Anion-exchange chromatography for EV fractionation

Liquid chromatography performed using a monolith anion-exchange column and applied to medium from cancer cell culture supernatant resulted in the separation of three distinct elution peaks (Fig. 1a). Apparently, all the three fractions contained EVs because they were positive for CD9 when analyzed by flow cytometry (Fig. 1). An EV enrichment step obtained submitting the cell culture supernatant to ultrafiltration and successive precipitation by means of the Total Exosome Isolation kit [35] induced a significant modification of the chromatographic elution profile (Fig. 1b). Four peaks were detected, with peaks 1 and 3 sharing the same retention volume of the homologous peaks eluted from the untreated cell culture supernatant. Peaks 2 and 4 have no direct correspondence in the elution profile from the untreated cell culture supernatant (the retention time of peak 2 differed in the two samples, peak 4 was detected only in the second sample) but were also positive for CD9 (Additional file 1: Figure S1). Strikingly, the precipitation step significantly reduced the protein content of peak 1 (from 60 to 3 mAU), while the protein content of peak 3 remained constant (15 mAU). The thorough characterization of the molecular content of the different fractions will be the object of a further project, but these preliminary results already enabled to establish that EV surface charge is sufficient to separate distinct sub-classes. Their further characterization relies on the availability of another orthogonal purification method, such as immunopurification exploiting antibodies for specific surface antigens.

Panning of nanobodies on EV-enriched fractions

An effective method for isolating antibodies able to recognize an antigen discriminating between two very close cell types is the blind in vitro selection known as differential panning. A pre-immune library is first incubated with the control population and all the antibodies bound to the membranes are discharged (depletion step). The unbound fraction is then challenged with the target cells and the antibodies that remain in solution are discharged, whereas the bound antibodies are recovered after elution (enrichment step) [30]. The approach has the further advantage of selecting binders that recognize the native conformation of the antigen because the selection is performed directly on whole cells and not on recombinant protein. We argued that it could be possible to adapt the protocol used for cells to EVs. We used the EV-enriched fraction corresponding to peak 1 obtained from chromatography of HEK293 supernatant for depletion and the material from peak 1 obtained from chromatography of SKBR3 supernatant for enrichment. In both cases, EV-enriched samples were bound to magnetic beads to simplify the washing steps. After two rounds of panning, 92 clones were analysed by phage ELISA on EVs recovered from both SKBR3 and HEK293 cells (Fig. 2a). This screening method uses directly culture media enriched in secreted phages and, despite the fact that the phage quantity is not normalized among the samples, is sufficient to discriminate between negative (14) and positive clones (78) which recognized EVs from both cell lines, usually with a preference for one of them. We did not identify binders with exclusive specificity for SKBR3 and selected 10 clones with high signal among those that apparently bound better to SKBR3-derived EVs. The sequencing results indicated the presence of five unique sequences. These underwent further validation by flow-cytometry in combination with commercial antibodies against CD9 (Fig. 2b). The data indicated two distinct binding behaviours for the nanobody clones. Whereas the two antibodies seemed to bind independently when the commercial anti-CD9 was tested with B1, F7, and D5, in the presence of the clones H1 and H6 the binding of the anti-CD9 appeared to be inhibited. Since CD9 is one of the canonical EV biomarkers, showing that we selected anti-CD9 nanobodies would indicate that direct panning on EVs is not only feasible and yields functional binders, but that at least part of them are directed against EVs, despite the low purity of the material used for the panning [19]. A consideration is necessary here to explain the lack of SKBR3-selective binders. We imputed it to the elevated loss of EVs from the beads during the washing steps that we observed. Under these conditions the library depletion was probably insufficient for eliminating clones able to recognize shared epitopes.

