Lysosome-related molecular subtypes reveal prognostic signatures and immunotherapeutic implications in ovarian cancer

Abstract

Lysosomes are critical organelles that act as degradation centers and signaling hubs within cells, playing a significant role in various cellular processes and human diseases, including cancer. However, the extent to which they influence the heterogeneity and clinical outcomes of ovarian cancer (OC) remains inadequately understood. In this study, we used consensus clustering to identify two distinct lysosome-related clusters (LCs) in OC by analyzing the expression profiles of OC patients from The Cancer Genome Atlas (TCGA) database. Further analyses revealed the functional characteristics and immune landscapes of these subgroups, providing valuable insights into the tumor immune microenvironment (TIME) and tumor responses to immunotherapy. Additionally, we developed and validated a prognostic model based on differentially expressed genes (DEGs) between the two LCs, demonstrating its effectiveness in predicting patient prognosis, TIME characteristics, and immunotherapy potential in OC. A further investigation explored the relationship between lysosome-associated risk scores, IC50 values of standard antitumor agents, and the expression levels of prognostic genes. Finally, in vitro experiments showed that inhibiting CRHR1, a lysosome-associated prognostic gene, significantly reduced OC cell proliferation, invasion, and migration. In conclusion, our study establishes a novel lysosome-based classification and prognostic framework for OC, offering a practical tool to predict clinical outcomes and guide personalized immunotherapy strategies.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-18271-9.

Introduction

Ovarian cancer (OC) is a formidable adversary in the realm of gynecological malignancies, consistently ranking as the most lethal with a sobering annual global mortality tally of approximately 200,000 lives¹. This aggressive cancer is often detected at an advanced stage, partly explaining the grim 5-year survival rate that lingers below the 50% threshold, a statistic heavily influenced by its relentless patterns of recurrence and metastasis². The heterogeneity of OC is underscored by its various subtypes, each with distinct molecular profiles and clinical behaviors. Approximately 90% of OC cases are of the epithelial ovarian cancer (EOC) variety, and this includes various subtypes such as high-grade serous ovarian cancer (HGSOC, about 80%), low-grade serous ovarian carcinoma (LGSOC), ovarian endometrioid carcinoma (OEC), ovarian clear cell carcinoma (OCCC), and mucinous ovarian cancer (MOC). Despite the advent of platinum-based chemotherapy and taxane regimens as the cornerstone of OC treatment, the specter of drug resistance looms large, leading to a high rate of treatment failure and disease recurrence³. The quest for more effective therapeutics is further complicated by the relative dearth of predictive biomarkers that can foretell the response to chemotherapy and immunotherapy, underscoring the urgent need for novel strategies to combat this deadly disease. The development of personalized medicine, underpinned by a deeper understanding of molecular intricacies, holds the promise of transforming outcomes for patients afflicted with this complex and challenging cancer.

Lysosomes, derived from the Golgi and endoplasmic reticulum networks, are pivotal for cellular recycling, maintaining a unique acidic environment of approximately pH 4.5-5.0, achieved through the action of the vacuolar ATPase (V-ATPase)⁴. This acidic milieu harbors a plethora of enzymes, including lipases, peptidases, nucleases, and glycosidases, which are adept at breaking down a vast array of macromolecules⁵. Lysosomes serve as the cellular central degradation and recycling stations, processing both extracellular and cytoplasmic materials delivered via endocytosis and autophagy. Beyond the traditional roles in degradation, lysosomes have emerged as key regulators of cellular function, with extra lysosomal enzyme activities, especially cathepsins, influencing cell biology in multiple ways⁵. These enzymes, once outside the lysosomal confines, can modulate cell signaling, proteolysis, and even cell death pathways. The role of lysosomes extends into the complex interplay between cellular metabolism and the tumor microenvironment (TME). In cancer, lysosomes exhibit heightened activity and an increased capacity for biogenesis, supporting the metabolic demands of rapidly dividing cancer cells. The enhanced lysosomal function facilitates stress response, nutrient recycling, and provides a survival advantage in the often harsh TME. Moreover, the secretion of lysosomal enzymes has been demonstrated to promote tumor invasion and metastasis by degrading the extracellular matrix and modulating the immune response⁶.

It is evident that lysosomal dysfunction plays a pivotal role in the development of cancer and other diseases, thus serving as a promising therapeutic target⁷. The delicate equilibrium of lysosomal activity is paramount for sustaining cellular homeostasis, and its dysregulation can precipitate a range of pathologies, encompassing lysosomal storage diseases and neurodegenerative conditions. A comprehensive understanding of the multifaceted roles of lysosomes in both health and disease is imperative for the development of novel strategies in cellular biology and medicine. However, the role of lysosomes in OC remains largely unexplored. In light of these findings, we propose that the expression of genes associated with lysosomal function could serve as a molecular classifier for OC, potentially aiding in the risk stratification and the prediction of prognosis for OC patients.

As shown in the flowchart in Fig. 1, in this study, we undertook a comprehensive bioinformatic analysis of the OC patients from the Cancer Genome Atlas (TCGA) cohort, with the objective of categorizing them into two distinct molecular subtypes. The analysis employed the expression patterns of lysosome-related genes as the primary discriminator. The lysosomoe-related molecular subtypes exhibited unique clinical, functional, and immunological attributes. Subsequently, a lysosome-associated risk model was formulated and validated, utilizing differentially expressed genes between these subtypes through rigorous statistical methods, including univariate Cox regression and least absolute shrinkage and selection operator (LASSO) regression. Furthermore, we investigated the interplay between this risk model and immune infiltration patterns, anticancer drug responsiveness, and immunotherapy outcomes, with the aim of elucidating its significance in OC progression. Finally, we investigated the role of CRHR1, one of the prognostic lysosome-related genes, in the proliferation, invasion, migration and apoptosis of OC cell lines. The findings of this study offer novel perspectives on the divergent prognostic and immunological landscapes across various OC molecular subtypes and introduce a robust framework for the stratification of patients to optimize therapeutic outcomes and survival rates.

