Directional cell movement through tissues is controlled by exosome secretion

Directional cell movement through tissues is critical for multiple biological processes and requires maintenance of polarity in the face of complex environmental cues. Here we use intravital imaging to demonstrate that secretion of exosomes from late endosomes is required for directionally persistent and efficient in vivo movement of cancer cells. Inhibiting exosome secretion or biogenesis leads to defective tumour cell migration associated with increased formation of unstable protrusions and excessive directional switching. In vitro rescue experiments with purified exosomes and matrix coating identify adhesion assembly as a critical exosome function that promotes efficient cell motility. Live-cell imaging reveals that exosome secretion directly precedes and promotes adhesion assembly. Fibronectin is found to be a critical motility-promoting cargo whose sorting into exosomes depends on binding to integrins. We propose that autocrine secretion of exosomes powerfully promotes directionally persistent and effective cell motility by reinforcing otherwise transient polarization states and promoting adhesion assembly.

D irectional cell movement involves the coordination of multiple subcellular processes. In order for cells to migrate effectively, polarization signals and protrusions must be stabilized to form a dominant leading-edge protrusion 1 . Adhesion assembly is likely to significantly contribute to signal reinforcement at the leading edge 1 ; however, the mechanisms by which cells spatially coordinate adhesion assembly during persistent cell migration are poorly understood. In tissues, the extracellular matrix (ECM) environment adds further complexity that could provide conflicting signals to migrating cells if it is not coordinated with chemical gradients.
One potential mechanism that could coordinate initial directional signalling with adhesion formation is vesicle trafficking. Polarization of and trafficking from the Golgi apparatus has been shown to regulate cell migration, although the cargoes are unknown [2][3][4] . Endocytic recycling of integrins has also been shown to impact cell motility 5,6 . Although early studies focused on long-and short-loop integrin-recycling pathways 6 , recent data have identified trafficking of the a5b1 integrin to and/or from late endosomal/lysosomal (LE/Lys) compartments as a key step in cell migration 7,8 . In addition, we found that the ECM molecule fibronectin (FN) is endocytosed and resecreted from a LE/Lys compartment to promote in vitro migration of fibrosarcoma, breast cancer 9 and epicardial 10 cells. Finally, although the cargoes are unknown, additional studies pinpoint secretion from LE/Lys compartments as important for leukocyte chemotaxis and epithelial migration 11,12 . Thus, current data suggest that a key decision step in cell migration may be whether otherwise degradatory LE/Lys compartments fuse with the plasma membrane to release important cargo, such as ECM components and their receptors.
Among the components of LE/Lys, compartments that might affect cellular migration are exosomes. Exosomes are small secreted vesicles that carry bioactive cargoes, including growth factors, angiogenic factors, transmembrane receptors, proteinases, ECM molecules and RNAs 13 . It has been shown that purified exosomes can promote adhesion and motility of cells 14,15 ; however, it is unclear how essential the process of exosome secretion is to cell migration or how it might affect underlying processes such as polarization. It is also unclear how exosome and/or LE/Lys secretion might impact cell migration through complex tissue environments.
To understand how LE/Lys secretion and exosomes control in vivo cell motility, we performed xenograft tumour cell motility studies in the chorioallantoic membrane (CAM) of chick embryos. This system is highly advantageous because it enables high-resolution live-imaging studies of cell migration through a physiologic collagen-rich stromal tissue environment [16][17][18] . We find that exosome secretion is critical for persistent directional migration of tumour cells in the chick CAM, likely due to stabilization of leading-edge protrusions. We further identify exosomes as critical carriers of ECM that promote adhesion assembly, a key step in leading-edge stabilization [19][20][21] . Targeting of the ECM molecule FN to exosomes depends on specific binding to integrin receptors, ensuring that ECM secreted on exosomes will match cellular receptors.

Endolysosomal secretion controls HT1080 migration in vivo.