H1 and H6 VHH-GFP constructs compete with anti-CD9 antibodies

To prove that H1 and H6 were anti-CD9 antibodies, these clones were first sub-cloned in an expression vector that allows the production of fusion proteins composed of nanobody, eGFP, and C-term 6xHis tag. The preliminary small-scale expression test showed that the constructs were produced in soluble form in E. coli (data not shown) and was used to set the optimal growth conditions. Purification of VHH-GFP constructs (42 kDa) from soluble bacterial fractions by metal-affinity chromatography resulted in non-homogenous preparations (Additional file 2: Figure S2) which needed an anion-exchange chromatographic step to separate contaminants from the fluorescent immunoreagents (Additional file 2: Figure S2). Final yields were in the range of 10 mg of homogeneous VHH-GFP per litre of medium and these purified constructs were used in competition experiments at flow-cytometry. Binding comparison to EVs derived from different cell lines confirmed that some clones (H1 and B1) had clearly differing binding preferences, an indication that they recognize different antigens on the EVs (Additional file 3: Figure S3A–C).

Latex beads coated with EV-enriched fractions recovered from HEK293, SKBR3, and Jurkat cells were tested for their CD9 positivity with a commercial monoclonal antibody (Additional file 3: Figure S3D), then were incubated either with anti-CD9–PE antibodies or with both anti-CD9–PE and VHH-GFP antibodies (Fig. 3). As seen by the shift in FL2 fluorescence intensity, the addition of H1 nanobodies significantly decreased the binding of anti-CD9 to beads coated with EVs. This was confirmed with EVs derived from all the three cell types and the H6 clone. In terms of labelled beads (Fig. 3), the presence of H1-GFP inhibited the CD9 binding from 36.86 ± 9.95 to 11.46 ± 1.82% (HEK293-EVs) and from 45.78 ± 8.42 to 12.18 ± 3.89% (SKBR3-EVs), while H6-VHH competition resulted in a decrease of anti-CD9 bound beads from 49.55 ± 4.21 to 13.40 ± 1.73% (HEK293-EVs) and from 52.86 ± 5.14 to 7.59 ± 0.86% (SKBR3-EVs). Controls such as an irrelevant nanobody (nbVHH) and the B1 clone which was assessed as a CD9-neg but EV-pos in the preliminary screening (Fig. 2b) did not affect significantly the CD9 binding to the vesicle-coated beads (Fig. 3). The capacity of the monovalent VHHs to compete successfully with the bivalent commercial monoclonal IgG was observed at conditions of almost 200 times molarity excess for the nanobodies in solution.

Nanobody-mediated EV capture on solid surfaces

The results of the competition experiments seemed to indicate that H1 and H6 are anti-CD9 nanobodies. This feature enables their use as specific binders for EVs for improving the stability of their binding (to beads, to ELISA plate, to biosensor surfaces, etc.) and their selective capture when contaminated EV fractions are available. As a proof-of-principle, H1/H6 nanobodies were used for immune-purification of EVs from both cell culture supernatant and human serum. The nanobodies were coated onto latex beads and used to bind the EVs present in the samples. EVs were then released by lowering the buffer pH. Beads were incubated with 20 µg of EV-fraction proteins recovered from supernatant of HEK293, SKBR3, and Jurkat cells. The yields (9–11 µg) were constant for all the samples. In the case of human plasma, immune-purification resulted in 42 µg of exosomal protein per 250 µL of undiluted plasma. A qualitative evaluation of the recovered EVs was performed by TEM (Fig. 4). This analysis identified vesicles with diameter between 50 and 200 nm in all the observed samples.

Next, we evaluated if the nanobody-mediated capture could increase the amount of EVs stably bound to the latex beads used for flow-cytometry. Both H1 and H6 coated beads were used in combinations with samples of cell culture supernatant collected from HEK293 and SKBR3 cells and the immune-captured EVs were used to bind anti-CD9–PE antibodies. The amount of captured EVs was finally assessed indirectly by measuring the PE-dependent fluorescence intensity (Fig. 5a). A clear increase was measured for both tested nanobodies, with constant higher fluorescence values obtained when using material from SKBR3 cell cultures. Both nanobodies bound to latex beads were successful in capturing EVs also from human plasma of healthy donors (Fig. 5b). Despite the elevated background signal and the low concentration of EVs in these samples [36], the fluorescence value increased from 5.37% (control) to 14.05% (H1-coated beads) and from 4.93% (control) to 14.89% (H6-coated beads). An irrelevant nanobody (nb-VHH) used as a control was not able to capture vesicles from human plasma. Purified H1 and H6 nanobodies were also able to bind strongly to the fraction 1 separated by IEX chromatography and linked to latex beads (Additional file 4: Figure S4), in agreement with the data previously obtained with monoclonal anti-CD9 that confirmed the presence of EVs in this sample.

