Cancer metastasis under the magnifying glass of epigenetics and epitranscriptomics

The metastatic cascade is defined by five different steps that allow tumor cells to leave the tissue of origin, spread to a distant area of the body, and finally colonize and grow in a different tissue (Fig. 1) [24, 25]. This process is extremely inefficient and can be aborted at several points. Indeed, the vast majority of cancer cells never leave the primary tumor, and of those that manage to enter into the circulation, most fail to colonize distant organs or are recognized by immune cells and abolished [26, 27]. Therapies that decrease the survival of metastatic cells in the vascular system have been shown to lower the risk of developing metastasis [28], but not all the steps of the metastatic cascade are amenable to therapeutic intervention, underlying the importance of identifying the drivers of the initial steps to develop effective interventions against the metastatic process [29].

Mechanistically, the first step, joining invasion and migration, consists of the separation of the tumor cells from their neighborhood cells in the primary tumor. One of the key mechanisms involved in the first steps of metastasis is the epithelial to mesenchymal transition (EMT), a biologic process that enables polarized epithelial cells to acquire a mesenchymal phenotype with enhanced migratory capacity, invasiveness, elevated resistance to apoptosis, and increased production of extracellular matrix components [30]. This process is mediated by transcription factors belonging to three distinct families: the Snail family (SNAIL1/SNAIL, SNAIL2/SLUG, and SNAIL3/SMUC), the Zeb family (ZEB1, ZEB2), and the b-HLH family (TWIST1, TWIST2) [31], as well as other factors that enhance cancer cell invasion through the crosstalk with the tumor microenvironment [32, 33]. These EMT transcription factors (EMT-TFs) repress epithelial markers like E-cadherin, occludins, claudins, or ZO-1 and activate mesenchymal markers like N-cadherin, vimentin, and fibronectin. EMT is considered an essential event in dissemination of cancer clones in epithelial tumors, as breast cancer [34]. Moreover, transforming growth factor beta (TGFβ)-induced EMT has been shown to promote the dendritic cell-like migration of breast cancer cells through the lymphatic vessels [35]. Also, the co-overexpression of Oct4 and Nanog transcription factors, essential to maintain the pluripotency and self-renewal of embryonic stem cells, positively regulates the EMT process in lung adenocarcinoma (LUAD) and breast cancer, contributing to metastasis and worse prognosis [36, 37]. Oct4/Nanog co-expression is also a strong independent predictor of tumor recurrence and unfavorable outcome in hepatocellular carcinoma (HCC) by promoting EMT through activation of Stat3/Snail signaling [38]. High expression levels of activated Stat3 (p-Stat3) and VEGF are also associated with lymph node involvement in esophageal squamous cell carcinoma (ESCC), indicating that Stat3/VEGF pathway promotes cancer cell lymphatic metastasis and is correlated with tumor-node-metastasis (TNM) stage [39]. From a therapeutic standpoint, the inhibition of Stat3 has been proposed as a relevant strategy to control stem cell-associated EMT phenotype. The recent development of small molecules targeting this pathway was deeply reviewed last year [40]. The main feature of EMT induction is the loss of cell–cell adherent junctions via inhibition of E-cadherin, encoded by the CDH1 gene. EMT inducers of the SNAIL, ZEB, and TWIST families act as direct E-cadherin repressors via binding to the E-box elements in the promoter region of the CDH1 gene [41, 42].

Following on the metastatic cascade, the second step is the intravasation, the transendothelial migration of cancer cells into blood or lymphatic vessels through the basal lamina fenestration. Third, during circulation, cells must survive in the circulatory system and deal with coagulation, shear stress, and the velocity of blood stream mechanic. Next is the extravasation or the second site seeding, through release of TGFβ by blood platelets that enable this process by disrupting cell-to-cell endothelial junctions [43]. Finally, distant tumor cells could undergo a mesenchymal to epithelial transition (MET) to settle down at the distant microenvironment and engage proliferation and metastasis outgrowth by adaptation to the new environment. This complex process is supported by the molecular communication of the tumor cells that permit their mobility, plastic transitions, adjacent angiogenesis, and immune escape.

