NPC2 deficiency as a potential target drives malignancy and chemo-resistance in DLBCL Authors: Yining Yuan, Qiqi Wu, Xinyu Li, Peipei Xu Published: 2026 ABSTRACT AbstractBackgroundDiffuse large B-cell lymphoma (DLBCL) remains challenging to treat, with a considerable proportion of patients exhibiting chemo-resistance and poor prognosis. A deeper understanding of the molecular mechanisms underlying lymphoma progression and therapy failure is urgently needed.MethodsUsing bioinformatic analysis of single-cell RNA-seq datasets, we identified NPC2 as a potential target associated with poor outcomes. Its function was explored in SU-DHL-4 and SU-DHL-2 cells by shRNA-mediated knockdown, followed by in vitro and in vivo experiments including qPCR, functional assays, flow cytometry, and xenograft models in NOD‑SCID mice.ResultsLow NPC2 expression was correlated with worse prognosis in DLBCL patients. NPC2 depletion did not affect cell proliferation, cell cycle, or apoptosis, but induced mitochondrial dysfunction, ROS accumulation, and enhanced migration and invasion. Importantly, NPC2 knockdown conferred resistance to vincristine both in vitro and in vivo, and promoted distant tissue damage in mouse models.ConclusionsOur study identifies NPC2 as a key regulator of DLBCL aggressiveness and chemo-resistance by modulating mitochondrial function and cell motility. NPC2 may serve as a novel potential therapeutic target for improving DLBCL treatment.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13062-026-00833-0.Keywords: Diffuse large B-cell lymphoma (DLBCL), NPC2, Single-cell RNA sequencing, Malignancy, Chemo-resistance, Vincristine Introduction Diffuse large B-cell lymphoma (DLBCL) is the most common subtype of non-Hodgkin lymphoma (NHL), accounting for approximately 30–40% of newly diagnosed cases worldwide [1]. Despite substantial progress in the understanding and management of lymphoma over the past decades, DLBCL remains a major clinical challenge due to its complex molecular heterogeneity and unpredictable clinical outcomes [2]. The introduction of rituximab combined with cyclophosphamide, doxorubicin, vincristine (Vinc), and prednisone (R-CHOP) has significantly improved the prognosis of patients with DLBCL and has become the standard first-line therapy [3, 4]. However, approximately 30–40% of patients eventually develop refractory or relapsed disease following initial treatment [5]. Patients with relapsed or refractory DLBCL often have limited therapeutic options and poor survival outcomes [6], highlighting the urgent need to identify novel molecular determinants that contribute to disease progression and treatment resistance. One of the key factors underlying the clinical heterogeneity of DLBCL is its molecular diversity. Gene expression profiling studies have led to the identification of two major molecular subtypes based on the cell-of-origin (COO) classification: germinal center B-cell-like (GCB) and activated B-cell-like (ABC) [7]. These subtypes exhibit distinct genetic alterations, signaling pathways, and clinical prognoses [8]. For example, ABC-DLBCL is frequently associated with chronic active B-cell receptor signaling and constitutive activation of the NF-κB pathway [9], whereas GCB-DLBCL often harbors genetic abnormalities affecting epigenetic regulators [10]. Subsequent genomic studies have further refined the molecular classification of DLBCL and revealed multiple genetically defined subtypes characterized by recurrent mutations in genes involved in immune signaling, chromatin modification, and B-cell differentiation [11–13]. Nevertheless, these classification systems only partially explain the extensive transcriptional diversity and variable clinical outcomes observed among patients. Recent advances in single-cell RNA sequencing (scRNA-seq) technologies have enabled high-resolution profiling of gene expression at the individual cell level, providing unprecedented opportunities to dissect tumor heterogeneity and cellular dynamics [14]. Unlike bulk transcriptomic approaches, which measure averaged gene expression across heterogeneous cell populations, scRNA-seq allows the identification of distinct cell states, lineage relationships, and transcriptional programs within tumors [15]. Using single-cell transcriptomic analysis, recent studies have revealed multiple malignant B-cell states and complex tumor ecosystems in DLBCL, suggesting that malignant cells may exist in dynamic transcriptional states that are associated with different differentiation stages and prognostic outcomes [16]. These findings indicate that transcriptional plasticity and cellular state transitions may play critical roles in lymphoma progression. Despite these advances, the molecular mechanisms that drive malignant progression and therapy resistance in DLBCL remain incompletely understood [17, 18]. In particular, the identification of genes that are specifically altered during the transition from early B-cell states to malignant tumor cell populations may provide valuable insights into lymphoma pathogenesis and reveal novel biomarkers for prognosis and therapeutic targeting [19]. Integrative analysis of multiple single-cell datasets offers a powerful strategy to systematically identify such candidate genes and uncover transcriptional programs associated with disease aggressiveness. Among genes involved in cellular homeostasis, Niemann-Pick disease type C2 protein (NPC2) is a soluble lysosomal protein that plays a critical role in intracellular cholesterol trafficking [20]. NPC2 functions together with NPC1 to mediate cholesterol transport from lysosomes to other cellular compartments, thereby maintaining lipid homeostasis and membrane integrity [20]. Dysregulation of cholesterol metabolism and lysosomal function has been increasingly implicated in cancer