Background: Chemoresistance is one of the leading causes that severely limits the success of osteosarcoma treatment. Evaluating chemoresistance before chemotherapy poses a new challenge for researchers. We established an effective chemoresistance risk scoring model for prechemotherapy osteosarcoma using single-cell sequencing.
Methods: We comprehensively analyzed osteosarcoma data from the bulk mRNA sequencing dataset TARGET-OS and the single-cell RNA sequencing (scRNA-seq) dataset GSE162454. Chemoresistant tumor clusters were identified using enrichment analysis and AUCell scoring. Its differentiated trajectory was achieved with inferCNV and pseudotime analysis. Ligand-receptor interactions were annotated with iTALK. Furthermore, we established a chemoresistance risk scoring model using LASSO regression based on scRNA-seq-based markers of chemoresistant tumor clusters. The TARGET-OS dataset was used as the training group, and the bulk mRNA array dataset GSE33382 was used as the validation group. Finally, the performance was verified for its discriminatory ability and calibration.
Results: Using bulk RNA data, we found that osteogenic expression was upregulated in chemoresistant osteosarcoma as compared to chemosensitive osteosarcoma. Then, we transferred the bulk RNA findings to scRNA-seq and noticed osteosarcoma tumor clusters C14 and C25 showing osteogenic cancer stem cell expression patterns, which fit chemoresistant characteristics. C14 and C25 possessed bridge roles in interactions with other clusters. On the one hand, they received various growth factor stimulators and could potentially transform into a proliferative state. On the other hand, they promote local tumor angiogenesis, bone remodeling and immunosuppression. Next, we identified a ten-gene signature from the C14 and C25 markers and constructed a chemoresistant risk scoring model using LASSO regression model. Finally, we found that chemoresistant osteosarcoma had higher chemoresistance risk score and that the model showed good discriminatory ability and calibration in both the training and validation groups (AUCtrain = 0.82; AUCvalid = 0.84). Compared with that of the classic bulk RNA-based model, it showed more robust performance in validation environment (AUCvalid-scRNA = 0.84; AUCvalid-bulk DEGs = 0.54).
Conclusions: Our work provides insights into understanding chemoresistant osteosarcoma tumor cells and using single-cell sequencing to establish a chemoresistance risk scoring model. The model showed good discriminatory ability and calibration and provided us with a feasible way to evaluate chemoresistance in prechemotherapy osteosarcoma.
Keywords: chemoresistance; heterogeneity; osteosarcoma; single-cell RNA sequencing; stemness.
Copyright © 2022 Zeng, Li, Zhang, Zhang, Jiang, Guo, Wang, Yang, Yan, Zhang, Wang, Huang, Zhao, Li, Hu and Gao.