Abstract

Background

   Ovarian cancer (OC) is diagnosed at advanced stages, resulting in
   limited treatment options for patients. While early detection of OC has
   been investigated, the invasiveness of approaches, high sample
   requirements, or false-positive rates undermined its benefits. Here, we
   present a “one-step” high-throughput microfluidic platform for
   epithelial ovarian cancer (EOC) detection that integrates small
   extracellular vesicle (sEV) capture, in situ lysis, and protein
   biomarker detection.

Results

   We identified 1,818 differentially expressed proteins (DEPs) through
   proteomic analysis of sEVs from patients’ serum, combined with cell
   lines. Through multi-step screening of DEPs, we identified EOC
   biomarkers to customize the microfluidic platform. We used the
   microfluidic platform to test the expression of EOC biomarkers with 2
   µL of serum from 209 participants in a prospective cohort. Based on the
   test results, an EOC detection model (P9) was constructed, which
   achieved a sensitivity of 92.3% (95% CI, 75.9–97.9%) for stage I, 90.0%
   (95% CI, 69.9–97.2%) for stage II at a specificity of 98.8% (95% CI,
   93.6–99.8%) in the training set. The specificities reached 98.8% (95%
   CI, 93.6–99.8%) in the training set and 100.0% (95% CI, 91.6–100.0%) in
   the validation set of a held-out group of 105 participants. A model
   combining the P9 and patient’s CA125 value exhibited 100.0% (95% CI,
   95.6–100%) specificity in both training and validation, without
   compromising sensitivity.

Conclusions

   We developed a non-invasive high-throughput microfluidic platform for
   EOC sEV-derived biomarker detection. It significantly reduced false
   positives and sample volume. Given its convenience and low cost, this
   platform could advance OC early detection to benefit of women.

Supplementary Information

   The online version contains supplementary material available at
   10.1186/s12951-025-03348-4.

   Keywords: Microfluid platform, Small extracellular vesicles (sEVs),
   Proteome, Epithelial ovarian cancer (EOC), Cancer early diagnosis

Background

   Ovarian cancer (OC) is the leading gynecologic malignancy responsible
   for high mortality rates, mainly due to its covert symptoms and
   consequently delayed diagnosis. Around 90% of OC cases are epithelial
   ovarian cancer (EOC), one of the most aggressive malignant neoplasms
   [[48]1], whose 5-year survival rate drops from 92% at the early stage
   to 29% at the late stage [[49]2–[50]4]. Despite advancements in cancer
   detection, the early diagnosis of EOC remains remarkably challenging.
   Hitherto, two multicenter clinical trials for OC screening were
   accomplished using transvaginal ultrasonography (TVS) and CA125 [[51]5,
   [52]6]. Intriguingly, a further study revealed that the serial changes
   of CA125 or the Risk of Ovarian Cancer Algorithm (ROCA), outperformed
   CA125 with a fixed cutoff in the same cohort [[53]7, [54]8],
   highlighting the significance of biomarker selection for the EOC
   screening. Using the ROCA, the United Kingdom Collaborative Trial of
   Ovarian Cancer Screening (UKCTOCS), the largest randomized clinical
   trial assessing the effect of screening on an average-risk population,
   exhibited a significant stage shift toward early stages [[55]5],
   rendering more patients the opportunity for curative treatments. It is
   notable that, due to the reported harms of invasive examination for TVS
   and overtreatment for considerable false-positive results, those
   conventional modalities are not recommended as standard of care. Hence,
   approaches that minimize invasiveness and false positives are key to
   benefiting EOC patients from cancer early detection.

   Small extracellular vesicles (sEVs) are present in the body fluids of
   all organisms, carrying primarily proteins and other substrates. sEVs
   derived from tumors contain pathological information and biomolecules
   that reflect the real-time state of living cancer cells, providing a
   relatively complete picture of cancer progression at all stages. This
   makes sEVs an attractive approach for early cancer detection
   [[56]9–[57]11]. sEV-based cancer early detection has been studied in
   lung cancer, hepatocellular carcinoma, pancreatic ductal
   adenocarcinoma, and others [[58]12–[59]16]. However, the enrichment and
   purification of sEVs from limited clinical samples, such as blood or
   urine, hinder the effective detection of sEV-derived biomarkers. Hence,
   more attention has been attracted to modalities for sEVs enrichment and
   purification, such as size-exclusion filtration, immunoaffinity, and
   acoustic-fluid-based technologies, to overcome clinical limitations
   [[60]17, [61]18]. Although some methods aimed to detect cancerous
   biomarkers from isolated sEVs [[62]19–[63]23], these approaches have
   primarily targeted surface proteins, thereby overlooking the potential
   of the enriched proteins within sEVs derived from living cancer cells.
   Therefore, there is a need for a device that can effectively enrich
   sEVs from limited clinical samples and detect targeted biomarkers in
   real-world scenarios.

   Herein, we developed a rapid, cost-effective microfluidic platform for
   early EOC detection, achieving “one-step” sEV cargo detection by
   integrating sEV capture, in situ lysis, and protein biomarker
   detection. Through proteomic analysis of sEVs from EOC patient serum
   and tumor cell lines, we identified 1,818 differentially expressed
   proteins (DEPs) compared to healthy controls. After functional analysis
   and antibody evaluation, 12 elevated proteins were selected as
   biomarker candidates linked to pathways such as Ras signaling and tumor
   metastasis. These biomarkers were used to construct the detection
   platform, which quantified serum biomarkers in 125 patients. A model
   built on 9 biomarkers achieved a specificity of 98.8% (95% CI,
   93.6–99.8%) in the training set and 100.0% (95% CI, 91.6–100.0%) in
   validation, with further improvements when combined with CA125. This
   study demonstrates a versatile platform for sEV-based biomarker
   detection, offering a practical tool for EOC early diagnosis and
   broader biomarker discovery.

Materials and methods

Sample collection

   Between March 2021 and February 2024, we enrolled consecutive OC
   patients who underwent radical surgery in the Department of Obstetrics
   and Gynecology of Peking University Third Hospital. A total of 200
   patients with EOC and 136 healthy controls were prospectively enrolled,
   and 10 mL of blood samples were obtained during pre-radical resection
   surgery. The demographic and clinical information of all participants
   in this study including age, CA125, International Federation of
   Gynecology and Obstetrics (FIGO) stage, grade, and histological
   subtypes was recorded in Table [64]S1.

   The inclusion and exclusion criteria were stated as follows. In brief,
   patients with OC who provided written consent and met the following
   criteria were enrolled: (1) aged between 18 and 75 years and (2) OC was
   newly clinically diagnosed by computed tomography and pathologically
   confirmed EOC. The exclusion criteria were as follows: (1) participants
   who received an organ transplant, prior non-autologous bone marrow
   transplant, or stem cell transplant, (2) participants who were
   diagnosed with other malignancies, (3) participants who had an immune
   disorder or autoimmune disease, (4) participants received any
   anti-cancer treatment 30 days before the expected enrollment date due
   to any non-cancer diseases, (5) participants who receive blood
   transfusion within 30 days before expected enrollment date, and (6)
   pregnant or lactating participants. For healthy individuals, the
   inclusion criteria were (1) provided written consent, (2) aged between
   18 and 75 years, and (3) no malignant diseases. The above exclusion
   criteria also applied to healthy individuals, and those who were
   suspected to have any type of OC were excluded.

