Background

Although cancer cells arise from genetic mutations in normal cells, the malignant phenotypes of tumours are influenced by their surrounding tumour microenvironment (TME) [1]. The TME comprises various cellular components, including fibroblasts, endothelial cells, and inflammatory cells, as well as noncellular components, such as the extracellular matrix [1]. Accumulating evidence has demonstrated that among these TME components, cancer-associated fibroblasts (CAFs) modulate tumour metastasis and tumour mechanics and influence angiogenesis, the immune system, and therapeutic resistance [2]. Although CAFs are complex and include subtypes with protumorigenic and antitumour effects, several reports have indicated that CAFs are potential therapeutic targets for cancer treatment [2, 3]. Direct cell-to-cell contact and humoral factors such as cytokines and chemokines are considered the principal lines of communication between cancer cells and CAFs. However, recent studies have revealed that extracellular vesicles (EVs) are also regulatory players in such communication.

EVs are vesicles enclosed by a lipid bilayer membrane that include various bioactive molecules, such as proteins, mRNAs, metabolites, and microRNAs (miRNAs). Historically, EVs were initially considered “garbage bins” for exporting unnecessary intracellular components [4]. However, in 2007, Valadi et al. proposed that mRNAs packaged in EVs can be utilized for intercellular communication upon translation in recipient cells [5]. This discovery accelerated research on the role of EVs as novel intercellular communication tools. In particular, over the past ten years, there has been a striking expansion of EV research in cancer-related fields [6]. Various cancer-related factors packaged into EVs are transferred to recipient cells within proximal sites and distal metastatic niches. Intriguingly, Hoshino et al. indicated that integrin marks on EV surfaces could be considered a “zip-code” for delivery to specific organs [7], suggesting that cancer cells utilize EVs to confer specific features on cells in distal organs to support metastasis. In addition, cells within the TME, particularly CAFs, also regulate tumour progression, such as metastasis, by transferring EVs to cancer cells [6]. These findings imply that EVs derived from cancer cells and host-derived CAFs are involved in the formation of a suitable TME that promotes cancer progression.

In addition to the striking functions of EVs in intercellular communication, these vesicles can be detected in various body fluids, such as blood, saliva, and urine [8]. Therefore, EVs are attractive targets for improving cancer treatment and diagnosis. How do cancer cells utilize EVs to dictate the function and phenotypes of surrounding TME components to support cancer progression? This review will focus on recent advances in the interplay between cancer cells and CAFs and summarize the current knowledge regarding the functional role of EVs in the interplay between cancer cells and fibroblasts within the TME.

Nomenclature of extracellular vesicles

There are various subtypes of EVs with differential components and biogenesis processes. They are traditionally classified into exosomes, ectosomes, and apoptotic bodies based on origin and size [4, 9]. Exosomes originate from the exocytosis of multivesicular bodies (MVBs) [9, 10]. Ceramide-dependent pathways and endosomal sorting complex required for transport (ESCRT) machinery partially regulate exosome release. Exosomes can contain the following proteins: membrane transport and fusion proteins (GTPase, annexins, and flotillin), tetraspanins (CD9, CD63, and CD81), heat shock proteins (Hsc70 and Hsp90), MVB-related proteins (Alix and TSG101), lipid proteins and phospholipases. Microvesicles are generated directly from budding of the plasma membrane [11]. Microvesicles can be enriched in some lipid components and phosphatidylserine [12]. Apoptotic bodies are generated during apoptosis. Thus, they contain intracellular fragments such as organelles, membranes, and cytosolic and nuclear fragments [13, 14]. However, it is still difficult to distinguish various types of EVs because of the lack of specific protein markers for each subtype. Thus, the expert consensus in EV research encourages the use of operational names unless specific and reliable EV markers of subcellular origin can be established [9]. (a) Physical characteristics, such as the size or density of EVs: small EVs (sEVs: < 100 nm or < 200 nm) and large and/or medium EVs (m/lEVs: > 200 nm). (b) Biochemical composition, such as positivity of staining for CD63, CD81, and Annexin A5. (c) Description of conditions or cells of origin, such as from podocytes and apoptotic bodies. In this review, to avoid confusion, we use the term EVs for all subtypes of membrane vesicles in the extracellular space rather than specific terms such as exosomes.

What are CAFs?