progression, tumor metabolism, and drug resistance [21]. However, the potential role of NPC2 in lymphoma biology and therapeutic response has not been well characterized. In the present study, we aimed to comprehensively characterize the gene expression landscape associated with adverse clinical outcomes in DLBCL by integrating multiple previously published single-cell RNA sequencing datasets. Through a large-scale integrative analysis of these datasets, we delineated two distinct DLBCL B-cell populations, designated as Early and Malignant. We then systematically investigated transcriptional differences between Early B-cell populations and Malignant DLBCL clones. Using advanced bioinformatics approaches including trajectory inference, differential expression analysis, and survival analysis, we identified candidate genes whose expression is significantly downregulated in malignant DLBCL cells and associated with poor patient prognosis. Among these candidates, NPC2 emerged as a potential target with significant clinical relevance. To further investigate the biological function of this gene, we performed a series of experimental validations. In vitro experiments using DLBCL cell line SU-DHL-4 was conducted to evaluate the effects of NPC2 depletion on cellular proliferation, migration, mitochondrial function, and drug response. In addition, in vivo xenograft models were established to examine the role of NPC2 in tumor progression and chemotherapy resistance. Collectively, our study integrates single-cell transcriptomic analysis with functional validation experiments to identify and characterize a novel gene NPC2 associated with malignant progression and chemoresistance in DLBCL. These findings provide new insights into the molecular mechanisms underlying DLBCL pathogenesis and may contribute to improved diagnostic, prognostic, and therapeutic strategies for this aggressive lymphoma. Method Single-cell data integration Data Acquisition and Processing To investigate the malignant transformation of B-cells in Diffuse Large B-cell Lymphoma (DLBCL), single-cell RNA sequencing (scRNA-seq) data were integrated from four independent cohorts: (1) Nodal B-cell lymphoma samples from Roider et al. [22]; (2) A comprehensive DLBCL cell state atlas from the Steen et al. [16]; (3) Primary central nervous system (PCNS) DLBCL samples from Liu et al. [23]; and (4) Tumor microenvironment profiling data from Zhang et al. [24]. Single-cell transcriptomic analysis Data Integration and B-cell Subsetting Standard QC was performed using Seurat (v5.0), retaining cells with number of detected genes (nFeature_RNA) between 200 and 6,000 and mitochondrial gene percentage (percent.mt) < 10%. Following log-normalization and identification of the top 2,000 HVGs, batch effects were mitigated using the Harmony algorithm. B-cell lineages were isolated via canonical markers (CD19, MS4A1) and subjected to iterative sub-clustering to resolve high-resolution cell states. Dimensionality Reduction and Proliferation Scoring PCA was performed on the integrated B-cell assay, and the top 30 PCs were utilized for UMAP construction. Proliferative potential was evaluated via CellCycleScoring, assigning cells to G1, S, or G2/M phases based on canonical marker expression. Trajectory Inference and Differential Expression Developmental trajectories were reconstructed using Monocle 2 (v2.22.0) with the DDRTree algorithm. The trajectory root was defined at the Pre-B-cell stage to model transition toward malignant-tendency clusters. Functional Enrichment Biological implications of DEGs were elucidated through GO, KEGG, and Reactome pathway analyses using clusterProfiler and ReactomePA. Significance was determined by a Benjamini-Hochberg adjusted P < 0.05. Kaplan-Meier curve (KM) To evaluate the prognostic significance of the differentially expressed genes, survival analysis was performed using the Gene Expression Profiling Interactive Analysis version 3 (GEPIA3) online platform (http://gepia3.cancer-pku.cn/) as described previously [25]. The Kaplan-Meier survival curves were generated based on gene expression levels. Patients of DLBC Tumor were stratified into high-expression and low-expression groups using the median expression value as the cutoff. Overall Survival (OS) was calculated, and log-rank test was used to determine the statistical significance of the differences between the two groups. P-value < 0.05 was considered statistically significant. Analysis of NPC2 expression profiles The tissue-specific expression profile of the NPC2 gene was retrieved from the Human Protein Atlas (HPA) database (https://www.proteinatlas.org/). The mRNA expression levels across different human tissues were obtained, and the data were presented as normalized Transcripts Per Million (nTPM). A bar plot was generated to visualize the distribution of NPC2 expression across various tissues using the nTPM values. RNA-seq data analysis Stratification and Differential Expression Patients from GSE datasets were dichotomized into NPC2-high and -low groups based on median expression. Differential expression analysis (DEA) was performed using the limma (v3.58.0) package by fitting linear models with empirical Bayes moderation. Significant differentially expressed genes (DEGs) were defined by |log2FC| > 0.58 and P-value < 0.05. Signature Visualization Transcriptomic signatures were visualized via volcano plots and Z-score-normalized heatmaps of the top 100 DEGs (50 up- and 50 down-regulated) using heatmap. Expression values were capped at saturation limits set at ± 2.5 to enhance visualization of core clusters. Functional Annotation Biological implications of NPC2-related signatures were elucidated through GO, KEGG, and Reactome pathway analyses using clusterProfiler (v4.10.0) and ReactomePA. Significance for all