   All human materials used in this study were approved by the Medical
   Science Research Ethics Committee of Peking University Third Hospital
   (IRB00006761-M2019291). All participants voluntarily signed a
   declaration of informed consent.

Preparation of sEV-capture microfluidic chips and sEV-detection microarray
chips

   To produce graphene oxide quantum dots (GOQDs)-assembled glass
   substrate, glass slides (CITOTEST, China, P/N. 80302-0001) were
   sequentially subjected to ultrasonic cleaning in deionized water,
   ethanol, and deionized water for 10 min each, followed by drying with
   nitrogen (N[2]) to maintain a clean glass surface. One side of the
   slides was treated with oxygen plasma for 10 min to form surface
   hydroxyl groups, followed by a 45-min immersion in a
   3-aminopropyltriethoxysilane (APTES) solution. The slides were then
   ultrasonically cleaned in deionized water and dried with N[2].
   APTES-treated slides were immersed in a 1 mg/mL GOQD solution for
   45 min, cleaned ultrasonically in deionized water, and dried with N[2].
   The GOQDs were immobilized on the glass surface via the reaction
   between their carboxyl groups and the amino groups on APTES.

   Microfluidic chips were fabricated via soft lithography technology.
   Negative photoresist SU8-2025 (KAYAKU, Japan, Y111069) was used for
   photolithography to create the silicon-based mold for the chip. The
   PDMS prepolymer and curing agent (Corning, USA, Sylgard 184) were mixed
   at a 10:1 ratio, poured into the mold, and cured in a drying oven at
   80 °C for 1 h. The GOQDs-assembled glass substrate was bonded to the
   antibody barcode chip, and the target antibody was introduced to form
   an antibody barcode array on the glass substrate. The antibody barcode
   glass substrate was combined with the microarray PDMS chip to form an
   antibody-barcode microarray chip. Each sample loading well covers a
   whole antibody array. The sEV-capture chip was obtained by bonding a
   through-hole PDMS chip and a GOQDs-assembled glass substrate, with the
   bottom substrate in the through-hole fixed with sEV-capture antibodies.
   For sEV capture, 4 µL of 100 µg/mL CD63 antibody (Sigma, Cat#CBL553)
   was added to each microwell and incubated overnight under a sealing
   film. Removal of incubation solution and followed by blocking with 3%
   BSA for 10 min and sequential washing with 1 × PBS and distilled water
   to remove unbound components. For sEV enrichment, 2 µL of serum sample
   was added to each well and incubated for 60 min, then removed
   supernatant and washed with 1 × PBS. The captured sEVs were lysed in
   situ by adding 5 µL of radio immunoprecipitation assay (RIPA) lysis
   buffer (Cat# 89900, Thermo Fisher Scientific, USA) to each well for
   30 min. For antigen detection, the PDMS microchamber of the microarray
   chip bound GOQDs slides with microprinted antibodies was blocked by 1%
   BSA for 10 min, then, by applying negative pressure and passing through
   the flow channels between the microwells, the sEV lysate was
   transferred to the wells on microarray chip for 40 min of incubation
   and then the supernatant was removed. For fluorescent labelling, the
   microarray chip was peeled off in 1% BSA, and the slide was incubated
   with biotin-containing detection antibody at a final concentration of
   10 µg/mL for 45 min, followed by 30 min of incubation with 100× diluted
   APC-conjugated streptavidin. After thorough washing with 1 × PBS and
   distilled water, the chip was spun to dry. The fluorescent intensity of
   antigens was measured using a fluorescence scanner (GenePix Microarray
   Scanner 4400 A, Molecular Devices LLC, San Jose, CA, USA). Table [65]S2
   summarizes all antibodies used for antigen capture and detection and
   protein standards for calibration.

Cell lines

   The immortalized human ovarian surface epithelial cell line (T1074) was
   purchased from Applied Biological Materials Inc. (Richmond, BC, Canada)
   and cultured in Prigrow I medium containing 10% fetal bovine serum
   (FBS; Biological Industries, Beit-Haemek, Israel). Ovarian cancer cell
   lines SKOV3 cell line and A2780 cell line were purchased from ATCC and
   Sigma and cultured in Dulbecco’s modified Eagle medium (DMEM; Gibco,
   Grand Island, NY, USA) and RPMI 1640 medium (Gibco, Grand Island, NY,
   USA), respectively, both of which were supplemented with 10% FBS.

sEV isolation

   To obtain sEVs from serum samples, the MagicOmics-EMB Exosomes
   Enrichment Kit (Qinglianbio, Beijing, Cat# MagicOmics-EMB8×) was used
   according to the manufacturer’s instructions. Briefly, 500 µL serum was
   mixed with an equal volume of Loading Buffer, and then added 10 µL of
   Reagent A followed by centrifuging at 3,000 g for 10 min at 4 °C.
   Transfer 1 mL of the supernatant to a new 1.5 mL centrifuge tube
   containing equilibrated exosome enrichment magnetic beads (EMB beads).
   Tubes were incubated on a rotator of 10 rpm for 1 h at room temperature
   and then were briefly centrifuged. Placed the tubes back on the
   magnetic separator for 3 min to clarify the liquid and then discarded
   the supernatant. The beads were washed by Wash Buffer 1 and 2
   sequentially. After the removal of the supernatant, the beads with sEVs
   were treated with RIPA lysis buffer (Beyotime, Haimen, China) to obtain
   sEV proteome for the following analysis.

   Differential ultracentrifugation was conducted to isolate sEVs from
   cell lines. In brief, sEVs from different cell lines were obtained from
   an equal volume of conditioned medium derived from the same number of
   cells. Specifically, the cells were initially cultured in complete
   medium until they reached 80% confluence in 15-cm culture dishes. The
   medium was then replaced with 20 mL of exosome-depleted serum
   (CELLiGENT, Uruguay) medium per dish, followed by an additional 48-hour
   incubation period. Finally, a total of 100 mL of conditioned medium was
   collected from five 15-cm dishes for subsequent sEV isolation. In the
   differential ultracentrifugation method, sEVs were isolated from the
   medium through a series of centrifugation steps performed at 4 °C. The
   initial centrifugation was conducted at 500×g for 15 min, followed by
   an additional centrifugation at 2,000×g for 10 min to remove dead
   cells. Cell debris was subsequently eliminated by centrifugation at
   10,000×g for 30 min. The sEVs were then pelleted via
   ultracentrifugation at 110,000×g for 70 min using a Beckman Ti70 rotor.
   The resulting pellet was resuspended in 1 mL of phosphate-buffered
   saline (PBS), filtered through a 0.22-µm membrane, and subjected to a
   final ultracentrifugation at 110,000×g for 70 min to remove soluble
   serum components and secreted proteins. The sEV pellet was resuspended
   in an appropriate volume of PBS, and the protein concentration was
   determined using the bicinchoninic acid (BCA) assay (Thermo Scientific,
   Cat# 23227). The final concentration of sEV-derived proteins was
   adjusted to 5 µg/mL for subsequent experiments.

