Integrated analysis of single-cell and bulk transcriptomics reveals cellular subtypes and molecular features associated with osteosarcoma prognosis.

Abstract

Background: Osteosarcoma (OS) is the most common primary bone malignancy with variable molecular biology and prognosis. However, our understanding of the association between cell types and OS progression remains poor.

Methods: We generated a human OS cell atlas by integrating over 110,000 single cells from 17 samples. Multiple machine learning algorithms were applied to develop tumor purity prediction models based on transcriptomic profile of OS. The Scissor algorithm and gene enrichment analyses were conducted to delve into cell-intrinsic molecular characteristics linked to OS prognosis. Moreover, the study investigated the impact of ATF6α in OS aggressiveness through genetic and pharmacological loss of function analyses. Lastly, the CellChat algorithm was employed to investigate cell-cell communications.

Results: Utilizing the high-quality human OS cell atlas, we identified tumor purity as a prognostic indicator and developed a robust tumor purity prediction model. We respectively delineated cancer cell- and immune cell-intrinsic molecular characteristics associated with OS prognosis at single-cell resolution. Interestingly, tumor cells with activated unfolded protein response (UPR) pathway were significantly associated with disease aggressiveness. Notably, ATF6α emerged as the top-activated transcription factor for this tumor subcluster. Subsequently, we confirmed that ATF6α was markedly associated with OS progression, while both genetic and pharmacological inhibition of ATF6α impaired the survival of HOS cells. Lastly, we depicted the landscape of signal crosstalk between the UPR-related subcluster and other cell types within the tumor microenvironment.

Conclusion: In summary, our work provides novel insights into the molecular biology of OS, and offers valuable resource for OS biomarker discovery and treatment strategy development.