enriched pathways was determined by a Benjamini-Hochberg (BH) adjusted P-value < 0.05. Cell culture and shRNA transduction The human diffuse large B-cell lymphoma cell line SU-DHL-2 (ABC phenotype) and SU-DHL-4 (GCB phenotype) were purchased from Nanjing DeepSpace Biotech. (China). The cells were cultured in RPMI-1640 medium (Sbjbio, China) supplemented with 10% heat-inactivated fetal bovine serum (FBS; Gibco, USA) and 1% penicillin-streptomycin solution (Sbjbio, China). The cell cultures were maintained in a humidified incubator at 37 °C with an atmosphere of 5% CO₂. SU-DHL-2 and SU-DHL-4 cells grow in suspension and were passaged every 2–3 days to maintain a density between 2 × 105 to 1 × 106 cells/mL. Both cell lines were treated with vincristine (Vinc) at a concentration of 3 nM, which corresponds to the previously reported IC50 [26]. To stably knockdown NPC2 gene expression in SU-DHL-2 and SU-DHL-4 cells, specific short hairpin RNAs (shRNAs) targeting NPC2 were designed and cloned into the pLKO.1 lentiviral vector. Two distinct shRNA sequences were used in this study sh1-NPC2 (target sequence 5’-CGGTTCTGTGGATGGAGTTAT-3’) and sh2-NPC2 (target sequence 5’-TCCAATGGTATCCAGTGATTC-3’). Lentiviral particles were produced by 293T cell via co-transfection of pLKO.1 vector with gag-pol and VSVG helper plasmids. The harvested lentiviruses were then used to transduce the SU-DHL-2 and SU-DHL-4 cell lines with 4 mg/L polybrene (Sangon, China). Transduced cells were selected with 1 mg/L puromycin (Sangon, China) to generate stable knockdown cell lines. Quantitative real-time PCR (RT-qPCR) Quantitative RT-qPCR was performed as described previously with minor modifications [27]. Total RNA was extracted from transduced SU-DHL-4 cells using RNA isolater (Vazyme, China). Isolated total RNA integrity was verified byethidium bromide staining and electrophoresis. 500 ng total RNA of each sample were then reverse transcribed with the HiScript III 1st Strand cDNA Synthesis Kit (Vazyme, China) using random hexamer primers according to the manufacturer’s instructions. Real-time quantitative qPCR was performed using the LightCycler 96 instrument (Roche, Switzerland) with ChamQ SYBR qPCR Master Mix (Vazyme, China). Specific oligonucleotide primers for target gene were performed to evaluate the mRNA expression levels of NPC2. The primers used for amplification were as follows: NPC2 forward primer (NPC2-qF) 5’-CAAAGGACAGTCTTACAGCGT-3’ and NPC2 reverse primer (NPC2-qR) 5’-GGATAGGGCAGTTAATTCCACTC-3’, GAPDH was used as an internal control, amplified with forward primer (GAPDH-qF) 5’-TCATTTCCTGGTATGACAACGA-3’ and reverse primer (GAPDH-qR) 5’-GTCTTACTCCTTGGAGGCC-3’. Cell counting Cell number at determined time points were using the trypan blue exclusion assay. Briefly, an aliquot of the cell suspension was mixed with an equal volume of 0.4% trypan blue solution (Sangon, China). Subsequently, 10 µL of the stained cell suspension was loaded onto a hemocytometer. Viable cells (unstained, refractile) were counted manually under a light microscope. Each experiment was performed in triplicate. Apoptosis staining Cell apoptosis was detected by Annexin V-FITC/PI double staining kit (Beyotime, China) in accordance with the manufacturer’s instructions. Briefly, SU-DHL-2 and SU-DHL-4 cells were resuspended in binding buffer. Annexin V and PI staining solution were added and incubated in the dark at room temperature for 15 min. The apoptotic rate was detected by flow cytometry (BD FACSCalibur, USA) within 1 h. Mitochondria membrane potential analysis The mitochondrial membrane potential was measured using the JC-1 staining kit (Solarbio, China). Both SU-DHL-2 and SU-DHL-4 cells were incubated with JC-1 staining working solution at 37 °C in a 5% CO₂ incubator for 20 min. Then were detected by flow cytometry. The emission wavelengths were 530 nm and 590 nm, data were analyzed with FlowJo software. ROS analysis Intracellular reactive oxygen species (ROS) levels were detected using the fluorescent probe DCFH-DA (Beyotime, China). Both SU-DHL-2 and SU-DHL-4 cells were stained in working solution and incubated at 37 °C in a 5% CO₂ incubator for 20 min in the dark. The fluorescence intensity was detected by flow cytometry, and the relative ROS level was calculated by comparing the mean fluorescence intensity (MFI) of the treated group with that of the control group. DLBCL mouse model NOD-SCID mice (6–8 weeks old) were purchased from Nanjing Qinglongshan Animal Breeding Farm (Nanjing, China) and cultured in specific pathogen-free (SPF) grade mice facility until sacrifice. The ethical review number of this experiment was 2020 ~ Gl18. For the in vivo therapeutic experiment, mice were randomly divided into three experimental groups: the PBS control group, shNC group, and shNPC2 group. To establish the DLBCL xenograft model, mice in the shNC and shNPC2 groups were intravenously injected via the tail vein with 0.1 mL cell suspension containing 107 SU-DHL-4 cells. Notably, the PBS control group received no tumor cell inoculation and was only subjected to Vinc treatment under the same administration regimen, which served specifically as a drug toxicity control to independently evaluate the in vivo cytotoxic effects of Vinc. At 12 days after cell inoculation, all three groups of mice received intraperitoneal injection of 3 mg/kg Vinc at two-day intervals for a total of four doses. All mice were euthanized on day 21, peripheral blood samples were collected and analyzed by flow cytometry, and major organs were harvested for subsequent histological examination. Survival curve For the survival assay, Vinc treatment was initiated on day 12 post-tumor cell inoculation. Mice were monitored daily for signs of morbidity, and body weight was recorded regularly. The humane endpoint was defined as a body weight