In situ exosome protein labeling strategy. APEX would be able to biotinylate exosome proteins if transported to the key subcellular location in exosome biogenesis, namely ILVs within MVEs. Construction of an APEX variant with an annexed ILV localization signal would enable such localization when the fusion construct is expressed in cells. We chose the catalytically improved APEX2 variant for our study15. Expression of an exogenous construct at a low level often helps it targeted to the desired subcellular location. We thus included an upstream open reading frame (uORF)16 in our construct to dampen the translation of the exosome-targetable APEX2, which we call “Exosome-Proxy APEX (EPA)” henceforward, from the downstream main open reading frame (Fig 1). The resulting APEX2 catalytic activity localized in the exosome-generating subcellular structure would enable a specific labeling of exosome-related proteins in situ in response to biotin-phenol and hydrogen peroxide. We have here applied this “Exosome-Proxy APEX Labeling (EPAL)” approach both to live cells (in vivo EPAL) and to the conditioned culture medium (in vitro EPAL) (Fig 1). In vivo EPAL would give rise to accumulation of biotinylated proteins specifically in ILVs, which can subsequently be purified from the cell lysates, leading to identification of proteins involved in exosome biogenesis, cargo selection, and cargo proteins. The in vivo EPAL-induced biotinylated proteins as well as the EPA itself could eventually be released in exosomes. If so, the in vivo EPAL-biotinylated proteins will also be purified from the conditioned medium and identified subsequently. On the other hand, if the catalytic activity of EPA is preserved and accessible after its release in exosomes, we would also be able to induce in vitro EPAL in the conditioned medium, allowing more direct determination of the cargo protein contents of secreted exosomes.
Establishing an Exosome-Proxy APEX, CD63-APEX2. We first tested whether the peptide called XPack (XP)17 would be appropriate for establishing an EPA. The XP peptide had previously been shown to confer the ability of exosomal secretion upon Green Fluorescence Protein (GFP). We thus ligated an XP-tagged GFP amino (N)-terminally to the APEX2 coding sequence (CDS). Since GFP tends to localize to the nucleus, we also included a nucleus export signal at the C-terminus of the fusion construct to increase its expression in the cytosol, where the exosome biogenesis vesicle trafficking occurs. When the resultant XP-GFP-APEX2 was expressed in HeLa cells, biotinylated proteins arose in response to biotin-phenol and hydrogen peroxide (Supplemental Fig S1a), indicating that APEX activity remains intact in this construct. We also noted that the proteins biotinylated by XP-GFP-APEX2 emerged in the conditioned media 30 hours after a pulse APEX labeling (Supplemental Fig S1b, left). The expression of XP-GFP-APEX2 did not significantly alter the number and size of nanoparticles in the culture medium (Supplemental Fig S1b, right), suggesting that the integrity of exosome biogenesis be maintained.
Immunofluorescence microscopy, however, showed little endosomal localization of XP-GFP-APEX2, which instead accumulated near the cell surface (Supplemental Fig S1c). XP-GFP-APEX2 did not appear to co-localize significantly with the late endosome marker LAMP2 (Supplemental Fig S1c), which is en route to exosome release and routinely used for identifying MVEs in the exosome field18. Immunogold transmission electron microscopy confirmed the limited association between XP-GFP-APEX2 and MVEs (Supplemental Fig S1d). To better visualize MVEs, we expressed a constitutively active form of the small GTPase Rab5 (Rab5QL), which is known to enlarge endosomes19. When analyzed by confocal immunofluorescence microscopy, XP-GFP-APEX2 was less represented on the endosomal membrane than the mCherry-conjugated Rab5QL (Supplemental Fig S1e). Furthermore, XP-GFP-APEX2 was hardly detectable within the luminal space of endosomes (Supplemental Fig S1e), contrary to other known MVE-traversing exosome cargoes19,20. Since an exosome protein typically shows high-level localization in MVEs at the steady state, reflecting the MVE origin of exosomes, XP-GFP-APEX2 appears to fall short of becoming an EPA.
Looking for an alternative EPA candidate, we next employed the well-characterized exosome cargo protein CD63. We made an expression construct named CD63-APEX2 by tandemly arranging the CD63 CDS, the APEX2 CDS, and the HA epitope from the amino (N)- to the Carboxy (C)- termini (Fig 2a). The tetra-transmembrane protein CD63 protrudes both the N- and the C-termini to the cytosolic side and is known to maintain protein topology during exosome biogenesis21. ILVs are thought to form by invaginating and pinching off the limiting membrane of the late endosome, converting the cytosolic side of the late endosome membrane to the lumen side of ILVs. Thus, if targeted as desired, CD63-APEX2 would insert its four transmembrane domains into the ILV membrane, expose the N and the C termini to the lumen side of ILVs, and induce biotinylation of the proteins in the luminal space bounded by the ILV membrane in response to biotin-phenol and hydrogen peroxide.
