Epithelial cells are initially attached to their neighboring cells through cell-cell adhesion complexes mediated by E-cadherin (19,20). EMT begins with the dissolution of epithelial cell-cell contacts essential for basement membrane integrity, such as tight junctions (at the apical surface of cells), adherens junctions (basal to the apical tight junctions), desmosomes (anchored to the intermediate filaments in the cytoskeleton) and gap junctions (channels between cells for passage of ions and small molecules), resulting in the downregulation of their associated proteins, claudins and occludins, cadherins, desmoplakin and plakophilin and connexins, respectively (21,22). The loss of E-cadherin leads to the translocation of β-catenin to the nucleus, where it can contribute to EMT through wingless-type MMTV integration site family (Wnt) signaling (23,24).

The disintegration of cell-cell junctions is also accompanied by the disruption of cell polarity complexes, consequently resulting in epithelial cells losing their apical-basal polarity and acquiring a front-rear polarity (25). In mammals, three protein complexes establish apical-basal cell polarity and membrane identity through their mutually antagonistic interactions (26). Partitioning-defective (PAR) complex [PAR6, PAR3 and atypical protein kinase C (aPKC)] (27) and Crumbs (CRB) complex [CRB, protein associated with Lin-7 1 (Pals 1) and Pals 1-associated tight-junction protein (PATJ)] (28) localize at the apical domain. Here, the PAR complex, which is required for tight junction formation (29), is stabilized by the CRB complex. The Scribble (Scrib) complex [Scrib, Disc large (dlg) and lethal giant larvae (lgl)] localize at the basolateral domain, opposite the PAR complex (30). During EMT, inducers such as Zinc finger E-box-binding homeobox (ZEB) and Zinc finger protein SNAI1 (commonly known as Snail) transcription factors disrupt the assembly and suppress the expression of these polarity complexes (26). This is enhanced by the downregulation of E-cadherin, which prevents the recruitment of Scrib complexes to the basolateral membrane (31). The overall loss of contact with the basement membrane reduces epithelial cell adhesion to eventually facilitate movement.

Cells reorganize their cytoskeletal architecture to enable migration. Peripheral F-actin fibers are replaced by stress fibers, and extracellular matrix (ECM) adhesion molecules such as integrins, paxillin, and focal adhesion kinase localize at the tips (32-34). Intermediate filaments, such as cytokeratins are replaced by vimentin, changing the cellular morphology from cuboidal or columnar to fibroblastic or spindle-shaped (35,36). The cell membrane projections [lamellipodia (thin sheet-like region at the leading edge of migrating cells) and filopodia (spike-like extensions at the edge of lamellipodia)] permit directional movement of the cell across its surroundings. Invadopodia, together with the increased expression of matrix metalloproteinases (MMP-1, -2, -3, -7, -9, -13 and -14), degrade the ECM, allowing the cells to detach from each other, penetrate the basement membrane, and invade the stroma and surrounding tissues (25,32,37).

Once mesenchymal cells reach the metastatic site, they may undergo EMT reversal or MET. Cells revert to an epithelial state, resulting in secondary tumors that resemble the histopathological phenotype of the primary tumor (11). MET may occur in the local environment of the distant organ, due to the absence of heterotypic signals that induced EMT at the primary tumor site (11). Additionally, tumor cells would need to induce signaling pathways to resume proliferation during colonization of the foreign environment (38,39).

EMT is driven by a marked change in transcriptional programming. The modulation of gene expression during EMT involves several signal transduction pathways and transcription factors that reprogram the cell towards a mesenchymal phenotype. The changes in gene expression are responsible for phenotypic manifestations, such as the loss of adhesion and polarity, and an increase in migration and invasiveness.

Several transcription factors that control EMT are dysregulated or aberrantly expressed in CRC. These include SNAI1/SNAI2 (Snail and Slug), ZEB1/ZEB2 (40,41) in the ZEB family, twist family basic helix-loop-helix transcription factors (TWIST) Twist1/Twist2 (42,43) and the forkhead box (FOX) family (44-46) of transcription factors. SNAIL and SLUG are responsible for the repression of E-cadherin (47,48). Slug is also required for Twist1-mediated EMT (48). Twist1 and Zeb1 are also known to drive the loss of E-cadherin and a pro-metastatic cellular phenotype of enhanced invasion (49). These transcription factors have been shown to be associated with cell migration, tumor progression and metastatic potential in CRC (25,50,51).

