At the first stage of metastasis, OS cells with an invasive phenotype migrate away from primary tumor and then invade into surrounding tissues. Such process of cancer invasion is critically dependent on the destruction and degradation of basement membrane and extracellular matrix (ECM), which is catalyzed by pericellular and extracellular proteases, mainly by the matrix metalloprotease (MMP) family. Remarkably, a series of studies have suggested that many proteins and microRNAs can promote OS metastasis via upregulating various MMPs, including MMP-9 (20, 21), MMP-13 (22, 23), membrane-associated MT1-MMP (also called MMP14) (24), and MMP-16 (25). Cathepsin is another proteolytic enzyme to be involved in OS metastasis (26). Furthermore, some inhibitors targeting MMPs have been explored to suppress OS aggressive behavior by blocking this stage, such as tissue inhibitor of metalloproteinases (TIMPs) (27) and Nobiletin (28). In addition, the interactions between tumor cells and microenvironment such as endothelial cells (29) and mesenchymal stem cells (MSCs) (30, 31) may promote tumorigenicity, whereas interactions with primed dendritic cells (32) and natural killer (NK) cells (33) have antitumor effects. For instance, MSCs promote OS metastasis by the interaction of CCL5 from MSCs and SDF-1 from OS cells via autocrine/paracrine communication (31). Tumor lysate-pulsed dendritic cells have been studied in combination with other candidate strategies, such as agonist antiglucocorticoid-induced tumor necrosis factor receptor (anti-GITR) antibody (32), anti-cytotoxic T-lymphocyte antigen-4 (anti-CTLA-4) antibody (33), and antitransforming growth factor-β (anti-TGF-β) antibody (34), to further improve OS treatment via enhancing antitumor immunity. In addition, NK cells can kill OS cells, including tumor-initiating cells, through the interaction between natural killer group 2 member D (NKG2D) receptor and its ligand (NKG2DL) (35). The significance of osteoclast-mediated bone destruction and resorption has been revealed throughout the OS development and progression (36). Nevertheless, the exact role of osteoclast in OS metastasis remains controversial and subject to further clarification. A more detailed discussion would be presented in the subsequent part of osteoclasts and metastasis in this review. Figure 1 A depicts the processes of OS cell invasion and migration at primary site.

Firstly, OS cells need to intravasate into microvasculature such as blood vasculature by crossing endothelial cells and basement membrane and then travel within blood flow. The circulating tumor cells arrest and eventually extravasate out from blood into the target secondary organ. Several studies have investigated the interactions between tumor cells and endothelia cells and identified a few related molecules, including runt-related transcription factor 2 (RUNX2), osteopontin (OPN), urokinase-type plasminogen activator (uPAR), and formyl peptide receptor type 1 (FPR1), all of which are further shown to facilitate metastasis in vivo. In order to survive in blood vessels, OS cells must acquire anoikis resistance property, which is regulated by many genes (37) such as FASN and ID1. Additionally, tumor cells also encounter with various physical hemodynamic forces (e.g., fluid shear stress) during transit (38). The time and intensity of circulating tumor cells exposure to fluids will affect their fates either survival or apoptosis (39). The development of bone adapts to mechanical load, suggesting that OS cells might be sensitive to their surround physical stimuli (40, 41). Moreover, the changes in physical stimuli could not only affect biological behaviors of OS cells but also influence their response to chemotherapy and radiation therapy (42, 43). The transfer of OS cells within vasculature is shown in Figure 1 B .

Under a poorly defined mechanism, the majority of circulating tumor cells arrive and arrest in lung microvasculature and subsequently extravasate into lung tissues, whereas only a minority of tumor cells can survive and eventually generate detectable metastasis (19, 44). Compared with primary site, the microenvironment at the secondary site presents a lot of differences, including oxygen tension, nutrition supply, and other physiochemical features. Lung is a foreign microenvironment, where tumor cells will undergo many challenges and face various fates, including apoptosis or death (e.g., immune clearance), dormancy, and proliferation into micrometastasis. In terms of micrometastases, they also have several fates either entering into angiogenic dormancy, or regression, or proliferation to form vascularized macrometastatic lesions (45, 46).