PKH26 sEV staining procedure

   The PBS solution containing enriched sEVs were mixed with 100 µL of
   Diluent C, followed by the addition of 0.4 µL of PKH26 stock solution.
   The mixture was incubated at 4 °C for 4 min, and the staining reaction
   was terminated by adding 200 µL of 0.5% BSA. To remove unbound dye, the
   mixture was centrifuged at 100,000 g for 1 h, and the supernatant
   containing the dye was discarded. The stained sEVs were resuspended in
   PBS solution.

Nanoparticle tracking analysis

   Nanoparticle tracking analysis was performed using the Nanosight NS 300
   system (NanoSight Technology, Malvern, UK), which is equipped with a
   488 nm laser and a highly sensitive sCMOS camera. sEVs were diluted
   100–500 times to achieve a range of 20–100 objects per frame. After
   being manually introduced into the sample chamber, each sample was
   measured in triplicate using a camera level of 13 (acquisition time:
   30 s; detection threshold: 7). A minimum of 200 complete tracks were
   analyzed per video and the analysis was carried out with NTA analysis
   software version 2.3.

Transmission electron microscopy

   sEV pellets were fixed with 2.5% glutaraldehyde at room temperature for
   10 min. Subsequently, a 5–10 µL drop of sEV suspension was loaded onto
   a copper mesh. After removing excess liquid via filter paper, the sEV
   samples were rinsed three times in PBS for 10 min each. Finally, the
   sEVs on the grid were negatively stained with 1% uranyl acetate (10 µL)
   for 1 min and observed using transmission electron microscopy JEM-1400.

Western blot

   T1074 cells and sEVs harvested from either cell lines or patients’
   blood were lysed in RIPA lysis buffer to extract total protein. Protein
   concentrations were quantified using the BCA protein assay, and 20 µg
   of protein was loaded into each well for Western blot analysis.
   Proteins were separated via sodium dodecyl sulfate polyacrylamide gel
   electrophoresis (SDS-PAGE) and subsequently transferred to
   poly-vinylidene fluoride (PVDF) membranes through electrical means.
   After transfer, the membranes were subjected to a blocking process in
   tris-buffered saline with Tween 20 (TBST) containing 5% skimmed milk
   for a duration of 2 h, followed by overnight incubation at 4 °C with
   primary antibodies in the indicated dilution. Following three washes in
   TBST, the membranes were incubated with 1:1000 diluted horseradish
   peroxidase (HRP)-conjugated secondary antibody at 37 °C for 1 h.
   Lastly, protein bands were detected by an enhanced chemiluminescence
   detection system (Amersham Biosciences). The antibodies used were
   Anti-CD63 antibody (Abcam, Cat# ab193349), Anti-TSG101 antibody (Abcam,
   Cat# ab125011), Anti-GM130 antibody (Proteintech, Cat# 66662-1-Ig),
   Anti-CD9 antibody (Cell signalling Technology, Cat# 13174), Anti-mouse
   secondary antibody (Proteintech, Cat# RGAM001), and anti-rabbit
   secondary antibody (Proteintech, Cat# RGAR001). The ECL
   chemiluminescence solution (Cat# A38554) was purchased from Thermo
   Fisher.

Mass Spectrometry-based proteomics

   For serum samples, 30 µL of lysis buffer was added to each sample well
   and incubated in a metal bath at 95 °C for 10 min. For enzyme
   reconstitution, 100 µL of ultrapure water was added to the enzyme vial
   and thoroughly mixed. Each sample was then supplemented with 100 µL of
   digestion buffer and 2 µL of the reconstituted enzyme solution. The
   samples were placed in a preheated shaking incubator at 37 °C,
   1000 rpm, for at least 4 h. Following digestion, an eight-strip tube
   was placed on a magnetic separator for 10 min until the solution became
   clear. The supernatant was carefully transferred to a new EP tube. To
   quench the reaction, 5 µL of quench buffer was added to the new tube
   and thoroughly mixed, followed by centrifugation at 14,000×g for
   10 min. Peptide desalting was performed using a C18 desalting column.
   The column was activated with 100% acetonitrile, equilibrated with 0.1%
   formic acid (FA), and loaded with the peptide sample. The column was
   then washed with 0.1% FA to remove impurities, followed by elution with
   70% acetonitrile. The eluate was collected and freeze-dried. The dried
   peptide samples were reconstituted in 0.1% FA in water and spiked with
   iRT standard peptides. A total of 1 µg of sample was injected for
   liquid chromatography-mass spectrometry (LC-MS) analysis. Peptides were
   separated using a nano-LC system with a C18 analytical column. The
   chromatographic gradient conditions were as follows: starting at 8%,
   increasing to 12% over 5 min, then to 30% over 35 min, followed by a
   gradual increase to 40% at 44 min. The proportion of mobile phase B was
   then raised to 95% at 45 min and held at 95% until 60 min. The analysis
   was conducted on a Q Exactive HF-X mass spectrometer (Thermo Fisher
   Scientific) equipped with a NanosprayFlex™ (ESI) ion source. The ion
   spray voltage was set to 2.4 kV, and the ion transfer tube temperature
   was maintained at 275 °C. Data were acquired using the data-independent
   acquisition (DIA) mode, with a full MS scan range of m/z 350–1500. The
   mass resolution was set to 60,000 for full MS scans, with a C-trap
   maximum capacity of 3 × 10⁶ and a maximum ion injection time of 50 ms.
   Peptide fragmentation was performed using higher-energy collisional
   dissociation (HCD) with stepped collision energies of 25.5, 27, and
   30%.

   For cell lines, protein samples were prepared according to previous
   studies [[66]24, [67]25]. In brief, sEVs were lysed with RIPA lysis
   buffer. The suspension was analyzed by SDS-PAGE and Coomassie G-250
   (Beijing Wokai Biotechnology Co., LTD, Beijing, China, Cat# [68]D10828)
   staining. After destaining with water, each lane of each sample was
   manually cut into eight bands. These cut gels subsequently underwent
   in-gel digestion and dehydration using a vacuum concentrator. The dried
   peptides were then reconstituted in 0.1% FA and loaded onto C18
   StageTips (Thermo Fisher Scientific, Waltham, USA) conditioned with 50%
   acetonitrile/0.1% FA. After two washes with 0.1% FA, the peptides were
   eluted with a solution of 50% acetonitrile and 0.1% FA and dehydrated
   in a vacuum concentrator. The extracted peptides were analyzed by an
   Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific).
   coupled with an Easy-nLC 1200 UPLC system (Thermo Fisher Scientific).
   Briefly, these peptides were dissolved in 25 µL of solvent A (0.1% FA
   in water) and loaded onto a homemade trap column (100 μm × 2 cm) packed
   with C18 reverse-phase resin (particle size, 3 μm; pore size, 120 Å;
   SunChrom, USA) at a maximum pressure of 220 bar with 12 µL of solvent
   A. The sample was subjected to chromatographic separation on a 150 μm ×
   15 cm homemade silica microcolumn (SunChrom, USA) with a particle size
   of 1.9 μm and a pore size of 120 Å, using a gradient of 11–95% mobile
   phase B (80% acetonitrile and 0.1% FA) at a flow rate of 600 nL/min for
   30 min. The gradient elution conditions were as follows: 11–13% mobile
   phase B for 2 min, 13–32% for 16 min, 32–42% for 7 min, 42–95% for
   1 min, and 95% for 4 min, for a total of 30 min. MS analysis was
   performed on an Orbitrap mass analyzer with full scans (350–1,550 m/z)
   followed by data-dependent acquisition and a mass resolution of
   120,000. In top-speed mode, the ions with the highest intensity were
   isolated in Quadrupole using a 1.6 m/z window and fragmented by higher
   energy collisional dissociation with a normalized collision energy of
   32%. Subsequently, these ions were detected in the Orbitrap at a mass
   resolution of 15,000, with maximum ion injection times of 50 and 22 ms.
   The automatic gain control (AGC) targets for full MS and MS/MS were set
   to 4e5 and 5e4, respectively, while dynamic exclusion time was set to
   30 s. Peptide match and isotope exclusion were enabled.

