Biogenesis and function of extracellular vesicles in cancer

Extracellular vesicles (EVs) are heterogeneous multi-signal messengers that support cancer growth and dissemination by mediating the tumor-stroma crosstalk. Exosomes are a subtype of EVs that originate from the limiting membrane of late endosomes, and as such contain information linked to both the intrinsic cell "state" and the extracellular signals cells received from their environment. Resolving the signals affecting exosome biogenesis, cargo sorting and release will increase our understanding of tumorigenesis. In this review we highlight key cell biological processes that couple exosome biogenesis to cargo sorting in cancer cells. Moreover, we discuss how the bidirectional communication between tumor and non-malignant cells affect cancer growth and metastatic behavior.


Introduction
Cells can communicate with each other by releasing extracellular vesicles (EVs), which interact with and modify the behavior of target cells at close or distant sites.
Based on their origin, EVs can be divided into subgroups of 'small' vesicles generated within endosomal compartments called "exosomes", and vesicles that bud directly from the plasma membrane. Regardless their origin, all EVs enclose or expose on their surface a multitude of biomolecules including RNA, lipids, proteins and possibly DNA. However, the mechanisms underlying the biogenesis of different EV subtypes and the sorting of these molecules are extremely difficult to define as EVs are generally analyzed in bulk and contaminated with protein, lipid and viral-like particles.
As critical mediators of cell-to-cell communication, EVs support cancer progression by mediating the crosstalk between tumor and stromal cells in their direct environment. However, EVs from cancer cells as well as from healthy tissues also seem to have systemic properties (Peinado, et al., 2012;Pucci, et al., 2016;Thomou, et al., 2017;Zomer, et al., 2015). Since exosomes derive from major sorting hubs at the limiting membrane of endosomes, they are at the center of a signaling network that connects external stimuli with tumor cell intrinsic features. Consistent with this notion, the difference in composition between exosomes and other EV subtypes and between EVs and parental cells indicates the existence of regulated sorting mechanisms determining the phenotypic and functional properties of these vesicles.
In this review we describe the key findings on the mechanisms underlying exosome biogenesis and cargo sorting. We discuss how these processes are deregulated in

A C C E P T E D M A N U S C R I P T
6 cancer and consequently change the functional properties of exosomes. However, because most of the current purification protocols and functional studies do not accurately discriminate between exosomes and plasma membrane-derived vesicles their respective role(s) in cancer remain elusive. We highlight the in vivo studies that establish EVs as crucial mediators in the tumor-to-stroma, stroma-to-tumor and tumor-to-tumor communication and discuss to what extent their effect can be attributed to the functional transfer of defined cargo molecules.

