In-situ polymerization-mediated glycan density measurement on extracellular vesicle surface for acute myeloid leukemia diagnosis

Abstract

Extracellular vesicles (EVs) have emerged as promising circulating biomarkers for diverse pathologies, yet clinical adoption remains limited due to inter-patient variability in blood EV concentrations, which introduces inconsistent biomarker signals and reduces diagnostic reliability. To address this, we developed an in-situ dopamine polymerization method for precise glycan density quantification on EVs expressing a specific membrane protein. EV membrane proteins were labeled using streptavidin-horseradish peroxidase-conjugated aptamers, while surface lipids and glycans were simultaneously tagged with fluorescent cholesterol and lectin, respectively. Controlled dopamine polymerization spatially restricted fluorescence quenching to EV membranes, enabling glycan density calculation via attenuation profiles of cholesterol- and lectin-bound fluorophores. The method was miniaturized into a microfluidic platform for rapid (< 1 h), wash-free point-of-care analysis. In 47 clinical specimens (16 patients with acute myeloid leukemia, 15 patients with benign hematological diseases, and 16 healthy donors), glycan density normalization reduced inter-patient variability compared to absolute measurements. Multi-lectin analysis of CD133⁺ EV glycan density achieved superior diagnostic accuracy (AUC = 0.904) in distinguishing acute myeloid leukemia from benign hematological diseases, outperforming direct glycan quantification. By integrating surface biophysical normalization, this platform enhances EV biomarker consistency, providing a rapid, blood-based diagnostic tool with clinical translatability for acute myeloid leukemia.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12951-026-04400-7.

Introduction

Acute myeloid leukemia (AML), as the most common type of leukemia, is a hematological malignancy that initiates from myeloid stem cells within the bone marrow and is characterized by the uncontrolled production of abnormal white blood cells [1, 2]. Its aggressive progression poses a significant threat to patient outcomes, necessitating timely diagnosis. Traditional diagnosis methods, such as bone marrow aspiration and biopsy, are often invasive, imposing additional physiological burdens on patients [3–5]. Consequently, there is an urgent need to search for blood biomarkers for non-invasive AML diagnosis, yet current biomarkers have shown limited accuracy and specificity.

Extracellular vesicles (EVs) are nanosized vesicular structures actively secreted by cells [6, 7]. They carry various biomarkers from their parent cells, including nucleic acids, proteins, and glycans [8–11]. Critically, unlike free-floating biomarkers, EV-associated cargo is shielded by a lipid bilayer membrane, conferring resistance to enzymatic degradation and significantly extending its half-life in blood circulation. This intrinsic protection underpins the enhanced stability, integrity, and detectability of EV biomarkers, thereby positioning EVs as robust and clinically translatable diagnostic biomarkers across diverse pathological conditions [12–15]. In AML, aberrant glycan patterns - whether displayed on the cell membrane or attached to glycoproteins - drive pathogenesis through multiple pathways. These include distorting chemokine and tyrosine kinase signaling to promote proliferation and bone marrow homing [16], altering stromal glycans in the marrow niche to foster an immunosuppressive microenvironment [17, 18], and strengthening leukemic adhesion to endothelium and extracellular matrix, thereby supporting survival, quiescence, and drug resistance [19, 20]. Since EVs inherit glycan patterns from their parental cells, EV‑surface glycans represent ideal circulating biomarkers for AML diagnosis, combining disease‑specific biological relevance with the protective stability provided by EV encapsulation.

The clinical application of EV glycans as diagnostic biomarkers for AML is limited by two interconnected issues: significant heterogeneity in EV concentration and a lack of suitable analytical methods. Baseline levels of total blood EVs differ drastically across individuals, spanning up to six orders of magnitude, while the abundance of disease-specific EV subpopulations also varies widely among patients [21, 22]. These combined sources of variability lead to inconsistent glycan expression patterns, even in patients with comparable clinical conditions. Meanwhile, current EV analysis typically relies on a “separate-then-detect” workflow, involving isolation steps such as ultracentrifugation or immunoaffinity capture, followed by separate concentration measurement and glycan profiling [23–25]. This process is not only time-consuming and equipment-dependent but may also compromise EV integrity during handling. Although a growing array of glycan analytical tools has been developed in recent years [26–32], no integrated diagnostic platform currently exists that can simultaneously quantify EV concentration and profiling their surface glycan composition.

