer (Agilent Technologies) following the manufacturer’s protocol. Briefly, transfected SH-SY5Y and SK-N-BE2 cells were seeded in XF96 microplates (1 × 10⁴ cells/well) and incubated overnight. Prior to measurement, the culture medium was replaced with Seahorse XF Base Medium supplemented with 10 mM glucose, 1 mM pyruvate, and 2 mM glutamine (pH 7.4), and cells were incubated at 37 °C in a non-CO₂ incubator for 1 h. OCR was recorded at baseline and after sequential injection of oligomycin (an ATP synthase inhibitor, 1.5 µM), FCCP (a mitochondrial uncoupler, 1.0 µM), and rotenone/antimycin A (complex I/III inhibitors, 0.5 µM each). Mitochondrial parameters including basal respiration, ATP production, maximal respiration, and spare respiratory capacity were calculated. Data were normalized to cell number and analyzed using Wave software (version 2.6.0, Agilent Technologies).

ATP content assay

ATP levels were measured using the Enhanced ATP Assay Kit (S0027, Beyotime Institute of Biotechnology) following the manufacturer’s instructions. Briefly, cells were harvested and lysed with lysis buffer, and the supernatant was collected after centrifugation at 12,000 ×g for 5 min. Equal volumes of the supernatant and ATP detection working solution were mixed and incubated at room temperature for 3–5 min. Luminescence was then measured using a microplate reader. ATP concentrations were normalized to total protein levels determined by the BCA protein assay.

Mitochondrial membrane potential assay

Mitochondrial membrane potential was evaluated using the JC-1 assay kit (C2006, Beyotime Institute of Biotechnology), and expressed as the ratio of JC-1 aggregate (red) to monomer (green) fluorescence intensity. Flow cytometric analysis was performed using a CytoFLEX flow cytometer (BD Biosciences).

Statistical analysis

Statistical analysis was performed using R software (version 4.3.0), with p-values < 0.05 considered statistically significant. Quantitative data are expressed as the mean ± standard deviation (SD) from three independent biological replicates. Data analysis and visualization were carried out using GraphPad Prism 8.4.2. The normality of data distribution was assessed using the Shapiro–Wilk test. Group comparisons were conducted using one-way ANOVA followed by Dunnett’s post hoc test or an unpaired two-tailed Student’s t-test, as appropriate.

Results

Single-cell data quality control

To ensure data quality across multiple samples, cells with fewer than 500 genes detected were removed according to the filtering formula: (nFeature_RNA > 200 & percent.mt < = median + 3MAD & nFeature_RNA < = median + 3MAD & nCount_RNA < = median + 3MAD & percent.ribo < = median + 3MAD). In this formula, nFeature_RNA refers to the number of genes, nCount_RNA indicates the total UMI counts per cell, percent.mt represents the percentage of mitochondrial reads, and percent.ribo represents the percentage of ribosomal reads. DoubletFinder was subsequently applied to filter out doublets, retaining a total of 68,572 cells, and generating violin and scatter plots after filtering (Figs. 1A-B). We searched for 2,000 hypervariable genes and selected the top 10 genes with the highest standard deviations for display (Fig. 1C). We then conducted standardization analysis (Fig. 1D), principal component analysis (PCA), and harmonization analysis (Figs. 1E) on the data.

Fig. 1
figure 1

Quality control and data standardization of scRNA-seq data. (A) Quality control for each cell in scRNA seq data, displaying the number of nCount_RNA, nFeature_RNA, percent.mt, and percent.ribo. (B) Scatter plots showing the correlation between filtered data nCount_RNA and nFeature_RNA, percent. mt, percent.ribo. (C) Variance analysis to select highly variable genes in the samples (red dots represent highly variable genes, black dots represent invariant genes). (D) Fragmentation maps for principal component screening. (E) Principal component analysis and Harmonization analysis.

Full size image

Annotation of cell types, quantification of metabolic reprogramming scores and differential analysis between high and low scores

After applying Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction, 13 distinct subclusters were obtained (Fig. 2A). Subsequent annotation assigned the 13 subclusters to 9 distinct cell types: Monocyte, CD8 + T cell, Hematopoietic stem cell (HSC), CD4 + T cell, B cell, NK cell, Erythrocyte, Dendritic cell, and Neuroblastoma (NB) cell (Fig. 2B). Bubble plots illustrating classical markers of the nine cell types (Fig. 2C) and bar charts depicting the cell type proportions between disease and control groups were subsequently produced. In comparison with the control group, the disease group showed elevated proportions of CD8 + T cells, CD4 + T cells, and NK cells, indicating a key role of cellular immunity in the body’s anti-tumor defense (Fig. 2D).

