Physisorption of Affinity Ligands Facilitates Extracellular Vesicle Detection with Low Nonspecific Binding to Plasmonic Gold Substrates

Abstract

Plasmonic biosensors are increasingly being used for the analysis of extracellular vesicles (EVs) originating from disease areas. However, the high nonspecific binding of EVs to the gold sensing surface has been a critical problem and hindered the true translational potential. Here, we report that direct antibody immobilization on the plasmonic gold surface via physisorption shows excellent capture of cancer-derived EVs with ultra-low nonspecific binding even at very high concentrations. Contrary to commonly used methods that involve thiol-based linker attachment and EDC/sulfo-NHS reaction, we showed a higher specific capture rate and >50-fold lower nonspecific on citrate-capped plain and nano-patterned gold surfaces. The method provides a simple, fast, and reproducible means to functionalize plasmonic gold surfaces with antibodies for robust EV biosensing.

Graphical Abstract

INTRODUCTION

Extracellular vesicles (EVs) are membrane-bound nano-sized vesicles actively shed by cells into the circulation.^1,2 EVs are abundantly present in biofluids, structurally stable, and carry biomolecules from their originating cells.³ These properties bring attention to EVs as attractive circulating biomarkers for the diagnosis of cancers,^4–6 as well as neurodegenerative,^7,8 cardiovascular,^9,10 inflammatory,^11,12 and infectious diseases.^13,14 In further exploiting EVs’ potential and accelerating their clinical translation, a critical technical challenge is developing sensitive, robust, and standardized assays that can determine EVs’ composition and molecular profiles in clinical samples.

Among various EV detection platforms^15–20, plasmonic sensors harnessing surface plasmon resonance (SPR) excited on gold substrates or nanostructures have gained interest. The plasmonic sensing approach is simple, sensitive and amenable to high throughput assays.²¹ Plasmonic sensors detect EVs by a resonance shift induced by the increase of local refractive index upon EV binding to a gold sensing surface, which allows for simple, rapid label-free detection.²² Furthermore, plasmonic sensors’ sensing range (typically 10 – 300 nm) matches well with the size of the majority of EVs, boosting the sensitivity for EV detection.²³ For reliable and robust EV detection, a key consideration in the plasmonic assay development is the surface chemistry that minimizes nonspecific bindings of non-target molecules to the sensing surface.²¹ Such nonspecific bindings also increase the local refractive index, causing false-positive signals and adversely affecting the sensor’s limit of detection.²⁴

Gold, the most popular material for plasmonic sensors, is vulnerable to nonspecific molecular adsorption by hydrophobic and electrostatic interactions.²⁵ Thus, self-assembly of thiol-based linkers (e.g., short carbon chains, polyethylene glycol/PEG, dextran polymers) with carboxyl functional groups have been widely used to immobilize affinity ligands on gold films through a 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide/N-hydroxysuccinimide (EDC/NHS) coupling reaction while passivating the surface to reduce nonspecific bindings.^21,26 Even with thiol-based linkers and passivation layers, nonspecific binding still occurs because of i) defective formation of linkers and ii) byproducts formation in the EDC/NHS reaction. The former can occur at defective sites (e.g., grain boundaries, step edges, and vacancies) on a gold surface and result in weak antibody attachment and surface passivation.^24,27 Regarding the latter, byproducts, such as hydrolyzed carboxyl and N-acylurea groups from an EDC/NHS reaction, often interfere with the covalent binding of antibodies to the linkers.^28,29 Such areas with defective linker layers and byproduct formation can be vulnerable to EVs’ nonspecific binding despite surface blocking with bovine serum albumin (BSA), fetal bovine serum (FBS), and other blocking reagents.

In gold nanoparticle-based immunosensors, physisorption of antibodies on citrate-capped gold nanoparticles is a standard method for antibody immobilization, presumably because of its simplicity, good robustness, and comparable sensitivity to chemisorption methods.^30–32 In gold films, however, physisorption is unexplored. Instead, most published reports use sulfhydryl mediated linkers, including mercaptoundecanoic acid (MUA), SH-PEGs, and other SH-linkers with carboxyl functional groups for EDC/NHS coupling.^33–38 These methods add complexity to the functionalization of gold substrate surfaces, are more expensive, and importantly, are not superior to preventing EVs’ nonspecific binding.