To determine the impact of LE/Lys secretion on in vivo motility, we inhibited two canonical regulators of LE/Lys secretion, Rab27a and Synaptotagmin-7 (Syt7), in HT1080 fibrosarcoma cells. Rab27a controls plasma membrane docking of multivesicular late endosomes (MVE) 22 whereas Syt7 controls fusion of LE/Lys compartments, including MVE, with the plasma membrane 23,24 . Syt7 and Rab27a expression were stably downregulated by the expression of specific targeting short hairpin RNAs (shRNAs; Fig. 1a). Green fluorescent protein (GFP)-expressing HT1080 cells carrying the scrambled control (HT1080 Scrambled ) or targetspecific shRNA (HT1080 Rab27aKD and HT1080 Syt7KD ), were imaged intravitally after local engraftment or intravenous (i.v.) injection into the CAM of chick embryos (Fig. 1b) 17,18 . Primary tumours formed after engraftment enable the visualization of cells migrating away from the tumour periphery, while i.v. injection enables high-resolution single-cell tracking of individually extravasated cells. After engraftment, the HT1080 Scrambled , HT1080 Rab27aKD or HT1080 Syt7KD cells all formed tumours in the CAM with little obvious difference in tumour size. Examination of the invasive front of each tumour at 4 days after inoculation revealed that fewer KD cells had migrated away from the tumour as compared with control cells (Fig. 1c,d). Following i.v. injection, cells that have extravasated from the bloodstream into the CAM grow to form colonies. Consistent with the previous studies 18 , individual control HT1080 cells had an elongated morphology and migrated rapidly through the CAM to form dispersed, loosely connected colonies (Fig. 1e,f). In contrast, extravasated HT1080 Rab27aKD or HT1080 Syt7KD cells exhibited a rounded morphology and formed few colonies that were overall larger in size, suggesting ongoing proliferation but defective migration (Fig. 1e,f) 18,25 .
Endolysosomal secretion promotes persistent and fast motility.
To directly measure cancer cell motility in vivo, we performed time-lapse imaging of cancer cells migrating within the CAM. GFP-expressing HT1080 Scrambled , HT1080 Rab27aKD and HT1080 Syt7KD were imaged in the CAM for 16 h during a period of 1-3 days after i.v. injection. Analysis of the movies revealed that HT1080 Scrambled exhibited polarized migration, allowing them to move in a directionally persistent manner. Conversely, HT1080 Rab27aKD and HT1080 Syt7KD cells that initially moved in one direction would frequently stop and reverse course (representative cell tracks shown in Fig. 2a, see also Supplementary Movies 1 and 2), leading to diminished overall displacement from the initial position (Fig. 2b,c). HT1080 Rab27aKD and HT1080 Syt7KD cells also exhibited an overall defect in the total path length travelled by the cells, regardless of the directionality (Fig. 2b,c). To determine whether there was a distinct defect in directionality in KD cells, separate from any defects in overall speed of movement, the persistence index was calculated by dividing the displacement rate by cell speed (Fig. 2d,e). Fully persistent movement would result in a straight migration track and give a persistence index of 1. Control cells had a persistence index of B0.4. By contrast, HT1080 Rab27aKD and HT1080 Syt7KD cells had persistence indexes of B0.2, indicating a strong defect in directionally persistent migration, independent of defects in cell speed (Fig. 2e). Wind-Rose plots of representative cell tracks further illustrate that endolysosomal secretion promotes persistent and efficient in vivo migration (Fig. 2f). A similar phenotype was observed in HEp3 squamous carcinoma cells ( Supplementary  Fig. 1).
Endolysosomal secretion controls protrusion dynamics. To further identify cellular features of migration regulated by endolysosomal secretion, we performed high-magnification livecell imaging of cells migrating in the CAM. Analysis of the cellular morphology revealed that HT1080 Scrambled cells were much more elongated than cells inhibited for LE/Lys secretion ( Fig. 3a,b, see also Supplementary Movie 3). It was also apparent that HT1080 Rab27aKD and HT1080 Syt7KD cells made more numerous but shorter-lived protrusions. Quantitation of the protrusion dynamics revealed that KD cells had a 50% increase in ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8164 the number of protrusions formed per frame during migration (Fig. 3c), along with a decrease in the length and persistence of protrusions (Fig. 3d,e). These data indicate that LE/Lys secretion strongly contributes to establishment of a persistent leading-edge protrusion in vivo, which is critical for maintaining the cell directionality during migration. Interestingly, these results are reminiscent of our previous in vitro finding that endolysosomal recycling of ECM promotes the stabilization of leading-edge  9 . To determine if Rab27a and Syt7 also control the stability of protrusions in vitro, we performed live imaging and kymography analysis of the movies. We found a similar in vitro phenotype, with HT1080 Rab27aKD and HT1080 Syt7KD cells making an increased number of short-lived protrusions ( Fig. 3f,g). The strong similarity between the in vitro and in vivo phenotypes suggests that we can exploit powerful available in vitro techniques to identify the mechanisms by which endolysosomal secretion controls cell motility.
