Exploring the Application of Surface-enhanced Raman Scattering-based Biosensing of Individual sEVs in Disease Diagnosis and Therapeutics

Summary

Here, we present a protocol to analyze individual small extracellular vesicles (sEVs) using label-free surface-enhanced Raman spectroscopy (SERS), enabling minimally invasive disease diagnostics and assessment of sEVs as therapeutic delivery vehicles.

Abstract

This protocol outlines a comprehensive, label-free platform that integrates surface-enhanced Raman spectroscopy (SERS) with machine learning (ML) to detect and molecularly profile individual small extracellular vesicles (sEVs) for diagnostic and therapeutic applications. The method begins with sEV isolation using either size exclusion chromatography or ultracentrifugation. Isolated vesicles are then analyzed on engineered plasmonic gold nanopyramid 2D array substrates capable of single-vesicle sensitivity. By leveraging intrinsic Raman biochemical fingerprints, the protocol enables high-specificity detection without external labels. Following spectral acquisition, data undergo preprocessing and are analyzed using trained machine learning algorithms (e.g., LDA, SVC) to classify disease states, successfully distinguishing gastric cancer from healthy controls using sEVs from tissue, plasma, and saliva with respective classification accuracies of 90.1%, 70.9%, and 60.7%. Additionally, its therapeutic application is shown by quantifying doxorubicin loading in single sEVs, a measurement enhanced by using graphene-coated substrates as an internal standard. This approach allows for high-throughput analysis that captures population heterogeneity essential for early disease detection and understanding drug loading efficiency at the single-vesicle level.

Introduction

Small extracellular vesicles (sEVs), typically ranging from 30 to 200 nm in diameter, are naturally secreted by cells into biological fluids and carry proteins, lipids, and nucleic acids indicative of their cell of origin^1,2,3. As non-invasive biomarkers present in saliva, urine, and other body fluids⁴, they hold promise for early disease diagnosis, prognosis, and therapeutic monitoring, especially in oncology^5,6,7. However, their clinical translation is hindered by their high heterogeneity and low abundance in complex biological matrices⁸. To overcome such challenges, this protocol leverages surface-enhanced Raman spectroscopy (SERS), a powerful analytical method capable of amplifying Raman scattering by up to 10¹⁰-10¹¹ times via plasmonic substrates^9,10,11. SERS captures intrinsic molecular "fingerprint" spectra of individual sEVs within seconds, enabling ultra-sensitive single-vesicle analysis while avoiding long assay times.

Our method synergistically combines label-free based SERS to achieve rapid, ultrasensitive characterization of individual sEVs^{12,13,14,15,16}. Label-free SERS captures intrinsic biochemical signatures and, when coupled with machine-learning models like LDA, SVC, has been shown to discriminate sEVs derived from healthy versus diseased cells with high specificity¹³. To further enhance reproducibility and quantitative capability of our platform, our method is fully compatible with graphene-covered Raman substrates¹⁶. Single layer graphene offers a chemically uniform surface that stabilizes plasmonic hot spots and enhances molecular adsorption^17,18, and more importantly, single layer graphene serves as the "built-in" gauge of SERS signals^12,19, which enables the quantification ability of molecules detected. Despite the Raman laser beam diameter (~ 1 µm) being larger than the typical sEV, our platform achieves effective single-vesicle resolution due to the unique interplay of physical and statistical mechanisms. The quasi-periodic gold nanopyramid substrate generates highly localized plasmonic "hot spots" of sizes smaller than 100 nm that amplify Raman signals in a super linear fashion¹². As a result, the signal from the sEV stochastically closest to the most intense hot spot within the beam dominates the spectrum. Additionally, SEM imaging confirms that sEVs are well spaced (> 1 µm apart), minimizing spectral overlap, while Raman intensity maps reveal signal localization that matches the size of individual vesicles^13,14. Together, these features ensure that each collected spectrum corresponds predominantly to a single sEV, enabling statistically robust, label-free single-vesicle analysis. Compared to conventional approaches, such as western blot, mass spectrometry, or regular Raman scattering that require large sample volumes and cannot resolve single vesicles²⁰, our single-vesicle SERS platform offers superior sensitivity, throughput, and molecular resolution. This workflow is appropriate for researchers aiming to develop minimally invasive, high-resolution diagnostic tools or to explore sEV heterogeneity and drug-loading efficiency in individualized therapeutic studies. In this protocol, we demonstrate how our SERS platform utilizes single sEVs to enable disease detection and therapeutic applications, particularly in cancer, including the whole workflow of vesicle isolation, plasmonic substrate preparation, Raman acquisition, and data analysis with machine learning.

Protocol

The collection of human specimens was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board (IRB) of UCLA. All donors provided written informed consent prior to their participation in the study.