Polymerization has been employed to develop rapid and sensitive analytical tools due to its fast reaction rate and significant physical/chemical differences between monomers and polymers [33–36]. Dopamine, a catecholamine, can oxidize and self-polymerize in ambient conditions to form a polydopamine (PDA) layer on nano-templates of various natures, morphologies, and sizes [37, 38]. Additionally, dopamine self-polymerization can be accelerated by horseradish peroxidase (HRP) to enable localized polymerization near HRP [39]. Motivated by these observations, we developed an analytical platform, named in-situ polymerization for EV glycan density evaluation (IPEG), to measure glycan and lipid content of EV expressing a specific membrane protein.

In this approach, EV membrane components were labeled with fluorescein isothiocyanate (FITC) conjugated cholesterol (CLS), rhodamine isothiocyanate (RITC) conjugated lectin, and HRP conjugated aptamer (APT-HRP). Then, dopamine and H₂O₂ were added directly to the above mixture - without prior removal of unbound fluorophores - to initiate HRP catalyzed dopamine oxidation and polymerization. Due to HRP’s spatially confined enzymatic activity and the template effect of EVs, dopamine polymerization preferentially occurred on the EV surface. This in-situ dopamine polymerization can quench fluorophores anchored to the membrane while leaving free-floating fluorophores unaffected (Fig. 1a). This selective quenching effectively suppresses background fluorescence arising from unbound fluorophores, thereby obviating the requirement for a post-labeling wash step. Consequently, glycan density at the EV surface can be quantified by the fluorescence decrement ratio between FITC and RITC (Fig. 1b). This tool was further integrated into a microfluidic platform for analyzing clinical samples in a point-of-care manner (Fig. 1c). Using this platform, we analyzed glycan density of EVs in AML patient plasma and observed reduced signal fluctuation compared to raw glycan signals. Importantly, CD133⁺ EV glycan density signals successfully distinguished AML patient samples from healthy controls, demonstrating the platform’s excellent diagnostic performance in biofluids. Overall, we developed a rapid, wash-free, and multi-dimensional analysis platform that mitigates EV biophysical and biomolecular heterogeneity in blood, showing promise for clinical translation.

Fig. 1.

(a) The schematic illustration of in-situ polymerization and fluorescence quenching processes. (b) The glycan density of EV was calculated by normalizing the fluorescence intensity reduction of lectin-RITC (ΔF_GLY) against that of CLS-PEG-FITC (ΔF_LIP). (c) Photograph of a microfluidic chip design for IPEG assay (Scale bar = 1.0 cm). The chip was used to analyze plasma samples of patients with AML and construct diagnostic models for AML

Materials and methods

Materials

DNA sequences were purchased from Sangon Biotech (Shanghai) Co., Ltd., and were stored under manufacturer`s guidance. The aptamer DNA sequences against CD63 (CD63-APT), nucleolin (NUC-APT), and CD133 (CD133-APT) were established by previous publications (Table S1) [40–42]. Diacrylate poly(ethylene glycol) (DAPEG, Mw = 1,000), polyethylenimine (PEI, Mw = 600), CLS-polyethylene glycol-FITC (CLS-PEG-FITC, PEG Mw = 2,000), RITC (98%), dopamine (98%), 1-(3-dimethylaminopropyl)-3-ethylcarbodiimide (EDC), N-hydroxysuccinimide (NHS), NHS-biotin (98%), H₂O₂ solution (30%), triethylamine (98%), N, N′-methylenebisacrylamide (99%), and (3-aminopropyl)triethoxysilane (APTES, 95%) were purchased from Aladdin Scientific Corp. Concanavalin A (ConA), Jacalin, Lens culinaris hemagglutinin (LCA), Phaseolus vulgaris leucoagglutinin (PHA-L), Peanut agglutinin (PNA), and Pisum sativum agglutinin (PSA) were purchased from Beijing InnoChem Science & Technology Co., Ltd. The binding moieties of lectins used in this article were listed in Table S2 [43]. Bicinchoninic acid assay (BCA), radio immunoprecipitation assay (RIPA) with protease inhibitors, streptavidin-HRP, TMB single-component substrate solution, and ELISA stopping solution were obtained from Shanghai Acmec Biochemical Technology Co., Ltd. and were used under manufacturer’s guidance. EV depleted fetal bovine serum (dFBS) was prepared by ultracentrifugation according to standard protocols [44].

