Figure 1 EV biogenesis and molecular components. (a) An overview of the biogenesis pathways for three major EV subtypes. (b) Molecular compositions of exosomes. Abbreviations: ESE, early sorting endosome; EV, extracellular vesicle; LSE, late sorting endosome; MVB, multivesicular body. Figure adapted from images created in BioRender; Wen Y. 2025. https://BioRender.com/4aacvz1 .#

传统生物发生机制将 EV 分为三类：

• 外泌体（Exosomes, 30–150 nm）：起源于内体系统，经早期分选内体（ESE）、晚期分选内体（LSE）及多囊体（MVBs）与质膜融合释放。
• 微囊泡/外胚体（Ectosomes/Microvesicles, 100–1000 nm）：通过质膜直接出芽与胞吞裂解形成，伴随胞内钙离子信号与磷脂酰丝氨酸（PS）外翻。
• 凋亡小体（Apoptotic Bodies, 1–5 μm）：细胞发生程序性死亡时膜泡化崩解产生的较大碎片。

EV 关键分子载荷全景

• 膜蛋白与胞质蛋白：包括四跨膜蛋白（CD9、CD63、CD81）、整合素（Integrins）、主要组织相容性复合体（MHC）、脂锚定蛋白（如胰腺癌标志物 Glypican-1）以及生物发生相关蛋白（Alix、TSG101、HSPs）。
• 核酸全谱：双链/单链 DNA（反映体细胞突变）、miRNA/mRNA（转录后调控与表达特征）、lncRNA、circRNA、tRNA 衍生小片段、piRNA 以及 snoRNA。
• 脂质与糖链：胆固醇、神经酰胺、鞘磷脂（维持膜结构与刚性）；高甘露糖型、唾液酸化与岩藻糖基化 N-糖链（免疫逃逸与器官靶向趋向性）。

Table 1 Summary of the molecular compositions of EVs#

1. 蛋白质定量与表征#

• 高灵敏荧光与超分辨：利用晶体管型半导体聚合物点（Pdots）实现单分子灵敏度的流动计数与超分辨拓扑成像；等离激元纳米荧光标签可实现荧光信号 6,700 倍增强，检出限（LOD）提升数千倍。
• 电化学与阻抗谱：多电极 iPEX 系统提供免标记阻抗检测，LOD 达 $500\text{ EVs/mL}$；DSPE-PEG 膜锚定氧化还原探针构建比率型电化学传感，无需复杂酶促级联即可测得 $811\text{ EVs/mL}$。
• 先进等离激元平台：KeyPLEX 利用电渗透与介电泳力克服扩散限制，10 分钟内完成稀有 EV 快速富集与无标记检测；DISEP（表面等离激元共振显微镜 SPRM 结合双低亲和力寡核苷酸探针）通过动态结合动力学区分肿瘤特异性 EV，仅需 $10\ \mu\text{L}$ 血浆即可实现 AUC 0.98 的精准诊断。
• 深度蛋白质组质谱：DDA 与 DIA（数据非依赖采集）质谱深度覆盖血浆与尿液 EV 蛋白质组，建立疾病特征图谱。

2. 核酸原位与免扩增检测#

• 原位无损分析：EV-CLIP（荷电脂质体融合系统）实现免提取、免扩增的数字 RNA 分析（LOD $79\text{ EVs/}\mu\text{L}$）；DNA 纳米笼热泳法选择性富集并检测成熟态 miRNA（LOD 达 $2.05\text{ fM}$）。
• CRISPR/Cas 生物传感：Cas13a 微室阵列实现飞摩尔级免扩增检测；膜融合系统（MFS-CRISPR）将 Cas13a 直接递送至囊泡内部；CLAMP 技术结合失活 dCas9-sgRNA 与光电化学传感，实现阿摩尔级（attomolar）多重 miRNA 单碱基特异性识别。

3. 蛋白质-核酸同步原位联检#

• 数字双 CRISPR-Cas 系统利用 18.8 万个微孔阵列，在单囊泡水平同步定量膜表面蛋白与内源性 miRNA（乳腺癌临床分类准确率 92%）。
• 微滴微流控技术并行捕获单个 EV 内部的 ERBB2 mRNA 与膜表面 HER2 蛋白，揭示基因转录与蛋白表达的异质性相关性。

Figure 4 Single EV analysis. (a) Schematic of SPIRFISH, a high-throughput method for single EV protein and RNA analysis. Panel adapted with permission from Reference 93 (CC BY-NC-ND 4.0). (b) Overview of the nanosurface microfluidic embedded with a MoSERS microchip for SERS identification of single EVs for glioblastoma cancer. Panel adapted with permission from Reference 98 (CC BY-NC-ND 4.0). (c) Schematic of high-throughput droplet-based EV analysis that enabled parallel droplet generation and incubation as well as EV detection and quantification. The droplet generation includes a parallelized flow-focusing droplet generator that encapsulates Ab-labeled fluorescent beads and enzyme substrates into droplets. After on-chip incubation to generate fluorescent signals, the droplets flow through a detection region that consists of 90 parallelized microfluidic channels, where their fluorescence is measured using two time domain–modulated laser diodes (one for bead fluorescence and another for substrate fluorescence) and a machine vision camera. Videos from the camera are processed either by a local computer or in the Cloud to quantify the EV concentration. Panel adapted with permission from Reference 111; copyright 2022 American Chemical Society. (d) Overview of iterative multiplexed analysis of single EVs based on cyclic staining and imaging. Colored Abs indicate the fluorophores conjugated to them (black = no fluorophore conjugation). Panel adapted from Reference 120 (CC BY 4.0). Abbreviations: Ab, antibody; AEVB, average EV per bead; AF555, Alexa Fluor 555; AF647, Alexa Fluor 647; AI, artificial intelligence; C2TCO, C2-symmetric trans-cyclooctene; EV, extracellular vesicle; GBM, glioblastoma; GPU, graphics processing unit; LOD, limit of detection; MB488, methyl thiazolyl blue 488; MoSERS, MoS2 surface-enhanced Raman spectroscopy; SERS, surface-enhanced Raman spectroscopy; smFISH, single-molecule fluorescence in situ hybridization; SPIRFISH, single-particle interferometric reflectance imaging sensor with single-molecule fluorescence in situ hybridization; SP-IRIS, single-particle interferometric reflectance imaging sensor; Tz, tetrazine.#

Figure 5 Machine learning facilitates EV data analysis. (a, b) Multigroup classification and diagnostic evaluation of EWS patients of different age groups using μTIP-dELISA. Dashed ellipses in panel a represent 95% confidence intervals for the means of predicted groups. Panels adapted with permission from Reference 123; copyright 2025 American Chemical Society. (c, d) Machine learning-assisted VTE risk evaluation. Colored ellipses in panel c represent 95% confidence intervals for the means of predicted groups. Panels adapted with permission from Reference 124; copyright 2023 American Chemical Society. (e) AI framework for one test-multicancer analysis using exosome-SERS-AI. Panel adapted from Reference 126 (CC BY 4.0). Abbreviations: AI, artificial intelligence; AUC, area under curve; CD99, cluster of differentiation 99; ENO-2, enolase 2; EV, extracellular vesicle; EWS, Ewing sarcoma; EZR, ezrin; MIL, multiple instance learning; μTIP-dELISA, topographically intensified, partition-less digital enzyme-linked immunosorbent assay; NGFR, nerve growth factor receptor; SERS, surface-enhanced Raman spectroscopy; TOO, tissue of origin; VTE, venous thromboembolism.#