Here, we report direct immobilization of antibodies onto plain and nanostructured gold films, showing highly reproducible, specific EV capture while significantly reducing nonspecific binding of EVs to the gold surface. Specifically, we show that direct physisorption of antibodies on citrate-treated plain and nanowell gold films can reduce the nonspecific EV binding by >50-fold compared to conventional chemisorption methods using MUA or SH-PEG-COOH linkers. The ultra-low nonspecific binding and good reproducibility are attributed to the uniform coverage of the surface with antibodies and blocking molecules and simple procedures minimizing variations in the immobilization process. The proposed method has experimentally proven to be not only simple but also more robust and efficient for EV detection.

RESULTS AND DISCUSSION

We first quantitatively evaluated the specific and nonspecific EV binding onto plain gold surfaces prepared by different methods. Physisorption onto a citrate-capped gold substrate (Figure 1A) immobilizes antibodies via hydrophobic and electrostatic interactions between antibodies and the gold surface. We compared this method to two standard antibody immobilization methods using thiol linkers immobilized on the gold surface and subsequent antibody immobilization via EDC and sulfo-NHS. Anti-CD63 and IgG isotype control antibodies were used to measure specific and nonspecific bindings of EVs, respectively. For better visualization and quantification, we isolated EVs from OV90, a human ovarian carcinoma cell line, and labeled the EVs with AZDye-555-tetrafluorophenyl (TFP) esters (please see EV labeling in Methods for details). After incubating EVs on the antibody-immobilized gold surfaces for 60 min and subsequent washing, we imaged and counted captured EVs using a fluorescence microscope (Figure 1B). On the anti-CD63 antibody-coated surface, there were no significant differences in captured EV counts among the three methods (P = 0.17, Kruskal-Wallis test; Figure 1C). However, interestingly, the gold surface with physisorbed IgG antibodies had a one-order lower number of nonspecifically bound EVs (mean EV count = 134.9, n = 9) compared to other chemisorbed IgG antibodies on gold substrates (mean EV counts = 2,430 for MUA and 2,088 for PEG, n = 9; Figure S1). Moreover, the physisorption method showed better reproducibility with a coefficient of variation (CV) of 6.4% (n = 9) compared to the chemisorption methods (18.1% with MUA, 8.6% with PEG, n = 9). We also tested variations in triplicate measurements showing the CV of 6.6% for the physisorption method and 7.5 - 20.4% for the chemisorption methods (Figure S2). We calculated specific capture ratios (i.e., the ratio of between specifically and nonspecifically captured EVs, Figure 1D), and the physisorption method showed over 200 ratios, 17.1 and 13.7 times higher than MUA- and PEG-based chemisorption methods, respectively. We also tested the chemisorption methods on citrate-treated gold surfaces to see if the citrate treatment contributes to the lower nonspecific binding. Figure S3 shows that the citrate treatment on the gold substrate before MUA or PEG immobilization does not significantly decrease the EVs’ nonspecific binding. In addition, we tested different antibody concentrations and found that 20 μg mL−1 showed the highest net EV binding (Figure S4).

Figure 1.

(A) Schematic illustration of physisorption and chemisorption procedures for antibody immobilization on gold substrates. (B) Representative fluorescence images of captured EVs on plain gold substrates coated with anti-CD63 and IgG control antibodies using physisorption and chemisorption methods. For chemisorption methods, mercaptoundecanoic acid (MUA) and 1kDa thiol polyethylene glycol (PEG) was used as linkers. EVs from OV90 cells were fluorescently labeled with AZDye 555. Scale bars = 50 μm. (C) The number of captured EVs on plain gold substrates prepared by different methods. (D) Specific capture efficiencies, defined by the ratio between specifically and nonspecifically captured EVs, are compared between the three methods tested.

