Background

Various types of cells are capable of secreting membrane vesicles, collectively termed extracellular vesicles (EVs), under both physiological and pathological states [1]. The amount and/or composition of released EVs change with external stimuli, such as pH, hypoxia and oxidative stress [2,3,4]. Based on their origin and size, EVs are broadly classified into two main classes: exosomes and microvesicles [5, 6]. Exosomes originate from intraluminal vesicles (ILVs) in multivesicular endosomes (MVEs), in which ILVs are generated by the inward budding and fission of endosomal membrane and then released upon fusion of MVEs with the plasma membrane (Fig. 1). Microvesicles, also called oncosomes in case of being released from cancer cells, shed directly from the plasma membrane or its extensions (for example, microvilli, filopodia) by an outward budding and fission (Fig. 1). Apoptotic bodies, derived from membrane blebbing during cell apoptosis, are another common subtype of EVs [7]. Although EVs were initially considered to dispose waste materials, their abilities in transferring cargoes between cells have attracted growing interests over the past decade [1]. The informative cargoes of EVs regulate biological functions at autocrine, paracrine and systemic levels and are transported in protected and directed manners to recipient cells. EV-mediated bidirectional communication between cells has played a key role in regulation of cancer initiation, development and progression [8, 9]. Increasing evidence indicates that enhanced EV secretion from cancer cells and dysregulation of their cargoes are associated with tumorigenesis [10]. Thus, tumor-derived EVs can serve as diagnostic and prognostic biomarkers of cancers as well as novel therapeutic targets and tools [11, 12]. Apart from proteins, metabolites and DNAs of EVs, EV-RNAs are also considered as important intercellular mediators affecting hallmarks of cancer [12]. Multiple RNA species are found in EVs, where non-coding RNAs (ncRNAs), particularly shorter RNA species, comprise the majority of EV-RNA transcripts [13]. The biological functions of these ncRNAs and their underlying mechanisms on recipient cells remain largely unknown and warrant further investigations. In this review, we summarize the cellular machineries and processes of EV formation, secretion and interaction with recipient cells; RNA sorting into EVs; biological roles of EV-ncRNAs, mainly including micro RNAs (miRNAs), long non-coding RNAs (lncRNAs) and circular RNAs (circRNAs), from various cells as well as their molecular mechanisms affecting phenotypes of recipient cells in premetastatic niches and the tumor microenvironment (TME).

Fig. 1
figure 1

Extracellular vesicle biogenesis and secretion in donor cells as well as its interaction with and intracellular fate in recipient cells. Microvesicles directly shed from the plasma membrane, where budding microdomains undergo phosphatidylserine translocation and remodeling of the actin cytoskeleton. By contrast, exosomes originate from endosomal pathway. Deriving from endocytosis, early sorting endosomes accumulate ILVs within the endosomal lumen and then mature into MVEs, where ESCRT components, ceramide, tetraspanins and syntenin could act in parallel or separately to recruit exosomal cargoes and generate ILVs. At this checkpoint, the MVEs can either enter into autophagy-lysosome pathway or exosomal secretion pathway. of note, amphisomes can either fuse with lysosomes or the plasm membrane. Upon secretion into extracellular space, exosomes and microvesicles can bind to the recipient cell surface via ligand-receptor or glycoprotein interactions and initiate signaling, uptake and fusion processes, contributing to transfer functional messages and cellular phenotypes. MVE (multivesicular endosome), EV (extracellular vesicle), PM (the plasma membrane), Ub (ubiquitin), ECM (extracellular matrix), ESCRT (endosomal sorting complex required for transport), SNARE (soluble N-ethylmaleimide-sensitive factor attachment protein receptor)

Full size image

Extracellular vesicle biogenesis

EVs have different modes of biogenesis according to their origin—endosome and the plasma membrane. However, clustered membrane microdomains, certain sorting machineries, membrane invaginations and fission processes are essential for inward-budding vesicles at the limiting membrane of the secretory MVEs (exosomes) and an outward-budding vesicle at the plasma membrane (microvesicles) [5].

