Functional heterogeneity of exosomes in hepatocellular carcinoma (Review)

1. Introduction

Primary liver cancer remains one of the most common malignancies and a leading cause of cancer-associated mortality worldwide, with hepatocellular carcinoma (HCC) accounting for the majority of cases. In 2020, ~905,700 new liver cancer cases and 830,200 liver cancer-related deaths occurred worldwide (1). The multifactorial pathogenesis of HCC involves chronic hepatitis B or C virus infection, alcohol-associated liver disease, metabolic liver disorders, particularly non-alcoholic fatty liver disease and aflatoxin exposure (2-8). Current therapeutic algorithms integrate tumor stage, hepatic functional reserve and patient comorbidities. Curative treatments are primarily available for early-stage disease, although, >85% of patients are diagnosed at intermediate or advanced stages, partly due to profound molecular and clinical heterogeneity (9,10). Patients with intermediate-stage disease may benefit from locoregional therapy, whereas advanced disease often requires systemic treatment, including tyrosine kinase inhibitors and immunotherapy-based regimens (5,11,12). Despite these advances, the overall survival of patients with HCC remains poor, largely due to limitations in early detection and durable systemic disease control (10,13). Therefore, robust biomarkers for early diagnosis and effective therapeutic targets are needed.

The emerging paradigm of exosome-mediated intercellular signaling has advanced the current understanding of HCC pathobiology. These 30- to 150-nm vesicles act as molecular carriers that transport proteins, lipids, RNA and DNA to recipient cells, thereby regulating cellular processes, including proliferation, migration, invasion, immune responses and metabolic reprogramming (14-16). Hu et al (17) reported that HCC-derived exosomes transfer circular RNA (circ)-cell cycle and apoptosis regulatory protein 1 (CCAR1) to CD8+ T-cells and induce T-cell exhaustion through STAT3 signaling. Conversely, preclinical models have demonstrated a tumor-suppressive role for microRNA (miR)-125b-enriched exosomes, which inhibit TGF-β1/SMAD2-mediated epithelial-mesenchymal transition (EMT), thereby suppressing metastasis and attenuating HCC progression (18). Given their key roles in HCC progression and metastasis, exosomes have emerged as promising candidates for early diagnosis and targeted therapy. In the present review, exosome biogenesis, isolation, characterization and functional heterogeneity are summarized, with a particular emphasis on mechanistic roles, biomarker evidence and therapeutic implications in HCC.

2. Overview of exosomes

According to the Minimal Information For Studies Of Extracellular Vesicles 2023 recommendations, the term ‘exosome’ should ideally be reserved for extracellular vesicles (EVs) whose endosomal origin has been experimentally demonstrated. However, the majority of HCC studies isolate vesicles primarily according to size, density or surface-marker expression, without directly establishing their biogenesis. Therefore, the present review retains the terminology used in the original studies when describing specific findings, whereas the more general terms ‘EVs’, ‘small EVs’ (sEVs) or ‘EV-enriched fractions’ are used when vesicle origin has not been conclusively demonstrated (19,20).

The extracellular environment contains diverse populations of membrane-bound particles collectively termed EVs. EVs arise through different biogenetic pathways, including endosomal sorting and plasma membrane budding mechanisms, and are released by numerous cell types, including tumor cells, immune cells, hepatocytes and stromal cells, into biological fluids such as blood, plasma, serum, urine and saliva (21-25). EVs participate in diverse physiological and pathological processes by transferring bioactive cargo, including proteins, lipids and nucleic acids, to recipient cells and modulating cellular signaling, immune responses and tissue homeostasis, thereby influencing processes such as immune regulation (26), cellular differentiation (27), tissue regeneration (28), viral pathogenesis (29), tumorigenesis and metastasis (30). This functional versatility makes EVs promising noninvasive biomarkers for disease characterization (31).

EVs comprise heterogeneous membrane-bound particles that differ in their cellular origin, biogenesis, size, molecular composition and biological activity. Previously, EVs have been classified as exosomes, microvesicles (ectosomes) and apoptotic bodies according to their presumed biogenetic pathways (19,32,33). However, vesicle size alone cannot reliably define biogenesis as the size distributions of these populations substantially overlap. Therefore, unless endosomal origin has been experimentally demonstrated, the International Society for EVs recommends the use of operational terms such as ‘sEVs’ and ‘medium/large EVs’ rather than assigning a specific biogenetic origin (20).

Exosomes are nanosized EVs surrounded by a phospholipid bilayer with an asymmetric distribution of lipids between the two leaflets (Fig. 1) (34). Their membranes contain sphingomyelin, phosphatidylcholine, phosphatidylethanolamine, phosphatidylserine, monosialotetrahexosylganglioside and phosphatidylinositol (35). Sphingomyelin is predominantly localized to the outer leaflet and phosphatidylcholine to the inner leaflet, whilst phosphatidylethanolamine exhibits a nonpolarized distribution. Notably, phosphatidylserine (PS) externalization, a well-established feature of apoptotic and malignant cells, facilitates macrophage recruitment and immune modulation through recognition by PS receptors, including T-cell immunoglobulin and mucin domain-containing protein 4, brain-specific angiogenesis inhibitor 1 and stabilin-2. These interactions activate downstream signaling pathways involved in efferocytosis, inflammatory regulation and immune remodeling. In addition, altered PS exposure on extracellular vesicles may influence their interaction with immune cells and supports the potential application of exosomal lipid profiling as a diagnostic approach in oncology (36-38).

Figure 1 Schematic illustration of exosome structure. Figure created with MedPeer (medpeer.cn). ssDNA, single stranded DNA; ESCRT, endosomal sorting complex required for transport; Hrs, hepatocyte growth factor-regulated tyrosine kinase substrate; TSG101, tumor susceptibility gene 101; ALIX, ALG-2-interacting protein X; HSP, heat-shock protein; LAMP1, lysosome-associated membrane protein 1.

The stable phospholipid bilayer protects luminal cargo from enzymatic degradation and may confer favorable biocompatibility, depending on cellular origin and cargo composition. These properties support the potential use of exosomes as nanoscale drug-delivery systems. Exosomes are commonly enriched in tetraspanins (such as CD9, CD63 and CD81), which participate in membrane dynamics; endosomal sorting complex required for transport (ESCRT)-associated proteins [such as tumor susceptibility gene 101 (TSG101), ALG-2-interacting protein X (ALIX) and hepatocyte growth factor regulated tyrosine kinase substrate] which are involved in multivesicular body (MVB) formation; and heat-shock proteins (such as Hsp70 and Hsp90), which are associated with cellular stress responses. CD9, CD63, CD81, lysosome associated membrane protein 1 (LAMP1) and TSG101 are commonly used as EV-associated markers for characterization (32,33,36).

Process of exosome formation

Exosome biogenesis predominantly occurs through the endocytic pathway (Fig. 2). Plasma membrane invagination first generates early endosomes, which subsequently mature into late endosomes. Inward budding of the endosomal membrane produces intraluminal vesicles (ILVs), which accumulate within late endosomes to form MVBs. The majority of MVBs fuse with lysosomes for cargo degradation, whereas a subset is transported to the plasma membrane through the coordinated actions of cytoskeletal elements (including actin and microtubules), motor proteins (including dynamin and myosin) and molecular regulators such as GTPases. Exosome release is regulated by Ras-associated binding (RAB) GTPases, cytoskeletal transport and soluble N-ethylmaleimide-sensitive factor attachment protein receptor (SNARE) proteins that mediate fusion of MVBs with the plasma membrane. LAMP1, LAMP2 and CD63 are associated with late endosomal and lysosomal compartments but are not components of the SNARE machinery (35,39). However, these proteins are closely related to exosome biogenesis and characterization, as they participate in endosomal maturation and are frequently used as markers of late endosomal compartments and EV-associated fractions. Therefore, distinguishing their roles from those of SNARE proteins is important for accurately describing the molecular regulation of exosome release.

Figure 2 Exosome formation pathway. LAMP, lysosome-associated membrane protein; MVB, multivesicular body; IPMCs, intracellular plasma membrane-connected compartments. Figure created with MedPeer (medpeer.cn).

ILV formation is coordinated by the ESCRT machinery and accessory proteins such as ALIX and TSG101. These components facilitate the sorting of ubiquitinated MVB membrane proteins and promote membrane curvature through sequential recruitment of the ESCRT-0, -I, -II and -III complexes, followed by vesicle scission mediated by the vacuolar protein sorting 4 homolog A (VPS4) ATPase. The ESCRT-III/VPS4-mediated membrane-remodeling process is a principal mechanism of ILV biogenesis (33,35,40).