The ultra-low nonspecific binding and high reproducibility are likely attributable to the fact that the gold surface was effectively blocked by physisorbed antibodies and BSA (Figure 2). The physisorption method is based on hydrophobic and electrostatic interactions directly between antibodies along with blocking molecules (e.g., BSA) and gold surface (Figure 2A). As a result, we obtained uniform antibody coating, confirmed by fluorescence imaging using AF647 dye-conjugated antibodies (Figure 2C) and good reproducibility for EV capture (Figure 1C). In contrast, chemisorption methods with thiol linkers have a few variables that can cause weak immobilization of antibodies and blocking proteins (Figure 2B) that EVs can displace as nonspecific binding. Chemisorption methods typically involve two steps: self-assembly of a linker layer and covalent antibody conjugation. In this process, non-ideal linker formation and non-covalent antibody attachment can occur for different reasons: First, the self-assembly of thiol linkers on the gold surface may possess defective formation of linkers.^24,27 Gold films prepared by evaporation or sputtering contain many defect sites, such as grain boundaries, step edges, and vacancy (Figure S5). Self-assembly of thiol linkers can be defectively formed at the defect sites (Figure 2B–i). As a result, antibodies can weakly bind to the linkers’ carbon (or polymer) chains (Figure 2B–ii) instead of covalent immobilization. Second, an EDC/NHS reaction is fast and unstable, which can result in the formation of amine-reactive groups (O-acylurea, Figure 2B–iii) and non-amine-reactive groups (Figure 2B–iv).^28,29,39,40 Specifically, rapid hydrolysis of reactive groups can occur during EDC/NHS activation in water, leading to forming carboxyl groups and urea derivatives such as N-acylurea.^28,29 Antibodies cannot be covalently attached to non-amine-reactive groups, and EVs can replace the weakly bound antibodies and blocking materials. Fluorescence imaging showed non-uniform immobilization of AF647-conjugated antibodies on gold films when chemisorption methods were used, possibly due to the defective formation of linkers and the byproduct production. In contrast, the physisorption method produced a uniform antibody coverage on the same gold surface (Figure 2C).

Figure 2.

Schematic illustration of interactions (A) between antibodies and gold film via physisorption and (B) between antibodies and linker/gold film in a chemisorption method. (C) Fluorescence images before and after immobilization of AF647 dye-labeled antibodies on gold substrates prepared by physisorption and chemisorption methods. The images show negligible background signals before immobilization and more uniform antibody coating on the substrate with the physisorption method than with chemisorption methods. Scale bars = 50 μm.

It is noteworthy that antibodies and BSA contains cysteine residues (i.g. free sulfhydryl group) that have a high affinity to gold.^41–43 Thus, in the case of direct antibody immobilization, in addition to hydrophobic and electrostatic interactions, cysteine residues can contribute to irreversible antibody immobilization on the gold surface (Figure S6).^44,45 This may explain why uniform and stable antibody immobilization was achieved using the physisorption method, which led to much lower nonspecific EV bindings.

We next tested the three methods on nano-patterned gold substrates for EV detection. We used gold nanowell structures with 200-nm hole size and 500-nm periodicity (Figure 3A). The nanoplasmonic structures have previously shown excellent EV detection sensitivity.^23,46,47 We applied direct physisorption and chemisorption with MUA or SH-PEG-COOH to immobilize anti-CD63 and IgG control antibodies and measured EV binding on the surfaces. In titrating EV concentrations, while the number of EVs captured on anti-CD63 antibody-coated gold nanowell substrates showed similarly regardless of antibody immobilization methods (Figure 3B), the direct physisorption method showed significantly lower amounts of nonspecific EV binding on IgG-coated gold nanowell substrates compared to those prepared by chemisorption methods (Figure 3C and S7). At a high EV concentration (~109 EV mL−1), the direct immobilization method showed a >50-fold lower nonspecific EV binding (363 ± 43, n = 6) compared to the other substrates prepared by chemisorption methods (18937 ± 3289, n = 6 with MUA; 29439 ± 957, n = 6; P < 0.0001, one-way ANOVA; Figure 3D,3E). With the lower nonspecific binding, the direct immobilization method showed significantly higher specific EV binding, calculated by subtracting EV counts on IgG from the counts on CD63 antibody-coated surfaces, than chemisorption methods (P < 0.0001, one-way ANOVA; Figure 3F). In comparison between plain and nanowell structures, all three methods showed slightly higher nonspecific binding on nanowell structures (a 2-fold increase with the direct physisorption method and a 10-fold increase with the chemical methods). This can be attributed to the curvatures around nanowell structures where the more defective formation of self-assembled linkers can occur, leading to more nonspecific EV bindings. Previous studies also reported that the high curvature surface of gold nanostructures led to the non-uniform linker formation.⁴⁸ Therefore, the physisorption method can be advantageous over chemisorption methods for antibody immobilization on gold nanostructures.