Fig. 1.

Flowchart of this study.

Materials and methods

Bulk rna‑seq data acquisition and preprocessing

The transcriptomic data and associated clinical information for ovarian cancer were sourced from TCGA database, available at cbioportal data base⁸. To ensure consistency between patient samples, we excluded samples with incomplete survival data. Additionally, we retrieved RNA-seq data and relevant clinical information from the Gene Expression Omnibus (GEO) database for ovarian cancer cohorts GSE32062⁹. We addressed issues related to duplicate gene symbols or multiple probes by selecting genes with the highest average expression levels. Of these, the TCGA-OV cohort included high-grade plasmacytoid carcinoma (HGSC, 80%), endometrioid carcinoma (12%), and clear cell carcinoma (8%). The GEO cohort (GSE32062) was limited to HGSC to minimise heterogeneity. The spatial transcriptomics data utilized in this study were obtained from the publicly accessible database (https://grswsci.top/).

A total of 914 lysosome-associated genes were compiled from the Molecular Signatures Database (MsigDB) database (Table S1).

Consensus clustering and differential gene expression

We selected prognostically relevant lysosome-related genes using univariate cox analysis (Table S2). Use of this part of the gene, we conducted consensus clustering via the “ConsensusClusterPlus” R package. Differential gene expression (DEG) analysis was performed using the “DESeq2” software package, employing an absolute log fold change (|logFC|) threshold of greater than 1 and a P-value of less than 0.05 as the criteria for significance¹⁰.

Construction of the lysosome‑related prognostic signature

To establish a lysosome‑related prognostic signature, we conducted a series of analyses including univariate and multivariate Cox regression, as well as ten-fold cross-validation using the LASSO regression¹¹. In the LASSO regression, we selected the “lambda.min” to prevent overfitting. A set of eight genes (PTGFR, ADH1C, CRHR1, CRLF1, SLITRK3, TH, NGFR and CD38) was utilized to construct the prognostic formula. Risk score = ∑ni(Coefi* Expi), where Coefi represents the coefficients of the genes and Expi represents relative expression of genes in the cohort.

Validation and performance assessment

We performed Kaplan-Meier analysis to compare the overall survival rates of high-risk and low-risk subgroups based on the median Riskscore. The accuracy of the Riskscore in predicting 1-year, 3-year, and 5-year survival rates was evaluated using ROC curves through the “timeROC” R package¹². Additionally, a prognostic nomogram incorporating the Riskscore and other clinical features was constructed using the “rms” R package. The performance of the nomogram was assessed through calibration curves and Decision Curve Analysis (DCA)¹³.

Functional enrichment analysis

Functional enrichment analysis was conducted using the “clusterProfiler” R package, focusing on Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways¹⁴. Additionally, Gene Set Enrichment Analysis (GSEA) and Gene Set Variation Analysis (GSVA) were employed to compare pathway activation between diverse groups¹⁵. The relevant pathways were sourced from MSigDB and associated research studies^16,17.

Immune infiltration analysis and immunotherapy responsiveness

The ESTIMATE (Estimation of Stromal and Immune cells in Malignant Tumor tissues using Expression data) algorithm was utilized to assess variations in immune scores, stromal scores, ESTIMATE scores, and tumor purity among samples¹⁸. To investigate immune cell infiltration, we employed the efficient algorithm TIMER2.0 to predict immune cell infiltration based on the gene expression data of tumors (http://timer.cistrome.org/). Additionally, the Tumor Immune Dysfunction and Exclusion (TIDE) algorithm was used to predict the response of ovarian cancer patients to immune checkpoint inhibitors (ICIs) treatment.

Single-cell transcriptome analysis

In this study, we procured single-cell RNA sequencing (scRNA-seq) data from seven ovarian tumor samples, which were sourced from the GEO repository under the accession number GSE184880. For our analysis, we utilized the Seurat package within the R programming environment to execute unsupervised clustering on a per-cell basis, with the read count matrix serving as the foundational input. Implementing stringent quality control procedures, we meticulously assessed each cell based on the number of genes detected and the proportion of mitochondrial gene counts. Cells were systematically filtered out if they exhibited fewer than 200 detected genes or if the mitochondrial gene counts exceeded 20% of their total gene counts. Additionally, to minimize the influence of spurious signals, we excluded genes that were detected in three or fewer cells. To mitigate batch effects and ensure data consistency across samples, we employed the Harmony algorithm for the integration and harmonization of multi-sample datasets. Guided by the Seurat workflow, we proceeded with dimension reduction clustering and differential expression analysis. This involved conducting principal component analysis (PCA) and uniform manifold approximation and projection (UMAP) for dimension reduction, focusing on the top 20 principal components to capture the most significant variance in the data. The assignment of cell cluster annotations was facilitated by the use of established gene markers, which served as reliable indicators for the identification and classification of distinct cell populations. Through this comprehensive approach, we aimed to elucidate the cellular heterogeneity within ovarian tumors and to uncover novel insights into the underlying biology of these complex malignancies.

Predicting drug responses

We used the R package oncoPredict to assess the predictive ability of risk score chemotherapeutic agents by calculating patients IC50 for various common chemotherapeutic agents¹⁹.

Cell culture

The human ovarian cell line (SK-OV-3) were acquired from the Chinese Academy of Sciences (Shanghai, China). Unless specified otherwise, cells were cultured in RPMI-1640 medium (Pricella, Wuhan, China). All culture media contained 10% fetal bovine serum (Pricella) and 1% penicillin/streptomycin (Pricella). Cells were maintained at 37 °C in a humidified 5% CO2 incubator.