To discuss the functional role of EVs in cancer-CAF interactions, we first need to define CAFs. CAFs are fibroblasts within the tumour stroma that can originate from many cell types, including resident fibroblasts, adipocytes, bone marrow-derived mesenchymal stem cells (BM-MSCs), and endothelial cells. The subtypes of CAFs include myofibroblast-like phenotypes similar to the features of activated fibroblasts during wound healing [15]. Many researchers, therefore, utilize adopted activated fibroblast markers such as alpha-smooth muscle actin (α-SMA) for the detection of CAFs [15]. However, recent promising studies and single-cell technologies have indicated that CAFs comprise diverse subtypes with distinct characteristics [16, 17]. Thus, conventional markers of CAFs are likely insufficient to detect all CAF subtypes within tumour tissue [18]. These findings are consistent with the fact that normal fibroblasts are also heterogeneous and lack unique markers that are not expressed in other cell types [2]. Although the precise definition of CAFs is still challenging, CAFs are practically considered morphologically elongated cellular components that are negative for epithelial, endothelial and leukocyte markers within tumour tissues [2]. To exclude the possibility of cancer cells undergoing epithelial–mesenchymal transition (EMT), CAFs are also defined as cells lacking genetic mutations found within the cancer cells [2].

The origin of CAFs and the signalling that mediates CAF heterogeneity remain controversial. Given the diverse subtypes in the normal fibroblast population, CAF subtypes may reflect their original tissue background. Indeed, in breast cancer, Raz et al. demonstrated that PDGFR-α-positive and PDGFR-α-negative subsets of CAFs are generated from BM-MSCs and resident fibroblasts, respectively [19]. They utilized a unique in vivo model involving BM implantation from GFP-expressing mice to MMTV-PyMT transgenic mice to track the cellular fate of BM-MSCs within the tumour mass. This study showed that BM-MSC-derived CAFs are negative for PDGFR-α expression and enhance angiogenesis and tumour growth. In contrast, in pancreatic ductal adenocarcinoma (PDAC), Tuveson's group showed that pancreatic stellate cells differentiate into two subsets of CAFs, inflammatory CAFs (iCAFs) and myofibroblastic CAFs (myCAFs) [20]. They also showed that IL-1 signalling induces iCAFs, but TGF-β signalling antagonizes this pathway, leading to myCAF differentiation [21]. Notably, these CAF subtypes are likely responsible for distinct functions for cancer progression, such as immunomodulatory functions and extracellular matrix (ECM) remodelling [19, 20]. These findings collectively imply that tissue environmental background and cancer-derived factors determine CAF complexity.

Diverse functions of CAFs in tumour progression have been reported. One substantial function of CAFs is as a source of various types of growth factors and chemokines, such as transforming growth factor-beta (TGF-β) [15, 22], CXC-chemokine ligand 12 (CXCL12)/SDF-1 [23], and HGF [24]. These CAF-derived factors promote an aggressive phenotype of cancer cells with invasive and metastatic behaviour, including the induction of EMT. The CD10 and GPR77 double-positive CAF subset secretes IL-6 and IL-8, sustaining cancer stemness and promoting chemoresistance [25]. In addition to direct effects on cancer cell behaviours, CAFs also generate VEGF to induce the angiogenesis of endothelial cells. CAFs also produce IL-6 and CXCL9 and regulate the immunosuppressive or immuno-promoting functions of inflammatory cells. The newly identified CAF subtype, antigen-presenting CAFs (apCAFs), can mediate CD4 + regulatory T-cell activation and CD8 + T-cell suppression [26]. These findings indicate that CAFs can mediate the function of both cancer cells and surrounding stromal cells within the TME.

Another essential function of CAFs is remodelling the extracellular matrix within the TME. CAFs produce ECM-related molecules, such as collagens and metalloproteases, which lead to the deposition and remodelling of the ECM. This remodelled ECM causes increased tissue stiffness, triggering proliferation signals in cancer cells. The increased tissue stiffness also promotes blood vessel collapse, reducing drug delivery. Matrix stiffening within tumour tissue also enhances the activation of YAP in normal fibroblasts and CAFs, which is involved in CAF generation and maintenance [27]. In addition, the metalloprotease produced by CAFs generates migratory tracks that allow cancer cell invasion [28]. Since CAFs can mediate the migration of cancer cells via E-cadherin/N-cadherin interactions [29], CAFs may be able to guide cancer cell invasion.

