Skip to main content
Download PDF

Identifying PSIP1 as a critical R-loop regulator in osteosarcoma via machine-learning and multi-omics analysis

Cancer Cell International volume 25, Article number: 159 (2025) Cite this article

Abstract

Dysregulation of R-loops has been implicated in tumor development, progression, and the regulation of tumor immune microenvironment (TME). However, their roles in osteosarcoma (OS) remain underexplored. In this study, we firstly constructed a novel R-loop Gene Prognostic Score Model (RGPSM) based on the RNA-sequencing (RNA-seq) datasets and evaluated the relationships between the RGPSM scores and the TME. Additionally, we identified key R-loop-related genes involved in OS progression using single-cell RNA sequencing (scRNA-seq) dataset, and validated these findings through experiments. We found that patients with high-RGPSM scores exhibited poorer prognosis, lower Huvos grades and a more suppressive TME. Moreover, the proportion of malignant cells was significantly higher in the high-RGPSM group. And integrated analysis of RNA-seq and scRNA-seq datasets revealed that PC4 and SRSF1 Interacting Protein 1 (PSIP1) was highly expressed in osteoblastic and proliferative OS cells. Notably, high expression of PSIP1 was associated with poor prognosis of OS patients. Subsequent experiments demonstrated that knockdown of PSIP1 inhibited OS progression both in vivo and in vitro, leading increased R-loop accumulation and DNA damage. Conversely, overexpression of PSIP1 facilitated R-loop resolution and reduced DNA damage induced by cisplatin. In conclusion, we developed a novel RGPSM that effectively predicted the outcomes of OS patients across diverse cohorts and identified PSIP1 as a critical modulator of OS progression by regulating R-loop accumulation and DNA damage.

Introduction

R-loops are dynamic, three-stranded nucleic acid structures composed of RNA-DNA hybrids [1]. During transcription, nascent RNA synthesized by RNA polymerase can invade the DNA template, forming the RNA-DNA duplexes [2]. R-loops can be broadly categorized into physiological and pathological R-loops [3]. Physiological R-loops play crucial roles in various biological processes, such as the regulation of cell proliferation, differentiation, RNA splicing, and DNA damage repair [3]. In contrary, pathological R-loops form in a non-procedural manner and pose great threats to genomic stability [4]. Under pathological conditions, the R-loops accumulate abnormally due to defects in topoisomerases and RNA damage [5]. These aberrant R-loops trigger a series of DNA damage responses, including double-strand breaks and genomic instability, ultimately leading to hypermutation and cell cycle arrest [6]. Recent studies have demonstrated that R-loops exert a profound impact on cancer progression through their complex interactions with tumor suppressors and oncogenes [7]. Alterations in R-loop frequency, stability, or genomic positioning have been linked to oncogenic processes, such as the activation of proto-oncogenes and the inactivation of tumor suppressor genes [8]. For instance, DHX9 alleviates R-loop-induced replication stress by resolving R-loops, contributing to drug resistance in breast cancer [9]. Additionally, Wang et al. reported that activation of estrogen receptor β (Erβ) induced R-loop accumulation in triple-negative breast cancer (TNBC) cells, leading to DNA damage [10]. Exploring the role and mechanisms of R-loops in tumors could facilitate the development of new tumor biomarkers and therapeutic targets.

Osteosarcoma (OS) is the most common primary malignant bone tumor, primarily affecting adolescents and young children [11]. It predominantly arises in the metaphyseal regions of long bones, particularly the distal femur, proximal tibia, and proximal humerus. Over the past four decades, advances in treatment, including surgical resection, chemotherapy, and immunotherapy, have significantly improved the 5-year survival rate of OS patients to approximately 80% [12]. However, the prognosis of patients with lung metastasis remains poor, with the 5-year survival rate of less than 20% [13]. In addition, 60% of patients may have the lung micro-metastasis before distant metastasis diagnosed [13]. Therefore, the development of more sensitive diagnostic methods and personalized treatment strategies for OS patients is crucial. Currently, research on R-loops in OS is limited, but given their known involvement in tumor progression and genomic instability, investigating the role of R-loops in OS is essential.

In recent years, high-throughput RNA sequencing (RNA-seq) and single-cell RNA sequencing (scRNA-seq) technologies have provided new insights into the cancer researches [14,15,16,17,18]. Furthermore, predictive models based on high-throughput data have significantly improved the accuracy of prognosis assessments across various cancers [19, 20]. Therapeutic targets identified through these technologies may also offer promising candidates for the development of novel chemotherapy agents [21, 22]. Currently, Zhang et al. first developed an R-loop-related prognostic evaluation model based on RNA-seq and scRNA-seq datasets, providing a comprehensive insight into the potential molecular mechanisms of R-loops, metabolic reprogramming, and T cell exhaustion in lung adenocarcinoma [23]. Additionally, other studies had utilized high-throughput data to investigate the regulatory mechanisms of R-loop-related genes in melanoma and hepatocellular carcinoma [24, 25]. Therefore, we proposed that utilizing high-throughput data to explore the relationship between R-loop-related genes and OS would provide valuable insights into the precise role and underlying mechanisms of R-loops in OS.

In this study, we first constructed an R-loop Gene Prognostic Score Model (RGPSM) using machine learning (ML) and identified 10 genes significantly associated with OS prognosis. Patients with high RGPSM scores exhibited poorer prognosis and lower levels of immune infiltration. The scRNA-seq analysis also revealed that cells with high-grade malignancy had elevated RGPSM scores and multiple signal transduction pathways. This suggested that R-loops were closely related to the progression of OS. At the same time, we screened out that PSIP1 was significantly over-expressed in osteoblastic OS cells based on scRNA-seq data. Furthermore, silencing PSIP1 significantly inhibited the proliferation and migration of OS cells, and induced accumulation of R-loops and related DNA damage in OS.

Methods

TARGET data downloaded

The bulk RNA-seq data of 85 OS patients, along with complete survival and prognostic information, was downloaded from the Xena website (https://xenabrowser.net/). The format of gene expression data was converted into Transcripts Per Million (TPM). And the corresponding clinical information of these patients were obtained from the TARGET website (https://www.cancer.gov/ccg/research/genome-sequencing/target).

GEO datasets preparation

To develop and validate the RGPSM, we collected Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/) datasets based on the following criteria: (1) Samples were pathologically diagnosed as OS; (2) The datasets had complete prognostic information and RNA-seq data [22]. And to investigate the expression differences of PSIP1 between OS tissues/cells and normal tissues/cells, we selected GEO datasets that included more than three tumor tissues/cells and adjacent normal tissues or cell lines, ensuring the statistical validity of the analysis. Therefore, the datasets, including GSE21257 (n = 52), GSE33382 (n = 82), GSE16091 (n = 34), GSE126209 (including complete RNA-seq data from 5 adjacent normal tissue [N] and 6 tumor tissue [T]) and GSE42352 (including 3 osteoblasts [OBs], 12 mesenchymal stem cells [MSCs] and 19 OS cell lines [OSs]) were acquired from the GEO using the “GEOquery” (version 2.66.0) R package. Additionally, the survival data of GSE33382 was obtained from R2: Genomics Analysis and Visualization Platform (http://r2.amc.nl) [22]. Then, gene symbols were annotated according to the corresponding GPL platforms. When duplicate gene symbols were annotated, the average expression value was used to represent the expression level.