Exosomes promote effective cell motility by carrying ECM. A specialized endolysosomal compartment that could potentially control cell migration is the MVE, which on fusion with the plasma membrane releases exosome vesicles to the extracellular milieu 13 . To confirm that exosome secretion is inhibited in HT1080 Rab27aKD and HT1080 Syt7KD cells, exosomes were purified from conditioned media by serial centrifugation and analysed by NanoSight particle analysis and immunoblotting for the exosome marker flotillin and the ECM molecule FN. NanoSight analysis revealed that HT1080 Rab27aKD and HT1080 Syt7KD cells secreted 2.2-to 3-fold fewer exosomes/cell than control cells (Fig. 4a). Consistent with the reduction in exosome secretion, the immunodetection of flotillin and FN was similarly reduced in exosome preparations from HT1080 Rab27aKD and HT1080 Syt7KD cells (Fig. 4b, top panel). However, when an equal number of vesicles was loaded on blots, there was no difference in FN or flotillin levels (Fig. 4b, bottom panel).
To determine if autocrine secretion of exosomes was required for in vitro cell motility, live imaging of HT1080 Scrambled and HT1080 Rab27aKD cells was performed in the absence or presence of purified exosomes. In the absence of exogenously provided exosomes, HT1080 Rab27aKD cells had a defect in the speed of migration (Fig. 4c,d, 'Exo 0' condition). However, unlike the migration in vivo, HT1080 Rab27aKD did not exhibit diminished persistence compared with HT1080 Scrambled , as quantitated by the persistence index (compare Fig. 4e, 'Exo 0', to in vivo migration results in Fig. 2e) and visualized in Wind-Rose plots (Fig. 4f). We speculate that this difference reflects the presence of directional cues in vivo that are not present in our in vitro motility environment.
Exosomes have the potential to control multiple aspects of cell motility, including stimulation of extracellular receptor signalling and as carriers of ECM to promote adhesion formation 14,15,26,27 . As a first test of these possible roles, exosomes purified from control HT1080 cells were added to the cells in two formatseither added into the media at the time of migration like a growth factor (Fig. 4c) or coated overnight onto tissue culture plates similar to ECM coatings (Fig. 4d). On addition of purified exosomes, HT1080 Scrambled cells exhibited increased migration speed, especially when exosomes were coated onto culture dishes. Interestingly, the speed defect of HT1080 Rab27aKD cells was fully rescued on exosome-coated plates ( Fig. 4d) but not with the addition of exosomes into media containing exosome-depleted serum (Fig. 4c). These data suggest that deposition of exosomes to the substrata where it is accessible to the basal cell surface is important for their ability to promote cell migration.
The presence of FN on exosomes (Fig. 4b), the lamellipodial stability defect of KD cells (Fig. 3) and the selective rescue of Rab27a-KD cells on exosome-coated plates (Fig. 4d) suggest that exosomes may serve a critical function in providing matrix attachment for migrating cells. By immunoprecipitation, FN is on the outside of exosomes, indicating that it is indeed of the expected topology to support adhesion and migration ( Supplementary Fig. 2). To determine whether the presence of FN on exosomes is important for their ability to support migration, we prepared FN-depleted exosomes by culturing control cells for 10 days in FN-depleted serum before collecting 48-h conditioned serum-free media for exosome purification. This procedure led to a 2.5-fold decrease in the FN content of exosomes, compared with control preparations (Fig. 4g). When coated onto culture dishes and tested in single-cell motility experiments, FN-depleted exosomes were less efficient at rescuing single-cell motility than control exosomes (compare Fig. 4h to Fig. 4d). Thus, while 10 mg ml À 1 coating of control exosomes fully rescued motility of Rab27a-KD cells, twice as many FN-depleted exosomes were required to rescue their motility. The rescue of Rab27a-KD cell motility by 20 mg ml À 1 The concomitant decrease in FN and the membrane marker flotillin in exosome preparations from Rab27a-and Syt7-KD cells that were loaded according to equal cell number (Fig. 4b, top blot) suggests that FN is not merely a contaminant of our ultracentrifugation (UC) preparation technique. Furthermore, the