Cancer-derived EVs can dictate preferable CAF characteristics for cancer progression

Over the past few years, there has been an increasing number of reports regarding the functional interplay of EVs in communication between cancer cells and CAFs (Tables 1 and 2). In this review, to better understand the intercellular crosstalk between cancer cells and CAFs via EVs, we focused on the papers that investigated EV components and revealed their functions on the recipient cells.

Table 1 Cancer-derived EV functions on CAF generation and property
Full size table
Table 2 CAF-derived EV functions on cancer cell property
Full size table

Many researchers have reported how CAFs are generated by cancer cells. Both direct contact and indirect interaction between cancer cells and CAFs are essential to maintain CAF characteristics. Among these interactions, TGF-β signalling is a well-established factor for activating fibroblasts and generating CAFs [2]. In addition to these mechanisms, cancer cells utilize EVs to generate CAFs (Fig. 1 and Table 1). The first report on cancer-derived EVs related to CAF induction was published by Webber et al., who showed that TGF-β was expressed on the surface of EVs derived from prostate cancer and mesothelioma cell lines [30, 31]. TGF-β on these cancer-derived EVs triggered the TGF-β/SMAD3 signalling pathway in recipient fibroblasts and induced myofibroblast-like phenotypes, including the expression of α-SMA and fibroblast growth factor 2 (FGF2) [30]. Recently, Huang et al. also showed that TGF-β in EVs generates CAFs via a noncanonical fibronectin-dependent pathway [32]. These reports demonstrated that EVs are responsible for some of the signalling previously thought to be carried out by growth factors. In studying triple-negative breast cancer, Sung et al. utilized an experimental model expressing integrin beta 4 (ITGB4) to determine the effect of EVs on the properties of CAFs [33]. Cancer-derived EVs transfer ITGB4 and induce BCL2 interacting protein 3 like (BNIP3L)-dependent mitophagy and glycolysis in CAFs. They also showed that the overexpression of ITGB4 in CAFs promotes breast cancer cell proliferation, EMT, and invasion [33]. Yan et al. demonstrated the reprogramming of glucose and glutathione metabolism in CAFs induced by miR-105 transferred by metastatic breast cancer cell-derived EVs [34]. Targeting of MXI1 by miR-105 activates MYC, the essential driver of metabolic reprogramming in CAFs. Long non-coding RNAs (lncRNAs) are also transferred from cancer cells to fibroblasts. Since the exchange of metabolites and amino acids is one mechanism of interplay between cancer cells and CAFs [35,36,37], it is highly plausible that cancer cells utilize EVs to dictate metabolic reprogramming in CAFs and thus construct a preferable microenvironment for their progression.

Fig. 1
figure 1

Summary of the intercellular crosstalk between cancer cells and CAFs via EVs. The interplay between cancer cells and cancer-associated fibroblasts (CAFs) generates a unique tumour microenvironment (TME) that supports cancer progression. Cancer cells utilize EVs to modify surrounding cells within the tumour microenvironment. Cancer cells secrete various bioactive molecules in EVs, which are transferred into the surrounding fibroblasts and CAFs; such molecules include transforming growth factor-beta (TGF-β), ITGB4, p53, mRNAs and miRNAs (Table 1). These cancer-derived EVs confer specific phenotypes, such as mitophagy and glycolysis, myofibroblast features, matrix remodelling, and inflammatory gene expression (blue arrows), to support cancer progressions (broken arrows, the functions for cancer progression) (Table 1). Cancer-derived EVs also induce distinct CAF subtypes and contribute to the heterogeneity of CAFs. On the other hand, to support cancer progression, CAFs utilize EVs to transfer various functional molecules, such as CD81, CD9, FAP, miRNAs, lncRNAs, transposable RNAs and metabolites. These molecules confer aggressive phenotypes on cancer cells (red arrows). However, some molecules are downregulated in CAFs and thus their transfer to cancer cells is prevented (Table 2). Interestingly, specific CAF subtype-derived EVs are also involved in cancer progression, suggesting that EV content may differ among CAF subtypes