Integration of datasets

The GSE21257 and GSE16091 were integrated to form the GEO-OS cohort. Batch effect between those two datasets was eliminated by using “ComBat” function in “sva” package (version 3.46.0) [26]. And the GSE21257, GSE16091, GSE33382 and TARGET datasets were merged into Meta-OS dataset for subsequent analyses.

Construction of RGPSM

The RGPSM model was constructed using an ML-based integrative framework comprising 101 algorithms combinations. This ensemble incorporated ten classical ML algorithms: elastic net regularization (Enet), least absolute shrinkage and selection operator (LASSO) regression, gradient-boosted decision trees (GBM), Cox proportional hazards model with adaptive boosting (CoxBoost), survival-optimized support vector machines (Survival-SVM), partial least squares Cox modeling (plsRcox), random survival forests (RSF), supervised principal component analysis (SuperPC), stepwise feature selection in Cox regression (StepCox), along with ridge regression and gradient boosting machine implementations [27]. A total of 1185 R-loop related genes were obtained for analysis (Sup list 1) [28]. Firstly, univariate-COX (uni-COX) regression analysis was performed on the TARGET-OS and GEO-OS datasets to identify the prognostic related genes. The p-value < 0.05 was considered statistically significant and the R-loop related prognostic genes (RPGs) shared in TARGET-OS and GEO-OS datasets were identified. Secondly, 101 ML procedures were trained and evaluated in the three datasets, including TARGET-OS (training dataset), GEO-OS (validation dataset) and GSE33382 (independent validation dataset) to construct the prognostic models [27]. Then, the Harrell’s concordance index (C-index) was used to evaluate the performance of all models [22]. The model with the highest average C-index and the simplest structure was selected as the RGPSM model for further analysis.

Assessing the prognostic performance of the RGPSM

Then, we calculated the risk score of each patient in the Meta-OS, TARGET-OS, GEO-OS, GSE33382, GSE21257 and GSE16091 datasets based on the RGPSM model. The prognostic accuracy of the RGPSM model was evaluated using the time-dependent receiver operating characteristic curve (tROC). And the areas under the ROC curves (AUCs) were calculated. To further validate the prognostic utility of RGPSM, patients in different cohorts were stratified into high- and low- RGPSM according to the optimal cutoff values, which was determined using the “surv_cutpoint” function from the “survminer” package (version 0.5.0). Kaplan-Meier (K-M) plots were then performed to visualize the clinical survival status of patients from different groups.

Valuation of independent predictive capability and construction of nomogram

The uni-COX and multivariate Cox (multi-COX) regression analyses were performed using “survival” package to assess whether RGPSM was an independent prognostic factor. The distribution of RGPSM risk scores across various clinicopathological subgroups was visualized using box plots. Furthermore, ROC curves were generated to evaluate the discriminatory power of the RPGSM scores in predicting Huvos grades. To enhance individualized prognostic assessment, a nomogram model was constructed by integrating RGPSM scores with relevant clinicopathological variables. The predictive accuracy of the nomogram model was assessed with a calibration curve using the “rms” (version 6.8.1) and “regplot” (version 1.1) package, and decision curve analysis (DCA) was conducted using the “ggDCA” function.

Identification of DEGs and functional enrichment analysis

Differentially expressed genes (DEGs) between high- and low-RGPSM patients in TARGET-OS dataset were identified by “limma” package (version 3.54.2). Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) analyses were performed for the functional enrichment analysis. Gene Set Enrichment Analysis (GSEA) was then conducted to decipher the biological function and enriched signaling pathways.

Analysis of tumor immune characteristics, immune checkpoint

Based on the gene expression characteristics of mesenchymal and immune cells, the single sample GSEA (ssGSEA) and Estimation of STromal and Immune cells in MAlignant Tumours using Expression data (ESTIMATE) algorithms were conducted to reveal the differences in TME between the high- and low-score groups in TARGET-OS dataset. In detail, the relative abundance of immune cells in patients was quantified using ssGSEA. Immune-related scores were then calculated using ESTIMATE analysis. And the expression levels of immune checkpoint genes were assessed using student’s t test and visualized using “ggplot2” package (version 3.5.1).

scRNA-seq data preprocessing

The GSE152048 dataset, which includes scRNA-seq data from 11 OS tissues, was obtained from the GEO website and analyzed using the Seurat v5.0 toolkit [29]. To enhance the accuracy of analysis, genes detected in fewer than 3 cells and with fewer than 200 features were initially excluded. We further filtered out cells with fewer than 300 or more than 4000 detected genes, as well as those with over 10% mitochondrial gene content. Additionally, potential doublets were identified and excluded using the “DoubletFinder” package (version 2.0.4) [30]. Ultimately, a total of 102,421 cells were utilized for the bioinformatic analysis. The top 2000 highly variable genes were identified by “FindVariableFeatures” function for principal component analysis (PCA). Subsequently, the batch effects were then mitigated using “harmony” package (version 1.2.3) [31]. Furthermore, “FindNeighbors” and “FindClusters” functions were used for clustering analysis. The t-distributed Stochastic Neighbor Embedding (tSNE) method was then used for visualization. The resolution was identified as 0.3 according to “clustree” method. Cell annotation was conducted using the established cellular markers [29]. Additionally, the “AUCell” package was used for calculating the RGPSM scores for each cell. And then we divided cells into high- and low-RGPSM groups according to threshold of AUCell method. The “monocle 2.0” (version 2.26.0) was applied for pseudotime trajectory analysis during osteoblastic OS development. The “CellChat” package (version 1.6.1) was performed to explore intercellular communication between different RGPSM groups [32, 33]. DEGs between high- and low- RGPSM cells were identified for KEGG and GO analyses.

Cell culture

hFOB1.19 cells and HOS cells were obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). And the U2OS and MG63 cells were purchased from the Procell (Wuhan, China). 143B cells were obtained from FuHeng Biology (Shanghai, China). The cells were cultured with mediums containing 10% fetal bovine serum (FBS) in the 37 °C incubator containing 5% CO2.

Lentivirus production

The shuttle plasmids, including pLV3-U6-PSIP1(human)-shRNA1-CopGFP-Puro (Cat. #P16694, shRNA-target1: GGAAGATACCGACCATGAAGA), pLV3-U6-PSIP1(human)-shRNA2-CopGFP-Puro (Cat. #P49329, shRNA-target2: GGAGTAGTGACAACAGCAACA), pLV3-CMV-PSIP1(human) -3×FLAG-Puro (Cat. #P62069) and packaging plasmids (psPAX2 and PMD2.G) were purchased from MiaoLingBio, China. HEK293T cells were cultured and transfected with the respective plasmids and packaging plasmids. After transfection with 6 h, new medium was replaced. The lentiviral supernatant was collected after 48 h.