SiRNA transfection

CRHR1 siRNA-1 (5′-GCAAAGTGCACTACCACAT‐3′) and CRHR1 siRNA-2 (5′‐GGGCCATTGGGAAACTTTACT‐3′) were used to knock down the expression of CRHR1, while a scramble siRNA was used as a control. The siRNAs were obtained from GenePharma Co. (Shanghai, China).

RNA extraction and quantitative real-time PCR (qPCR)

Total RNA was extracted using the TRIzol reagent (Takara, #9108), followed by reverse transcription into cDNA using the Hifair^® III 1st Strand cDNA Synthesis SuperMix (Yeasen, 11141ES60) according to the manufacturer’s instructions. SYBR (Yeasen, 11201ES08), primers, and cDNA were mixed in the recommended proportions as per the manufacturer’s instructions. The qPCR was conducted using Bio-Rad CFX96. Gene expression levels were normalized with actin used as an internal reference.

Cellular proliferation, invasion, and migration assays

For the measurement of cell proliferation, a Cell Counting Kit-8 (CCK-8) assay was used according to the manufacturer’s instructions (CK04, Dojindo, Japan). The colony formation assay was performed by inoculating 200 cells/well in a 6-well plate. Approximately 14 days later, colonies were visible and fixed. Cells were stained with 0.1% crystal violet and photographed. Transwell assays with Matrigel (plain RPMI-1640:Matrigel = 9:1) were performed to measure cell invasion and migration. Equal amounts of cells were seeded in the upper chambers (Corning, USA) in serum-free medium, and RPMI-1640 with 10% FBS was added to the lower chambers. After 24 h, the invasive cells were fixed and stained with 0.1% crystal violet.

Statistical analysis

All data analyses were performed using R software (version 4.1.3). Unless otherwise specified, the two-tailed Wilcoxon test was used for comparisons. Fisher’s exact test was utilized for contingency table analysis. Pearson correlation analysis was conducted to assess the correlations between variables. Survival differences were evaluated using Kaplan-Meier (K-M) survival curves and log-rank tests. A P-value of less than 0.05 was considered statistically significant.

Results

Identification of lysosome-related molecular subtypes in OC

Despite substantial advancements in understanding the intertumoral heterogeneity of OC, the identification of novel molecular subtypes continues to be a critical unmet need, imperative for mitigating disease recurrence and drug resistance. Initially, we curated OC patients from TCGA database to delve into the expression heterogeneity of lysosome-related genes. Employing an unsupervised consensus clustering analysis on the variable expression profiles of lysosome-related genes, two distinct lysosome-related clusters (LCs), designated as LC1 and LC2, were discerned (Fig. 2A and Figures S1A, B). Further validation through PCA revealed significant transcriptional differences between these two LC clusters, underscoring their distinct molecular identities (Fig. 2B). Given the significance of cancer subtype classification in elucidating tumorigenic mechanisms, we embarked on a comparative analysis of essential molecular and clinical attributes between the LC clusters. Notably, we observed distinct expression patterns of variable lysosome-related genes among OC patients within the LC1 and LC2 clusters (Fig. 2C). When examining clinical characteristics, we found that the LC2 cluster harbored a higher proportion of patients with advanced disease stages and older age (Fig. 2C). Furthermore, survival analysis revealed that LC2 patients exhibited inferior prognoses compared to LC1 patients (Fig. 2D). The additional independent clustering validation set also provided supplementary evidence for the stability of this clustering pattern (Figures S1C, D). In summary, our study identified two novel molecular subtypes of OC based on the variable expression of lysosome-related genes, which not only exhibited distinct molecular and clinical features but also correlated with differential prognoses.

Fig. 2.

Consensus clustering based prognostic lysosome-related genes in OC. (A) Consensus matrix depicting the clustering results when k (cluster number) is set to 2. (B) Point plots of PCA analyses for clustered groupings. (C) Heatmap displaying the expression levels of prognostic lysosomal-related genes, along with clinical characteristic annotations for each cluster. (D) Kaplan-Meier curves illustrating the OS differences between the two clusters (P = 0.00022).

Functional characteristics of different LC subtypes

In order to explore the disparities in potential biological functionalities between the two clusters in greater depth, we initiated our investigation with DEG analysis (Fig. 3A). Subsequent to this, we employed GSEA, leveraging the fold changes identified in the DEG analysis, which unveiled that LC2 cluster patients exhibited heightened enrichment scores in interferon (IFN) response, inflammatory response, and oxidative phosphorylation signaling pathways (Fig. 3B). Conversely, OC patients belonging to the LC1 cluster manifested an elevated enrichment in epithelial-mesenchymal transition (EMT), hedgehog signaling, and myogenesis signaling pathways (Fig. 3B). Furthermore, we conducted comprehensive KEGG and GO functional enrichment analyses, which revealed a prominent upregulation of numerous neuro-related signaling pathways in LC2 cluster patients. These pathways encompassed neuroactive ligand-receptor interaction, axon development and regulation of nervous system development (Fig. 3C, D). In contrast, LC1 cluster patients displayed a heightened enrichment in a spectrum of immune-related pathways, particularly cytokine-cytokine receptor interaction, antigen processing and presentation, and leukocyte mediated immunity (Figure S1E, F). To gain a nuanced understanding of the transcriptional heterogeneity within OC patients, we implemented the GSVA algorithm to quantify 16 recurrent cancer cell states that interface with the tumor microenvironment (TME)^16,17, forming organized systems conducive to immune evasion, metastasis, and drug resistance. Our findings indicated that LC1 subtype patients were enriched in diverse gene modules, such as IFN and oxidative phosphorylation modules. In stark contrast, LC2 cluster patients demonstrated a higher score for astrocyte (AC)-like, oligodendrocyte progenitor cell (OPC)-like and neural progenitor cell (NPC)-like modules (Fig. 3E). Collectively, our systematic exploration of the functional state disparities between the two molecular subtypes of OC not only underscores their distinct biological characteristics but also offers fresh perspectives into the intricate mechanisms that underpin OC progression.

