Xu Jianying, Shi Wenjie, Zhu Yi, Zhang Chao, Nagelschmitz Julia, Doelling Maximilian, Al-Madhi Sara, Mukund Mahajan Ujjwal, Pech Maciej, Rose Georg, Croner Roland Siegfried, Zheng Guoliang, Kahlert Christoph, Kahlert Ulf Dietrich
Department of Medicine II, Hospital of the LMU, Munich, Germany.
Molecular and Experimental Surgery, Clinic for General-, Visceral -, Vascular- and Transplantation Surgery, Medical Faculty and University Hospital Magdeburg, Otto-von-Guericke University, Magdeburg, Germany.
Elife. 2025 Aug 8;14:RP103737. doi: 10.7554/eLife.103737.
Pancreatic cancer (PC) is a highly aggressive malignancy in humans, where early diagnosis significantly improves patient outcomes. However, effective methods for accurate and early detection remain limited. In this multiethnic study involving human subjects, we developed a liquid biopsy signature based on extracellular vesicle (EV)-derived microRNAs (miRNAs) linked to radiomics features extracted from patients' tumor imaging. We integrated eight datasets containing clinical records, imaging data of benign and malignant pancreatic lesions, and small RNA sequencing data from plasma-derived EVs of PC patients. Radiomics features were extracted and analyzed using the limma package, with feature selection conducted via the Boruta algorithm and model construction through Least Absolute Shrinkage and Selection Operator regression. Radiomics-related low-abundance EV miRNAs were identified via weighted gene co-expression network analysis and validated for diagnostic accuracy using 10 machine-learning algorithms. Three key EV miRNAs were found to robustly distinguish malignant from benign lesions. Subsequent molecular clustering of these miRNAs and their predicted targets identified two PC subtypes, with distinct survival profiles and therapeutic responses. Specifically, one cluster was associated with prolonged overall survival and higher predicted sensitivity to immunotherapy, while the other indicated high-risk tumors potentially amenable to targeted drug interventions. This radiogenomic EV miRNA signature in human plasma represents a promising non-invasive biomarker for early diagnosis and molecular subtyping of PC, with potential implications for precision treatment strategies.
胰腺癌（PC）是一种在人类中具有高度侵袭性的恶性肿瘤，早期诊断能显著改善患者预后。然而，准确早期检测的有效方法仍然有限。在这项涉及人类受试者的多民族研究中，我们基于细胞外囊泡（EV）衍生的微小RNA（miRNA）开发了一种液体活检特征，这些miRNA与从患者肿瘤成像中提取的放射组学特征相关联。我们整合了八个数据集，其中包含临床记录、良性和恶性胰腺病变的成像数据以及来自PC患者血浆来源的EV的小RNA测序数据。使用limma软件包提取和分析放射组学特征，通过Boruta算法进行特征选择，并通过最小绝对收缩和选择算子回归进行模型构建。通过加权基因共表达网络分析鉴定与放射组学相关的低丰度EV miRNA，并使用10种机器学习算法验证其诊断准确性。发现三种关键的EV miRNA能够可靠地区分恶性病变和良性病变。随后对这些miRNA及其预测靶点进行分子聚类，确定了两种PC亚型，它们具有不同的生存特征和治疗反应。具体而言，一个聚类与总体生存期延长以及对免疫治疗的较高预测敏感性相关，而另一个则表明是可能适合靶向药物干预的高危肿瘤。这种人类血浆中的放射基因组EV miRNA特征代表了一种有前景的非侵入性生物标志物，可用于PC的早期诊断和分子亚型分类，对精准治疗策略具有潜在意义。