Cell proliferation assay

CCK8 assay and colony formation assay were performed to assess cell proliferation abilities. In brief, cells were digested, resuspended and counted. The cell suspension was diluted to 1 × 104 cells/ml. Then, 100 μL of cell suspension was added into 96-well plate or 6-well plate, respectively. The CCK8 kit (Cat. #40203ES76, YEASEN, China) was used for detect the optical density (OD) values of cells in 96-wall plates. And after 10–14 days of culture, cells in 6-well plates were fixed with 4% paraformaldehyde for 15 min, followed by crystal violet staining, and images were captured using a camera and analyzed using image J.

Migration assays

The migration assays were performed using the transwell (Cat. #3422, Corning, the USA). 2 × 104 cells were plated on the upper layer of the chamber without matrix gel, and culture medium containing 20% serum was added to the lower layer. After culturing for 24 h, the chamber was collected. The cells were fixed with paraformaldehyde and stained with crystal violet. After removing the cells on the upper layer, the chambers were photographed. ImageJ software was used to analyze migration and invasion ability.

Western blot assay

Total proteins were extracted using strong RIPA lysis buffer (Cat. #P0013B, Beyotime, China) supplemented with protease inhibitor cocktail (Cat. #HY-K0010, MedChemExpress, China). BCA kit (Cat. #A55864, Thermo Scientific™, the USA) was used for detecting protein concentration. Equal amounts of proteins were separated by SDS-PAGE and then transferred onto PVDF membranes (Miliipore, the USA). Then, the membranes were blocked with 5% nonfat milk and incubated with primary antibody overnight at 4 °C. The antibodies used was as follows: anti-PSIP1 antibody (1:1000, Cat. #25504-1-AP, Proteintech), anti-GAPDH antibody (1:1000, Cat. #10494-1-AP, Proteintech, Wuhan, China). And the membranes were washed with TBST for 3 times followed by incubating with secondary antibodies for 1 h at temperature. After washing with TBST, the membranes were visualized using chemiluminescence kit (Bio-rad, the USA).

Immunofluorescence

The cells were plated and cultured in confocal microplates. After treatments, the cells were washed with PBS, fixed with 4% paraformaldehyde for 15 min and thereby permeabilized with 0.1% Triton X-100 for 10 min. Afterwards, the samples were blocked with 5% BSA at room temperature and then incubated with primary antibody (S9.6, 1:200, Kerafast; Cat. #ENH001, the USA, and γH2AX, 1:200, Sigma, Cat. # 05–636, the USA) for overnight at 4 °C. Then, the cells were washed with TBST for 3 times and incubated with fluorescent secondary antibodies for 2 h at room temperature followed by stained with DAPI for 10 min. Images were observed used the confocal microscopy (STELLARIS 5, Leica, Germany).

Animal experiments

Three-week-old BCLB/C-nude mice were used for the subcutaneous xenograft assays. 1 × 106 143B cells in 100 μL of PBS were injected into the mice subcutaneously. The tumor formation of mice was observed regularly, and the tumor size of mice was measured according to the following formula: Tumor size = (long diameter × short diameter2) / 2.

Hematoxylin-eosin (H&E) staining and immunohistonchemistry (IHC)

The tumor was fixed with 4% paraformaldehyde, then dehydrated, embedded, and sliced. Further H&E staining and IHC experiments were then performed according previous studies [34]. The anti-bodies used in IHC were as followed: anti-PSIP1 antibody (1:100, Cat. #25504-1-AP, Proteintech) and anti-Ki-67 antibody (1:100, Cat. GB111499-100, Servicebio).

Statistical analysis

All experiments were repeated three times independently. All the bioinformatic analyses were performed by R (version 4.2.0). And GraphPad Prism (version 10.0.0) software was used for processing and visualizing the data. Independent sample t-test was used for comparison between two samples, and one-way ANOVA was used for analysis of more than two groups. P < 0.05 was considered statistically significant.

Results

Development and validation of the RGPSM

The RGPSM was constructed following the schematic shown in Figure S1A. Initially, the GSE21257 and GSE16091 datasets were integrated using the “Combat” function form the “sva” R package (Fig. S1B). This merged dataset was named as GEO-OS dataset. Then, we identified 10 RPGs shared in both TARGET-OS and GEO-OS datasets (Fig. S1C), which were subsequently used for ML-based prognostic modeling. All models were then constructed and evaluated across the training set (TARGET-OS), validation set (GEO-OS), and independent validation set (GSE33382). The random survival forest (RSF) model and the stepCox + RSF model achieved the highest average C-index (0.754) across the three datasets (Fig. 1A). The RSF model was selected for constructing the RGPSM due to its simplicity and robust performance (Sup Fig. S1D). Furthermore, we evaluated the prognostic predictive performance of the RGPSM across various datasets. The tROC analysis revealed that the 1-, 3-, and 5-year AUCs in the TARGET-OS dataset were 0.951, 0.977, and 0.984, respectively (Fig. 1B). In the GEO-OS dataset, the AUCs for 1, 3, and 5 years were 0.776, 0.754, and 0.696, respectively (Fig. 1C). In the GSE33382 dataset, the corresponding AUCs were 0.737, 0.726, and 0.701 (Fig. 1D). Patients in different datasets were divided into high- and low-RGPSM groups according the cut-off identified by “surv_cutpoint” function (Fig. S1E). K-M plots demonstrated that patients in the high-RGPSM group had significantly worse progress across the training set (P < 0.0001), validation set (P < 0.0001), and independent test set (P = 0.00015) (Fig. 1E-F). Additionally, we generated tROC curves and K-M plots for the two GEO datasets (GSE21257 and GSE16091) and the Meta-OS dataset (Fig. 1H-K and Fig. S1F-G). These results further confirmed the high robustness of the RGPSM.

