Multi-modal characterization of metabolic and immune gene clusters in adrenocortical carcinoma treatment.

Abstract

Adrenocortical carcinoma (ACC) is an uncommon and aggressive endocrine malignancy, characterized by limited therapeutic options and considerable variability in patient outcomes. The challenge is to combine the complex information of ACC with artificial intelligence (AI) and clinical and pathology data to achieve precision medicine and improve patient prognosis. We developed the Steroid-related Immune Score (SIS) using multi-modal analysis of genomics, digital pathology, and artificial intelligence and validated it in external datasets. In addition, we conducted single-cell RNA sequencing (scRNA-seq) of small samples and in vitro functional experiments. SIS delivered a stable performance with an AUC of 0.8 ± 0.01 in the ResNet50 and Vision Transformer-B16 models. We validated the best model in external ACC cohorts. Using Class Activation Maps (CAMs) technology revealed that SIS was associated with lymphocyte infiltration, establishing it as a new feature in addition to the Weiss scoring system. Patients in the high SIS group responded well to immunotherapy, while the low SIS group showed adaptability to hormone inhibition therapy. Single-cell RNA sequencing data revealed the relationship between the tumor microenvironment and drug resistance in ACC. In vitro functional assays demonstrated that elevated DHCR7 gene expression correlated with unfavorable prognosis and treatment sensitivity, identifying it as a prospective therapeutic target. Furthermore, there are similarities between the metabolic characteristics of ACC and schizophrenia, such as calcium and iron ion levels. Our multi-modal analysis comprehensively characterizes the immune microenvironment of ACC, emphasizing the synergistic regulation of metabolic and immune gene clusters that influence ACC patients' responses to immune and hormone therapies.