NOTE: A total of 4 cell lines, AGS (CRL-1739), A549 (CCL-185), NCI-N87 (CRL-5822), Hs 738.St/Int (CRL-7869), all purchased from a commercial vendor, were used in this study. These cell lines were cultured and passaged strictly according to the vendor's instructions and reagents. The conditioned media were collected and stored at -20 °C before being processed. At Samsung Medical Center in Korea, donors voluntarily contributed tissue, plasma, and saliva specimens. Gastric cancer tissue was obtained during surgical resection, while biopsies from non-cancer controls were collected via endoscopy. Plasma was gathered in EDTA tubes according to standard clinical protocols. About  5 mL of unstimulated whole saliva was collected from each subject, then centrifuged at 2,600 × g for 15 min at  4 °C. To preserve extracellular RNA, the supernatant was supplemented with SUPERase-In RNase inhibitor at 20 U/mL. All specimens were kept at -80 °C until further processing. sEVs were isolated from conditioned media and clinical specimens following established protocols²¹.

1. sEV isolation and characterization

1. Prepare sEV samples
   1. Thaw the cleared cell culture supernatant completely at room temperature.
   2. Pass the supernatant through a sterile 0.22 µm syringe filter to remove debris and larger contaminants.
   3. Equilibrate a 35 nm size-exclusion chromatography (SEC) column with 1× PBS solution.
   4. Apply 500 µL of the filtered sample to the column.
   5. Elute the sample with 1x PBS according to the manufacturer's instructions.
   6. Collect the EV-enriched fractions and store them in clean vials at −20 °C.
2. Determine particle concentration and size
   1. Dilute sEVs samples in sterile DPBS to a final particle concentration of 10⁷-10⁹ particles/mL.
   2. Using a nanoparticle tracking analysis instrument, record five 60 s videos per sample. Use a camera level of 14 and a constant flow rate setting of 25.
   3. Analyze the videos with the instrument's software, applying the dilution factor to determine the final particle concentration (particles/mL) and mean diameter (nm).
3. Characterize tetraspanin expression
   1. Pre-scan human tetraspanin chips (1 cm × 1 cm, pre-coated with anti-CD9, anti-CD63, and anti-CD81) to establish a baseline signal.
   2. Load diluted sEV samples onto the chips in a sealed 24-well plate and incubate overnight at room temperature.
   3. Add 35 µL of detection antibodies (anti-CD9 CF488, anti-CD63 CF647, anti-CD81 CF555) and incubate for 1 h at room temperature.
   4. Wash the chips according to the manufacturer's protocol.
   5. Measure vesicular tetraspanin expression using a biomarker characterization reader and its associated scanning software.
4. Confirm sEV morphology with TEM
   1. Fix ~100 µL sEVs with ~100 µL of 4% paraformaldehyde for 30 min.
   2. Load approximately 100 µL of fixed sEVs onto 300-mesh copper grids and incubate for 5 min.
   3. Rinse the grids with deionized water.
   4. Stain the grids with a 1% solution of uranyl acetate (~ 20 µL) in distilled water.
   5. Perform a final rinse with water and allow the grids to air-dry completely.
   6. Image the samples via transmission electron microscopy (TEM) to confirm sEV size and morphology.

2. SERS plasmonic substrate fabrication ^12,18,19

1. Create nanosphere monolayer
   1. Spread 500 nm diameter polystyrene spheres on a water surface within a Langmuir-Blodgett trough.
   2. Gently compress the spheres to form a close-packed hexagonal monolayer.
   3. Transfer the sphere monolayer onto a 4 inch Si wafer (with ~50 nm thermal SiO₂) by dipping the wafer into the trough and pulling it upward slowly.
2. Etch nanosphere array
   1. Reduce the sphere diameter from 500 nm to ~250 nm using a reactive-ion etching system with oxygen plasma (O₂ gas flow at 100 sccm, 50 s, 200 W power). For more details of the fabrication process, please refer to the publication from our previous work^12,18,19.
   2. Deposit a 50 nm Cr layer over the shrunken sphere array via electron-beam evaporation at 1.5 Å per s.
   3. Lift off the polystyrene spheres using acetone to leave a perforated Cr mask on the SiO₂ surface.
   4. Etch the exposed SiO₂ layer in the Cr openings using a mixture of 25 sccm Ar, 25 sccm CHF₃, with RIE power at 200W for 2 min.
3. Fabricate inverted pyramids
   1. Immerse the wafer in 60 wt% KOH at 60 °C for 2 min to etch inverted pyramidal pits into the silicon.
   2. Remove the Cr mask using a Cr etchant for 5 min.
   3. Remove the SiO₂ layer in 6:1 buffered oxide etchant until fully cleared.
   4. Deposit a 200 nm Au film conformally over the inverted pyramids using electron-beam evaporation.
4. Finalize plasmonic substrate
   1. Apply fast-curing epoxy (~1 mL) to a separate SiO₂/Si substrate. Press the gold-coated wafer onto the epoxy to form a sandwich structure with the gold layer in the middle.
   2. Peel the epoxy away from the silicon mold to form a free-standing Au plasmonic substrate after the epoxy has fully cured.