Fig. 1
figure 1

Development and validation of the R-loop Gene Prognostic Score Model (RGPSM). (A) C-indices of 101 combinations of multiple machine-learning in three cohorts. (B-D and H-I) Time-dependent ROC curve analysis of the RGPSM in different cohorts. (E-G and J-K) Kaplan-Meier survival analyses of RGPSM in different cohorts

Full size image

Independent prognostic evaluation and nomogram model construction

Subsequently, we evaluated the relationships between clinicopathological characteristics of OS patients and the RGPSM scores. Uni-COX regression analysis was firstly performed in the Meta-OS dataset to identify the prognostic related parameters. The results indicated that the RGPSM, Huvos grade, age, metastasis status and primary tumor site were associated with the patients’ overall survival (Fig. 2A). Subsequently, we analyzed the RGPSM scores in different clinical subgroups. Notably, higher Huvos grade is generally correlated with better response to chemotherapy in OS patients [35]. In the Meta-OS and GEO-OS datasets, the RGPSM scores of patients in the Huvo III/IV group were lower than the patients in the Huvos I/II group (Fig. 2B). The results in TARGET-OS and GSE33382 datasets were consistent although without statistically different (Fig. 2B). Patients with metastases showed higher RGPSM scores (Fig. 2C). And the tROC demonstrated that the RGPSM was effective in predicting the Huvos grades of OS patients (Fig. 2D). These findings suggested that RGPSM scores might be linked to chemotherapy resistance and metastasis in OS. Additionally, there were no differences in RGPSM scores between the patients with different age or tumor location in all cohorts (Fig. S2A-B). Next, multi-COX regression analysis was conducted to assess the independent prognostic value of these factors identified in the uni-COX analysis. Patients with missing data were excluded, and one patient with a primary axial tumor was excluded due to small sample size limitations. Ultimately, a total of 128 patients were included in the analysis. The results indicated that the RGPSM, metastasis status and huvos grades were the independent predictive factors for predicting the prognosis of OS patients (Fig. 2E). Meanwhile, to enhance the predictive performance of RGPSM, we constructed a nomogram incorporating RGPSM scores along with other clinicopathological parameters (Fig. 2F). The ROC curves demonstrated the strong predictive capability of this nomogram model, with AUCs of 0.944, 0.904, and 0.933 for 1-year, 3-year, and 5-year predictions, respectively (Fig. 2G). Moreover, the AUCs of nomogram were significantly higher than RGPSM and other clinical parameters (Fig. S2C). The calibration curve of the nomogram closely aligned with the ideal diagonal line (Fig. 2H), and the decision curve demonstrated that the model provided a substantial net benefit (Fig. 2I).

Fig. 2
figure 2

Independent prognostic evaluation and nomogram model construction. (A) Uni-COX analysis of RGPSM and clinical features in predicting the OS prognosis. (B) The RGPSM scores in different huvos grades (I/II and III/IV). (C) The RGPSM scores in different metastasis status (M0, without metastasis; M1, with metastasis). (D) The tROC analyses of RGPSM in predicting the Huvos grades. (E) Muti-COX analysis of RGPSM and clinical feratures in predicting the OS prognosis. (F) The nomogram model constructed by incorporating RGPSM and clinical factors. (G-I) The tROC analyses (G), calibration curve (H) and decision curve (I) of nomogram in predicting prognosis. *, p < 0.05, **, p < 0.01, ***, p < 0.001

Full size image

Functional enrichment and immune analyses

To elucidate the regulatory role of R-loops in OS, we identified the DEGs between high- and low-RGPSM groups in the TARGET-OS dataset. A total of 980 differentially DEGs were identified, including 478 upregulated and 502 downregulated genes (Fig. 3A, fold change > 1.2, p < 0.05). Subsequent KEGG and GO enrichment analyses highlighted significant enrichment in pathways associated with tumor progression and immune regulation. Key pathways included the regulation of cell-cell adhesion, T cell activation, negative regulation of cell adhesion, T cell differentiation, Wnt signaling pathway, and Th17 cell differentiation (Fig. S3A-B). Further analyses using GSEA-KEGG, GSEA-GO, and GSEA-Reactome also revealed marked differences in numerous immune regulation-related pathways between the two groups (Fig. 3B-C and Sup. Fig. S3C). Given these findings, we further assessed the immune infiltration between the high- and low-RGPSM groups. Patients in the high-RGPSM group showed higher tumor purity scores but lower ESTIMATE, Immune, and Stromal scores (Fig. 3D). Moreover, an increase in RGPSM scores was associated with a decrease in immune-related scores (Fig. 3E). Concurrently, we analyzed the differences in immune cell populations and immune checkpoint-related gene expression levels between the two groups. The results demonstrated a significant reduction in several tumor-killing immune cell types, including activated B cells, activated CD8+ T cells, and natural killer T cells, in the high-RGPSM group (Fig. 3F). Additionally, three key immunosuppressive checkpoint genes were found to be downregulated in the high RGPSM group (Fig. S3D). In conclusion, R-loops might play a crucial role in shaping the TME, potentially influencing immune evasion and tumor progression.

Fig. 3
figure 3

Functional enrichment and immune analysis. (A) Different expressed genes (DEGs) between the high- and low-RGOSM groups in the TARGET-OS dataset. (B-C) GSEA-KEGG (B) and GSEA-GO (C) analyses of DEGs. (D-E) ESTIMATE analysis of different RGPSM groups. (F) Immune cell populations of different RGPSM groups

Full size image

RGPSM-based scRNA-seq data analysis

The scRNA-seq data from 11 OS patients (GSE152048) was downloaded to investigate the role of R-loop-related genes in OS. After batch correction using the “harmony” algorithm, dimensionality reduction and clustering were performed using PCA, t-SNE, and UMAP, revealing no significant batch effects among the patients (Fig. S4A). Cells were then identified as 13 clusters at a resolution of 0.3 (Fig. S4B-C). The cell types were then annotated as chondroblastic cells, myeloid cells, osteoblastic cells, cancer-associated fibroblasts (CAFs), proliferative osteoblastic cells (osteoblastic_proli), osteoclasts, tumor-infiltrating lymphocytes (TILs), endothelial cells, and pericytes (Fig. 4A-B). The expression levels of marker genes were visualized using dot plots and violin plots (Fig. 4B and Fig. S4D). Additionally, the RGPSM model-related genes were then used to score all cells via the “AUCell” package for further analysis (Fig. 4C). According to the optimal cutoff value calculated by the ‘AUCell_exploreThresholds’ function, the cells were then divided into high- and low-RPGSM score groups (Fig. S4E). Analysis of cellular composition revealed a significantly higher proportion of malignant cells (chondroblastic, osteoblastic, osteoblastic_proli, and osteoclast cells) in the high-RGPSM group, whereas the proportion of immune cells was markedly lower (Fig. 4D). Given the high malignancy of osteoblastic OS cells, we conducted pseudotime trajectory analysis to explore their differentiation dynamics. As osteoblastic OS cells differentiated into proliferative osteoblastic OS cells, the RGPSM scores significantly increased, suggesting that R-loops might be associated with the malignant progression of OS cells (Fig. 4E). Meanwhile, we analyzed the DEGs between the two groups of cells. KEGG and GO analyses were performed and revealed numerous differences in immune-related pathways between the two groups of cells, which is consistent with the previous analysis results (Fig. S5A-B). In addition, we used “CellChat” to analyze the cell signaling communication within each group. In general, the high-RGPSM group exhibited both stronger and more frequent cell-cell signaling interactions compared to the low-RGPSM group (Fig. 4F and G). We also identified the top 16 signaling pathways with the highest information flow based on statistical significance (Fig. S5C). Significant differences in the intensity and directionality of these signaling pathways, encompassing both incoming and outgoing signaling patterns, were observed among various cell types between the high-RGPSM and low-RGPSM score groups (Fig. 4H).