Next we assessed the possibility to improve the reproducibility of EV ELISA by using H1- and H6-mediated nanobody immune-capture instead of direct EV coating on microplate surface to reduce the vesicle removal during the washing steps. In the experimental setting, EVs from cell culture supernatant were first immobilized in the microplate wells – directly or by means of anti-CD9 nanobodies-and later used to screen phage displayed nanobodies according to their affinities for such EVs (Fig. 5c). Bound phages were visualized using anti-M13 HRP antibodies. The false negative (EVs from HEK293 cells) and the high standard deviation (EVs from SKBR3) values obtained by direct EV coating indicated highly unstable conditions and/or consequent low data reproducibility. H1/H6-mediated EV immune-capture eliminated the variability and enabled to collect reproducible data. Probably, EVs directly coated on the microplate plastic are apparently prone to be washed away and adopting this methodology resulted in misleading results. In detail, the absorbance (A₄₅₀) values describing the E10 clone binding capacity to HEK293-derived EVs passed from 0.18 ± 0.05 (direct coating) to 2.60 ± 0.35 and 2.49 ± 0.53 for H1- and H6-mediated EV capture, respectively. In the case of the SKBR3 EVs, direct coating yielded an A₄₅₀ value of 1.16 ± 1.46, whereas nanobody-mediated EV coating resulted in A₄₅₀ values of 2.49 ± 0.53 and 1.73 ± 0.31.

Cell culture

Human embryonic kidney (HEK-293), breast adenocarcinoma (SKBR3), and T cell lymphoma (Jurkat) cells were obtained from the American Type Culture Collection (ATCC) (Rockville, MD, USA). HEK-293 and SKBR3 cells were grown in Dulbecco’s modified minimal essential medium (DMEM) supplemented with 10% v/v heat inactivated fetal bovine serum (FBS), 100 U/mL penicillin and 100 µg/mL streptomycin (Gibco, Thermo-Fisher Scientific) at 37 °C under a 5% CO₂/95% air atmosphere at constant humidity. Jurkat cells were grown in RPMI medium supplemented with 10% v/v heat inactivated fetal bovine serum (FBS), 100 U/mL penicillin and 100 µg/mL streptomycin (Gibco, Thermo-Fisher Scientific) at 37 °C under a 5% CO₂/95% air atmosphere at constant humidity.

Purification of EV-enriched fractions from cell culture supernatant

SKBR3 and HEK293 cells were grown in DMEM medium supplemented with FBS (not EV-free) until they reached 70–80% confluence. Jurkat cells were grown in RPMI medium supplemented with FBS (not EV-free) for 3 days. The culture medium was then removed and the cells rinsed with PBS before the addition of FBS-free DMEM medium in the case of HEK-293 and SKBR3, or FBS-free RPMI in the case of Jurkat cells. After 24 h, the conditioned cell culture supernatant was transferred into 50 mL polypropylene tubes and centrifuged 30 min at 300×g at 4 °C to pellet floating cells and filtered through 0.45 µM syringe filter to remove cell debris and aggregates. The supernatant was collected into a sterile glass bottle and used for chromatographic separation using an ÄKTA pure 25 system (GE Healthcare) in combination with a 1 mL CIM QA-1-monolith column with pores 6 µm in diameter (BIAseparations) equilibrated in 20 mM Tris–HCl, pH 8.03. Two millilitre of exhausted media were used, corresponding to a culture of roughly 1 million cells, diluted 1:10 in 20 mM Tris–HCl buffer, pH 8. After a 10 mL washing step with the equilibration buffer, a 20 mL linear salt gradient (0–100%) was obtained by adding 20 mM Tris–HCl, pH 8.03, 2 M NaCl at a flow rate of 5 mL/min. The eluted material was monitored detecting the absorbance at 280 nm. As an alternative, an EV-enriched fraction was purified using the Total Exosome Isolation kit (Life Technologies, Thermo-Fisher Scientific), according to the manufacturer’s instruction and ultrafiltration using an ultra-centrifugal cartridge with cut-off of 100 kDa (Amicon, Merck).