Figure 3.

(A) A scanning electron micrograph of a gold nanowell substrate. (B-C) Representative fluorescence images showing (B) specifically and (C) nonspecifically bound OV90 EVs on anti CD63 antibody- and IgG control antibody-immobilized nanowell substrates, respectively, which were prepared by physisorption and chemisorption methods (scale bars = 25 μm). (D-F) The numbers of captured EVs with titrating concentrations of OV90 EVs on gold nanowell substrates coated with (D) anti-CD63, (E) IgG control antibodies. (F) Net EV counts after subtracting the numbers of nonspecifically bound EVs from those from anti-CD63 coated substrates.

We finally tested the specificity to capture and detect cancer-derived EVs on the nanowell immunogold substrate. We and others previously showed that EpCAM is highly expressed in EVs from ovarian cancer patients differentiated and thus could be used as a diagnostic biomarker.^15,23,49 We aimed to test nanowell plasmonic substrates prepared by the proposed physisorption method that can specifically detect cancer-derived EVs in high backgrounds of EVs from benign cells or normal plasma. We first mixed EVs from ovarian cancer (OV90) and benign (TIOSE4) cell lines in a 1:1 ratio and applied them on nanowell substrates coated with anti-CD63, anti-EpCAM, and IgG control antibodies by the physisorption method. For quantification, we fluorescently labeled OV90 with AZDye 555 and TIOSE4 EVs with AZDye 647, respectively (Figure 4A). We can see that both OV90 and TIOSE4 EVs were captured on the anti-CD63 coated surfaces as CD63 is a universal marker for EVs (Figure 4B). However, on the substrates coated with anti-EpCAM antibody for cancer-derived EV detection, a significantly higher number of OV90 EVs were captured (P = 0.0002, Mann-Whitney t-test). In contrast, the captured TIOSE4 EVs counts were comparable with those on the IgG control antibody-coated substrates (Figure 4C). The result demonstrates specific capture and detection of cancer-derived EVs with little nonspecific binding on the gold nanowell substrates; this can be used to differentiate ovarian cancer from benign cases. We also tested the assay for OV90 and CaOV3, ovarian cancer-derived EVs, mixed with host cell-derived EVs from a healthy normal plasma sample (Figure 4D and E). The results showed that cancer-derived EVs can be effectively captured on the nanowell substrates coated with anti-EpCAM antibodies with little nonspecific binding of host cell-derived EVs below the counts on the IgG control antibody-coated surface. For substrates coated with anti-CD63 antibodies, both cancer cell- and host cell-derived EVs were captured.

Figure 4.