Full size image

Cancer cell properties such as TP53 gene mutation also affect EV components and have consequences on CAF generation [38,39,40,41]. Novo et al. demonstrated that TP53-mutant cancer cells mediate integrin trafficking in fibroblasts via EVs and promote the deposition of a pro-invasive ECM [38]. Although they did not focus on molecules encapsulated in EVs, this finding implies that EVs are involved in matrix remodelling within the TME via CAF generation and activation. Interestingly, Ma et al. showed the direct transfer of gain-of-function p53 proteins from cancer cells to fibroblasts via EVs. This gain-of-function p53 isoform induced a CAF phenotype in fibroblasts through the Nrf2-dependent pathway [39]. Since Nrf2 is the essential regulator of ECM production and deposition [42], transferring the gain-of-function p53 isoform to CAFs might involve a proinvasive ECM construction. In addition, Vu et al. found that miR-125b in breast cancer cell-derived EVs generates CAFs from resident fibroblasts [43]. Downregulation of TP53INP1 by EV-transferred miR-125b induced a CAF state in fibroblasts to promote tumour growth [43]. Given that aberrant p53 function can confer a CAF phenotype on fibroblasts [38, 39, 44], the expression status of the TP53 gene in both cancer cells and CAFs may reflect the features of CAF subtypes. Recently, Bhome et al. demonstrated that the EMT phenotype of colorectal cancer (CRC) cells influences CAF generation [45] CRC cells with an epithelial phenotype secrete miR-200 family members via EVs to attenuate TGF-β-mediated CAF features by targeting ZEB1 in normal fibroblasts. However, EVs derived from CRC cells with a mesenchymal phenotype contain less miR-200 family, thus allowing TGF-β-mediated CAF feature induction. An aggressive subtype of CRC based on the consensus molecular subtypes (CMSs) is associated with mesenchymal gene signatures. Their study provides one of the essential clues as to why cancer cells with aggressive phenotypes possess CAF-abundant stroma.

EVs also generate specific CAF subtypes from resident fibroblasts, promoting CAF heterogeneity. Our group reported that highly metastatic gastric cancer (GC) cells secrete EVs and induce the expression of chemokines such as IL-1 and IL-8 in stomach fibroblasts, which have similar properties as iCAFs [46]. Regarding the mechanism, these GC-derived EVs transport miRNAs, including miR-193b, to fibroblasts and induce chemokine expression. On the other hand, these GC cell-derived EVs do not have the capacity to regulate the expression of α-SMA and collagen, which are myCAF features. Notably, compared with GC cells with low metastatic potential, those with high metastatic potential can more strongly induce both α-SMA and chemokine expression in fibroblasts. These data suggest that cancer-derived EVs selectively confer iCAF-like features on fibroblasts. IL-8-expressing CAFs within the tumour stroma are closely associated with poor outcomes in patients with GC. These findings collectively imply that GC generates a protumorigenic microenvironment to support disease progression. Other research groups have also shown that cancer-derived EVs affect CAF heterogeneity [47, 48]. Goulet et al. showed that bladder cancer-derived EVs induce iCAF features in primary bladder fibroblasts [48]. Compared with TGF-β treatment, these cancer-derived EVs have a weaker effect on the induction of α-SMA in fibroblasts [48]. Although these researchers did not clarify the contribution of cancer-derived EVs to iCAF differentiation, they showed that EV-induced iCAFs express IL-6 and activate the STAT3 signalling pathway that leads to EMT in bladder cancer [48]. Lahav et al. demonstrated that metastatic melanoma cell-derived EVs activate proinflammatory signalling in fibroblasts within metastatic niches [47]. They also found that melanoma cell-derived EVs contain RNAs capable of instigating inflammatory signalling in fibroblasts, such as high-mobility group box 1 (Hmgb1), thymic stromal lymphopoietin (Tslp), and interferon regulatory factor 1 (Irf1) [47]. Interestingly, these EVs cannot induce myCAF features, suggesting that metastatic cancer cell-derived EVs may preferentially encourage iCAF features rather than myCAF features [49]. These findings indicate that EVs deliver messages from cancer cells, instigate various features in CAFs, and affect the feedback from CAFs for cancer progression.