In addition to ESCRT-dependent mechanisms, Wei et al (41) described an alternative pathway involving RAB31 activation. EGFR-dependent phosphorylation activates RAB31, which subsequently recruits flotillins within lipid rafts to promote inward budding of the endosomal limiting membrane and ILV formation. RAB31 also protects MVBs from lysosomal degradation by inhibiting their fusion with lysosomes through modulation of lysosomal fusion machinery, including SNARE-dependent membrane fusion components. This regulation preserves MVB availability for plasma membrane docking and exosome release, thus illustrating the mechanistic complexity of exosome production. Complementing endosomal pathways, sEVs can also bud from specialized plasma-membrane domains. These vesicles should not automatically be classified as exosomes unless their endosomal origin is demonstrated (Fig. 2). CD4+ T lymphocytes exemplify this mechanism, releasing exosomes from specialized plasma membrane domains enriched with exosomal markers (including CD63 and CD81). Macrophages employ a distinct strategy through the intracellular plasma membrane-connected compartment (IPMC), which maintains persistent extracellular communication through a membranous neck. Unlike transient MVB-plasma membrane fusion, the IPMC permits continuous small molecule exchange while generating exosomes, which is a unique feature with implications for tumor microenvironment (TME) modulation.

Methodological advances in exosomal isolation

Common strategies for exosome isolation include ultracentrifugation (UC), ultrafiltration, size-exclusion chromatography (SEC), immunoaffinity capture (IAC), polymer-based precipitation (32,42) and emerging microfluidic platforms (43). These methods exploit differences in particle size, buoyant density, surface-antigen expression and solubility to separate EV-enriched fractions.

UC has previously been one of the most widely used methods for EV isolation (36,42,44). Although well-established and reagent-free, UC remains time-consuming, requires specialized equipment and may cause vesicle aggregation or co-isolation of non-vesicular contaminants. These limitations restrict its use in large-scale and clinical applications (42,44).

Differential UC (DUC) and density-gradient UC (DGUC) are two commonly used UC-based approaches. DUC uses sequentially increasing centrifugal forces and spin durations to remove cells, cellular debris and macromolecular contaminants. Following removal of the supernatant, exosomes are typically recovered by UC at 100,000-150,000 x g. Despite being operationally simple, DUC remains labor-intensive and time-consuming and can cause notable co-sedimentation of particles with similar sizes, particularly those in the 50-150-nm range. DGUC uses inert media, such as sucrose or cesium chloride, to establish a density gradient for isopycnic separation of vesicular components. Density-specific band collection can yield purer EV preparations than DUC. However, the need for precise gradient preparation and the technical complexity of the procedure limit throughput and practical use in resource-constrained settings (40,45,46).

Ultrafiltration is another widely used method for exosome separation (33,47). Membrane-based ultrafiltration separates particles by forcing biological samples through nanoporous barriers with pore sizes of 30-200 nm. Common membrane materials include track-etched polycarbonate and polyethersulfone. Comparative studies have shown that ultrafiltration is faster than UC (4-6 vs. >10 h) and reduces dependence on specialized UC equipment (33,47-49). However, shear stress may deform or rupture vesicles, membranes may become clogged by adherent particles and biomolecules of similar size may remain as contaminants. Tangential-flow filtration partially mitigates these limitations through continuous-flow operation and improved membrane performance. SEC, also termed gel filtration, separates particles according to hydrodynamic volume using a porous stationary-phase matrix. Macromolecular species that cannot enter the pores elute earlier, whereas smaller particles elute later after entering the pores. SEC generally improves preservation of native vesicle morphology and structure compared with UC, although its moderate resolution may require additional purification or concentration steps, including the use of qEV or EVSecond columns (50-53). SEC is also less costly compared with affinity-based methods and avoids antibody-dependent selection bias.

IAC exploits specific binding between EV surface epitopes, such as the tetraspanins CD9, CD63 and CD81 or epithelial cell-adhesion molecules, as well as immobilized antibodies or other ligands (14,54-56). Samples are typically incubated with antibody-coated magnetic beads, after which a magnetic field is used to selectively recover the captured vesicles. IAC requires only a small sample volume, can achieve high purity and enables enrichment of specific EV subpopulations. However, broader application is limited by low yield, variable recovery, relatively high cost and poor scalability. Inefficient release of vesicles from magnetic beads further reduces overall recovery and consequently, IAC is often combined with other methods to improve purity. For example, combining IAC with UC may help distinguish target vesicles from other EVs of similar size and density.

Polymer-based precipitation reduces exosome solubility through macromolecular crowding. Commonly used polymers include polyethylene glycol, dextran sulfate and dextran acetate. Protocols commonly involve incubation for 12-16 h at 4˚C, followed by low-speed centrifugation (10,000 x g for 30 min) to recover a crude pellet. This approach is simple, cost-effective and requires only basic laboratory equipment. However, purity remains a major issue as non-exosomal contaminants and residual polymers may coprecipitate with vesicles (57-59). Subsequent SEC or density-gradient purification can improve sample purity (60).

Microfluidic-based exosome isolation

Microfluidic techniques have emerged as promising approaches for exosome isolation. These platforms enable multiparametric EV separation by integrating physical and biological discrimination mechanisms, such as size exclusion, acoustic focusing and dielectrophoresis, within microfabricated devices. Emerging platforms, including immunomicrofluidic chips (including ExoChip and ExoSearch), acoustofluidic devices and other field-driven systems, can provide rapid processing of small sample volumes and integrate isolation with downstream analysis. Despite their potential, widespread adoption is limited by technical complexity, manufacturing costs and a lack of operational standardization. Recent advances in mass-producible thermoplastic devices, automated detection and data-analysis workflows may help overcome these scalability limitations (46,52).

In summary, each available exosome-isolation method exhibits distinct advantages and limitations and no single approach fully meets the requirements of both experimental and clinical applications. Multimodal workflows, such as UC-SEC-IAC cascades, may improve purity. Future efforts should aim to prioritize standardized protocols with interlaboratory reproducibility, cost-effective affinity ligands such as DNA aptamers or peptide mimetics and point-of-care microfluidic systems with integrated on-chip characterization. Progress in these areas will facilitate clinical translation, particularly for early disease detection and precision therapy.

Characterization and quality control of EVs

Rigorous EV characterization is key as isolation methods based on size, density, precipitation or affinity may co-isolate lipoproteins, protein aggregates, ribonucleoprotein complexes and other non-vesicular particles. No single analytical method is sufficient to establish the identity and purity of an EV preparation. Complementary approaches should therefore be used to assess particle morphology, size distribution, concentration, molecular composition and potential contaminants (19,20).

Transmission electron microscopy or cryo-electron microscopy can be used to examine particle morphology, whereas nanoparticle-tracking analysis, tunable resistive-pulse sensing or associated techniques can determine particle size and concentration. Protein characterization should include EV-associated proteins from different cellular compartments, such as CD9, CD63, CD81, TSG101, ALIX and selected cytosolic proteins. Potential contaminants and non-EV components, including albumin, apolipoproteins and proteins associated with intracellular organelles, should also be evaluated when appropriate (20,61).

Preanalytical variables are particularly important in clinical studies. Serum and plasma may yield different vesicle and contaminant profiles, whereas platelet activation, hemolysis, delayed sample processing, storage duration and repeated freeze-thaw cycles can alter EV abundance and cargo measurements. Accordingly, clinically oriented HCC studies should aim to report sample collection, processing, storage, isolation, characterization and normalization procedures in sufficient detail to permit interstudy comparisons and independent replication (62).

3. Functional heterogeneity of exosome-mediated communication in HCC

Diverse liver cell populations secrete exosomes, including parenchymal cells, hepatocytes and nonparenchymal cells. The latter group includes immune cells (such as T-cells, B-cells, dendritic cells, natural killer cells and macrophages), hepatic stellate cells (HSCs), adipocytes, mesenchymal stem cells (MSCs) and fibroblasts. Hepatocyte-derived exosomes exert paracrine effects by facilitating communication with neighboring hepatocytes and nonparenchymal cells and thereby contribute to hepatic regeneration and tissue repair (63). Nonparenchymal cells are also active sources of EVs. For example, HSCs (64), hepatic sinusoidal endothelial cells (65) and cholangiocytes (66) generate functionally distinct vesicles that participate in pathophysiological processes during liver disease. HCC cells release oncogenic exosomes that establish tumor-stroma crosstalk and remodel the TME, thereby promoting tumor-cell proliferation, angiogenesis and metastatic dissemination (17). This vesicle-mediated communication network not only contributes to hepatocarcinogenesis but also creates clinical opportunities for liquid-biopsy biomarkers and engineered therapeutic nanocarriers (67).

Collectively, hepatic exosomes act as dual mediators, whereby they support physiological homeostasis and tissue repair, but also drive disease progression when dysregulated. This context-dependent activity highlights their potential as theranostic targets in HCC (Fig. 3). The biological effects of exosomes in HCC are determined not simply by their presence, but by heterogeneity in cellular origin, cargo enrichment, surface markers, recipient-cell tropism and disease stage-dependent release. This heterogeneity aids in explaining the opposing effects exerted by HCC-associated exosomes, ranging from tumor suppression and immune activation to angiogenesis, metastasis, immune escape and therapeutic resistance.

Figure 3 Exosomal cargo affects HCC progression. HCC, hepatocellular carcinoma; circRNA, circular RNA; miRNA, microRNA; lncRNA, long non-coding RNA; TME, tumor microenvironment; HSC, hepatic stellate cell. Figure created with FigDraw (https://www.figdraw.com).