Fig. 2
figure 2

Annotation of cellular subtypes of single-cell samples. (A) Distribution of samples within the cell clusters. (B) Annotation assigned the 13 cell clusters as 9 cell types. (C) Bubble chart of classic markers of 9 cell types. (D) Histogram of cell proportions corresponding to groups. (E) Employed AUCell to score metabolic reprogramming-related genes’ activity at the single-cell level. Metabolic reprogramming scores of Neuroblastoma cells were significantly elevated in the disease group compared to the control group. (F) A total of 136 differentially expressed genes. Ten genes with upregulated and ten with downregulated expression were displayed.

Full size image

We retrieved metabolic reprogramming-related genes from the literature25 and employed AUCell to score their activity at the single-cell level. Within the disease group, NB cells exhibited significantly higher metabolic reprogramming scores than other cell subtypes, and were thus selected as the key cell type for further analysis (Fig. 2E). Differential expression analysis was performed within the NB cell population, by dividing cells into high and low metabolic score groups, to explore transcriptional differences related to metabolic states. A total of 136 differentially expressed genes were identified based on the criteria: p_val < 0.05 and |log₂(FC)| > 0.585 (Fig. 2F, Supplementary Table 1).

Ligand-receptor interaction and transcriptional regulatory analysis

Ligand-receptor interactions were analyzed using the CellChat package based on features from the single-cell expression profiles of the disease and control groups. A network of complex interactions was identified among the various cell subtypes (Fig. 3A). Statistical analysis revealed that Monocytes exhibited more extensive potential interactions with other cell types (Fig. 3B). Furthermore, ligand and receptor molecules mediating the interactions between NB cells (with high or low metabolic reprogramming scores) and other cell types were also characterized. The results indicated that “Midkine (MDK) - Nucleolin (NCL)” serves as an important ligand-receptor pair in the intercellular communication across groups (Fig. 3C).

Fig. 3
figure 3

Ligand-receptor interaction and transcriptional regulatory analysis. (A) Overview of intercellular communication across different cell types. Lines indicate interactions between cell types based on the expression of corresponding receptors. The thickness of each line is proportional to the number of ligands expressed by the interacting (ligand-producing) cell type, reflecting the strength or frequency of potential communication. (B) Monocytes exhibited more extensive potential interactions with other cell types. (C) Ligand-receptor pairs mediating the interactions between high- and low-metabolic-score Neuroblastoma cells and other cell populations. (D) A heatmap displaying the regulon activity scores (RAS) for cells with high and low metabolic reprogramming scores was constructed. Rows represent different regulons, columns represent individual cells, and redder colors indicate higher RAS values, reflecting greater activity of the regulon in that group. (E) Scatter plots to visualize the transcription factor specificity rankings for cells with high versus low metabolic reprogramming scores. In these plots, the x-axis shows the rank while the y-axis indicates the regulon specificity scores (RSS), with higher RSS values suggesting stronger association of a regulon with cell type specificity.

Full size image

The NB cell subcluster was chosen for SCENIC analysis, through which all regulatory units (regulons) in this subpopulation were identified (Supplementary File 2). A heatmap displaying the regulon activity scores (RAS) for cells with high and low metabolic reprogramming scores was constructed. Rows represent different regulons, columns represent individual cells, and redder colors indicate higher RAS values, reflecting greater activity of the regulon in that group (Fig. 3D, Supplementary File 4). Additionally, we employed scatter plots to visualize the transcription factor specificity rankings for cells with high versus low metabolic reprogramming scores. In these plots, the x-axis shows the rank while the y-axis indicates the regulon specificity scores (RSS), with higher RSS values suggesting stronger association of a regulon with cell type specificity. Based on the RSS scores, we identified JUND, JUNB, FOS, E2F1, and KLF16 as key transcription factors associated with metabolic reprogramming in NB (Fig. 3E, Supplementary File 5).