A few studies have attempted to elucidate the potential mechanisms by which OS cells conquer the selective pressures to successfully survive and proliferate in lungs. The response of OS cells to different stresses encountered in lungs is diverse. Endoplasmic reticulum (ER) stress can alter environment in ER lumen and disturb protein folding (47). The unfolded protein response (UPR) and UPR-related signaling pathways are also identified to dysregulated in OS (48, 49). Compared with low metastatic counterparts, OS cells with high metastatic potential express a higher level of ER chaperone protein HSPA5, whose inhibition can reduce lung metastasis (50). The ER stress may even partly account for the resistance of chemotherapy in OS (51). Moreover, the mechanism remains poorly understood by which OS cells preferentially metastasize to lungs. Based on current research, the mechanical restriction of circulating tumor cells within lung microvasculature might play a crucial role in lung tropism (52, 53). Another possible explanation of lung tropism is the notion of premetastatic niche (54). Extracellular vesicles, specifically exosomes released from cancer and stromal cells provide a favorable scenario for initiating organ-directed metastasis in several cancers, including melanoma (55), gastric cancer (56), pancreatic cancer (57), and cervical squamous cell carcinoma (58). A few reviews have discussed the role of exosomes in metastatic organotropism (59, 60), among which one publication focused on bone sarcomas (61). Moreover, Hoshino et al. revealed that tumor-derived exosomes could be uptaken by organ-specific cells to prepare the premetastatic niche, which is mediated by different exosomal integrins depending on the metastatic organs (62). Although their results were mainly from breast cancer, they also profiled the proteome of OS-derived exosome. Recently, several research groups have demonstrated that OS-derived exosomes enhance OS lung metastasis and some circulating plasma exosomal biomarkers are detected to be dysregulated in metastatic OS (63–66). Another study further confirmed the preferential seeding of OS-secreted EVs in lung tissue by fluorescence microscopy with lifetime imaging in vivo (67). Thus, we speculate that exosomes derived from metastatic OS cells may preferentially retain in lung to create a premetastatic niche by interacting with resident cells, and thus attract OS cells to migrate to lung and support their survival. More studies are needed to investigate the biodistribution of OS-secreted exosomes and whether molecules within exosomes direct OS cells to lung. A brief description of OS lung metastasis is presented in Figure 1 C .

Bone is a type of connective tissue with complex components, including various cells, soluble factors, and extracellular matrix (ECM). These components interact and influence with each other, by which the bone tissue maintains homeostasis in physiological conditions. The appropriate balance between bone formation and destruction determines normal structure of bone. Indeed, OS occurrence and metastatic initiation arise in such a complex bone environment. Additionally, metastatic cells also interact with surrounding microenvironment in each step of metastatic process, primarily in lung. Therefore, study on the interactions between microenvironment and metastatic cells will expand our understanding of OS biology. Increasing evidence has indicated that metastatic cells elicit and receive signals to and from microenvironment, which lead to metastasis inhibition or promotion. Figure 2 provides a brief overview of the interactions between metastatic OS cells and microenvironment.

TP63 is a member of the well-known tumor suppressor gene p53 family, and its splice variant ΔNTP63 is characterized by the lack of N-terminal transactivation domain. Bid and colleagues have shown that ΔNTP63 overexpression in OS cells increases phosphorylation of signal transducer and activator of transcription 3 (STAT3) by enhancing interleukin-6 (IL-6) and interleukin-8 (IL-8) secretion (123). In addition, phosphorylation of STAT3 can stabilize hypoxia-inducible factor 1-alpha (HIF1-α) and induce vascular endothelial growth factor (VEGF) secretion. Combined with clinical data, these results reveal a prometastatic role of ΔNTP63 in OS. Furthermore, the suppression of above cytokine/chemokine signaling pathways can reduce OS metastasis. In a mouse model, the inhibitions of IL-6 and C–X–C motif chemokine 8 (CXCL8, also called IL-8) significantly prolong survival by decreasing their deaths from metastasis (121). Moreover, the combination of inhibitors against IL-6 and CXCL8 achieves an intensive antimetastatic efficiency whereas each inhibitor alone only shows a modest effect. Another laboratory has reported that the metastatic OS cells expressing ΔNTP63α disseminate in a transforming growth factor beta (TGFβ)-rich microenvironment by upregulation of Smad4 and TβRII via suppressing miRNA-527/665 (120). In addition, TGF-β expression on the surface of extracellular vesicles (EVs), derived from OS cells, can induce the IL-6 secretion from mesenchymal stem cells (MSCs), which in turn facilitated proliferation of metastatic cells by activating STAT3. The administration of anti-IL-6 antibody disturbs this cross-talk signaling to reduce metastasis in mice (153). A research group recently has reported that myofibroblastic reprogramming of OS cells contribute to the formation of lung metastasis. This fibrotic reprogramming could be initiated by the activation of fibroblast growth factor (FGF) signaling and sustained by the resultant fibronectin (FN) deposition. They also demonstrated the efficacy of nintedanib in disrupting lung metastasis, but not in primary bone lesion, by blocking fibrotic reprogramming through inhibiting pan FGFR-FN axis (199).