Identification of differentially expressed proteins

   To identify differentially expressed proteins between EOC patients and
   healthy controls, raw gene counts were identified and normalized with
   HTSeq. Using the “DESeq” package, the corrected p-value < 0.05 and|fold
   change| > 3 for cell lines and p-value < 0.05 and|fold change| > 1.5
   for blood samples were set as the significance criteria, respectively,
   to identify elevated DEP, which were displayed with volcano plots.

KEGG pathway and gene ontology enrichment analyses

   KEGG pathway and Gene Ontology enrichment analyses were performed using
   the “clusterProfiler” R package to identify functions and pathways
   enriched among differentially expressed proteins between EOC samples
   and healthy controls.

EOC diagnostic model construction

   The 9-protein model was constructed employing ridge regression through
   the “glmnet” R package, which included five-fold cross-validation on
   the training dataset to enhance model validity and performance.
   Additionally, to assess the model’s diagnostic accuracy, a receiver
   operating characteristic (ROC) curve was generated using the “pROC” R
   package.

Statistical analysis

   Data normality was assessed using the Kolmogorov-Smirnov test, while
   homogeneity of variance was evaluated using Levene’s test. Normally
   distributed data were presented as the mean ± standard deviation, while
   the other data were presented as the median and interquartile range for
   continuous variables and counts and percentages for categorical
   variables. The one-way Student t-test was used for the comparison of
   normally distributed data while Mann-Whitney U test for the other
   continuous data. The consistency between blood- and cell line-derived
   sEVs proteomes was assessed using Pearson correlation analysis. The
   linear relationship between fluorescence intensity or optical density
   (O.D.) values and target antigen concentrations was measured by
   R-squared (R^2). The p-value < 0.05 was considered statistically
   significant.

Results

Develop a “one-step” microfluidic platform for sEV capture, in situ lysis,
and biomarker detection

   To efficiently capture protein features of sEVs and enable
   high-throughput analysis for early EOC detection, we developed a
   microfluidic platform. This platform integrates sEVs capture, in situ
   lysis, and biomarker detection, expanding the analysis scope beyond the
   sEVs surface. Its performance was validated using a prospectively
   enrolled EOC case-control cohort. A schematic diagram of the study is
   shown in Fig. [69]1.

Fig. 1.

   [70]Fig. 1
   [71]Open in a new tab

   Schematic diagram for microfluidic platform and study design.

   The microfluidic platform mainly contains two functional chips, a
   microfluidic chip for capturing sEVs and in situ lysis and a microarray
   chip for targeted sEV-derived protein detection. The sEV-capture
   microfluidic chip is successively assembled with GOQDs and CD63
   antibodies, and the microarray chip is microprinted with antibody
   microstrip arrays to capture targeted proteins. To detect the target
   antigens in human blood samples, approximately 2 µL of serum is added
   to the platform in each microwell, up to 60 samples per test. The sEVs
   in samples are captured and lysed, and the target biomarker is then
   assessed on a microarray chip. The amount of target biomarkers is
   determined based on its fluorescent intensity detected on the chip by a
   separate scanner. In this study, we constructed an EOC-screening
   microfluidic platform, with 12 biomarkers identified by analyzing LC-MS
   proteomes of sEVs harvested from cell lines and human serum. This
   constructed platform assessed clinical serum samples of a prospective
   cohort. An EOC early detection model was developed and validated based
   on the levels of 12 biomarkers in the cohort

   Due to the scarcity of sEVs in the blood, the power of capturing sEVs
   is the initial “checkpoint” in evaluating the performance of any
   sEV-focused microfluidic device. GOQDs have demonstrated their great
   potential in biosensors, bioimaging, and other biomedical applications
   [[72]26]. We wondered whether GOQDs could facilitate sEV capture and be
   suitable for sequentially in situ sEV lysis. As illustrated in the
   schematic diagram (Fig. [73]2A), we constructed the sEV-capture
   microfluidic chip by bonding a sample-loading polydimethylsiloxane
   (PDMS) chip to a GOQDs-assembled glass slide, allowing for the
   simultaneous analysis of up to 60 samples. To specifically capture
   sEVs, CD63 antibodies were used, as they can better differentiate sEVs
   from other extracellular vesicles [[74]27]. Theoretically, a
   GOQDs-assembled chip immobilizes CD63 antibodies that bind CD63 on the
   sEV surface, thereby efficiently trapping sEVs. To examine this
   hypothesis, we first assessed the distinct concentrations of CD63
   antibodies and incubation periods on the chip. Using fluorescent
   secondary antibodies to label CD63 immobilized on the chip, we
   demonstrated that the maximum sEV capture capacity of the chip was
   achieved when the microwell was incubated with 4 µL of 100 µg/mL CD63
   antibody for 12 h (Figures [75]S1A, B), which was established as the
   standard protocol for the sEV microfluidic chip preparation.

Fig. 2.

   [76]Fig. 2
   [77]Open in a new tab

   Design and validation of microfluidic chips and microarray chips.

   A, Design of microfluidic and microarray chips. The microfluidic chip
   for sEV capture and lysis was comprised of a glass slide assembled with
   GOQDs and a 60-well PDMS microchamber layer, allowing for the
   simultaneous analysis of up to 60 samples per test. CD63 antibody
   solution was used as sEV capture solution in this study, which was
   added to every microwell on the chip and CD63 antibodies would be
   immobilized by GOQDs with incubation. After washing off extra
   antibodies with DPBS buffer, the chip was added with liquid samples
   containing sEVs and incubated for sEV enrichment. After thoroughly
   washing with DPBS buffer, captured sEVs were lysed with lysis solution.
   Then, the lysate of each sample was automatically transferred from the
   microfluid to a distinct microwell on a microarray chip for target
   antigen detection by applying negative pressure through the flow
   channels between the microwells. On the other hand, a microarray chip
   for antigen detection was also fabricated on a GOQDs-assembled glass
   slide substrate with a 60-well PDMS microchamber layer. The antibodies
   targeting biomarkers of interest were previously defined and
   microprinted on the GOQDs slide as a position-encoded array, each of
   whose unit contained a certain number of microstrips to specifically
   capture corresponding biomarkers. Once target biomarkers immobilized
   from liquid samples, such as sEV lysate, a cocktail solution of
   biomarker detection antibodies was loaded and well covered the whole
   chip. With removing unbonded antigens with DPBS buffer, fluorescence
   intensities of the detection antibody targeting each biomarker were
   detected to calculate the level of biomarkers in each sample.