drop to 75% of the initial baseline weight, which served as a surrogate indicator for terminal disease. Survival curves were generated using the Kaplan-Meier method. Cell cycle analysis Cell cycle detection was performed using PI staining kit (Yeasen, China). Cells were seeded and treated as indicated, harvested, fixed, treated with RNase A (Yeasen, China), and stained with PI working solution in the dark, detected by BD FACSCalibur flow cytometry, and analyzed with FlowJo software to determine cell cycle distribution. Cell migration and invasion assay Cell migration and cell invasion were tested by transwell assay as described previously [28] with the following minor modifications. For both migration and invasion assay, SU-DHL-4 were seeded in an 8 μm transwell (Nest, China) at a density of 5 × 104 cells/well. For migration assay, transwells were collected after being in a complete RPMI-1640 (Sbjbio, China) medium for 24 h. For invasion assay, transwells were coated with 50 uL Matrigel (Corning, USA) before seeding cells. Cells were collected after being in a complete RPMI-1640 medium for 24 h. Samples were then stained with 0.2% crystal violet (Sangon, China) and incubated for 5 min at room temperature. The number of penetrated cells from each sample was counted by ImageJ software (NIH, USA). Histopathology Tissue sections were prepared by Nanjing DeepSpace Biotech. (China) and were stained with hematoxylin and eosin (H&E) or stained with anti-human CD45 antibody (Proteintech, China) as previously report [29]. Images were captured by a bright-field microscope (Olympus, Japan) and were evaluated by a pathologist who was blinded of the samples’ identity. Scoring criteria of different organs were described as previous reports [30–35]. Cardiac evaluation focused on myocardial injury and hemorrhage; hepatic assessment included steatosis, neutrophil accumulation, thrombosis, bacterial colonies, portal vein injury, sinusoidal injury, and central lobular vein injury; splenic scoring evaluated necrosis, neutrophil accumulation, thrombosis, and bacterial colonies; and renal criteria encompassed glomerular and tubular morphology, inflammation, and fibrosis, as well as interstitial inflammation and fibrosis. The proportion of CD45‑positive signals in tissue sections was quantified using ImageJ software as previous reports [36]. The positive area ratio was calculated in randomly selected microscopic fields for quantitative comparison. Flow cytometry Blood samples of DLBCL mice (SU-DHL-4 transplanted) were collected from the mice via the submandibular vein using a 25-gauge needle, with a collection volume of 30 µL per mouse. Then, samples are stained with antibodies against mouse and ABflo® 488 anti-human CD45 (Abclonal, China) on ice for 30 min. Cell was then fixed using lysing buffer solution 10X (BD bioscience, USA) at room temperature for 15 min. Data were acquired by LSRFortessa (BD bioscience, USA) and analyzed with FlowJo V10.62 (TreeStar, USA). Statistical analysis All statistical analyses were performed using R (v4.4.3) and GraphPad Prism (v8.0.2). For all the Student’s t-test, we note p-values with indications within histograms in each figure: (****) p < 0.0001, (***) p < 0.001, (**) p < 0.01, (*) p < 0.05, or not significant (ns) compared between samples. Results Integration of DLBCL single-cell RNA-seq datasets and identification of early and malignant cell populations To comprehensively investigate the tumor microenvironment and cellular heterogeneity in Diffuse Large B-Cell Lymphoma (DLBCL), we performed an integrative analysis of single-cell RNA sequencing (scRNA-seq) data from four independent cohorts: Roider et al. [22], Steen et al. [16], Liu et al. [23], and Zhang et al. [24] (Fig. 1a). B-cell lineages were isolated via canonical markers (CD19, MS4A1). The Uniform Manifold Approximation and Projection (UMAP) of all B-cells from the integrated datasets is shown (Fig. 1b), where cells are color-coded according to their dataset of origin. Which demonstrating successful batch correction and data merging. We subsequently classified the DLBCL B-cells based on the Cell of Origin (COO), a critical determinant for DLBCL stratification as previously reported [37]. As depicted in Fig. 1c, the clustering results categorized the samples into two major subtypes, activated B-cell (ABC) and germinal center B-cell (GCB), which is matched with previously report (Supplementary Fig. 1a, individual patient data highlights in the UMAP). Given the importance of proliferation in lymphoma biology, we evaluated the cell cycle status of the B-cells. Using the CellCycleScoring method based on the expression of marker genes, we assigned cells to specific cell cycle stages (Fig. 1d). To further validate these assignments and visualize proliferative activity, we examined the expression of genes associated with the S-G2-M phases. The expression of four representative cell cycle-related genes CDK1, CDC20, MKI67, and PCNA (Fig. 1e). The expression patterns of these markers confirmed the active cell cycle states identified within the integrated datasets. Integration of DLBCL single-cell RNA-seq datasets. (a) Schematic diagram of the data integration and analysis workflow. (b) UMAP projection of all cells from the integrated datasets. The four individual datasets are color-coded. (c) Clustering results based on the cell of origin (COO) classification; samples are categorized into activated B-cell (ABC) and germinal center B-cell (GCB) subtypes. (d) Cell-cycle phase assignment of B-cells using CellCycleScoring based on canonical marker gene expression. (e) UMAP projections colored by the z-scored expression of four representative genes associated with the S-G2-M phases of the cell cycle To resolve the heterogeneity within the integrated DLBCL scRNA-seq datasets, we