Cell culture and EVs harvesting

THP-1, HepG2, and HUVEC cell lines were obtained from the National Collection of Authenticated Cell Cultures in Shanghai. THP-1 cells were cultured in RPMI 1640 medium (Gibco) supplemented with 10% dFBS and 1% penicillin-streptomycin (Gibico). HepG2 and HUVEC cell lines were cultured in DMEM medium (Gibco) supplemented with 10% dFBS and 1% penicillin-streptomycin. For harvesting EVs, 5.0 × 10⁶ cells were cultured with 10 mL culture medium for 48 h at 37 °C. Then, the cell culture medium was filtered through a 0.22 μm filter (Millipore), followed by sequentially centrifugation at 1,000 g for 5 min, 10,000 g for 10 min, and 100,000 g for 2 h. The collected EVs were stored at -80 °C for future use. The EV concentrations were analyzed by a nanoparticle tracking analysis system with identical settings for all samples (Fig. S1).

Lectin and DNA modification

For labeling lectin with RITC, 300 µL lectin solution (10.0 mg/mL) was incubated with 200 µL RITC solution (1.0 mg/mL) at 25 °C for 24 h in dark. For preparing biotin conjugated lectin, 100 µL lectin solution (10.0 mg/mL) were incubated with 10 µL NHS-biotin (0.1 mg/mL) at 25 °C for 12 h. The obtained lectin-RITC and lectin-biotin were purified by an ultra-centrifugal filter (Amicon, MWCO = 3,000). Note the conjugation of lectin to glycans on the EV surface was validated by incubating EVs with lectin for 30 min, followed by dynamic light scattering measurement of resulting EV-lectin complexes (Fig. S2).

For conjugating HRP to aptamer DNA sequence, 50 µL solution of aptamer with biotin at the 5` end (1.0 µM) was incubated with 50 µL solution of streptavidin-HRP (0.1 µg/mL) for 2 h at 37 °C. Then, the mixture was treated by a Zeba™ spin desalting column (Thermo, MWCO = 40,000) to remove residuals. The obtained APT-HRP was concentrated by an ultra-centrifugal filter (Amicon, MWCO = 10,000).

For labeling aptamer sequence with RITC, 500 µL aptamer solution (10.0 µM, with primary amine at 5` end) was incubated with 50 µL N, N′-methylenebisacrylamide solution (10.0 mM) and 2.5 µL triethylamine for 24 h at 37 °C. Then, aptamer sequence was further reacted with PEI (Mw = 600, 5.0 mM) for 24 h at 37 °C. Finally, the PEI modified aptamer sequence was incubated with RITC (4.0 mg/mL) for 12 h at dark to give RITC labeled aptamer (APT-RITC). After each step, the aptamer solution was treated by an ultra-centrifugal filter (MWCO = 3,000) to remove residuals.

EV membrane labeling process monitoring

The EV membrane was labeled by intercalating CLS-PEG-FITC into the lipid layer [45, 46]. For monitoring the kinetics of EV membrane labeling process, 10 µL CLS-PEG-FITC solution (1.0 µM) and 2 µL CD63-APT-RITC (0.5 µM) were incubated with 50 µL dFBS containing 0.67 ⋅ 10¹²/mL EVs for different time intervals. Then, the mixture solution was treated by a Zeba™ spin desalting column (Thermo, MWCO = 40,000) to remove residuals. The FITC and RITC fluorescence intensities of the eluted solution were recorded by a plate-reader to monitor the labeling process of CLS-PEG-FITC and APT-RITC (Fig. S3).

Glass slide modification

Circular glass slides (diameter = 6 mm) were immersed in boiling piranha solution [7 : 3, H₂SO₄ (98%) : H₂O₂ (30%)] for 30 min, followed by washing with ethanol, water, and acetone repeatedly. Then, the glass slides were immersed in 50 mL APTES solution (5% in ethanol) supplemented with 0.25 mL triethylamine for 4 h. After incubation, the glass slides were washed by ethanol, water, and acetone repeatedly, followed by drying in an oven. For anchoring CD63-APT to glass slide surface, the APTES modified glass slides were sequentially incubated with DAPEG solution (10.0 µM in PBS) and CD63-APT (10.0 µM in PBS, with primary amine at 5` end) for 24 h at 37 °C. After each step, the glass slides were washed by PBS.