   B, Qualitative validation of microfluidic chips for sEV capture. A
   serial of sEV solution with concentration gradients (10^10 sEVs/mL to
   10^3 sEVs/mL) was added to the chip. The captured sEVs on the chip were
   detected by fluorescent staining of APC-labelled CD63 antibodies (upper
   panel) and SEM (lower panel, 40,000× magnification).

   C, Qualitative validation of microfluidic chips for sEV enrichment
   efficiency. A solution of PKH26-labelled sEVs at 10^9 sEVs/mL was
   loaded onto the chip for validation. When reaching the set incubation
   times, the solution was transferred to a new microwell. The number of
   PKH26-labelled sEVs was counted in 3 areas (100 μm × 100 μm) randomly
   selected in each well.

   D, sEV lysis efficiency analysis. sEVs were labelled by CD63-APC and
   treated with 5 µL lysis solution on the slice for different treatment
   time.

   E, Comparison of antigen detection between microarray chip and
   traditional ELISA assay. The left panel is the fluorescent image of the
   chip. The OPN antigen was diluted according to a 10-fold gradient
   concentration. The fluorescence values were quantified to be fitted to
   a linear curve. LOD for each antigen was calculated

   To assess the sEV capture power of the GOQDs-CD63 antibody-based
   microfluidic chip, a series of solutions with varying concentrations of
   APC-labeled sEVs were introduced into distinct microwells. The captured
   sEVs were detected using a high-resolution fluorescence scanner and
   scanning electron microscope (SEM). A linear correlation was observed
   between the sEV concentrations and fluorescence intensity (Figure
   [78]S1C). Even at a concentration of 10^3 sEVs/mL, the fluorescence
   signals of sEVs were detected, which is significantly lower than the
   concentration of sEVs in serum from healthy individuals and patients
   with cancers [[79]28]. Consistently, chip-captured intact sEVs were
   also observed by SEM in all microwells with various sEV concentrations,
   including 10^3 sEVs/mL (Fig. [80]2B and Figure [81]S1D). These results
   indicate that the chip is highly effective in capturing sEVs.

   Given that immunoaffinity is also affected by interaction time between
   antibodies and antigens, we further investigated the optimal incubation
   time of the sample in the microwells for the maximal sEV capture
   efficiency of the chip. Considering the sEV concentration in healthy
   individuals and the counting of sEV fluorescence under high-resolution
   microscopy, we employed a PKH26-stained sEV suspension with a
   concentration of 10^9 sEVs/mL. To evaluate the efficacy of sEV
   enrichment for the chip, we loaded the PKH26-stained sEV solution to
   several microwells. Once reaching the designated incubation period, the
   residual eluent was transferred to a new microwell. The microwells
   prior to and post liquid-transfer were allowed to air dry. Then, we
   measured the fluorescence signals for sEVs and compared the
   fluorescence intensities of captured sEVs in the microwell and its
   corresponding transferred microwell across different incubation
   durations. As anticipated, with increasing incubation time, more
   fluorescent signals were detected in the microwells, while the
   fluorescence signals in the microwells of transferred eluent
   correspondingly decreased. The fluorescence signals from the incubation
   microwells reached a plateau after 60 min, where prolonged incubation
   time did not significantly improve sEV enrichment (83.6% for 60 min vs.
   84.1% for 120 min; Fig. [82]2C). Therefore, 60 min of incubation is
   suggested for capturing sEVs from human serum samples.

   For sEV lysis, the second “checkpoint” of the sEV-capture chip is the
   efficiency of in situ sEV lysis, which also determines whether there
   will be sufficient biomarkers in sEV lysate for subsequent
   investigations. To assess whether sEVs captured in the chip were lysed
   efficiently, lysis buffer was loaded to the microwells with captured
   sEVs and sequentially cleaned by the buffer. The reduction of CD63-APC
   fluorescence signal labelled on sEVs reflected the lysis process. The
   fluorescent signals of the sEVs weakened over time and vanished after
   30 min (Fig. [83]2D and Figure [84]S1E). This result demonstrated the
   efficacy of in situ sEV lysis process on the chip and suggests the
   optimal lysis treatment time, approximately 30 min, for future
   implementation.

   To integrate a customizable biomarker detection function into the
   automatic sEV analysis system, enabling project-specific analyses.
   Antibodies to specifically capture the target proteins were directly
   microprinted on the GOQDs-assembled chip and formed a position-encoded
   microarray chip. Here, we evaluated the biomarker detection performance
   of the chip by comparing it with a conventional method, ELISA analysis.
   Explicitly, osteopontin (OPN) monoclonal antibodies were microprinted
   on the chip, and the antigen solutions of OPN with varying
   concentrations were added to the chip, whose fluorescent signals were
   presented as a fine line on the microfluidic chip. As illustrated in
   Fig. [85]2E, the intensity decreased in a linear manner with the
   lowered concentration of the antigen. Intriguingly, the chip detected
   the antigen at a concentration of 0.02 pg/mL, whereas the limit of
   detection (LOD) of the ELISA test was 12.29 pg/mL. Notably, there was a
   considerable difference in the requirement of sample amounts between
   the microfluidic chip (2 µL) and ELISA (50–100 µL) per test. These
   findings reflect not only the precision of the chip but also a
   significantly lower demand for the sample amount for analysis using the
   chip. Given the clinical reality of limited patient samples, this chip
   could fulfil essential clinical test requirements without an additional
   burden.

Differential abundance of sEV-derived proteins in patients with EOC and
healthy individuals revealed candidate biomarkers for diagnosis

   To identify the ovarian cancer biomarkers carried in sEVs, we used
   prospectively collected blood samples to capture the signals of
   patients’ status and cell lines reflecting the stable features of EOC.
   We conducted proteomic analyses on sEVs from patients with EOC and
   ovarian cancer cell lines. (Fig. [86]1 and Methods). To harvest sEVs,
   we retrieved the blood samples from 12 patients with EOC and 10 healthy
   individuals and culture supernatant from cell lines of Skov3 (ovarian
   serous cystadenocarcinoma), and A2780 (ovarian endometroid
   adenocarcinoma), and T1074 (ovarian epithelial cells, for control).
   After collecting sEVs from patients’ serum and cell culture
   extracellular fluid, the morphology of collected particles was examined
   by transmission electron microscope (TEM). sEVs in all samples
   demonstrated the presence of rounded, cup-shaped, and double
   membrane-bound vesicle-like structures (Fig. [87]3A). Moreover,
   nanoparticle tracking analysis (NTA) exhibited the majority of the
   particles ranging from 100 to 200 nm in both patients’ serum and cell
   lines, suggesting that the collected particles were mainly sEVs
   (Fig. [88]3B). To further assess the purity of the collected sEVs,
   western blot analysis was performed targeting three sEV-specific
   markers, CD63, CD9, and TSG101 (tumor suppressor gene 101).
   Additionally, GM130, a Golgi marker, was included as a negative control
   to confirm the absence of cellular contamination. This analysis
   confirmed the collection of sEVs from both patients with EOC and cell
   lines (Fig. [89]3C).