first characterized the cellular states based on established classifications. The UMAP projections colored by DLBCL B-cell states as defined by Steen et al. [16] (Fig. 2a and Supplementary Fig. 1b). Concurrently, we mapped the B-cell developmental stages according to Lee et al. [38] (Fig. 2b and Supplementary Fig. 1c). These annotations provided a foundational context for distinguishing different cellular identities within the tumor microenvironment. To infer the differentiation trajectories and cellular progression, we performed pseudo-time analysis. When cells were classified by DLBCL B-cell states, the analysis revealed a distinct trajectory where specific stages such as S1, S4, and S5 were prominently featured along the path, while other stages were depicted in grey (Fig. 2c and Supplementary Fig. 1b). Similarly, pseudo-time analysis based on B-cell developmental stages highlighted the Naïve B, Immature B, and Pre-B stages, illustrating the continuum of B-cell development within the datasets (Fig. 2d and Supplementary Fig. 1c). COO of B-cell in DLBCL revealed that B-cells at different stages of differentiation are hijacked by different oncogenic mechanisms, and the signaling pathways they rely on determine their sensitivity to treatment and invasive behavior [39]. Based on these classifications and trajectory analyses, we aimed to separate early-stage B-cells from the malignant population (Fig. 2c and d right side). We defined the ‘Early’ population by intersecting specific subsets identified in both the DLBCL B-cell states and the B-cell developmental stages (Supplementary Fig. 2a). The remaining cells were categorized as the ‘Malignant’ group (Fig. 2e and Supplementary Fig. 2a). Displayed clustering results based on these newly defined Early and Malignant groups showing a clear spatial separation in the UMAP projection (Fig. 2e). Identification of Early and Malignant cell clusters from integrated scRNA-seq datasets. (a) UMAP projections colored by DLBCL B-cell states according to Steen et al. [16]. (b) UMAP projections colored by B-cell developmental stages according to Lee et al. [38]. (c-d) Pseudo-time analysis of cells classified by DLBCL B-cell states (c) or B-cell developmental stages (d). In (c), only S1, S4, and S5 stages are colored, while others are shown in grey; in (d), only Naïve B, Immature B, and Pre-B stages are colored, while others are shown in grey. See also Supplementary Fig. 1b-c for all stages. (e) Clustering results based on the newly defined Early and Malignant groups. (f) Volcano plot showing differentially expressed genes (DEGs) in the Malignant group compared to the Early group To elucidate the molecular differences between these two populations, we conducted a gene differential expression analysis. The volcano plot comparing the Malignant group to the Early group identified a distinct set of differentially expressed genes (DEGs), highlighting the transcriptional signatures that distinguish the malignant lymphoma cells from the early-stage B-cell subsets (Fig. 2f). NPC2 is a key prognostic gene associated with DLBCL survival To elucidate the molecular distinctions between the newly defined Malignant and Early groups, we performed a differential expression analysis using a pseudo-bulk approach. The top 25 upregulated and downregulated genes in the Malignant group compared to the Early group were displayed (Fig. 3a). These genes may reveal distinct transcriptional profiles associated with malignancy. We next investigated the clinical relevance of these differentially expressed genes (DEGs) by conducting Kaplan-Meier survival analyses. Strikingly, among the genes examined, NPC2 was the only gene found to be significantly associated with DLBCL patient survival (Fig. 3b and Supplementary Fig. 3). The survival analysis indicated that NPC2 expression levels correlated with patient outcomes significantly (P = 0.0234). Comprehensive survival analyses for other genes are provided in Supplementary Fig. 3. To understand the biological context of NPC2 gene, we examined its RNA expression profiles across various human organs and tissues using the Human Protein Atlas (HPA) database (Fig. 3c). NPC2 exhibited a specific expression distribution across different tissues such as blood vessel, spleen, gallbladder, kidney, bone marrow, heart, and liver. Furthermore, we visualized the expression of NPC2 within our single-cell dataset (Fig. 3e). Most of the NPC2 expressed cell are located in Early group but not Malignant, which is consistent with our previous finding (Fig. 3e). To characterize the biological processes driving the Malignant phenotype, we performed Gene Ontology (GO) enrichment analysis. The top 10 enriched GO terms in the Malignant group compared to the Early group shown that mostly related with the cell division process (Fig. 3d). The Malignant group, characterized by lower NPC2 expression, exhibits a transcriptional profile associated with higher proliferative potential. Additionally, KEGG and Reactome pathway analyses of the top enriched gene sets confirmed that cell cycle-related signaling pathways are predominantly activated in these malignant cells (Supplementary Fig. 2b and 2c). Survival association analysis identifies NPC2 as a prognostic candidate gene in DLBCL. (a) Top 25 upregulated and downregulated genes in the Malignant group compared to the Early group based on pseudo-bulk analysis. (b) Kaplan-Meier survival analysis of NPC2 expression. The P-value and Hazard ratio (HR) are indicated. See also Supplementary Fig. 3 for other genes. (c) RNA expression profiles of NPC2 across different human organs and tissues. Data retrieved from the Human Protein Atlas (HPA) database. (d) Top 10 enriched Gene Ontology (GO) terms in the Malignant group compared to the Early group. (e) UMAP projections colored by