Kleinerman and colleagues have indicated a noteworthy connection between OS cells and resident cells within lung, that is, the metastatic cells expressing Fas will be eliminated by binding to ligand (FasL), which is consistently expressed in lung (76). Under such selective pressure, only those Fas-negative OS cells or cells with nonfunctional Fas signaling can evade this defense mechanism and survive in lung (200). Thus, it is feasible to find agents that can induce Fas expression on OS cells as an alternative therapeutic approach against lung metastasis. This research group has identified two agents, chemotherapeutic agent gemcitabine and histone deacetylase inhibitor entinostat, both of which induce the regression of lung metastasis in wild-type mice by upregulating Fas expression (201, 202). However, such therapeutic efficacy was not observed in FasL-deficient mice since a FasL+ lung microenvironment is a prerequisite for this treatment (78, 203, 204). Moreover, another study confirmed this result again in a canine model with lung metastasis (205). These findings suggest that incorporating lung microenvironment as part of the therapy strategy may benefit patients with established lung metastasis. As an important factor in microenvironment, EVs like tumor-derived exosomes contain various components (e.g., proteins and nucleic acids) and play an essential role in intercellular communication and metastatic progression in a variety of cancers (206). Notably, the expression profiles of EVs from OS samples with low metastatic potential are significantly distinguished from that in high metastatic potential samples by transcriptome analysis (207). A study further indicated that culture medium supplemented with exosomes would change secretome of OS cells and affect their aggressive properties (208). Interestingly, proteomic analysis of the EVs derived from highly metastatic OS cells shows the enrichment of metastasis-related proteins. The EVs from highly metastatic clonal variants of OS can even be internalized into low metastatic cells and thereby endow them with metastatic ability through horizontal phenotypic transfer (67). Also, an in vivo metastatic model further confirmed the potential of OS-derived EVs to promote metastasis (65). The exosomes released by OS cells, carrying programmed death-ligand 1 (PD-L1) and N-cadherin, could also stimulate pulmonary metastasis (63), and a recent study found that the plasma exosomal sentrin SUMO-specific protease 1 (SENP1) level is closely related to pulmonary metastasis in OS patients (66). As such, OS cells preferentially migrate and localize to lung directed by the EVs they have secreted, which may partly account for lung tropism of metastasis. The role of cancer-derived exosomes in cancer development and progression was systematically discussed in a review by Kok et al., mainly elaborating on the trafficking of enriched genetic signals carried by exosomal cargo in fostering cancer progression in several tumor types, including OS (209). In addition to exosomes derived from the tumor itself, exosomes from other cells such as bone marrow mesenchymal stem cells (210), adipose-derived mesenchymal stem cells (211) and tumor-associated macrophages can also affect metastasis of osteosarcoma (212, 213). OS cells and non-OS cells in tumor microenvironment could influence themselves or each other by releasing EVs through autocrine/paracrine pathways, shaping tumor microenvironment, modulating cell biological behaviors, especially the aggressiveness of OS cells. The secreted CXCLs within microenvironment selectively recruit different types of cells through binding to their transmembrane receptors. Tumor cells and leukocytes expressing CXCRs migrate following CXCL gradient (214). The CXCL/CXCR axis plays an essential role in leukocyte trafficking, immune homeostasis maintenance like T-cell homing, directional migration of tumor cells. It is well documented that the binding of CXCL12 (also known as SDF-1) to its receptor like CXCR4 or CXCR7 remarkably promotes tumor progression including OS (100, 118, 215). In primary bone site, OS epigenetically downregulates CXCL12 expression by DNMT1, impairs cytotoxic T-cell homing to the tumor site and this chemokine gradient of CXCL12 drives the metastasis of OS cells to the lung ( Figure 3 A ), where CXCL12 is highly expressed. The constitutive expression of CXCL12 in lung may largely determine lung as the main site of OS metastasis ( Figure 3 B ). In addition to CXCR4, CXCR7 is another receptor for CXCL12 and also participate in OS lung metastasis (110). The coexpression of CXCR7 and CXCR4 on OS cells could enhance their metastatic ability, since a chemokine gradient of CXCL12 between bone and lung is produced through CXCR7-mediated CXCL12 scavenging in primary bone site (116).Moreover, osteoprotegerin could induce SDF-1 secretion from endothelial cells, which further promotes OS development by increasing neovascularization via SDF-1/CXCR4 axis (216). Nigris et al. indicated that transcriptional repressor Yin Yang 1 protein (YY1) enhances metastatic potential of OS cells by activating VEGF/CXCR4 axis (112). Another study also suggested that VEGF secreted from MSCs promotes CXCR4-mediated metastasis (108). In addition, IL-8 produced by MSCs increases anoikis resistance and metastasis of OS cells by regulating CXCR1/Akt signaling pathway (104). Gozo and colleagues has demonstrated that forkhead box protein C2 (FOXC2) maintains OS cells in a stem-like state and promotes metastasis by increasing CXCR4 (114). More importantly, the inhibition of CXCR4 by its antibody (98, 106) or inhibitor (like AMD300) (217) can successfully reduce metastasis. Due to the prominent role of CXCL/CXCR axis in metastasis and its close association with angiogenesis, anoikis, and immune response, novel therapy targeting this axis might be promising for treating metastasis.