Key gene selection through random survival forest analysis

A random survival forest analysis was conducted on 136 metabolism-related differentially expressed genes. Genes with relative importance scores greater than 0.4 were considered final markers, and their ranking by importance was visualized (Fig. 4A, Supplementary File 6). Subsequent survival analysis was performed on these highly ranked genes. The results showed that five genes—Mitochondrial ribosomal protein L21 (MRPL21), NHP2 ribonucleoprotein (NHP2), Ribosomal Protein L13 (RPL13), Ribosomal Protein L18A (RPL18A), and Ribosomal Protein L27a (RPL27A)—were significantly associated with survival (p < 0.05) (Fig. 4B-F). These five genes were selected as core targets for further investigation.

Fig. 4
figure 4

Identification of key metabolic reprogramming-related genes. (A) The random survival forest analysis of metabolic reprogramming-related genes in neuroblastoma (NB). The variable relative importance of genes > 0.4 were identified as key genes in NB. (B-F) Kaplan-Meier survival analysis of NB patients with high- and low-expression of MRPL21, NHP2, RPL13, RPL18A, and RPL27A.

Full size image

Correlation between metabolic gene expression and immune infiltration

The tumor microenvironment (TME), composed of fibroblasts, immune cells, extracellular matrix components, cytokines, and growth factors, plays a pivotal role in tumor progression, prognosis, and therapeutic response. To explore the relationship between metabolic gene expression and immune infiltration in the bone marrow, we stratified samples based on the expression levels of five key metabolic genes (MRPL21, NHP2, RPL13, RPL18A, and RPL27A) and systematically analyzed immune cell abundance.

Compared to the low MRPL21 expression group, samples in the high MRPL21 expression group showed significantly higher levels of Macrophages M0 and Plasma cells, and significantly lower levels of resting Mast cells, Monocytes, and resting CD4 memory T cells (Fig. 5A-B).

Fig. 5
figure 5

Immune-related analysis. (A, B) The difference in immune cell infiltration between the high and low expression groups of MRPL21 in NB. (C) Correlation analysis of MRPL21 with various types of immune cells. (D, E) The difference in immune cell infiltration between the high and low expression groups of NHP2 in NB. (F) Correlation analysis of NHP2 with various types of immune cells. (G, H) The difference in immune cell infiltration between the high and low expression groups of RPL13 in NB. (I) Correlation analysis of RPL13 with various types of immune cells. (J, K) The difference in immune cell infiltration between the high and low expression groups of RPL18A in NB. (L) Correlation analysis of RPL18A with various types of immune cells. (M, N) The difference in immune cell infiltration between the high and low expression groups of RPL27A in NB. (O) Correlation analysis of RPL27A with various types of immune cells.

Full size image

In comparison with the low NHP2 expression group, the high NHP2 expression group exhibited markedly elevated levels of memory B cells, Macrophages M0, and Plasma cells, and a significantly reduced level of Monocytes (Fig. 5D-E).

Compared to the low RPL13 expression group, the high RPL13 expression group exhibited significantly higher levels of memory B cells, Macrophages M0, and activated Mast cells, while Macrophages M1, resting CD4 memory T cells, and gamma delta T cells were significantly decreased (Fig. 5G-H).

The high RPL18A expression group, relative to the low expression group, demonstrated elevated levels of memory B cells, Macrophages M0, and resting NK cells, along with decreased levels of Macrophages M1, CD4 memory T cells (activated and resting), and gamma delta T cells (Fig. 5J-K).

In comparison with the low RPL27A expression group, samples with high RPL27A expression exhibited elevated levels of memory B cells and Neutrophils, while levels of Macrophages M1, Monocytes, and resting CD4 memory T cells were markedly reduced (Fig. 5M-N).

We conducted further analysis to investigate the associations between key genes and immune cell types. The results showed that MRPL21 was significantly positively correlated with Plasma cells, and negatively correlated with resting CD4 memory T cells and Monocytes (Fig. 5C).

NHP2 showed significant positive correlations with Plasma cells and Macrophages M0, while showing negative correlations with resting CD4 memory T cells and Monocytes (Fig. 5F).

RPL13 was significantly positively correlated with memory B cells, Plasma cells, Macrophages M0, activated Mast cells, and Eosinophils, and negatively correlated with resting and activated CD4 memory T cells, Macrophages M1, and resting Mast cells (Fig. 5I).

RPL18A was positively related to Plasma cells, Macrophages M0, and activated Mast cells, and negatively related to CD4 memory T cells (both resting and activated), Macrophages M1, and resting Mast cells (Fig. 5L).