Specificity tests of nanowell substrates prepared by the physisorption method using EVs from ovarian cancer cells mixed with EVs from benign cells or normal human plasma. (A) Representative fluorescence images of nanowell substrates functionalized with anti-CD63, anti-EpCAM, and IgG control antibodies. EVs from OV90 ovarian cancer cells were fluorescently labeled by AZDye 555 (shown green in the images). EVs from TIOSE4 benign cells were fluorescently labeled by AZDye 647 (shown red in the images). Scale bars = 50 μm. (B) A mixture of OV90 and TIOSE4-derived EVs were applied on anti-CD63, anti-EPCAM, and IgG coated nanowell substrates, and the captured EV counts were detected by fluorescence imaging. (C) Specific capture efficiencies defined by the ratio between specifically and nonspecifically captured EVs for OV90 and TIOSE4-derived EVs. (D) The number of EVs captured on anti-CD63, anti-EpCAM, and IgG antibody-coated nanowell substrates. Significantly higher numbers of cancer-derived EVs (OV90 and CaOV3) were captured on anti-EpCAM coated surfaces than those from normal plasma, while their levels on IgG-coated surfaces are comparable. (E) Specific capture efficiencies for EVs from OV90 and CaOV3 ovarian cancer cells and healthy human plasma samples.

CONCLUSION

In summary, we have qualitatively compared different antibody immobilization methods for specific capture and detection of cancer-derived EVs. Interestingly, direct physisorption of antibodies onto the gold surface showed ultra-low nonspecific binding of EVs on plasmonic gold sensor surfaces with excellent reproducibility. Compared to protein and nucleic acid targets, EVs are much larger in size with negative charges of lipid membranes. These properties make EVs easily stuck on gold surfaces by nonspecific interactions. Chemisorption methods with thiol-based linkers have been widely used to immobilize antibodies on plasmonic sensing surfaces. However, the chemisorption surface chemistry involves multiple steps, and the reproducibility can be hampered by multiple factors, such as the purity, concentrations, incubation time, humidity, pH, among others, which can produce weak non-covalent attachment of antibodies. These become vulnerable spots that EVs can bind nonspecifically.

Direct antibody immobilization via physisorption has been considered weak conjugation and thus less popular than chemisorption methods for gold films. However, recent studies showed that the physisorption of antibodies on gold nanoparticles was indeed irreversible in most biological fluids.⁵⁰ In addition, recent reports have shown that cysteine residues exist at antibodies and BSA.^41–43 and can strengthen the direct physisorption to a gold surface.^44,45 Motivated by these new reports and the fact that direct antibody physisorption has been commonly used for citrate-capped gold nanoparticles,^51,52 we tested directed antibody immobilization on citrated-capped gold substrates and performed quantitative analyses in comparison with conventional chemical methods. We showed that direct antibody physisorption can significantly reduce nonspecific EV binding by a >50 fold at a high EV concentration (>109 mL−1). Considering the concentration of EVs in human clinical samples, typically in the range of 109 – 1011 mL−1, this ultra-low nonspecific binding is critical in clinical assays using plasmonic sensors. In a proof-of-concept demonstration, we showed sensitive and specific detection of cancer-derived EVs from those from benign and host cells. The direct physisorption method is simple, fast, and reproducible for antibody immobilization and EV detection with low nonspecific binding using gold plasmonic biosensors. We envision that this method can be more widely used in various clinical testing and thus, accelerate the clinical translation of plasmonic sensors for cancer diagnosis through EV analysis.