Multiple signal transduction pathways have been implicated as modulators of the EMT transcriptional program. Transforming growth factor β (TGF-β) is one of the key inducers of EMT in CRC, as well as other types of cancer. Mothers against decapentaplegic homolog (SMAD) family member 4 (SMAD4) is a pro-epithelial factor whose loss leads to heightened EMT through aberrant signal transducer and activator of transcription 3 (STAT3) activation and increased signaling via the TGF-β/bone morphogenetic protein (BMP) pathways (52). Apart from the SMAD pathways, TGF-β also signals via SMAD-independent pathways, such as the Ras/mitogen-activated protein kinase (MAPK) and phosphoinositide 3-kinase (PI3K)/Akt pathways (53,54).

The Wnt/β-catenin pathway is another crucial driver of EMT in CRC. The canonical Wnt pathway leads to SNAIL upregulation and SLUG stabilization. Consequently, Wnt3a expression is associated with mesenchymal markers in clinical samples, and Wnt inhibition can partially reverse EMT expression. In addition, the non-canonical Wnt pathway has also been implicated in metastatic CRC (55).

Transcriptional reprogramming during EMT drives the characteristic switch towards mesenchymal marker expression. The common markers of EMT are typically proteins involved in the architecture of cell junctions and the cytoskeleton, controlling cell morphology and interactions with neighboring cells, as well as the extracellular environment, key physical features that are altered in EMT. Thus, the EMT transcriptional program capacitates the cancer cell for metastatic spread by altering the expression of key structural proteins to facilitate pro-metastatic characteristics such as loss of adhesion, cytoskeletal remodeling, migration and invasion.

One hallmark of EMT is the cadherin switch from E-cadherin to N-cadherin. The epithelial marker, E-cadherin, is a cell-surface protein that mediates cell-cell adhesion, and also binds β-catenin at its cytoplasmic domain. The downregulation of E-cadherin results in the loss of adhesion, an important phenotype for cells progressing towards metastasis, and also frees β-catenin for activation of its signaling cascade. On the other hand, N-cadherin, which is upregulated in mesenchymal cells, facilitates dynamic adhesion contacts crucial for migration, and confers an affinity for other N-cadherin-expressing mesenchymal cells. Thus, the E-to-N-cadherin switch results in the loss of adhesion and increased migration (25).

Claudins and occludins are crucial cell-cell adhesion proteins found in tight junctions, and are subsequently downregulated in EMT to dissolve epithelial apical-basal polarity (56). Conversely, the N-cadherin ligand, neural cell adhesion molecule (NCAM), is upregulated in EMT, modulating the activity of N-cadherin-associated receptor tyrosine kinases, such as Fyn to facilitate focal adhesions (57).

As regards cytoskeletal proteins, EMT is marked by the downregulation of cytokeratin and the upregulation of vimentin, which are both components of intermediate filaments (IFs). The altered composition of IFs facilitates differential trafficking of organelles and proteins, as well as enhanced cellular motility (56). Finally, cell-matrix interactions are also modulated via the altered regulation of integrins, upregulation of fibronectin, and increased expression and secretion of matrix metalloproteinases MMP2 and MMP9 (25).

Similar to normal stem cells, CSCs also have the capacity for self-renewal and differentiation, and can initiate new tumors, such as in seeding metastatic lesions, or in cancer recurrence following therapy. CSCs are distinguishable by specific expression marker profiles, and can apparently arise from the acquisition of oncogenic alterations in normal stem cells, or from the dedifferentiation of differentiated cancer cells (58). In fact, pro-stemness factors are often proto-oncogenic molecules that are dysregulated in cancer. Thus, CSCs are often viewed as an important component of emerging therapeutic resistance and subsequent progression and metastasis, as CSCs that escape treatment are able to expand and initiate new lesions at a later stage.

EMT has also been implicated in the emergence and maintenance of CSCs. Cancer cells that have undergone EMT often exhibit a CSC phenotype, and conversely, CSCs often express EMT markers, signifying a mesenchymal shift. The close association between EMT and CSCs underscores the important role of both in the overall process of metastatic spread. TGF-β/Smad signaling is a key link, as it modulates both EMT and stemness phenotypes (59).

A summary of the overall process of EMT in the context of metastasis is illustrated in Fig. 1. Pro-EMT signaling, such as through the TGF-β or Wnt/β-catenin pathways, drives transcriptional reprogramming facilitated by EMT transcription factors which include SNAIL, ZEB and TWIST. The cancer cell undergoes changes in morphology towards a more motile mesenchymal phenotype, accompanied by a switch in expression from epithelial to mesenchymal markers. With enhanced ECM degradation, the cell is free to intravasate into the circulation and extravasate at a distant metastatic site. Once there, the cell reverts to an epithelial-like phenotype. Angiogenic induction and the action of stromal cells, such as tumor-associated macrophages (TAMs) or cancer-associated fibroblasts (CAFs) further modify the niche to favor metastatic growth.

