An extracellular vesicle targeting ligand that binds to Arc proteins and facilitates Arc transport in vivo

Communication between distant cells can be mediated by extracellular vesicles (EVs) that deliver proteins and RNAs to recipient cells. Little is known about how EVs are targeted to specific cell types. Here, we identify the Drosophila cell-surface protein Stranded at second (Sas) as a targeting ligand for EVs. Full-length Sas is present in EV preparations from transfected Drosophila Schneider 2 (S2) cells. Sas is a binding partner for the Ptp10D receptor tyrosine phosphatase, and Sas-bearing EVs preferentially target to cells expressing Ptp10D. We used co-immunoprecipitation and peptide binding to show that the cytoplasmic domain (ICD) of Sas binds to dArc1 and mammalian Arc. dArc1 and Arc are related to retrotransposon Gag proteins. They form virus-like capsids which encapsulate Arc and other mRNAs and are transported between cells via EVs. The Sas ICD contains a motif required for dArc1 binding that is shared by the mammalian and Drosophila amyloid precursor protein (APP) orthologs, and the APP ICD also binds to mammalian Arc. Sas facilitates delivery of dArc1 capsids bearing dArc1 mRNA into distant Ptp10D-expressing recipient cells in vivo.


Introduction
Extracellular vesicles (EVs) are mediators of cell-cell communication that transport specific protein and RNA cargoes. They are a heterogeneous collection of vesicular structures that are exported from cells by a variety of mechanisms. Exosomes are 30-150 nm in diameter and are released into cell supernatants via fusion of multivesicular bodies (MVBs) with the plasma membrane. Exosomes and other EVs carry specific proteins and RNAs, and EVs derived from different cell types contain different cargoes. EV cargoes are biomarkers for specific diseases. Because EVs can encapsulate RNAs and protect them from degradation, and then deliver those RNAs to recipient cells, they represent a promising new type of therapeutic agent (O'Brien et al., 2020;Teng and Fussenegger, 2020).
Little is known about mechanisms involved in EV targeting to specific cell types. EVs are internalized into cells after receptor binding using a variety of endocytic mechanisms, resulting in the delivery of their cargoes into the recipient cells. They can also directly activate intracellular signaling without endocytosis by interacting with cell surface receptors. In this paper, we identify Stranded at second (i-iii) Brain lobes. GFP labels cell bodies (arrowheads in iii) and axon tracts. V5 labels cell bodies only weakly (arrowheads in ii), strongly labels some axon tracts, and localizes to the periphery (sheath) of the brain lobes (arrows in ii). (iv-vi) Ventral nerve cord. GFP strongly labels Ap VNC cell bodies (arrowheads in vi) and weakly labels Ap axons (arrow in vi). V5 weakly labels cell bodies (arrowheads in v), and strongly labels segments of axons (arrows in v) and areas adjacent to the axons. These may be extracellular matrix and/ or glial sheaths. Scale bar, 10 µm. (c) Western blot, showing that Sas FL localizes to EVs. EVs were prepared from supernatants from S2 cells expressing the indicated protein, and equal amounts of cell lysate proteins and EV proteins were loaded on the gel. Top panel, anti-V5 blot. Sas FL migrates slightly above the 250 kD marker, and Sas short slightly below it. Middle panel, Wg (EV marker). Almost all of the Wg is in EVs. Bottom panel, anti-β-actin (cytoplasmic marker) blot. There is no signal in EVs. In these samples, 60% of Sas FL and 10% of Sas short is in EVs; the average over 4 experiments was 46% in EVs for Sas FL and 10% for Sas short . The absence of β-actin signal in the EVs shows that they are not heavily contaminated by cytosol. (d) Localization of endogenous Sas isoforms in the embryonic gut. Wild-type (wt) late stage 16 embryos were triple-stained for Sas FL (using the [Schonbaum et al., 1992] antiserum, which primarily recognizes the EVT region; green), Sas short (using our antipeptide antibody; red), and Crumbs (apical marker; blue). (i-iv) Foregut; (v-viii), hindgut. Note that in both gut regions Sas FL colocalizes with Crumbs at the apical (luminal) cell surfaces (arrows), while anti-Sas short labels the entire width of the gut wall (brackets). See   between the VWFC and FN-III domains that is lacking in the Sas short (PA/PC) isoform ( Figure 1a). We expressed Sas FL tagged with an N-terminal V5 epitope tag (inserted immediately after the signal sequence) in embryonic late stage 16 Apterous (Ap) neurons, which consist of paired neurons (one per hemisegment) in the ventral nerve cord (VNC) and scattered neurons in the brain lobes. It was expressed together with mCD8-GFP, which is also a transmembrane CSP. The GFP signal was restricted to Ap neuron cell bodies, with faint staining on the axons. However, V5-Sas FL was observed in sheaths around brain lobes and in areas adjacent to axons in the VNC, as well as in puncta throughout the VNC and brain (Figure 1b). The V5 signal in cell bodies was very weak, especially in the brain.
We also expressed V5-Sas short in Ap neurons, and observed that V5 staining was restricted to cell bodies and axons in the VNC, and to cell bodies in the brain (Figure 1-figure supplement 1). Thus, although Sas FL and Sas short have the same TM and ICD, they differ in subcellular localization, with Sas short being retained in expressing cells and Sas FL moving away from these cells and into the extracellular matrix.
Movement of V5-Sas FL , and presumably of endogenous Sas FL , away from its source could occur through cleavage of the Sas ECD from the cell surface or by release of intact Sas in EVs. To distinguish between these possibilities, we expressed V5-tagged Sas FL and Sas short in transiently transfected Drosophila Schneider 2 (S2) cells, which express endogenous Sas at almost undetectable levels. We prepared EVs from S2 cell supernatants using the Invitrogen Exosome Isolation Kit or by ultracentrifugation, and analyzed their contents by western blotting. EVs contained the Wg protein, which is a marker for EVs in S2 cells and is present at very low levels in lysates (Koles et al., 2012;Figure 1c, Figure 1-figure supplement 1). They lacked β-actin, showing that they are not heavily contaminated by cytosol. We found that about 50% of V5-Sas FL localized to EVs, while ~90% of V5-Sas short was retained in the cell lysate ( Figure 1c). We did not observe any proteolytic cleavage products in EVs or unpurified supernatants.
EV preparations made using the Exosome Isolation kit have similar characteristics to those generated by ultracentrifugation (Skottvoll et al., 2019), and the kit requires much less material, allowing EVs to be purified from small-scale transfections. However, to confirm results obtained with the kit, we also purified EVs from S2 cells expressing V5-tagged Sas FL using a standard ultracentrifugation protocol (Théry et al., 2006), and showed that Sas FL and Wg are also present in these EVs (Figure 1figure supplement 1).
The commonly used rabbit antiserum against Sas primarily recognizes the EVT region, so embryo staining reveals the localization of endogenous Sas FL (Schonbaum et al., 1992). To visualize Sas short , we made an anti-peptide antibody against a sequence spanning an exon junction in the PA/PC isoforms. This recognizes Sas short , and staining with the antibody is eliminated in sas mutant embryos ( Figure 1-figure supplement 1). Double-staining of the foregut and hindgut with the two Sas antibodies showed that Sas FL localizes to apical cell surfaces, while Sas short is distributed across the entire cell membrane (Figure 1d). These data imply that the EVT sequence lacking in Sas short is required for both apical localization and targeting to EVs. Polarized cells can release EVs with different cargoes from their apical and basolateral surfaces (Matsui et al., 2021), so EV targeting could be downstream of apical localization in vivo. S2 cells are unpolarized, however, so this mechanism is unlikely to apply to trafficking of Sas FL to EVs in cultured S2s.