Effects of exosomal RNAs on HCC. Effects of exosomal miRNAs on HCC

miRNAs are evolutionarily conserved small noncoding RNAs that regulate gene expression post-transcriptionally by binding to complementary sequences in target mRNAs and are estimated to influence >60% of human protein-coding genes (68,69). In HCC, miRNAs can either promote tumor initiation and progression through pathway dysregulation or exert therapeutic effects by restoring regulatory networks (68,69). Experimental evidence has demonstrated cell type-specific changes in exosomal miRNAs during HCC progression. For example, activated HSC-derived exosomes with reduced miR-148a-3p levels were shown to promote HCC progression by activating the integrin-α 5 (ITGA5)/PI3K/Akt pathway (70). The phospholipid bilayer protects exosomal miRNAs from enzymatic degradation, supporting their potential use as biomarkers. A number of studies have evaluated serum exosomal miRNAs for HCC diagnosis and prognosis (71,72). Cho et al (73) reported that serum exosomal miR-10b-5p showed potential for early-stage HCC detection, achieving an area under the curve (AUC) of 0.934, with a sensitivity of 90.7% and a specificity of 75.0%, whereas serum exosomal miR-215-5p was associated with disease-free survival and may serve as a prognostic indicator. Rui et al (74) profiled serum exosomal miRNAs from 124 patients with HCC and 46 non-tumor controls and identified distinct miRNA base-distribution patterns between HCC and control groups. A three-miRNA signature comprising miR-122-5p, let-7d-5p and miR-425-5p demonstrated diagnostic potential, particularly for low-stage HCC detection. However, external validation and standardized analytical pipelines remain necessary before clinical translation.

HCC remains a substantial global health burden. According to global estimates, ~905,700 new liver cancer cases and 830,200 liver cancer-related deaths occurred worldwide in 2020. The global burden of primary liver cancer is projected to increase substantially by 2040, with the annual number of new cases expected to reach ~1.4 million and annual deaths ~1.3 million, representing increases of 55.0 and 56.4%, respectively, compared with 2020 levels (75). Clinical management is complicated by frequent recurrence, metastatic dissemination and drug resistance. As summarized in Table I, exosomal miRNAs from diverse cellular sources regulate immune evasion, EMT, metastasis, angiogenesis and chemoresistance (76-85). These findings also suggest that selected exosomal miRNAs could help overcome therapeutic resistance. For example, miR-199a-3p is carried by exosomes derived from adipose-derived MSCs (68). mTOR signaling is a key regulator involved in chemoresistance in HCC. Exosomal miR-199a-3p suppresses mTOR signaling by reducing mTOR expression and decreasing the phosphorylation levels of downstream effectors, including 4E-binding protein 1 (4E-BP1) and ribosomal protein S6 kinase β-1 (S6K1). In preclinical HCC models, miR-199a-3p-enriched exosomes increased the sensitivity of HCC cells to doxorubicin treatment, suggesting their potential application in overcoming chemotherapy resistance (68).

Table I Functional effects of exosomal miRNAs in HCC.

Table I

Functional effects of exosomal miRNAs in HCC.

First author, year	Donor cell/EV source	Exosomal miRNA/s	Recipient cell/s	Biological effect in HCC	Proposed mechanism	(Refs.)
Kim et al, 2021	Huh-7 cells	miR-125b	SK-HEP-1 and SNU449 cells	Suppresses EMT and metastatic progression	Exosomal miR-125b targets SMAD2 and inhibits MMP expression and TGF-β1/SMAD signaling	(18)
Lou et al, 2020	AMSCs	miR-199a-3p	PLC/PRF/5 cells	Enhances doxorubicin sensitivity	AMSC-derived exosomal miR-199a-3p inhibits mTOR signaling and reduces phosphorylation of 4E-BP1 and S6K1, thereby increasing HCC cell sensitivity to doxorubicin	(68)
Zhang et al, 2022	HSCs	miR-148a-3p	PLC, HCCLM3 and SMMC-7721 cells	Suppresses HCC progression as depletion from activated HSC-derived EVs favors tumor progression	Restored exosomal miR-148a-3p targets ITGA5 and inhibits PI3K/Akt signaling, thereby reducing HCC cell proliferation, migration and invasion	(70)
Ding et al, 2023	BMSCs	miR-374c-5p	HepG2, BEL-7402, SMMC-7721, HCCLM3 and SK-Hep-1 cells	Suppresses proliferation, migration, invasion and EMT	BMSC-derived exosomal miR-374c-5p targets LIMK1 and inhibits Wnt/β-catenin signaling	(76)
Dong et al, 2021	HCCLM3, MHCC97L and PLC/PRF/5 cells	miR-3682-3p	HUVECs	Suppresses endothelial migration, invasion and tube formation	Exosomal miR-3682-3p targets ANGPT1 and inhibits Ras-MEK1/2-ERK1/2 signaling, with depletion in EVs from highly metastatic HCC cells favoring angiogenesis	(77)
Hu et al, 2024	HCC cells	miR-21-5p	Macrophages	Promotes HCC progression	Exosomal miR-21-5p regulates the SP1/XBP1 axis and promotes M2 macrophage polarization	(78)
Nakano et al, 2023	HCC cells (dete ctedin circulating EV fractions)	miR-4669	HCC cells and macrophages	Promotes tumor aggressiveness, sorafenib resistance and an immunosuppressive microenvironment	miR-4669 increases SIRT1 and lncRNA MVIH expression in HCC cells, with associated GAPDH secretion contributing to M2 macrophage polarization	(79)
Qi et al, 2022	CAFs	miR-20a-5p	Huh-7 cells	Enhances proliferation, migration and invasion	CAF-derived exosomal miR-20a-5p suppresses the tumor suppressor LIMA1, stabilizes BMI1 and activates Wnt/β-catenin signaling	(80)
Wang et al, 2022	M1 macrophages	miR-628-5p	Huh7 and MHCC97H cells	Suppresses HCC cell proliferation and malignant progression	Exosomal miR-628-5p targets METTL14 and inhibits the circFUT8/CHMP4B regulatory axis	(81)
Wang et al, 2021	HepG2 and SMMC-7721 cells	miR-1290	HUVECs	Promotes angiogenesis	Exosomal miR-1290 suppresses SMEK1, thereby relieving inhibition of VEGFR2 phosphorylation and enhancing endothelial angiogenic activity	(82)
Xu et al, 2023	HCC cells	miR-200b-3p	Macrophages	Promotes HCC cell proliferation and metastasis	Exosomal miR-200b-3p downregulates ZEB1, increases IL-4, activates JAK/STAT signaling in M2 macrophages and increases PIM1 and VEGFA expression	(83)
Yugawa et al, 2021	CAFs	miR-150-3p	Huh-7 and Hep3 B-cells	Suppresses HCC cell migration and invasion	Exosomal transfer of miR-150-3p suppresses HCC cell motility (the direct downstream target was not fully established in the original study)	(84)
Zongqiang et al, 2022	Huh-7 cells	miR-452-5p	THP-1-derived macrophages	Promotes HCC cell migration, invasion and tumorigenesis	Exosomal miR-452-5p targets TIMP3 and induces M2 macrophage polarization	(85)
Han et al, 2023	BM-EVs	miR-3190-5p	Primary HCC cells	Promotes primary-tumor progression and further metastatic dissemination	BM-EV-derived miR-3190-5p downregulates ALKBH5 and alters gene expression through m6A-dependent and m6A-independent mechanisms	(86)
Yang et al, 2020	Highly metastatic 97 hm and Huhm cells	miR-92a-3p	97 h and Huh-7 cells	Promotes EMT, migration and metastasis	Exosomal miR-92a-3p suppresses PTEN and activates Akt/Snail signaling, transferring a more aggressive phenotype to recipient HCC cells	(87)
Yokota et al, 2021	Highly metastatic Huh-7M cells	miR-638, miR-663a, miR-3648 and miR-4258	HUVECs	Increases endothelial permeability and promotes intrahepatic metastasis	Transferred miRNAs downregulate VE-cadherin and ZO-1 in endothelial cells, disrupting the vascular barrier and facilitating metastatic dissemination	(93)

[i] HCC, hepatocellular carcinoma; miR/miRNA, microRNA; AMSC, adipose-derived mesenchymal stem cell; 4E-BP1, 4E-binding protein 1; S6K1, ribosomal protein S6 kinase B1; VE-cadherin, vascular endothelial-cadherin; ZO-1, zonula occludens-1; METTL14, methyltransferase 14; circ, circular RNA; FUT8, fucosyltransferase 8; CHMP4B, charged multivesicular body protein 4B; HSC, hepatic stellate cell; ITGA5, integrin subunit α 5; EV, extracellular vesicle; CAF, cancer-associated fibroblasts; SIRT1, sirtuin 1; lncRNA, long non-coding RNA; MVIH, lncRNA associated with microvascular invasion in hepatocellular carcinoma; SP1, specificity protein 1; XBP1, X-box binding protein 1; HUVEC, human umbilical vein endothelial cell; ANGPT1, angiopoietin 1; BMSC, bone marrow mesenchymal stem cell; EMT, epithelial-mesenchymal transition; LIMK1, LIM domain kinase 1; BM-EV, bone metastasis-derived EV; ALKBH5, AlkB homolog 5; LIMA1, LIM domain and actin binding 1; BMI1, BMI1 proto-oncogene; TIMP3, TIMP metallopeptidase inhibitor 3; ZEB1, zinc finger E-Box binding homeobox 1; PIM1, Pim-1 proto-oncogene; SMEK1, suppressor of MEK1 homolog 1.

