探索基于表面增强拉曼散射的生物感测技术在疾病诊断和治疗中的应用
March 13th, 2026
本文介绍了利用无标记表面增强拉曼光谱（SERS）分析单个小型细胞外囊泡（sEV）的方案，实现微创疾病诊断和评估sEV作为治疗载体。
我们结合表面增强拉曼散射技术和机器学习，解决癌症早期诊断，同时提供生物医学应用。一个关键的实验挑战是外泌体样品固有的异质性、高度复杂、高维的光谱数据，这又因低信噪比而加剧。首先，将大约五微升的小型细胞外囊泡样本移液器移液到指定的等离激元底物上。
将底材放入干燥器中，让液滴完全干燥约15分钟。然后将干燥的基底转移到共聚焦拉曼显微镜上。使用仪器软件 WiRE 4.4 版，开始数据收集。
将激发激光器设置为785纳米，功率为5毫瓦，并校准系统。要从金基底获取光谱，需在300×300微米范围内进行侦察扫描，使用静态模式，步长10微米，曝光0.1秒，激光功率为50%。定位囊泡后，在五乘五网格内以一微米步长、每点0.2秒曝光和50%激光功率进行静态模式的高分辨率映射。
接下来，通过以下扫描从石墨烯基底获取光谱。筛选初始原始光谱，排除任何显示过多噪声或异常光谱形状的光谱。对每个光谱应用萨维茨克-戈莱滤光片，以减少荧光背景并平滑光谱数据。
使用峰值到基线法或标准差法计算每个频谱的信噪比。对于每个频谱，计算信噪比，即在代表性拉曼波段基线减去后峰值强度与噪声标准差的比值。通过从原始频谱中减去Savitsky-Golay平滑版本的噪声，使用窗口大小为11，多项式阶为3。
按信噪比由高到低对所有频谱进行排序，并评估诊断峰值在排名频谱中的持续性，以确定阈值。然后在这些峰值消失为基线噪声的点设定阈值，对应信噪比为28。排除低于信噪比阈值28的频谱，以确保仅保留高质量的光谱供后续分析使用。
并将剩余的每个频谱归一化为其最大峰值强度或曲线下的总面积。根据每个光谱测量的来源（如胃癌患者或健康对照）为每个光谱测量分配标签。将标记的谱数据输入带有线性核的支持向量分类器，并进行分层洗牌分割进行五重交叉验证，以确保每个重叠的平衡表示。
现在，对所有折叠的准确率指标取平均，以评估整体模型表现。最后，使用线性判别分析（LDA）降低表面增强拉曼散射数据集的维数以便可视化。使用训练有素的线性判别分析模型，直接对样本组进行分类。
拉曼光谱与对应的平均光谱显示了每种样本类型中从组织、血液和唾液样本中分离出的单个小细胞外囊泡中截然不同的分子特征。线性判别分析显示细胞外囊泡数据根据来源组织、血液或唾液明显聚集。使用支持载体机分类器进行二元分类，组织样本准确率最高，达到90.1%，其次是血液70.9%，唾液为60.7%，基于LDA的分类在组织来源的细胞外囊泡数据中清晰区分健康癌和胃癌样本。
类似基于LDA的血液和唾液分类分离显示，健康样本与癌症样本之间的分离度较差。多柔比星培养的囊泡无石墨烯的拉曼光谱显示，峰值一致地位于约1,081、1、206和1,440反向厘米。对于石墨烯，观察到一个约1,350反向厘米的D峰和约1,580反向厘米的G峰，作为内部标准。
多柔比星峰在442厘米处与石墨烯G峰的比值随着药物浓度升高和潜伏时间延长而增加。我们建立了单一囊泡生化指纹识别，以区分具有早期诊断潜力的疾病来源。我们技术的三大关键优势。
第一，它非常敏感。第二，它是非侵入性的。第三，它不需要溶解，也不需要物理分解粒子。
未来，我们将谱与蛋白质组组谱相关联，蛋白质组组图谱反过来有助于将外显子含量与细胞通讯和运输等特定功能相关联。
View the full transcript and gain access to thousands of scientific videos
Overview
This protocol presents a label-free analytical platform that integrates surface-enhanced Raman spectroscopy (SERS) with machine learning to detect and molecularly profile individual small extracellular vesicles (sEVs). The method enables high-specificity detection and classification of disease states, such as distinguishing gastric cancer from healthy controls, and quantifies drug loading in single vesicles, supporting both diagnostic and therapeutic applications.
Key Study Components
Area of Science
- Analytical chemistry
- Biomedical diagnostics
- Nanotechnology
- Machine learning applications in biosensing
Background
- Small extracellular vesicles (sEVs) are promising biomarkers for disease diagnosis and therapeutic monitoring.
- Traditional detection methods often require labeling or lysis, which can limit sensitivity and throughput.