RPL27A showed a significant positive correlation with Macrophages M0, and significant negative correlations with resting and activated CD4 memory T cells, Monocytes, Macrophages M1, and resting Mast cells (Fig. 5O).

Collectively, these results demonstrate that metabolic reprogramming in NB cells may actively modulate the immune landscape, suggesting a coordinated tumor–immune interaction that favors immune evasion.

GSEA analysis.

Next, we investigated the specific signaling pathways associated with the key genes to explore the potential molecular mechanisms by which they influence NB progression. According to GSEA results, RPL18A was mainly enriched in the Ribosome, DNA replication, and Mismatch repair signaling pathways (Fig. 6A, Supplementary File 7). NHP2 showed enrichment in Base excision repair, Homologous recombination, and RNA polymerase pathways (Fig. 6B, Supplementary File 8).

Fig. 6
figure 6

Gene set enrichment analysis. (A-E) Signaling pathways involved in different key genes.

Full size image

RPL27A was mainly associated with enrichment in Spliceosome, Nucleotide excision repair, and RNA polymerase pathways (Fig. 6C, Supplementary File 9).

RPL13 was found to be enriched in the PI3K-Akt signaling pathway, p53 signaling pathway, and RNA degradation pathway (Fig. 6D, Supplementary File 10).

MRPL21 showed significant enrichment in Oxidative phosphorylation, mRNA surveillance, and Basal transcription factor pathways (Fig. 6E, Supplementary File 11).

GSVA analysis

The GSVA results revealed that RPL18A was primarily enriched in MYC_TARGETS_V1 and MYC_TARGETS_V2 signaling pathways (Fig. 7A, Supplementary File 12).

Fig. 7
figure 7

Gene set variation analysis (GSVA). (A-E) GSVA pathway with significant enrichment of key genes: top indicates high expression and bottom indicates low expression based on hallmarker background set.

Full size image

NHP2 showed enrichment in pathways including DNA_REPAIR and MTORC1_SIGNALING (Fig. 7B, Supplementary File 13). Enrichment of RPL27A was observed in the MTORC1_SIGNALING and P53_PATHWAY signaling pathways (Fig. 7C, Supplementary File 14).

RPL13 was enriched in MYC_TARGETS_V2, MYC_TARGETS_V1, and other related pathways (Fig. 7D, Supplementary File 15).

MRPL21 mainly enriched in MYC_TARGETS_V2 and OXIDATIVE PHOSPHORYLATION signaling pathways (Fig. 7E, Supplementary File 16). The results indicate that these key genes might affect the progression of NB by regulating these signaling pathways.

Expression abundance of key genes, disease-related gene co-expression network, and immune-metabolic pathway analysis

The VlnPlot and FeaturePlot functions in the SeuratR package were used to examine the expression of key genes in single cells. The results showed that all five genes were predominantly expressed in the neuroblastoma cell cluster, with particularly high enrichment observed for RPL13, RPL18A, and MRPL21 (Fig. 8A). The violin plots further confirmed this trend, revealing a marked elevation in expression levels of MRPL21 and RPL13 specifically in the neuroblastoma cell population (Fig. 8B). These findings suggest that these genes may play essential roles in the metabolic regulation of neuroblastoma cells and hold potential as prognostic biomarkers or therapeutic targets.

Fig. 8
figure 8

Expression abundance of key genes, disease-related gene co-expression network, and immune-metabolic pathway analysis (A, B) The VlnPlot and FeaturePlot functions were used to examine the expression of key genes in single cells in NB. (C) The differences in key gene expression across immune, metabolic, and proliferation-related pathways were visualized using bubble plots.

Full size image

We then retrieved NB-related regulatory genes from the GeneCards database (https://www.genecards.org/). High-ranking genes based on Relevance score were selected to analyze the co-expression network between key genes and NB regulators TP53, NRAS, and NKIF1B. NHP2 and MRPL21 exhibited strong positive correlations with TP53, NRAS, and NKIF1B (Supplementary Figs. 1–5). In addition, AUCell was used to quantify immune- and metabolism-related pathway gene activity at the single-cell level. The differences in key gene expression across immune, metabolic, and proliferation-related pathways were visualized using bubble plots. The results showed high activity of the 5 key genes in the MYC_TARGETS_V1 and MYC_TARGETS_V2 pathways. NHP2 and MRPL21 were particularly active in the Oxidative phosphorylation pathway (Fig. 8C, Supplementary File 17).

