Canine E-cadherin inserted into the peGFP-N1 vector was kindly provided by W. James Nelson (Adams et al., 1998). Cadherin domains 2 to 5 (amino acids 272–671) were deleted by two-step mutagenesis PCR using the primer pair 5′- CAC​CCA​GGC​AGT​CTT​CCA​AGG​ATA​TCT​CAA​GCT​CAC​AGA​TAA​CC -3′ and 5′- GGT​TAT​CTG​TGA​GCT​TGA​GAT​ATC​CTT​GGA​AGA​CTG​CCT​GGG​TG -3′ to generate plasmid pE-cadherinΔE2-5GFP_PTAP. Mutation of the E-cadherin PTAP motif into ASAA was induced by overlap extension PCR with inside primers 5′- CG GAC ACT GAC gCT Agc GCT gCT CCT TAT GAC-3′ and 5′- GTC ATA AGG AGc AGC gcT AGc GTC AGT GTC CG-3’ (mutated nucleotides are depicted in small letters). Successful generation of plasmid constructs was validated by sequence analysis.

The following monoclonal and polyclonal antibodies or nanobodies were used in this study: anti-E-cadherin (C-terminus: BD Transduction Laboratories, 61018; N-terminus: Genetex/Biozol GTX134997), anti-Tsg101 (Abcam, 4A10), anti-GFP (Takara, 632592), anti-GFP-nanobodies (Chromotek), anti-actin (BD Transduction Laboratories, 612656), anti-Hrs (Enzo, A-5; GeneTex, GTX89364), anti-GAPDH (Abcam, 6C5), anti-giantin (Covance, PRB-114C), anti-PDI (BD Transduction Laboratories, 610946), anti-TOM20 (Santa Cruz, Sc11415).

E-cadherin sequences were aligned with ClustalOmega (McWilliam et al., 2013). Sequence logos aligned to the decapeptide (765)DTPTAPPYD(773) were generated by WebLogo (Crooks et al., 2004).

MDCK (Madin-Darby Canine Kidney) type II cells were cultured in MEM high glucose supplemented with 2 mM glutamine, 100 U/ml penicillin, 100 mg/ml streptomycin and 10% FCS at 37°C in humidified atmosphere containing 5% CO₂. Transfections were performed with Lipofectamine 2000 (Invitrogen). For generation of stable cell lines, MDCK cells were split in high ratios 2 days after transfection and selected in MEM medium containing 0.4 mg/ml Zeocin or an equivalent antibiotic for selection. We transferred single clones to 24 well plates with Trypsin/EDTA-soaked Whatman slices. Subsequently, the clones were analyzed for expression of the exogenous proteins by immunoblot and fluorescence microscopy. Only those clones were selected that exhibited a transfection efficiency of at least 90% transfected cells.

Cells were transfected as previously described (Bänfer et al., 2018) at day 1 after seeding with the following siRNA duplexes (Invitrogen): Tsg101: 5′-GGU UAC CCG UUU AGA UCA A [dT][dT]-3′, 5′-UUG AUC UAA ACG GGU AAC C [dT][dT]-3’; Hrs: 5′-UUC UUC UCC CAG UAG UUC C [dT][dT]-3′, 5′-GGA ACU ACU GGG AGA AGA A [dT][dT]-3′ and 5′-GGA ACG AGC CCA AGU ACA A [dT][dT]-3′, 5′-UUG UAC UUG GGC UCG UUC C [dT][dT]-3’.

The cells were washed three times with PBS and incubated overnight with MEM and 10% fetal calf serum (FCS). To avoid contamination of the exosomal fraction by bovine serum EVs, cell culture media were subjected to centrifugation at 100,000 g for 2 h prior to overnight incubation. Medium was collected and submitted to a series of centrifugation steps as previously described (Bänfer et al., 2018). Cell culture supernatants were centrifuged at 100,000 g for 1 h. The resulting pellet was washed in PBS++ (PBS supplemented with 1 mM CaCl2 and 1 mM MgCl2), repelleted again at 100,000 g for 1 h and then resuspended in either PBS++ or in SDS-PAGE sample buffer for further use. All steps were performed at 4°C.