Fig. 3.

Differences of biological functions between LC1 and LC2 subgroups. (A) Volcano plot of DEGs between clusters, with LC1 as control (| logFC |>1, P < 0.05). (B) Bar plot showing different pathways enriched between LC1 and LC2. (C) KEGG pathways enriched in LC2. (D) GO analysis highlighting the biological processes (BP) enriched in LC2. (E) Boxplots showing the signature score of 16 cancer cell states between LC1 and LC2.

Distinct immune landscapes of LC1/2 subtypes

We next explored whether variations in lysosome-related gene expression patterns signified intertumoral immune heterogeneity and microenvironmental differences in OC, by detecting a panel of immunologically pertinent signatures. Our findings illuminated that a substantial proportion of immune-associated pathways exhibited augmented enrichment within the LC1 cohort (Fig. 4A). To further substantiate our findings, we leveraged the ESTIMATE algorithm to analyze the TME scores across the two LC clusters. Intriguingly, the LC1 cluster exhibited higher TME scores, suggesting a more pronounced immune presence (Fig. 4B). Recognizing the TME heterogeneity between these two clusters, we employed CIBERSORT to assess the relative abundance of distinct immune cell subpopulations in OC. Notably, the LC1 cluster was characterized by higher infiltration levels of M1-like macrophages, naïve B cells, and T follicular helper (Tfh) cell, whereas memory resting CD4⁺ T cells and M1-like macrophages were predominantly enriched in the LC2 cluster (Fig. 4C). Thus, the LC1 cluster exhibited an immune “hot” phenotype, partly explaining patients in this group had better outcomes compared to those in the LC2 cluster. Moreover, through TIDE analysis, we found the LC2 cluster exhibited notably elevated TIDE scores, exclusion scores, and dysfunction scores, hinting at a heightened likelihood of immune evasion among these patients (Fig. 4D-F). Additionally, we discerned a diminished immunotherapy response in LC2 cluster patients compared to LC1 cluster patients (Fig. 4G). Overall, our study provides a comprehensive portrayal of the immune landscape and immunotherapy responsiveness within distinct OC subgroups, underscoring the complexity and heterogeneity of the immune microenvironment in OC.

Fig. 4.

Immune Infiltration and responsiveness to immunotherapy across Clusters. (A) Boxplots showing the signature score of immune-related pathways between LC1 and LC2. (B) The mean value of scaled estimate scores between LC1 and LC2. (C) Boxplots showing the proportion of 22 immune cells in LC1 and LC2 of OV estimated by CIBERSORT. (D–G) TIDE analysis including TIDE score, Exclusion score, Dysfunction score and potential immunotherapy responder.

Development of a lysosome-related prognostic model in OC

The aforementioned findings solidified the link between lysosome-related genes and both the prognoses and responses to immunotherapy of OC patients, indicating the potential of lysosome-related classification for evaluating prognosis and treatment efficacy. Consequently, we intended to build a novel lysosome-related prognostic mode based on the DEGs between these two OC clusters. By utilizing univariate Cox regression and LASSO analysis, we narrowed down our focus to eight genes, including PTGFR, ADH1C, CRHR1, CRLF1, SLITRK3, TH, NGFR and CD38, and developed a risk assessment model using the OC patients among the TCGA cohort as the training set (Figs. 5A and Figures S2A-C).

Fig. 5.

Establishment and validation of the lysosomal-related prognostic signature. (A) Multivariate Cox coefficients for eight genes (PTGFR, ADH1C, CRHR1, CRLF1, SLITRK3, TH, NGFR and CD38) in the prognostic signature. (B) Riskscore distribution among OV patients, sorted from lowest to highest. (C) Survival status categorized by Riskscore for each OV patient. (D) Sankey diagram correlating clusters, Riskscore groups, and OV survival status. (E) Heatmap displaying expression levels of seven genes in different Riskscore groups. (F) Kaplan–Meier analysis comparing overall survival between high and low Riskscore groups in OV (P < 0.0001). (G) Receiver Operating Characteristic (ROC) curves depicting Riskscore signature’s predictive performance for 1, 3, and 5-year overall survival in BLCA. (H,I) Kaplan-Meier analysis and time-dependent ROC curves in external validation sets: GSE32062.

Following the computation of the lysosome-related risk score for each patient within the OC cohort, we stratified them into high-risk and low-risk groups based on the median risk score (Fig. 5B). In parallel, we presented the survival status of the patients (Fig. 5C) and the expression patterns of the eight genes comprising the prognostic signature (Fig. 5D). A noteworthy trend emerged: patients with higher lysosome-related risk scores exhibited poorer survival outcomes, which was mirrored by increased expression levels of the “risk” genes, specifically PTGFR and ADH1C. We then employed a Sankey diagram, to visually illustrate the intricate interplay between the identified clusters, lysosome-related risk subgroups, and the survival outcomes in OC (Fig. 5E). This diagram vividly demonstrated that patients with elevated lysosome-related risk scores were disproportionately represented in LC2 and were more likely to face a less favorable prognosis. This observation was corroborated by Kaplan-Meier survival analysis (Fig. 5F), which underscored the significant survival advantage enjoyed by patients belonging to the low-risk group, further validating the prognostic value of the lysosome-related prognostic model in stratifying OC patients based on their overall survival (OS) prospects.

To evaluate the predictive prowess of our prognostic signature, we crafted Receiver Operating Characteristic (ROC) curves for OS at 1, 3, and 5 years (Fig. 5G). The areas under the curve (AUC) yielded values of 0.61, 0.64, and 0.71, respectively, underscoring the commendable predictive performance of our model, particularly in long-term survival predictions. Additionally, we verified the accuracy of our signature in a distinct external validation set (GSE32062), yielding satisfactory results with 1, 3, 5-year AUC values of 0.64, 0.62 and 0.63, respectively (Fig. 5H-I). These findings not only affirm the robustness of our prognostic model but also highlight its potential for clinical application.