Intraluminal vesicle generation and endosomal sorting in MVEs

Within the endosomal system, early sorting endosomes carry membrane cargoes that are internalized from the plasma membrane or originated from the trans-Golgi network, and then they mature into MVEs when ILVs accumulate within the lumen of endosomes [14]. The membrane cargoes could serve as regulators of selective recruitment of sorting machineries [15, 16]. Sorting machineries are required for cargo segregation on microdomains and subsequent inward budding and fission of ILVs.

The membrane remodeling role of the endosomal sorting complex required for transport (ESCRT) machinery in ILV biogenesis is of major importance to the formation of MVEs and exosome secretion. The siRNA-mediated depletion of multiple subunits of the ESCRT machinery and accessory proteins reveals their respective roles in modulation of the secretion and protein composition of exosomes [17]. Interestingly, simultaneous knockdown of key components of the ESCRT machinery still allows the formation of a few enlarged ILVs devoid of EGFR in distinct MVEs upon EGF stimulation [18]. As the best-characterized mechanism, the ESCRT machinery harbors four biochemically distinct protein complexes (ESCRT-0, −I, −II, and -III), which together with accessory proteins perform in a stepwise manner to sequester MVE cargoes in endosomes and induce the inward budding of endosomal membrane to form ILVs [19]. More specifically, early-acting ESCRT components segregate ubiquitinated membrane cargoes and probably initiate membrane bending on discrete endosomal microdomains, subsequently recruiting ESCRT-III; ESCRT-III subunits, together with the ATPase VPS4, further induce the budding and fission of the microdomains away from the cytosol [20].

In addition to ubiquitin-dependent endosomal sorting, syndecan–syntenin-ALIX axis modulates loading of ILVs with specific cargoes and production of the distinct subpopulations of exosomes. Cytosolic adaptor syntenin connects transmembrane protein syndecan to ESCRT accessory component ALIX, which could bridge the gap between syndecan and the ESCRT-III subunits, thereby facilitating the exosome secretion as well as exosomal release of syntenin, cleaved syndecans and syndecan cargoes (for example, FGF-FGFR complexes) [16]. Syntenin also interacts with CD63 on endosomal membranes and significantly influences exosomal release of CD63. Tetraspanin-6 is another syntenin-interacting membrane protein, and their interaction promotes exosome secretion [21]. The regulators of syndecan-syntenin-ALIX pathway include heparanase, small GTPase ADP ribosylation factor 6 (ARF6) and its effector phospholipase D2 (PLD2). Specifically, heparanase could induce efficient clustering of syndecans by trimming their heparan sulfate chains and allow enhanced binding of syntenin to endosomal syndecans, thereby promoting intraluminal budding and syntenin exosome secretion [22, 23].

The first exosome biogenesis pathway independent of ESCRT requires the generation of ceramide on MVEs [24]. With the cone-shaped structure and the self-association capability through hydrogen bonding, ceramide could induce a spontaneous curvature on the membranes and trigger the coalescence of ceramide microdomains into macrodomains or membrane platforms [25, 26]. Moreover, ceramide-dependent endosomal sorting requires activation of inhibitory G protein (Gi)-coupled sphingosine 1-phosphate (S1P) receptors on MVEs. Specifically, S1P, the ceramide metabolite, constitutively activates the Gi-coupled S1P receptors in an autocrine manner, thereby activating the Rho family GTPases Cdc42 and rac1 and forming F-actin networks on MVEs [27, 28]. Certain cargoes, such as proteolipid protein, CD63, CD81 and flotilin-2, are sorted into the ILVs of MVEs in a ceramide-dependent manner.

Proteins of the tetraspanin family have also been shown to mediate endosomal sorting, ILV formation and production of different exosome subpopulations. During melanogenesis, tetraspanin CD63 regulates a balance between the ESCRT-independent and -dependent endosomal sorting of the PMEL luminal domain, and Apolipoprotein E (ApoE) regulates the loading of the PMEL luminal domain into ILVs in the CD63-dependent sorting pathway [29, 30]. Interestingly, CD63 and Hrs mediate competing mechanisms that promote the formation of different sized ILVs [31]. Moreover, tetraspanin CD82 (and likely CD9) could form complexes with E-cadherin at the plasma membrane, which sorts cellular β-catenin to MVEs and exosomes [32].