As expected from the use of uORF, CD63-APEX2 proteins were expressed at a much-reduced level relative to endogenous CD63, as shown in Western blot analysis with an anti-CD63 antibody (Fig 2a, anti-CD63). Western blotting with an anti-HA antibody confirmed the expression of CD63-APEX2, which migrated slower in SDS-PAGE than endogenous CD63 (Fig 2a, anti-HA). Streptavidin blotting further showed that biotinylated proteins arose from the CD63-APEX2-expressing cells after both biotin-phenol and hydrogen peroxide had been added (Fig 2b), indicating that APEX2 is catalytically intact in the fusion construct. We next examined whether the proteins biotinylated by CD63-APEX2 could be released into the culture medium. Streptavidin blotting showed that little biotinylated proteins were left at a detectable level inside the cells 30 hours after a pulse labeling (Fig 2c, cell lysates). The biotinylated proteins were instead recovered from the conditioned medium after centrifugally removing large particulates as pellets (Fig 2c, serum-free (s.f.) conditioned media supernatant), indicating that most of the biotinylated proteins had been released out to the medium. Nanoparticles released from the CD63-APEX2-expressing cells did not display a noticeable difference in size and number from those the control cells (Fig 2d), suggesting that the production and physical proprieties of exosomes remain intact in the CD63-APEX2-expressing cells.
In contrast to XP-GFP-APEX2, CD63-APEX2 displayed a significant co-localization with the late endosome marker LAMP2 in immunofluorescence microscopy (Fig 2e). To confirm this co-localization biochemically, cell homogenates were fractionated by density gradient centrifugation. CD63-APEX2 was co-segregated with the late-endosome marker LAMP2 but not with the early endosome marker EEA122 (Supplemental Fig S2a). The segregation pattern of CD63-APEX2 was indistinguishable from that of endogenous CD63, which is known to traffic through LAMP2-positive MVEs towards exosomal release. Notably, CD63-APEX2 appears to accumulate specifically inside MVEs, as revealed by transmission electron microscopy of the cells stained with a gold-conjugated-anti-HA antibody (Fig 2f). Confocal Immunofluorescence analysis further confirmed this spatial specificity when endosomes were enlarged by Rab5QL expression: both CD63-APEX2 and the mCherry-conjugated Rab5QL accumulated on the ILV-MVE membrane (Fig 2g). We therefore classify CD63-APEX2 as an EPA.
In vivo EPAL inducible by CD63-APEX2. As we noted above, CD63-APEX2 in the presence of biotin-phenol and hydrogen peroxide can induce biotinylation of proteins in cells and their subsequent release into the culture medium (Figs 2b, c). We thus asked whether the accumulation of CD63-APEX2 in MVEs (Figs 2e-g) is associated with the induced biotinylation. When CD63-APEX2-expressing cells were stained with an anti-HA antibody for CD63-APEX2 and Streptavidin for biotinylated proteins, Streptavidin-positive structure emerged after biotin-phenol and hydrogen peroxide had been added together, demonstrating the specificity of APEX biotinylation (Supplemental Fig S2b). At a higher resolution, the CD63-APEX2 staining in endosomes appeared to surround the Streptavidin staining, presumably in the ILV lumen (Fig 2h), consistent with the proposed topology of CD63-APEX2 with the APEX2 portion protruding to the ILV lumen. The ILV membrane may limit the diffusion of the biotin-phenolic radical produced by CD63-APEX2, thereby confining protein biotinylation to the ILV interior. Of note, Streptavidin staining appears to be sparser than that of CD63-APEX2 (Fig 2h and Supplemental Fig S2b), suggesting rather low accessibility of either biotin-phenol or hydrogen peroxide or both across several membranous structures to the ILV interior.
The spatially restricted co-localization of CD63-APEX2 and biotinylated proteins at last demonstrates an in vivo EPAL, which is inducible by the EPA CD63-APEX2. Streptavidin pull-down of in vivo EPAL-induced cell lysates would enrich a set of proteins involved in exosome biogenesis, cargo selection as well as cargo proteins. Moreover, the biotinylated proteins released in exosomes could also be collected by their affinity to Streptavidin from the conditioned medium that the in vivo EPAL-induced cells produce. Thus, in vivo EPAL will have broad utility.