- SERS provides molecular fingerprinting capabilities with high sensitivity.
- Machine learning enhances the analysis of complex, high-dimensional spectral data.
Purpose of Study
- To develop a label-free, high-throughput platform for single-vesicle analysis using SERS and machine learning.
- To distinguish between healthy and disease states (e.g., gastric cancer) using sEVs from various biofluids.
- To quantify drug (doxorubicin) loading in individual sEVs for therapeutic assessment.
Methods Used
- Isolation of sEVs via size exclusion chromatography or ultracentrifugation.
- Deposition of sEVs onto engineered plasmonic gold nanopyramid 2D array substrates.
- SERS data acquisition using a confocal Raman microscope with specific laser settings and mapping protocols.
- Preprocessing of spectra (noise filtering, normalization, signal-to-noise thresholding).
- Machine learning analysis using support vector classifiers (SVC) and linear discriminant analysis (LDA) for classification and visualization.
- Use of graphene-coated substrates as internal standards for drug quantification.
Main Results
- Distinct molecular signatures were observed for sEVs from tissue, blood, and saliva.
- Classification accuracies for distinguishing gastric cancer from healthy controls were 90.1% (tissue), 70.9% (blood), and 60.7% (saliva).
- LDA revealed clear clustering of sEVs by source and disease state, especially for tissue-derived samples.
- Quantification of doxorubicin loading in single sEVs was achieved, with graphene peaks serving as internal standards.
Conclusions
- The platform enables sensitive, non-invasive, and label-free detection of sEVs at the single-vesicle level.
- It allows for high-throughput analysis and captures population heterogeneity, aiding early disease detection.
- The method supports both diagnostic and therapeutic applications, including drug loading assessment.
Frequently Asked Questions
Executive Industry Relevance
Single-vesicle SERS biosensing, integrated with machine learning, addresses the challenge of molecular heterogeneity in extracellular vesicle (EV) populations, enabling high-specificity disease state classification and drug loading quantification. This platform enhances predictive confidence at the early diagnostic and therapeutic assessment stages, supporting risk-adjusted portfolio decisions in oncology and precision medicine. Its label-free, high-throughput capability positions it as a reusable asset for translational biomarker discovery and mechanistic de-risking.
Strategic Applications in Biopharma R&D
Early Discovery & Target Validation
- Enables molecular fingerprinting of individual sEVs to clarify disease-associated pathways.
- Supports functional target validation by distinguishing disease from healthy states using intrinsic Raman signatures.
- Facilitates predictive confidence in biomarker selection and triage of candidate targets.
Screening & Assay Development
- Prepares validated, label-free biosensing systems for downstream screening workflows.
- Delivers quantitative, reproducible spectral outputs for robust assay standardization.
- Enables high-throughput, single-vesicle analysis to capture population heterogeneity relevant for screening readiness.
Translational & Preclinical Research
- Aligns molecular EV profiles with disease states, supporting translational biomarker continuity.
- Quantifies therapeutic payloads (e.g., doxorubicin) in sEVs, informing preclinical drug delivery studies.
- Provides mechanistic de-risking by linking spectral profiles to functional outcomes.
Pipeline & Workflow Integration
This SERS-ML platform integrates from early discovery through translational research, bridging hypothesis testing, screening, and preclinical validation for EV-based diagnostics and therapeutics.
- Discovery Biology: Supports hypothesis testing and pathway clarification via single-vesicle molecular profiling.
- Screening: Provides reproducible, quantitative readouts for assay development and compound evaluation.
- Analytics: Delivers high-dimensional spectral data and robust statistical classification for condition comparison.
- Translational Research: Enables continuity from discovery to preclinical validation by quantifying disease and therapeutic markers in sEVs.
- Enterprise Reuse: Establishes a scalable, label-free biosensing capability applicable across multiple disease and therapeutic contexts.
Operational & Enterprise Impact
- Scientific Value: Increases predictive confidence and reduces mechanistic ambiguity in EV-based biomarker and drug delivery research.
- Operational Value: Standardizes high-throughput, label-free workflows with robust data quality controls.
- Strategic Value: Improves go/no-go decisions and capital efficiency by enabling early, accurate disease and therapeutic assessments.
- Portfolio Impact: Supports risk-adjusted prioritization and advancement of EV-based diagnostic and therapeutic programs.
Implementation Considerations
- Requires expertise in Raman spectroscopy, machine learning, and EV biology.
- Needs access to confocal Raman microscopes, engineered plasmonic substrates, and computational infrastructure for ML analysis.
- Demands rigorous cross-team standardization of spectral acquisition and preprocessing protocols.
- Adaptation across different biofluids and disease models may require substrate and algorithm optimization.
- Signal-to-noise thresholds and spectral quality controls are critical for reliable downstream analysis.