The progression of bone tumor leads to osteolysis (218), which in turn promotes the dissemination of tumor cells and thus forms by a vicious cycle. Breaking this vicious cycle between osteoclasts and tumor cells may be one of the promising ways to treat OS metastasis. Several studies have used RANK-Fc to perturb TNFRSF11A/TNFSF11 (also called RANK/RANKL) axis and successfully reduced metastasis by inducing anoikis and apoptosis of OS cells (107, 109, 111). In addition, other agents that can restrict bone resorption are also expected to become potential candidates to inhibit metastasis, such as inhibitors against transcription factor Sp7, TNFSF10 and TGF-β/Smad signaling. Anti-osteoclast drug like zoledronic acid could be synergistically used to enhance the therapy efficacy for OS progression (219). One publication has discussed the prospects of novel therapeutic strategy targeting osteoclast activity for OS (220). However, Munoz et al. suggested the tendency of mutual restrain between metastatic OS cells and osteoclasts and that metastasis-competent OS cells induce the loss of ACP5+ osteoclasts, which in turn facilitate metastasis (119). Notably, EVs derived from OS could suppress osteoclastogenesis and further enhance its metastasis (65). An in vitro study was performed to directly explore the reciprocal modulation between OS and osteoclastic cells by a co-culture system (221). More studies are needed to further clarify the exact role of osteoclast in OS metastasis. The interaction between osteoclasts and OS cells was displayed in Figure 3 C .

Angiogenesis is an essential component for tumor growth and progression by supplying adequate blood and nutrients (75, 222). Additionally, these new blood vessels provide the principal route by which cancer cells exit the primary site and enter circulation (223). Angiogenesis is also required for tumor colonization at the site of metastasis. In OS metastasis, intensive investigation has explored the role of angiogenesis and interaction between proangiogenic and antiangiogenic factors, in order to develop potential optimum targets for antiangiogenic therapy ( Figure 3 A ). A meta-analysis including nine articles revealed that VEGF is positively associated with tumor metastasis and a higher tumor grade (70). Two publications also suggested that angiopoietin-like protein 2 and YY1 accelerate metastasis via VEGF-mediated angiogenesis (112, 224). Furthermore, VEGF knockdown (73) restricts OS metastasis. Also, the blockade of VEGF receptor 2 (VEGFR2) by anlotinib, a tyrosine kinase inhibitor, results in metastasis suppression in a preclinical study (225). In contrast to VEGF-related angiogenesis, pigment epithelium-derived factor (PEDF) or PEDF-derived synthetic peptides both exhibited antimetastasis activity by inhibiting angiogenesis (77, 79, 81, 226). Antiangiogenesis therapy seems to be an appealing and attractive alternative strategy to manage OS metastasis. However, the safety and efficacy of antiangiogenic therapy has not been confirmed in clinical trials, more work is needed to achieve clinical transformation of this treatment. Current research progress and application limitation in antiangiogenesis therapy for OS can refer to other reviews (227, 228).