Extrahepatic metastasis refers to the dissemination of HCC cells beyond the liver to distant organs and represents a characteristic feature of advanced HCC. The lungs and bones are among the most common metastatic sites. At initial diagnosis, 16.1-38.5% of patients with HCC have bone metastases, with 11.7% developing bone metastases after curative resection. The prognosis of patients with bone metastases is particularly poor, with a median overall survival time of 4.6 months (75). Recent studies regarding intercellular communication have provided mechanistic insights into this metastatic process. Han et al (86) identified a bidirectional bone-liver communication axis mediated by bone metastasis-derived extracellular vesicles (BM-EVs). These EVs entered the systemic circulation and returned to primary hepatic tumors, where they promoted tumor progression and additional metastatic lesion formation through delivery of miR-3190-5p to primary HCC cells. The investigators also developed an HCC-targeted liposomal platform containing miR-3190-5p antagonists. In preclinical models, this approach attenuated the prometastatic effects of BM-EVs and slowed disease progression in vivo. This study provided mechanistic evidence for a bone-liver communication axis and suggested a potential targeted strategy (86). Similarly, Yang et al (87) demonstrated that highly metastatic HCC cells transferred metastatic capacity to less metastatic cells through exosomal miR-92a-3p, thereby regulating the PTEN/Akt pathway in recipient cells.

EMT is a key process in tumor invasion and metastasis. During EMT, epithelial tumor cells progressively lose adhesive properties and acquire mesenchymal features that increase motility and invasiveness. The process involves numerous coordinated structural and molecular changes including: i) Disassembly of epithelial junctional complexes; ii) transition from apicobasal to front-rear polarity; iii) dynamic reorganization of the actin cytoskeleton and associated morphological remodeling; iv) downregulation of epithelial markers, such as epithelial-cadherin) and zonula occludens-1 (ZO-1), with concomitant upregulation of mesenchymal markers, including vimentin, fibronectin, N-cadherin and β1/β3 integrins; v) increased membrane protrusion and motility; and vi) acquisition of extracellular matrix-degrading capacity, which facilitates invasive dissemination (88-91). The central role of EMT in HCC metastasis has prompted investigation of exosomal miRNA-mediated regulatory mechanisms and their potential diagnostic and therapeutic applications (39,92). Yokota et al (93) reported that miR-638, miR-663a, miR-3648 and miR-4258 in exosomes secreted by highly metastatic HuH-7M cells disrupted the endothelial barrier by suppressing vascular endothelial-cadherin and ZO-1, thereby increasing vascular permeability and promoting intrahepatic metastasis. Disruption of exosomal miR-23a-3p-mediated intercellular crosstalk has also been reported to inhibit Smad signaling and suppress EMT in HCC (94).

Collectively, exosomes are important mediators of tumor-stroma communication and coordinate oncogenesis, metastatic-niche formation and therapeutic resistance. Their stability in biofluids enables noninvasive isolation and molecular profiling, including at early disease stages (55). These properties support the development of exosome-based diagnostic and therapeutic strategies for precision management of HCC.

These findings demonstrate that exosomal miRNAs do not exhibit uniformly oncogenic or tumor-suppressive effects in HCC. Their functions depend on the cellular source and molecular state of the donor cells, the identity of the recipient cells and the disease context (95). For example, miRNAs transferred from highly metastatic tumor cells or stromal cells frequently promote EMT, angiogenesis, immune escape or drug resistance, whereas selected miRNAs derived from MSCs or other cell types may suppress oncogenic signaling (87). Therefore, evaluating an exosomal miRNA without considering its donor cell origin and recipient cell context may obscure its biological and clinical importance.

Effects of exosomal circRNAs on HCC. circRNAs are primarily generated by back-splicing of precursor mRNAs and are covalently closed, single-stranded RNA molecules that lack 5'-3' polarity and polyadenylated tails. Their subcellular localization and sequence-specific interactions with DNA, RNA and proteins confer diverse regulatory functions, including mRNA stabilization, transcriptional and post-transcriptional regulation, modulation of translation and interference with signaling pathways. A number of circRNAs can also serve as templates for functional peptides under specific pathophysiological conditions. Mechanistically, circRNAs regulate downstream processes by acting as miRNA or protein sponges, encoding peptides and forming RNA-protein complexes (96-99).

The competing endogenous RNA (ceRNA) model proposes that non-coding RNAs can compete for binding to shared miRNAs and thereby regulate gene expression post-transcriptionally (100,101). Accordingly, circRNAs, long noncoding RNAs (lncRNAs) and other RNA species can function as miRNA sponges through shared miRNA-response elements and influence expression, tumorigenesis and cancer progression. Growing evidence has indicated that ceRNA networks regulate HCC cell proliferation, metastasis, EMT and chemoresistance (100,101). For example, macrophage-derived exosomal hsa_circ_0004658 has been shown to suppress HCC progression by sequestering miR-499b-5p and restoring junctional adhesion molecule 3 (JAM3) expression (102). Conversely, circTGFBR2 (hsa_circ_0005224) in exosomes from starvation-stressed hepatocytes has been shown to promote HCC progression by sequestering miR-205-5p and activating autophagy-related protein 5 (ATG5)-dependent protective autophagy (103).

Exosomal circRNAs can also exert opposing effects on HCC metastasis. For example, circ_0072088 inhibits metastasis through the miR-375/MMP16 axis (104), whereas circ-MMP2 (hsa_circ_0039411) promotes metastasis by sequestering miR-136-5p and upregulating MMP2(105). Zhu et al (106) reported that exosomal hsa_circ_0004277 promoted EMT by binding human antigen R, destabilizing ZO-1 mRNA and disrupting epithelial integrity. Collectively, these findings identify exosomal circRNAs as important regulators of ceRNA networks in HCC and as potential diagnostic biomarkers or therapeutic targets.

circRNAs also contribute to the immunosuppressive TME that supports HCC tumor growth, invasion and metastasis. Despite immunotherapy having improved HCC treatment (107), certain patients do not benefit from current regimens or experience immune-associated adverse events, in part due to an immunosuppressive TME (108,109). Hu et al (17) demonstrated that HCC cell-derived exosomal circCCAR1 promoted CD8+ T-cell dysfunction and resistance to anti-programmed cell death protein 1 (PD-1) therapy through STAT3-associated signaling. Hypoxia and tumor necrosis can trigger marked extracellular ATP (eATP) release within the TME. Through the ATP-adenosine pathway, eATP is hydrolyzed to immunosuppressive adenosine, which impairs natural killer cell and T-cell function and accelerates tumor progression (110). HCC cell-derived exosomal circ-transmembrane protein (TMEM)-181 has been reported to upregulate CD39 expression in macrophages (111). This macrophage-HCC cell crosstalk activated the ATP-adenosine axis, increased adenosine concentrations in the TME, suppressed CD8+ T-cell activity and conferred resistance to anti-PD-1 therapy. In addition, the Hippo pathway exhibits important roles in tumor suppression and immune regulation. HCC-derived exosomal circ- p21-activated kinase 1 (PAK1) has been shown to promote HCC progression by inactivating Hippo signaling and thereby enhancing tumor-cell proliferation, migration and invasion. Furthermore, circPAK1-enriched exosomes may mediate the intercellular transfer of lenvatinib resistance among HCC cells (112).

In summary, exosomal circRNAs regulate key oncogenic processes in HCC through ceRNA networks and modulation of the immunosuppressive TME. Their central mechanistic roles support further evaluation as diagnostic biomarkers and therapeutic targets, including targets for overcoming resistance to immunotherapy.

Evidence also illustrates the functional heterogeneity of exosomal circRNAs. circRNAs, including those that regulate overlapping molecular targets, exert different effects depending on their cellular origin and microenvironmental context. For example, macrophage-derived circRNAs may exert tumor-suppressive effects, whereas HCC cell-derived circRNAs often promote autophagy, EMT, immune escape or treatment resistance (102). However, the majority of reported mechanisms have been established in individual cell-line or animal models, and independent validation across HCC etiologies and clinical cohorts remains limited due to the lack of large-scale prospective studies, standardized detection methods and multicenter clinical validation (113).