Expression analysis of 5 key genes in NB and normal astroglial cell lines

We performed RT-qPCR to assess the expression levels of five key genes—MRPL21, NHP2, RPL13, RPL18A, and RPL27A—in human NB cell lines (SH-SY5Y, SK-N-BE2, and SK-N-SH) and a normal human astroglial cell line (SVGp12). MRPL21 expression was significantly elevated in SH-SY5Y and SK-N-BE2 cells compared to SVGp12 (P < 0.0001). In contrast, MRPL21 levels were low in SVGp12 cells, suggesting a potential association with altered metabolic activity or mitochondrial dysfunction in tumor cells. Among the other genes, NHP2, RPL13, RPL18A, and RPL27A were upregulated only in SH-SY5Y cells, but either downregulated or unchanged in SK-N-BE2 and SK-N-SH cells relative to SVGp12 (Fig. 9). Collectively, the consistently high expression of MRPL21 in NB cell lines implies its potential involvement in tumor initiation and progression, highlighting it as a promising candidate for further functional investigation.

Fig. 9
figure 9

Comparative analysis of MRPL21, NHP2, RPL13, RPL18A, and RPL27A mRNA expression in neuroblastoma and normal astroglial cell lines.

Full size image

MRPL21 knockdown suppresses proliferation, migration, and invasion of NB cell lines

To assess the role of MRPL21 in the biological behavior of NB cells, SH-SY5Y and SK-N-BE2 cells were transfected with si-MRPL21 (#1 and #2) or negative control siRNA (si-NC). RT-qPCR and Western blot analyses confirmed that MRPL21 expression was significantly downregulated at both the mRNA (Fig. 10A) and protein levels (Fig. 10B, Supplementary File 3) in the si-MRPL21-1 and si-MRPL21-2 groups compared to the si-NC group.

Fig. 10
figure 10

MRPL21 knockdown inhibited cell viability, migration, and invasion in neuroblastoma cells. RT-qPCR (A) and Western blot (B) analyses confirmed the successful knockdown of MRPL21 by siRNA in SH-SY5Y and SK-N-BE2 cells. NC indicates the negative control group. The effects of MRPL21 knockdown on cell viability, migration, and invasion were evaluated using CCK-8 (C, D), wound healing (E, F), and transwell invasion (G) assays, respectively. Data are presented as mean ± standard deviation (SD). *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Full size image

CCK-8 assays demonstrated that MRPL21 knockdown significantly suppressed the proliferation of SH-SY5Y (Fig. 10C) and SK-N-BE2 cells (Fig. 10D) at 48 and 72 h post-transfection, indicating that MRPL21 promotes NB cell proliferation.

Wound healing assays further showed that MRPL21 silencing markedly impaired cell migration in both SH-SY5Y (Fig. 10E) and SK-N-BE2 cells (Fig. 10F), with significantly reduced migration rates observed at 24 and 48 h compared to the control group. These results suggest that MRPL21 is essential for maintaining the migratory capacity of NB cells.

In addition, transwell invasion assays revealed that MRPL21 knockdown significantly reduced the number of invasive cells in both SH-SY5Y and SK-N-BE2 lines compared to the si-NC group (Fig. 10G), supporting the notion that MRPL21 enhances the invasive potential of NB cells.

MRPL21 knockdown inhibited oxidative phosphorylation and induces mitochondrial dysfunction in NB cells

To evaluate the impact of MRPL21 on mitochondrial respiration, the OCR was measured in SH-SY5Y and SK-N-BE2 cells using the Seahorse XFe96 extracellular flux analyzer after transfection with si-NC, si-MRPL21-1, and si-MRPL21-2. In SH-SY5Y cells, MRPL21 knockdown markedly reduced basal OCR, FCCP-stimulated maximal respiration, and ATP production-coupled oxygen consumption compared with the si-NC group, indicating impaired mitochondrial oxidative phosphorylation (Fig. 11A). Similarly, in SK-N-BE2 cells, both si-MRPL21-1 and si-MRPL21-2 led to a significant reduction in OCR parameters, although the overall respiratory capacity was lower than in SH-SY5Y cells (Fig. 11B). These findings suggest that MRPL21 is essential for maintaining mitochondrial respiratory function and energy metabolism in neuroblastoma cells. To further investigate the impact of MRPL21 on cellular energy metabolism, ATP levels were measured in two NB cell lines. The results revealed that ATP levels in the si-NC group were significantly higher than those in the si-MRPL21-1 and si-MRPL21-2 groups in both SH-SY5Y and SK-N-BE2 cells. These findings suggest that MRPL21 knockdown markedly impairs mitochondrial energy metabolism, leading to a reduction in ATP production (Fig. 11C).