3. Graphene transfer to plasmonic substrate

1. Prepare graphene film
   1. Prepare a 1 M FeCl₃ solution (A 1 M stock solution of anhydrous ferric chloride was prepared by dissolving 81 g of FeCl₃ in deionized water and diluting to a final volume of 500 mL) to etch the copper substrate from a PMMA-coated monolayer graphene sheet.
   2. Place the graphene-coated copper sheet face-down on the surface of the FeCl₃ solution until the copper is completely dissolved.
2. Rinse graphene film
   1. Using a clean glass slide, carefully transfer the floating graphene film to a deionized water bath to rinse off etchant residues.
   2. Repeat this rinsing step three times.
3. Transfer graphene to substrate
   1. Float the SERS plasmonic substrate beneath the graphene film in the final DI water bath.
   2. Gently lift the substrate to settle the graphene evenly on its surface.
   3. Air-dry the graphene-coated substrate at room temperature for 1 h.
   4. Transfer it to a vacuum chamber and dry under reduced pressure at 180 °C for 1 h.
4. Clean final substrate
   1. Immerse the dried substrate in an acetone bath at 50 °C for 1 h to dissolve the PMMA.
   2. Rinse the substrate three times in isopropyl alcohol.
   3. Dry the cleaned, graphene-transferred plasmonic substrate in a desiccator before use.

4. SERS sample preparation and spectra acquisition

1. Prepare sample on substrate
   1. Pipette approximately 5 µL of the sEV sample onto the appropriate plasmonic substrate.
      NOTE: For quantification of drug-loaded sEVs, apply vesicles to the gold-graphene substrate. For all other experiments, use a plain gold plasmonic substrate.
   2. Place the substrate in a desiccator and allow the droplet to dry completely (~15 min).
2. Calibrate Raman system
   1. Transfer the substrate to a confocal Raman microscope.
   2. Initiate data collection using the instrument's software (WiRE version 4.4).
   3. Set the excitation laser to 785 nm at 5 mW.
   4. Calibrate the system by aligning the silicon reference peak to 520 cm^-1.
3. Acquire spectra from gold substrate (diagnostics)
   1. Perform a scouting scan over a 300 µm × 300 µm area to locate single vesicles using static mode with the following settings: 50% power, 0.1 s exposure, 10 µm step size.
   2. After locating sEVs, conduct high-resolution mapping in static mode within a 5 × 5 grid using the following settings: 1 µm step size, 50% power, 0.2 s exposure per point.
4. Acquire spectra from graphene substrate (therapeutics)
   1. Perform a scouting scan over a 300 µm × 300 µm area to locate individual vesicles using static mode with the following settings: 50% laser power, 0.1 s exposure, 10 µm step size.
   2. Perform high-resolution mapping in extended mode with a 2 × 2 grid using the following settings: 1 µm step size, 1% laser power, 10 s exposure.

5. Data preprocessing

1. Perform spectral quality control
   1. Screen initial raw spectra to exclude highly noisy signals or spectra with abnormal shapes.
   2. Remove isolated cosmic ray spikes using threshold-based detection or local deviation filtering.
2. Filter and normalize spectra
   1. Apply a Savitzky-Golay filter to each spectrum to reduce fluorescence background and smooth the data.
   2. Calculate the signal-to-noise ratio (SNR) of each spectrum, typically using either the peak-to-baseline ratio or standard-deviation method. To explain further, typically the SNR for each spectrum was calculated as the ratio of the peak intensity (after baseline subtraction) at a representative Raman band to the standard deviation of the noise. The noise was estimated from the residual spectrum obtained by subtracting a Savitzky-Golay-smoothed version of the spectrum (window size = 11, polynomial order = 3) from the original spectrum, thereby isolating high-frequency fluctuations. To empirically determine an appropriate SNR threshold, all spectra were ranked in ascending order of SNR, and the persistence of characteristic biochemical peaks - specifically the phenylalanine band (~1002 cm^-1 , amino acid), the amide I band (~1650 cm^-1 , protein), and the CH₂ deformation band (~1440 cm^-1 , lipid) was evaluated across the ranked spectra. The threshold was set at the point where any one of these diagnostic peaks consistently disappeared into the baseline noise. For our gold-nanopyramid SERS substrate and acquisition parameters (785 nm excitation, 5 mW laser power, 0.2 s integration), this empirically corresponded to an SNR threshold of 28.
   3. Exclude any spectra that fall below the SNR threshold as decided according to 5.2.2 above. In our workflow, spectra with a SNR below 28 were excluded from the training dataset, since setting the threshold at this level ensures that only high-quality spectra are retained for downstream analysis and the improvement of the model robustness.
   4. Normalize each spectrum to either the maximum peak intensity or the total area under the curve.