Direct panning on EV-enriched fractions

Panning was performed following a modified version of previously described methods [29]. Nanobody-displaying phages (3 × 10¹¹)—diluted in 1 mL of PBS containing 2% skimmed milk—were first depleted by incubating them twice 30 min in the presence of 25 µL of milk-blocked naked beads and each time the bound fraction was discharged. The unbound phage fraction was transferred to the HEK293 EV-coated beads and incubated for 1 h. EV-coated beads were prepared using epoxy magnetic beads (25 µL; Life Technologies, Thermo-Fisher Scientific). Epoxy magnetic beads were washed in PBS and finally resuspended in 25 µL of the same buffer before adding 2 µg of the EV-enriched fraction recovered by chromatographic separation of the material present in HEK293 cell culture supernatant. After overnight incubation, the still reactive sites on bead surfaces were blocked 30 min at room temperature with 1 mL of PBS, 2% milk powder. Pre-selected phages were incubated with EV-coated beads for 1 h at room temperature and the unbound fraction was recovered. The depletion procedure as repeated and finally the unbound phage fraction was incubated with beads coated with EV-enriched fraction recovered by chromatographic separation of SKBR3 culture media (see above). After 20 washing cycles in PBS, bound phages were eluted by adding 1 mL of 200 mM glycine, pH 2.2 containing 1 mg/mL BSA and neutralized in 150 µL of 1 M Tris–HCl, pH 9.1. The phages were used to infect TG1 cells that were spread over 2xTY Petri dishes. The resulting colonies were recovered the successive day and the corresponding rescued phages used for a second panning cycle.

Ev elisa

The phages produced by the colonies recovered after the second panning cycle were screened by ELISA for their capacity to bind EVs. An irrelevant clone was used as a negative control. Microtiter plates (96-well, MaxiSorp, Nunc™) were coated by incubating overnight at 4 °C with 2 μg/well (100 µL/well in PBS) of total protein from EV-enriched fractions. Plates were blocked (1 h at 37 °C in PBS buffer plus 5% (w/v) skimmed milk) and phages pre-incubated 1 h at 37 °C in PBS buffer plus 1% (w/v) skimmed milk. One hundred microliter per well of blocked phages were added to the plates and incubated 1 h at room temperature. After washing (4 times × 5 min in PBS), 100 μL of mouse anti-M13 antibodies conjugated with HRP (dilution 1:5000, GE Healthcare) were added to each well and incubated 1 h at 37 °C. Reporter signal was obtained by adding 100 µL of TMB solution (Sigma Aldrich) after another washing step (4 times × 5 min in PBS). The colour reaction was stopped by 50 μL of 2 NaH₂SO₄. Absorbance at 405 nm was recorded by using a HTS7000 Bioassay reader (Perkin Elmer).