Intraluminal vesicle budding and MVB formation
Exosomes are nanosized ( Multiple mechanisms for MVB biogenesis have been described. A critical role for lipid raft microdomains in MVB formation was established in 2008 by Trajkovic and colleagues. The authors demonstrated that ceramide, generated from sphingomyelin A C C E P T E D M A N U S C R I P T 7 hydrolysis by neutral sphingomyelinase 2 (nSMase2), induces negative membrane curvature via its cone-shaped structure leading to ILV budding into MVBs (Trajkovic, et al., 2008). Subsequent studies into the genesis of endosome-derived vesicles revealed that, apart from the lipid composition and physical characteristics of endosome membrane microdomains, exosomes biogenesis is closely connected to the sorting of cargo molecules.
The Endosomal Sorting Complex Required for Transport (ESCRT) machinery is the most extensively described pathway of MVB biogenesis, responsible for the sorting of ubiquitinated proteins into ILVs (Henne, Buchkovich, & Emr, 2011). This process is initiated by ESCRT-0, which recognizes and retains ubiquitinated proteins in the late endosomal membrane. After initial involution of the limiting membrane into the MVB lumen triggered by ESCRT-I/II, ESCRT-III forms a spiral-shaped structure that constricts the budding neck and the ATPase VPS4 drives membrane scission. Despite extensive research into the role of ESCRT in MVB biogenesis, whether MVBs generated by ubiquitin-dependent mechanisms are mainly destined to lysosome for cargo degradation or can fuse with the plasma membrane to release exosomes is still unclear (Christ, Raiborg, Wenzel, Campsteijn, & Stenmark, 2017; A C C E P T E D M A N U S C R I P T 9 Borger, & Powis, 2014; Wei, et al., 2017). This type of shift may also occur in nontransformed cells such as antigen-presenting cells, which are known to release large amounts of exosomes upon stimulation (Muntasell, Berger, & Roche, 2007 Pu, & Bonifacino, 2016;Pu, et al., 2015) as well as RAB7-dependent protein complexes (Raiborg, et al., 2015). The multiple late endosomal transport mechanisms might be subject to differential regulation or control the trafficking of distinct MVB subtypes. Alternatively, the mechanisms responsible for secretory MVB transport might be cell type-specific, as demonstrated for the RABs that regulate MVB docking to the plasma membrane. RAB35 mediates MVB docking in oligodendroglial cells (Hsu, et al., 2010), whereas RAB27 controls this process in several cancer cell lines both in vitro (D. Hoshino, et al., 2013;Ostrowski, A C C E P T E D M A N U S C R I P T 10 et al., 2010) and in vivo (Peinado, et al., 2012), possibly by stabilizing the branched actin filaments that form the MVB docking site (Sinha, et al., 2016 for small GTPases such as RAL-1 in this process (Hyenne, et al., 2015).
Over the last decade, several SNARE proteins, such as YKT6, Syntaxin-1a and Syntaxin-5 have been implicated in exosome release from different cell types and organisms (Gross, Chaudhary, Bartscherer, & Boutros, 2012;Hyenne, et al., 2015;Koles, et al., 2012;Ruiz-Martinez, et al., 2016). Moreover, we and others demonstrated a role for the SNARE proteins SNAP23, Syntaxin-4 and VAMP7 in the release of exosomes from tumor cells (Fader, Sanchez, Mestre, & Colombo, 2009;Verweij, et al., 2018;Wei, et al., 2017). These proteins form an exocytic SNAREcomplex that has previously been reported to promote invadopodia formation and tumor cell invasion (Williams, McNeilly, & Coppolino, 2014), processes that indeed have been linked to cancer exosome release (D. Hoshino, et al., 2013). SNARE complex formation and membrane fusion is tightly controlled by multiple regulatory mechanisms (Snyder, Kelly, & Woodbury, 2006;Sudhof & Rothman, 2009). In the A C C E P T E D M A N U S C R I P T 11 context of tumor exosome release, Hoshino and colleagues demonstrated a role for the calcium-sensing SNARE interacting protein synaptotagmin-7 (D. Hoshino, et al., 2013). Furthermore, SNARE activity is partly controlled by their phosphorylation profile, which influences their localization or the interaction with SNARE partners (Snyder, et al., 2006). Indeed, recent independent studies revealed that phosphorylation of SNAP23 promotes exosome secretion (Verweij, et al., 2018;Wei, et al., 2017). Interestingly, SNAP23 phosphorylation at Ser95 by PKM2 mediates constitutive release in tumor cells, whereas, as we recently demonstrated, phosphorylation of Ser110 triggers exosome release in response to histamine stimulation (Verweij, et al., 2018;Wei, et al., 2017). These findings are consistent with the idea that multiple regulatory mechanisms control SNARE activity through phosphorylation of different residues and support a role for the microenvironment in the regulation of exosome release.
Finally, when MVBs fuse with the plasma membrane, the exosomes are released into the extracellular space where they interact with the extracellular matrix, influence cells in the microenvironment, but can also enter the circulation via lymph or blood.
Importantly, several studies report that in certain cell-types a proportion of exosomes remains attached to the cell surface, where they might function as signaling platforms for juxtacrine communication (Edgar, Manna, Nishimura, Banting, & Robinson, 2016;Mobius, et al., 2002;Verweij, et al., 2018).
Altogether, the studies into exosome release mechanisms indicate that it is a tightly controlled process that is regulated at multiple levels between ILV budding and MVB

ACCEPTED MANUSCRIPT
A C C E P T E D M A N U S C R I P T 12 fusion. This realization implies that exosomes are unlikely merely an alternative method for 'waste' disposal, but suggests that exosomes play a physiological role.
Accordingly, the regulation of exosome release in vivo is likely a dynamic process where cells adapt the magnitude of release for each exosome subpopulation in response to defined external and internal stimuli. It is important to note that current methods to study exosome biogenesis mechanisms are based on the isolation of EVs from culture supernatant, which requires long-term cell culture (24-72 hours), thus making it impossible to capture the dynamics of exosome release. To overcome this limitation we recently developed tetraspanin-based reporters that can be used to study exosome release in real time by visualizing the fusion of MVBs with the plasma membrane (Verweij, et al., 2018). Using this method we revealed a role for G protein-coupled receptor signaling in the regulation of exosome secretion. Future studies using these reporters could provide further knowledge on how dynamic changes in the tumor microenvironment might contribute to cancer exosome release.