Tumors are known to have a predisposition to metastasize to specific organs. This phenomenon, however, remained poorly understood for almost a century. The intrinsic properties of cancer cells, including genes and pathways implicated in the colonization of new metastatic niches, were invoked and constituted the predominant explanation for the phenomenon known as metastatic organotropism (60-62). Indeed, EMT itself has been shown to mediate metastatic colonization and metastatic organotropism both directly and indirectly. In pancreatic ductal adenocarcinoma (PDAC) models, the loss of p120ctn, binding partner and stabilizer of E-cadherin at adherens junctions, has been shown to cause a shift in the metastatic load from the liver to the lung. Furthermore, this could be reversed by the transfection of PDAC cells with p120ctn isoform 1A (63). EMT has also been shown to affect organ-specific metastasis by influencing metabolic reprogramming (64,65) and by regulating the tumor immune microenvironment (66). Whether these can be extrapolated to other epithelial tumors, such as CRC warrants further investigation.

While events that render the pre-metastatic niche favorable for dissemination and the growth of tumor cells upon arrival were being characterized, among these being angiogenesis and immunosuppression (67-69), the question of what dictates metastasis to specific organs remained unanswered. The discovery that exosomal integrins in tumor-released exosomes directed organ-specific colonization helped explain the induction of the pre-metastatic niche (7). Together with integrins, connexins, proteins that function as an integral component of channels at gap junctions, were found to be embedded in exosomal membranes (70). Furthermore, resident stromal cells in metastatic target organs have been shown to internalize these exosomes and their cargo, further influencing expression of genes implicated in the preparation of the pre-metastatic niche (7).

Exosomes can be differentiated from other extracellular vesicles based on their mode of biogenesis: microvesicles (100 to 1,000 nm) originate from the cell surface and are formed through the direct outward budding of the plasma membrane (79), while apoptotic bodies (800 to 5,000 nm) are generated from the outward blebbing and fragmentation of the cell membrane as cells undergo apoptosis (80). Exosomes (40-120 nm in diameter) are initially formed as intraluminal vesicles (ILVs) within the endosomal system. In this pathway, the invagination of early endosomes during maturation sequesters cytoplasmic genetic material, such as RNA, DNA, lipids and proteins into ILVs. The late endosomes, now referred to as multivesicular bodies (MVBs), are either targeted to the lysosomes for degradation, serve as temporary storage compartments, or fuse with the plasma membrane, releasing ILVs into the extracellular space as exosomes (81-83).

The formation of intraluminal vesicles occurs via multiple mechanisms. The most well-described pathway is dependent on the endosomal sorting complex required for transport (ESCRT) budding machinery, where ESCRT-0 recognizes and forms domains of ubiquitylated proteins, which are later confined when the endosomal membrane is deformed by the ESCRT-I and ESCRT-II complexes. ESCRT-III then cleaves the bud neck to form intraluminal vesicles (84). On the other hand, the syndecan-syntenin-ALG-2-interacting protein X (ALIX) pathway recognizes any material bound to the heparan sulfate of syndecan. Syndecans are transmembrane proteins whose cytosolic domain can bind to and recruit syntenin, a soluble cytoplasmic protein, to the plasma membrane. Syntenin also binds to ALIX, an accessory protein to the ESCRT-III, in turn linking the syndecans to the ESCRT complexes for ILV formation (85,86). Alternatively, membrane budding can occur through an ESCRT-independent pathway involving the sphingolipid ceramide. Ceramide induces the coalescence of lipid microdomains found in the endosomal membrane, and its cone-shaped structure causes spontaneous negative curvature between the membrane leaflets, resulting in the inward budding of endosomes to produce ILVs (87). A recent study also reported another ESCRT-independent pathway using Rab31, which prevents the fusion of the MVB to the lysosome to instead promote ILV secretion (88). In these pathways, Rab guanosine triphosphatases (Rab GTPases) mediate the fusion of MVBs with the plasma membrane. Rab27a and Rab27b influence the localization (89), whereas Rab35 regulates the docking and tethering (90) of the MVBs at the plasma membrane for subsequent secretion of the intraluminal vesicles into the extracellular fluid as exosomes.

The major cargo residing within exosomes are proteins, lipids, RNA and a minimal amount of DNA. The contents of exosomes are highly heterogeneous, depending on the tissue, cell type and physiological or pathological context. Notably, exosome profiles do not fully reflect that of parent cells, signifying a selective mechanism for sorting cellular genetic material into exosomes (9,91,92). There is also evidence to indicate that some RNA transcripts can be found exclusively in exosomes and not in the cells of origin (9,92) which suggests that certain genes are destined for export and communication with other cells (93).