Establishment of a nomogram to forecast survival in OC

In addition, we conducted both univariate and multivariate Cox regression analyses, which revealed that the lysosome-related prognostic signature score emerged as an independent prognostic factor for OC patients, regardless of age, pathological stage, and histologic grade (Fig. 6A). Subsequently, we devised a predictive nomogram to enhance the prognostic efficacy of the lysosome-related risk model. This nomogram serves as a quantitative and visualization tool for predicting 1-, 3-, and 5-year OS rates (Fig. 6B). To assess the performance of the nomogram, we plotted calibration curves, which demonstrated that the prediction curves of the model closely aligned with the ideal curve, indicating good agreement between predicted and observed outcomes (Fig. 6C-E). Moreover, the nomogram exhibited a significant positive net benefit in predicting the risk of death, surpassing the traditional age and TNM staging system in this regard (Fig. 6F). In summary, these findings underscore the substantial clinical value of the nomogram model in predicting survival outcomes for OC patients.

Fig. 6.

Survival prediction nomogram based on riskscore. (A) Univariate Cox regression analysis of clinical characteristics and Riskscore. Factors with P < 0.05 were included in subsequent multivariate Cox regression analysis. (B) Nomogram incorporating age, stage and Riskscore, utilized for 1, 3, and 5-year survival predictions. (C–E) Calibration curves at 1, 3, and 5 years, respectively, demonstrating nomogram’s predictive accuracy. (F) Decision curve analysis (DCA) evaluating the clinical utility of the nomogram.

Functional and characteristics features of high/low-risk group patients

Subsequently, we delved into DEG analysis and identified 793 upregulated genes and 448 downregulated genes in high-risk group patients compared to their low-risk counterparts (Fig. 7A). GO and KEGG enrichment analyses of the upregulated genes revealed their involvement in pivotal pathways such as neuroactive ligand-receptor interaction, PI3K-AKT signaling pathway and extracellular matrix organization (Fig. 7B, C). Conversely, the downregulated genes were associated with multiple immune-related pathways like natural killer (NK) cell mediated cytotoxicity, antigen processing and presentation and complement activation (Figure S3A, B). To gain further insights, we conducted GSEA analysis, which revealed that EMT, myogenesis, and angiogenesis were enriched in high-risk patients, suggesting a more aggressive phenotype. In contrast, IFN alpha and gamma response, oxidative phosphorylation and IL6-JAK-STAT3 signaling pathway were highly enriched in low-risk patients, indicative of a more indolent disease state (Fig. 7D). By leveraging GSVA to calculate the signature scores of 16 recurrent cancer cell states, we found that the AC, basal, and mesenchymal gene modules were significantly overexpressed in high-risk patients compared to those in low-risk group (Fig. 7E). Furthermore, we depicted the mutation profiles of high- and low-risk OC patients and observed that most frequently mutant genes showed comparable mutation frequency, except for higher mutation frequency of FLG and NF1 (Figure S3C). Collectively, these findings underscore the existence of distinct functional and genomic characteristics between high- and low-risk OC patients, providing valuable insights into the underlying biological mechanisms that differentiate these two risk strata.

Fig. 7.

Differences in biological functions between risk subgroups. (A) Volcano plot of DEGs between low/high Risk groups, using the low risk group as control (| logFC |>1, P < 0.05). (B) KEGG pathways enriched in high risk group. (C) GO analysis highlighting the BP enriched in high risk group. (D) Bar plot showing different pathways enriched between low/high risk group calculated. (E) Boxplots showing the signature score of 16 cancer cell states between low/high risk group. (F) Boxplots showing the signature score of 17 immune pathways between low/high risk group.

Two risk groups showed distinct immune microenvironments and immunotherapy responses

Despite the scarcity of research delving into the intricate relationship between lysosome-related genes and the tumor immune microenvironment (TIME) in OC, elucidating the TIME landscape in distinct risk groups is paramount. Consequently, we embarked on an examination of the differences in TIME between high- and low-risk OC patients. Notably, our ESTIMATE analysis unveiled a significant elevation of immune scores and decreased stromal scores among patients with higher risk scores (Fig. 8A and Figures S4A-C). To further dissect the intricacies of TIME variations, we harnessed multiple sophisticated algorithms, including CIBERSORT, CIBERSORT-abs, and MCP-counter, which collectively illuminated the diverse TIME compositions between the two risk groups (Fig. 8B and Figures S4D, E). Our findings revealed a notable enrichment of M2-like macrophages and, along with resting memory CD4⁺ T cells in the high-risk BLCA patient group, contrasted by a depletion of M1-like macrophages, regulatory T cells (Tregs) and Tfh cells (Fig. 8B-E).

Fig. 8.

Immune Infiltration and responsiveness to immunotherapy across risk group. (A) The value of Immunescore between low/high risk group. (B) Boxplots showing the proportion of 22 immune cells in low/high risk groups of OC estimated by CIBERSORT. (C–E) Scatter plots showing the correlation between the risk score and the proportion of M1-like macrophages, M2-like macrophages and CD4⁺ T cells. (F–G) TIDE analysis including TIDE score, Exclusion score, Dysfunction score and potential immunotherapy responder.

