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

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.

Keywords: Machine-learning; Multi-omics analysis; Osteosarcoma; PSIP1; R-loop.

Fig. 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

Fig. 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

Fig. 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

Fig. 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

Fig. 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

Fig. 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

Fig. 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