Deep sequencing has revealed a variety of exosomal RNA cargo, including mRNAs, miRNAs, long non-coding RNAs (lncRNAs), ribosomal RNAs, transfer RNAs, piwi-interacting RNAs, small nuclear RNAs, small nucleolar RNAs and even circular RNAs, each present at varying levels of abundance (74,91,94-96). Among these, miRNAs are generally the most predominant RNAs encapsulated in exosomes (91,96). Several mechanisms regulating the selective export of miRNAs into exosomes have been proposed, from intrinsic sequences present in the 3′ end of miRNAs to cellular lipids and proteins that guide their incorporation into exosomes. First, heterogeneous nuclear ribonucleoproteins (hnRNPs) directly bind to 4-bp sequence motifs in the 3′ end of miRNA sequences to package them into exosomes. Sumoylated hnRNP A2/B1 can recognize the RNA motif GGAG in miR-198 and miR-601 (97), while hnRNP-Q can recognize the GGCU motif in miR-3470a, miR-194-2-3p, miR-6981-5p, miR-690 and miR-365-2-5p (98). Other members of the hnRNP family, such as hnRNPA1 and hnRNPC (97) and miRNAs enriched with the GUUG motif (99) were also sorted into exosomes, although their cognate sequence motifs and RNA binding proteins, respectively, have not yet been identified. Second, 3′ end post-transcriptional modifications can also influence cargo loading. The RNA sequencing of B cells and urine samples has revealed that adenylated miRNAs are retained in the cells, while those that are uridylated are preferentially sorted into exosomes (100). Third, membrane lipids involved in EV biogenesis also control the loading of miRNAs into exosomes, indicative of the ceramide-dependent secretory pathway independent of the ESCRT machinery (101). The overexpression of neural sphingomyelinase 2 (nSMase2), which regulates ceramide biosynthesis, has been shown to increase the quantity of miRNA exported into exosomes, whereas nSMase2 inhibition reduces the quantity of exosomal miRNAs (102). Similarly, the inhibition of sphingomyelin phosphodiesterase 3 (SMPD3), which catalyzes the hydrolysis of sphingomyelin to ceramide, has been shown to result in a concentration-dependent increase in cellular miR-638 levels and a decrease in exosomal miR-638 levels, using the SW480 human colorectal and HuH-7 human hepatocellular cancer cell line (103). In patients with CRC, the expression levels of miR-638 have also been found to be downregulated in serum exosomes and to be associated with a poor overall and disease-free survival (104). Fourth, Argonaute 2 (Ago2), a key component of the RNA-induced silencing complex (RISC) appears to play a role in exosomal miRNA secretion. The knockdown of Ago2 has been shown to decrease the expression of miRNAs (miR-451, miR-150 and miR-142-3p) that are most commonly exported from different cell types (105). In this regard, KRAS mutations have also been implicated in miRNA export, negatively regulating the secretion of miRNAs into exosomes, since downstream MEK/ERK signaling inhibits the association of Ago2 with multivesicular bodies (106). Notably, all essential RISC components, such as Dicer, transactivation response RNA-binding protein (TRBP) and Ago2 are present in cancer-derived exosomes and this enables processing of pre-miRNAs into mature miRNAs within the nanovesicles, unlike exosomes derived from normal cells (107).

Apart from cell state, exosome secretion is influenced by several factors, such as cellular stress, heat, ischemia, pH, the loss of cellular attachment and the accumulation of intracellular calcium (108). Of note, low pH levels and hypoxia are common features of solid tumors. The low pH of the tumor microenvironment results in increased exosome release and subsequent uptake by recipient cells, which explains the abundance and effects of exosomes in cancer (109). Moreover, hypoxia increases the release of exosomes, which can potentially facilitate angiogenesis and metastasis, further sustaining the growth and survival of cancer cells (110).

Exosome internalization by recipient cells highlights the critical role of exosomes in cell-to-cell crosstalk, with the unique advantage of targeting specific locations compared to cytokines and hormones in the systemic circulation. Exosomes can transfer genetic information to neighboring or distant cells through three principal mechanisms: i) Direct fusion of the exosomal lipid membrane with the cellular membrane of recipient cells, releasing the exosome cargo into the cytosol (111); ii) proteins on the exosomal membrane serve as ligands for receptors on the surface of recipient cells; and iii) endocytosis either mediated by the clathrin protein, which is associated with engulfment of partner receptors, or micropinocytosis, which is associated with membrane ruffles that are induced by receptor tyrosine kinases (112). The specificity of these receptor-ligand interactions suggests that exosomes are targeted to particular cells (7).