size distribution of exosomes and microvesicles in our UC fractions were as expected (Fig. 5a) and approximately threefold fewer MVs are secreted by our cells, suggesting that it is exosomes rather than microvesicles that promote cell motility in our system. However, to test the specificity of exosomes in carrying FN and promoting cell motility, we further purified our UC exosome preparations by sedimentation into a density gradient of Optiprep (DG exosomes). Western blot analysis of the fractions revealed that FN was purified in two peaks-one that included the exosome markers HSP70 and flotillin along with the FN-binding integrin a5 subunit and another peak that did not have any of those markers and probably represents FN fibrils not associated with exosomes (Fig. 5b). Microvesicle preparations contained no FN or HSP70 but did contain a small amount of flotillin and a5 integrin, which are both plasma membrane markers and may be found in both types of vesicles. To determine whether our Optiprep-purified exosomes could rescue cell motility defects of Rab27a-KD cells, we coated plates and performed the same live-imaging assay as we had performed with UC-exosomes. Compared with one-step UC-exosomes from the same conditioned media preparation, Optiprep-double purified exosomes were equipotent in their ability to rescue cell motility (Fig. 5d). By comparison, the microvesicle fraction from the same conditioned media was unable to rescue motility. As with the UC-exosomes (Fig. 4h), FN-depleted DG exosomes were less efficient in rescuing cell motility compared with control DG exosomes (Fig. 5c,d).
These data are consistent with our recent finding that recycling of endocytosed FN from a LE/Lys compartment drives cell motility 9 and suggest that exosomes are key delivery vehicles for that FN. Consistent with this model, we tested whether endocytosed fluorescent FN was likewise secreted by HT1080 cells migrating in the chick CAM. Indeed, 24 h after i.v. injection of DyLight550-FN-loaded HT1080 cells into the chick CAM, we observed incorporation of the endocytosed FN into streak-like structures reminiscent of adhesions associated with cellular protrusions (Supplementary Fig. 3 and Supplementary Movie 4).
Secreted exosomes promote assembly of nascent adhesions. Our data suggest that a major role of exosomes is to transport ECM and presumably regulate adhesions. To investigate whether exosome-containing MVE are transported to and secreted at specific cellular locations, we performed live confocal imaging of cells expressing the MVE marker mCherry-CD63. To simultaneously visualize adhesions, we expressed GFP-paxillin. As previously reported, CD63-positive tubules and vesicles were observed moving dynamically in cells ( Fig. 6a and Supplementary Movie 5). In some cases, these vesicular structures appeared to target peripheral adhesions ( Fig. 6a and Supplementary Movie 5, arrows); however, the majority of the CD63 was in endosomes away from the plasma membrane and some CD63-positive endosomes also carried paxillin ( Fig. 6a and Supplementary Movie 5, arrowhead).
To definitively identify exosome secretion sites, we constructed a novel imaging probe, which has the pH-sensitive GFP-pHLuorin moiety inserted into the first extracellular loop of CD63 (CD63-pHLuorin). Because pHLuorin-GFP fluorescence is quenched at the acidic late endosomal pH, but fluoresces brightly at the neutral pH of the extracellular environment 30 , it is ideal to track proteins that are exocytosed from endosomes. To validate this construct, we co-expressed it with mCherry-Rab5-Q79L, which both marks and leads to the enlargement of early-to-late-transitioning endosomes 31 , in control and Rab27a-KD cells. Consistent with the construct marking secretory MVE, pHLuorin-CD63 fluorescence was present on the cell surface of control cells, but not of Rab27a-KD cells ( Supplementary Fig. 5). In addition, Bafilomycin treatment of Rab27a-KD cells to inhibit acidification of endosomes reveals its presence in internal vesicles. Finally, since many Rab5-positive endosomes will not be fully acidified yet, note the internal co-localization of pHLuorin-CD63 with some Rab5-Q79L endosomes in control cells but less co-localization in Rab27a-KD cells, which should have an accumulation of secretory MVE downstream of early endosomes due to the docking defect 22 .