Microvesicle biogenesis
In addition to the MVB-derived exosomes, extracellular vesicles can directly bud and pinch of from the plasma membrane. These vesicles are often generically referred to as microvesicles although they are extremely heterogeneous in size, ranging from exosome-like EVs of 50nm to EVs as large as 10μm. Consistent with their heterogeneity, microvesicles can be generated via multiple distinct mechanisms that partially overlap with those involved in exosome biogenesis. For instance, the ESCRT machinery is involved in the production of nano-sized vesicles that are enriched in

A C C E P T E D M A N U S C R I P T
13 cell surface proteins, reflecting their plasma membrane origin (Nabhan, Hu, Oh, Cohen, & Lu, 2012;Q. Wang & Lu, 2017). Furthermore, similar to the role of nSMase2 in ILV formation, acid sphingomyelinase induces ceramide-dependent microvesicle production, indicating a general role for sphingomyelinases in the biogenesis of MVB-and plasma membrane-derived EVs (Bianco, et al., 2009).
Another mechanism of microvesicle biogenesis is linked to non-apoptotic plasma membrane blebs that are frequently observed in highly aggressive cancer cells.
These blebs continuously expand and retract at the cell surface and contribute to cell motility via a currently unknown mechanism (Fackler & Grosse, 2008

Extracellular vesicle cargo
The functional properties of EVs in the tumor microenvironment are dictated by their cargo and by their release and uptake dynamics. Identifying the components responsible for the pro-tumorigenic effects of cancer EVs and the pathways leading

A C C E P T E D M A N U S C R I P T
14 to their incorporation into vesicles is currently one of the major challenges in the field.

RNA cargo
In 2006, Ratajczak et al were first to propose horizontal transfer of RNA between donor and recipient cells (Ratajczak, et al., 2006). Subsequently, Valadi and Skog independently demonstrated that mRNA molecules transported by EVs can be actually translated into protein, providing evidence of virus-independent horizontal transfer of genetic material between cells (Skog, et al., 2008;Valadi, et al., 2007).
This finding opened the way to intense efforts to address the functionality of EVassociated RNA molecules in vivo. Elegant studies using reporter systems under the control of the CRE recombinase demonstrated that vesicles-enclosed CRE mRNA can

A C C E P T E D M A N U S C R I P T
15 be transferred in vivo over long distances (Ridder, et al., 2014;Ridder, et al., 2015;Zomer, et al., 2015). Using this methodology in combination with intravital imaging, Zomer et al. showed that EV released by aggressive tumor cells can induce the acquisition of malignant traits into less aggressive ones, suggesting a role for EV-RNA in this process (Zomer, et al., 2015).