Collectively, cargo sorting and ILV budding in MVEs are closely-related processes and mediated by both ESCRT-dependent and ESCRT-independent pathway. It is plausible that in these processes, different pathways can function on same or different MVEs, and collaborative pathways can participate in different steps or certain steps. Hence, different MVE or ILV subpopulations could coexist and contain distinct compositions and morphology.

Intracellular fate, transport and extracellular release of MVEs

The matured MVEs are targeted either to lysosomes or autophagosomes for the degradation and recycling of their contents or to plasma membrane for release of ILVs, referred as exosomes (Fig. 1). Although the main fate of MVEs is to fuse with lysosomes, the balance between the degradative and secretory pathways of MVEs could be reversed, resulting in increased exosome secretion. The exosome secretion is considered as a homeostatic response to counteract fluctuant lysosomal or autophagic activity [33,34,35]. Due to the impaired autophagy and lysosomal pathways, MVEs with superfluous or defective cargoes, such as cholesterol and self-aggregating proteins could be rerouted to the plasma membrane for exocytosis [36, 37] Notably, there is another scenario that defected autophagic and lysosomal functions promote loading of degradative cargoes into ILVs and exosomes but do not increase exosome secretion, suggesting the MVE fate remains unchanged [38]. Macroautophagy starts with sequestration of waste or damaged cellular components into autophagosomes, and they can fuse with MVEs to form amphisomes, which can subsequently fuse with lysosomes [39]. Upon inhibition of degradation, amphisomes can reroute to the plasma membrane and extracellularly release their ILVs with autophagy-associated proteins (Fig. 1). Altogether, autophagy-exosome and lysosome-exosome crosstalks can influence the fate of MVEs and their cargoes towards secretion or degradation. The underlying mechanisms of this balance are still under investigation but possibly involve the components of the exosome biogenesis and autophagy pathway. ISGylation of TSG101, one of ESCRT-I components, promotes TSG101 aggregation and degradation by inducing MVE fusion with lysosomes, thereby repressing MVE numbers and exosome secretion [40]. Tetraspanin-6 increases exosome secretion by activating syntenin pathway, which is correlated with impaired autophagosome-lysosomal fusion [21]. Prion protein induces caveolin-1 (CAV-1) internalization and subsequent inhibition on ATG12–ATG5 complex, leading to autophagy inhibition and exosome secretion [35]. Alcohol-induced miR-155 disrupts autophagic and lysosomal functions to enhance exosome secretion by targeting LAMP1 and LAMP2 [34]. SIRT1 promotes ATP6V1A mRNA stability and then disrupts function of V-ATPase proton pump, thereby reducing lysosomal acidification and increasing exosome secretion [41]. PIKfyve depletion, which inhibits PI(3,5)P2 synthesis, reduces autophagic flux and autophagic degradation, which consequently promotes the secretion of exosomes containing autophagy-related proteins [42]. In addition, independent of controlling autophagy–exosome balance, autophagy machineries have recently been reported to modulate MVE formation and targeting. ATG12-ATG3 axis promotes exosome secretion, endolysosomal trafficking and basal autophagy by interacting with Alix and thereby inducing its active ‘open’ conformation [43]. Independent of ATG7 and canonical autophagy, ATG5 decreases acidification of MVEs by removing ATP6V1E1 from V1V0-ATPase, thereby increasing exosome production [44].

As for MVEs destined for secretion or degradation, their intracellular transport, docking and final fusion with intracellular membranes are indispensable and tightly regulated. MVEs are transported along cytoskeleton to the membranes, which is modulated by multiple molecular motors and switches (small GTPases )[45, 46]. Various Rab GTPases, which shift from GDP- to GTP-bound states for activating effectors, participate in modulation of MVE targeting. Rab7-dependent transport of MVEs to lysosomes has been well documented, and exosome secretion could depend on ubiquitylation status of Rab7 and endosomal cholesterol levels, which modulate dynein motor-mediated MVE transport [47,48,49]. Moreover, Rab24 is involved in Rab7-mediated endolysosomal degradation possibly by interacting with Rab7 and its effector RILP [50]. In the transport and docking of MVEs, Rab27a and Rab27b perform different roles at distinct locations possibly by activating their respective effectors Slp4 and Slac2b [15]. Moreover, Rab11 and Rab35-induced releases of exosomes rely on intracellular Ca2+ levels. Munc13–4, as a Rab11a effector, promotes trafficking of Rab11+ endosomes to CD63+ MVEs, in order to increase the size and secretion competence of MVEs [51]. Rab35, mediated by TBC1D10A–C, acts on the plasma membrane for docking or tethering of MVEs [52].