Fig. 4
figure 4

RGPSM-based scRNA data analysis. (A) Cell types identified in scRNA-seq. (B) Dot plot of markers used in cell annotation. (C) The tSNE plot to display the RGPSM scores of cells. (D) The cell proportions in different RGPSM scoring groups. (E) The pseudotime analysis and RGPSM scores of osteoblastic and osteoblastic_proli cells. (F) The number of inferred interactions between two RGPSM scoring groups. (G) Cell communication weight map of each cell in different RGPSM scoring groups, and the size of the dots and the thickness of the lines represent the communication intensity. (H) The incoming and outcoming signaling patterns in different RGPSM scoring groups

Full size image

Identifying therapeutic targets for OS patients

To identify potential R-loop-associated therapeutic targets in OS, we examined the expression levels of ten RGPSM model genes in both bulk RNA-seq and scRNA-seq datasets. In the bulk RNA-seq analysis, a comparison between high- and low-RGPSM score groups revealed that RBM34, BLOC1S1, DDX21, PSIP1, RPL10A, and RPS27A were significantly upregulated in the high-RGPSM group (Fig. 5A). Subsequently, we analyzed the single-cell expression patterns of these six upregulated genes across different cell types. The results demonstrated that RBM34 and NOLC1 were expressed at low levels in OS cells (Fig. 5B). Conversely, DDX21, RPL10A, and RPS27A exhibited high expression across various cell types, including tumor and immune cells, indicating a lack of specificity for OS cells (Fig. 5B). Notably, PSIP1 was found to be highly expressed specifically in osteoblastic and proliferative osteoblastic OS cells, suggesting its potential as a therapeutic target (Fig. 5B). Moreover, high expression levels of PSIP1 were significantly associated with poor prognosis in OS patients across different datasets (Fig. 5C). Further analysis of PSIP1 mRNA expression between OS cells/samples and normal cells/samples using GEO datasets revealed that PSIP1 was highly expressed in OS cells and tissues (Fig. 5D). Additionally, PSIP1 protein (including the p75 and p52 isoforms) levels were markedly upregulated in OS cells compared to osteoblasts (hFOB1.19 cells) (Fig. 5E).

Fig. 5
figure 5

Identifying therapeutic targets for OS patients. (A) The heatmap of ten RGPSM-related genes in the high- and low-RGPSM patients. (B) The feature plots of six high expressed genes. (C) The K-M plots of PSIP1 in predicting the prognosis of OS patients in various cohorts. (D) The PSIP1 mRNA expression levels between normal samples (normal, N)/cells (osteoblast, OB; mesenchymal stem cell, MSC) and tumor (T)/OS cells. (E) The PSIP1 protein levels in hFOB1.19 and OS cells. *, p < 0.05; **, p < 0.01

Full size image

Silencing PSIP1 inhibited proliferation, invasion and migration of OS

To explore the functional role of PSIP1 in the proliferation and migration of OS cells, we used lentivirus to silence and overexpress PSIP1 in 143B and HOS cells (Fig. S6). The results of CCK-8 and colony formation assays showed that knockdown of PSIP1 significantly inhibited cell proliferation in both 143B and HOS cells, while overexpression of PSIP1 enhanced OS cell proliferation (Fig. 6A-B). Additionally, knockdown of PSIP1 impaired the migration ability of OS cells, as demonstrated by transwell assays (Fig. 6C). Furthermore, in vivo experiments indicated that xenograft tumors in the sh-PSIP1 groups grew more slowly than those in the negative control group, with a significant reduction in tumor size and weight (Fig. 6D-F). Moreover, PSIP1 inhibition led to a significant reduction in Ki-67 protein expression, a marker of proliferation (Fig. 6G). These findings suggested that PSIP1 might be a potential therapeutic target for OS patients.

Fig. 6
figure 6

Silencing PSIP1 inhibited proliferation, invasion and migration of OS. (A-B) CCK8 (A) and colonies plate information assays (B) were used to measure the effects of silencing or up-regulating PSIP1 in OS cells. (C) Transwell migration assays were used to detect the migratory ability of OS cells with silencing or up-regulating PSIP1. (D) The subcutaneous tumor morphology after knockdown of PSIP1. (E-F) The size (E) and weight (F) of tumors. (G) The HE results and IHC results of PSIP1 and Ki-67 of subcutaneous tumors

Full size image

Silencing PSIP1 induced R-loop accumulation and DNA damage

Dysregulation of R-loop homeostasis has been implicated in the progression of various tumors. Jayakumar S et al. found that PSIP1 directly interacted with R-loops and participated into their resolution, thereby inhibiting the R-loop-associated DNA damage regulation [36]. We then examined the levels of R-loops in OS cells with up-regulating or silencing PSIP1. Knockdown of PSIP1 led to an increase of R-loops, while overexpression of PSIP1 significantly reduced cisplatin (CDDP)-induced R-loop accumulation (Fig. 7A). Meanwhile, we observed an elevated γ-H2AX level in the cells with knocking down of PSIP1 (Fig. 7B). And the overexpression of PSIP1 alleviated CDDP-induced increase in γ-H2AX fluorescence intensity (Fig. 7B), suggesting that silencing PSIP1 might inhibit OS proliferation, migration and invasion by inducing R-loop accumulation and its associated DNA damage.

Fig. 7
figure 7

Silencing PSIP1 induced R-loop accumulation and DNA damage. (A) The fluorescence intensity of S9.6 after different treatments. (B) The relative γ-H2AX foci intensity (γ-H2AX/DAPI) after different treatments. *, p < 0.05; **, p < 0.01

Full size image

Discussion

Over the past four decades, significant progress has been made in OS research. However, the intrinsic heterogeneity of OS limits the effectiveness of current treatment strategies. While surgery combined with chemotherapy remains the standard treatment strategy for early-stage OS patients, but they often fail to achieve favorable outcomes in patients with advanced-stage disease [37]. Nowadays, the major clinical challenge in OS management is the lack of reliable biomarkers for early detection and personalized treatment for OS patients [38]. Recent studies have highlighted the critical role of R-loops in tumor progression. Aberrant R-loop accumulation or defects in their resolution have been linked to an increased risk of cancer development and progression. Besides, R-loop scores have been identified as promising prognostic biomarkers, predicting patient survival and treatment response in lung cancer [23]. Despite these insights, the role of R-loops in OS remains poorly understood. Whether R-loop-related genes could serve as reliable prognostic markers and therapeutic targets for OS patients remain to be further studied.

