Early predictors of cancer evolution under therapy have been identiﬁed using an artiﬁcial intelligence program to analyze data from tumor samples of patients with glioma. The model shows signs of signiﬁcantly better performance than current grading and diagnostic models.
The international research team was led by Hong Kong University of Science and Technology (HKUST), and by Prof. WANG Jiguang, Padma Harilela Associate Professor of Life Science in the HKUST’s Division of Life Science and Department of Chemical and Biological Engineering, and reported in Science Translational Medicine.
Malignant diffuse gliomas are the most common primary brain tumors in adults. Currently, the main treatment of gliomas is a combination of surgery, radiotherapy, and the chemotherapy agent temozolomide (TMZ). This type of treatment can usually prolong patients’ overall survival time for around three months; however, patients almost inevitably suffer a relapse of aggressive gliomas. It also leads to a worse prognosis for patients. Until now, the molecular mechanisms driving glioma evolution remain elusive.
The authors performed a comprehensive analysis of TMZ-treated adult tumors in prior trials, as well as newly collected samples from Beijing Tiantan Hospital (CGGA), Samsung Medical Center (SMC), and Prince Whales Hospital of Chinese University of Hong Kong (CUHK), to identify genomic and transcriptomic early predictors of tumor evolution in each molecular subtype.
A machine learning model was developed named CELLO2 (Cancer EvoLution for LOngitudinal data version 2), to predict tumor behavior under TMZ treatment based on clinical and genomic features collected at the initial diagnosis. The model deﬁned a new strategy to grade glioma risk, and successfully identiﬁed a subgroup of patients with signiﬁcantly worse prognosis.
In all glioma subtypes, MYC gain or MYC-targets activation at diagnosis was associated with TMZ-resistance. c-MYC, a piece of genetic code that regulates the cell cycle, was identiﬁed as an early predictor of treatment-induced hypermutation, a classic pathway by which tumors evolve toward TMZ resistance. It was subsequently demonstrated that c-MYC, binding to open chromatin and transcriptionally active genomic regions, increases the vulnerability of key mismatch repair genes to TMZ-induced mutagenesis, thus triggering hypermutation.
The team further discovered that East Asian brain tumors have remarkable differences with those of Caucasians in terms of chromosomes 7-gain and 10-loss (+7/-10), and molecular mechanisms of MYC activation. Particularly, East Asian gliomas carried more MYC gain as one of the early brain tumor formations.
To allow public use of the predictive mode, the team developed an interactive, open available web portal (CELLO2, www.wang-lab-hkust.com:3838/cello2) for patients and doctors to explore the longitudinal glioma data resource and make predictions of developing treatment-induced hypermutation and grade progression based on clinical and genomic features. The platform will be useful for patients for their better understanding of the aggressiveness of their brain tumors.
The authors acknowledge collaboration from the teams of Tao Jiang at Beijing Tiantan Hospital, Dohyun Nam at Samsung Medical Center, Antonio Iavarone at University of Miami, Danny Chan, Ho Keung Ng and Wai Sang Poon at Prince Whales Hospital of CUHK, Zhaoqi Liu at Beijing Institute of Genomics, and Angela Wu at HKUST.Back To Top
Machine-Learning Model Identifies Early Predictors of Aggressive Brain Cancer. Appl Rad Oncol.