The metabolic program of primary tumor cells is quite different from that of metastatic cells in many ways such as nutritional availability, energy demand, oxygenation level, metabolites, and metabolic pathways, all of which participate in tumor metastasis (229). An advanced metabolomics has shed light on the study of biochemical status and reprogrammed metabolism during metastasis. Metabolomics covers a wide range of metabolites, mainly including sugars, amino acids, and fatty acids. Giang et al. have observed the Warburg effect in several human OS cell lines (230), a common phenomenon of aerobic glycolysis in cancer cells that means pyruvate is transformed into lactic acid even in the presence of oxygen (231). Further study in mice by Hua and colleagues found the dynamic metabolic reprogramming throughout tumor occurrence and progression (232). They identified a number of differentially expressed metabolic biomarkers in serum prior and postmetastasis. Their results suggested the metabolic reprogramming in OS metastasis, characterized by lowered carbohydrate and amino acid metabolism, while an elevated lipid metabolism. Moreover, the serum metabolic profile of lung metastasis is distinct from that of primary tumor. Mice developing lung metastasis have a higher level of lipid metabolites in serum compared with mice without metastasis (232). Consistently, a global analysis of lipidomic reveals the alteration of lipid profiles in metastatic OS cells (233). Previous studies using synvinolin (inhibitor of de novo cholesterol synthesis) observed metastasis reduction, which further supported a critical role of lipid metabolism in tumor metastasis (234, 235). The inositol pathway is significantly downregulated in highly metastatic OS cells (236). Functional study both in ex vivo lung culture and in vivo mouse model has indicated that the administration of inositol hexaphosphate, which will be converted to inositol once entering cells, can reduce lung metastasis by suppressing MAPK and PI3K signaling pathways.

HIF1-α expression is positively associated with metastasis while negatively with survival. Naggar et al. has shown that Y-box-binding protein 1 (YB-1) facilitates metastasis by direct translational activation of epithelial-to-mesenchymal transition (EMT) and HIF1-α, which then induces CXCR4 expression (194). Functionally, HIF1-α enhances invasion of OS cells in hypoxia via increasing VEGF-A (237). Another pathway, HIF1-α/CXCR4 is also proven to facilitate metastasis in vitro and in vivo (238). Additionally, HIF-1α binds to AP-1-binding motif within Cyr61 (also called CCN1) promoter and induces Cyr61 expression, which plays a prometastatic role in human melanoma cells under hypoxia (239). Hypoxia is one of the prominent features of many malignant tumors and also provides an opportunity to develop agents that target hypoxic region. For instance, hypoxia-activated prodrug TH-302 can reduce OS metastasis as a single agent or in combination with chemotherapy (240). The high expression of Cyr61 indicates poor survival in OS patients (91) and promotes metastasis via activating PI-3K/Akt/GSK3β, IGF1/IGFR signaling pathways, facilitating angiogenesis characterized by an increase in VEGF, FGF2, PECAM, and a decrease in TSP-1 and SPARC (93) and promoting EMT-like process (95). Of note, WW domain containing oxidoreductase (WWOX) as tumor suppressor has been revealed to maintain mitochondrial respiration and attenuate Warburg effect by inhibiting HIF1-α (241) and c-Jun (242) by physical interaction with them. Recently, a research group further found that WWOX inhibits OS metastasis in vitro and in vivo through downregulation of RUNX2 (243). The suppression of c-Jun activity can inhibit metastasis by increasing apoptosis (85, 89) and chemosensitivity in OS cells (87). Tang and colleagues demonstrated that cadherin-4 (CDH4) overexpression would activate c-Jun via the JNK pathway (83). In addition to c-Jun, proteins from several families like c-Fos, ATF, and MAF, can form transcriptional complex AP-1, and their activation may be one of the mechanisms underlying metastasis (244). Leaner et al. have shown a higher activity of AP-1 in highly metastatic OS cells, compared with low metastatic counterparts (245). In terms of OS, AP-1 promotes metastasis by upregulating podoplanin and TGF-β (246). Another study also reported that FGFR1 silence, a downstream target of c-Fos/AP-1 complex, could significantly reduce lung metastasis (247). The growth and progression of malignant tumors largely rely on glycolysis for energy, resulting in an acidic tumor microenvironment, which may in turn affect the biological behaviors of tumor cells. Several studies have investigated how OS cells adapt to acidic microenvironment. Brief exposure of OS cells to acidic condition results in cell death, whereas prolonged exposure reverses cell death, and even fosters tumor invasiveness (40). The adaption of OS cells to acidosis leads to a metabolic reprogramming with epigenetic stability (248). The underlying mechanism has been further explored, by which acid microenvironment activates a stress-regulated switch to support cell survive. Studies found that cIAP proteins and NF-κB pathway are activated in OS cells in response to acid tumor microenvironment (249, 250). Metabolism-related regulatory networks are shown in Figure 4 .