Fig. 3.

   [90]Fig. 3
   [91]Open in a new tab

   Proteomic analysis of sEVs harvested from EOC cell lines and patients’
   blood.

   A, sEVs isolated from cell line culture supernatant and human blood.

   B, Nanoparticle tracking analysis of the size distribution and
   concentration for sEVs from cell lines and human blood.

   C, Western blot analysis was performed targeting three sEV-specific
   markers, CD9, CD63, and TSG101 in the cell lysate of T1074 cells and
   sEVs harvested from cell lines and human serum. The Golgi marker GM130
   was used as a negative marker to assess cellular contamination.

   D, Heatmap of differentially expressed proteins among EOC patients,
   cell lines, and healthy controls. Color scale ranges from red to blue
   representing protein levels from high to low. Proteins unidentified in
   the sample are shown in grey.

   E, PCA analysis of proteomic profiles of sEVs isolated from blood
   samples of 12 patients with EOC and 10 healthy controls.

   F, PCA analysis of proteomic profiles of sEVs of two EOC cell lines and
   a control cell line.

   G, Pearson correlation analysis of sEV protein levels between blood
   samples and cell lines.

   H, Volcano plots of DEP in EOC patients and cell lines compared with
   their respective healthy controls.

   I, Venn diagram showing the overlap between blood samples and cell
   lines and all the sEV proteins identified in these two resources by
   LC-MS.

   J, KEGG enrichment analysis of 75 proteins elevated in sEVs derived
   shared between EOC patients and cell lines.

   K, Heatmap of elevated proteins for EOC detection biomarker
   exploration, including the respective top 20 highly expressed proteins
   in EOC cell lines and patients and shared elevated proteins excluding
   those enriched in KEGG pathways of multiple neurodegenerative diseases.
   Color scale ranges from red to blue representing protein levels from
   high to low. Proteins unidentified in the sample are shown in grey.

   L, GO enrichment analysis of the proteins described in (K)

   Implementation of mass spectrometry-based proteomic analysis,
   approximately 3629 and 1088 proteins were identified from EOC cell
   lines and patients’ serum, respectively (Tables [92]S3, [93]S4). Data
   visualization of these proteins was demonstrated in a heatmap
   (Fig. [94]3D). A two-dimensional scatterplot of principal component
   analysis (PCA) revealed that the sEV proteomic profiles of patients
   with EOC generally were clustered apart from those of healthy
   individuals (Fig. [95]3E), consistent with the observation in cell
   lines (Fig. [96]3F). Furthermore, Pearson correlation analysis
   indicated that protein levels do not seem to correlate between these
   two sourses (r = 0.064, p = 0.12; Fig. [97]3G).

   To screen for EOC-associated proteins, we identified 332 and 1321
   differentially upregulated proteins of sEVs from blood samples and cell
   lines, respectively, compared to their corresponding healthy controls
   (Fig. [98]3H). We observed 75 elevated proteins that were shared
   between EOC cell lines and patients (Fig. [99]3I). The KEGG pathway
   enrichment analysis demonstrated these proteins involved in platelet
   activation, intercellular junctions, various degenerative diseases,
   etc. (Fig. [100]3J). These findings are consistent with previous
   studies [[101]29]. Given the low Pearson correlation between patients
   and cell lines as shown above, to better capture sEV proteome
   signatures, we collectively explored EOC-associated biomarkers from 3
   sources: (1) the top 20 upregulated proteins in two EOC cell lines, (2)
   the top 20 upregulated proteins in patient serum samples, and (3)
   proteins enriched in the above KEGG pathways except for multiple
   neurodegenerative diseases. A total of 72 proteins, enriched in
   platelet activation, cell-substrate junctions, cadherin binding,
   vesicle lumen, etc., were selected as a pool of EOC potential
   biomarkers for further investigation. The expression levels of these
   proteins were summarized in a heatmap (Figs. [102]3K, L).

Develop customized EOC biomarker detection microarray chip

   As described above, the biomarker detection chip of this microfluidic
   platform is designed to accommodate custom biomarker-capture antibodies
   for microprinting. Here, we use commercialized antibodies, the most
   common and convenient source of antibodies for most studies. Of the 72
   initially identified proteins, 12 were selected for further analysis
   based on the availability of commercial paired antibodies (capture and
   detection antibodies) that recognize distinct epitopes without steric
   hindrance, as well as corresponding recombinant proteins for
   quantitative calibration. These biomarker-capture antibodies were
   microprinted onto the GOQDs-assembled substrate as a microstrip array.
   The antibody barcode microarray of these 12 antibodies was integrated
   with a 60-microwell PDMS chip, where each microwell contained a unit of
   12 microstrips, one microstrip per target biomarker. This setup allows
   the simultaneous analysis of 12 biomarkers for 60 samples in a single
   run. Before detection, the fluorescent secondary antibody IgG-AF488 was
   added to microstrips lying out of the detection zoom to assess whether
   biomarker-specific capture antibodies had been immobilized on the GOQD
   chip. Only microstrips with immobilized antibodies displayed
   fluorescence. The microarray chip with the microstrip pattern is
   demonstrated in Fig. [103]4A. To ascertain the detection performance of
   this antibody-microprinted microarray, concentration gradient solutions
   for each of the protein standards of these 12 antigens were prepared
   and loaded into distinct wells on microprinted chips, thereby
   determining the specificity, sensitivity and the LODs for each
   sEV-capture antibody. With the corresponding biomarker-detection
   antibody, fluorescent signals were detected only in the microprinted
   lines of the antibody corresponding to the loaded antigens, thereby
   confirming the reliability of the chip in simultaneously detecting
   multiple biomarkers (Figure [104]S3). Consistently, quantitative
   assessment of fluorescently labeled detection antibody for each
   biomarker with standards showing that the fluorescence intensity was
   linearly correlated with antigen concentrations for all the selected
   antibodies, with correlation R^2 ranging from 0.925 (VASP) to 0.9990
   (RAB11B) (Fig. [105]4B). The LOD of each antibody is shown in Table
   [106]S5. The lowest LOD (0.02 pg/mL) was observed in RAB11B, which
   reflects the precision of this chip. Collectively, the above results
   indicate that the microprinted chip of 12 antibodies could
   quantitatively analyze EOC biomarkers in patients’ serum.

Fig. 4.

   [107]Fig. 4
   [108]Open in a new tab

   Construction and validation of microarray chip customized for EOC
   detection.

   A, Photograph of a microarray chip. Blue lines were non-detection ends
   of the 12-biomarker capture antibody for each microwell on the chip,
   labeled with fluorescent secondary antibody IgG-AF488 for pre-detection
   quality control.