Exosomal pellets were resuspended in PBS++, pooled, and subsequently split into three identical aliquots. Proteinase digestion was then performed with 0.5 mg/ml to 1 mg/ml proteinase K (Fermentas) in presence or absence of 1% Triton X-100 for 30 min at 37°C. As control, one of the aliquots was incubated without proteinase K.

MDCK cells stably expressing eGFP or eGFP fusion proteins were washed with PBS⁺⁺, followed by mechanical detachment of the cells in PBS⁺⁺. The cells were pelleted by centrifugation at 500 g for 3 min at 4°C and rinsed twice with PBS⁺⁺. Cell lysis was achieved by application of lysis buffer (10 mM HEPES, 150 mM NaCl, 0.5 mM EDTA, 1% Triton X-100, 0.5% SDS and proteinase inhibitor cocktail, pH 7.5). Cell lysates (17,000 g, 15 min, 4°C) were then incubated with GFP-nanobody agarose (GFP-Trap, Chromotek) or control beads without GFP-nanobodies for 1.5 h at 4°C. Finally, beads were rinsed four times with Co-IP washing buffer (10 mM HEPES, 150 mM NaCl, 0.5 mM EDTA) and boiled in SDS-PAGE loading buffer for Western blot analysis.

Immunofluorescence analysis was performed essentially as previously described (Bänfer et al., 2018). The cells were grown on cover slips and fixed with 4% paraformaldehyde for 20 min. Afterwards, cells were permeabilized with 0.1% Triton-X-100 for 20 min and blocked in 5% BSA/PBS⁺⁺ for 1 h. Primary antibodies were added in blocking reagent for 2 h or overnight. Secondary antibodies labelled with the indicated Alexa Fluor dyes were applied in PBS⁺⁺ for 1 h. Nuclei were stained with Hoechst 33342. Following incubation, cells were washed with PBS⁺⁺ and mounted with Mowiol. Confocal images were acquired on a Leica STELLARIS microscope equipped with a ×93 glycerol planapochromat objective (Leica Microsystems). Processing of images was done with Leica LAS X and Volocity 5 (PerkinElmer). We calculated co-localization between markers as Manders’ coefficient using the Volocity software package. Structures with coefficients <0.5 were classified as “not-colocalized”.

For nanoFCM, a Nano Analyzer (NanoFCM Co. Ltd., Nottingham, United Kingdom) equipped with a 488 nm laser, was calibrated using 200 nm polystyrene beads (NanoFCM Co.) with a defined concentration of 2.08 × 108 particles/ml, which were also used as a reference for particle concentration. In addition, monodisperse silica beads (NanoFCM Co. Ltd.) of four different sizes served as size reference standards to calibrate the size of EVs. Freshly filtered (0.1 µm) 1 × PBS was analyzed as background signal and subtracted from the other measurements. Each dot plot was derived from data collected approximately 4,000 events with a sample pressure of 1.0 kPa. For immunofluorescence staining, the following antibodies were used (BioLegend): FITC-conjugated mouse anti-human/canine CD9 antibody (clone HI9a), and anti-human Ecad, with secondary PE-conjugated donkey anti-rabbit IgG antibody (clone Poly4064); as isotype controls, FITC-conjugated mouse IgG1, κ (clone MOCP-21), and PE-conjugated donkey anti-rabbit IgG antibody (clone Poly4064); 1 ng/µl of each antibody in 50 µL 1 × PBS. After removing antibody aggregates by centrifugation at 12,000× g for 10 min, the supernatant was added to 5 × 108 purified EVs, followed by incubation for 90 min at 25°C under constant shaking. Stained EV were diluted 1:100 in 1xPBS for NanoFCM analysis.