Depending on the cell types and cellular status, MVEs are transported towards the plasma membrane for multidirectional secretion or polarized secretion [53]. The delivery of MVEs to specific membrane locations depends on microtubule and branched actin network, as reported for immune synapse (IS) between immune cells and invadopodia in cancer cells [54, 55]. The oriented transport of MVEs is controlled by positioning of the microtubule-organizing center (MTOC), which is redirected towards IS during antigen presentation. The MTOC reorientation requires phospholipase C-γ-mediated accumulation of diacylglycerol (DAG) at the IS [56]. In lymphocytes, DAG kinase α, transforming DAG into phosphatidic acid, serves as a negative regulator of maturation and polarized traffic of MVEs by reducing PKD1/2, DAG effector, recruitment to and activation at the IS [57, 58]. Within invadopodia, cortactin binds with the Arp2/3 complex to promote actin nucleation, which stabilizes the branched actin and allows more docking sites of MVEs at the plasma membrane [59].

As the final step of exosome release, MVE fusion with the plasma membrane is governed by soluble N-ethylmaleimide-sensitive factor attachment protein receptor (SNARE) proteins and their regulators. Membranes have their distinct set of SNARE proteins, target-membrane SNAREs (t-SNAREs) and vesicle-membrane SNAREs (v-SNAREs) anchored to the acceptor and vesicle membranes respectively, and their pairing and assembly into the SNARE complexes bring the membranes into close proximity, thereby driving membrane-fusion events. As a v-SNARE, VAMP7 is required for exosome secretion, owing to its ability to modulate MVE fusion with the plasma membrane [60]. The exosome secretion of tumor cells has been shown to rely on PKM2-mediated and H1HR-mediated phosphorylation of SNAP23, which could promote the formation of the SNARE complex to facilitate the docking and fusion between MVEs and the plasma membrane [61, 62]. Other SNARE proteins also participate in exosome secretion, such as Ykt6 and VAMP5 [63, 64]. Notably, GTPase RAL-1 regulates not only the formation of MVEs, but also their fusion with the plasma membrane by colocalizing with t-SNARE SYX- 5[65].

Microvesicle biogenesis and release

Compared with exosome biogenesis, microvesicle biogenesis is not fairly well understood. Microvesicle release can be initiated by increased Ca2+ concentration, which results in disruption of membrane asymmetry and actin cytoskeleton rearrangements by mediating Ca2+-dependent enzymes [66]. This enzymatic pathway includes flippases, floppases and scramblases that mediate the translocation of phosphatidylserine from the inner leaflet to the cell surface, as well as calpain and gelsolin that cleave actin filaments and cap** proteins respectively [67]. Moreover, Peptidylarginine deiminases (PADs), Ca2+-dependent enzymes, convert protein-bound arginine to citrulline for deiminating proteins; PADs stimulate microvesiculation through deimination of cytoskeletal actin [68]. Loss of membrane lipid asymmetry imposes local membrane curvature during microvesicle formation, followed by actin-myosin-based contraction that promotes microvesicle fission and release [69]. Of note, phosphatidylserine exposure does not occur in entire microvesicle population, suggesting the involving of other mechanisms in microvesicle budding, including clustering of transmembrane proteins with spontaneous curvature as well as changes in lipid composition (for example, cholesterol and galactosylsphingosine) and related domains [70]. The releasing process of microvesicles requires ATP-dependent actomyosin contractile machinery composed of actin and myosin, which facilities contraction at microvesicle necks [71]. Interestingly, this contractile machinery propels sliding of apical membrane towards the microvillus tip and leads to membrane vesiculation and microvesicle shedding at the tip [72, 73].