To visualize exosome secretion in migrating cells, wide-field epifluorescence movies of HT1080 cells expressing pHLuorin-CD63 and mCherry-paxillin migrating in two-dimensinal were performed ( Fig. 6b and Supplementary Movie 6). Overall, pHLuorin-CD63 fluorescence was observed at the front of the cell in thick ruffles and at the back of the cell at attachment sites. At the back of the cell, a striking feature was prominent 'slime trails' of fluorescence that were frequently left behind the migrating cells (Fig. 6b,c). Examination of phase-contrast images from the same movies reveals that the rear of the cell is far in advance of the 'slime trails' and does not reveal any apparent bits of torn-off plasma membrane (Fig. 6b). Furthermore, examination of mCherry-paxillin fluorescence gathered at the same time shows that only green fluorescence is present in the trail (Fig. 6b and Supplementary Movie 6). These data suggest that the primary source of the slime trail is exosomes and not torn-off plasma membrane or large adhesions left behind. These trails appeared to serve an adhesive function based on attachment to membrane at the rear of the cell that is particularly evident in the movies (Supplementary Movie 6). To further analyse the localization of pHLuorin-CD63 with respect to leading-edge protrusions, higher speed movies were taken. Kymographs showing protrusion at the leading edge over time were obtained and revealed that pHLuorin-CD63 was not enriched in the initial flat lamellipodial protrusions ( Fig. 6c and Supplementary Movie 7). Instead there was an increased fluorescence in membranes that ruffled back, which could be due to membrane thickness or adhesion of exosomes after the initial membrane protrusion.
To further define the relationship between exosome secretion and adhesion formation, we performed live total internal reflected fluorescence (TIRF) microscopy of cells expressing both pHLuorin-CD63 and mCherry-paxillin ( Fig. 7a and Supplementary Movie 8). To visualize newly forming adhesions, movies were taken in actively spreading cells. Interestingly, bursts of pHLuorin-CD63 closely preceded adhesion formation. Analysis of these data revealed that 93% of adhesions were preceded by a burst of CD63-pHLuorin, with a median of 2 min lag between CD63 appearance and adhesion formation (Fig. 7b,c).
To specifically test the role of exosome secretion in adhesion formation, we analysed the dynamics of adhesions in GFP-paxillin-expressing control and Rab27a-KD cells. Confocal movies taken in actively spreading cells revealed that Rab27a-KD cells have a selective defect in the rate of adhesion assembly (Fig. 7d,e and Supplementary Movie 9). There was no difference in the rate of adhesion disassembly (Fig. 7f) or in the overall number of peripheral adhesions per frame of the movies (25.16±4.37 for HT1080 Scrambled , 27.65±3.08 (P ¼ 0.3229 compared with scrambled) for HT1080 Rab27aKD1 and 30.40±4.24 (P ¼ 0.2112 compared with scrambled) for HT1080 Rab27aKD2 , P by Student's t-test). Overall, the slow adhesion assembly time led to a significant prolongation of the overall adhesion lifetime (average lifetime t 1/2 for HT1080 Scrambled adhesions ¼ 7.0 ± 0.4 min, compared with HT1080 Rab27aKD1 ¼ 10.7 ± 0.8 min (Po0.002, Student's t-test) and HT1080 Rab27aKD2 ¼ 10.4 ± 1.0 min (Po0.002, Student's t-test). These data are consistent with the known relationship between dynamic adhesions and fast migration speeds [32][33][34] .

FN is targeted to exosomes by interactions with integrins.
Our data indicate that FN becomes specifically targeted to exosomes in a form that promotes efficient adhesion formation. Soluble FN becomes converted to a fibrillar, adhesion-promoting form in association with cellular integrin receptors 35 , which are also frequent exosome cargos. To test whether targeting of FN to secreted exosomes depends on binding to integrins, we inhibited FN-integrin interactions and assessed the resulting FN content of exosomes. Treatment of HT1080 cells with integrin-binding RGD peptide for 48 h greatly decreased the FN content of exosomes compared with cells treated with control RGE peptide (Fig. 7g). Likewise, inhibition of the FN-binding integrin a5b1 by KD of a5 also decreased the FN content of exosomes (Fig. 7g). These data indicate that FN is targeted to exosomes by binding to cognate cellular integrin receptors, and then secreted in adhesive form to promote adhesion formation and effective cell motility.