In vitro EPAL inducible by CD63-APEX2. While most APEX studies thus far focusing on intracellular organelles, APEX has also proven useful in analyzing the proteome of the synaptic cleft, an open, extracellular side of the cell surface23,24, demonstrating that APEX methodology can be extended to defining extracellular proteomes. We thus asked whether extracellular CD63-APEX2 can biotinylate proteins in situ in secreted exosomes when biotin-phenol and hydrogen peroxide are added to the conditioned medium. Of note, such in vitro EPAL induction requires that the ascorbate peroxidase activity of CD63-APEX2 be maintained after its release in exosomes and that both biotin-phenol and hydrogen peroxide be accessible to CD63-APEX2 across the exosome membrane. If so, the induction of in vitro EPAL would be more direct and efficient than that of in vivo EPAL as biotin-phenol and hydrogen peroxide must cross just a single lipid bilayer to induce in vitro EPAL.
In vitro EPAL would further require that the culture medium contain little endogenous peroxidase activity, which could otherwise interfere with downstream capturing steps. We noted that treating a cell-free, fetal bovine serum (FBS)-containing medium with biotin-phenol and hydrogen peroxide generated a substantial level of biotinylated proteins in proportion to the added amount of FBS (Fig 3a), indicating an endogenous peroxidase activity present in FBS. A previous study by Takahashi et al. (1987) indeed reported a similar activity in human serum, which would be reducible by zinc chloride via an unknown mechanism25. We were likewise able to decrease to some degree the peroxidase activity from the FBS-containing medium by adding zinc chloride (Fig 3b, ZnCl). To better eliminate the endogenous peroxidase activity, we subjected the FBS-containing medium to filtration through a 500 kDa-cutoff membrane. To our surprise, the peroxidase activity was disproportionately fractionated into the filtrated while little left in the infiltrated (Fig 3b, 500 kDa Retentate), suggesting that the infiltrated fraction could be used for in vitro EPAL induction and subsequent capturing of the biotinylated proteins.
To set up in vitro EPAL, we next examined whether the 500-kDa-retained fraction of the conditioned medium contains the EPA CD63-APEX2. CD63-APEX2 was recovered not only in the 100,000 x g pellet fraction of conventional ultracentrifugation (Fig 3c, 100K x g pellets) but also in the 500-kDa-retained fraction (Fig 3c, 500 kDa retentate), consistent with previous reports showing the presence of exosomes in both of these fractions26,27. Moreover, adding biotin-phenol and hydrogen peroxide to the 500-kDa-retained fraction induced biotinylation of a wide mass range of proteins (Fig 3d, IB: strep-HRP), demonstrating an in vitro EPAL inducible by the EPA CD63-APEX2 with coupled filtration.
The biotinylated proteins in the 500-kDa-retained fraction contained known exosome cargo proteins such as Flotilin-1 and SDCBP28 as shown by Western blot analysis of Streptavidin-captured materials (Figs 3d-e). Of note, whereas the 500-kDa-infiltrated fraction contained both Flotilin-1 and SDCBP, the filtrate fraction contained SDCBP but not Flotilin-1 (Fig 3e), suggesting enrichment of exosomes in the retained as well as heterogeneous distributions of exosome proteins in the extracellular milieu. A component of the RNA-induced silencing complex, Ago2 can be secreted extracellularly via an unclear mechanism10,29. We were not able to capture Ago2 with Streptavidin matrix as a protein biotinylated by in vitro EPAL in the conditioned medium of CD63-APEX2-expressing cells (Fig 3f, Bound), suggesting that Ago2 be either absent or unreactive to the biotin-phenolic radical in the CD63-APEX2-containing exosomes. The conditioned medium per se contained a substantial level of Ago2 as revealed by Western blot analysis of the proteins that had not been captured by Streptavidin matrix (Fig 3f, Unbound). Detergent addition to the conditioned medium prior to in vitro EPAL induction, however, caused Ago2 to be biotinylated by CD63-APEX2, presumably through the release of CD63-APEX2 from the exosome (Fig 3g), suggesting that Ago2 be secreted into the cell culture medium via a mechanism other than release from the CD63-APEX2-containing exosomes in our experimental system.
As with CD63-APEX2, XP-GFP-APEX2-expressing cells also produced Streptavidin-reactive biotinylated proteins in the conditioned medium in response to biotin-phenol and hydrogen peroxide (Supplementary Fig S3). However, these biotinylated proteins did not include a detectable level of Flotilin-1 and SDCBP (Supplementary Fig S3), again ruling out the use of XP-GFP-APEX2 as an EPA in inducing in vitro EPAL.
