MRI-derived radiomics assessing tumor-infiltrating macrophages enable prediction of immune-phenotype, immunotherapy response and survival in glioma
Di Chen,Rui Zhang,Xiaoming Huang,Chunxia Ji,Wei Xia,Ying Qi,Xinyu Yang,Lishuang Lin,Jing Wang,Haixia Cheng,Weijun Tang,Jinhua Yu,Dave S. B. Hoon,Jun Zhang,Xin Gao,Yu Yao
DOI: https://doi.org/10.1186/s40364-024-00560-6
2024-01-31
Biomarker Research
Abstract:Abstract Background The tumor immune microenvironment can influence the prognosis and treatment response to immunotherapy. We aimed to develop a non-invasive radiomic signature in high-grade glioma (HGG) to predict the absolute density of tumor-associated macrophages (TAMs), the preponderant immune cells in the microenvironment of HGG. We also aimed to evaluate the association between the signature, and tumor immune phenotype as well as response to immunotherapy. Methods In this retrospective setting, total of 379 patients with HGG from three independent cohorts were included to construct a radiomic model named Radiomics Immunological Biomarker (RIB) for predicting the absolute density of M2-like TAM using the mRMR feature ranking method and LASSO classifier. Among them, 145 patients from the TCGA microarray cohort were randomly allocated into a training set ( N =101) and an internal validation set ( N =44), while the immune-phenotype cohort ( N =203) and the immunotherapy-treated cohort ( N =31, patients from a prospective clinical trial treated with DC vaccine) recruited from Huashan Hospital were used as two external validation sets. The immunotherapy-treated cohort was also used to evaluate the relationship between RIB and immunotherapy response. Radiogenomic analysis was performed to find functional annotations using RNA sequencing data from TAM cells. Results An 11-feature radiomic model for M2-like TAM was developed and validated in four datasets of HGG patients (area under the curve = 0.849, 0.719, 0.674, and 0.671) using MRI images of post contrast enhanced T1-weighted (T1CE). Patients with high RIB scores had a strong inflammatory response. Four hub-genes (SLC7A7, RNASE6, HLA-DRB1 and CD300A) expressed by TAM were identified to be closely related to the RIB, providing important evidence for biological interpretation. Only individuals with a high RIB score were shown to have survival benefits from DC vaccine [DC vaccine vs. Placebo: median progression-free survival (mPFS), 10.0 mos vs. 4.5 mos, HR=0.17, P =0.0056, 95%CI=0.041-0.68; median overall survival (mOS), 15.0 mos vs. 7.0 mos, HR=0.17, P =0.0076, 95%CI=0.04-0.68]. Multivariate analyses also confirmed that treatment by DC vaccine was an independent factor for improved survival in the high RIB score group. However, in the low RIB score group, DC vaccine was not associated with improved survival. Furthermore, a radiomic nomogram based on the RIB score and clinical factors could efficiently predict the 1-, 2-, and 3-year survival rates, as confirmed by ROC curve analysis (AUC for 1-, 2- and 3-year survival: 0.705, 0.729 and 0.684, respectively). Conclusions The radiomic model could allow for non-invasive assessment of the absolute density of TAM from MRI images in HGG patients. Of note, our RIB model is the first immunological radiomic model confirmed to have the ability to predict survival benefits from DC vaccine in gliomas, thereby providing a novel tool to inform treatment decisions and monitor patient treatment course by radiomics.
oncology,medicine, research & experimental