the z-scored expression of NPC2 in total cells, the Early group, and the Malignant group, respectively Validation of NPC2 association with malignancy in bulk RNA-seq cohorts To further validate the clinical significance of NPC2 and its association with malignancy in DLBCL, we utilized 2 published large cohort of bulk RNA-seq from the GSE10846 [40, 41] dataset. Patients were stratified into NPC2-high and NPC2-low groups based on the median expression level of NPC2 (Fig. 4a). This stratification allowed for a comparative analysis of the transcriptional landscapes between the two groups. We subsequently performed gene differential expression analysis to identify genes associated with NPC2 expression levels. The volcano plot illustrated the differentially expressed genes (DEGs) between the NPC2-high and NPC2-low groups (Fig. 4b). The distinct expression patterns of these DEGs are further visualized in the heatmap (Fig. 4c), confirming a clear transcriptional separation between the two patient groups. To elucidate the biological pathways associated with NPC2 expression, we conducted Gene Ontology (GO) enrichment analysis. The top 10 enriched GO terms in the NPC2-high group compared to the NPC2-low group were displayed (Fig. 4d). The results show that NPC2 gene expression was predominantly associated with chemotaxis and leukocyte migration or activation in DLBCL patients. Furthermore, to ensure the robustness of our findings, we validated these results using an independent GeneChip dataset GSE87371 [42, 43] (Supplementary Fig. 4), which yielded consistent results. Association of NPC2 with malignancy validated in a large cohort of GeneChip data. (a) Patients from the GSE10846 dataset were stratified into NPC2-high and NPC2-low groups based on median expression. (b) Volcano plot of the NPC2-high group compared to the NPC2-low group (735 genes up, 28 genes down, ≥ 1.5-fold threshold). (c) Heatmap of differentially expressed genes (DEGs) between the two groups. (d) Top 10 enriched GO terms in the NPC2-high group compared to the NPC2-low group. Gene sets are from (b), logFC cutoff = 0.58. Only upregulated genes are displayed. See also Supplementary Fig. 4 for validation results using the GSE87371 dataset NPC2 knockdown promotes mitochondrial dysfunction, ROS accumulation and aggressiveness in DLBCL cells The SU-DHL-4 (GCB) and SU-DHL-2 (ABC) cell line are well-characterized and widely used human DLBCL cell line for functional studies of DLBCL pathogenesis and chemo-resistance which has been applied to investigate molecular mechanisms underlying drug-resistance and B-cell lymphoma aggressiveness in previous reports [44–46]. We first explored the intrinsic role of NPC2 in SU-DHL-4 cells, two short hairpin RNAs (sh1 and sh2) targeting NPC2 were designed and constructed (Fig. 5a). qPCR analysis verified that both shRNAs efficiently knocked down endogenous NPC2 expression (Fig. 5b). Functional assays revealed that NPC2 depletion exerted no significant effect on cell proliferation (Fig. 5c). Similarly, cell cycle distribution and apoptotic rate remained unchanged following NPC2 knockdown, indicating that NPC2 does not regulate these cellular processes (Supplementary Fig. 5a and 5b). NPC2 knockdown enhances malignant phenotypes in SU-DHL-4 cells. (a) Schematic of NPC2 shRNA target design and qPCR primer locations. (b) RT-qPCR analysis of NPC2 mRNA expression levels after shRNA transduction. (c) Cell number assessed by trypan blue staining. The cell number at the initial time point was normalized to 1. (d-e) JC-1 staining analyzed by flow cytometry. Representative images (d) and quantitative histograms (e) are shown. (f-g) ROS levels detected by DCFH-DA staining and analyzed by flow cytometry. Representative images (f) and quantitative histograms are shown. (h-i) Migration and invasion assays of NPC2-knockdown cells. Representative images (h) and quantitative histograms (i) are shown. Scale bar, 200 μm Notably, in SU-DHL-4, NPC2 knockdown resulted in an increased proportion of green fluorescence in mitochondrial membrane potential analysis (Fig. 5d and e), suggesting the occurrence of mitochondrial depolarization and dysfunction. Meanwhile, intracellular ROS levels were significantly elevated, implying oxidative stress induced by NPC2 deficiency (Fig. 5f and g). Furthermore, trans-well assays demonstrated that NPC2 knockdown markedly enhanced cell migration and invasion, and these phenotypes were consistent in both shNPC2 cell lines (Fig. 5h and i). In summary, NPC2 knockdown does not affect proliferation, cell cycle or apoptosis in SU-DHL-4 cells, but induces mitochondrial depolarization and ROS accumulation. Loss of NPC2 significantly enhances cell migration and invasion, suggesting that NPC2 may act as a negative regulator of lymphoma cell aggressiveness. NPC2 knockdown reduces sensitivity to vincristine in DLBCL cells Given the correlation between low NPC2 expression and poor prognosis, we examined the response of NPC2‑knockdown cells to vincristine (Vinc). Vinc markedly inhibited proliferation in shNC control cells in both SU-DHL-2 (ABC phenotype) and SU-DHL-4 (GCB phenotype) cell lines, whereas this growth suppression was partially alleviated upon NPC2 depletion (Fig. 6a-b and Supplementary Fig. 6a-b), indicating that low NPC2 levels conferred partial chemo-resistance. NPC2 knockdown confers vincristine resistance in SU-DHL-4 cells. (a) Cell number assessed by trypan blue staining. The cell number at the initial time point was normalized to 1. (b) Quantitative analysis of cell numbers at the endpoint (72 h). (c-d) Apoptosis analysis by Annexin V/PI staining and flow cytometry. Representative images (c) and quantitative histograms (d) are shown. (e-f) Cell cycle analysis by PI staining and flow cytometry. Representative images (e) and quantitative