In this study, we developed an R-loop-related prognostic models using the 101 combinations of ML algorithms across three independent datasets. A ten-gene RGPSM was constructedusing the RSF algorithm, which was selected for its highest average C-index (0.754) and relatively simple structure. The model demonstrated robust predictive performance for patient survival and chemotherapy response, with higher RGPSM scores associated with poorer survival outcomes and reduced chemotherapy sensitivity. Moreover, the nomogram intuitively demonstrated the strong concordance between RGPSM-predicted survival probabilities and actual patient survival rates. In addition, scRNA-seq data further revealed that highly malignant OS cells exhibited increased RGPSM scores, and these scores progressively increased as the osteoblastic OS cells differentiating into proliferative osteoblastic OS cells. These findings highlighted the potential of R-loop-related genes as biomarkers for risk assessment and as guiding factors for personalized treatment strategies in OS patients.

We conducted a further analysis of DEGs between the high- and low-RGPSM patients, revealing significant differences in immunity-related pathways. The TME is a dynamic and complex system composed of various cellular components, including stromal cells, endothelial cells, immune cells, and non-cellular components [39]. Interactions of cells in the TME are crucial for OS progression, metastasis, and chemoresistance. Among the immune cells in the TME, regulatory T cells (Tregs) play vital roles in maintaining immune homeostasis and inhibiting excessive immune activation, potentially contributing to tumor progression [40]. In contrast, CD8+ T cells and NK cells are capable of recognizing and eliminating tumor cells by producing various cytokines and chemokines [41, 42]. Recent studies have increasingly suggested that R-loops are involved in the regulation of the TME. Aberrant R-loop formation can increase the proportion of CD8 + T cells in the TME, thereby enhancing the anti-tumor immune response through activation of the STING pathway [43]. Additionally, RNA-DNA hybrids generated by R-loops are immunogenic, boosting innate immunity and being closely associated with tumor immunity [44]. In our study, tumor purity was positively correlated with high RGPSM scores, while immune scores were decreased. The high-RGPSM scores were associated with the exhaustion of anti-cancer immune cells, such as CD8+ T cell, memory T cell and NK cells. In the scRNA-seq data, the proportion of TILs was higher in the low-RGPSM group. In patients with high RGPSM scores, the TME appeared to be more inclined to be suppressed, thereby promoting survival and proliferation of OS cells. Therefore, the status of R-loops might be useful to assess the immune landscape in OS patients, providing valuable insights for guiding tumor treatment strategies [23].

Apart from introducing an RPGSM for stratifying high-risk OS patients, we also identified PSIP1 as a potential therapeutic target for OS. PSIP1 is a chromatin-associated protein that functions as an RNA-binding protein, participates in RNA splicing and transcriptional regulation [45]. The PSIP1 gene encodes two protein isoforms, PSIP1/p75 and PSIP1/p52, which share 325 amino acids at their N-terminus, including the critical PWWP domain [46]. The PWWP domain of PSIP1 can bind to free DNA and mediates chromatin association, with other proteins interacting with PSIP1 to participate in or regulated chromatin association [47]. Previous studies shows that PSIP1 is highly expressed in multiple cancers, functioning as a proto-oncogene that promotes cell proliferation, survival and chemotherapy resistance [48,49,50]. In our study, we found the mRNA expression level of PSIP1 was significantly upregulated in OS cells/tissues, as evidenced by scRNA-seq analysis and GEO dataset validation. Moreover, the protein levels of PSIP1 (including p75 and p52 isoforms) in OS cells were significantly higher than in hFOB1.19 cells. The results were consistent with previous studies [51]. Functionally, silencing PSIP1 significantly inhibited OS cell proliferation and migration, whereas overexpression of PSIP1 enhanced these processes, confirming its pro-tumorigenic role in OS. Meanwhile, overexpression of PSIP1 cloud enhance the resolution of R-loops and decrease associated DNA damage, that might be primarily attributed to the PSIP1/p75 isoform [36].

To sum up, we developed an RGPSM model using ML methods to classify high-risk subsets of OS patients in three independent datasets. By integrating ML, scRNA-seq analysis, and experimental validation, we identified PSIP1 as a regulator of R-loop resolution that promoted OS progression, highlighting it as a potential therapeutic target.

Data availability

Bioinformatic data analyzed in this study can be downloaded from public databases. Original data are available in the additional files.

References

  1. Crossley MP, Bocek M, Cimprich KA. R-loops as cellular regulators and genomic threats. Mol Cell. 2019;73(3):398–411.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  2. Crossley MP, Song C, Bocek MJ, Choi JH, Kousouros JN, Sathirachinda A, Lin C, Brickner JR, Bai G, Lans H, et al. R-loop-derived cytoplasmic RNA-DNA hybrids activate an immune response. Nature. 2023;613(7942):187–94.

    Article  CAS  PubMed  Google Scholar 

  3. Xu Y, Jiao Y, Liu C, Miao R, Liu C, Wang Y, Ma C, Liu J. R-loop and diseases: the cell cycle matters. Mol Cancer. 2024;23(1):84.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  4. Chakraborty P. New insight into the biology of R-loops. Mutat Res. 2020;821:111711.

    Article  CAS  PubMed  Google Scholar 

  5. Chakraborty P, Huang JTJ, Hiom K. DHX9 helicase promotes R-loop formation in cells with impaired RNA splicing. Nat Commun. 2018;9(1):4346.

    Article  PubMed  PubMed Central  Google Scholar 

  6. Li X, Manley JL. Inactivation of the SR protein splicing factor ASF/SF2 results in genomic instability. Cell. 2005;122(3):365–78.

    Article  CAS  PubMed  Google Scholar 

  7. Elsakrmy N, Cui H. R-loops and R-loop-binding proteins in cancer progression and drug resistance. Int J Mol Sci. 2023;24(8).

  8. Santos-Pereira JM, Aguilera A. R loops: new modulators of genome dynamics and function. Nat Rev Genet. 2015;16(10):583–97.

    Article  CAS  PubMed  Google Scholar 

  9. Huang TT, Chiang CY, Nair JR, Wilson KM, Cheng K, Lee JM. AKT1 interacts with DHX9 to mitigate R loop-induced replication stress in ovarian cancer. Cancer Res. 2024;84(6):887–904.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  10. Wang D, Tang M, Zhang P, Yang K, Huang L, Wu M, Shen Q, Yue J, Wang W, Gong Y, et al. Activation of ERβ hijacks the splicing machinery to trigger R-loop formation in triple-negative breast cancer. Proc Natl Acad Sci U S A. 2024;121(13):e2306814121.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  11. Ritter J, Bielack SS. Osteosarcoma. Ann Oncol. 2010;21(Suppl 7):vii320–325.

    Article  PubMed  Google Scholar 

  12. Wittig JC, Bickels J, Priebat D, Jelinek J, Kellar-Graney K, Shmookler B, Malawer MM. Osteosarcoma: a multidisciplinary approach to diagnosis and treatment. Am Fam Physician. 2002;65(6):1123–32.