Exosomes were once considered to serve only as a means of cellular waste disposal when they were first described for removal of plasma membrane proteins during reticulocyte maturation (81,113). Later on, Raposo et al (114) isolated exosomes from B lymphocytes and demonstrated their involvement in antigen presentation, capable of inducing an immune response. Exosomal messenger RNAs were then found to be internalized and translated into functional proteins (9,92), while exosomal miRNAs and lncRNAs can regulate the translation of target mRNAs (75,107,115,116) in recipient cells, concretely establishing a role in intercellular communication. This links exosomes to several biological processes as well as disease pathogenesis.

In the context of cancer, cumulative evidence suggests that exosomes can promote tumorigenesis through the horizontal transfer of oncogenic material to recipient cells. Likewise, cancer cells can utilize exosomes to discard tumor-suppressive genetic material not beneficial for tumor growth so as to increase their own oncogenicity (117). For instance, in CRC, miR-100 is a tumor suppressor that inhibits cellular migration and invasion by targeting Leucine-rich repeat-containing G protein-coupled receptor 5 (Lgr5) (118). Mutant KRAS CRC cells have been reported to secrete exosomes enriched in miR-100 as a strategy to sustain low intracellular levels (117). Tumor-derived exosomes have been shown to induce (119) or suppress (120) the immune response, promote the formation of a pre-metastatic tumor niche (7,121), regulate angiogenesis (122), enhance migration (76) and cell proliferation (77), and induce epithelial to mesenchymal transition (123), among others (Fig. 2).

Exosomes can also promote tumor resistance by encapsulating drugs and their metabolites into exosomes for export, as a drug efflux mechanism (124). Similarly, drug-resistant cancer cells can induce chemoresistance in other cancer cells through exosome-mediated transfer of efflux transporters (78).

CAFs are essential components of the TME, with roles in matrix deposition and remodeling (125). Up to 80% of stromal fibroblasts in a tumor are believed to acquire an activated phenotype and become CAFs (126). These cells can modulate tumor progression by releasing cytokines such as TGF-β, tumor necrosis factor α (TNF-α), and interleukin (IL)-6 and IL-8 (127). Tumor cells and CAFs in the TME maintain extensive reciprocal signaling interactions (125). Recently, primary CRC cells, as well as HCT116 cells were shown to release exosomes containing miR-10b which are taken up by surrounding fibroblasts. miR-10b was found to increase TGF-β and α-smooth muscle actin (α-SMA) through the inhibition of PI3K activity, which is indicative of fibroblast activation into pro-tumorigenic CAFs (127). In a similar manner, CAFs have been shown to deliver exosomal miRNA to tumor cells. In patients with CRC, CAF-derived exosomes have been found to be enriched with miR-92a, which is delivered to CRC cells (128). The uptake of exosomal miR-92a enhances EMT and metastatic capability by downregulating F-Box and WD repeat domain containing 7 (FBXW7) and modulator of apoptosis 1 (MOAP1), and by activating the Wnt/β-catenin pathway (128). Of note, cellular miR-92a has also been characterized as an oncogenic miRNA promoting migration, F-actin remodeling and EMT marker expression in CRC cells via the downregulation of neurofibromin (NF2)/Merlin, a tumor suppressor gene that controls contact-dependent inhibition, adhesion and migration (129). Other miR-92a targets include Dickkopf-3 (Dkk-3) and claudin-11, resulting in enhanced angiogenesis and disruption of tight junctions, respectively (130). In patients with CRC, the loss of claudin-11 has been shown to be associated with increased metastasis and a poor prognosis (131). In the CRC microenvironment, CAFs were previously shown to deliver miR-21 to cancer cells in exosomes. miR-21 was identified as the most abundant and most enriched miRNA in an exosomal cancer-associated fibroblast signature that also includes miRNAs 329, 181a, 199b, 382 and 215. Orthotopic xenografts derived from fibroblasts overexpressing miR-21 exhibited an increased liver metastases relative to the controls (132).

The exosomal cargo of CAFs in CRC may also include the lncRNA H19. H19 can attenuate the inhibitory effects of miR-141 and can then activate the Wnt/β-catenin pathway to promote stemness (133). Thus, exosomal delivery of miRNAs and/or lncRNAs from CAFs may drive EMT through redundant mechanisms. This is in addition to other forms of stromal-tumor crosstalk that may exist in parallel, notably the secretion of CAF-derived soluble factors such as TGF-β and juxtacrine signaling (134,135). Increasing evidence suggests that CAFs are heterogenous and may assume different functional roles (136,137). It is highly likely that different groups of CAFs are involved in exosomal miRNA exchange with tumor cells.