Discussion
Cell movement through tissues is a complex process that requires a coherent response to potentially divergent tissue cues. Here we ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms8164 identify exosome secretion as a key autocrine mechanism to reinforce directional cues and support productive migration (see model in Fig. 8). In vivo, inhibition of exosome formation, MVE docking or MVE fusion led to similar defects in cell migration speed and persistence. In vitro, purified exosomes, but not shed microvesicles, were able to rescue speed defects of exosome-inhibited cells. Both in vitro and in vivo, secretion of exosomes stabilizes nascent protrusions, a phenotype associated with cellular adhesion [19][20][21] . In vitro live-imaging experiments further demonstrated that a key function of exosomes is to promote adhesion assembly. A critical exosome cargo that drives the adhesion and speed of migrating cells is ECM, as demonstrated by the reduction in speed supported by FN-depleted exosomes and full rescue of speed defects in exosome-inhibited cells by purified FN. Altogether, our results demonstrate that autocrine exosome secretion promotes both efficient and directional cell motility and indicates that a major function of ECM in exosomes is to promote adhesion formation and enhance cell speed.
While exosomes have been implicated in cell motility previously 14,15,26 , it has not been clear how essential to the process they are or what specific steps are regulated. Our data now demonstrate that autocrine secretion of exosomes controls both directional persistence and speed of cells migrating in vivo. To understand in depth the role of exosome-delivered ECM in cell motility, we carried out a variety of live-imaging experiments to both visualize the spatiotemporal relationship between exosome secretion and motility structure formation and demonstrate specific defects of exosome-inhibited cells in adhesion assembly and cell speed but not in protrusion  formation. Altogether, these data reveal a much more important role for exosome secretion in adhesion formation than previously appreciated 36,37 , and particularly highlight the key role of autocrine exosome secretion in motility processes.
A key question raised by our findings is why does the cell need to place FN on the outside of an exosome to promote adhesion assembly? FN is initially secreted from Golgi-derived vesicles in a soluble non-adhesion-promoting form and then assembled into fibres through interaction with integrins on the cell surface 35 . These fibres may not be optimally positioned to promote migration in one direction and their assembly may not be dynamic enough to respond to fluctuating directional cues. By contrast, endocytosis and resecretion of FN-integrin complexes on exosomes likely facilitates dynamic placement of highly competent adhesive substrata that promote cellular motility. In a previous study, we found that resecretion of fluorescent FN at the bottom of cells occurred within 1 h of its addition to the culture medium and was rapidly incorporated into nascent streak-like adhesion structures 9 . Presumably this endocytosis-resecretion process is happening continuously so that at any given time there may be a pool of matrix-carrying exosomes in MVE that could be rapidly secreted and used for migration.
FN is known to be endocytosed in association with integrins and targeted to MVE 8 (Fig. 7g). During the intraluminal vesiculation process that creates exosomes within the MVE, the FN would be placed on the outside of the exosome (Supplementary Fig. 2) bound to integrins in the correct topology to interact with both the outside of the cell and with other ECM such as the collagen fibres in the chick CAM (see model, Fig. 8). In vitro, the major source of FN is serum. In the tumour microenvironment, the potential sources of soluble FN are many, including fibroblasts, endothelial cells and tumour cells, as well as leakage from the vasculature. It is also likely that additional matrix molecules could play a similar role, depending on the receptor profile of the migrating cells. In addition, matrix proteolysis could liberate adhesive fragments that could be internalized and then recycled on the outside of exosomes to promote dynamic adhesion formation and motility.
In addition to ECM, exosomes contain numerous cargoes that could impact cell motility in diverse contexts, including proteinases, growth, chemotactic and other extracellular factors 23,26,27,38 . Indeed, our in vivo data indicate that exosomes not only regulate cell speed but also directionality. Because Rab27a-KD cells did not have a defect in persistence in our random motility in vitro assays, we were not able to address exosome control of directionality in that experimental set-up. One possibility is that proteinases could facilitate persistent movement through tissues by remodelling impeding ECM. Another possibility is that exosomes mediate directional movement via chemokines or other factors. Specific identification of which exosome cargoes affect distinct aspects of cell motility is an important future area of research for both the exosome and motility fields. In addition, juxtaposition of multiple motility-promoting factors in a single exosome or closely apposed exosomes could synergistically modulate complex motility behaviours. Understanding these multivariate interactions will be an ongoing challenge for the field.