Fig. 11
figure 11

MRPL21 knockdown inhibited oxidative phosphorylation and induces mitochondrial dysfunction in neuroblastoma cells. MRPL21 knockdown reduces mitochondrial oxygen consumption in SH-SY5Y (A) and SK-N-BE2 (B) cells. The oxygen consumption rate (OCR) was measured using the Seahorse XFe96 analyzer. Oligomycin, FCCP, and rotenone/antimycin A (ROT/AA) were sequentially added to assess mitochondrial function. (C) MRPL21 knockdown reduces ATP production in SH-SY5Y and SK-N-BE2 cells. MRPL21 knockdown impairs mitochondrial membrane potential in SH-SY5Y (D) and SK-N-BE2 (E) cells.

Full size image

To investigate the effect of MRPL21 on mitochondrial function in NB cells, JC-1 staining combined with flow cytometry was performed to assess changes in mitochondrial membrane potential (ΔΨm). In both NB cell lines (SH-SY5Y and SK-N-BE2), the JC-1 aggregate/monomer fluorescence ratio—an indicator of ΔΨm—was significantly reduced in the si-MRPL21-1 and si-MRPL21-2 groups compared with the si-NC group under basal conditions, indicating that MRPL21 knockdown impairs mitochondrial membrane potential. Upon treatment with oligomycin and antimycin A (OA), the membrane potential further declined across all groups, suggesting that MRPL21 deficiency enhances cellular susceptibility to mitochondrial stress (Fig. 11D, E). Collectively, these findings demonstrate that MRPL21 knockdown markedly decreases mitochondrial membrane potential under both basal and OA-induced conditions, underscoring its critical role in maintaining mitochondrial homeostasis in NB cells.

Discussion

Cell metabolism is a basic process vital for maintaining life, involving energy generation, biosynthesis, and redox balance. Under normal physiological conditions, cellular metabolism heavily depends on external nutrient supply and endogenous signal regulation. The primary task of cellular metabolism is to convert nutrients (such as glucose) into energy (such as ATP) to support normal cellular functions. This is accomplished mainly through the tricarboxylic acid (TCA) cycle, coupled with the electron transport chain and oxidative phosphorylation (OXPHOS). Proliferating cells demand not only energy but also the biosynthesis of macromolecules like nucleic acids, proteins, and lipids to facilitate growth and the production of daughter cells26. To adapt to the demands of sustained proliferation, tumor cells frequently activate abnormal metabolic pathways to optimize the use of energy and synthetic substrates, a phenomenon known as “metabolic reprogramming”. This process serves not only as a crucial basis for tumor cell phenotypic transformation but also offers novel intervention targets for cancer therapy26.

This study sought to investigate the functions and mechanisms of critical metabolic genes in NB. Initially, nine NB-specific cellular subclusters were identified using single-cell transcriptome data. CellChat analysis uncovered intricate ligand-receptor communications between “Neuroblastoma cells” and other cellular subgroups. Among them, the “MDK-NCL” axis was recognized as a significant ligand-receptor pair, with MDK being a growth-promoting factor abundantly expressed during embryogenesis27. NCL, a primary receptor for MDK, is a multifunctional ribonucleoprotein localized in the nucleus, cytoplasm, and cell membrane. Through binding to NCL on the tumor cell membrane, MDK activates various signaling cascades including AKT, ERK, and STAT3, enhancing tumor cell proliferation, resistance to apoptosis, and invasiveness28. Notably, this axis had also been reported in the original publication from which the dataset was derived9. Our findings therefore represent an independent computational confirmation, reinforcing the robustness and potential biological relevance of this interaction under a different analytical framework. However, we acknowledge that no further experimental validation of the MDK–NCL axis was conducted in this study. Since our central aim was to elucidate the regulatory mechanisms of metabolic reprogramming in NB and identify metabolism-related therapeutic targets, the MDK–NCL axis was not prioritized for downstream functional investigation. Future studies will be necessary to determine whether this axis contributes to metabolic remodeling or immune modulation in bone marrow–infiltrating NB cells.