RNAs via exosomes was provided in 2010, when using viral miRNAs endogenously produced in Epstein Barr virus-infected B cells Pegtel et al demonstrated that small
RNAs are secreted via exosomes and can regulate gene expression in recipient dendritic cells (Pegtel, et al., 2010). While certain EV-associated miRNAs have a demonstrated role in cancer (Le, et al., 2014;Zhou, et al., 2014), conclusive evidence that their transfer via EVs is responsible for promoting cancer progression in vivo is still lacking. Indeed, it is difficult to discriminate EV-mediated miRNA transfer from endogenous miRNA expression in recipient cells. A compelling demonstration of EVmediated miRNA transfer in vivo was recently provided by Thomou and colleagues.
Using an adipose tissue-specific Dicer knockout mouse model, the authors demonstrated that brown fat is a major source of EV-associated circulating miRNAs.
Transplantation of normal adipose tissue into the engineered mice not only restored the circulating levels of EV miRNAs, but could regulate gene expression in distant organs, an effect that could be reproduced administering normal serum EVs (Thomou, et al., 2017).
A number of defined RNA binding proteins (RBP) have been suggested to mediate RNA sorting into exosomes Mukherjee, et al., 2016;Santangelo, et al., 2016;Shurtleff, Temoche-Diaz, Karfilis, Ri, & Schekman, 2016;Teng, et al., 2017;Villarroya-Beltri, et al., 2013), yet how these proteins connect to the endosomal system is far from resolved, and the stoichiometry between a given RBP and RNA molecules has not been established. In fact, two independent studies showed that certain pol III transcripts are sorted into exosomes after uncoupling from their binding partners (Baglio, et al., 2016;Nabet, et al., 2017). Thus there are presumably multiple sorting mechanisms functioning in parallel.
The localization of AGO2 and other components of the RNA-induced silencing complex in close proximity of MVBs suggests their involvement in exosomal sorting of miRNAs (Gibbings, Ciaudo, Erhardt, & Voinnet, 2009). However, because AGO2 has not been consistently found in EVs, its role in miRNA sorting remains controversial (McKenzie, et al., 2016;Melo, et al., 2014;Shurtleff, et al., 2016;Van Deun, et al., 2014), in particular since AGO is a stable protein found outside of vesicles in human plasma (Arroyo, et al., 2011). A debated study from Melo et al. proposes that breast cancer derived exosomes possess the protein machinery to process pre-miRNAs into mature miRNAs. Interestingly, the authors report that mature miRNAs generated within these cancer exosomes influence the transcriptome of non-malignant target cells promoting transformation (Melo, et al., 2014). Though highly provocative, this study awaits independent confirmation.

A C C E P T E D M A N U S C R I P T
18 remain to be elucidated, the fact that the EV gDNA fragments are equally distributed over the whole genome suggests a random process (Kahlert, et al., 2014;Thakur, et al., 2014 Interestingly, secretion of DNA fragments via EVs prevents activation of cytoplasmic DNA sensors, thereby contributing to cellular homeostasis .
This aspect might be particularly relevant in cancer, where elevated levels of DNA damage might require efficient removal of cytoplasmic DNA via EVs. Consistent with this hypothesis, cancer cells seem to display enhanced EV production.

A C C E P T E D M A N U S C R I P T
19

Deregulation of EV release in cancer
A large body of evidence suggests that cancer cells release higher amounts of EVs compared to non-malignant cells, which makes the EV biogenesis machinery or components thereof attractive targets for anticancer therapy. Indeed, overexpression of ESCRT components as well as syntenin and heparanase has been observed in various tumors (Koliopanos, et al., 2001;Koo, et al., 2002;R. T. Liu, et al., 2002;Oh, Stanton, West, Todd, & Wagner, 2007;Toyoshima, et al., 2007;Vlodavsky, et al., 1999). In addition, hyperactivation of RalB in pancreatic carcinoma and colorectal cancer, overexpression of YKT6 in non small-cell lung cancer and elevated Rho-ROCK signaling observed in various tumor types might enhance EV production in tumor cells (Lim, et al., 2006;Martin & Der, 2012;Morgan-Fisher, Wewer, & Yoneda, 2013;Ruiz-Martinez, et al., 2016).
Enhanced EV release in cancer cells can be determined by both cell-intrinsic and environmental signals. Activation of oncogenic signaling pathways, for instance by EGFRvIII and H-RAS V12 (Al-Nedawi, et al., 2008;Lee, et al., 2014;Takasugi, et al., 2017), can increase EV production in cancer cells. Moreover, it was recently shown that the proto-oncogene SRC specifically stimulates the syntenin exosome biogenesis pathway by phosphorylating the cytosolic domain of syndecan and syntenin (Imjeti, et al., 2017). Enhanced cancer EV release can also depend on the (de)regulation of the membrane fusion machinery. For instance, overexpression of PKM2, a glycolytic enzyme implicated in the Warburg effect, enhances EV secretion by phosphorylating the tSNARE SNAP23 (Wei, et al., 2017).

A C C E P T E D M A N U S C R I P T
20 Apart from cell-autonomous mechanisms, microenvironmental conditions such as hypoxia can contribute to the regulation of both microvesicle and exosome release from cancer cells (King, Michael, & Gleadle, 2012;L. Li, et al., 2016;. Interestingly, while hypoxia-induced microvesicle production requires HIF1adependent expression of RAB22 (T. Wang, et al., 2014), enhanced exosome release mainly involves the phosphorylation of the Akt-substrate PRAS40 (Guo, et al., 2017), which provides the opportunity to specifically evaluate the contribution of different hypoxia-induced EV subtypes to cancer progression.