Sandwich enzyme-linked lectin assay (ELLA) and BCA

Sandwich ELLA was carried out by capturing EVs onto glass slide surface by CD63-APT and probing glycans by lectin-HRP activated reaction. In a typical procedure, 50 µL THP-1 cell culture medium was incubated with CD63-APT coated glass slides for 1 h at 37 °C. Then, the glass slides were sequentially incubated sequentially with 50 µL lectin-biotin solution (1.0 µg/mL) for 30 min and streptavidin-HRP solution (5.0 mg/mL) for 30 min. After each step, the glass slides were washed repeatedly by PBS buffer. Finally, TMB single-component substrate solution and ELISA stopping solution were used to treat the glass slides under manufacturer`s guidance. For measuring the protein content of EVs, BCA was applied to the EVs captured at glass slide surface under manufacturer`s guidance.

IPEG assay

The IPEG assay was performed by the triplet labeling of EVs and subsequent in-situ dopamine polymerization. In the first step, 50 µL THP-1 cell culture medium was incubated with 12.0 µL solution of NUC-APT-HRP (0.01 µg/mL) for 45 min at 25 °C. Then, the mixture was further incubated with 10.0 µL CLS-PEG-FITC solution (1.0 µM) and 10.0 µL ConA-RITC solution (50.0 µg/mL) for 45 min at 37 °C. In the second step, 23.0 µL dopamine solution (10.0 mg/mL), 1.0 µL H₂O₂ solution (30%), and 13.0 µL TRIS solution (10 mM) were incubated together with the mixture solution to trigger the in-situ dopamine polymerization. After 30 min of incubation, the FITC (λ_ex = 495, λ_em = 520 nm) and RITC (λ_ex = 570, λ_em = 595 nm) fluorescence intensities of the mixture solution were recorded by a plate-reader. For negative control, the scramble sequence of NUC-APT was also conjugated with HRP (NUC-SCR-HRP) and was used to treat THP-1 cell culture medium under the same procedure. The glycan and lipid content of NUC⁺ EVs were defined as ΔF_GLY and ΔF_LIP, respectively. ΔF_LIP was calculated as: ΔF_LIP = 1 – F_APT−FITC/F_SCR−FITC, where F_APT−FITC and F_SCR−FITC were the FITC fluorescence of cell culture medium treated by NUC-APT-HRP and NUC-SCR-HRP, respectively. ΔF_GLY was calculated as: ΔF_GLY = 1 – F_APT−RITC/F_SCR−RITC, where F_APT−RITC and F_SCR−RITC were the RITC fluorescence of cell culture medium treated by NUC-APT-HRP and NUC-SCR-HRP, respectively. The glycan density at EV membrane surface was calculated as: glycan density = ΔF_GLY/ΔF_LIP.

Microfluidic chip fabrication

The cover layer of the microfluidic chip was fabricated from PMMA sheet by a CO₂ laser engraver (Atomstack K60). For the microchannel layer, the SU-8 photoresist was spin-coated onto a silicon wafer and patterned under UV exposure to obtain the casting mold. PDMS and crosslinker were mixed at a ratio of 10 : 1 and were poured on the casting mold, followed by curing at 60 °C overnight. Subsequently, the microchannel layer was peeled off the casting mold and was punched by a hole puncher to generate inlet and outlet ports. The cover and substrate layers were functionalized with APTES, followed by oxygen plasma treatment of all layers to form a sealed device. The microfluidic chip was designed with dimensions of 65 mm (length) × 30 mm (width) × 6 mm (height). The exploded view of the fabricated chip was shown in Fig. S4.

Device manipulation

The detailed work-flow of microfluidic chip was illustrated in Fig. S5. Briefly, 10.0 µL cell culture medium (or plasma sample) and 2.4 µL solution of APT-HRP (0.01 µg/mL) were introduced via inlet 1 and inlet 2, respectively. The solution flowed through the serpentine channel at a flow rate of 2.0 µL/min and was incubated in the microfluidic channel to facilitate the labeling of EV membrane protein by HRP, with a total duration of 15 min for both processes. Then, a mixture solution containing 2.0 µL CLS-PEG-FITC solution (1.0 µM) and 2.0 µL lectin-RITC (50.0 µg/mL) solution was introduced via inlet 3, flowed through the second serpentine channel at 2.0 µL /min, and incubated in the microfluidic channel for lipid and glycan labeling respectively, with a total duration of 15 min for both processes. Finally, a mixture solution containing 4.6 µL dopamine solution (10.0 mg/mL), 0.2 µL H₂O₂ solution (30%), and 2.6 µL TRIS solution (10 mM) was introduced via inlet 4 (flow rate 4.0 µL/min) and was incubated in detection chamber to enable the in-situ dopamine polymerization, with a total duration of 10 min for both processes. Finally, the fluorescence intensities of reaction solution were recorded by a plate-reader. The workflow of IPEG was completed within 40 min.