Use of the pHLuorin-CD63 probe to observe only extracellular CD63 allowed us to achieve a strong signal-to-noise ratio and observe dynamic relationships between exosome secretion and cellular structures. Our observation of adhesive trails left behind migrating cells is reminiscent of slime trails left by lower organisms, including snails, slugs and slime moulds [39][40][41] . Such slime trails contain complex mixtures of components and serve a variety of purposes, including providing adhesive traction without gluing the organism in place 39,40 , and, in the case of slime moulds, as a chemorepellant to prevent recrossing of the same track 41 . We speculate that trails of exosomes from mammalian cells may provide a similar mixture of functions, including not only adhesion but also the ability to modulate other cellular behaviours via diverse cargo molecules.
In this study, we identified a specific cellular organelle, the MVE, as critical to directional cell motility. As polarization is an intrinsic part of directional movement, this finding is consistent with recent findings that targeted MVE secretion is important for formation of the posterior pole of polarized Dictyostelia 42 . In addition, a recent study found that fibroblast-induced exosomes could mobilize Wnt11 from breast cancer cells to induce planar cell polarity signalling 27 . Interestingly, several molecules that have been previously shown to promote directional cell motility are sorted into exosomes and/or regulate their cargo content. Thus, syndecans were recently shown to control exosome biogenesis 43 and also separately to control directional migration 1 . Likewise, Wnt molecules are major components of exosomes 44 and together with syndecan-4 regulate directional cell motility in vivo 1,27,45 . On the basis of these studies and our findings, we speculate that a number of molecules known to control directional cell motility may be doing so via control of exosome biogenesis or cargo content, or as biological cargoes of exosomes.
In summary, we have shown that exosome secretion is essential for directional and efficient cell migration and enhances migration speed by delivering ECM to promote adhesion formation. These data identify a novel feedback pathway that enhances cell motility via local autocrine secretion.
The CAM model. Use of the chick CAM to model the formation of metastatic colonies and primary tumours is described in detail elsewhere 17,18 . In brief, fertilized eggs decanted into sterilized weigh boats were allowed to mature ex ovo for 9 days. GFP-expressing cells were injected either i.v. or within the CAM tissue on developmental day 10. At indicated times, the ex ovo chick embryos were placed in a custom intravital imaging chamber 46,47 . Images of tumour cells in the CAM were acquired using a Lumar V12 fluorescence stereomicroscope (Zeiss) or an Olympus BX61 equipped with a digital camera controlled with Volocity image acquisition software (PerkinElmer).
In vivo analysis of cell migration and protrusion dynamics. GFP-expressing HT1080 cells (1 Â 10 5 ) were injected i.v. into the CAM of ex ovo chicks on developmental day 10. At 24 h post injection, the ex ovo chicks were transferred to an intravital imaging chamber and placed under a fully automated temperaturecontrolled microscope (99°F, BX61, Olympus 18 ). Time-lapse migration movies were captured every 15 min for 16 h using Â 10/0.40 UPlanSApo objective lens, whereas protrusion dynamics movies were captured at every 2 min for 1 h using Â 20/0.70 W UApo/340 water objective lens. To analyse in vivo cell migration, the nuclear position of each cell was tracked and cell speed, displacement rate and persistence index were analysed using Volocity. For protrusion dynamics analysis, the aspect ratio was calculated using Fiji, the number of protrusions formed was counted manually and length and persistence of protrusions were measured using Metamorph software (Molecular Devices).

Isolation of exosomes.
To collect conditioned media, 80% confluent cells were cultured for 48 h in Opti-MEM. Exosomes were isolated from conditioned media by serial centrifugation at 300g for 10 min, 2,000g (4,000 r.p.m. in Ti45 rotor) for 30 min, 10000g for 30 min (9,300 r.p.m. in Ti45) and 100,000g (30,000 r.p.m. in Ti45) for overnight to, respectively, sediment live cells, dead cells, debris and microvesicles and exosomes (B3.5 mg exosomes from 1 Â 10 6 HT1080s). For further purification of UC-exosomes, the discontinuous iodixanol gradient was prepared 48 . Solutions (40% (w/v), 20% (w/v), 10% (w/v) and 5% (w/v)) of iodixanol were made by diluting OptiPrep (60% (w/v) aqueous iodixanol, Axis-Shield PoC) with 0.25 M sucrose/10 mM Tris, pH 7.5 from the bottom to the top of a 14 Â 89 mm polyallomer tube. UC-exosomes were added on a top of the gradient and continuous gradient was made through UC at 100,000g (24,000 r.p.m. in SW40 Ti rotor) for 18 h. Twelve fractions were collected and each fraction was diluted in PBS. Each fraction was pelleted through another round of ultracentrifugation at 100,000g for 3 h, washed with PBS and resuspended in PBS. To deplete FN in exosomes, cells were cultured in DMEM supplemented with 10% FN-depleted BGS 9 for 10 days before collecting conditioned medium for exosome isolation. Exosomes were quantitated by NanoSight tracking analysis.