A PROSITE database search of the UniProtKB/Swiss-Prot database revealed 866 candidate polypeptides that contain at least one P(S/T)AP late domain motif (Figure 1A, Supplementary Table S1). 242 of these protein candidates are listed in the ExoCarta database as EV cargo proteins including 16 transmembrane proteins. Among them 5 candidates belong to the cadherin family. Primary sequences of E-cadherin orthologs revealed a conserved PS/TAP motif in the cytosolic domains of e.g., human (… P₇₆₇T₇₆₈A₇₆₉P₇₇₀ … ) and dog (… P₈₂₅T₈₂₆A₈₂₇P₈₂₈ … ) E-cadherin (Figures 1B,C). To assess if E-cadherin is recruited into ILVs of MVEs that can be released as EVs, epithelial MDCK cells were immunostained for the ESCRT-0 protein Hrs, which labels MVEs (Bache et al., 2003). About 6% of these Hrs-positive structures were co-stained with endogenously expressed E-cadherin, which is significantly above the co-staining efficiency of E-cadherin with mitochondrial TOM20 as negative control (Figure 2). These observations document that a small fraction of E-cadherin is closely related to MVEs, which are the source compartment for exosomal release.

We then isolated EVs from MDCK cell medium by differential centrifugation as previously published (Bänfer et al., 2018) and monitored the presence of E-cadherin in the EV fraction by immunoblot. Antibodies directed against Tsg101 were used as positive control for the validation of successful EV isolation. Figure 3A indicates that E-cadherin as well as Tsg101 are enriched in isolated EVs. Antibodies directed against endoplasmic reticulum protein disulfide isomerase (PDI), the Golgi component giantin and mitochondrial TOM20 were used as negative controls to verify the purity of isolated eEVs. To clarify the orientation of E-cadherin with its single transmembrane domain on the EV membrane, purified EVs were treated with the unspecific protease proteinase K in the presence or absence of Triton X-100 (Figure 3B). Antibodies directed against E-cadherin did not detect full length E-cadherin after proteinase K-treatment. This indicates, that the N-terminal extracellular domain of E-cadherin was accessible for the protease and therefore degraded. Others have shown that a significant fraction of E-cadherin can be proteolytically shed into the extracellular space through cleavage by secretases and caspases into 80 kDa soluble E-cadherin (sE-cadherin) and the remaining membrane-attached cytoplasmic tail of about 22 kDa (David and Rajasekaran, 2012). We also detected this smaller band with an antibody directed against the cytoplasmic domain of E-cadherin and this band was insensitive to proteinase K digestion. It was solely degraded if EV-membrane lipids were solubilized by Triton X-100. It is important to note that Tsg101 and the designed eGFP variant containing the late domain motif at the C-terminus (eGFP-PSAP), which reside in the lumen of EVs (Bänfer et al., 2018), showed a similar sensitivity pattern in this assay. Thus, E-cadherin is oriented with the large extracellular domain facing the outer surface of EVs while the cytoplasmic tail points into the EV lumen (Figure 3C). In essence, the cytoplasmic tail of E-cadherin, which harbors a PS/TAP motif, soluble eGFP-PSAP and the ESCRT-component Tsg101 are found in the EV lumen.

Next, we sought to study putative interaction of E-cadherin with Tsg101. Therefore, we incubated MDCK_Ecad-GFP and MDCK_Tsg101-GFP cells stably expressing E-cadherin-GFP or Tsg101-GFP (Bänfer et al., 2018). GFP-Trap beads were used for precipitation of the GFP fusion proteins from cell lysates. Indeed, Tsg101 was pulled down by E-cadherin-GFP (Figure 4A). Moreover, Tsg101-GFP precipitated E-cadherin. This is a first hint for interaction between E-cadherin and Tsg101 in MDCK cells. However, we cannot conclude from these experiments, whether the interaction is direct or indirect.