Clinical analysis

Blood samples were collected from 16 patients with AML (M2 and M3, FAB classification), 15 patients with benign hematological diseases (BHD), and 16 healthy donors (patients baseline characteristics summarized in Table S3). Blood were collected in EDTA tubes and were immediately centrifuge at 1,000 g for 5 min to harvest upper plasma layer. The obtained plasma samples were treated by a 0.22 μm filter and were centrifuged at 1,000 g for 10 min and 10,000 g for 10 min. Plasma sample were stored at -80 °C until use. For analyzing clinical sample with IPEG, 10 µL plasma sample was treated with the same labeling and polymerization steps described in “Device manipulation”.

Western blotting

RIPA assay was used to lyse EVs according to manufacturer`s guidance, followed by using BCA to measure the protein concentration. The obtained EV proteins were denatured by boiling for 10 min and were resolved through SDS-polyacrylamide gel electrophoresis. After being transferred to a poly(vinylidene fluoride) membrane, the proteins were labeled with primary antibodies against CD63 (Thermo), ALIX (Bioss), HSP70 (Thermo), and APOB (BOSTER). Finally, the proteins were treated with HRP-conjugated goat anti-mouse IgG antibody (Biomass) for chemiluminescence analysis with a BIORAD ChemiDOC™ XRS imaging system.

Instrument

Solution absorbance and fluorescence were recorded by a plate-reader (Thermo Fisher Scientific). Hydration diameters of nanostructures were analyzed by a dynamic light scattering system (NanoBrook 173Plus, Brookhaven). The ultracentrifugation for EVs harvesting was performed on an Avanti JXN-30 (Beckman) at 4 °C. EV concentration was analyzed by a nanoparticle tracking analysis system (ZetaView, Particle Metrix). Transmission electron microscopy (TEM) analysis was performed by a Talos L120C G2 (Thermo) without staining.

Statistics

All experiments were performed in triplicate and presented as mean ± standard deviation. The two-tail Student’s t-test was applied to evaluate the significance of the data, with P < 0.05 considered as significantly difference. Variance inflation factors were computed to assess the severity of multicollinearity among clinical data (Table S4). Sample size calculation was performed with α = 0.05 and power (1 – β) = 0.95 (Table S5). Recovery rate analysis was performed to evaluate the reliability of IPEG (Table S6). Leave-one-out cross-validation was employed to construct multiple linear regression scoring models based on the following equation:

where α_i is the regression coefficient, V_i is the measured value, and β is the intercept (Table S7). Receiver operating characteristic (ROC) curves were generated for individual glycan density and combinations of six glycan densities using the constructed multiple linear regression scoring models. Euclidean distance was employed to calculate the cut-off point on the ROC curve by utilizing the closest-to-(0,1) criterion (Table S8).

Results and discussion

In-situ polymerization and fluorescence quenching

Following a 30-min dopamine polymerization (Fig. S6), the hydration diameters of nanostructures in the EV solution increased with rising dopamine concentration, reaching a maximum of 155.9 ± 45.6 nm at 120 mM dopamine (Fig. 2a). In contrast, PDA nanoparticles formed in PBS under the same dopamine concentration range showed significantly lower hydration diameters (< 50 nm). This demonstrated those large nanostructures formed in EV solution was due to the formation of PDA layer at EV surface. Moreover, a marked reduction in nanostructure sizes was observed in EVs solution upon increasing dopamine concentration to 150 mM and 200 mM, resulting from EV template saturation and the formation of smaller PDA nanoparticles [47]. To optimize the balance between EV-templated polymerization efficacy and suppression of non-templated PDA nanoparticle formation, a dopamine concentration of 120 mM was selected for subsequent experiments. The TEM image of EVs before in-situ polymerization showed typical vesicular structures with a high-contrast periphery and a bright center (Fig. 2b). Following polymerization, the EVs maintained their spherical morphology with enhanced peripheral electron density, confirming successful PDA layer deposition. Control polymerization in PBS alone generated small PDA nanoparticles (14.1 ± 3.1 nm), unequivocally demonstrating the essential templating role of EVs. Notably, localized dopamine polymerization on the EV surface was further corroborated by TEM visualization of a PDA layer enveloping CD63-immunogold-labeled EVs (Fig. S7).

Fig. 2.