Effects of exosomal lncRNAs on HCC. lncRNAs are non-protein-coding transcripts longer than 200 nucleotides that regulate gene expression through epigenetic mechanisms, chromatin remodeling and post-transcriptional processes. The function of each lncRNA depends partly on subcellular localization, with nuclear lncRNAs regulating transcriptional programs and RNA processing and cytoplasmic lncRNAs influencing mRNA stability, translation and post-translational modification. A number of lncRNAs also function as ceRNAs by sequestering miRNAs, thereby altering the expression of miRNA-regulated oncogenes (114-116). Accumulating evidence has established lncRNAs as important regulators of carcinogenesis through dynamic interactions with diverse biomolecules (115-117). Systematic characterization of their upstream regulators and downstream effectors is therefore important for elucidating oncogenic mechanisms and developing targeted therapies.

lncRNAs can regulate tumor behavior and metabolism by sequestering miRNAs. For example, Xu et al (118) demonstrated that exosomal lnc-M2 macrophage polarization-associated (MMPA) was transferred from tumor-associated macrophages (TAMs) to HCC cells, where it sequestered miR-548s, upregulated aldehyde dehydrogenase 1 family member A3 (ALDH1A3), reprogrammed glucose metabolism and promoted tumorigenesis. HCC cell-derived exosomal zinc finger protein, FOG family member 2 (ZFPM2)- antisense RNA 1 (AS1) has also been reported to sequester miR-18b-5p, thereby increasing pyruvate kinase muscle expression and promoting tumor stemness and glycolysis (119). In addition, cancer-associated fibroblast (CAF)-derived exosomal taurine up-regulated 1 (TUG1) acted as a ceRNA for miR-524-5p and increased SIX homeobox 1 (SIX1) expression. SIX1 subsequently activated the glycolytic enzymes hexokinase 2 (HK2) and lactate dehydrogenase A (LDHA), enhancing HCC cell migration, invasion and metabolic reprogramming (120).

HCC is a hypervascular malignancy that relies on neovascularization for tumor growth and metastatic dissemination. Exosomal lncRNAs can promote HCC progression by activating proangiogenic signaling. For example, HCC-derived exosomal small nucleolar RNA host gene 16 (SNHG16) was reported to act as a ceRNA for miR-4500, increase polypeptide N-acetylgalactosaminyltransferase 1 (GALNT1) expression and activate the PI3K/Akt/mTOR pathway (121). You et al (121) reported that HCC-derived exosomal long intergenic non-protein coding RNA (LINC)-00161 promoted angiogenesis and metastasis by sequestering miR-590-3p and activating Rho-associated protein kinase (ROCK) signaling. Elevated circulating LINC00161 may also have prognostic relevance. Exosomal lncRNAs contribute not only to angiogenesis but also to other aspects of HCC pathogenesis (122). For example, m6A-modified lncRNA m6A-modified lncRNA MAPKAPK5-AS1 (MAAS) was found to be enriched in exosomes from M2 macrophages in hepatitis B (HBV)-associated HCC. These exosomes transferred MAAS to HCC cells, where it stabilized c-Myc mRNA and accelerated tumor-cell proliferation (122).

Collectively, these findings identify exosomal lncRNAs as important regulators of HCC initiation and progression through ceRNA-mediated regulation, epigenetic reprogramming, metabolic remodeling and intercellular communication. Further elucidation of these pathways may support the development of improved diagnostic strategies and targeted therapies for HCC (122).

The functions of exosomal lncRNAs are similarly context-dependent. TAM-, CAF- and HCC cell-derived lncRNAs can converge on metabolic reprogramming, angiogenesis, stemness or proliferative signaling through distinct regulatory networks. Despite this, the current literature is dominated by mechanistic studies regarding individual lncRNAs and relatively few candidates have been independently replicated or prospectively evaluated in patient cohorts. The primary reasons include limited external validation, heterogeneous sample processing and insufficient clinical-scale studies. These limitations should be addressed before exosomal lncRNAs are considered clinically actionable biomarkers or therapeutic targets (122). Representative exosomal circRNAs and lncRNAs, together with their cellular sources, major targets or signaling pathways and functions in HCC, are summarized in Table II.

Table II Functional roles of exosomal circRNAs and lncRNAs in HCC.

Table II

Functional roles of exosomal circRNAs and lncRNAs in HCC.

First author, year	Class	Exosomal molecule	Donor cell/EV source	Major target or signaling pathway	Function in HCC	(Refs.)
Hu et al, 2023	circRNA	circCCAR1	HCC cells	CD8+ T-cells and STAT3-associated signaling	Induces CD8+ T-cell dysfunction and contributes to anti-PD-1 resistance	(17)
Zhang et al, 2022	circRNA	hsa_circ_0004658	Macrophages with RBPJ overexpression	miR-499b-5p/JAM3	Suppresses HCC cell proliferation and promotes apoptosis	(102)
Wang et al, 2023	circRNA	circTGFBR2 (hsa_circ_0005224)	Starvation-stressed hepatocytes	miR-205-5p/ATG5-dependent autophagy	Promotes stress adaptation and HCC progression under nutrient deprivation	(103)
Lin et al, 2021	circRNA	circ_0072088	HCC cells	miR-375/MMP16	Suppresses HCC cell migration and invasion	(104)
Liu et al, 2020	circRNA	circ-MMP2 (hsa_circ_0039411)	HCC cells	miR-136-5p/MMP2	Promotes migration, invasion and metastatic progression	(105)
Zhu et al, 2020	circRNA	hsa_circ_0004277	HCC cells	HuR binding and destabilization of ZO-1 mRNA	Promotes EMT and a malignant phenotype by disrupting epithelial integrity	(106)
Lu et al, 2021	circRNA	circTMEM181	HCC cells	Macrophage CD39 and ATP-adenosine axis	Suppresses CD8+ T-cell activity and contributes to anti-PD-1 resistance	(111)
Hao et al, 2022	circRNA	circPAK1	HCC cells	Hippo/YAP signaling and 14-3-3ζ interaction	Promotes HCC progression and may mediate intercellular transfer of lenvatinib resistance	(112)
Xu et al, 2022	lncRNA	lncMMPA	Tumor-associated macrophages	miR-548s/ALDH1A3	Reprograms glucose metabolism and promotes HCC tumorigenesis	(118)
Ji et al, 2023	lncRNA	ZFPM2-AS1	HCC cells	miR-18b-5p/PKM	Promotes stemness, glycolysis, metastasis and macrophage polarization	(119)
Lu et al, 2022	lncRNA	TUG1	Cancer-associated fibroblasts	miR-524-5p/SIX1	Promotes HCC cell migration, invasion and glycolytic reprogramming	(120)
You et al, 2021	lncRNA	LINC00161	HCC cells; detected in serum EV fractions	miR-590-3p/ROCK axis	Promotes angiogenesis and metastasis; elevated circulating levels have potential prognostic relevance	(121)
Li et al, 2021	lncRNA	SNHG16	HCC cells	miR-4500/GALNT1/PI3K-Akt-mTOR	Promotes angiogenesis	(140)
Tao et al, 2022	lncRNA	MAAS	M2 macrophages in HBV-associated HCC	Stabilization of c-Myc mRNA	Accelerates HCC cell proliferation	(141)

[i] circRNA/circ, circular RNA; lncRNA, long non-coding RNA; HCC, hepatocellular carcinoma; RBPJ, recombination signal binding protein for immunoglobulin κ J region; JAM3, junctional adhesion molecule 3; PD-1, programmed cell death 1; ATG5, autophagy related 5; ZO-1, zonula occludens-1; EMT, epithelial-mesenchymal transition; TMEM, transmembrane protein; PAK1, p21 (RAC1) activated kinase 1; YAP, Yes-associated protein; ALDH1A3, aldehyde dehydrogenase 1 family member A3; ZFPM2-AS1, ZFPM2 antisense RNA 1; PKM, pyruvate kinase muscle; TUG1, taurine up-regulated 1; SIX1, SIX homeobox 1; SNHG16, small nucleolar RNA host gene 16; GALNT1, polypeptide N-acetylgalactosaminyltransferase 1; EV, extracellular vesicle; ROCK, Rho-associated protein kinase; HBV, hepatitis B; MAAS, m6A-modified lncRNA MAPKAPK5-AS1.

Effects of exosomal proteins on HCC. Exosomal proteins are key mediators of intercellular communication in HCC (123). Numerous studies have evaluated serum exosomal proteins as candidate biomarkers for HCC diagnosis and prognosis (124,125). Mechanistically, HCC-derived exosomes delivered smoothened protein (SMO) to HSCs, whereby SMO activated Hedgehog signaling. This activation increased GLI family zinc finger 1 (Gli1)-mediated transcription of the miRNA let-7b host gene (MIRLET7BHG), which subsequently enhanced SMO expression by competitively binding miR-330-5p, thereby promoting HSC activation, hepatic fibrosis and HCC progression (126). Li et al (127) described an additional exosome-mediated regulatory mechanism involving general transcription factor IIH subunit 2 (GTF2H2). GTF2H2 was packaged into exosomes and transferred from Huh7 cells to human umbilical vein endothelial cells, where it suppressed endothelial cell proliferation and migration, suggesting that exosomal GTF2H2 may inhibit HCC angiogenesis.