The profound variations observed in TIME profiles among patients underscored their profound implications for immunotherapy responsiveness²⁰. To predict the therapeutic outcomes, we performed TIDE analysis, a computational method designed to predict the efficacy of immunotherapy, where a higher TIDE score signifies reduced therapeutic benefits and an increased potential for immune evasion. The results revealed startlingly higher TIDE and exclusion scores, but comparable dysfunction scores among high-risk patients (Fig. 8F-H). Remarkably, the high-risk group exhibited a decreased proportion of individuals predicted to favorably respond to immunotherapy (Fig. 8I). These results not only underscore the distinct TIME compositions between high- and low-risk patient groups but also hint at the potential for heightened immunotherapy benefits among low-risk OC patients, providing novel insights for risk-benefit assessment in immunotherapy.

Single-cell RNA-seq analyses reveal an immune-activated microenvironment in low-risk patients

scRNA-seq analysis of solid tumors has provided valuable insights into the heterogeneity across various subtypes and the intricate cellular ecosystems that include tumor cells alongside immune and stromal cells²¹. This technology enables a detailed examination of individual cell gene expression profiles, highlighting the diverse interactions within the tumor microenvironment. Understanding these complexities is crucial for developing targeted therapies and improving treatment outcomes in OC. Herein, we reanalyzed a public scRNA-seq dataset of OC patients and identified plenty of canonical cell types, including diverse immune cell subpopulations, stromal cell subsets and epithelial cells (Fig. 9A-B). Based on the calculated lysosome-related risk score of each OC sample, we assigned seven OC patients into high- and low-risk groups (Fig. 9C). Obviously, we noticed dramatically distinct cellular components in two groups, with abundant CD8⁺ T cells present in the low-risk group (Fig. 9D-F). Furthermore, in the CD8⁺ T cell compartment, the low-risk group showed an elevated enrichment score of IFNα and IFNγ response, and multiple T cell mediated immune pathways (Fig. 9G, H). Besides, several T cell effector molecules were highly expressed in low-risk group (Fig. 9I). Summarily, the above single-cell RNA-seq analyses of OC patients reveal an immune-activated microenvironment in the low-risk group patients.

Fig. 9.

scRNA-seq analyses of the TIME based on the lysosome-related prognostic signature. (A) UMAP plot showing the major cell subpopulations in OC. (B) Bubble heatmap showing expression levels of selected signature genes in OC. Dot size indicates fraction of expressing cells, colored based on normalized expression levels. (C) The rank of risk scores based on the pseudobulk RNA-seq expression of OC samples. (D) Relative proportions of diverse cell types across high/low-risk OC tumors. (E) UMAP plot showing the major cell subpopulations of high-risk samples. (F) UMAP plot showing the major cell subpopulations of low-risk samples. (F) UMAP plot showing the diverse subsets of myeloid cells in breast cancers. (G) GSEA analysis of IFNα and IFNγ pathways in CD8⁺ T cells of high-risk versus low-risk tumors. (H) GSEA analysis of various T cell-related pathways in CD8⁺ T cells of high-risk versus low-risk tumors. (I) Bubble heatmap showing expression levels of selected signature genes in CD8⁺ T cells. Dot size indicates fraction of expressing cells, colored based on normalized expression levels.

Anti-cancer drug sensitivity analysis

Tumor heterogeneity significantly impacts patient outcomes by causing varying responses to different treatments, ultimately leading to treatment failure and tumor recurrence²². To gain insights into the interplay between drug sensitivity, lysosome-related risk score, and candidate prognostic genes, we determined the half maximal inhibitory concentration (IC50) values for each drug in the training set samples. Our analysis of BLCA patients revealed crucial associations between these factors (Fig. 10A). Specifically, the risk score positively correlated with the IC50 values of erlotinib, GSK1904529A, and leflunomide, suggesting a reduced sensitivity to these drugs in patients with higher risk scores. Conversely, the risk score negatively correlated with the IC50 values of AZD1332, AZD2014, and IGF1R_3801, indicating increased sensitivity to these agents in high-risk patients. Furthermore, we observed that the IC50 values of BMS.536,924, NU7441 and staurosporine were inversely related to the expression levels of PTGFR. High expression of NGFR was positively correlated to the IC50 of erlotinib. This suggests that the expression levels of these candidate prognostic genes may modulate drug sensitivity. Besides, the correlation between model genes and classical therapeutic agents was comprehensively illustrated (Fig. 10B-M). These findings provide valuable insights for personalizing treatment strategies in OC patients, as they indicate that drug selection should be tailored based on individual patient risk scores and gene expression profiles.

Fig. 10.

High- and low-risk group patients differ in drug sensitivity. (A) Bubble plot showing the relationship between IC50 of drugs, risk score, and model genes. (B,C) Boxplot showing the comparison of IC50 of Erlotinib between high/low-risk groups, and scatter plot showing the correlation between the IC50 of drug and risk score. (D,E) Boxplot showing the comparison of IC50 of Leflunomide between high/low-risk groups, and scatter plot showing the correlation between the IC50 of drug and risk score. (F,G) Boxplot showing the comparison of IC50 of Fulvestrant between high/low-risk groups, and scatter plot showing the correlation between the IC50 of drug and risk score. (H,I) Boxplot showing the comparison of IC50 of Pictilisib between high/low-risk groups, and scatter plot showing the correlation between the IC50 of drug and risk score. (J,K) Boxplot showing the comparison of IC50 of Staurosporine between high/low-risk groups, and scatter plot showing the correlation between the IC50 of drug and risk score. (L,M) Boxplot showing the comparison of IC50 of Foretinib between high/low-risk groups, and scatter plot showing the correlation between the IC50 of drug and risk score.