(a) Hydration diameters of THP-1 EV solution and PBS after being reacted with different concentrations of dopamine for 30 min. (b) The TEM images of THP-1 EV before and after in-situ dopamine polymerization and the TEM image of PBS buffer after dopamine polymerization. Scale bar = 100 nm. (c) The fluorescence spectra of EV solution after being reacted with different amount of dopamine. EV was labeled with CD63-APT-HRP, CLS-PEG-FITC, and ConA-RITC before polymerization. (d) The FITC fluorescence intensity of THP-1 EV (1.0 ⋅ 10¹⁰/mL) and CLS-PEG-FITC mixture in dFBS under the treatment with PBS, CD63-APT-HRP (APT), and CD63-SCR-HRP (SCR). # indicated EV were not spiked into dFBS. (e) The RITC fluorescence intensity of THP-1 EV (1.0 ⋅ 10¹⁰/mL) and ConA-RITC mixture in dFBS under the treatment with PBS, CD133-APT-HRP (APT), and CD133-SCR-HRP (SCR). # indicated CD133 (0.1 µg/mL) was spiked into EV solution. * indicated significant difference, P < 0.05. Experiments were conducted in triplicate (n = 3)

Due to its π-π conjugation, PDA exhibits strong absorbance in the visible and near-infrared regions, thus potentially exerting a strong quenching effect on various fluorophores [48, 49]. To utilize this unique optical property of PDA layer, EVs were further labeled with CLS-PEG-FITC and lectin-RITC. Upon dopamine polymerization, gradual reductions in the fluorescence intensities of CLS-PEG-FITC and ConA-RITC were observed for EV solution upon reaction with increasing dopamine concentration, which indicated successful quenching of fluorophores by PDA layer (Fig. 2c). The fluorescence reduction was specifically observed in CD63-APT-HRP treated EVs, whereas CD63-SCR-HRP treated EVs maintained FITC fluorescence intensities comparable to PBS treated controls. This confirmed the PDA layer’s quenching effect on EV surfaces and validates its specificity toward CLS-PEG-FITC intercalated in EV membranes (Fig. 2d). Critically, CD63-APT-HRP treated dFBS showed negligible fluorescence reduction, underscoring the necessity of EV templates for dopamine polymerization-mediated quenching. Similar quenching specificity was observed against ConA-RITC bound to glycan at EV surface, for fluorescence intensity reduction was only exhibited in EV solution treated by CD133-APT-HRP, while CD133-SCR-HRP treatment cannot induce fluorescence quenching effect against ConA-RITC (Fig. 2e). To further investigate the influence of free-floating glycoprotein, CD133, a glycoprotein enriched in D-mannose [50], was supplemented to above THP-1 EV containing dFBS. After dopamine polymerization, the same RITC quenching effects were observed for EV solutions with/without free-floating CD133, which proved the dopamine polymerization cannot quench ConA-RITC conjugated to free-floating glycoprotein.

Collectively, the wash-free nature of the IPEG assay is achieved through spatially controlled polymerization-induced fluorescence quenching. Unlike conventional methods requiring multiple washing steps to remove unbound probes, our platform leverages a selective quenching effect enabled by the localized deposition of PDA on EV membranes. This selectivity ensures that only fluorophores anchored to EVs (e.g., CLS-PEG-FITC in lipids and lectin-RITC on glycans) are quenched, while free-floating fluorophores in solution remain unaffected. As illustrated in Fig. 2d, e, the fluorescence attenuation is negligible in controls using scramble sequences or non-EV environments (e.g., dFBS or CD133-spiked solutions).

IPEG assay optimization and validation

To optimize the quenching efficacy of in-situ polymerization, systematic investigations were conducted on the influence of H₂O₂ and APT-HRP concentrations on fluorescence intensity differences between APT-HRP and SCR-HRP processed samples. Experimental data from Fig. 3a, b demonstrated maximal fluorescence variance at optimized reagent quantities: 1.0 µL of 30% H₂O₂ and 12.0 µL of 0.01 µg/mL APT-HRP. Since both H₂O₂ and APT-HRP can accelerate dopamine polymerization, these results suggested a moderate polymerization rate benefits the quenching effect on the EV surface.

Fig. 3.