Analysis of Sas FL EVs by electron microscopy
To demonstrate that Sas is actually on EVs, we used immuno-EM and EM tomography to analyze purified EV preparations from V5-Sas FL -expressing S2 cells. The tomographic images showed that the EVs span a range of sizes, from ~30 nm in diameter to >100 nm, and that they are a mixture of single and double-membrane vesicles (Figure 2c and d, Figure 2-figure supplement 1). For immuno-EM, we incubated EVs with anti-V5, followed by gold-labeled anti-mouse secondary antibody. Figure 2a shows a typical image, in which an EV is associated with multiple 10 nm gold particles. The distance between the EV membrane (yellow bracket: diameter of the vesicle) and a gold particle (white bracket: distance between membrane and a particle) can be more than 40 nm. This likely reflects the large size of the Sas ECD, in which the N-terminal V5 epitope is separated by 1590 aa from the TM domain. The distance is variable, however, because the Sas ECD, which is composed of a chain of domains separated by linkers, is likely to be flexible and able to adopt many different conformations. The region outside the membrane boundary is of higher density, probably because it represents the protein sheath around the EV membrane. To further characterize Sas localization, we performed an experiment in which EVs were incubated with both mouse anti-V5 and rabbit anti-Sas, which primarily recognizes the EVT region in the middle of the ECD, followed by 10 nm gold particle-labeled antimouse secondary antibody and 5 nm gold particle-labeled anti-rabbit secondary antibody. Figure 2b shows an EV that is associated with multiple 10 nm (arrow) and 5 nm (arrowhead) gold particles.
To analyze the numbers and sizes of EVs from Sas FL -expressing and control S2s, we examined purified EVs using Nanoparticle Tracking Analysis with a NanoSight instrument (NTA, System Biosciences, LLC). We observed that the distribution of EV diameters was shifted toward smaller values in cells expressing Sas FL (mean diameter = 102 nm, vs. 129 nm for control cells) (Figure 2e). Expression of Sas FL increased the number of EVs per cell in the exosome size range (30-150 nm in diameter) by 44%, and the number of EVs per cell of <100 nm in diameter by 72%, suggesting that the presence of high levels of Sas FL increases the rate of EV production.
Many of the EVs from V5-Sas FL -expressing S2 cells EVs had denser objects within their boundaries ( Figure 2d, Figure 2-figure supplement 1). These were typically 30-40 nm in diameter. Such objects were also found within EVs from untransfected S2 cells (Ashley et al., 2018), so their presence does not require expression of Sas FL .