Inhibition of CRHR1 impedes OC progression

Among the candidate prognostic genes, we focused on CRHR1, which has been studied only minimally in OC. To investigate the role of CRHR1 in OC cells, we designed two siRNAs (CRHR1-si1 and CRHR1-si2) to silence CRHR1 expression in SKOV3 cell line. CRHR1 expression was verified using RT-qPCR, confirming that both siRNAs effectively reduced CRHR1 levels (Fig. 11A). Subsequently, we performed CCK8 and colony formation assays on SKOV3 cells transfected with CRHR1-si1 and CRHR1-si2. The results demonstrated that CRHR1 knockdown significantly inhibited the proliferation of OC cells (Fig. 11B-D). Additionally, transwell assays indicated that the absence of CRHR1 impeded the invasion and migration of SKOV3 cells (Fig. 11E, F). Western blotting revealed that CRHR1 knockdown downregulates Snail and Vimentin while upregulating E-cadherin, suggesting EMT inhibition (Fig. 11G). Overall, these findings suggest that CRHR1 inhibition effectively impedes tumor progression in OC.

Fig. 11.

Knockdown of CRHR1 inhibits OV cell proliferation. (A) Relative levels of CRHR1 in SK-OV-3 cells. (B) SK-OV-3 cells were treated with con siRNA, CRHR1 siRNA-1, or CRHR1 siRNA-2. Cell viability was measured by CCK8 assay. (C,D) A total of 300 cells were treated with con siRNA, CRHR1 siRNA-1, or CRHR1 siRNA-2. Colony formation was measured after 14 d of cultivation. (E,F) Transwell assays showing that CRHR1 silencing decreased cell invasion and migration. Quantitative analysis of the invasion ratio is shown. (G) The protein levels of CRHR1, Snail, Vimentin, E-cadherin and Actin were measured by Western blotting.

Discussion

In the present study, we conducted a comprehensive bioinformatics analysis of the role of lysosome-related genes in OC patients, thereby unveiling a robust correlation between lysosome-related genes and OC subtypes. This finding suggests that OC patients can be stratified into two distinct clusters based on the variable expression profiles of lysosome-related genes. Subsequent interrogation of the molecular, functional state, immune landscape, and immunotherapy response between these two LC subtypes was then conducted. By detecting the DEGs between LC1 and LC2 subtypes, we built and validated a novel lysosome-related prognostic signature. Patients were divided into high- and low-risk groups, and it was observed that these groups exhibited divergent survival outcomes, immune cell infiltrations, immunotherapy responses and drug sensitivities. The identification of lysosome-related molecular subtypes and the establishment of a prognostic model have yielded significant insights, provided a more precise classification of OC patients, and aided in the development of effective therapeutic strategies.

In the realm of personalized medicine, there is a growing emphasis on the precise targeting of subcellular structures, particularly lysosomes, as a promising strategy for cancer treatment. Lysosomes are dynamic organelles that serve as hubs in numerous endosomal trafficking pathways, playing a pivotal role in cellular homeostasis and metabolism. Their dysregulation has been implicated in various aspects of cancer biology, including tumor growth, progression, and resistance to therapeutic interventions⁷. Moreover, the potential role of lysosomes in anticancer therapies has garnered significant interest in the field of immuno-oncology recently. By targeting lysosomes, it triggers apoptotic pathways, inhibits cytoprotective autophagy, and activates a distinct form of apoptosis known as immunogenic cell death (ICD). This mechanism is particularly intriguing as it stimulates both a local and systemic immune response against antigens released from dying cells²³. Stressors that can lead to ICD often involve the generation of reactive oxygen species (ROS), which, in excess, can induce lysosome membrane permeability (LMP)²⁴. In turn, LMP triggers the release of lysosomal contents, including damage-associated molecular patterns (DAMPs) and other immunostimulatory molecules, into the cytoplasm. These molecular signals act as potent activators of the immune system, attracting and activating immune cells such as dendritic cells, which then present tumor-associated antigens to T cells, thereby initiating an adaptive immune response against the tumor. Previous study reported a prognostic model based on lysosome-related genes to serve as a robust tool for accurately predicting the prognoses of acute myeloid leukemia (AML) patients and offer valuable insights into the disease progression and treatment response in AML²⁵. Nevertheless, there is limited research targeting the role of lysosome-related genes in OC.

Our study initially developed novel OC lysosome-based molecular subtypes through consensus clustering analysis, including LC1 and LC2, each characterized by unique clinical, survival, biological, and immunological profiles. The most striking discovery was that the LC2 OC patients had significantly inferior prognoses compared to those in the LC1 group. In addition, the LC2 showed enhanced activity of EMT, a crucial process in cancer progression that endows tumor cells with enhanced migratory, invasive, and survival capabilities. The process of EMT is characterised by the loss of cell polarity and intercellular adhesion, resulting in a mesenchymal phenotype that contributes to metastasis and therapy resistance²⁶. Our findings underscore the pivotal role of EMT in driving the aggressive phenotype observed in LC2 patients. Furthermore, we found multiple neuron-associated pathways were enriched among OC patients in the LC2 subgroup. The intricate interplay between the nervous system and cancer can exert profound influences on various aspects of tumorigenesis and progression. This intricate crosstalk modulates critical processes such as oncogenesis, tumor growth, invasion, metastatic dissemination, development of treatment resistance, stimulation of tumor-promoting inflammation, and impairment of anti-cancer immune responses²⁷. The results indicated that the nervous system played a vital role during the progression of LC2 subtype OC patients. Furthermore, the LC2 group exhibited diminished IFN signaling. A mounting body of research has underscored that quiescent and unresolved type I IFN responses contribute to tumor progression and therapeutic resistance. These suboptimal IFN signaling patterns have been shown to mediate cytoprotective effects, enhance stemness characteristics, foster tolerance to chromosomal instability, and facilitate the establishment of an immunologically exhausted TME²⁸. The inadequate activity of IFN signaling in LC2 patients may offer a partial explanation for the unfavorable prognoses observed in this cohort. Tumor cells have been shown to survive within their microenvironment, evading immune surveillance and resisting drug interventions²⁹. In principle, the LC2 cohort should exhibit a decline in the expression levels of immune checkpoints and a reduction in the infiltration of antitumor immune cells. This collective observation implies a comprehensive impairment of immune functions, which further empowers tumor cells to thrive. The findings of this study indicate that the LC1 patients exhibited an elevated immune score and a multitude of immune-related modules, suggesting an immune “hot” phenotype. Furthermore, the LC1 patients were predicted to be more sensitive to immunotherapy compared to those in the LC2 group.