(a and b) The FITC fluorescence intensities of EV solution after in-situ dopamine polymerization with different volumes of H₂O₂ (30%) (a) and APT-HRP solution (0.01 µg/mL) (b). (c and d) The titration curves of EV lipid (c) and glycan (d) obtained by titrating against known concentrations of EVs. The FITC and RITC fluorescence intensity were recorded for the titration of lipid and glycan, respectively. (e) The correlation analysis between EV lipid content measured by IPEG assay and the EV protein content measured by BCA. (f) The correlation analysis between EV glycan content measured by IPEG assay and that measured by ELLA. (g) The comparison of singleplex and duplex IPEG assay. The lipid (upper) and glycan (lower) of EVs in the culture media of THP-1, HepG2, and HUVEC were analyzed by IPEG assay individually (singleplex) or simultaneously (duplex). * indicated significant difference, P < 0.05. Experiments were conducted in triplicate (n = 3). CD63⁺ EV, NUC⁺ EV, and CD133⁺ EV represented EVs expressing CD63, nucleolin, and CD133, respectively

Through calibration with these optimized parameters, the IPEG assay demonstrated enhanced sensitivity with detection limits of 0.377 × 10⁷ /mL for lipid (Fig. 3c) and 0.538 × 10⁷ /mL for lectin targets (Fig. 3d). Additionally, the IPEG assay demonstrated linear detection ranges of 1.0 ⋅ 10⁶ − 1.0 ⋅ 10⁹ EV/mL for ΔF_LIP measurement (Fig. S8). Validation studies revealed strong correlation coefficients between lipid-specific fluorescence differences (ΔF_LIP) and conventional BCA protein quantification (R² = 0.9701), confirming ΔF_LIP’s utility for EV enumeration in complex biological matrices (Fig. 3e). Parallel analysis of glycan-specific signals (ΔF_GLY) exhibited R² = 0.9224 concordance with established ELLA methodology, verifying the platform’s capacity for precise glycan profiling (Fig. 3f). The ΔF_LIP/ΔF_GLY signals obtained from IPEG exhibited a linear relationship with the EV protein concentration-normalized ELLA signals (Fig. S9), further validating the precision of IPEG in quantifying glycan density on the EV surface. Comparative evaluation of assay configurations demonstrated equivalent ΔF_LIP/ΔF_GLY outputs between singleplex and duplex IPEG formats, while duplex detection achieved significant temporal efficiency gains in signal acquisition (Fig. 3g and Fig. S10). Notably, the IPEG method exhibited enhanced analytical reliability for both EV lipid and glycan analyses. Recovery rates ranged from 92.29% to 109.03% for lipid analysis and from 87.39% to 114.91% for glycan analysis (Table S6), while relative standard deviations were 4.72% and 5.40% for lipid and glycan analyses, respectively (Fig. S11). Furthermore, since viruses (10–500 nm) fall within the detectable size range of our method [28], IPEG is amenable to viral glycan profiling, provided that appropriate experimental adjustments are implemented.

Clinical sample analysis

As one of the most aggressive leukemia, the non-invasive diagnosis of AML has drawn a great attention in the past decade. To investigate the performance of IPEG assay in identifying circulating biomarkers for non-invasive AML diagnosis, IPEG assay was applied to analyze the lipid and glycan content of plasma samples collected from 16 patients with AML, 15 patients with BHD and 16 healthy donors. NUC-APT-HRP and CD133-APT-HRP were used to label NUC⁺ EV and CD133⁺ EV, respectively, as NUC and CD133 were two typical transmembrane proteins used in leukemia theragnostic applications [51–53]. The ΔF_LIP and ΔF_GLY values strongly correlated with the BCA and ELLA results, respectively (R² = 0.9425 and 0.9119; Fig. S12), confirming that the IPEG assay yields quantitative data consistent with established clinical methods. Drastic fluctuation in ΔF_LIP signals were observed for both NUC⁺ EV and CD133⁺ EV in the plasma samples of AML, BHD, and healthy donor groups (Fig. 4a, b). This corroborated previous reports of broad plasma EV concentration ranges observed in cancer patients. The ΔF_LIP signal fluctuation was also observed in the plasma of healthy donors, which may be attributed to the different expression levels of NUC and CD133 by healthy donors under different physiological conditions. Due to the huge signal fluctuations in ΔF_LIP, tremendously varied EV glycan expression levels were observed in the plasma of AML patients, BHD patients, and healthy donors (Fig. 4c), making it meaningless to further evaluate their diagnostic performance.

Fig. 4.