High-mobility group box 1 (HMGB1) is an evolutionarily conserved damage-associated molecular pattern protein that is widely expressed in eukaryotic cells. HMGB1-enriched exosomes derived from HCC cells have been reported to promote regulatory B-cell accumulation through the HMGB1-TLR2/4-MAPK axis. These immunosuppressive B-cells secreted IL-10 and impaired CD8+ T-cell cytotoxicity, thereby establishing an immune-tolerant microenvironment (128).

Emerging evidence has indicated that metastatic potential can be transferred horizontally between HCC cell subpopulations not only through exosomal miRNAs (87), but also through exosomal proteins. Sun et al (129) showed that highly metastatic HCC cells packaged S100 calcium-binding protein A4 (S100A4) into exosomes that were subsequently taken up by less metastatic cells. Exosomal S100A4 induced metastatic reprogramming through STAT3 signaling, increased osteopontin (OPN) expression and enhanced the metastatic capacity of recipient cells. Combined measurement of plasma exosomal S100A4 and soluble OPN also exhibited potential in postoperative prognostic assessment. An additional study demonstrated that carboxypeptidase E, a bioactive protein enriched in exosomes from highly metastatic HCC cells, promoted the proliferation and invasion of less metastatic HCC cells (130).

In summary, exosomal proteins, whether located within the vesicle lumen or anchored to the membrane, are functionally important cargo molecules with potential clinical applications (131). They can reflect donor-cell pathophysiology and actively regulate HCC progression through intercellular communication. By influencing processes, from tumor initiation to metastatic dissemination, exosomal proteins may serve as biomarkers for HCC diagnosis and prognosis and as targets for therapeutic development (132).

These findings indicate that exosomal proteins may promote or restrain HCC progression depending on their molecular identity, cellular source and recipient-cell target. However, evidence for numerous candidate proteins remains preclinical and proteins initially identified in HCC cell-derived exosomes are not necessarily equivalent to proteins measured in unfractionated patient serum. Therefore, future studies should aim to distinguish proteins directly quantified in circulating EVs from soluble serum proteins identified through exosome-guided discovery approaches (133).

4. Summary and outlook

Current knowledge regarding EV biogenesis, isolation, characterization and cargo-mediated communication in HCC was summarized in the present review. Particular attention was given to the context-dependent functions of EV subpopulations and cargo molecules, as well as the methodological and clinical factors that currently limit their translation into reproducible biomarkers and therapeutic tools.

Functional heterogeneity of HCC-associated exosomes: Sources, subpopulations and pathological consequences

Despite exosomes often being discussed as a single class of vesicles, HCC-associated EVs are heterogeneous in biogenesis, size, density, surface composition, molecular cargo, cellular origin, recipient-cell tropism and biological activity. This multidimensional heterogeneity provides a framework for understanding the manner in which EVs can exert opposing effects in HCC, including tumor suppression or promotion, immune activation or escape, angiogenesis, metastasis and therapeutic resistance (134).

A specific source of heterogeneity may be the existence of physically and molecularly distinct extracellular nanoparticle subpopulations (135). Using asymmetric-flow field-flow fractionation, studies in non-HCC cancer-cell models operationally separated extracellular nanoparticles into large exosome vesicles, small exosome vesicles and smaller nonmembranous particles termed exomeres. These fractions differed in molecular composition, biophysical properties and organ biodistribution (136,137).

Cellular origin is another key determinant of EV function in the HCC microenvironment. HCC cells, hepatocytes, HSCs, TAMs, CAFs, endothelial cells, MSCs and lymphocytes can release vesicles with distinct cargo profiles and recipient-cell preferences (138). For example, HCC cell-derived vesicles may promote T-cell exhaustion, stromal activation, angiogenesis, metastatic reprogramming or drug resistance. By contrast, vesicles enriched in selected tumor-suppressive miRNAs may inhibit EMT and metastatic progression. Thus, the biological effect of an EV population is determined not only by its cargo, but also by the state of the donor cell and the molecular characteristics of the recipient cell (17,139).

To further clarify cell type-dependent functional heterogeneity, HCC-associated EVs should be considered as donor cell-defined communication units rather than as a uniform vesicle population. Despite the majority of HCC studies having characterized EV preparations by size distribution and canonical EV-associated markers rather than by comprehensive single-vesicle structural profiling, the available experimental evidence indicates that EV cargo composition, target-cell preference and biological consequences differ markedly according to the cellular source (134-136).

HCC cell-derived EVs represent one of the best-characterized vesicle populations in the HCC microenvironment. These vesicles can contain oncogenic or immune-regulatory miRNAs, circRNAs, lncRNAs and proteins, with their targets including tumor cells, endothelial cells, macrophages, HSCs and cytotoxic T-cells. For example, HCC cell-derived exosomal circCCAR1 was shown to induce CD8+ T-cell dysfunction through STAT3-associated signaling and to contribute to resistance to anti-PD-1 therapy (17). Similarly, HCC-derived exosomal circTMEM181 was shown to upregulate CD39 expression in macrophages, activate the ATP-adenosine axis, increase immunosuppressive adenosine production and suppress CD8+ T-cell activity, thereby promoting resistance to anti-PD-1 therapy (111).

HCC-derived exosomal circPAK1 was shown to promote tumor progression through Hippo/Yes-associated protein (YAP) pathway inactivation and was implicated in the intercellular transfer of lenvatinib resistance (112). In addition, HCC cell-derived exosomal SNHG16 activated the miR-4500/GALNT1/PI3K-Akt-mTOR axis to promote angiogenesis (140), whereas exosomal LINC00161 enhanced angiogenesis and metastasis through the miR-590-3p/ROCK axis (140). Protein cargo also contributes to this heterogeneity. HCC-derived exosomal HMGB1 was shown to promote regulatory B-cell accumulation through the HMGB1-TLR2/4-MAPK axis, leading to IL-10 secretion and impaired CD8+ T-cell cytotoxicity (128). Highly metastatic HCC cells also transferred S100A4-containing exosomes to less metastatic HCC cells, activating STAT3 signaling, increasing OPN expression and enhancing metastatic capacity (129). These findings indicate that HCC cell-derived EVs can coordinate immune escape, angiogenesis, metastatic reprogramming and therapeutic resistance.

Non-malignant hepatocytes and stress-adapted hepatocytes may also generate functionally distinct vesicles. Under nutrient-deprivation conditions, starvation-stressed hepatocytes released exosomes enriched in circTGFBR2. These vesicles promoted HCC progression by sequestering miR-205-5p and activating ATG5-dependent protective autophagy in recipient HCC cells (103). This illustrates that hepatocyte-derived vesicles are not necessarily homeostatic; under microenvironmental stress, their cargo can support tumor-cell survival and adaptation.

Stromal-cell-derived vesicles provide another layer of heterogeneity. Activated HSC-derived exosomes with reduced miR-148a-3p levels promoted HCC progression through activation of the ITGA5/PI3K/Akt pathway (70). Conversely, HCC-derived exosomal SMO targeted HSCs and activated Hedgehog signaling, thereby increasing Gli1-mediated MIRLET7BHG transcription, enhancing SMO expression through miR-330-5p competition and promoting HSC activation, fibrosis and HCC progression (126). These findings demonstrate reciprocal EV-mediated communication between malignant hepatocytes and HSCs, demonstrating an association between tumor progression and fibrogenic remodeling of the liver microenvironment.

Macrophage- and TAM-derived vesicles may exert either tumor-promoting or tumor-suppressive functions depending on their activation state and cargo. Macrophage-derived exosomal hsa_circ_0004658 suppressed HCC progression by sponging miR-499b-5p and restoring JAM3 expression (102). By contrast, TAM-derived exosomal lncMMPA was transferred to HCC cells, where it sequestered miR-548s, upregulated ALDH1A3, reprogrammed glucose metabolism and promoted tumorigenesis (118). In HBV-associated HCC, M2 macrophage-derived exosomal lncRNA MAAS stabilized c-Myc mRNA and accelerated HCC-cell proliferation (141). Collectively, this shows that macrophage-derived EVs are highly context-dependent: Selected macrophage EV cargo may suppress tumor growth, whereas TAM- or M2 macrophage-derived vesicles may enhance metabolic reprogramming, proliferation and immune escape.

CAF-derived EVs primarily influence tumor-cell metabolism, motility and invasiveness. CAF-derived exosomal TUG1 acted as a ceRNA for miR-524-5p and increased SIX1 expression. SIX1 subsequently activated glycolytic enzymes including HK2 and LDHA, thereby enhancing HCC-cell migration, invasion and glycolytic reprogramming (120). Thus, CAF-derived vesicles provide a mechanism through which stromal fibroblasts can reshape tumor-cell metabolism and promote aggressive behavior.

MSC-derived vesicles can have therapeutic or tumor-suppressive effects depending on their cargo. Exosomes derived from adipose-derived MSCs delivered miR-199a-3p to HCC cells, inhibited mTOR signaling, reduced phosphorylation of 4E-BP1 and S6K1, and increased HCC-cell sensitivity to doxorubicin (68). In addition, exosomal miR-125b inhibited TGF-β1/SMAD2-mediated EMT and suppressed metastatic progression (18). These findings suggest that selected MSC-derived or tumor-suppressive miRNA-enriched vesicles may counteract oncogenic signaling and improve therapeutic response.