CAF-derived EVs change the properties of cancer cells during disease progression

Stromal fibroblasts “educated” by cancer cells also secrete EVs and establish intercellular communication that ultimately benefits cancer progression (Fig. 1 and Table 2). Luga et al. first reported the functional role of CAF-derived EVs in cancer progression. They showed that CD81-positive CAF-derived EVs activate the Wnt-planar cell polarity (PCP) signalling pathway and enhance breast cancer cell motility and metastasis [50]. CD81 on the surface of EVs may support the endocytic trafficking of Wnt11 in cancer cells. This mechanism mediated by CAF-derived EVs activates the Wnt-PCP pathway in cancer cells to drive invasive behaviour [50]. CD81, a tetraspanin family member protein, is a conventional marker of EVs [9]. Other tetraspanin family proteins may also support the internalization of CAF-derived EVs. Miki et al. showed that CAFs secrete CD-9-positive EVs and stimulate diffuse-type GC migration. In addition, they showed that a CD9 neutralizing antibody inhibits EV uptake by GC cell lines [51]. Consistent with this finding, Nigri et al. also demonstrated that CD9 neutralizing antibody impaired CAF-derived ANXA6-positive uptake by PDAC cells [52]. ANXA6 can interact with CD9 in CAFs, suggesting that CD9 might regulate EV-specific ANXA6 shuttling. ANXA6- and CD9-double-positive CAF-derived EVs facilitate p38 MAPK signal activation to enhance migratory ability in PDAC cells. Although these reports did not clarify how the suppression of CD9 expression inhibits EV uptake in cancer cells, they identified CD9 function as a mediator of the internalization of CAF-derived EVs into cancer cells.

CAF-derived EVs support the development of cancer cell therapeutic resistance. Minn's group reported the precise mechanisms by which breast cancer cells achieve therapeutic resistance through CAF-derived EVs [53, 54]. CAF-derived EVs contain transposable RNAs that stimulate RIG-I recognition to activate STAT1 in cancer cells. Activated STAT1 cooperates with juxtacrine-activated NOTCH3 to mediate NOTCH target gene transcription that supports resistance to chemotherapy and radiation [53]. They also found that NOTCH-MYC signalling induces RN7SL1 expression in CAFs. RN7SL1 generally possesses 5′ppp capping that is regulated by SRP9 and SRP14, which may prevent its recognition by RIG-I. However, NOTCH-MYC signalling in CAFs induces RN7SL1 without 5′ppp capping, which is encapsulated into CAF-derived EVs and acts as a damage-associated molecular pattern (DAMP) to activate RIG-I in recipient cancer cells [54]. Kunou et al. demonstrated the function of CAF-derived EVs in the gemcitabine resistance in malignant lymphoma [55]. These CAF-derived EVs suppress the expression of equilibrative nucleoside transporter 2 (ENT2) in lymphoma cells. They also found that miR-4717-5p is one of the most abundant miRNAs in CAF-derived EVs and directly targets ENT2 expression [55]. In colorectal cancer (CRC), Deng et al. showed that FAP-positive CAFs secrete lncRNAs via EVs [56]. These CAF-derived lncRNAs promote oxaliplatin resistance in CRC cells in vitro and in vivo by stabilizing human antigen R (HuR) mRNA to increase β-catenin expression. The FAP-positive CAF subset is associated with regulatory T-cell-mediated immunosuppression, which results in poor outcomes in breast cancer patients [57]. Thus, it is possible that the FAP-positive CAF subset also secretes functional EVs to facilitate the malignant behaviour of cancer cells. Indeed, Dou et al. reported the immunomodulatory functions of CAF-derived EVs [58]. Although they did not refer to CAF subsets, they showed that CAF-derived miR-92 directly targets LATS2 expression and induces PD-L1 expression through YAP activation in breast cancer cells [58].

The functional role of CAF subset-derived EVs in cancer progression was recently reported. Gao et al. identified CD63-positive CAF subsets using single-cell analysis of oestrogen receptor α (ERα)-negative tumours in the MMTV-PyMT mouse mammary gland carcinoma model [59]. These CAF subsets secrete miR-22 via EVs to directly suppress the expression of ERα and PTEN in ERα-positive breast cancer cells and organoids, resulting in the induction of tamoxifen resistance. Interestingly, TIMP1, a CD63 ligand, can induce the STAT3 signalling pathway to regulate the expression of miR-22 and CD63 to generate a CD63-positive CAF phenotype. TIMP1 is upregulated in CD63-positive CAFs, suggesting a TIMP1-mediated positive feedback loop in the CD63-positive CAF phenotype. In addition, they showed that SFRS1 mediates miR-22 loading into CD63-positive CAF-derived EVs [59]. Although it is still unclear whether EV components differ among all CAF subsets, these findings indicate that CAF heterogeneity affects EV function in cancer progression.