histograms (f) are shown Apoptosis assays of two cell lines showed that Vinc induced a strong apoptotic response in control cells, increasing from ~ 2% to 40%, while apoptosis in shNPC2 cells only rose to ~ 15% (Fig. 6c-d and Supplementary Fig. 6c-d). This protective phenotype against Vinc-induced cell death was consistently observed in both SU-DHL-2 and SU-DHL-4 cell lines. Cell cycle analysis revealed a prominent G2/M arrest in Vinc‑treated control cells, consistent with effective drug‑induced cell cycle blockade. In contrast, shNPC2 cells failed to undergo G2/M arrest, further supporting impaired chemotherapeutic response (Fig. 6e-f and Supplementary Fig. 6e-f). Together, these results demonstrate that NPC2 knockdown desensitizes both SU-DHL-2 and SU-DHL-4 cells to Vinc by attenuating apoptosis and G2/M arrest. NPC2 knockdown confers vincristine chemo-resistance and enhances tissue injury in vivo For in vivo studies, NOD‑SCID mice were intravenously injected with negative control shRNA (shNC) or NPC2 shRNA targets (shNPC2) into SU-DHL-4 cells (Fig. 7a). After 21 days, the proportion of human CD45⁺ cells in peripheral blood were comparable between the two groups, indicating similar engraftment efficiency and consistent with in vitro findings that NPC2 does not affect cell proliferation (Fig. 7b and c). NPC2 expression is associated with vincristine resistance in vivo. (a, d) Schematic illustration of the SU-DHL-4 xenograft mouse model without (a) or with (d) vincristine treatment. Mice were treated with vincristine every two days for four doses starting 12 days post-transplantation. (b-c, e-f) Flow cytometry analysis of human CD45⁺ cells in peripheral blood of xenograft mice. Representative images (b, e) and quantitative histograms (c, f) are shown. (g) Survival curve of transplanted mice following the same treatment as described in (d). (h) Representative images of hematoxylin and eosin (H&E) staining of heart, liver, spleen, and kidney tissues. Scale bar, 100 μm. (i) Histopathological scoring of heart, liver, spleen, and kidney tissues For in vivo therapeutic assessment, NOD‑SCID mice were intravenously injected with shNC or shNPC2 SU-DHL-4 cells. An additional PBS control group was included in the in vivo experiment, in which mice received no tumor cell inoculation but were only treated with Vinc; this group served specifically as a drug toxicity control to independently evaluate the in vivo adverse effects caused by Vinc monotherapy. After 12 days, mice from all three groups were treated with Vinc at two-days interval for four doses. All mice were sacrificed on day 21 for peripheral blood flow cytometry analysis and organ histological analysis (Fig. 7d). In the therapeutic trial, Vinc treatment significantly reduced hCD45⁺ cells in control mice, whereas this effect was largely blunted in shNPC2 group, with hCD45⁺ proportion nearly 1.5‑fold higher than in controls (Fig. 7e and f). These data confirmed that NPC2 depletion causes resistance to Vinc in vivo, which may contribute to the poor prognosis of patients with low NPC2 expression (Fig. 7g). At the endpoint, major organs were harvested for histological analysis. Severe tissue damage was observed in the liver, spleen, and kidney, but not in the heart, which may be attributed to enhanced migration and invasion induced by NPC2 knockdown (Fig. 7h and i). To evaluate DLBCL cell infiltration in tissues, IHC staining for hCD45 was performed on mouse spleen and liver samples. The results demonstrated that the extent of hCD45-positive cell infiltration was relatively low in the shNC group, whereas markedly increased infiltration was observed in the shNPC2 group (Supplementary Fig. 7a and b). This consistent infiltration pattern was observed in both spleen and liver tissues. Together, these results demonstrate that NPC2 knockdown desensitizes SU-DHL-4 cells to Vinc, and promotes systemic tissue injury in vivo. Discussion DLBCL remains a highly heterogeneous disease with substantial unmet needs in relapsed or refractory cases, where chemo-resistance is critical barriers to treatment success. Consequently, it is essential to clarify the molecular processes that drive the chemo-resistance of DLBCL. NPC2, a lysosomal protein that modulates intracellular cholesterol and lipid homeostasis by mediating cholesterol transport in the late endosomal and lysosomal system, has been implicated in multiple pathological processes, yet its biological functions and underlying molecular regulatory mechanisms in DLBCL remain poorly understood. In this study, we performed integrative scRNA-seq analysis of multiple DLBCL cohorts and identified distinct early-stage and malignant B-cell subpopulations, and further screening confirmed NPC2 as a key prognostic gene significantly downregulated in malignant cells and closely correlated with patient poor survival. Functional enrichment analyses revealed that malignant DLBCL cells were predominantly enriched in cell cycle-related pathways, while low NPC2 expression was closely associated with tumor migration, chemotaxis and immune activation. While the proliferation or cell cycle related GO terms reflect the general state of the malignant population, NPC2 expression itself is not directly linked to this process. Our experiments reveal that NPC2 knockdown in SU-DHL-4 cells abrogates vincristine (Vinc) induced G2/M cell cycle arrest and apoptotic response (Fig. 8). Previous reports have demonstrated that NPC2 modulates chemotherapy sensitivity by suppressing PI3K/AKT pathway in LUAD, while reduced NPC2 expression enhances MAPK/AKT signaling to attenuate sorafenib efficacy in hepatocellular carcinoma [47]. Although we did not directly verify these signaling pathways in the present DLBCL model, this published evidence provides a potential mechanistic reference for