We further conducted SCENIC transcriptional regulation analysis on the “Neuroblastoma cell” cluster, identifying JUND, JUNB, FOS, E2F1, and KLF16 as critical transcription factors linked to metabolic reprogramming in NB. JUND, JUNB, and FOS are members of the AP-1 (activator protein-1) transcription factor complex, acting as major responders to stress, cytokine, and growth factor signaling29,30. Studies have demonstrated that the JUN/FOS complex plays a regulatory role in neurodevelopment and neural crest cell fate decisions31, implying its possible participation in the early phases of NB pathogenesis. E2F1 is a classical regulator of the cell cycle, primarily promoting the G1/S phase transition and controlling tumor cell proliferation32. Research has revealed that in NB, E2F1 enhances the transcription of BMCC1 (BCH motif-containing molecule at the carboxyl terminal region 1), with elevated BMCC1 levels promoting apoptosis and suppressing tumor progression32. KLF16 (Krüppel-like factor 16), belonging to the KLF family, has been scarcely studied in NB; however, it has been reported to suppress human glioma cell proliferation and tumorigenic potential via targeting mitochondrial transcription factor A (TFAM)33.

Following the retrieval of metabolic reprogramming-related genes from the literature, AUCell analysis was performed to quantify gene scores at the single-cell level. The “Neuroblastoma cell” cluster exhibited significantly elevated metabolic reprogramming scores in the disease group, leading to its designation as the key subpopulation for subsequent analyses. Through the integration of differential expression analysis, random survival forest modeling, and survival analysis, five core genes—MRPL21, NHP2, RPL13, RPL18A, and RPL27A—were identified as significantly associated with NB prognosis. Notably, these genes were highly expressed within the “Neuroblastoma cell” cluster, implicating their critical roles in metabolic regulation and the maintenance of tumor cell functionality.

MRPL21 encodes a ribosomal protein integral to mitochondrial function, forming part of the mitochondrial large subunit (39 S)34. Accumulating evidence has highlighted a strong association between mitochondrial dysfunction and the initiation and progression of various malignancies. Dysregulation of MRPL21 expression may impair mitochondrial protein synthesis and oxidative phosphorylation, thereby perturbing cellular energy metabolism, apoptosis, and proliferation34. Emerging evidence indicates that elevated MRPL21 expression in lung adenocarcinoma correlates with unfavorable prognosis and may facilitate tumor progression through regulation of the cell cycle and modulation of immune cell infiltration34. Moreover, MRPL21 has been implicated in promoting the proliferation of hepatocellular carcinoma cells by inducing TP53 mutations and concurrently enhancing resistance to apoptosis35.

NHP2 encodes a ribonucleoprotein component of the H/ACA small nucleolar ribonucleoprotein (snoRNP) complex, which is indispensable for ribosome biogenesis and the maintenance of telomere integrity36. In a hepatocellular carcinoma model induced by hepatitis B virus X protein (HBX), silencing of NHP2 impaired telomerase activity by downregulating the expression of telomerase reverse transcriptase (TERT), leading to suppressed tumor cell proliferation and enhanced apoptosis36. These observations highlight the pivotal role of NHP2 in preserving telomere function. Dysregulation or mutation of NHP2 may therefore contribute to tumorigenesis and cancer progression. A deeper understanding of NHP2’s biological functions and oncogenic mechanisms may provide novel insights into cancer pathogenesis and offer potential targets for diagnostic and therapeutic intervention.

RPL13, a member of the ribosomal protein L13 family, encodes a structural component of the 60 S large ribosomal subunit. It plays a fundamental role in maintaining the architecture and function of ribosomes, the cellular machinery responsible for protein synthesis37. Highly conserved across eukaryotic species, RPL13 is ubiquitously and abundantly expressed, reflecting its classification as a housekeeping gene essential for sustaining basic cellular functions. The critical involvement of RPL13 in core biological processes underscores its potential importance not only in normal cell physiology but also in pathological conditions where translational control is dysregulated37.