Cancer-derived EVs reprogram energy metabolism in CAFs, as discussed in the previous section. CAFs also support the metabolic properties of cancer cells [60, 61]. Zhao et al. showed that CAF-derived EVs contain intact metabolites, such as amino acids, and change the metabolic properties of prostate cancer cells [60, 61]. Oxidative phosphorylation is inhibited in the presence of CAF-derived EVs. In contrast, glycolysis and lactate levels in prostate cancer cell lines are increased by these CAF-derived EVs. In addition to the direct transfer of metabolites via EVs, Li et al. also showed that the lncRNA SNHG3 affects the metabolic reprogramming of breast cancer cells. SNHG3 in EVs acts as a molecular sponge of miR-330-5p, resulting in increased lactate production in breast cancer cells [62]. Breast cancer-derived EVs transfer miR-122 and modify glucose consumption by fibroblasts at metastatic sites by targeting pyruvate kinases [63]. Moreover, metabolic dysregulation of CAFs may alter immunoregulation through IL-6 production [36]. Collectively, these findings imply that communication networks involving EVs contribute to cancer survival in environments with limited oxygen and nutrient supplies.

Perspectives on the therapeutic and diagnostic application of CAF-derived EVs

As described above, EV-mediated interactions between cancer cells and CAFs have been implicated in several malignant behaviours of cancer cells. Therefore, targeting EVs and CAFs is being explored as a fascinating strategy for cancer therapy. Kong et al. suggested that the inhibition of vitamin D receptor (VDR) signalling could target miR-10 secretion via EVs from CAFs and inhibit its cancer-promoting functions in PDAC [64]. This study was motivated by the two reports demonstrating that VDR signalling reprograms CAFs, thus diminishing their cancer-promoting role in PDAC, and is associated with better clinical outcomes in CRC [65, 66]. The inhibitory effects of two compounds on the interplay between cancer cells and CAFs have also been demonstrated. Lee et al. showed the impact of the multitarget small drug HNC0014 on the function of cancer spheroid-derived EVs. HNC0014 inhibits CAF generation by these EVs and may contribute to the antitumour immune response [67]. Chen et al. utilized ovatodiolide (OV), the bioactive component of Anisomeles indica, to suppress CAF generation and tumour sphere formation induced by cancer-derived EVs [68]. Interestingly, proton-pump inhibitors (PPIs), which are used to eradicate the oncogenic pathogen Helicobacter pylori, may also inhibit CAF generation [69].

Some studies have identified the molecules specific to CAF-derived EVs and these molecules have been proposed as diagnostic biomarkers for cancer. Ganig et al. performed proteome analysis of EVs derived from CAFs and normal fibroblasts and identified 11 proteins with differential abundance in CAF-derived EVs [70]. Among these candidate proteins, QSOX1 was selected for further study and was found to be decreased in CAF-derived EVs compared with normal fibroblast-derived EVs. The abundance of QSOX1 in plasma samples was reduced in CRC patients compared to healthy donors and patients with benign disease [70]. Although this study used small cohorts to investigate QSOX1 in EVs, the findings suggest the utility of CAF-derived EVs in cancer diagnosis. The secretome of fibroblasts influenced by cancer-derived EVs has also been investigated as a biomarker for lung cancer detection. Zhang et al. compared the secretome profile of lung cancer-derived EV treated and untreated MCR5 fibroblasts to find candidates for a diagnostic marker [71]. Based on the multiple datasets comprising 1897 patient samples, they showed that the gene expression signatures of five candidates are independent prognostic factors for identifying patients who may require adjuvant therapy.

There remain challenges in the clinical application of EVs because of their diversity and the difficulty in selectively targeting cancer-associated EVs [8]. However, further investigations into targeting intercellular communication via EVs will provide several avenues for cancer therapy and diagnosis.

Conclusion

The precise mechanisms of intercellular communication between cancer cells and CAFs remain obscure because diverse pathways modulate multiple EV components and non-EV factors, such as growth factors, cytokines and chemokines. In addition, CAFs are surprisingly heterogeneous, which is the major obstacle to understanding their biology and developing related therapeutics. However, as described above, remarkable progress in EV research has led to the elucidation of the novel mechanism underlying the intrinsic cancer cell–stromal cell interplay during cancer initiation and progression. EVs derived from cancer cells and CAFs have impressively diverse functions that contribute to tumour progression. Cancer cells secrete EVs and thus dictate specific properties in CAFs. Against such “education” by cancer cells, CAFs “respond” via EVs to suppress cancer initiation and progression. Understanding the precise mechanisms of EVs in cancer cell–CAF interactions may provide breakthroughs in the development of diagnostic and prognostic tools and therapeutic strategies for cancer.