A C C E P T E D M A N U S C R I P T
21 Deregulation of cancer EV release and cargo critically affect the crosstalk between tumor and stromal in the tumor microenvironment and even at distant (pre-) metastatic sites. In the following section we will discuss the consequences of such alterations on cancer development and progression.
Although cancer EVs are carriers of tumor-antigens that could in principle mediate antitumor immunity (Robbins & Morelli, 2014;Wolfers, et al., 2001), extensive evidence suggests that they mainly act by suppressing the immune response.
Independent studies demonstrated that tumor-derived EVs can inhibit proliferation and activation of CD8+ cells while promoting the expansion of regulatory T cells (Clayton, et al., 2007;Wieckowski, et al., 2009) through mechanisms that have not yet been defined. In addition to their effect on the adaptive immune system, cancer EVs can 'educate' innate immune components towards a pro-tumorigenic phenotype (Chalmin, et al., 2010;Haderk, et al., 2017;Shinohara, et al., 2017). For instance, Hsp72 on the surface of cancer EVs activates myeloid-derived suppressor cell (MDSC) by inducing the IL-6/Stat3 signaling pathway through a TLR2-dependent mechanism (Chalmin, et al., 2010). Moreover, EV-associated miRNAs have been reported to polarize tumor-associated macrophages towards a pro-tumorigenic M2 phenotype (Y.-L. Shinohara, et al., 2017). Interestingly, a recent study by Haderk and coworkers showed that CLL EV-enclosed Y RNA induces the expression of the immune checkpoint protein PD-L1 as well as the production of pro-tumorigenic cytokines in monocytes by triggering TLR7/8 (Haderk, et al., 2017). However, since Y RNAs are found abundantly in many types of cancer exosomes but also in exosomes from healthy cells, urine and plasma, additional factors must play a role.
Nevertheless, this study adds to the existing evidence that exogenous EV-protected

A C C E P T E D M A N U S C R I P T
23 non-coding RNAs can act through mechanisms independent from gene regulation, by activating pattern recognition receptors (PRR) triggering an innate immune response (Baglio, et al., 2016;Fabbri, et al., 2012;Nabet, et al., 2017).  Webber, et al., 2015). Consistent with this finding, we recently demonstrated that EVs released by malignant bone tumor cells carry high levels of membrane-associated TGFβ, which induces IL-6 release in mesenchymal stem cells (MSCs) (Baglio, et al., 2017). Importantly, injection of EVeducated MSCs in mice bearing bone tumors was associated with STAT3 activation and increased metastatic dissemination, suggesting that the inflammatory loop initiated by cancer EVs within the tumor microenvironment enhances the metastatic behavior of tumor cells. In accordance with the observations by Webber and colleagues, soluble TGFβ failed to induce a prometastatic phenotype in MSC. We conclude that EV-bound growth factors might have different or highly enhanced biological activities compared to their soluble counterparts, as suggested by a recent independent study (Feng, et al., 2017). Whether this is dependent on a different conformation acquired by the growth factor when bound to the EV surface, or on the presence of co-stimulatory signals needs further investigation.

A C C E P T E D M A N U S C R I P T
24

Effects of tumor EVs on extracellular matrix
Apart from their effects on local or recruited tumor-associated stromal cells, EVs  (Sung, Ketova, Hoshino, Zijlstra, & Weaver, 2015). The same group later demonstrated that tumor exosomes not only enhance cell migration speed, but also promote directional movement towards a chemotactic gradient, although the exact mechanisms remain unclear (Sung & Weaver, 2017). Importantly, tumor exosomes contribute to invasive cell behavior by stimulating the formation and function of invadopodia. The branched actin filaments in invadopodia are important docking and fusion sites for MVBs (Sinha, et al., 2016), and exosome secretion not only increases invadopodia formation and stability, but also promotes matrix degeneration by transporting the proteinase MT1-MMP to the plasma membrane (D. Hoshino, et al., 2013).