In vitro single-cell migration assay. For coating experiments, tissue culture 24-well plates were incubated with appropriately exosome or human plasma FN (Gibco BRL) suspensions in PBS at 4°C overnight before washing and using. Trypsinized cells (3,000 in each well) were then cultured overnight on the coated plates before capturing and analysing time-lapse migration movies using a Nikon Eclipse TE2000E microscope equipped with a 37°C chamber as previously described 9 . The nuclear position of each cell was tracked and cell speed and persistence index was calculated using Metamorph software. For in vitro migration with addback of exosomes in the media, cells were allowed to adhere to tissue culture well plates for 5 h before changing the media to DMEM supplemented with 10% exosome-depleted BGS (prepared by ultracentrifugation for 12 h at 100,000g)± purified exosomes. Then, live-cell imaging was performed as described above.
Adhesion assembly and disassembly assay. GFP-paxillin-expressing cells were plated on glass-bottom dishes (MatTek) coated with 1 mg ml À 1 FN and allowed to adhere for 1 h. Images were acquired with an LSM 510 laser scanning confocal microscope (CarlZeiss) equipped with a Â 63/1.40 NA Plan Apo oil objective lens at 37°C with 5% CO 2 . Images were taken every 30 s for 45 min. The rate constants for assembly and disassembly of adhesions were measured through the Focal Adhesion Analysis Server 49 .
Live imaging of CD63 and focal adhesions. Either mCherry-CD63 was stably expressed in stable GFP-paxillin-expressing HT1080 or pHLuorin-CD63 was transiently co-transfected into HT1080 along with mCherry-paxillin or mCherry-Rab5A Q79L. A day after transfection, cells were replated on glass-bottom dishes coated with FN (1 mg ml À 1 ). Images of mCherry-CD63/GFP-paxillin-or mCherry-Rab5A Q79L/pHLuorin-CD63-expressing cells were acquired with LSM 710 META laser scanning confocal microscope equipped with a Â 63/1.40 Plan-Apochromat oil objective lens at 37°C with 5% CO 2 . Epifluorescence images of pHLuorin-CD63-expressing cells were acquired with Nikon Eclipse TE2000E microscope equipped with a 37°C chamber using a Â 40/0.60 Plan Fluor lens. For migration movies, images were taken every 2 min for 4 h. For lamellipodial protrusion analysis by kymography, images were taken every 6 s for 20 min. Images of pHLuorin-CD63/mCherry-paxillin-expressing cells were taken with a Nikon Eclipse TiE TIRF microscope equipped with a Perfect Focus System using a TIRF Â 100/1.49 NA oil-immersion lens and Neo 5.5 cMOS camera (Andor). For each cell, images were taken every 30 s for 45 min, yielding 91 frames per movie.
Numbers and statistics. For both quantitated data and representative images from experiments, the n values and independent experiment numbers are listed in the figure legends. Imaging data were acquired from at least three independent experiments and multiple images or movies on multiple cells were captured. Cell numbers to be quantitated for motility experiments were determined by our experience and analyses were excluded only if there is an obvious reason for poor data, such as dead or sick-looking cells. For non-quantitated western blots (for example, checking knockdown), they were generally performed a single time. All data sets were tested for normality using the Kolmogorov-Smirnov normality test in SPSS software (IBM). Non-parametric data groups were compared using the Mann-Whitney U-test and plotted as box and whiskers, in which the box indicates the 25-75th percentile and the whiskers indicate the 5-95th percentile, with the line indicating the median. Parametric data were compared using Student's t-test and plotted as mean±s.e. in bar graphs.