   B, Quantitative validation of microarray chips in detecting 12 EOC
   biomarkers. Each antigen was diluted according to a 10-fold gradient
   concentration. The fluorescence values were quantified and fitted to a
   linear curve. LOD for each antigen was calculated

Construct the integrated sEV-based microfluidic platform for EOC biomarker
detection

   Based on the performance of the microfluidic chip for sEV capture and
   lysis, as well as the microarray chip for the detection of 12 EOC
   biomarkers, we developed an integrated sEV-based microfluidic platform
   for EOC biomarker detection.

   To evaluate the overall performance of the platform as a potential EOC
   sEV detection device for clinical utility, we used a cohort of serum
   samples from 125 EOC patients and 84 healthy controls (Table [109]S1).
   Their serum samples were loaded into the ready-to-use sEV-capture
   microfluidic chips for sEV capture and lysis. Then respective sEV
   lysates were transferred to distinct microwells of a microarray chip
   for biomarker capture. The APC-labeled detection antibody cocktail was
   loaded onto the whole chip to identify the captured biomarkers of
   interest. The fluorescence intensity of each biomarker per sample was
   detected simultaneously (Figure[110] S4A). Distinct clusters of
   patients and controls were also observed based on the levels of 12
   biomarkers (Figure [111]S4B). As demonstrated in Figs. [112]5 and 9 out
   of the 12 biomarkers (HEL2, RAB11B, HSPA4, ACTG1, CD147, GLUT3, RHOA,
   CD46, and RAN) exhibited higher signals in the majority of EOC patients
   compared with healthy controls (Figure [113]S4C). In contrast, we did
   not detect SANP23 and VASP in most patients with EOC and controls. The
   average concentration of sEV proteins in patients and the overall LOD
   of each antibody were calculated collectively. Notably, the LOD of GARS
   was higher than its respective average concentration in serum sEVs
   (Fig. [114]5). Consequently, these three biomarkers were excluded to
   ensure more accurate prediction, resulting in nine candidate
   biomarkers. These results also demonstrate the efficacy of the
   integrated microfluidic platform for capturing adequate sEVs with
   efficient lysis to provide sufficient targeted protein biomarkers for
   future analyses.

Fig. 5.

   [115]Fig. 5
   [116]Open in a new tab

   Using the microfluidic platform to detect the expression levels of 12
   EOC biomarkers in serum sEVs from a prospective cohort.

   The expression level of 12 EOC biomarkers in 125 patients with EOC and
   84 healthy controls. Data are presented as mean ± standard deviation of
   three replicates. For each biomarker, the LOD and the average
   concentration of EOC patients are graphed on the right

EOC diagnostic model development and validation

   The conventional biomarkers currently used for EOC diagnosis, CA125 and
   HE4, are not optimal, which exhibit relatively low specificity and
   sensitivity [[117]30]. With this microfluidic platform, we attempted to
   develop an EOC detection model based on the expression levels of the 9
   biomarkers from the aforementioned 125 EOC patients and 84 healthy
   controls, as a training set. To predict EOC in this training set, an
   area under the curve (AUC) of the receiver operating characteristic
   curve (ROC) for each biomarker ranged from 0.488 (RAN; 95% CI,
   0.408–0.568) to 0.950 (GLUT3; 95% CI, 0.921–0.979) (Fig. [118]6A).
   Additionally, we constructed a model (referred to as P9) by combining
   all 9 candidate biomarkers. This model exhibited an AUC of 0.991 (95%
   CI, 0.981–1.00), which was higher than the AUC of any single biomarker
   (Fig. [119]6A). The sensitivity of the P9 model was 94.4% (95% CI,
   88.9–97.3%) at a specificity of 98.8% (95% CI, 93.6–99.8%).
   Furthermore, we evaluated the performance of the P9 model at different
   FIGO stages of EOC. The sensitivity was 92.3% (95% CI, 75.9–97.9%) for
   stage I, 90.0% (95% CI, 69.9–97.2%) for stage II, 95.6% (95% CI,
   85.2–98.8%), and 97.1% (95% CI, 85.1–99.5%) (Fig. [120]6B). High-grade
   serous carcinoma (HGSC) is the most prevalent type of EOC and is
   challenging to diagnose at its early stages. The sensitivity of the P9
   model reached 88.9% (95% CI, 71.9–96.1%) for the early stage (stage
   I–II) and 96.5% (95% CI, 88.1–99.0%) for the advanced stage (stage
   III–IV). (Figure [121]S5A).

Fig. 6.

   [122]Fig. 6
   [123]Open in a new tab

   Construction and validation of EOC detection models.

   A, ROC of each biomarker and their combination (P9) for prediction of
   EOC in the training set.

   B, Performance of P9 model, total and stage-wise in both training and
   validation sets.

   C, Expression level of 12 EOC biomarkers in a validation set of 63
   patients with EOC and 42 healthy controls. Data are presented as
   mean ± standard deviation of three replicates. For each biomarker, the
   LOD and the average concentration of EOC patients are graphed on the
   right.

   D, ROC of the model combining the expression levels of all 9 biomarkers
   in serum sEVs and the level of CA125 in human blood.

   E, Performance of CAP9 model, total and stage-wise in both training and
   validation sets

   To validate the performance of the P9 model, we used another set of 63
   patients with EOC and 42 healthy controls. The expression of each of
   the 12 biomarkers was obtained (Fig. [124]6C). The overall expression
   patten of the 12 biomarkers in this held-out group resembles that in
   the training set. The sensitivity of the P9 model in the validation set
   was 93.7% (95% CI, 84.8–97.5%) when the specificity reached 100.0% (95%
   CI, 91.6–100.0%) and the sensitivity was 91.7% (95% CI, 64.6–98.5%) for
   stage I, 91.7% (95% CI, 64.6–98.5%) for stage II, 95.0% (95% CI,
   76.4–99.1%) for stage III, and 94.7% (95% CI, 75.4–99.1%) for stage IV
   (Fig. [125]6B), and the specificity of the P9 model was 89.5% (95% CI,
   68.6–97.1%) for stage I–II of HGSC and 93.1% (95% CI, 78.0–98.1%)
   (Figure [126]S5A).

   Given the CA125 cut-off as the current EOC clinic biomarker, we used
   the CA125 with a fixed cut-off of 35 U/mL to predict EOC in the
   training set. The AUC of CA125 was 0.910 (95% CI, 0.871–0.948) and when
   specificity was 78.6% (95% CI, 68.7–86.0%), a sensitivity was 85.6%
   (95% CI, 78.4–90.7%) (Figure [127]S5B). Visualizing the predictions of
   EOC positives and negatives of EOC made by CA125 and P9 model
   separately, only 1 control was identified as EOC positive by the P9
   model versus 18 false positives by CA125 solely, so that 17 (94.4%)
   false negatives of CA125 were correctly predicted by the P9 model.
   Additionally, only 1 patient was not correctly predicted by CA125 or P9
   models (Figure [128]S5C). Therefore, the P9 model could dramatically
   increase prediction accuracy for both EOC and non-EOC populations.
   Given that a model based on the CA125 value, rather than a cut-off,
   showed better performance, we developed a CAP9 model combining the
   CA125 value and 9 biomarkers, whose AUC surpassed that of the P9 model
   (Fig. [129]6D). As shown in Fig. [130]6E, the specificity of the CAP9
   model reached 100.0% in both training and validation sets and higher
   sensitivity in stage I and IV in the validation set. These findings
   indicate that an EOC diagnostic model based on the levels of nine
   sEV-derived biomarkers identified and measured by our integrated
   platform may benefit patients with EOC by enabling early detection with
   minimal false positives, particularly when combined with CA125.