EVs in pre-metastatic niche formation
The tumor-promoting activity of tumor-derived EVs is not restricted to the tumor microenvironment. Rather, cancer EVs enter the circulation and reach distant organs, where they can generate favorable environmental conditions enabling the outgrowth of disseminated tumor cells. This process, known as pre-metastatic niche (PMN) formation, requires a series of predefined steps involving induction of

A C C E P T E D M A N U S C R I P T
25 vascular leakiness, alteration of stromal components and immune suppression (Peinado, et al., 2017).
The first evidence for the EV involvement in pre-metastatic niche formation was provided by Peinado and colleagues (Peinado, et al., 2012). Soon after these elaborate studies, the same group demonstrated that EVs released from metastatic cells not only preferentially target and reprogram stromal cells in their predicted destinations, but also can drive organ-specific metastatic outgrowth.
Intriguingly, the authors found that EVs from lung-tropic metastatic breast cancer cells could redirect metastasis formation of bone-tropic tumor cells to the lung (A. Hoshino, et al., 2015). They concluded that this effect was dependent on the integrin profile of the vesicles, which dictates EV organotropism and promotes an inflammatory response in recipient cells. While this fascinating study strengthen the

A C C E P T E D M A N U S C R I P T
26 role of cancer EVs in the pre-metastatic niche formation, whether the organ-specific tropism of cancer EVs is solely determined by their integrin repertoire remains an open question. Moreover, future studies should be conducted to clarify the relative contribution of EV-mediated signaling and cancer cell-intrinsic factors to the organotropic behavior of malignant cells (A. Hoshino, et al., 2015).
Not only cancer EV-associated proteins, but also EV-enclosed small RNAs can contribute to the generation of the pre-metastatic niche through distinct mechanisms (Fong, et al., 2015;Tominaga, et al., 2015;Zhou, et al., 2014). MiR-105 secreted by breast cancer cells induces vascular leakiness in distant organs by targeting the tight junction protein ZO-1 (Zhou, et al., 2014) while miR-122 inhibits glucose uptake by stromal cells in the PMN, thus increasing glucose availability and metastatic outgrowth (Fong, et al., 2015). Furthermore, non-coding small nuclear RNAs transferred by cancer EVs can trigger TLR3 signaling in lung epithelial cells, which results in neutrophil recruitment and lung metastatic niche formation (Y. Liu, et al., 2016).
While these studies indicate a critical role for cancer EVs in the sequence of events preceding the arrival of cancer cells to the future metastatic sites, the physiological relevance of these findings awaits confirmation. Indeed, these studies have been mainly conducted by exposing stromal cells to arbitrary amounts of cancer EVs or by directly injecting tumor vesicles in the circulation. In physiological conditions the amount of cancer EVs entering the circulation and reaching defined (pre-metastatic) organs might be limited for instance by internalization by phagocytic cells. In line

A C C E P T E D M A N U S C R I P T
27 with this notion, Pucci and colleagues demonstrated that melanoma EVs are trapped by a barrier of subcapsular sinus macrophages in the tumor-draining lymph nodes, which prevents their dissemination (Pucci, et al., 2016). Importantly, the authors show that cancer progression as well as certain anticancer treatments can disrupt the macrophage barrier, thereby allowing cancer EVs to enter the lymph node cortex and trigger a tumor-promoting humoral immunity. This study demonstrates that many context-dependent factors may influence the systemic function of tumor EVs in vivo, and highlights the need for preclinical models that faithfully reproduce the dynamics of EV release, tropism and stroma education.

Stroma-to-tumor communication
Early studies on EV-mediated interaction in cancer described the tumor-stroma communication mostly as a unidirectional process where tumor EVs alter the behavior of stromal cells. This concept was reconsidered in 2012, when a provocative study by Luga and colleagues showed that fibroblast-derived vesicles can be modified and recycled by cancer cells to activate an autocrine Wnt-planar cell polarity signaling enhancing tumor cell motility and metastasis (Luga, et al., 2012).