Sas FL EVs target to cells expressing Ptp10D
Having shown that Sas FL moves away from expressing neurons in the embryo and is an EV component, we then asked whether it can be incorporated into distant cells in vivo, presumably through Immuno-EM images of EVs from a purified EV prep from V5-Sas FL -expressing S2 cells. EV outline (membrane) diameters are indicated by yellow brackets. White brackets, separation between EV outline and a gold particle. (a) Immuno-EM with 10 nm anti-V5 gold particles (arrows). (b) Immuno-EM with both 10 nm anti-V5 (large gold, arrow) and 5 nm anti-Sas (small gold, arrowhead). (c) EM tomogram of an empty double-membrane vesicle (arrows). Apparent EV sizes differ between immuno-EM and tomography, which use very different preparation methods. A low-mag view of a single slice from an EM tomogram of an EV preparation is shown in  (e) Nanoparticle Tracking Analysis (conventional NTA with NanoSight) of purified EV preparations from untransfected (green curve) and Sas FL -expressing (blue curve) S2 cells. The mean size distribution is indicated. Standard error indicated by red color around curves. Source data files include an Excel file of raw data for the NTA analysis, the conversion of the numbers from numbers of EVs per sample to numbers of EVs per cell, based on cell counts, and plots of the data.
The online version of this article includes the following video, source data, and figure supplement(s) for figure 2: Source data 1. NTA raw data.  https://elifesciences.org/articles/82874/figures#fig2video1 endocytosis of EVs. We expressed V5-Sas FL in 3 rd instar larval salivary glands (SGs) using an SG-specific GAL4 driver, Sage-GAL4. SG-specific expression of dsRed from this driver in whole larvae is shown in Figure 3-figure supplement 1. We then visualized V5 staining in other tissues. We found that V5-Sas FL made in SGs is present in imaginal discs, which are separated from SGs by larval hemolymph (wing and haltere discs shown in Figure 3a-c). There was no V5 staining in imaginal discs from driver-alone larvae (compare Figure 3a and b). We also expressed V5-Sas short in SGs, and this was not observed in imaginal discs (Figure 3c). This is consistent with the observation that V5-Sas short does not move away from expressing Ap neurons and is restricted to cell lysates in S2 cells (Figure 1b and c). wt discs express the Sas receptor Ptp10D at low levels (see Figure 3-figure supplement 1), but we do not know whether Ptp10D is required for Sas FL binding to discs. Overexpressed Ptp10D can stimulate Sas FL accumulation in discs, however (see  (g) Transfer of Sas FL from EVs into recipient S2 cells. Supernatants from S2 cells expressing V5-Sas FL were incubated with cultures of untransfected S2 cells or cells expressing Ptp10D, Numb, or both, and cell lysates analyzed by western blotting with anti-V5. Note that V5-Sas FL levels were elevated relative to control cells by expression of either Ptp10D or Numb, and that levels were further increased by coexpression of Ptp10D and Numb coexpression. (h) Quantitation of results from panels (d-f) and (g). Levels of transferred V5-Sas FL were increased by ~fourfold relative to untransfected controls by Ptp10D+Numb coexpression in S2 cells (n=6), and by ~threefold relative to ap-GAL4/+control by Ptp10D+Numb coexpression in wing discs (n=5). Quantitation was done using Image J.
The online version of this article includes the following source data and figure supplement(s) for figure 3: Source data 1. Source data files include raw and labelled images for the western blots shown in panel (g), and an Excel file of the quantitation of the western blot and disc immunofluorescence signals used to generate panel (h). To examine the mechanisms involved in specific targeting of Sas EVs, we added supernatants (EV fraction) from V5-Sas FL -expressing S2 cells to S2 cell cultures and analyzed recipient cell lysates by western blotting. We observed that expression of the Sas receptor Ptp10D in recipient cells increased V5-Sas FL levels in these cells, as did expression of Numb, a regulator of endocytosis that binds to the Sas ICD (Chien et al., 1998). Expression of both Ptp10D and Numb produced a synergistic effect, increasing V5-Sas FL by ~fourfold relative to untransfected recipient cells (Figure 3g-h). We speculate that binding of Numb to the Sas ICD increases Sas uptake and/or protects endocytosed Sas from degradation.
We then developed an assay to examine the effects of Ptp10D and Numb on Sas targeting in larval cells by incubating dissected 3 rd instar wing imaginal discs with V5-Sas FL supernatants. Wing discs from driver-alone (control) larvae displayed weak V5 staining after incubation with V5-Sas FL EVs. Staining was increased in discs from larvae expressing Ptp10D in imaginal discs, and further elevated (~threefold increase relative to driver control) by expression of both Ptp10D and Numb (

Sas binds to dArc1 and mammalian Arc via a conserved tyrosine motif
To examine whether Sas interacts with specific EV cargoes, we made EV preparations from S2 cells expressing V5-Sas FL and from untransfected control cells, lysed them with nonionic detergent, incubated the lysate with anti-V5-coupled magnetic beads, and analyzed bead-bound proteins by mass spectrometry (Figure 4a, Supplementary file 1). We ranked the identified proteins by their degree of enrichment in the V5-Sas FL sample relative to the control. Proteins that are present in the V5-Sas FL sample should include EV cargoes that bind to Sas FL and are therefore present in V5 IPs, plus proteins that nonspecifically bind to V5 beads. Proteins in the control sample would be only those that nonspecifically bind to V5 beads. We observed that the most highly enriched protein (after Sas itself) is dArc1 (22-fold) ( Figure 4b). dArc2 is #7 on the list (6-fold). dArc1 mRNA, presumably encapsulated within dArc1 capsids, is known to be a prominent mRNA component of EVs from Drosophila cultured cells (Lefebvre et al., 2016;Ashley et al., 2018). Our data suggest that dArc1, and possibly dArc2, associate with Sas FL , since they are enriched in V5 IPs from cells expressing V5-Sas FL . We then went on to show that dArc1 binds directly to the Sas ICD (see below).
The presence of Arc proteins in EVs is consistent with published data from both mammalian and Drosophila cell cultures. EVs from media of short-term cultures of mouse cortical neurons were shown to contain denser objects whose size (~30 nm in diameter) was consistent with mammalian Arc capsids, and which were associated with anti-Arc gold particles (Pastuzyn et al., 2018). For dArc1, capsid-like structures that bound to anti-dArc1 gold particles were detected in lysed preparations of EVs from S2 cells (Ashley et al., 2018). The dArc1 capsid is 37 nm in diameter (Erlendsson et al., 2020;Hallin et al., 2021). As described above, EM tomography of EVs from Sas FL -expressing S2 cells showed denser objects of 30-40 nm in diameter within many of them ( Other proteins observed in the mass spectrometric analysis that were present at higher levels in the IP from V5-Sas FL -expressing EVs included small ribonucleoproteins (SmE and SmF), a ribosomal protein (NHP2), and a collagen (Vkg; Figure 4b). Proteins in these categories were found to be major EV components in a proteomic analysis of S2 and Kc167 cell EVs (Koppen et al., 2011). We think it likely that some or all these proteins are abundant contaminants that do not actually interact with Sas but happened to be present at higher levels in the IP from Sas-expressing cells vs. the IP from control cells. We did not further examine any of these proteins.
Since dArc1 was enriched in Sas FL preparations purified from EV lysates with anti-V5, we then investigated whether it binds to the Sas ICD (which would be in the EV interior) by co-IP in S2 cells. We coexpressed Myc epitope-tagged dArc1 with a fusion protein in which the V5-tagged ECD of mouse CD8 was attached to the TM domain and the 37 aa ICD of Sas. We then IP'd cell lysates with anti-Myc, and detected V5-mCD8 ECD -Sas TM-ICD and dArc1-Myc by western blotting. We observed that dArc1 co-IP'd with the Sas ICD fusion protein (Figure 4c).
The Sas ICD sequence contains the sequence motif YDNPSY, which is a PTB-binding motif (NPXY) that overlaps by two amino acids with an SH2-binding motif (YXXP) that is also a potential Abl tyrosine Purified EVs from control S2 cells or S2 cells expressing V5-Sas FL were lysed and IP'd with anti-V5, followed by protease digestion and mass spectrometry analysis. (b) Mass spectrometry results. The 7 proteins present at the highest levels in IPs from V5-Sas FL EVs relative to IPs from control EVs (>6 fold ratio) are listed. Sas itself was the most highly enriched protein, as expected. dArc1 and dArc2 were enriched by 22-fold and 6-fold, respectively. (c) Co-IP/western blot analysis of association between Sas and Arc fusion proteins in transfected S2 cells. S2 cells were transfected with the V5-mCD8 ECD -Sas TM-ICD fusion protein construct, or with equivalent constructs in which the Sas ICD was replaced by the Appl or APP ICD, with or without Myc-tagged dArc1 or mammalian (rat) Arc (rArc FL ) constructs. Lysates (Input 1%) were blotted with anti-Myc and anti-V5 (left), and IP'd with anti-Myc nanobody and blotted with anti-V5 and anti-Myc (right). Anti-V5 bands of the correct size were observed in anti-Myc IPs when dArc1-Myc was expressed with Sas or Appl ICD constructs (red symbols and boxes), and when Myc-rArc FL was expressed together with Sas or APP ICD constructs (blue symbols and boxes)(n=6). (d) Direct binding of purified GST-dArc1 and GST-rArc FL fusion proteins to Sas, APP, and Appl ICD peptides, as well as to scrambled and YDNPSY deletion mutant Sas peptides. Biotinylated peptides were bound to streptavidin magnetic beads, which were incubated with GST-Arc proteins, followed by western blotting of bead-bound proteins with anti-GST. GST-dArc1 bound to the wild-type, but not to scrambled or YDNPSY deletion mutant Sas ICD peptides, while GST-rArc FL bound to wild-type Sas and APP ICD peptides. (e) Sequences of the complete Sas, APP, and Appl ICDs, corresponding to biotinylated peptide sequences. The conserved tyrosine motif is boxed, with tyrosines in red. *, stop codons.
The online version of this article includes the following source data for figure 4: Source data 1. Source data files include raw and labelled images for the western blots shown in panels (c) and (d) and an Excel file of the table in panel (b).
Source data 2. Co-IP analyses raw data.
Source data 3. Peptide binding assay raw data.