MATERIALS AND METHODS

Materials

Tri-Sodium citrate dihydrate (United States Pharmacopeial/USP standards), sodium bicarbonate (ACS reagent, ≥99.7%), Phosphate buffer (PB) solution (0.1 M), Bovine serum albumin (heat shock fraction, protease-free, fatty acid-free, essentially globulin free, pH 7, ≥98%), RPMI 1640 Complete Medium, Sepharose® CL-4B, and 11-mercaptoundecanoic acid (MUA) were obtained from Sigma Aldrich (St. Louis, MO, USA). Alexa Fluor® 647 Mouse IgG1, κ Isotype Ctrl (FC) was purchased from BioLegend (San Diego, CA, USA). Anti-CD63 antibody was obtained from Ancell Corporation (Minnesota, USA). Thiol PEG acid (SH-PEG-COOH, 1,000 Da) and thiol functionalized methoxyl polyethylene glycol, PEG thiol (mPEG-SH, 350 Da) were purchased from Nanocs Inc. (New York, USA). Phosphate buffered saline (PBS, pH 7.4), Tris buffer (1 M, pH 8.0, RNase-free), MES buffer (BupH™ MES Buffered Saline Packs), EpCAM Monoclonal Antibody (323/A3), Mouse IgG1 kappa Isotype Control (P3.6.2.8.1) antibody, Zeba™ Micro Spin Desalting Columns (40K MWCO, 75 μL), 1-ethyl-3-(3-dimethyl aminopropyl) carbodiimide hydrochloride (EDC), N-hydroxysulfosuccinimide (sulfo-NHS), dimethyl sulfoxide (anhydrous, 99.8+%) (DMSO), ethanol (99.5%, ACS reagent, absolute), Fetal Bovine Serum (exosome-depleted) (FBS), and ProLong™ Gold Antifade Mountant were purchased from Thermo Fisher Scientific (Waltham, MA, USA). Sterile human plasma in sodium EDTA (number: D519-04-0050) was obtained from Rockland (Pottstown, PA, USA). Nanodimple (200-nm hole size and 500-nm periodicity) Si3N4 substrate was purchased from LumArray, Inc. (Massachusetts, USA). AZDye 555 DBCO and AZDye 647 DBCO were obtained from Click Chemistry Tools (Arizona, USA). Azido-dPEG®₁₂-TFP ester was purchased from Quanta Biodesign (Ohio, USA)

Preparation of gold substrates

100-nm-thick gold films with 5-nm titanium as an adhesion layer were deposited using an e-beam evaporator on a silicon wafer and a nanowell patterned Si3N4 substrate to obtain plain and nanowell gold substrates, respectively.

Physisorption of antibody on gold substrates

The gold substrate was cleaned with ethanol and deionized water. Then the substrate was immersed in a 100 mM citrate solution for 30 min. An antibody solution (20 μg mL−1 in 10 mM PB, pH 7.0) was pipetted onto the cleaned gold substrate and incubated for 1 h. Please note that the longer incubation time for antibody immobilization with physisorption (1 h in the physisorption vs. 30 min for chemisorption methods described below) is preferred as the direct antibody physisorption is slower than the antibody immobilization with EDC/NHS coupling in the chemisorption methods. Then, the surface of the substrate was blocked with a BSA solution (final concentration: 10 mg mL−1 in 1X PBS) and incubated for 30 min followed by washing. Finally, the substrate was washed using 1X PBS.

Chemisorption of antibody on gold substrates

Two different linkers (MUA or SH-1k-PEG-COOH) were used for the chemisorption of antibodies on gold substrates. For MUA, the gold substrate was cleaned with ethanol, and then, the substrate was immersed in 10 mM MUA solution (in ethanol) for 12 h. It should be noted that the MUA solution was sonicated to disperse aggregated MUA in ethanol before it was used. After the formation of the MUA layer on the gold surface, the substrate was washed with ethanol under sonication. For SH-PEG-COOH, 1 mM SH-PEG-COOH solution (in pH 3.0 Tris buffer), 0.5 mM mPEG-SH (in pH 3.0 Tris buffer), and 0.1% SDS were mixed with the ratio of 1:1:1 and dropped onto the cleaned gold substrate. After incubation for 12 h, the PEG acid-coated gold substrate was washed using deionized water. After linker formation, a mixture solution containing EDC and NHS (final concentrations of EDC and NHS were 0.5 M in pH 6.0 MES buffer.) was freshly prepared and dropped onto MUA-coated gold substrate. After incubation for 10 min, the substrate was washed with deionized water. Then, 20 μg mL−1 of antibody solution (diluted in pH 7.0 PB) was pipetted onto the substrate, and the substrate was incubated for 30 min. For blocking, a blocking solution (10 mg mL−1 BSA and 0.5 M glycine in PBS) was added to the substrate and incubated for 30 min followed by washing. Then, the substrate was washed using PBS.