EVs derived from metastatic niches may also contribute to systemic disease progression. Bone metastasis-derived EVs entered the circulation and returned to primary hepatic tumors, where they delivered miR-3190-5p to HCC cells and promoted primary-tumor progression and additional metastatic lesion formation (86). In preclinical mouse models, an HCC-targeted liposomal platform carrying miR-3190-5p antagonists attenuated the prometastatic effects of these EVs and slowed disease progression (86). This highlights that EV heterogeneity is influenced not only by primary-tumor and liver microenvironmental cells, but also by distant metastatic niches.

Overall, these studies demonstrate that HCC-associated exosomal heterogeneity is determined by the donor-cell lineage, activation state, stress condition, cargo-sorting program and recipient-cell context. HCC cell-derived vesicles frequently promote immune evasion, angiogenesis, metastasis and drug resistance; HSC- and CAF-derived vesicles remodel fibrotic and metabolic components of the TME; macrophage-derived vesicles can exert either tumor-suppressive or tumor-promoting effects depending on polarization state; and selected MSC-derived vesicles may restore tumor-suppressive signaling or improve drug sensitivity. Therefore, future studies should avoid analyzing bulk EVs alone and should instead aim to integrate donor-cell tracing, vesicle subpopulation profiling, cargo-resolved analysis and functional validation in relevant recipient cells. The cell type-dependent EV cargo profiles, target cells, mechanisms and functional consequences of HCC-associated EVs are summarized in Table III.

Table III Cell type-dependent heterogeneity of HCC-associated EVs.

Table III

Cell type-dependent heterogeneity of HCC-associated EVs.

First author, year	EV-producing cell type/source	Representative EV cargo or profile	Main target cells	Mechanism/pathway	Experimental or functional consequence	(Refs.)
Hu et al, 2023	HCC cells	circCCAR1	CD8+ T-cells	STAT3-associated signaling	Induced CD8+ T-cell dysfunction and contributed to anti-PD-1 resistance	(17)
Kim et al, 2021	Huh-7 cells	miR-125b	SK-HEP-1 and SNU449 cells	SMAD2 targeting, MMP and TGF-β1/SMAD signaling inhibition	Suppressed EMT and metastatic progression	(18)
Lou et al, 2020	Adipose-derived MSCs	miR-199a-3p	PLC/PRF/5 cells	mTOR/4E-BP1/S6K1 inhibition	Increased HCC-cell sensitivity to doxorubicin	(68)
Zhang et al, 2022	HSCs/activated HSC-derived EVs	miR-148a-3p	PLC, HCCLM3 and SMMC-7721 cells	ITGA5/PI3K/Akt pathway	Restored exosomal miR-148a-3p reduced HCC-cell proliferation, migration and invasion, with depletion from activated HSC-derived EVs favoring tumor progression	(70)
Han et al, 2023	Bone metastasis-derived EVs	miR-3190-5p	Primary HCC cells	ALKBH5 downregulation and altered gene expression through m6A-dependent and m6A-independent mechanisms	Promoted primary-tumor progression and further metastatic dissemination	(86)
Zhang et al, 2022	Macrophages with RBPJ overexpression	hsa_circ_0004658	HCC cells	miR-499b-5p/JAM3 axis	Suppressed HCC-cell proliferation and promoted apoptosis	(102)
Wang et al, 2023	Starvation-stressed hepatocytes	circTGFBR2	HCC cells	miR-205-5p/ATG5-dependent protective autophagy	Promoted stress adaptation and HCC progression under nutrient deprivation	(103)
Lu et al, 2021	HCC cells	circTMEM181	Macrophages and CD8+ T-cells	Macrophage CD39 upregulation and ATP-adenosine axis activation	Increased adenosine production, suppressed CD8+ T-cell activity and promoted anti-PD-1 resistance	(111)
Hao et al, 2022	HCC cells	circPAK1	HCC cells	Hippo/YAP signaling and 14-3-3ζ interaction	Promoted HCC progression and mediated intercellular transfer of lenvatinib resistance	(112)
Xu et al, 2022	Tumor-associated macrophages	lncMMPA	HCC cells	miR-548s/ALDH1A3 axis	Reprogrammed glucose metabolism and promoted HCC tumorigenesis	(118)
Lu et al, 2022	Cancer-associated fibroblasts	TUG1	HCC cells	miR-524-5p/SIX1 axis, HK2 and LDHA activation	Enhanced HCC-cell migration, invasion and glycolytic reprogramming	(120)
You et al, 2021	HCC cells, detected in serum EV fractions	LINC00161	Endothelial cells and HCC cells	miR-590-3p/ROCK axis	Promoted angiogenesis and metastasis, elevated circulating levels showed potential prognostic relevance	(121)
Ye et al, 2018	HCC cells	HMGB1 protein	Regulatory B-cells and CD8+ T-cells	HMGB1-TLR2/4-MAPK axis	Induced IL-10-producing regulatory B-cells and impaired CD8+ T-cell cytotoxicity	(128)
Sun et al, 2021	Highly metastatic HCC cells	S100A4 protein	Less metastatic HCC cells	STAT3 activation and OPN upregulation	Transferred metastatic capacity to recipient HCC cells and enhanced metastatic capacity	(129)
Li et al, 2021	HCC cells	SNHG16	Endothelial/tumor-associated cells	miR-4500/GALNT1/PI3K-Akt-mTOR axis	Promoted angiogenesis	(140)
Tao et al, 2022	M2 macrophages in HBV-associated HCC	lncRNA MAAS	HCC cells	Stabilization of c-Myc mRNA	Accelerated HCC-cell proliferation	(141)

[i] HCC, hepatocellular carcinoma; EV, extracellular vesicle; PD-1, programmed cell death 1; circ, circular RNA; CCAR1, cell division cycle and apoptosis regulator 1; TMEM, transmembrane protein; YAP, Yes-associated protein; PAK1, p21 (RAC1) activated kinase 1; EMT, epithelial-mesenchymal transition; SNHG16, small nucleolar RNA host gene 16; GALNT1, polypeptide N-acetylgalactosaminyltransferase 1; ROCK, Rho-associated protein kinase; HBV, hepatitis B; TLR, toll-like receptor; HMGB1, high mobility group box 1; OPN, osteopontin; ATG5, autophagy-related protein 5; HSC, hepatic stellate cell; ITGA5, integrin α-5 protein; ALDH1A3, aldehyde dehydrogenase 1 family member A3; lnc, long non-coding RNA; MMPA, M2 macrophage polarization-associated; JAM3, junctional adhesion molecule 3; RBPJ, recombination signal binding protein for immunoglobulin κ J region; TUG1, taurine up-regulated 1; SIX1, SIX homeobox 1; HK2, hexokinase 2; LDHA, lactate dehydrogenase A; MSC, mesenchymal stem cell; 4E-BP1, 4E-binding protein 1; S6K1, ribosomal protein S6 kinase B1; ALKBH5, AlkB homolog 5; MAAS, m6A-modified lncRNA MAPKAPK5-AS1.

Disease stage and microenvironmental conditions may further alter EV release and cargo sorting. Hypoxia, nutrient deprivation, inflammation, viral infection, metabolic stress and exposure to anticancer treatment can reshape the RNA and protein content of EVs (142). Consequently, vesicles released from early lesions, advanced tumors, metastatic sites or treatment-resistant clones may transmit different biological signals. This context-dependent cargo selection may contribute to metastatic-niche formation, immune remodeling and intercellular transmission of treatment resistance (135).

Methodological variation can also create apparent heterogeneity. UC, SEC, precipitation, ultrafiltration and IAC enrich different particle populations and differ in the extent to which they co-isolate lipoproteins and non-vesicular macromolecules. Inconsistent isolation and normalization methods may therefore partly explain the variable results of candidate exosomal biomarkers across studies (143).

Clinically, functional heterogeneity presents both opportunities and challenges. Specific EV subpopulations may provide more precise information about tumor burden, metastatic potential, immune status or therapeutic response than unfractionated circulating biomarkers. Conversely, the lack of standardized subpopulation markers and analytical procedures limits reproducibility. Future studies should aim to combine standardized EV characterization with subpopulation-resolved or single-vesicle analyses, multiomics cargo profiling and prospective clinical validation to identify EV populations that are biologically meaningful and clinically actionable in HCC (144).

Current clinical utility and validation gaps of exosome-based biomarkers in HCC

Exosome-associated molecules are attractive liquid-biopsy candidates as vesicular encapsulation may stabilize molecular cargo and capture signals derived from both tumor cells and the TME. Among the reported candidates, serum exosomal miR-10b-5p, serum exosomal miRNA panels and selected EV-associated protein panels currently appear to be the most promising, although the strength of evidence differs notably among studies. Cho et al (73) evaluated serum exosomal miR-10b-5p and miR-215-5p in a cohort including 28 healthy controls, 60 patients with chronic liver disease and 90 patients with HCC. This design was clinically relevant because it included chronic liver disease controls rather than only healthy individuals. Serum exosomal miR-10b-5p exhibited strong diagnostic performance for stage I HCC, with an area under the curve (AUC) of 0.934, 90.7% sensitivity and 75.0% specificity, whereas miR-215-5p was associated with poor disease-free survival. The main strength of this study was its inclusion of a liver-disease comparator group and its focus on early-stage HCC. However, the findings were derived from a single-center cohort and require prospective external validation before clinical application.