In patients with metastatic CRC, serum exosomal miR-106b-3p has been found to be upregulated and correlated with metastatic progression (138). Cellular and exosomal miR-106b-3p has been found to enhance migration and invasion, and drive pro-EMT expression, as well as promote lung metastasis in a mouse xenograft model, by targeting the tumor suppressor, deleted in liver cancer 1 (DLC-1) (138). DLC-1 is a GTPase activating protein (GAP), thus a negative regulator, for Rho GTPases, which regulate actin remodeling and thus cellular motility (139).

Exosomal miR-106b-3p from highly metastatic CRC cells also has profound effects on cell adhesion. They are released to less metastatic CRC cells and are then able to upregulate N-cadherin and downregulate E-cadherin protein expression. This is also achieved via the downregulation of DLC-1, effectively promoting metastasis in situ via coordinated effects on both cell adhesion and cytoskeletal rearrangement (138). Exosomes harboring miR-210 are also secreted by adherent HCT-8 colon cancer cells and cause them to detach and grow in suspension, indicative of metastatic capacity and anoikis resistance. Furthermore, these detached cells have been shown to be E-cadherin-negative and vimentin-positive, in contrast to adherent colonies that are E-cadherin-positive and vimentin-negative (123).

Exosomal miR-1246 in CRC, on the other hand, has been linked to the degradation of the ECM. Gain-of-function mutations in the p53 gene found in CRC cells have been shown to increase miR-1246 levels in exosomes, which are in turn able to reprogram macrophages into TAMs, the major component of tumor-infiltrating immune cells (140-142). The TAM phenotype is a classic phenotype of solid tumors with poor prognosis, and is characterized by heightened stimulation of ECM degradation, as well as enhanced migratory and invasive capacities (142).

Tumor-derived exosomal miRNAs may promote EMT reversal or MET by suppressing effectors, inducers, transcription factors, and other players involved in EMT. An increased expression of miR-200c and miR-141 in exosomes is indicative of MET in CRC cells. In a previous study, upon treatment with the drug decitabine, SW480 (primary CRC) cells did not exhibit any significant differences, while the metastatic cell lines SW620 (derived from lymph node metastasis) and SW620/OxR (derived from lymph node metastasis with acquired resistance to oxaliplatin) exhibited decreased migration and invasion properties. This was accompanied by the upregulation of E-cadherin and exosomal miR-200c and miR-141, which together suggest the acquisition of epithelial characteristics through the reversal of EMT (143).

CRC-derived exosomal miR-200c, miR-141 and miR-429 can also inhibit EMT by directly targeting the ZEB family in endothelial cells. In a previous study, in 3D spheroid models, co-culture with naïve CRC cells did not demonstrate disruption of lymphatic (exomiR-200c) (144) and blood (exomiR-200c, -141 and -429) (145) endothelial cell layers. By contrast, co-culture with 5-fluorouracil-resistant CRC cells which release exosomes without these miRNAs, resulted in enhanced disruption of endothelial cell layers. This suggests that exosomal miR-200c, -141 and -429 contribute to maintaining epithelial barrier integrity and prevents EMT, which explains the increased metastasis in chemoresistant CRC (144,145).

Another tumor suppressor exosomal miRNA is miR-1255b-5p, which has been found to be an EMT inhibitory miRNA downregulated in serum exosomes of CRC patients (146). CRC-derived exosomal miR-1255b-5p, which downregulates human telomerase reverse transcriptase (hTERT) and inhibits Wnt/β-catenin signaling, has been found to be downregulated under hypoxic conditions, providing a link between hypoxic regulation, telomere maintenance and EMT (146).

CSC and EMT plasticity can also be modulated by miRNA action, involving the deregulation of key tumor suppressor miRNAs, such as miR-200c, miR-203 and miR-183 repressed by TGF-β/Zeb1 (147); and the let-7 downregulation of high mobility group A2 (HMGA2), implicated in TGF-β/Smad/SNAIL, SLUG transcriptional activation (148).

As exosomes are considered to play a role in CSC homeostasis (149), exosomal miRNA cargoes are also implicated in modulating stemness. Bone marrow-derived mesenchymal stem cells have been found to produce exosomes enriched with miR-142-3p, which enlarges the colorectal CSC population by targeting Numb, an inhibitor of Notch signaling (150). Certain CSC-associated miRNAs have been found to be possible CRC tissue and serum biomarkers, particularly miR-18a as a metastatic serum marker (151). Exosomal miR-92a-3p secreted from CAFs has also been found to promote both stemness and EMT in CRC by downregulating FBXW7 and MOAP1 (128). Exosomal miR-128-3p is a tumor suppressor that targets B lymphoma Mo-MLV insertion region 1 homolog (Bmi1), upregulating E-cadherin and inhibiting EMT, as well as the multidrug resistance-associated protein 5 (MRP5) drug transporter, re-sensitizing resistant cells to oxaliplatin (152). Drug resistance is another phenotype closely associated with both EMT and CSCs, particularly given the important role of CSCs in mediating tumor recurrence and metastasis. CAF-derived exosomes have similarly been implicated in promoting chemoresistance through the priming of CSCs (153).