Nanobody subcloning and production

Nanobody sequences were subcloned into a modified pET14b vector using NcoI and NotI enzymes to obtain fusion constructs in which nanobodies are linked at their C-terminal with both eGFP and 6xHis tag [27]. These vectors were transformed into E. coli BL21 (DE3) hosting the plasmid for the expression of sulfhydryl oxidase and DsbC [44, 45]. Antibodies were produced as previously described with some modifications [27]. Briefly, 2 mL of over-night pre-culture were used to inoculate 500 mL of LB broth in the presence of the 100 µg/mL ampicillin and 25 µg/mL chloramphenicol. Bacteria were grown at 37 °C until OD_600nm reached 0.4. Sulfhydryl oxidase and DsbC expression was induced by adding 0.5% (g/mL culture) of arabinose and the temperature was lowered to 30 °C. After 30 min, 0.1 mM of IPTG was added to induce antibody expression, the bacteria were grown overnight at 21 °C, harvested, and frozen. Bacterial pellet was resuspended in 20 mL of 100 mM Tris–HCl, pH8, 500 mM NaCl, 2.5 mM MgCl₂. Lysozyme (0.5 mg/mL) and DNAse (3U) were added and the lysate was kept 30 min at room temperature. Samples were sonicated and finally centrifuged at 18,000×g for 20 min at 4 °C. Supernatant was filtered (0.45 µM) and loaded on a 5 mL HiTrap Talon Crude column (GE Healthcare) connected to a chromatographic ÄKTA pure 25 system (GE Healthcare). The column was equilibrated with 50 mM phosphate buffer, pH 7.4, containing 500 mM NaCl and 5 mM imidazole, while bound proteins were eluted using 50 mM phosphate buffer, pH 7.4, containing 500 mM NaCl and 150 mM imidazole. Fractions containing VHH-GFP constructs were pooled and buffer was exchanged into 30 mM Tris–HCl buffer, pH 8.3, by using a 5 mL HiTrap Desalting column (GE Healthcare) in combination with a chromatographic ÄKTA pure 25 system (GE Healthcare). Further purification of the VHH-GFP construct was achieved using a Mini Q 4.6/50 ion exchange column (GE Healthcare) mounted on a chromatographic ÄKTA purifier ten system (GE Healthcare). Samples were loaded on a column pre-equilibrated with loading buffer (30 mM Tris buffer, pH 8.3) and eluted using 30 mM Tris buffer, pH 8.3, 1 M NaCl. Elution was achieved by using a linear 0–100% gradient (10 column volumes). Antibody concentration was determined by Bradford colorimetric assay [46] while the presence of contaminants was assessed by SDS-PAGE (14%, denaturing conditions).

Flow-cytometry of EV fractions

Flow-cytometry was performed by using either phages or purified nanobodies. In the first case, the protocol started with the coating of aldehyde/sulphate latex beads 4% w/v, 4 µm (Thermo Fisher Scientific) with EVs. Specifically, 25 µL of latex beads were coated overnight at 4 °C with 2 µg of total protein from EV-enriched fraction in PBS. Beads were washed 3 times with PBS and blocked first for 30 min at room temperature with 200 mM glycine and then 30 min in PBS plus 5% (w/v) skimmed milk. Phages were pre-incubated 1 h at 37 °C in PBS buffer plus 1% (w/v) skimmed milk and 100 µL of the blocked phages were incubated with 25 µL beads for 45 min at room temperature. Beads were washed 3 times in PBS before adding 100 μL of mouse anti-M13 antibodies (dilution 1:100, GE Healthcare). After 30 min at room temperature, beads were washed 4 times in PBS and 100 μL of anti-mouse PE-labelled antibodies (dilution 1:100, BD Biosciences) were added. After 30 min at room temperature and another washing step as above, 100 μL of anti-CD9 APC-labelled antibodies (BD-Biosciences, dilution 1:100, final concertation 16.7 nM) were added. Beads were incubated 20 min at room temperature before washing and flow-cytometry analysis by using a FACSCanto II instrument (BD Biosciences) collecting around 1000 events/s. A blue solid state 200 mW laser at 488 nm and a 100 mW red laser at 640 nm were used for excitation. The emission was detected with 561 nm (FL2, PE) and 660 nm (FL3, APC) filter. The positive beads were gated on the FL2-PE, FL3-APC plot. APC and PE specific fluorescence was assessed as the signal increase with respect to negative control (autofluorescence of beads without coated EVs).

When purified VHH-GFP constructs were used for detection of EVs, 25 µL of aldehyde/sulfate latex blocked beads containing EVs were incubated with 100 µg of nanobodies (2.5 μM) in 1% BSA solution in PBS for 1 h at room temperature. Beads were washed 3 times in PBS before adding 100 μL of mouse anti-CD9 PE-labelled antibodies (dilution 1:100, GE Healthcare) and were then incubated for 45 min at room temperature. After a washing step as described before, beads were analysed using FACS Calibur (BD Biosciences). A blue solid state 200 mW laser at 488 nm was used for excitation. The emission was detected with 525 nm (FL1, GFP) and 561 nm (FL2, PE) filter. The positive beads were gated on the FL1-GFP, SSC plot and FL2-PE, SSC plot. GFP and/or PE specific fluorescence was assessed as the signal increase with respect to negative control (autofluorescence of beads without coated EVs).