    PubMed  Google Scholar 

  13. Isakoff MS, Bielack SS, Meltzer P, Gorlick R. Osteosarcoma: current treatment and a collaborative pathway to success. J Clin Oncol. 2015;33(27):3029–35.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  14. Abedi S, Behmanesh A, Mazhar FN, Bagherifard A, Sami SH, Heidari N, Hossein-Khannazer N, Namazifard S, Kazem Arki M, Shams R, et al. Machine learning and experimental analyses identified MiRNA expression models associated with metastatic osteosarcoma. Biochim Biophys Acta Mol Basis Dis. 2024;1870(7):167357.

    Article  CAS  PubMed  Google Scholar 

  15. Jiang YC, Xu QT, Wang HB, Ren SY, Zhang Y. A novel prognostic signature related to programmed cell death in osteosarcoma. Front Immunol. 2024;15:1427661.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  16. Li S, Zheng Z, Wang B. Machine learning survival prediction using tumor lipid metabolism genes for osteosarcoma. Sci Rep. 2024;14(1):12934.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  17. Wang T, Wang L, Zhang L, Long Y, Zhang Y, Hou Z. Single-cell RNA sequencing in orthopedic research. Bone Res. 2023;11(1):10.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  18. Wu C, Gong S, Duan Y, Deng C, Kallendrusch S, Berninghausen L, Osterhoff G, Schopow N. A tumor microenvironment-based prognostic index for osteosarcoma. J Biomed Sci. 2023;30(1):23.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  19. Zhang N, Han Y, Cao H, Wang Q. Inflammasome-related gene signatures as prognostic biomarkers in osteosarcoma. J Cell Mol Med. 2024;28(9):e18286.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  20. Cheng D, Zhang Z, Mi Z, Tao W, Liu D, Fu J, Fan H. Deciphering the heterogeneity and immunosuppressive function of regulatory T cells in osteosarcoma using single-cell RNA transcriptome. Comput Biol Med. 2023;165:107417.

    Article  CAS  PubMed  Google Scholar 

  21. Nie J, He C, Shu Z, Liu N, Zhong Y, Long X, Liu J, Yang F, Liu Z, Huang P. Identification and experimental validation of Stearoyl-CoA desaturase is a new drug therapeutic target for osteosarcoma. Eur J Pharmacol. 2024;963:176249.

    Article  CAS  PubMed  Google Scholar 

  22. Wang Y, Ma X, Xu E, Huang Z, Yang C, Zhu K, Dong Y, Zhang C. Identifying squalene epoxidase as a metabolic vulnerability in high-risk osteosarcoma using an artificial intelligence-derived prognostic index. Clin Transl Med. 2024;14(2):e1586.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  23. Zhang S, Liu Y, Sun Y, Liu Q, Gu Y, Huang Y, Zeng Z, Tang F, Ouyang Y. Aberrant R-loop-mediated immune evasion, cellular communication, and metabolic reprogramming affect cancer progression: a single-cell analysis. Mol Cancer. 2024;23(1):11.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  24. Ouyang Y, Gu Y, Li S, Wei X, Liu Y, Wang Z, Tang F, Zhang S. Dysregulation of R-loop homeostasis shapes the immunosuppressive microenvironment and induces malignant progression in melanoma. Apoptosis. 2025;30(1–2):131–48.

    Article  CAS  PubMed  Google Scholar 

  25. Chen L, Yang H, Wei X, Liu J, Han X, Zhang C, Liu Y, Zhang Y, Xu Y, Li Y, et al. Integrated single-cell and bulk transcriptome analysis of R-loop score-based signature with regard to immune microenvironment, lipid metabolism and prognosis in HCC. Front Immunol. 2024;15:1487372.

    Article  CAS  PubMed  Google Scholar 

  26. Leek JT, Johnson WE, Parker HS, Jaffe AE, Storey JD. The Sva package for removing batch effects and other unwanted variation in high-throughput experiments. Bioinformatics. 2012;28(6):882–3.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  27. Liu Z, Liu L, Weng S, Guo C, Dang Q, Xu H, Wang L, Lu T, Zhang Y, Sun Z, et al. Machine learning-based integration develops an immune-derived LncRNA signature for improving outcomes in colorectal cancer. Nat Commun. 2022;13(1):816.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  28. Lin R, Zhong X, Zhou Y, Geng H, Hu Q, Huang Z, Hu J, Fu XD, Chen L, Chen JY. R-loopBase: a knowledgebase for genome-wide R-loop formation and regulation. Nucleic Acids Res. 2022;50(D1):D303–15.

    Article  CAS  PubMed  Google Scholar 

  29. Zhou Y, Yang D, Yang Q, Lv X, Huang W, Zhou Z, Wang Y, Zhang Z, Yuan T, Ding X, et al. Single-cell RNA landscape of intratumoral heterogeneity and immunosuppressive microenvironment in advanced osteosarcoma. Nat Commun. 2020;11(1):6322.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  30. McGinnis CS, Murrow LM, Gartner ZJ. DoubletFinder: doublet detection in single-cell RNA sequencing data using artificial nearest neighbors. Cell Syst. 2019;8(4):329–e337324.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  31. Korsunsky I, Millard N, Fan J, Slowikowski K, Zhang F, Wei K, Baglaenko Y, Brenner M, Loh PR, Raychaudhuri S. Fast, sensitive and accurate integration of single-cell data with harmony. Nat Methods. 2019;16(12):1289–96.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  32. Qiu X, Mao Q, Tang Y, Wang L, Chawla R, Pliner HA, Trapnell C. Reversed graph embedding resolves complex single-cell trajectories. Nat Methods. 2017;14(10):979–82.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  33. Jin S, Guerrero-Juarez CF, Zhang L, Chang I, Ramos R, Kuan CH, Myung P, Plikus MV, Nie Q. Inference and analysis of cell-cell communication using cellchat. Nat Commun. 2021;12(1):1088.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  34. Wang C, Yang X, Wang T, Wang Y, Shi J, Sun Q, Wang Y, Wu H. Construction of hematoxylin-eosin, immunohistochemistry, and EBER-ISH methodology after trichloroisocyanuric acid treatment in melanin-containing tissues. Sci Rep. 2022;12(1):16062.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  35. Meyers PA, Schwartz CL, Krailo M, Kleinerman ES, Betcher D, Bernstein ML, Conrad E, Ferguson W, Gebhardt M, Goorin AM, et al. Osteosarcoma: a randomized, prospective trial of the addition of ifosfamide and/or muramyl tripeptide to cisplatin, doxorubicin, and high-dose methotrexate. J Clin Oncol. 2005;23(9):2004–11.

    Article  CAS  PubMed  Google Scholar 

  36. Jayakumar S, Patel M, Boulet F, Aziz H, Brooke GN, Tummala H, Pradeepa MM. PSIP1/LEDGF reduces R-loops at transcription sites to maintain genome integrity. Nat Commun. 2024;15(1):361.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  37. Zhu N, Ni H, Guo S, Shen YQ, Chen Q. Bone complications of cancer treatment. Cancer Treat Rev. 2024;130:102828.