As mentioned above, redox stress is a metabolic challenge in lung microenvironment for tumor cells. Studies shown that the excess reactive oxidative species (251, 252) and reactive nitrogen species (253, 254), both impair mitochondrial function. In order to overcome such oxidative pressures, metastatic OS cells trigger their antioxidant response by upregulating redox-related enzymes or glutathione-related metabolic pathways (255, 256). However, the previous mentioned study by Hua et al. reported the contradictory result that glutathione pathway is suppressed at the metastasis phase (232). More studies are urgently needed to further elucidate the exact role of metabolism especially glucose metabolism in tumor cells during dynamic metastatic process. Thereby, the exploration of metabolism reprogramming and metabolic vulnerability may provide novel targets for treating metastasis.

It is well known that immune system both innate and adaptive immunity plays a key role in tumorigenesis and progression, and even to a large extent determines the fate of tumor cells. With the advance in basic and preclinical research, increasing immune-based therapies are being approved for clinical use in various cancers, which yield encouraging outcomes in those patients who are resistant to conventional treatment. Nevertheless, the application of immunotherapy in metastatic OS is far from satisfactory to date. Recently, a series of studies identified several immune cells, including innate immunocytes (e.g., macrophages and dendritic cells) (257–259) and adaptive immunocytes (e.g., T lymphocytes) (260), all of which putatively participate in immune response during OS metastatic progression ( Figure 5 ). The number, function, state of immune cells, and their interactions with each other collectively determine whether they promote or inhibit metastasis. Researchers have also recognized some immunocytes as diagnostic or prognostic biomarkers. For instance, the increase of M2-polarized tumor-associated macrophages (TAMs) maintains OS stemness and metastatic property (261). In addition, TAMs activated by mifamurtide (262) or M2-type macrophages suppressed by all-trans retinoic acid (ATRA) (263, 264) both can inhibit tumorigenicity and progression of OS cells. However, the inconsistent results among studies are available, which require more research to further elucidate the dynamic and complex role of TAMs in OS. In addition to TAMs, tumor-infiltrating lymphocytes (TILs) within OS microenvironment also affect OS progression (261, 265). The increased CD8⁺ TILs as well as the high ratio of CD8⁺/FOXP3⁺ TILs both predict a better prognosis in OS patients (266, 267). However, the higher density of TILs is found to be within metastatic sites compared with primary tumor, which constitutes a special immune niche (82, 267). It should be noted that the interaction between PD-L1 (expressed in metastatic OS cells) and PD-1 (expressed in tumor-infiltrating cytotoxic T lymphocytes) limits antitumor function of T cells and thus promotes OS metastasis by evading immune surveillance. The dynamic and intricate regulation of immune system seems to be decisive for metastasis with respect to synthesis of immune factors, spatiotemporal expression of surface markers and phenotypic transition of immunocytes, and their interactions with surrounding compositions. With increasing understanding of immunomodulatory mechanism underlying OS metastasis, researchers are constantly exploring the key immune checkpoints to develop potential immunotherapy strategies for OS with the aim of improving prognosis. A recent review has discussed the mechanisms and status of immunotherapy for OS (268).

