Integrated Pipeline of Rapid Isolation and Analysis of Human Plasma Exosomes for Cancer Discrimination Based on Deep Learning of MALDI-TOF MS Fingerprints

Plasma exosomes have shown great potential for liquid biopsy in clinical cancer diagnosis. Herein, we present an integrated strategy for isolating and analyzing exosomes from human plasma rapidly and then discriminating different cancers excellently based on deep learning fingerprints of plasma exosomes. Sequential size-exclusion chromatography (SSEC) was developed efficiently for separating exosomes from human plasma. SSEC isolated plasma exosomes, taking as less as 2 h for a single sample with high purity such that the discard rates of high-density lipoproteins and low/very low-density lipoproteins were 93 and 85%, respectively. Benefitting from the rapid and high-purity isolation, the contents encapsulated in exosomes, covered by plasma proteins, were well profiled by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MS). We further analyzed 220 clinical samples, including 79 breast cancer patients, 57 pancreatic cancer patients, and 84 healthy controls. After MS data pre-processing and feature selection, the extracted MS feature peaks were utilized as inputs for constructing a multi-classifier artificial neural network (denoted as Exo-ANN) model. The optimized model avoided overfitting and performed well in both training cohorts and test cohorts. For the samples in the independent test cohort, it realized a diagnosed accuracy of 80.0% with an area under the curve of 0.91 for the whole group. These results suggest that our integrated pipeline may become a generic tool for liquid biopsy based on the analysis of plasma exosomes in clinics.血浆外泌体在临床癌症诊断的液体活检领域展现出巨大应用潜力。本研究提出一种整合策略，可快速分离并分析人血浆来源的外泌体，并基于血浆外泌体的深度学习指纹精准区分不同癌症类型。本研究高效开发了连续尺寸排阻色谱（Sequential Size-Exclusion Chromatography, SSEC）技术，用于从人血浆中分离外泌体：该技术单次处理单份样本仅需2小时，且分离产物纯度极高，高密度脂蛋白与低密度/极低密度脂蛋白的去除率分别达93%与85%。得益于快速且高纯度的分离流程，被血浆蛋白遮蔽的外泌体包裹内容物可通过基质辅助激光解吸电离飞行时间质谱（Matrix-Assisted Laser Desorption/Ionization Time-of-Flight Mass Spectrometry, MS）实现良好的组分表征。本研究共分析220份临床样本，涵盖79例乳腺癌患者、57例胰腺癌患者及84名健康对照个体。经质谱数据预处理与特征筛选后，提取得到的质谱特征峰被用作输入，构建多分类人工神经网络（记为Exo-ANN）模型。优化后的模型有效避免了过拟合问题，在训练队列与测试队列中均表现优异。针对独立测试队列中的样本，该模型整体诊断准确率达80.0%，曲线下面积（area under the curve）为0.91。上述结果表明，本研究开发的整合流程有望成为临床中基于血浆外泌体分析的通用液体活检工具。