Capture of EVs on solid phase using purified nanobodies

EVs were captured using purified VHH-GFP constructs in both ELISA and flow-cytometry format assays. For ELISA, microtiter plates (96-well, MaxiSorp, Nunc™) were coated by incubating 10 μg/well of VHH-GFP (100 µL/well in PBS) overnight at 4 °C. Plates were blocked (1 h at 37 °C in PBS buffer plus 5% (w/v) skimmed milk) after which 0.1 µg of total protein from EV fraction in PBS with 1% milk (w/v) was added and incubated 1 h at 21 °C. After washing (4 × 5 min, PBS), 100 µL/well of blocked phages were added to the plates and incubated 1 h at room temperature. After washing (4 × 5 min in PBS), 100 μL of mouse anti-M13 antibodies conjugated with HRP (dilution 1:5000, GE Healthcare) were added to each well and incubated 1 h at 37 °C. Signal was developed after another washing step by the addition of 100 µL of TMB solution (Sigma Aldrich) and 1 h incubation at 37 °C. The colour reaction was stopped by 50 μL of 2 NaH₂SO₄. Absorbance was monitored at 405 nm in a HTS7000 Bioassay reader (Perkin Elemer).

For flow-cytometry, 25 µL of latex beads were coated overnight at 4 °C with 50 µg of VHH-GFP. Beads were washed three times with PBS and blocked first for 30 min at room temperature with 200 mM glycine and then 30 min in PBS plus 5% (w/v) skimmed milk. Beads were incubated with 10 µg of total protein from EV enriched fraction in PBS with 1% milk (w/v)/10 µL of VHH-coated beads or with 1:2 diluted plasma pool from healthy volunteers for 1 h at room temperature. Beads were washed 3 times in PBS before adding 100 μL of mouse anti-CD9 PE-labelled antibodies (dilution 1:100, GE Healthcare) and incubated for 45 min at room temperature. After washing steps as described before, beads were analysed using FACS Calibur (BD Biosciences). The analysis rate was around 1000 events s-1. A blue solid state 200 mW laser at 488 nm was used for excitation, the emission was detected at 561 nm (FL2, PE). The positive beads were gated on the FL2-PE, SSC plot. Specific fluorescence was assessed as the signal increase with respect to negative control (autofluorescence of beads without coated EVs).

EV immune-purification and TEM analysis

Twenty microliter of latex beads were coated overnight at 4 °C with 100 µg of H1/H6-GFP. Beads were washed three times with PBS and blocked first for 30 min at room temperature with 200 mM glycine and then 30 min in PBS plus 5% (w/v) skimmed milk. Beads were incubated 1 h at room temperature with: (i) 20 µg of total protein from cell supernatant EV enriched fraction in PBS with 1% milk (w/v); (ii) plasma pool from healthy volunteers diluted 1:2 in the same buffer. Beads were washed 5 times in PBS before elution in 100 µL of 200 mM glycine, pH 2.2 and immediate neutralization in 15 µL of 1 M Tris–HCl, pH 9.1. Eluted vesicles were stabilized by the addition of 25 mM trehalose and their morphology was evaluated by TEM after sample negative staining. Four microliter of vesicle suspension were adsorbed on carbon/formvar coated 400 mesh nickel grids (Electron Microscopy Sciences, Fort Washington, USA). After 10 min, the sample excess was removed with filter paper and immediately replaced by 4 μL of staining agent (uranyl acetate diluted 1:3 in distilled water), which was allowed to settle 10 min. Grids were washed 5 times in distilled water, the water excess was removed with filter paper, and then observed in a Philips CM 10 (FEI, Eindhoven, The Netherlands) TEM, operated at 80 kV.

Statistical analysis

Statistical analysis was performed using GraphPad Prism v7 for Windows (San Diego). A significance level of p ≤ 0.05 was used for analysis of variance, implemented using the Kruskal–Wallis test followed by the Tukey’s post hoc test (p ≤ 0.05). Correlation between different parameters was performed at significance level of p ≤ 0.05.