Source data 4. MS analysis result table.
kinase substrate sequence (Colicelli, 2010;Figure 4e). The NPXY motif is the target for binding of the Numb PTB (Li et al., 1998). This suggests that an SH2 protein and a PTB protein might compete for binding to this sequence, if the first tyrosine was phosphorylated to create an SH2 docking site. The PTB domain of Numb does not require tyrosine phosphorylation to bind to its NPXY target. Interestingly, in an earlier mass spectrometric analysis, we found that the Shc protein, which contains a phosphotyrosine-binding SH2 domain, was associated with Sas purified from S2 cells treated with pervanadate to induce high-level tyrosine phosphorylation.
We searched for other Drosophila CSPs containing a sequence with similar properties in their ICDs, and found only one, Appl, which has the sequence YENPTY but is otherwise unrelated to the Sas ICD. Human APP, the mammalian ortholog of Appl, contains the same sequence in its short ICD (Figure 4e), as do the two APP paralogs, APLP1 and APLP2. We then replaced the Sas ICD in the V5-mCD8 ECD -Sas TM-ICD construct with the Appl and APP ICDs, and found that the Appl ICD protein co-IP'd with dArc1-Myc (Figure 4c), implicating the Y(D/E)NP(S/T)Y sequence in binding to dArc1. This sequence contains the consensus motif for binding of mammalian Arc to TARPγ2, CaMKII, and NMDA receptor peptides, which is X-P-X-(Y/F/H) Nielsen et al., 2019). Arc binds to the NMDA receptor as a monomer (Nielsen et al., 2019). The TARPγ2 Arc-binding peptide is RIPSYR, which is similar to the sequences in Sas (PSYK) and APP (PTYK). Accordingly, we expressed Myc-tagged mammalian Arc (rArc FL ) in S2 cells and examined whether it could co-IP with the V5-mCD8-ICD fusion proteins. We observed that Myc-rArc FL was able to co-IP with the Sas and APP ICDs (Figure 4c).
The co-IP data indicate that the Sas, APP, and Appl ICDs associate with dArc1 and Arc, but does not show that the proteins bind directly to each other. To evaluate this, we made the complete Sas, APP, and Appl ICDs (Figure 4e), as well as a scrambled version of the Sas ICD and a deletion mutant of the Sas ICD that lacks the YDNPSY sequence, as biotinylated peptides, and bound these to streptavidin-coupled magnetic beads. To make purified Arc proteins for binding, we expressed dArc1 and mammalian Arc as GST fusion proteins in E. coli. We mixed the beads with purified GST-dArc1 and GST-rArc FL proteins and examined whether we could observe specific binding. As a positive control, we made purified Numb PTB domain, and showed that it bound as expected to the Sas, APP, and Appl peptides, which all contain the NPXY PTB-binding motif, but not to the scrambled Sas peptide or the YDNPSY deletion mutant. In the peptide binding assay, we observed that dArc1 directly bound to the Sas ICD peptide, but not to the other peptides. Mammalian Arc also bound to the Sas ICD peptide, as well as to the APP ICD peptide (Figure 4d).
Mammalian and Drosophila Arc are not orthologs, and are apparently derived from independent Ty3/gypsy retrotransposon lineages (Ashley et al., 2018;Pastuzyn et al., 2018;Hantak et al., 2021). The fact that both proteins mediate intercellular communication suggests that they may be products of convergent evolution. Our results implicate the Y(D/E)NP(S/T)Y sequence as a determinant of binding to Arcs (Figure 4e), and show that fly and mammalian Arcs can bind to the same peptide sequences.
Our data also suggest that APP might be a CSP that has a relationship to Arc which is similar to that of Sas to dArc1. APP also localizes to EVs (Laulagnier et al., 2018;Pérez-González et al., 2020). The connection between APP and Arc will be of interest to explore in future studies, since Arc has been implicated in AD pathogenesis (Wu et al., 2011;Landgren et al., 2012;Bi et al., 2018). The first Y in the YENPTY motif in APP has been reported to be a substrate for the Abl tyrosine kinase (Zambrano et al., 2001). If YENP was phosphorylated, it would become a docking site for a class of SH2 domain proteins, and binding of this protein(s) could occlude Arc binding to the adjacent PTYK sequence. The Abl inhibitor imatinib (Gleevec), which would be expected to block phosphorylation of this site, inhibits formation of β-amyloid peptide (Aβ) (Netzer et al., 2003), and binding of Arc to APP could be relevant to this effect.