The transmembrane protein TMEM16F, which has scramblase activity, induces phosphatidylserine exposure and platelet-derived microvesicle release [74]. The ARF and RHO families of small GTPases are important regulators of actin dynamics and facilitate actin cytoskeleton-based fission of microvesicles in tumor cells. ARF1 modulates the activation of RhoA and RhoC, which leads to myosin light-chain (MLC) phosphorylation and actomyosin contraction [75]. Activation of RhoA, and its downstream effector RHO-associated protein kinase (ROCK), has been shown to activate Lim kinase (LIMK) that can phosphorylate cofilin and inhibit its actin-severing activity [76]. In addition to participating ARF1/Rho/MLC and RhoA/ROCK/LIMK/cofilin signaling, RhoA/ROCK signaling activates extracellular signal-regulated kinase (ERK) and then inhibits myosin light chain phosphatase (MLCP), which inactivates MLC, thereby promoting microvesicle secretion; and this Rho signaling are promoted by ARF6 activation and Rac1 downregulation [77]. Interestingly, activation of ARF6 promotes phosphorylation of MLC by activating phospholipase D (PLD) and then recruiting ERK to the plasma membrane for phosphorylating MLC kinase (MLCK), resulting in microvesicle release; whereas its inactivation induces the opposite effect through PKC-mediated phosphorylation of MLC and then decreased MLC activity [71]. These observations suggest that ARF6 harbors two distinct downstream pathways (RhoA/ERK/MLCP/MLC and PLD/ERK/MLCK/MLC) to promote MLC activity and microvesicle shedding. Apart from the small GTPases, other cytoskeletal regulators are available for microvesicle secretion. DIaPh3 suppresses membrane bleb formation and microvesicle secretion, which is associated with phosphorylation state of cofilin [78]. Activation of protease activated receptor 2 (PAR2) by trypsin induces AKT phosphorylation and then activates Rab5a at the plasma membrane, resulting in polymerization of actin and microvesicle secretion [79]. In a later report, Activation of PAR2 regulates actomyosin rearrangements to enhances microvesicle secretion via three independent pathways, including MAPK/MLCK/MLC, P38/MK2/HSP27 and RhoA/ROCK signaling [80].

Additional cell surface receptors are involved in microvesicle secretion, including G protein-coupled receptor 30 [81], α-2-Macroglobulin receptor [82], transient receptor potential vanilloid type 1 [88] and tissue factor [83]. In addition, microvesicle secretioin is also induced by intratumoral hypoxia, which transcriptionally regulates the expression of the small GTPase Rab22A that colocalizes with budding MVs [118]. Upon loss of Ago2, highly secreted miR-451 is the most affected, probably because its Dicer-independent maturation requires only Ago2-mediated cleavage. Ago2 knockdown also decreases the exosomal content of small RNAs, indicating that Ago2 may serve as an important transferring machinery for EV-miRNAs [119]. This notion was further strengthened by the fact that Ago2 can be sorted to exosomes and control the sorting of specific miRNAs (for example, let-7a) into exosomes. However, the presence of Agos in EVs is still questionable as the regulation of Agos on endosomal membranes may differ according to cell type or cellular state [112]. GW182 knockdown decreases the release of exosomal miRNA, such as miR-146a and miR-155, probably by making Ago-loaded miRNAs more vulnerable to ribonucleas e[120]. Interestingly, as a negative regulator of miRNA function, HuR can replace Ago2 from target mRNAs and capture miR-122 from Ago2, and ubiquitination of HuR on MVEs promotes miR-122 unloading and then the extracellular export of miR-122 [121].

Fig. 2
figure 2

RNA incorporation into EVs. Various RBPs and membrane-associated proteins are required for the different steps of EV-RNA sorting. First, intracellular RNAs can interact with RBPs or motif-specific RBPs, which may prevent RNAs from degradation. Second, RNA-loaded RBPs undergoing post-translational modification can be recruited to the sites of EV budding via binding to membrane-associated proteins; otherwise, RNAs and RNA-loaded RBPs are incorporated passively into EVs. Third, upon reaching the budding membrane, the RBPs can be co-sorted with loaded RNAs into EVs or unload RNAs into EVs. In addition, 3′ end RNA tailing, such as adenylation and uridylation, controls RNA distribution between cells and EVs. MVE (multivesicular endosome), EV (extracellular vesicle), RBP (RNA binding protein).