    Article  CAS  PubMed  Google Scholar 

  38. Ma Q, Sun J, Liu Q, Fu J, Wen Y, Zhang F, Wu Y, Zhang X, Gong L, Zhang W. Identification of a biomarker to predict doxorubicin/cisplatin chemotherapy efficacy in osteosarcoma patients using primary, recurrent and metastatic specimens. Transl Oncol. 2024;49:102098.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  39. Gajewski TF, Schreiber H, Fu YX. Innate and adaptive immune cells in the tumor microenvironment. Nat Immunol. 2013;14(10):1014–22.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  40. Ji D, Song C, Li Y, Xia J, Wu Y, Jia J, Cui X, Yu S, Gu J. Combination of radiotherapy and suppression of Tregs enhances abscopal antitumor effect and inhibits metastasis in rectal cancer. J Immunother Cancer. 2020;8(2).

  41. van der Leun AM, Thommen DS, Schumacher TN. CD8(+) T cell states in human cancer: insights from single-cell analysis. Nat Rev Cancer. 2020;20(4):218–32.

    Article  PubMed  PubMed Central  Google Scholar 

  42. Philip M, Schietinger A. CD8(+) T cell differentiation and dysfunction in cancer. Nat Rev Immunol. 2022;22(4):209–23.

    Article  CAS  PubMed  Google Scholar 

  43. Maxwell MB, Hom-Tedla MS, Yi J, Li S, Rivera SA, Yu J, Burns MJ, McRae HM, Stevenson BT, Coakley KE, et al. ARID1A suppresses R-loop-mediated STING-type I interferon pathway activation of anti-tumor immunity. Cell. 2024;187(13):3390–e34083319.

    Article  CAS  PubMed  Google Scholar 

  44. Cytoplasmic R-loop, Processing-Derived. RNA-DNA hybrids are immunogenic. Cancer Discov. 2023;13(3):526.

    Article  Google Scholar 

  45. Ge H, Si Y, Roeder RG. Isolation of cDNAs encoding novel transcription coactivators p52 and p75 reveals an alternate regulatory mechanism of transcriptional activation. EMBO J. 1998;17(22):6723–9.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  46. Baude A, Aaes TL, Zhai B, Al-Nakouzi N, Oo HZ, Daugaard M, Rohde M, Jäättelä M. Hepatoma-derived growth factor-related protein 2 promotes DNA repair by homologous recombination. Nucleic Acids Res. 2016;44(5):2214–26.

    Article  CAS  PubMed  Google Scholar 

  47. Pradeepa MM, Grimes GR, Taylor GC, Sutherland HG, Bickmore WA. Psip1/Ledgf p75 restrains hox gene expression by recruiting both trithorax and polycomb group proteins. Nucleic Acids Res. 2014;42(14):9021–32.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  48. Singh DK, Gholamalamdari O, Jadaliha M, Ling Li X, Lin YC, Zhang Y, Guang S, Hashemikhabir S, Tiwari S, Zhu YJ, et al. PSIP1/p75 promotes tumorigenicity in breast cancer cells by promoting the transcription of cell cycle genes. Carcinogenesis. 2017;38(10):966–75.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  49. Chan TS, Hawkins C, Krieger JR, McGlade CJ, Huang A. JPO2/CDCA7L and LEDGF/p75 are novel mediators of PI3K/AKT signaling and aggressive phenotypes in medulloblastoma. Cancer Res. 2016;76(9):2802–12.

    Article  CAS  PubMed  Google Scholar 

  50. El Ashkar S, Schwaller J, Pieters T, Goossens S, Demeulemeester J, Christ F, Van Belle S, Juge S, Boeckx N, Engelman A, et al. LEDGF/p75 is dispensable for hematopoiesis but essential for MLL-rearranged leukemogenesis. Blood. 2018;131(1):95–107.

    PubMed  PubMed Central  Google Scholar 

  51. Fan Q, Wang Y, Cheng J, Pan B, Zang X, Liu R, Deng Y. Single-cell RNA-seq reveals T cell exhaustion and immune response landscape in osteosarcoma. Front Immunol. 2024;15:1362970.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

Download references

Acknowledgements

None.

Funding

This work was supported by the National Natural Science Foundation of China (NO. 82360543), Jiangxi Provincial Natural Science Foundation (NO. 20232ACB206043), Jiangxi Province Postgraduate Innovation Special Fund Project (NO. YC2023-B067).

Author information

Author notes

  1. Jiangbo Nie and Shijiang Wang contributed equally to this work.

Authors and Affiliations

  1. Department of Orthopedic Surgery, The First Hospital of Nanchang, The Third Affiliated Hospital of Nanchang University, Nanchang, China

    Jiangbo Nie & Zhili Liu

  2. Department of Orthopedic Surgery, The First Affiliated Hospital, Jiangxi Medical College, Nanchang University, Nanchang, China

    Jiangbo Nie, Shijiang Wang, Yanxin Zhong, Feng Yang, Jiaming Liu & Zhili Liu

  3. Jiangxi Provincial Key Laboratory of Spine and Spinal Cord Diseases, Nanchang, China

    Jiangbo Nie, Shijiang Wang, Yanxin Zhong, Feng Yang, Jiaming Liu & Zhili Liu

Authors
  1. Jiangbo Nie
    View author publications

    You can also search for this author inPubMed Google Scholar

  2. Shijiang Wang
    View author publications

    You can also search for this author inPubMed Google Scholar

  3. Yanxin Zhong
    View author publications

    You can also search for this author inPubMed Google Scholar

  4. Feng Yang
    View author publications

    You can also search for this author inPubMed Google Scholar

  5. Jiaming Liu
    View author publications

    You can also search for this author inPubMed Google Scholar

  6. Zhili Liu
    View author publications

    You can also search for this author inPubMed Google Scholar

Contributions

NJB, LZL and WSJ: conception and design. NJB, WSJ, YF, LJM and LZL: review and revision of the manuscript. NJB, WSJ and ZYX: investigation, visualization and experiments. NJB: writing and bioinformatic analysis. NJB, YF and LJM: funding. All authors reviewed the manuscript.

Corresponding author

Correspondence to Zhili Liu.

Ethics declarations

Ethics approval and consent to participate

Animal experiments were in compliance with ethical standards and had been ethically approved (CDYFY-IACUC-202407QR005).

Consent for publication

The authors declare no competing interests.

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary Material 1

Supplementary Material 2

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Nie, J., Wang, S., Zhong, Y. et al. Identifying PSIP1 as a critical R-loop regulator in osteosarcoma via machine-learning and multi-omics analysis. Cancer Cell Int 25, 159 (2025). https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12935-025-03775-1

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12935-025-03775-1

Keywords

Cancer Cell International

ISSN: 1475-2867

Contact us