The downregulation of tumor suppressive exosomal miRNAs, as well as disposal via cargo sorting are also resorted to by CRC cells to promote invasion and metastasis. The tumor suppressive miR-149 and miR-96-5p are both downregulated in exosomes of CRC cells, while expression levels of glypican-1 (GPC1), its direct in vitro and in vivo target, and which induces EMT and promotes invasion, is increased (131).

The tumor suppressive miR-486-5p is downregulated in CRC tissues, in part because of promoter hypermethylation. miR-486-5p is a negative regulator of pleiomorphic adenoma gene-like 2 (PLAGL2), a transcription factor for β-catenin and insulin-like growth factor 2 (IGF2) with roles in promoting proliferation, cell survival and metastasis, as well as decreasing E-cadherin and increasing N-cadherin expression. Interestingly, miR-486-5p was found to be particularly enriched in plasma exosomes (154), suggesting that preferential exomiR cargo loading may be at play to evade its tumor suppressive effects.

Similar observations have been reported for the tumor suppressive miR-8073 and miR-193a. While miR-8073 has invariant expression intracellularly and in exosomes of normal colorectal cells, it is preferentially sorted into exosomes at up to 60 times the concentration found inside CRC cells. Thus, its oncogenic mRNA targets are effectively de-repressed. These include, among others, FOXM1, which is involved in cancer growth and metastasis, as well as Methyl-CpG-binding domain protein 3 (MBD3) which is known to induce pluripotent stem cells (155). In metastatic colon cancer in the liver, tumor cells selectively sort miR-193a out of cells via exosomes. Thus, its direct target inside the cell, Caprin1, is de-repressed and leads to G1 cell cycle arrest, and consequently, cell proliferation. Likewise, miR-193a has been found to be enriched in the exosomal fraction of serum from patients with CRC, particularly in advanced stages of the disease with higher risks of metastasis (156).

The journey of metastatic cells continues way after they have detached from the primary tumor, remodeled their cytoskeletal architecture, and breached tissue boundaries. The remaining steps of the metastatic cascade are fraught with further hurdles that include intravasation into the circulation, extravasation into the secondary site, and preparation of the pre-metastatic niche prior to colonization. The latter entails forming new blood vessels as well as immune-proofing of the new microenvironment.

Tumor growth and metastasis depend on blood vessels for the supply of oxygen and nutrients, for the removal of waste products, and as routes for cancer cells to be able to migrate to a different site (157). However, to stimulate new vessel growth, there must be a balance between activators and inhibitors of angiogenesis (158). Tumor-derived exosomes can shuttle such cargo, including miRNAs which can target anti-angiogenic and pro-angiogenic genes, between cancer cells and endothelial cells (159).

Among the pro-angiogenic exosomal miRNAs in CRC are miR-25-3p, miR-92a, miR-1229, miR-183-5p and miR-1246. CRC cells are able to transfer the metastasis-promoting miR-25-3p to endothelial cells via exosome transfer. By targeting Kruppel-like factor (KLF)2, vascular endothelial growth factor (VEGF) receptor 2 (VEGFR2) expression is upregulated, promoting angiogenesis (160). Targeting KLF4, on the other hand, regulates endothelial integrity through the activation of tight junction proteins, such as claudin, occludin and zonula occludens-1 (ZO-1) (161). CRC exosome-mediated transfer of miR-25-3p is thus able to induce vascular leakiness and can promote pre-metastatic niche formation in secondary sites, such as the liver and lungs. miR-25-3p has also been shown to be enriched in the circulating exosomes of CRC patients with metastasis when compared to those from patients with non-metastatic CRC (162).

Exosomal miR-92a-3p facilitates tumor angiogenesis by inducing partial EMT in endothelial cells (130). Exosomes derived from colon cancer cells and from plasma derived from murine xenograft models which were enriched with miR-92a-3p have been found to stimulate tube formation in human umbilical vein endothelial cells (HUVECs) upon transfer. miR-92a-3p promotes angiogenesis through the downregulation of Dkk-3 and claudin-11 (130,163).