Sas can facilitate intercellular transfer of dArc1 and its mRNA in vivo
Sas is not required for loading of dArc1 capsids into EVs, since dArc1 mRNA is a normal component of EVs from cell lines that do not express Sas. If it behaves like mammalian Arc in its interactions with peptides (Nielsen et al., 2019), dArc1 might bind to Sas as a monomer. Perhaps Sas recruits dArc1 monomers (possibly bound to mRNA via their Zn 2+ fingers) to nascent EVs during their biogenesis, and they then assemble into capsids. Binding of Sas to dArc1 may help to increase the probability that Sas-bearing EVs contain dArc1 capsids bearing dArc1 mRNA. One function of Sas might then be to deliver the EVs and their dArc1 capsid cargo, including encapsulated RNA, to specific recipient cells that express the Sas binding partner Ptp10D.
Having shown that Sas FL can move within larvae and that it binds to dArc1, which is a known component of EVs that mediates intercellular communication, we then examined whether it can cause dArc1 to move from source cells into recipient cells in vivo. To establish an assay system for dArc1 capsid movement in embryos, we first expressed V5-Sas FL in late stage 16 embryonic SGs together with RFP, and observed that V5 signal moved to the gut and tracheae, while RFP was retained in the SGs as expected. When V5-Sas short was expressed in SGs, however, V5 staining was only observed in the SGs (Figure 5-figure supplement 1). This is consistent with the failure of V5-Sas short to move from SGs to imaginal discs in larvae (Figure 3c).
To examine dArc1 protein transport, we needed to express untagged dArc1 and visualize it with antibody against dArc1 (Ashley et al., 2018), because we were unsuccessful in detecting movement of tagged versions of dArc1. dArc1 is made at very low levels in embryos. In late stage 16 control embryos (Sage-GAL4/+ or Sage-GAL4 >Sas FL ), we observed faint staining throughout the embryo, with higher levels in the gut ( Figure 5-figure supplement 2). We then expressed dArc1 from a UAS construct that contained only the dArc1 open reading frame (ORF), flanked by heterologous 5' and 3' UTR sequences. The 3' UTR was derived from SV40. When we expressed dArc1 alone in SGs, we observed bright anti-dArc1 staining in the SGs and increased staining relative to controls in the gut and in dots in the body wall. When Sas FL and dArc1 were expressed together, dArc1 staining in the gut was further increased ( Figure 5-figure supplement 2).
To localize dArc1 staining in the body wall and compare it to Ptp10D staining, we examined dissected 'fillets' at high magnification. For reference, Figure 5-figure supplement 1 shows the evolution of Ptp10D expression in fillets from stage 14 to late stage 16. VNC expression continuously increases during this time period, while tracheal expression begins in stage 14, decreases in stage 15, and re-emerges at stage 16, at which time Ptp10D is expressed in the main tracheal trunk and major tracheal branches. Figure 5d' shows that, in late stage 16 embryos expressing both Sas FL and dArc1 in SGs, there were many bright puncta stained with anti-dArc1 in the dorsal tracheal trunk, which expresses Ptp10D. These puncta appeared similar to those previously observed at larval NMJs (Ashley et al., 2018). They were not detectable in control embryos (Sage-GAL4/+; Figure 5a').
There were lower numbers of fainter dArc1 puncta in tracheal trunks of the two other genotypes (Sage >dArc1 and Sage >Sas FL ) (Figure 5b' and c'). Endogenous Sas is expressed at low levels in SGs, and endogenous dArc1 mRNA is also present in SGs (Figure 5g), although dArc1 protein is not detectable. Endogenous Sas FL may be able to transport some of the overexpressed dArc1, and overexpressed Sas FL might transport some endogenous dArc1, giving rise to the observed puncta. It is also interesting that dArc1 (and dArc1 mRNA; see below) is observed in tracheal cells, but not in VNC neurons, which also express Ptp10D at high levels. There is a glial sheath around the VNC at late stage 16, and this might block access of EVs to Ptp10D-expressing neurons. Alternatively, perhaps there are cofactors required for EV binding and/or internalization that are not expressed in neurons.
Dramatic effects of Sas FL on dArc1 expression and localization were observed when endogenous dArc1 mRNA was examined by fluorescence in situ hybridization (FISH) in embryos expressing the UAS-dArc1 ORF construct in SGs. To detect mRNA, we used the 700 nt antisense 3' UTR probe employed in the (Ashley et al., 2018) paper to visualize dArc1 mRNA puncta at the NMJ. Note that this probe does not recognize overexpressed dArc1 mRNA made from the UAS construct, because that contains only the dArc1 ORF and no dArc1 3' UTR sequences. In late stage 16 control embryos (Sage-GAL4/+), we observed faint FISH signals in the SGs and a few puncta elsewhere in the embryo (Figure 5e). A similar pattern was seen in Sage >Sas FL embryos (Figure 5g). When dArc1 was expressed from the UAS-dArc1 ORF construct, we observed bright FISH signals in SGs with the endogenous dArc1 3' UTR probe (Figure 5f). There were also scattered puncta in other parts of the embryos. This shows that exogenous dArc1 induces expression of endogenous dArc1 mRNA (or stabilizes the mRNA). No signal was observed when a sense dArc1 3' UTR probe was used for FISH ( Figure 5-figure supplement 2). Finally, when Sas FL and dArc1 were expressed together, we observed a completely different pattern, in which the entire tracheal system is lit up by the FISH signal for the endogenous dArc1 3' UTR ( Figure 5h). The foregut and esophagus also stain brightly. By contrast, in embryos expressing Sas short There is weak expression of dArc1 mRNA in the SGs in controls. When dArc1 (from an ORF construct) is expressed alone, bright SG staining is observed, indicating that exogenous dArc1 increases expression of endogenous dArc1 mRNA. There are also scattered dArc1 mRNA puncta elsewhere in the embryo. When Sas FL and dArc1 are expressed together, bright dArc1 mRNA FISH staining of the entire tracheal system is observed (arrows indicate dorsal tracheal trunks), as well as the foregut (arrowhead) and esophagus (n>100 embryos examined for each genotype; representative results are shown). i, i', high-magnification views of dArc1 mRNA in the tracheae in an obliquely mounted (anterior to the left, dorsal up) embryo expressing Sas FL and dArc1 in SGs. i' is a higher-magnification inset (yellow dotted outline) from (i). Arrow in (i), SG; arrowhead, foregut loop. Arrow in i', dorsal tracheal trunk; arrowhead, transverse connective. Scale bar in (e) (applies to e-h), 50 µm; scale bar in (a) (applies to a-d), 10 µm; scale bar in (i), 50 µm; scale bar in (i'), 50 µm.
The online version of this article includes the following figure supplement(s) for figure 5:  and dArc1 in SGs, the dArc1 FISH signal is observed in the SGs, with only a few puncta elsewhere in the embryo ( Figure 5-figure supplement 2), consistent with the fact that Sas short cannot move within the embryo (Figure 5-figure supplement 1).
When an antisense probe for the SV40 3' UTR sequence in the UAS-dArc1 construct was used for FISH, we observed signal only within the SGs, even when Sas FL was coexpressed ( Figure 5-figure  supplement 2). This suggests that dArc1 mRNA lacking the endogenous dArc1 3' UTR sequences is not efficiently loaded into dArc1 capsids that can move elsewhere in the embryo. This is consistent with the findings of Ashley et al., 2018, who showed that the dArc1 3' UTR is required for dArc1 mRNA transfer at the NMJ. No signal was observed when a sense SV40 3' UTR probe was used for FISH ( Figure 5-figure supplement 2). Figure 5i and i' show the tracheae and SGs at higher magnification in side views of an embryo expressing both Sas FL and dArc1 in SGs. The dorsal tracheal trunk (arrow) and the transverse connective (arrowhead) both display bright dArc1 FISH signals. Note that, because this is a confocal image (optical section), the cells at the edges of the tracheal trunk are bright, while the hollow lumen is dark. The brightness of the tracheal FISH signal suggests that it represents not only dArc1 mRNA transferred from capsids, but dArc1 mRNA synthesized in these cells in response to dArc1 protein made from the dArc1 mRNA transported in the capsid. If this is correct, it would represent an amplification Figure 6. Schematic diagram of the processes involved in movement of EVs bearing Sas FL and dArc1 capsids from salivary glands to tracheal cells. Steps 1 and 2, the presence of the dArc1 protein made from a UAS-dArc1 ORF construct increases expression of endogenous dArc1 mRNA in embryonic SGs. EVs with Sas FL on their surfaces bearing dArc1 capsids containing endogenous dArc1 mRNA diffuse or are transported through the hemolymph (Step 3) and bind to Ptp10D-expressing tracheal cells (Step 4). The EVs internalize into the tracheal cells and release dArc1 mRNA (Step 5), and dArc1 protein translated from that mRNA induces expression of more endogenous dArc1 mRNA (Steps 5 and 6). Nuclei, orange circles in 1-2 and blue circles in 4-6. mechanism in which translated dArc1 mRNA from EVs can induce expression of much more dArc1 mRNA in the recipient cells. Finally, we examined whether the Sas ICD is required for dArc1 mRNA transport by expressing dArc1 together with a protein (Sas ECD-TM -GFP) in which the Sas ICD was replaced by GFP. This protein is present in EVs when expressed in S2 cells, but it does not produce any dArc1 FISH signal outside of the SGs (Figure 5-figure supplement 2), indicating that it cannot facilitate transport of dArc1 capsids to tracheal cells.