6. Spectral analysis with machine learning

1. Train the Model
   1. Assign a label to each spectral measurement corresponding to its origin (e.g., gastric cancer patient vs. healthy control).
   2. Input the spectral data into a Support Vector Classifier (SVC) with a linear kernel.
   3. Use a stratified shuffle split for 5-fold cross-validation.
   4. Average the accuracy metrics from each fold to assess model performance.
2. Reduce dimensionality and classify
   1. Use Linear Discriminant Analysis (LDA) to visualize high-dimensional SERS data in reduced dimensions.
   2. Use the LDA model for direct classification between sample groups.
3. Evaluate the Model
   1. Evaluate the trained models on a holdout set of blinded patient samples to simulate a real-world diagnostic scenario. For more information, please refer to these publications from our group^13,15,22.
      NOTE: Perform all steps with clean tools in a dust-free environment. Monolayer graphene was grown or sourced on a copper substrate and coated with a PMMA layer to protect it during transfer.

Results

Successful implementation of the single-vesicle SERS protocol yields distinct Raman spectra from individual sEVs, allowing downstream diagnostic classification and therapeutic drug loading monitoring. Figure 1 depicts the workflow of the diagnostic study for gastric cancer and the drug-loaded sEVs monitoring application using our protocol reported here, respectively.

Discussion

A pivotal aspect of this protocol is the substrate preparation and sEV purification. The performance of label-free SERS depends strongly on producing uniform, high conformal plasmonic substrates with strong enhancement "hot spot". Likewise, high-purity sEV isolation via size-exclusion chromatography is essential to render reliable and reproducible SERS spectra. Impurities such as protein aggregates or lipoproteins not only weaken vesicle signal but can confound downstream machine-learning classification ...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
1x PBS	Thermo Fisher Scientific	J61196.AP
300-mesh copper grid	Electron Microscopy Sciences	EMS200-Cu
4% paraformaldehyde solution	Sigma Aldrich	1004960700
4″ Si wafers	MSE Supplies	WA0810	(100) oriented
A549 cell line	ATCC	CCL-185
AGS cell line	ATCC	CRL-1739
Centrifuge Rotor F-35-6-30, 5430	Eppendorf	22654306
CHA Mark 40	CHA Mark 40	https://chaindustries.com/mark-40-system/	e-beam evaporation
Cr etchant solution	Sigma Aldrich	651826-500ML
Doxorubicin D-4000	LC Laboratories	D-4000_3g
ExoView R100 system and ExoView Human Tetraspanin Kit	NanoView Biosciences	https://www.nowmedicalstudios. com/portfolio/nanoview- biosciences-3d-product-illustrations
FeCl3 solution	Sigma Aldrich	451649
HF solution	Sigma Aldrich	695068-500ML
Hs 738.St/Int cell line	ATCC	CRL-7869
inVia confocal Raman microscope	Renishaw	https://www.renishaw.com/en/ invia-confocal-raman-microscope --6260?srsltid=AfmBOoqbZlyW qv2KDdcXV0UZnp8tdLNgYZ0 GbZvGOWI6z1OKi48rg5-b
KOH solution	Sigma Aldrich	417661
MaxQ 4000 Benchtop Incubating/Refrigerating Shakers	Thermo Scientific	SHKA4000 (4320)
MCO-19AIC incubator	SANYO	8380-30-0035
Millex-GP Filter	MilliporeSigma	SLGPM33RS	0.22 µm
NanoSEM 230 microscope	FEI	SEM-FEI-014
Nanosight NS300	Malvern Panalytical	https://www.malvernpanalytical .com/en/support/product-support /nanosight-range/nanosight-ns300
NCI-N87 cell line	ATCC	CRL-5822
Optima TLX Ultracentrifuge	Beckman Coulter	8043-30-1197
Oxford Plasmalab 80 Plus	Oxford Plasmalab	https://plasma.oxinst.com/ products/rie/plasmapro-800-rie
Polystyrene spheres	Alfa Aesar	AA42711AB	500 nm
qEVoriginal/35 nm	Izon	ICO-35
Single-layer graphene	MSE Supplies	ME0408
T-75 flask	VWR	156800
TF20 High-resolution EM	FEI	https://eicn.cnsi.ucla.edu/project/fei-tf20-tem/
uranyl acetate solution	Electron Microscopy Sciences	22400-1