Rui et al (74) analyzed serum exosomal miRNAs in 124 patients with HCC and 46 non-tumor donors and performed internal training and validation analyses. A three-miRNA model comprising miR-122-5p, let-7d-5p and miR-425-5p achieved an area under the receiver operating curve (AUROC) of 0.905 in the training analysis and an AUROC of 0.923 for low-stage HCC. This study supports the potential value of multi-miRNA panels rather than single-molecule biomarkers and provides evidence for early-stage disease detection. However, comparison with clinically representative surveillance populations, including patients with cirrhosis, chronic hepatitis or indeterminate liver nodules, remains necessary. Independent multicenter validation and standardized exosome-isolation and normalization procedures are also required.

Protein-based studies provide complementary but less mature evidence. Cho et al (125) first identified chaperonin containing TCP1 subunit 8 (CCT8) and cofilin 1 (CFL1) through proteomic analysis of HCC cell-derived EVs and then evaluated these proteins in serum test and validation cohorts. The clinical cohort included a small serum test cohort and a larger serum validation cohort of 224 individuals. CCT8 and CFL1 achieved AUC values of 0.698 and 0.677, respectively, whereas combining CCT8, CFL1 and AFP increased the AUC to 0.838. Higher levels were associated with vascular invasion, tumor stage, disease-free survival and overall survival. The main strength of this study was its combination of EV-guided discovery with clinical serum validation and its assessment of prognostic associations. However, the clinical measurements were performed in serum rather than in purified circulating patient-derived EVs. Therefore, this study supported EV-guided biomarker discovery but did not directly validate circulating exosomal CCT8 or CFL1 as clinical biomarkers.

Zhao et al (124) evaluated a serum EV protein panel including von Willebrand factor, galectin 3 binding protein, TGFB1, serpin family C member 1, hemopexin, haptoglobin, hemoglobin subunit α 1, fibrinogen α chain, fibrinogen γ chain and fibrinogen β chain. The experimental design included construction of a spectral library from pooled sera of 10 patients with HCC and 10 normal controls, followed by DIA analysis of 20 patients with HCC and 10 independent normal controls. The 10-protein profile separated HCC from normal-control samples in the discovery dataset. However, independently validated ROC-based clinical performance was not reported. In addition, the study used a small healthy-control comparison and did not include clinically important control groups such as patients with cirrhosis or chronic hepatitis. The Cancer Genome Atlas analysis assessed tissue gene expression rather than circulating EV protein concentrations, which further limits direct clinical interpretation. The experimental designs, exosome sources, clinicopathological information, major findings, strengths and limitations of representative EV-associated biomarker studies are summarized in Table IV.

Table IV Representative studies evaluating EV-associated biomarkers for HCC diagnosis and prognosis.

Table IV

Representative studies evaluating EV-associated biomarkers for HCC diagnosis and prognosis.

First author, year	EV source and experimental design	Cohort and clinicopathological information	Biomarker/s	Main findings	Strengths	Limitations	(Refs.)
Cho et al, 2020	Serum exosomal miRNA analysis	28 healthy controls, 60 patients with chronic liver disease and 90 patients with HCC, with analysis including early-stage HCC and postoperative prognostic assessment	Serum exosomal miR-10b-5p and miR-215-5p	miR-10b-5p exhibited an AUC of 0.934 for stage I HCC, with 90.7% sensitivity and 75.0% specificity and miR-215-5p was associated with poor DFS	Included chronic liver-disease controls and evaluated early-stage HCC detection	Single-center cohort, therefore prospective external validation is required	(73)
Rui et al, 2023	Serum exosomal miRNA profiling with internal training and validation analyses	124 patients with HCC and 46 non-tumor donors, with analysis including low-stage HCC	Serum exosomal miR-122-5p, let-7d-5p and miR-425-5p	A three-miRNA model achieved an AUROC of 0.905 in the training analysis and an AUROC of 0.923 for low-stage HCC	Supported a multi-miRNA panel strategy and provided evidence for low-stage disease detection	Requires comparison with clinically representative surveillance populations and independent multicenter validation	(74)
Zhao et al, 2021	Serum EV proteomic profiling using pooled sera for spectral-library construction and DIA analysis	Spectral library from pooled sera of 10 patients with HCC and 10 normal controls, with DIA analysis of 20 patients with HCC and 10 independent normal controls	Serum EV protein panel including VWF, LGALS3BP, TGFB1, SERPINC1, HPX, HP, HBA1, FGA, FGG and FGB	A 10-protein profile separated HCC from normal-control samples in the discovery dataset, thus independently validated ROC-based clinical performance was not reported	Provided preliminary EV-protein signature discovery data	Small cohort, healthy-control comparison without cirrhosis or chronic-hepatitis controls and TCGA analysis assessed tissue gene expression rather than circulating EV protein concentrations	(124)
Cho et al, 2021	HCC cell-derived EV proteomic discovery followed by serum test and validation analyses	HCC cell EV discovery (serum test cohort n=8 and serum validation cohort n=224), with higher biomarker levels being associated with vascular invasion, stage, DFS and OS	Serum CCT8 and CFL1, identified through HCC cell-derived EV proteomics (combined with AFP)	CCT8 AUC: 0.698, CFL1 AUC: 0.677 and CCT8 + CFL1 + AFP AUC: 0.838	Combined EV-guided discovery with clinical validation and assessed prognostic associations	Clinical measurements were performed in serum rather than purified patient-derived EVs, thus direct validation as circulating exosomal biomarkers remains incomplete	(125)

[i] OS, overall survival; DFS, disease-free survival; EV, extracellular vesicle; HCC, hepatocellular carcinoma; AUC, area under the curve; miRNA/miR, microRNA; AUROC, area under the receiver operating characteristic curve; CCT, chaperonin containing TCP1 subunit; CFL1, cofilin 1; AFP, α-fetoprotein; DIA, data-independent acquisition; VWF, von Willebrand factor; LGALS3BP, galectin 3 binding protein; SERPINC1, serpin family C member 1; HPX, hemopexin; HP, haptoglobin; HBA1, hemoglobin subunit α 1; FGA, fibrinogen α chain; FGG, fibrinogen γ chain; FGB, fibrinogen β chain; TCGA, The Cancer Genome Atlas.

Collectively, current biomarker studies suggest that serum exosomal miRNA panels and selected EV-associated protein signatures exhibit potential value in HCC detection and prognostic stratification (73,74,124,125). However, the evidence remains insufficient for routine clinical translation. The most consistent limitations include retrospective or single-center design, relatively small sample size, limited external validation, heterogeneous control groups, incomplete reporting of clinicopathological variables and differences in sample processing, EV-isolation methods, analytical platforms or normalization strategies. Future studies should aim to prioritize prospective multicenter cohorts, clinically appropriate surveillance controls, standardized EV workflows and direct comparison with AFP, imaging and established clinical variables.

Despite these promising findings, to the best of our knowledge, no exosome-derived biomarker has entered routine clinical practice for HCC surveillance or diagnosis. The majority of studies have been retrospective, single-center investigations involving relatively small cohorts and a number of these investigations have compared patients with HCC with healthy individuals rather than clinically relevant control groups, such as patients with cirrhosis, chronic viral hepatitis, metabolic liver disease or indeterminate liver nodules. Differences in sample type, preanalytical handling, EV-isolation procedures, RNA extraction, normalization, disease etiology and tumor stage further limit cross-study comparisons. Consequently, no exosomal miRNA, lncRNA, circRNA or protein has demonstrated a consistently improved performance across independent cohorts.

Clinical translation will therefore require standardized EV workflows, prospective multicenter studies, clinically appropriate control groups and independent external validation. Candidate biomarkers should also demonstrate incremental value beyond AFP, imaging findings and established clinical variables. Multimarker panels may be more clinically realistic compared with individual biomarkers, but adoption will ultimately depend on whether they improve early detection, recurrence prediction, treatment selection or patient outcomes.

Acknowledgements

Not applicable.

Funding

Funding: The present study was supported by the Guiding Project of Fujian Provincial Department of Science and Technology (grant no. 2023D034), the Natural Science Foundation of Xiamen (grant no. 3502Z202373106) and Scientific Research Foundation of State Key Laboratory of Vaccines for Infectious Diseases, Xiang'an Biomedicine Laboratory (grant no. 2024XAKJ0102013).

Availability of data and materials

Not applicable.

Authors' contributions

SH, XW and ZW conceived the present study and drafted the manuscript. XZ critically revised the manuscript. JL contributed to drafting, editing and critical revision. JW supervised the project and contributed to drafting, editing and critical revision. All authors read and approved the final version of the manuscript. Data authentication is not applicable.

Ethics approval and consent to participate

Not applicable.

Patient consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

References