A C C E P T E D M A N U S C R I P T
28 dictated by the exposure of the producing cells to cancer-derived signals. For instance, polarization of tumor-associated macrophages towards the M2 phenotype determines the secretion of EV-associated oncogenic miRNAs regulating breast cancer cell invasiveness (Yang, et al., 2011). Interestingly, Roccaro and colleagues showed that EVs released by mesenchymal stem cells can have opposite effects on MM progression (Roccaro, et al., 2013). non-immunogenic delivery system that could be used for transferring therapeutics to tumors, and have been already used in the clinic to treat graft versus host disease (Kordelas, et al., 2014). Moreover, as these vesicles retain the surface expression profile of the MSCs (H. S. Kim, et al., 2012), they might share with their parental cells a similar tropism for tumor sites (Baglio, Pegtel, & Baldini, 2012).
However, the mechanisms regulating endogenous RNA recognition or evasion from PRRs remained unclear. This question was addressed by two recent studies (Baglio, 29 et al., 2016;Nabet, et al., 2017). We and others independently demonstrated that within the producing cell pol III transcripts are mainly associated with their RNAbinding partners that shield them from recognition by intracellular sensors. Under specific conditions leading to the alteration of the RNA-RBP stoichiometry, naked RNA molecules can be incorporated into exosomes and, upon delivery into recipient cells, activate PRR and ISG expression. Nabet and colleagues showed that this mechanism is strongly implicated in the tumor-stroma communication promoting cancer progression (Nabet, et al., 2017). Specifically, the authors demonstrated that stroma activation by tumor cells increases the transcription of the non-coding RNA RN7SL1, which is then sorted in its naked form into stromal exosomes. The antiviral immune response triggered by unshielded RN7SL1 upon delivery in cancer cells enhances tumor growth, metastasis formation and therapy resistance, thereby linking stromal activation to EV-mediated inflammation and cancer progression.

A C C E P T E D M A N U S C R I P T
Strengthening the role of EV-associated RNA in stroma-to-tumor communication, an impressive study by Zhang et al. revealed that EV-mediated transfer of astrocytederived miRNAs support the outgrowth of breast cancer brain metastases . The authors show that both mouse and human breast cancer cells reversibly lose the expression of the tumor suppressor gene PTEN when disseminating to the brain, but not to other metastatic sites. They then combine a series of elegant in vivo approaches, including astrocyte-specific miRNA depletion and adenoviral-mediated inhibition of EV release, to demonstrate that the transfer of astrocyte-derived miR-19a is responsible for the reversible PTEN downregulation leading to metastatic outgrowth. This study establishes a role for stromal EVs in the

A C C E P T E D M A N U S C R I P T
30 co-adaptation between tumor cell and host environment leading to the development of organ-specific metastasis. It would be interesting to investigate whether the astrocyte-EV content changes in response to cancer-released signals and how this would influence the outgrowth of brain metastasis. Moreover, future research should be aimed to understand whether this model can be extended to the formation of metastatic lesion in other organs.
Finally, a novel form of EV communication based on the transfer of mitochondrial DNA has been recently implicated in the development of drug resistance (Sansone, et al., 2017). Sansone and colleagues found that cancer-associated fibroblast derived EVs, enclosing the full mitochondrial genome, restore defects in oxidative phosphorylation of hormone therapy-induced dormant breast cancer stem cells, leading to escape from dormancy and development of hormone therapy-resistance (Sansone, et al., 2017). This original work raises multiple questions: which EV subclass is responsible for the transfer of the mitochondrial genome? How is mitochondrial DNA incorporated into EVs? Most importantly, how can the mitochondrial DNA be functionally transferred into the correct subcellular location and restore the metabolic activity in recipient cells? Although many aspects need to be clarified, this study reveals a novel modality of metabolic communication in cancer.

ACCEPTED MANUSCRIPT
A C C E P T E D M A N U S C R I P T 31

Tumor-to-tumor communication
The  EVs not only transfer malignant traits between cancer cell subpopulations, but can also propagate tumor heterogeneity (Godlewski, et al., 2017;Ricklefs, et al., 2016).

ACCEPTED MANUSCRIPT
A C C E P T E D M A N U S C R I P T 32

Concluding remarks
It is now fully established that EVs and exosomes have functional properties in vivo that can be applied in clinical settings. The role of exosomes, as defined by their intracellular endosomal origin, remains unclear as in vivo tools to follow exosomes release and biogenesis are still lacking. The only way to conclusively identify a role for exosomes in cancer is by interfering with their biogenesis, cargo-loading and/or function. This is not easy since, as we described in this review, multiple mechanisms exist that presumably act in parallel, and the nature and dynamics of exosome subpopulations are far from understood. How to block one exosome pathway?
While much has been learnt from in vitro characterization and administration of EVs to (tumor bearing) mice, specific inhibition of exosome biogenesis in cancer cells will provide defining answers into the actual role of cancer released exosomes in oncogenesis and progression.

Conflict of Interest statement
The authors declare that there are no conflicts of interest.

Acknowledgements
This work has received funding from the Dutch Cancer Society (