Full size image

In addition, ESCRT proteins have been shown to regulate miRNA sorting into EVs. Alix knockdown reduces loading of secreted miRNAs into EVs but not the release of EVs probably by interacting with Ago2 [122]. However, the other study reported that Alix knockdown does not influence the extracellular export of miR-146a [123]. Another ESCRT protein, Vps4A, mediates the release of oncogenic miRNAs in exosomes [124].

Additional RBPs have been shown to be involved in the sorting of specific miRNAs into EVs. Upon nuclear export and dissociated from Ran-GTP, Exportin-5 dictates pre-miRNA complex by interacting ARF6-GTP-GRP1 complex that transfers miRNAs to microvesicle biogenesis sites [125]. MVP presents in exosomes and promotes the sorting of miR-193a into exosomes via binding to miR-193a [126]. YBX1 interacts with miR-223 and promotes miR-223 sorting into exosomes, and it is also involved in TICAM-1-mediated sorting of miR-21 into EVs [127, 128]. MEX3C can be targeted to endolysosomal pathway through interaction with AP-2 complex and associated with an RBP of miR-451a, thereby allowing the sorting of miR-451a into exosomes [129].

Certain RBPs have been proposed to perform miRNAs sorting by recognizing specific RNA motifs. hnRNPA2B1 controls exosomal sorting of miRNAs with the GGAG motif, and sumoylation of hnRNPA2B1 promotes its binding to the miRNAs and localization into exosomes [130]. Cav-1, a membrane-bound protein, forms complex with hnRNPA2B1 and induces hnRNPA2B1 O-GlcNAcylation via its tyrosine-14 phosphorylation, thereby directing hnRNPA2B1-bound miR-17/93 into microvesicles; O-GlcNAcylation of hnRNPA2B1 enhances its binding to specific miRNAs and incorporation into microvesicles [131]. Similarly, SYNCRIP displays the GGCU-motif-specific exosomal sorting capacity for miRNAs [132]. By contrast, ANXA2 mediates the sorting of miRNAs into EVs in a sequence-independent manner and binds EV miRNAs in the presence of Ca2+ [133].

In addition to RBPs, sphingomyelinase pathway has been shown to involve in exosomal export of miRNAs. Inhibition of neutral sphingomyelinase 2, and therefore the ceramide generation, prevents the sorting of multiple miRNAs into EVs, such as miR-451a, miR-122 and miR-146a [121, 123, 129]. As another example, inhibition of sphingosine kinase 2, and therefore Sphyngosine-1-phosphate generation, reduces exosomal loading of miRNA-21 [134].

Selection of miRNAs for exosomal release is also tightly associated with their 3′ end post-transcriptional modifications. miRNAs distribution relies on 3′ end uridylation and adenylation, which promotes miRNAs exosomal release and cellular retention respectively [113]. For example, in cancer cells, miR-2909 is targeted to or excluded from exosomes in a manner dependent on its 3′-end adenylation to uridylation ratio, which seems to be linked to different distribution of adenosine kinase between cells and exosomes [135].

Although it is still lack of evidence that how lncRNAs are targeted to EV production site, they are likely to share common cis-acting signals and sorting machineries (trans-acting proteins) with mRNAs. mRNAs have been shown to differentially sorted to EVs mostly depending on their specific sequences and secondary structures in the 3′-untranslated regions. The presence of three motifs (ACCAGCCU, CAGUGAGC and UAAUCCCA) in mRNAs and lncRNAs is associated with their exosomal secretion, and YBX1 could be involved in the sorting process by specifically binding with these motifs [136,137,138]. Interestingly, miRNAs could also regulate mRNA targeting into EVs by specifically binding with zipcode RNA sequence motifs. miR-1289 directly binds with the inserted zipcode on EGFP mRNA and then enhances the efficiency of zipcode-mediated EGFP mRNA sorting into microvesicles [139]. The potential roles of miRNAs in transferring mRNAs into EVs are also illustrated by the fact that miRNA binding sites are found in the predicted motifs enriched in EV mRNAs [140].