Conclusions
Our results on movement of Sas EVs containing dArc1 capsids are summarized in the diagram of Figure 6. These findings contribute to the understanding of intercellular communication mechanisms by showing that Sas FL is an EV targeting ligand that can direct internalization of EVs into cells expressing the Sas receptor Ptp10D. Endogenous Sas FL is likely to move between cells in vivo, because tagged Sas FL moves away from both neuronal and non-neuronal cells when ectopically expressed.
dArc1 is related to retrotransposon Gag proteins, and it forms a capsid that contains dArc1 mRNA and is loaded into EVs (Ashley et al., 2018). Overexpressed Sas FL can facilitate transfer of dArc1 capsids into distant Ptp10D-expressing recipient cells in vivo. Whether endogenous Sas FL also transfers dArc1 between cells cannot be determined from our data, and would be difficult to evaluate, since both the sas and dArc1 genes are broadly expressed. The Sas ICD binds directly to dArc1.
Mammalian Arc also forms capsids that are transported via EVs (Pastuzyn et al., 2018), and it binds to the Sas and APP ICDs, which share a tyrosine motif. Full-length APP and some of its proteolytic products are localized to EVs, and EVs from N2a cells bearing tagged APP are internalized into cultured neurons, but not into glia (Laulagnier et al., 2018). It will be interesting to determine if APP EVs contain Arc capsids, and if the presence of APP on Arc-containing EVs causes Arc to be preferentially delivered to a specific population of neurons. This could have implications for Alzheimer's disease research, because APP is the source of β-amyloid peptide and Arc has been linked to β-amyloid accumulation and AD pathogenesis (Wu et al., 2011;Landgren et al., 2012;Bi et al., 2018).

Fly stocks and genetics
The following stocks were used: yw for wild-type control, ap-GAL4 (Bloomington 50156), UAS-mCD8::GFP (Bloomington 5130), UAS-myr::mRFP (Bloomington 7118), UAS-mCherry.NLS (Bloomington 38424), sas15 (null mutant)(Bloomington 2098), Sage-GAL4 (a gift from Deborah J. Andrew), Ptp10DEP1172 (Bloomington 11332), UAS-dArc1 (Bloomington 37532), UAS-Numb (a gift from Yuh Nung Jan), UAS-SasFL and UAS-V5-SasFL , Arc1esm18 (Bloomington 37530). Crosses and embryo collections were performed at room temperature. For overexpression experiments, embryos were shifted to 29 °C for at least 120 min prior to fixation and staining and 3rd instar larvae were shifted to 29 °C for overnight for further analysis. For the EV targeting experiments, imaginal discs from 3rd instar larvae were harvested at room temperature and incubated in 200 µl of S2 supernatant overnight at 29 °C before fixation and staining. There are 10,000-50,000 cells in a 3rd instar imaginal disc. Given the results from the NTA analysis, we can conclude that ~140,000 EVs are present in 200 µl of supernatant from V5-Sas FL -expressing S2 cells. We used 5 wing discs per incubation, so the ratio of EVs to cells is ~0.5 to~2. The relative V5 signal intensities on the imaginal discs were measured by densitometry analysis using ImageJ software.