Indexing of the CD63-APEX2-containing exosome proteome content in mouse proximal tubule-derived cells. We next assessed whether the CD63-APEX2 in vitro EPAL strategy could be applied to indexing exosome proteome contents through mass spectrometry analysis of the biotinylated proteins. In doing so, we further sought to extend our in vitro EPAL method established above in HeLa cells to another cell line and to a physiological setting as well. The kidney proximal tubules are known to be exposed to a variety of oxidative insults. We therefore chose the mouse kidney proximal tubule-originated BUMPT cells30 to stably express CD63-APEX2 (Fig 4a). Three independent lines of CD63-APEX2-expressing BUMPT cells were subjected to an oxidative stress. After inducing in vitro EPAL in the collected conditioned media (Fig 4a), the resulting biotinylated proteins were captured with Streptavidin matrix and subjected to mass spectrometry analysis.
We noted that the level of biotinylation, as revealed by the numbers of peptide spectrum matched (PSM), was largely invariable not only among replicates but also regardless of the oxidative stress treatment (Fig 4b). This suggests that the CD63-APEX2-mediated in vitro EPAL can induce comparable biotinylation reactions in the culture media from both stressed and unstressed cells. However, the in vitro EPAL of the unstressed cell medium yielded more species of biotinylated proteins than that of the stressed cell medium (Fig 4c), indicating that the oxidative stress somehow reduced the range of the protein species biotinylated by CD63-APEX2 in the medium. The mean level of biotinylation of the identified proteins, on the other hand, was comparable between the stressed and the unstressed cells (Fig 4d).
The proteins identified by our in vitro EPAL approach showed a high-degree correlation when any two biological replicates of a given treatment were compared (Fig 4e), demonstrating reproducibility and robustness of our mass spectrometry approach in combination with the CD63-APEX2 in vitro EPAL. We noted that the high-confident set of 287 proteins recovered from the in vitro EPAL of unstressed BUMPT cells shares 56 proteins with the set of the top 100 proteins in the Exocarta database31 (Supplemental Fig S4a), confirming the utility of our in vitro EPAL approach in identifying exosome cargo proteins. The remainder 44 proteins in the top 100 protein set of the Exocarta database were not detected in our method, however (Supplemental Fig S4a). The in vitro EPAL dataset, on the other hand, contained 231 proteins that were absent in the Exocarta top 100 protein set (Supplemental Fig S4a; Supplemental Table). Different cell origins and varying mass spectrometry data depths between our dataset and the Exocarta dataset might account for the limited overlap between the two protein sets.
Comparisons between the stressed and the unstressed treatment data in the CD63-APEX2 in vitro EPAL of BUMPT cells gave rise to lesser correlation values than those between biological replicates of a given treatment (Fig 4e), suggesting that specific sets of proteins be enriched in exosomes per each treatment group. Indeed, proteins enriched in the stressed cell samples were distinguished from those from unstressed cell samples in fold change analysis (Fig 4f). Thbs1 and Hsp90, for example, were enriched in the medium of the unstressed cells (Fig 4f), which were confirmed by orthogonal, independent in vitro EPAL experiments (Supplemental Fig S4b).
Notably, proteins uniquely identified from the stressed cell medium included various components of both the large and small subunits of the ribosome, including Rps6, Rpl13, Rps9, Rpl19, and so on (Fig 4f). Indeed, functional enrichment analysis of the stress cell dataset revealed that protein production-related gene ontology (GO) terms32 ranked high among various biological functions (Fig 4g). Orthogonal, independent in vitro EPAL experiments confirmed the enrichment of the ribosomal small subunit protein Rps6 specifically in the conditioned medium of the stressed cells (Supplemental Fig S4b). Addition of proteinase K to the conditioned medium did not significantly alter the recovery of the biotinylated Rps6 protein, suggesting that ribosomal proteins such as Rps6 be extracellularly secreted as enclosed in a vesicle in which CD63-APEX2 also resides (Supplemental Fig S4c).
To conclude, we have demonstrated that the CD63-APEX2 in vitro EPAL in combination with mass spectrometry and proteomics analyses can define a reliable set of proteins released in the exosomes harboring an EPA such as CD63-APEX2. We further showed that the CD63-APEX2 in vitro EPAL can be executed in diverse cell types and physiological settings, indicating broad utility of our approach. Our data also revealed that kidney proximal tubule-derived cells can accumulate ribosomal proteins in exosomes in response to oxidative stress, suggesting that these proteins may either play a role as biomarkers for cellular stress or mediate oxidative stress responses between cells.