MiRNAs contain 18–25 bases and most of them fell within 21–23 bases. Typically, miRNAs bind to 3′untranslated region (3′UTR) of mRNAs by complementarily pairing to destabilize mRNAs or block translation and thus reduce their expressions (349–351). A single miRNA can target multiple mRNAs and vice versa, one mRNA can be regulated by various miRNAs. Thereby, miRNAs may indirectly affect various biological processes such as proliferation, differentiation, apoptosis, and angiogenesis. The unique signatures of miRNA expression profile have been studied in OS metastasis as diagnostic and/or prognostic biomarkers. More importantly, an enormous growing body of studies have explored potential mechanisms, by which miRNAs participate in OS metastasis.

A minority of the miRNAs that we retrieved from literature have been demonstrated for their effects on OS metastasis in vivo. Also, only few miRNAs show prometastatic activity in OS, while the majority of miRNAs inhibit metastasis. A study reported that miR-195 suppressed metastatic potential by targeting FASN (352). However, other studies found a high level of miR-195 in serum from OS patients (353, 354). Thus, a single miRNA might play a dual role in different stages of OS progression or the seemingly contradictory results are probably caused by small number of participants. Further studies are needed to confirm the exact function of such miRNAs like miR-195. Whether miRNAs promote or inhibit metastasis depends on proteins they target. Intriguingly, miR-27a* is the first passenger miRNA strand and shows prometastatic activity in OS cells, which is encoded by MIR27 and directly connects with CBFA2T3, the same as miR-27a (355). Furthermore, the injection of OS cells expressing miR-27a to mice can generate metastatic nodules within lung and bone in vivo (356, 357). In turn, miRNAs are also regulated by metastasis-associated genes such as MYC, TP53, and TGFB1. The epigenetic modulation affects the miRNA expression as well. One research team found that apurinic/apyrimidinic endodeoxyribonuclease 1 (APEX1) participated in multiple biological processes of OS by shifting miRNA expression profiles (358). Some drugs used to treat OS-like epirubicin (359) and diallyltrisulfide (360) can also alter miRNA expression profiles.

Unfortunately, miRNA-mediated therapy still remains the first phase of clinical trial at best despite so much basic and preclinical research. The main issue needed to be solved is how to deliver miRNAs efficiently and precisely to the sites of interest. Current methods generally show the following deficiencies, including low transfection efficiency, rapid degradation, and abnormal accumulation in nonspecific tissues and organs. Using hyaluronic acid-associated liposomes as carrier may accelerate the entry of miRNAs into cells and prevent them from degradation (361). An alternative approach is to apply viral vectors with hairpin molecules that will be processed into mature miRNAs (362).

LncRNAs are a large group of noncoding RNAs with more than 200 nucleotides and emerging evidence demonstrates its crucial role in multiple biological processes, particularly in tumorigenesis (363, 364). Briefly, lncRNAs in nucleus can regulate gene expression both at genetic and epigenetic levels by affecting transcription factors and chromatin-modifying complexes to bind specific gene loci. In cytoplasm, lncRNAs modulate the stability and translation activity of mRNAs either directly or indirectly. The majority of lncRNAs included in our study compete with miRNAs to bind target mRNAs as sponges. Recently, lncRNAs have engaged much attention in the field of tumor aggressiveness, including migration, invasion, and metastasis. Unlike miRNAs, most of lncRNAs involved in OS are positively associated with metastatic property. We outlined those lncRNAs that have been widely studied in OS metastasis and confirmed in vivo.