The lipid contents (ΔF_LIP) of NUC⁺ EV (a) and CD133⁺ EV (b) in plasma samples of AML, BHD, and healthy donor groups. (c) The heatmaps of EV glycan expression levels (ΔF_GLY) of NUC⁺ EV (upper) and CD133⁺ EV (lower) in plasma samples of AML, BHD, and healthy donor groups

To minimize the EV glycan signal fluctuation caused by the aberrant EV numbers in plasma, the glycan signals were divided by EV lipid signals for analyzing EV glycan density. As shown in Fig. 5a, the observed fluctuations in ΔF_GLY/ΔF_LIP were significantly attenuated for both NUC⁺ EV and CD133⁺ EV. Moreover, variance inflation factors for the ΔF_GLY/ΔF_LIP ratio were below 3.0 across all six lectins (Table S4), indicating negligible multicollinearity among the glycan density signals. Notably, a clear trend of higher ΔF_GLY/ΔF_LIP values was observed for CD133⁺ EVs in the plasma of AML patients compared to that of BHD patients and healthy donors. The ROC curves were plotted for evaluating the diagnostic performance of ΔF_GLY/ΔF_LIP. For NUC⁺ EVs, the area under the curve (AUC) values for individual lectins ranged from 0.663 to 0.722 in distinguishing hematological disease patients from healthy donors, and from 0.571 to 0.638 in distinguishing AML patients from BHD patients (Fig. 5b). Meanwhile, the combined AUC for the six-lectin panel was 0.734 for distinguishing hematological disease from healthy donors and 0.650 for distinguishing AML from BHD, respectively. In comparison, CD133⁺ EV demonstrated enhanced diagnostic performance: ConA, LCA, PNA achieved AUC values exceeding 0.800 in distinguishing hematological disease patients from healthy donors, whereas ConA and Jacalin achieved AUC values above 0.800 in differentiating AML patients from BHD patients (Fig. 5c). When applied as a six-lectin panel, the mix AUC of CD133⁺ EV reached 0.923 for hematological disease versus healthy donors and 0.904 for AML versus BHD. Furthermore, the optimal cut-off points for the six-lectin panel of CD133⁺ EV yielded a sensitivity of 0.871 and specificity of 0.875 in distinguishing hematological disease patients from healthy donors, and a sensitivity of 0.813 and specificity of 0.867 in distinguishing AML patients from BHD patients (Table S8). These results demonstrated ΔF_GLY/ΔF_LIP outperformed absolute ΔF_GLY in AML diagnosis and implied the glycans associated to CD133⁺ EV were potential circulating biomarkers for AML diagnosis.

Fig. 5.

(a) Heatmaps depicting EV glycan densities (ΔF_GLY/ΔF_LIP) of NUC⁺ EV (upper) and CD133⁺ EV (lower) in plasma samples from patients with AML, patients with BHD, and healthy donors. The ROC curves for glycan densities (ΔF_GLY/ΔF_LIP) of NUC⁺ EV (b) and CD133⁺ EV (c) in distinguishing plasma sample from: (i) hematological disease vs. healthy donor (left); and (ii) AML vs. BHD (right)

Conclusion

This study developed a microfluidic platform to analyze the glycan density of EVs by utilizing EV template-based and HRP-controlled in-situ dopamine polymerization. This method generates a PDA layer specifically on the EV surface, inducing fluorophore quenching at the EV membrane. By fluorescently labeling EV membrane lipids and glycans, the platform quantifies EV lipid and glycan content in AML patient plasma, deriving glycan density signals through normalization of glycan to lipid signals. These normalized signals demonstrate reduced variability compared to raw glycan data and improved diagnostic performance for AML. While promising, limitations of this work include a small cohort and reliance on specific aptamers/lectins. Future work requires validation in larger cohorts and expansion to diverse glycan/protein targets. The platform represents a rapid, sensitive point-of-care diagnostic tool with potential for diverse cancer applications.

Author contributions

X.W., H.Z, J.X., and X.S. conceived the project and designed the experiments. X.W., W.L., and L.Z. conducted experiments and analyzed data. C.L., X.H. and L.T. performed cellular analysis. J.X. collected clinical samples and performed clinical investigations. X.W., H.Z, J.X., and X.S. wrote the manuscript.

Funding

We acknowledge the financial support by National Natural Science Foundation of China (Grant No. 22464007, 82260044, and 52375577) and Guizhou Provincial Natural Science Foundation (Grant No. ZK[2021]482).

Data availability

The datasets generated and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics statement

The clinical experiments were carried out in accordance with the ethical guidelines of the Helsinki Declaration and were approved by the Ethics Committee of Guizhou Medical University (No. 2020-10). Informed consent was obtained from all participants involved in this study.

Competing interests

The authors declare no competing interests.