In order to find out if Tsg101 plays a functional role in recruitment of E-cadherin into EVs, which could be in analogy to the mechanism published for soluble galectin-3 (Bänfer et al., 2018), we performed experiments were siRNA was used to specifically deplete the cellular content of Tsg101. Figures 4B,C show specific depletion to about 25% of residual Tsg101. Under these conditions EV recruitment of E-cadherin significantly declined by about 50% (Figures 4B,D). The EV-pool of actin, which has also been listed as EV cargo molecule (Xu et al., 2016), was not dramatically affected by Tsg101-depletion indicating that the cells still secrete representative EV quantities. These data suggest that expression of Tsg101 is linked to efficient incorporation of E-cadherin into secreted EVs, most likely by recognition of a cytoplasmic binding motif and the initiation of E-cadherin transport into budding ILVs.

We thus addressed the question if the cytoplasmic PTAP late domain motif of E-cadherin mediates ILV-recruitment of E-cadherin and interaction between E-cadherin and Tsg101. To increase the transfection efficiency and to facilitate the generation of stable cell clones, we first generated a GFP-tagged variant of E-cadherin with a deletion of cadherin domains two to five (E-cadherinΔE2-5GFP_PTAP) (Figure 5A). In a second step, the cytoplasmic PTAP motif of this variant was mutated to ASAA (E-cadherinΔE2-5GFP_ASAA), mimicking a mutation that is known to abrogate the release of Marburg virus-like particles (Dolnik et al., 2010). The two constructs were transiently transfected into MDCK cells. 48 h after transfection the cells were fixed and immunostained for Hrs and Tsg101 as MVE-markers (Figure 5B). Quantification using Manders’ correlation coefficients revealed that 9.61 ± 1.7% of cytoplasmic E-cadherinΔE2-5GFP_PTAP structures localized to Tsg101-positive vesicles representing the MVB formation site (Figures 5B,C). In contrast, co-staining of E-cadherinΔE2-5GFP_ASAA resulted in a significantly reduced overlap of 2.30 ± 0.7%, thus supporting the idea of a relevant role of the cytoplasmic PTAP late domain in ILV-sorting of this transmembrane polypeptide. Consequences of PTAP late domain mutation on EV recruitment of E-cadherin were studied in MDCK cells stably expressing E-cadherinΔE2-5GFP (MDCK_{E-cadherinΔE2-5GFP}) or E-cadherinΔE2-5GFP_ASAA (MDCK_{E-cadherinΔE2-5GFPASAA)}. Here indeed, we found that E-cadherinΔE2-5GFP was enriched in EVs, whereas the quantities of E-cadherinΔE2-5GFP_ASAA were drastically reduced in isolated EVs (Figure 5D). Evidence for a central role of the cytosolic E-cadherin PTAP motif in Tsg101-interaction was provided by diminished Tsg101 precipitation of E-cadherinΔE2-5GFP_ASAA (Figure 5E). In conclusion, the PTAP motif in the cytoplasmic tail is essential for efficient Tsg101 interaction and EV-mediated release of E-cadherin.

Finally, we monitored the spectrum of EV populations released by MDCK cells using nano-flow cytometry (nFCM). Therefore, EVs were collected from the supernatants of MDCK or MDCK_Gal3-GFP cells and exposed polypeptides were fluorescently stained with antibodies directed against CD9, CD63 or E-cadherin to discriminate between distinct EV cargoes by flow cytometry (Figure 6). EV analysis at single particle level then revealed a strong correlation between the presence of E-cadherin and galectin-3 on extracellular vesicles. Galectin-3 is recruited by PSAP-mediated sorting into EVs (Bänfer et al., 2018) and shows the highest overlap with E-cadherin on these vesicles. Nearly all galectin-3-positive vesicles contain E-cadherin (20.1 % versus 1.3%). However, not all E-cadherin-positive vesicles are also loaded with galectin-3, which can be explained by a cargo-specific EV loading efficiency and modulation of this process by additional cellular factors. Less stringent correlation was detected between vesicles positive for the tetraspanins CD9-or CD-63- and E-cadherin. Altogether, these observations indicate that significant quantities of E-cadherin and galectin-3 are sorted into identical extracellular vesicles.