RNA content of EVs varies depending on the EV subpopulation, cell type and the physiological or pathological state of producing cells as well as their received stimuli. Origin of EV-RNA diversity can be attributed to cellular RNA profile and different RNA sorting and protection mechanisms. EV-RNA loading can occur by either active or passive mechanisms and largely depend on RBPs and their partners as well as RNA motifs and modifications, with combined effect on stabilization and/or subcellular localization of EV-RNAs. Chaperone RBPs can be co-sorted with intracellular RNAs and present on exosomes, whereas shuttling RBPs can transfer RNAs to membrane-bound RBPs in MVEs and exclude themselves from exosomes. Post-translational modification of RBPs is associated with their affinity for MVEs and RNAs, suggesting an additional layer of regulation of exosomal sorting. RBP-mediated RNA incorporation into EVs has been shown to depend on ceramide generation, indicating RBPs are likely to be recruited to the ceramide-enriched microdomains that will bud in selective RNA-loading processes. 3′-end of RNA sequence appears to be a primary site that contains RNA sorting signals for EV secretion. Specific motifs and structures of RNAs play important roles in EV-RNA secretion by mediating RNA-RBP and RNA-RNA interaction. Nontemplated nucleotide additions have an impact on RNA distribution between EVs and cells probably by controlling RNA metabolism.

Biological roles of RNA-containing extracellular vesicles in the TME and premetastatic niches

The deregulation of EV-RNAs among different cancer types and their cell-type-specific functions have recently started to be uncovered. These EV-RNAs carry genetic messages of donor cells to neighboring or distant stromal and tumor cells, and contribute, at least in part, to bidirectional communication within the TME. Once reaching the recipient cells, EV-RNAs can trigger molecular and phenotypic reprogramming of recipient cells. The underlying mechanisms of EV-RNAs affecting cellular functions are different according to the type of RNA. In recipient cells, mRNAs delivered by EVs can be translated into functional proteins, whereas ncRNAs delivered by EVs can engage complex networks of ncRNA interactions and serve as important regulators of gene expression in cellular processes [141, 142]. EV-RNAs have been considered as oncogenic drivers or tumor suppressors in various types of cancers. The proliferation, apoptosis, migration, invasion, dormancy, stemness and therapy resistance of cancer cells are actively mediated by EV-RNAs from their malignant counterparts, educated noncancerous cells and normal cells (Figs. 3 and 4). EV-RNAs from cancer cells also continuously reprogram stromal cells to support tumor development and progression, establishing a feed-forward or -back loop of intercellular communication (Figs. 3 and 4). The reprogramming of stromal cells and immune cells results in stromal activation, vascular restructure and immune evasion, further driving tumor growth, invasion, metastasis and therapy resistance.

Fig. 3
figure 3

EV-RNA mediated crosstalk within tumors and between tumors and stroma modulating malignant behaviors of cancer cells. Cancer initiation, development and progression are attributed to sophisticated and multidirectional communication between various cells. Tumor-derived EV-RNAs can elicit oncogenic, prometastatic, proangiogenic and differentiated phenotypes of stromal cells in the tumor microenvironment or prometastatic niches. Tumor-derived EV-RNAs also drive normal and tumor cell subpopulations towards malignant phenotypes. EV-RNAs from cancer-reprogrammed stromal or normal cells also contribute to malignant behaviors of cancer cells, thereby affecting the growth, migration, invasion and survival of primary and metastatic cancer cells. Of note, EV-RNAs from normal cell can also restrain the malignant behaviors of cancer cells.

Full size image
Fig. 4
figure 4

EV-RNA mediated crosstalk between cancer cells and immune cells and within immune cells modulating malignant behaviors of cancer cells. Tumor-derived EV-RNAs can contribute to the immunosuppressive and decreased anti-tumoral activities of various immune cells and induce immunoinhibitory phenotype of CAFs and normal cells. EV-RNA-mediated communication between immunes also leads to cancer progression.