understanding the pro-resistance role of NPC2 deficiency. Hypothetical model of NPC2 deficiency in DLBCL progression and chemo-resistance. Reduced NPC2 expression in malignant B-cells induces mitochondrial dysfunction characterized by increased mitochondrial membrane potential (MMP) and elevated reactive oxygen species (ROS) production. These changes promote cell cycle dysregulation, enhanced migration, and invasion, leading to increased tumor aggressiveness. NPC2-deficient DLBCL cells consequently develop resistance to vincristine treatment. In-vivo, this resistance contributes to sustained lymphoma progression and organ damage in xenograft models The mitochondrial depolarization and excessive intracellular ROS accumulation observed in NPC2-depleted SU-DHL-4 cells further reinforce this resistance mechanism, ROS overproduction is known to activate pro-survival signaling pathways and induce genomic instability in tumor cells [48, 49], thereby diminishing chemo-efficacy and promoting treatment tolerance, highlighting NPC2 as a conserved mediator of chemo-resistance across solid tumors and hematological malignancies. However, the association between NPC2 expression and these signaling pathways in DLBCL still requires further study. Based on the DEG results in Fig. 3a (Supplementary Fig. 2d), our findings suggest that the transition from Early to Malignant B-cell states is not solely driven by intrinsic proliferative signals, but is also characterized by a broad transcriptional reprogramming aimed at immune evasion and suppression. Chemotaxis and cell mobility are pivotal for DLBCL dissemination, and our data demonstrate that NPC2 acts as a negative regulator of these processes. Transwell assays confirm that NPC2 knockdown markedly enhances DLBCL cell migration and invasion capabilities, which is in contrast to observations in glioblastoma (GBM) [50]. In GBM, studies using U87-MG and U251 cell lines have clearly shown that NPC2 depletion exerts the opposite effect, specifically inhibiting tumor cell motility, migration, and invasion [50]. This divergent role of NPC2 in DLBCL versus GBM underscores its context-dependent functional specificity across different tumor types. Mechanistically, NPC2’s role in cholesterol transport—critical for membrane integrity and signaling—may underpin this effect. In gastrointestinal cancers and LUAD, NPC2 has also been shown to correlate with chemotaxis and leukocyte migration-related pathways [47, 51], further supporting its conserved involvement in directing tumor cell movement and tissue infiltration, albeit with distinct regulatory patterns in different malignancies. Interestingly, our data indicate that NPC2 knockdown did not significantly increase the invasion of DLBCL cells into the liver, but rather selectively enhanced their accumulation in the spleen (Supplementary Fig. 7). This organ-specific pattern is consistent with our H&E staining results (Fig. 7h), though the underlying mechanisms warrant further investigation. Despite these findings, this study has several limitations. Previous studies have established that PI3K/PTEN dysregulation and immune alterations, including CD8⁺ T cell exhaustion, profoundly dictate tumor behavior and clinical outcomes [52, 53]. However, we did not explore the role of NPC2 in the DLBCL tumor microenvironment (TME), which is a critical regulator of tumor progression; previous studies in gastrointestinal cancers, GBM and LUAD have highlighted NPC2’s correlation with immune cell infiltration [47, 54], suggesting its potential involvement in TME remodeling that may further modulate DLBCL progression. In addition, while we observed that NPC2 knockdown abrogates Vinc-induced G2/M arrest, the specific molecular mechanisms by which NPC2 regulates the cell cycle remain unelucidated. Conclusion We identify NPC2 as a potential target and negative regulator of DLBCL aggressiveness and vincristine (Vinc) resistance. NPC2 deficiency induces mitochondrial dysfunction or ROS accumulation and abrogates Vinc-induced G2/M arrest or apoptosis. This highlights NPC2 as a potential target to improve outcomes in relapsed or refractory DLBCL (Fig. 8). Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements Not applicable. Abbreviations Activated B-cell Cell-of-origin Differentially expressed genes Diffuse large B-cell lymphoma Glioblastoma Germinal center B Immunohistochemistry Non-Hodgkin lymphoma Niemann-Pick disease type C2 Normalized transcripts per million Primary central nervous system Rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone Reactive oxygen species Quantitative Real-Time PCR Single-cell RNA sequencing Short hairpin RNA Tumor microenvironment Uniform manifold approximation and projection Vincristine Author contributions Conceptualization, YY and PX; methodology, YY, LX and PX; validation, YY and WQ; formal analysis, YY and WQ; investigation, YY, WQ and LX; data curation, YY; writing and original draft preparation, YY; writing—review and editing, YY and PX; supervision, PX; funding acquisition, PX. All authors read and approved the final manuscript. Funding This work was supported by the National Natural Science Foundation of China (82370207 for PX), Natural Science Foundation of Jiangsu Province (BK20240120 for PX). Data availability All the data generated and analyzed during this study are included in the manuscript. Declarations Ethics approval and consent to participate All animal experiments conducted in this study were approved by the Animal Ethics Committee of Nanjing Drum Tower Hospital (Approval No. 2020 ~ Gl18). The study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals. All efforts were made to minimize suffering. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement All the data generated and analyzed during this study are included in the manuscript.