To further explore the prognostic value of lysosome-related genes, we interrogated the DEGs between LC1 and LC2 groups and built a robust lysosome-related prognostic model via univariate Cox regression and LASSO regression analyses. The present lysosome-related prognostic signature encompassed eight DEGs between two LC subgroups, including PTGFR, ADH1C, CRHR1, CRLF1, SLITRK3, TH, NGFR and CD38. CRLF1 is found to promote proliferation and metastasis of papillary thyroid carcinoma^30,31, but inhibit tumor stemness, tumorigenesis and metastasis of colorectal cancer³². However, there is little research exploring the role of CRLF1 in OC. Corticotropin releasing hormone receptor 1 (CRHR1) regulates immuno-escape of ovarian cancer cell via FasL modulation and might be a potential therapeutical target for ovarian cancer³³. Besides, CRH/CRHR1 signaling promotes cancer cell proliferation via IL-6/JAK2/STAT3 signaling pathway and VEGF-induced tumor angiogenesis in colon cancer³⁴. In our study, we first revealed that inhibition of CRHR1 significantly impedes OC cell proliferation, invasion, and migration, but promoted OC cell apoptosis. EMT is a key determinant of the most lethal features of cancer, including metastasis formation and chemoresistance, making it an attractive target in oncology³⁵. Herein, we found that CRHR1 induces EMT in OC cells as demonstrated by transwell. However, the mechanism by which CRHR1 drives EMT remains unclear and requires further investigation. Alcohol dehydrogenase 1 C (ADH1C) is found to attenuate colorectal cancer progression via the ADH1C/PHGDH/PSAT1/serine metabolic pathway³⁶, but is narrowly studied in ovarian cancer. CD38 exhibits a positive association with prognostic outcomes and the presence of immune cells within the microenvironment of epithelial ovarian cancer, playing a role in modulating the immune response against tumors³⁷.

The lysosome-related prognostic signature demonstrated notable efficacy in predicting the prognosis of OC patients across various datasets. Notably, patients in the high-risk group exhibited significantly worse outcomes compared to those in the low-risk group. Integration of this risk score with clinical parameters led to the development of a nomogram that exhibited substantial clinical predictive value for OC. Further analysis revealed that the high-risk group was associated with the EMT process, characterized by loss of epithelial cell polarity, decreased intercellular adhesion, and acquisition of mesenchymal traits. When comparing the TIME between high- and low-risk groups, the high-risk group exhibited decreased immune cell infiltration overall. The findings of the present study indicated a significant positive correlation between the ratio of M2-like to M1-like macrophages and the lysosome-related prognostic risk score in OC patients. Macrophages in their quiescent state, known as M0, can be differentiated into two distinct phenotypes: M1-like and M2-like. These polarized macrophages play a pivotal role in mediating inflammatory processes. Specifically, M1-like macrophages are predominantly implicated in driving pro-inflammatory reactions, in contrast to M2-like macrophages, which are primarily engaged in modulating anti-inflammatory responses³⁸. Furthermore, the low-risk group exhibited elevated levels of Tfh cells, which facilitate B cell function and antibody-mediated immune responses, frequently associated with improved prognoses in various solid tumors³⁹. The TIDE analysis indicated that low-risk patients were more likely to respond favorably to immunotherapy. These extensive analyses underscore the potential of the lysosome-related prognostic signature as a robust predictor of OC prognosis and immunotherapy response. Finally, the study examined the correlation between the risk score and the IC50 values of commonly used chemotherapy drugs. This analysis provided valuable insights into the potential for combining therapies or selecting OC patients who may be more sensitive to specific chemotherapy drugs. Overall, the present study highlights the potential clinical utility of the lysosome-related prognostic signature in guiding personalized treatment strategies for OC patients.

Whilst the innovative molecular classification based on lysosome-related genes and the robust prognostic model they inform are noteworthy, there are inherent limitations that must be addressed. The practical application of these lysosome-centric approaches in clinical settings could encounter a range of challenges. It is therefore essential to engage in close collaboration with clinical experts with a view to refining and standardizing the processes, thereby improving their clinical applicability and ease of use. Furthermore, the retrospective nature of OC patient recruitment has the potential to compromise the validity of the results. To address this, it is essential to validate these findings through rigorous, multicenter, randomized controlled trials. These trials should be characterized by robust methodologies, extensive patient samples, and thorough follow-up periods to ensure the reliability and generalizability of the outcomes.

Conclusion

In conclusion, our study has delineated two distinct extracellular clusters of OC patients, characterized by significant disparities in survival rates and the infiltration of immune cells, among several other clinical and molecular variables. Furthermore, we have presented a novel prognostic signature anchored in lysosome-related genes, which stands as an independent predictor of OC prognosis. Inhibition of CRHR1 prominently impedes OC cell proliferation, invasion, and migration. Additionally, CRHR1 induces tumor cell apoptosis and EMT. This lysosome-driven molecular stratification and prognostic signature not only serves as a reliable gauge of tumor progression but also holds promise as a valuable instrument for the formulation of individualized clinical strategies. They are particularly pertinent to the field of immunotherapy for OC, where personalized treatment approaches are paramount.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

ZG contributed to the conception and design of this study. ZF, YL and YC analyzed the data and performed the experiments. ZF and YL drafted the original manuscript. ZG polished and revised the manuscript. This manuscript has been read and approved by all authors.

Data availability

The datasets generated and analyzed during the current study are available in the GEO (https://www.ncbi.nlm.nih.gov/geo/) repository.

Declarations

Competing interests

The authors declare no competing interests.