Full size image

Regulation of malignant phenotypes of cancer cells by tumor and stromal EV-RNAs

During cancer development, there are cell competition between cancer cells and neighboring normal cells [9]. As a homeostatic mechanism, abundant noncancerous cells can release tumor-suppressive miRNAs to inhibit malignant phenotypes of adjacent cancer cells [12]. EV-miR-143 from normal epithelial prostate cells suppresses the proliferation of adjacent prostate cancer cells [143]. EV-miR-145 from tumor-associated stroma cells impairs the viability and induces the apoptosis of adjacent pancreatic ductal adenocarcinoma cells [144]. EV-miRNAs from liver stem cells inhibit the proliferation and promote the apoptosis of hepatocellular carcinoma (HCC) cells in vitro and in vivo [145]. EV-miR-145 from adipose tissue-derived mesenchymal stem cells (AMSCs) inhibits the proliferation and promotes the apoptosis of prostate cancer cells [146]. EV-lncRNA-PTENP1 from normal cells induces the apoptosis of bladder cancer cells and inhibits their proliferation, migration and invasion by targeting miR-17, thereby reducing tumor growth in vivo [147]. Of note, EV-RNAs from normal cells also contribute to the malignant behaviors of cancer cells. EV-circRNA-DB from adipocytes promotes the proliferation and migration of HCC cells and reduces their DNA damage by targeting miR-34a, resulting in tumor growth and metastasis in vivo [148]. Mesenchymal stem cells (MSCs) are multi-potent stromal cells derived from bone marrow, adipose tissue, umbilical cord or placental, with potential to act as protumoral components of the TME under both normoxic and hypoxic states. EV-miR-410 from umbilical cord MSCs promotes the proliferation and inhibits the apoptosis of lung adenocarcinoma cells by targeting PTEN in vitro and in vivo [149]. EV-miRNAs, including miR-193a-3p, miR-210-3p and miR-5100, from hypoxic bone marrow mesenchymal stem cells (BMSCs) promote the mesenchymal-to-epithelial transition (EMT), migration and invasion of lung cancer cells by activating STAT3 pathway [150]. EV-miR-21-5p from hypoxic BMSCs increases the proliferation, survival, migration and invasion of lung cancer cells as well as M2 macrophage polarization, a protumorigenic shift, by targeting PTEN, PDCD4 and RECK, leading to tumor growth and intratumoral angiogenesis in vivo [151]. EV-miR-142-3p from BMSCs promotes the stemness, doxorubicin resistance, invasion and adhesion of colon cancer cells by targeting Numb [152]. EV-miR-23b from BMSCs reduces the proliferation, sensitivity to docetaxel and CD44, a stem cell marker, of bone marrow–metastatic breast cancer (BC) cells by targeting MARCKS, thereby contributing to dormancy of BC stem cells in metastatic niches [153].

The EV-mediated transfer of tumor-suppressive miRNAs to cancer cells indicates that manipulation of EV-RNAs could have therapeutic benefits in cancers. For example, stellate cell-derived EVs loaded with miR-335-5p inhibit the proliferation and invasion of HCC cells and HCC tumor growth in vivo [154]. In another case, EVs from miR-195-transfected fibroblasts induce tumor shrinkage and improve the survival in a rat cholangiocarcinoma model [155]. In addition, EVs from miR-122-transfected AMSCs sensitize HCC cells to sorafenib both in vitro and in vivo [156]. EVs from miR-126-3p-transfected BMSCs suppress the proliferation, migration, and invasion and increase the apoptosis of pancreatic cancer cells by targeting ADAM9, reducing tumor growth in vivo [254]. In this context, the EVs concentration is an important factor for the dose-dependent effect of RNAs on target cells. It is hypothesized that diverse RNAs could work together to simultaneously mediate cancer hallmarks by jointly regulating a single pathway or mRNAs [255]. Therefore, corporation between EV-RNAs may be another important factor for augmenting their ability affecting tumorigenesis. The complementary roles of EV-RNAs in cancer biology remain largely unexplored and require further research. Of note, most published papers of EV-RNAs have focused on RNA content and functions of exosomes rather than microvesicles in cancers. Different types of EVs possess distinct biological properties and RNA content, which could affect their distribution and functions in the TME. Therefore, characterization of RNA content, delivery and functions of different EV subpopulations contributes to expand our knowledge of EV-RNA-mediated cell-cell communication in cancer biology.