Metastasis-associated lung adenocarcinoma transcript 1 (MALAT1), also named noncoding nuclear-enriched abundant transcript 2 (NEAT2), is located in chromosome 11q13 with a length of 8.7kb (365). The MALAT1 overexpression is closely correlated with advanced clinical stage and distant metastasis in OS patients (366). Furthermore, a meta-analysis confirmed the predictive value of MALAT1 for metastasis and poor prognosis in six different types of cancer, including OS (367). Consistent with this result, another meta-analysis only including OS also showed MALAT1 as a valuable biomarker for prognosis (368). MALAT1 knockdown inhibits OS metastasis in vitro and in vivo by affecting various downstream pathways, such as downregulation of PI3K/Akt/MMP-9 signaling pathway (334, 335) and upregulation of E-cadherin and several miRNAs (336, 369). Also, high dose of 17β-estradiol inhibits metastatic potential of OS cells by suppressing MALAT1 while low dose presents the opposite effect (370, 371). It is noteworthy that the most widely studied lncRNAs is the large family of small nucleolar RNA host gene (SNHG), of which all members are considered to promote OS progression and metastasis by sponging various miRNAs. SNHG1, for example, activates Nin one binding protein (NOB1) by sponging miRNA-326 and thus promotes migration/invasion in OS cells and metastasis in murine models (341). Consistently, additional research further shows that SNHG1 is positively related to advanced clinical stage, distant metastasis, and poor survival of OS patients. In addition, SNHG1 facilitates cell proliferation, migration, and invasion via activating PI3K/AKT and Wnt/β-catenin pathways by sequestrating various miRNAs (341, 342, 372). Besides, SNHG12 promotes metastasis by increasing angiomotin (373), Notch2 (374) and IGF1R (375) through sponging miRNA-195-5p. There are two research groups that independently demonstrated the role of lncRNA DANCR in OS progression and possible underlying mechanisms. As a competitive endogenous lncRNA, DANCR either increases rho-associated protein kinase 1 (ROCK1) through decoying miRNA-335-5p and miRNA-1972 (376), or enhances AXL/PI3K/Akt signaling pathway by sponging miRNA-33a-5p (377), and thereby promotes OS metastasis.

CircRNAs are a newly identified class of endogenous noncoding RNAs characterized by closed-loop structures. Unlike typical linear RNA, circRNAs do not contain 5′-3′ polarity or polyadenylated tail and is thus resistant to enzyme degradation (378, 379). Similar to lncRNAs, circRNAs can competitively bind miRNAs as sponges to indirectly regulate gene expression. Currently, increasing evidence indicates that circRNAs have an important role in many biological behaviors and diseases, particularly in tumorigenesis and tumor progression. The dysregulation of some circRNAs has also been identified in OS metastatic process (380). Almost all circRNAs, like lncRNAs, act as sponges of miRNAs to disinhibit specific prometastatic pathways that normally are inhibited by miRNAs in OS. However, it is not always the case. It has been reported that low expression of circ-HIPK3 is observed in OS cell lines, tissues, and plasm and associated with lung metastasis (381). Overexpression of circ-HIPK3 can inhibit proliferation, migration, and invasion in OS cells. These findings indicate that circ-HIPK3 may have great value in clinical practice as a tumor suppressor. Intriguingly, circ-HIPK3 plays the opposite role in tumor occurrence and progression in other tumors, which strongly suggests that whether a single circRNA acts as an oncogene or a tumor suppressor gene depends on tumor types and which miRNA it sponges (382–386). In addition, circRNAs hsa_circ_0002052 and circ-ITCH are also found to be downregulated in OS cell lines and tissues. Functional studies further confirm that their overexpression can suppress OS cell progressiveness by inhibiting Wnt/β-catenin pathway via targeting miRNA-1205/APC2 axis (387), and by suppressing PTEN/PI3K/AKT and Sp1 pathways via sponging miRNA-22 (388), respectively. So far, the knowledge and research on circRNAs in OS metastasis are much less than that of miRNAs or lncRNAs. Nevertheless, the special annular structure gives circRNAs unique advantages with stronger stability and higher abundance.

In addition to tumor tissues, collecting noncoding RNAs from blood is convenient in clinical practice. Notably, a number of studies have reported that exosomes could carry noncoding RNAs (212, 213) and even transport them into OS cells (210), thus foster OS metastasis. With the deeper understanding of the relationship between noncoding RNAs and clinicalpathological characteristics, noncoding RNAs will help early diagnosis of metastasis, monitoring of treatment, and prediction of prognosis in OS patients. Although there is no therapy targeting noncoding RNAs used for OS yet, further research will make significant progress to develop novel therapeutic strategies and improve outcomes for OS patients with metastasis.