Discussion

   Invasiveness and high false positive rates have been reported as the
   main concerns for current large clinical trials of EOC screening. New
   approaches could be the solution for it. Tumor-derived sEVs have long
   been recognized as a promising modality for cancer early detection,
   with their utility demonstrated in several types of cancer [[131]31].
   Despite advancements in microfluidic devices for cancer diagnosis based
   on sEV proteomes, the devices continue to target the antigens on the
   sEV surface, rather than the cargos inside sEVs, which could severely
   undermine the significance and potential of the various cargoes inside
   the sEVs [[132]23, [133]32]. In this study, we developed a microfluidic
   platform integrating a microfluidic chip for “sEV capture and lysis”
   and a microarray chip for “sEV cargo of interest detection and
   analysis” for EOC early detection in prospectively collected EOC blood
   samples. Based on EOC biomarker levels detected by the “one-step”
   microfluidic platform, we developed an EOC diagnostic model, whose
   performance surpassed that of the conventional serum EOC biomarker,
   CA125.

   To develop the microfluidic and microarray platform, we used the
   glass-slices assembled with nano-material GOQDs, which have been
   demonstrated as good substrates for high-efficient antibody
   immobilization [[134]33]. Sample-loading PDMS chip with detection units
   bound to the GOQDs chip allows the device to simultaneously process
   serum samples of up to 60 individuals and 12 biomarkers per sample.
   Meanwhile, we simplified the recipe of sEV capture antibodies by only
   targeting CD63, since CD63 has been reported not only to be more sEVs
   specific but also overexpressed in OC [[135]34]. The consequent sEV
   capture and enrichment analyses have proved the high efficiency of sEV
   capture of microfluidic chips with the sole CD63 antibody. The proposed
   platform is a universal tool to analyze sEV-derived proteins in
   different diseases, organoids or other research systems. More
   importantly, for the first time, this microfluidic platform realized
   sEV collection, in situ sEV lysis, and sEV-derived protein screening
   with extremely small sample volumes. In addition, these sEV capture and
   lysis microarray chips were stable at − 80 °C for 10 months at aspects
   of structure and function, which allows for their promising
   application.

   To calibrate the fluorescent intensity of antibodies printed on the
   microarray chips, the commercialized protein standards were used, and
   its LOD was predicted accordingly. However, 3 out of 12 proteins showed
   quite low levels of expression when analyzing blood samples in the
   microfluidic platform, whose LODs were higher than their average
   protein levels in the samples. This could be due to differences in
   methods, immunofluorescence versus mass spectrum. Meanwhile, these low
   expression levels were consistent among the samples in both training
   and validation sets, where the microfluidic and microarray chips were
   produced at different times, reflecting the robustness of the
   microfluidic platform. Certainly, the cause of this discrepancy is
   worthy of consideration and can be avoided by a pre-test. The 9
   proteins used to construct the EOC diagnostic model have been reported
   in several types of cancer, except for HEL2, an E3 ubiquitin-protein
   ligase. It has been reported to play a key role in ribosome quality
   control [[136]35], while the direct link of HEL2 to cancer remains
   unclear, which might be worthy of future investigation. As a factor
   closely related to the cancer hallmark, the Warburg effect, GLUT3 have
   been reported in a variety of cancer types [[137]36] but has not yet
   been reported as a biomarker in sEVs. In this study, we noticed that
   the sensitivity of GLUT3 was 85.6% (95% CI, 78.4–90.7%) at a
   specificity of 100.0% (95% CI, 95.6–100.0%) in the training set,
   showing its great potential in cancer diagnosis.

   Furthermore, based on the expression level of 9 proteins detected by
   this microfluidic platform, we constructed an EOC diagnosis model, P9,
   the sensitivity of which was 99.4% (95% CI 88.9–97.3%) at a specificity
   of 98.8% (95% CI, 93.6–99.8%). In the validation set, the specificity
   reached 100.0% (95% CI, 91.6–100.0%) with 93.7% (95% CI, 84.9–97.8%)
   sensitivity. The model’s performance in the training and validation
   sets reflects the robustness and reliability of our non-invasive
   “one-step” EOC detection microfluidic platform, which is much higher
   than those of CA125 in our studies, while the performance of CA125 in
   EOC detection in our study already showed better performance than those
   reported previously [[138]37]. The high specificity of the P9 model can
   systematically reduce false positives. To assess the possibility of
   integrating the P9 model into the current clinic operation, the
   training cohort was predicted using the model and the sole CA125
   cut-off (35 U/mL). We observed extremely high accuracy when test
   results were concordant in both methods. However, when results
   differed, the proportion of false results was much smaller in the P9
   model, thereby providing a meaningful reference for diagnosis. The
   model of combined CA125 value and P9 biomarkers reached 100.0%
   specificity. Given the limited sample size in this study, a larger
   cohort is needed to further evaluate this combined model.

   The occult nature of ovarian cancer is the main reason for most
   patients with EOC are diagnosed at advanced stages. Liquid biopsy has
   been developed rapidly for biomarker detection and the amount of sample
   has been a long concern for both medical personnel and patients,
   particularly for blood samples [[139]38, [140]39]. Inevitably, an
   inadequate sample can lead to questionable results such as false
   negative. However, patients under certain physiological conditions are
   not suitable for blood drawn. In this scenario, the dilemma of whether
   the test will benefit or harm the patient must be addressed. In our
   study, we used approximately 2 µL of serum per patient and tested for
   12 biomarkers as needed, which significantly lowered the sample burden
   for more tests to benefit patients. With a proper schedule of medical
   examinations, no extra samples will be required from patients to
   conduct the test. The volume of serum samples makes this microfluid
   platform an ideal candidate for clinical tests in the real world. In
   addition, the cost of the test for each individual is less than that of
   a conventional CA125 test. Therefore, this EOC early detection test
   could be suitable for integration into clinical operations in the
   future.

Conclusions

   In conclusion, this study introduced a one-step high-throughput
   sEV-based microfluidic platform, customized for effective EOC detection
   with low requirement of serum volume and cost. Our findings also
   demonstrate the great potential of the platform for other clinical
   applications by targeting their sEV cargos.

Electronic supplementary material

   Below is the link to the electronic supplementary material.
   [141]12951_2025_3348_MOESM1_ESM.docx^ (12.9MB, docx)

   Supplementary Material 1: Figure S1–S4, Table S1–S2, Table S5
   [142]Supplementary Material 2: Table S3^ (757.6KB, xlsx)
   [143]Supplementary Material 3: Table S4^ (495.5KB, xlsx)

Acknowledgements