Cell culture and preparation of EVs and cell lysates
EVs and cell lysates were prepared from S2 cells (ATCC, CRL-1963) that were cultured for four days at 22℃ in Schneider's medium (Gibco) supplemented with 10% exosome-free FBS (#EXO-FBSHI-50A-1, SBI) to avoid contamination from Bovine serum exosomes. S2 cells were authenticated and confirmed to be free of mycoplasma by ATCC. DNA constructs were transiently transfected into S2 cells using Effectene (Qiagen). EVs for western blot analysis and electron microscopy were collected using Total Exosome Isolation reagent (#4478359, Invitrogen) from the supernatants of S2 cultures. This kit has been found to produce exosomes of equivalent quality from mammalian cells (with respect to the presence of exosome markers and the depletion of non-exosome proteins) to those generated using ultracentrifugation (Skottvoll et al., 2019). One part of the reagent and two parts of supernatant were mixed and incubated at 4℃ overnight. Pellets of EVs were collected after centrifugation at 10,000 x g for 60 min at 4℃. The EV pellets were resuspended in PBS for western blot analysis.
To prepare EVs using the ultracentrifugation protocol of Théry et al., 2006, S2 cells were cultured in a medium with 10% exosome-free FBS (SBI) and the cultures were grown in 100 mm Petri dishes at 22℃. Cells were collected at 1-1.5×10 6 cells/ml. The supernatant from a 50 ml culture was filtered using a 0.22 µm filter to eliminate dead cells and large debris prior to further purification by ultracentrifugation (Optima MAX-XP, Beckman Coulter). The filtered supernatants were spun at 100,000 x g for 70 min and the collected EV pellets were resuspended in PBS. Then, the samples were spun again at 100,000 x g for 70 min. The final EV pellet was resuspended in 60 µL of PBS and processed for further analyses or stored at -80℃.
For the EV targeting experiments between S2 cells, supernatants from transiently transfected donor cells were collected and filtered using 0.22 µm PVDF membranes before resuspension and incubation with the recipient cells. Two days before the supernatant swap between EV donor and recipient cell cultures, the recipient cells were transiently transfected with DNA constructs. The recipient cells were incubated in the supernatants with EVs from donor cells for 2 hr at 22℃. For western blot analyses, cell lysates were prepared using RIPA cell lysis buffer. To measure the size and number of EV particles from S2 cell culture, collected EV pellets were subjected to NTA by System Biosciences, LLC (Palo Alto, CA, USA) using a NanoSight instrument. The NTA measurements rely on light scattering to extract particle size and the number of particles in a sample and the NTA software (Version 2.3) collects data on multiple particles to calculate the hydrodynamic diameter of each particle using the Stokes-Einstein equation (System Biosciences, LLC).

Mass spectrometry analysis
Samples were lyophilized and proteins were trypsin-digested as previously described (Pierce et al., 2013). A total of 200 ng of digested peptides were analyzed as previously described (Sung et al., 2016). Briefly, peptides were loaded onto a 26 cm analytical HPLC column (75 µm inner diameter) packed with ReproSil-Pur C 18AQ 1.9 µm resin (120 Å pore size; Dr. Maisch, Ammerbuch, Germany). Peptides were separated with a 120 min gradient at a flow rate of 350 nl/min at 50 °C (column heater) using the following gradient: 2-6% solvent B (7.5 min), 6-25% B (82.5 min), 25-40% B (30 min), 40-100% B (1 min), and 100% B (9 min), where solvent A was 97.8% H 2 O, 2% ACN, and 0.2% formic acid, and solvent B was 19.8% H 2 O, 80% ACN, and 0.2% formic acid. Samples were analyzed using an EASY-nLC 1000 coupled to an Orbitrap Fusion operated in data-dependent acquisition mode to automatically switch between a full scan (m/z=350-1500) in the Orbitrap at 120,000 resolving power and an MS/MS scan of higher-energy collisional dissociation fragmentation detected in the ion trap (using TopSpeed). The automatic gain control (AGC) targets of the Orbitrap and ion trap were 400,000 and 10,000.

Mass spectrometry data
Raw data were searched using MaxQuant (version 1.5.3.30) (Cox and Mann, 2008;Wagner et al., 2011) against the Uniprot D melanogaster database. Fragment ion tolerance was 0.5 Da. Precursor mass tolerance was 4.5 ppm after automatic recalibration. Searches were permitted up to two missed tryptic peptide cleavages. Cysteine carbamidomethylation was designated as a fixed modification while Methionine oxidation and N-terminal acetylation were designated as variable modifications. False discovery rates were estimated to be <1% using a target-decoy approach. Complete data are in Supplementary file 1.

Protein expression and purification
To express and purify Arc proteins in the E. coli system, the cDNAs of dArc1, dArc2 and rArc were subcloned into the pGEX-4T-1 vectors together with GST-6xHis-tags and TEV protease cleavage site. Arc proteins were expressed in E. coli strain BL21 (DE3) grown in LB broth by induction of log-phase cultures with 1 mM isopropyl-β-D-thiogalactopyranoside (IPTG) and incubated overnight at 23 °C. Cells were pelleted and resuspended in B-PER lysis buffer (#78243, Thermo Fisher Scientific) before centrifugation to collect cell lysates.
Tagged Arc proteins were pulled down using Ni-NTA resin column and the eluates with GST-6xHis-dArc1 and GST-6xHis-rArc proteins were used for peptide binding assays.