Cell culture and EV isolation

The human ovarian carcinoma cell lines (OV90 and CaOV3) were purchased from American Type Culture Collection (ATCC), and the ovarian benign cell line (TIOSE4) was obtained from transfection of hTERT into NOSE cells maintained in 1:1 Media 199:MCDB 105 with gentamicin (25 μg mL−1), 15% heat-inactivated serum, and G418 (500 μg mL−1).⁵³ All cell lines were maintained in RPMI-1640 (Hyclone) complete medium supplemented with 10% FBS, 100 U mL−1 penicillin, and 100 μg mL−1 streptomycin (Cellgro) at 37°C in 5% CO2. EV isolation was conducted as previously reported.⁴⁷ Briefly, the cells were cultured until 80-90% in a conditioned medium supplemented with 1% Exosome-depleted FBS, 100 U mL−1 penicillin, and 100 μg mL−1 streptomycin for 48 h. Next, the supernatant was collected and concentrated with Centricon Plus-70 Centrifugal Filter (MWCO = 10 kDa). The concentrate was loaded onto a size-exclusion chromatography column which was packed with Sepharose CL-4B.⁵⁴ The fractions of 4 and 5 of 1 mL were collected and concentrated with the Amicon Ultra-2 Centrifugal Filter (MWCO = 10 kDa). The isolated EVs were stored until use at −80°C.

Plasma EV isolation

Healthy normal plasma samples were obtained from MGH Biobank (IRB# 2019P003472). The plasma samples were centrifuged at 300 g for 10 min at 4°C to eliminate the dead cells and debris. After that, they were centrifuged at 2,000 g for 20 min at 4°C again to separate the apoptotic bodies. The supernatant was passed through with size-exclusion chromatography for taking the 4th and 5th fractions. To concentrate the EV solution, Amino Ultra-2 Centrifugal Filter was used and centrifuged at 3,500 x g for 30 min at 4 °C.

Nanoparticle tracking analysis (NTA) for EV concentration measurements

The concentrations of isolated EVs were measured using the NanoSight LM10 microscope (Malvern). The measurements were conducted with a 642 nm laser module at room temperature. Each sample was diluted 500 fold in PBS and manually placed in the chamber. Briefly, each sample was recorded with a camera for 30 seconds in a quadruplicate and analyzed with NTA software.

Fluorescent labeling of EVs

Isolated EVs were labeled with AZDye 555-PEG-TFP ester and AZDye 647-PEG-TFP ester.⁵⁵ First, AZDye-PEG-TFP ester was prepared by mixing 5 μL of 25 mM AZDye 555 (or 647) DBCO in anhydrous DMSO with 5 μL of 27.5 mM Azido-dPEG also in anhydrous DMSO. The mixture was incubated at room temperature for 2 h. For EV labeling, 3 μL of EV solution (1.3 × 109 EVs mL−1) was mixed with 2 μL of 0.3 M sodium bicarbonate and 0.2 μL AZDye-PEG-TFP ester (10 mg mL−1 in DMSO) and incubated for 1 h. The fluorophore-labeled EVs were washed twice and collected using Zeba™ microspin desalting columns at 1,500 g for 2 min.

Fluorescence EV detection

Fluorophore-labeled EV solution was dropped onto an antibody-immobilized gold substrate. After incubation for 60 min, the substrate was washed with PBS. Then, antifade mountant was dropped onto the substrate. Fluorescence images of captured EVs on gold substrates were obtained using an upright automated epifluorescence microscope (Zeiss AX10, Axio Imager.M2) equipped with a CMOS camera (C15440-20UP, Hamamatsu, Japan). The full size of a fluorescence image is about 374 μm × 374 μm. For EV detection and counting, we used the ComDet plugin in Image J. In the ComDet analysis, we set the approximate particle size (pixel) to 3 and the intensity threshold (particle brightness) to 5. The analysis identifies all fluorescence dots and outputs their location, size, and intensity (Figure S8). For the selectivity tests, OV90 and TIOSE4 (or OV90 and EVs from normal human plasma, or CaOV3 and EVs from normal human plasma) were mixed in 1× PBS (The final concentration of each EV sample was 4×108 mL−1.). The mixed EV solution was used for the selectivity test.