CRC-derived exosomal miR-1229 promotes tube formation in HUVECs through the inhibition of homeodomain-interacting protein kinase 2 (HIPK2) and the subsequent activation of the VEGF pathway. In patients with CRC, serum exosomes harbor increased levels of miR-1229 which correlate with tumor size, lymphatic metastasis, 'tumor, nodes, metastases' (TNM) stage and poor survival (164). The upregulation of miR-183-5p in CRC cell-derived exosomes has been found to enhance angiogenesis by the repression of FOXO1 (165). A pro-angiogenic role has also been demonstrated for miR-1246, which has been found to be contained in microvesicles secreted by CRC cells, and activates Smad1/5/8 signaling via the direct targeting of promyelocytic leukemia (PML) mRNA (166).

Anti-angiogenic exosomal miRNAs in CRC include miR-126, miR-125a-3p and miR-125a-5p. miR-126 has been reported to target VEGF, an activator of angiogenesis (167) as well as its negative regulators, such as Sprouty-related, EVH1 domain-containing protein 1 (SPRED1) (168,169). However, in vitro studies on CRC have yielded contradicting results, wherein both the overexpression (170) and silencing (171) of miR-126 have been shown to lead to neo-vessel formation. Nonetheless, patients with metastatic CRC exhibit low miR-126 expression levels which are associated with poor survival, confirming the tumor suppressive role of miR-126 in CRC (167). Increased levels of extracellular miR-126 in the plasma of patients with metastatic CRC could also be a predictive biomarker for resistance to anti-angiogenic treatment using bevacizumab, a monoclonal anti-VEGF antibody (172).

miR-125a-3p targets fucosyltransferase (FUT)5 and FUT6, which regulate the PI3K/Akt signaling pathway. The overexpression of miR-125a-3p has been shown to result in the downregulation of FUT5 and FUT6, subsequently inhibiting the proliferation, migration, invasion and angiogenesis of CRC cells (173). The significant upregulation of miR-125a-3p levels in plasma exosomes is being considered as a useful biomarker for the detection of early-stage colon cancer (174).

miR-125a-5p is a known tumor suppressor in CRC, since it directly targets: i) VEGFA, resulting in reduced tube formation in HUVECs and a suppressed cell proliferation, migration and invasion in CRC (175); ii) Tafazzin (TAZ), a key transducer of the Hippo tumor-suppressor pathway, resulting in inhibited migration, invasion and EMT (176); and iii) B-cell lymphoma 2 (Bcl-2), Bcl-2-like protein 12 (BCL2L12) and myeloid cell leukemia 1 (Mcl-1), resulting in the inhibition of cell proliferation and the promotion of apoptosis (177) in colon cancer cells. However, as a biomarker, miR-125a-5p is barely detectable in plasma exosomes due to its low expression in CRC tissues (174).

Tumor-derived exosomes act as intercellular messengers between cancer cells and immune cells to either activate or inhibit immune response and/or escape recognition by the immune system. In tumors, an immunosuppressive microenvironment is mainly induced by inflammation (178). CRC-derived exosomes harboring miR-203 have been shown to differentiate monocytes into TAMs following internalization (121). It has been hypothesized that tumor-derived exosomal miR-203 targets suppressor of cytokine signaling 3 (SOCS3) (179), which is crucial for the activation of M1 macrophages characterized by a pro-inflammatory phenotype. Consequently, the loss of SOCS3 expression in the macrophages results in their anti-inflammatory M2 TAM characteristics (180). In line with this finding, circulating miR-203 has been found to promote liver metastasis in murine xenograft models, suggesting their role in hepatic pre-metastatic niche formation. In patients with CRC, a high miR-203 expression in serum exosomes and a low miR-203 expression in tumor tissues has been shown to be associated with metastasis and a poor prognosis (121).

CRC cell-derived exosomal miR-934 has been found to induce M2 polarization in macrophages, enabling the induced TAMs to promote liver metastasis via secretion of the chemokine C-X-C motif chemokine ligand 13 (CXCL13) to remodel the premetastatic niche (181). Exosomal miR-25, miR-130b and miR-425 have been similarly implicated in TAM polarization and C-X-C motif chemokine 12 (CXCL12)/C-X-C motif chemokine receptor 4 (CXCR4)-dependent liver metastasis (182). On the other hand, CRC cell-derived exosomes have been found to diminish the migration of THP-1 monocytes in vitro via the delivery of let-7d, which can downregulate the chemokine C-C motif chemokine ligand 7 (CCL7) (183), suggesting that exosomal miRNAs can aid immune evasion by interfering with immune cell chemotaxis. A list of exosomal miRNAs that have been implicated in different steps of the metastatic cascade is presented in Table I.