Lipidome-based rapid diagnosis with machine learning for detection of TGF-β signalling activated area in head and neck cancer

• Published in: British journal of cancer 2020-03, Vol.122 (7), p.995-1004
• Main Authors: Ishii, Hiroki, Saitoh, Masao, Sakamoto, Kaname, Sakamoto, Kei, Saigusa, Daisuke, Kasai, Hirotake, Ashizawa, Kei, Miyazawa, Keiji, Takeda, Sen, Masuyama, Keisuke, Yoshimura, Kentaro
• Format: Article
• Language: English
• Subjects: 631/67/1665/3016; 692/308/575; 692/4028/546; Biomedical and Life Sciences; Biomedicine; Cancer Research; Cell Line, Tumor; Diagnosis; Drug Resistance; Epidemiology; Head & neck cancer; Head and Neck Neoplasms - diagnosis; Head and Neck Neoplasms - pathology; Humans; Learning algorithms; Lecithin; Lipidomics - methods; Machine learning; Machine Learning - standards; Mass spectroscopy; Medical diagnosis; Molecular Medicine; Oncology; Phosphatidylcholine; Signal Transduction; Smad2 protein; Squamous cell carcinoma; Transforming Growth Factor beta - metabolism; Transforming growth factor-b; Transforming growth factor-b1; Tumor microenvironment; Tumors
• ISSN: 0007-0920; 1532-1827
• DOI: 10.1038/s41416-020-0732-y

Background Several pro-oncogenic signals, including transforming growth factor beta (TGF-β) signalling from tumour microenvironment, generate intratumoural phenotypic heterogeneity and result in tumour progression and treatment failure. However, the precise diagnosis for tumour areas containing subclones with cytokine-induced malignant properties remains clinically challenging. Methods We established a rapid diagnostic system based on the combination of probe electrospray ionisation-mass spectrometry (PESI-MS) and machine learning without the aid of immunohistological and biochemical procedures to identify tumour areas with heterogeneous TGF-β signalling status in head and neck squamous cell carcinoma (HNSCC). A total of 240 and 90 mass spectra were obtained from TGF-β-unstimulated and -stimulated HNSCC cells, respectively, by PESI-MS and were used for the construction of a diagnostic system based on lipidome. Results This discriminant algorithm achieved 98.79% accuracy in discrimination of TGF-β1-stimulated cells from untreated cells. In clinical human HNSCC tissues, this approach achieved determination of tumour areas with activated TGF-β signalling as efficiently as a conventional histopathological assessment using phosphorylated-SMAD2 staining. Furthermore, several altered peaks on mass spectra were identified as phosphatidylcholine species in TGF-β-stimulated HNSCC cells. Conclusions This diagnostic system combined with PESI-MS and machine learning encourages us to clinically diagnose intratumoural phenotypic heterogeneity induced by TGF-β.