Electron tomography and immuno-EM
For imaging of EVs by electron tomography (ET), EVs were prepared using the Exosome Isolation Kit as described above. Supernatant was removed and replaced with ~10 ml 10% Ficoll, 5% sucrose in 0.1 M sodium cacodylate trihydrate with minimal disturbance of the pellet. Pellets were transferred to brass planchettes (type A/B; Ted Pella, Inc) and ultra-rapidly frozen with a HPM-010 high-pressure freezing machine (Bal-Tec/ABRA). Vitrified samples were transferred under liquid nitrogen to cryo-tubes (nunc) containing a frozen solution of 2.5% osmium tetroxide, 0.05% uranyl acetate in acetone and placed in an AFS-2 Freeze-Substitution Machine (Leica Microsystems, Vienna). Samples were freeze-substituted at -90 °C for 72 hr, warmed to -20 °C over 12 hr, held at -20° for 12 hr, then warmed to room temperature. Samples were rinsed 3 x with acetone and infiltrated into Epon-Araldite resin (Electron Microscopy Sciences). Resin was polymerized at 60 °C for 24 hr.
Serial semi-thin (170 nm) sections were cut with a UC6 ultramicrotome (Leica Microsystems) using a diamond knife (Diatome Ltd., Switzerland). Sections were collected onto Formar-coated copper/ rhodium slot grids (Electron Microscopy Sciences) and stained with 3% uranyl acetate and lead citrate. Colloidal gold particles (10 nm) were placed on both surfaces of the grid to serve as fiducial markers for subsequent image alignment. Grids were placed in a dual-axis tomography holder (Model 2040; Fischione Instruments, Inc) and imaged with a Tecnai T12 transmission electron microscope (Thermo-Fisher Scientific) at 120 k eV. For dual-axis tomography, grids were tilted +/-62° and images acquired at 1° intervals. The grid was rotated 90° and a similar tilt-series was recorded about the orthogonal axis. Tilt-series data was acquired automatically using the SerialEM software package. Tomographic data was calculated, analyzed and modeled on iMac Pro and M1 computers (Apple, Inc) using the IMOD software package.
For immuno-EM, EV pellets were prepared as per above. Supernatant was removed and pellets fixed with 4% paraformaldehyde in PBS for 1 hr. Pellets were then infiltrated with 2.1 M sucrose in PBS over 24 hr, with >3 changes of the infiltration solution during that time. Pellets were placed onto aluminum sectioning stubs, drained of excess liquid and frozen in liquid nitrogen. Cryosections (100 nm) were cut at -140 °C with a UC6/FC6 cryoultramicrotome (Leica Microsystems) using cryodiamond knives (Diatome Ltd). Cryosections were collected with a wire loop containing 2.3 M sucrose in PBS and transferred to Formvar-coated, carbon-coated, glow-discharged 100-mesh copper/ rhodium grids (Electron Microscopy Sciences) at room temperature. Nonspecific antibody binding sites were blocked by incubating the grids with 10% calf serum in PBS for 30'. Sections were then labeled with 1° antibodies (diluted in 5% calf serum/PBS) for 2 hr, rinsed 4 x with PBS, then labeled with 10 nm and/or 15 nm gold-conjugated 2° antibodies (diluted in 5% calf serum/PBS) for 2 hr. Grids were rinsed 4 x with PBS, 3 x with dH 2 O then simultaneously negatively stained and stabilized with 1% uranyl acetate, 1% methylcellulose in dH 2 O. Immuno-EM samples were imaged as per the tomography samples, above.

Peptide binding assay
For the peptide binding assay, biotinylated peptides (wt Sas ICD , Sas ICD variations (scrambled and ΔYDNPSY), APP ICD and Appl ICD ) made by RS Synthesis, Inc, were incubated with Streptavidin magnetic beads (#88817, Pierce) for 45 min at 4 °C and the beads were extensively washed with TBST. Purified GST-6xHis-dArc1 and rArc proteins were added to the beads with bound biotinylated peptides and incubated at 4 °C overnight. Similar experiments were performed with Numb PTB domain protein purified from E. coli. The beads were carefully washed with TBST and eluates prepared for western blot analysis following the standard protocol described above.

Fluorescent in situ hybridization (FISH)
The FISH protocol was a modification of protocols from Kosman et al., 2004. Fixed L16 whole embryos were prepared using standard protocols and rinsed with ethanol quickly four times. Then the embryos were permeabilized twice with a mixture of xylenes and ethanol (1:2, v/v) and washed three times with ethanol for 5 min each. To rehydrate the embryos, the embryos were washed with 100%, 50% and 0% methanol in PBT sequentially for 30 min each step. The rehydrated embryos were permeabilized again using proteinase K (20 µg/mL in PBT) for exactly 7 min and washed three times for 5 min each in PBT followed by a second fixation (5% paraformaldyhyde and 1% DMSO in PBT) for 25 min and washed three times in PBT for 5 min each. Then the embryos were prepared for pre-hybridization by incubation in 50% hybridization buffer (50% formamide, 5 x SSC, 100 μg/ml fragmented salmon testes DNA, 50 μg/ml heparin, 0.1% Tween-20) in PBT for 5 min. For pre-hybridization, embryos were incubated in hybridization buffer for more than 90 min at 55℃ while changing the buffer every 30 min. The prehybridized embryos were incubated in DIG-tagged dArc1 mRNA probe for 18 hr at 55℃ for annealing. The embryos were washed with hybridization buffer three times for 30 min each at 55℃, after which the buffer was replaced with replaced the buffer with PBT containing rhodamine-conjugated sheep anti-DIG antibody (#11207750910, SigmaAldrich) overnight at 4℃. Then the embryos were washed and mounted for confocal microscopy.

Probe preparation
Probes for detection of endogenous dArc1 mRNA were designed against a 760 nt region of the dArc1 mRNA 3' UTR sequence, which was used for FISH in a previous study (Ashley et al., 2018). Probes for detetion of dArc1 mRNA from the UAS-dArc1 ORF construct were designed against a 616 nt region of SV40 3' UTR sequence. To generate antisense and sense probes for dArc1 mRNA, cDNA sequences from dArc1 were PCR amplified and purified to use as positive and negative probe templates. The same procedure was used for antisense and sense probes for the SV40 3' UTR. The DNA templates were heated to 55℃ for 2 min and then put back on ice. Transcription reactions were set up to label probes with digoxigenin (DIG, # 11277073910, Roche) and incubated at 37℃ for 2 hr. Probes were precipitated and resuspended in hybridization buffer and stored at -20℃.

Additional files
Supplementary files • Supplementary file 1. Sas IP MS analysis raw data.

• MDAR checklist
Data availability Key data generated or analysed during this study are included in the manuscript and supporting files.