RPL27A, a member of the L27AE ribosomal protein family, encodes a structural component of the 60 S large ribosomal subunit essential for ribosome assembly and protein synthesis38. Elevated RPL27A expression has been observed in triple-negative breast cancer (TNBC) tissues, and its knockdown in MDA-MB-231 cells significantly impaired cancer cell migration and invasion, highlighting its role in tumor progression and metastasis39. In this study, high expression levels of RPL13 and RPL27A were associated with poor prognosis in NB. Enrichment analysis suggested that these genes may drive NB progression through the PI3K-Akt, p53, MTORC1, and MYC_TARGETS signaling pathways.

RPL18A encodes the 60 S ribosomal large subunit protein L18a, which is essential for protein synthesis. Previous studies have reported that elevated RPL18A expression is associated with favorable prognosis in glioma40. However, in the present study, high RPL18A expression correlated with poor prognosis in NB patients. Enrichment analysis suggested that RPL18A may contribute to NB progression, potentially through the MYC_TARGETS signaling pathway.

Mounting evidence indicates that tumor metabolic reprogramming plays a pivotal role in reshaping the immune microenvironment. Enhanced mitochondrial biogenesis, ribosomal activity, and OXPHOS in tumor cells can lead to increased production of reactive oxygen species (ROS), nutrient depletion, and lactate accumulation, all of which contribute to the suppression of immune cell infiltration and function41. In this study, we found that high expression of five mitochondrial ribosomal genes—MRPL21, NHP2, RPL13, RPL18A, and RPL27A—was significantly associated with reduced infiltration of monocytes, M1-polarized macrophages, and CD8⁺ T cells in the bone marrow microenvironment. These findings suggest a potential link between mitochondrial-driven metabolic rewiring and the establishment of an immunosuppressive niche in metastatic NB.

Although our results are based on correlation, they are supported by existing literature demonstrating the role of tumor metabolism in modulating antitumor immunity. For instance, oxidative metabolic phenotypes have been shown to impede dendritic cell activation and limit T cell proliferation within the tumor microenvironment42,43. Further experimental validation, such as gene perturbation studies coupled with immunophenotyping and metabolic flux assays, will be essential to fully elucidate the causal relationships between metabolic gene expression and immune landscape remodeling in NB.

To strengthen the biological relevance of our findings, we performed in vitro functional validation of MRPL21, one of the five core metabolic reprogramming genes identified through single-cell analysis. Notably, the two NB cell lines used in these experiments—SK-N-BE2 and SH-SY5Y—were both derived from bone marrow metastatic lesions of NB patients, which aligns closely with the origin of our single-cell transcriptomic data. This concordance provides a biologically relevant model system for experimental validation.

Our results showed that MRPL21 knockdown led to significant suppression of NB cell proliferation and migration, as well as impaired oxidative phosphorylation and mitochondrial function, consistent with the gene’s predicted role in NB metabolic reprogramming. These findings offer direct experimental support for the computational inference that MRPL21 contributes to NB malignancy in a metabolism-dependent manner, thereby bridging the gap between transcriptomic data and functional outcomes.

Conclusion

In this study, we conducted a comprehensive analysis of the metabolic landscape of NB cells from bone marrow metastatic samples using single-cell transcriptomic profiling. We identified five key metabolic genes—MRPL21, NHP2, RPL13, RPL18A, and RPL27A—that are significantly associated with prognosis in NB patients. These genes were found to not only be linked to metabolic dysregulation but also to correlate with the development of an immunosuppressive microenvironment, suggesting a complex interplay between metabolic reprogramming and immune evasion in NB progression.

Our functional validation experiments confirmed that MRPL21 plays a central role in NB cell survival, proliferation, migration, and invasion. Specifically, MRPL21 silencing impaired mitochondrial function and oxidative phosphorylation, indicating that it drives NB progression through its regulation of mitochondrial bioenergetics and metabolic reprogramming. Importantly, the validation experiments were conducted in SK-N-BE2 and SH-SY5Y cell lines, which are derived from bone marrow metastatic lesions in NB patients, providing biological relevance to our computational findings.

Collectively, our study highlights the importance of metabolic reprogramming in bone marrow–derived metastatic NB and reveals MRPL21 as a potential therapeutic target. Future investigations into the downstream signaling pathways regulated by MRPL21, as well as combined strategies targeting both tumor metabolism and immune modulation, may provide new therapeutic avenues to improve the clinical outcomes of NB patients.