It is important to remark that a delay in the metastatic outgrowth of disseminated cancer cells (DCCs) that reside in metastatic niches can occur, a state known as metastatic dormancy. This phenomenon could explain the fact that a significant proportion of metastatic diseases present years-to-decades following initial diagnosis and treatment [44]. During metastatic dormancy, the tumor cells stop proliferating and stay in a quiescent state due to unfavorable proliferative conditions. This process has important clinical implications as dormant cells can be responsible of disease recurrence, as non-proliferative cells are more resistant to therapies and evade immune attacks. Neophytou and colleagues recently reviewed mechanisms that promote or impede dormancy, and how drugs could target specifically these cells [45]. Metastatic dormancy is dynamically regulated to retain their self-renewal capacity and adopt potential new phenotypes depending on the environment. Thus, dormancy requires extensive crosstalk between DCCs and the cells that comprise their microenvironments, including endothelial cells, tissue-resident and circulating immune cells, mesenchymal stem cells, and stromal fibroblasts [44]. Interestingly, epigenetic regulation by DNA methylation and histone marks has been implicated in the dormancy of the DCCs and could constitute therapeutic options [44], but with the risk of reawakening dormant cells and promoting tumor progression even after a long latency period [46]. Consequently, a deeper understanding of the interplay between dormancy and epigenetic modulations seems crucial to identify novel targets to prevent and treat metastatic disease.

As previously mentioned, no specific genetic mutation seems to be determinant in metastatic progression, but some mutational combinations could enhance the metastatic ability. This topic has been reviewed few years ago by Patel et al. [8]. Although metastases have similar mutational profiles to their respective primary tumors, they commonly bear driver alterations at higher frequency than primary tumors. Priestley et al. characterized 2520 samples of metastatic tumors from 22 solid cancer types and they found no evidence of driver mutations that were specific to metastases [47], but they found that alterations in the MLK4 gene (which encodes Mixed lineage kinase 4) are frequently associated with metastasis. Interestingly, MLK4 upregulation has previously been linked to migratory and invasive phenotypes in breast cancer cells [48]. In addition, germline variants of APOE have been shown to be associated with different outcomes in melanoma. Mice expressing the human APOE4 variant exhibited reduced metastasis compared to those expressing APOE2, suggesting that pre-existing hereditary genetic variants can modulate progression and clinical outcomes due to variations in metastatic competences [49]. Using mouse models, it has been shown that a combination of four mutations in genes commonly found in primary human cancers (Apc^fl/fl, Kras^LSL−G12D, Tgfbr2^fl/fl, and Trp53^fl/fl) promotes metastases, but the lack of one of them hinders the metastatic ability [50]. Besides, a study of multiregion biopsies of primary and liver metastatic regions from colorectal cancers with whole-exome sequencing and copy number profiling suggested that the genetic intratumoral heterogeneity is a potential driving force to generate metastasis-initiating clones [51]. Copy number aberrations and concomitant overexpression of MYC, YAP1, or MMP13 also increased the incidence of brain metastasis from lung cancer [52]. This observation suggests that overexpression of genes by other mechanisms like gene-promoter hypomethylation or enhanced chromatin accessibility could result into similar outcomes. Moreover, genetic aberrations can impact central pieces of the epigenetic or epitranscriptomic machineries. To cite one example, gene amplification of the m⁶A reader YTHDF3 has been described in breast cancer brain metastases. Resultant overexpression of YTHDF3 increases translation of m⁶A-enriched transcripts as ST6GALNAC5 and GJA1, key brain metastatic genes [53].

Epigenetic modifications refer to heritable changes in gene activity that do not involve changes in the underlying DNA sequence. Epigenetic factors fine-tune gene expression programs and act as master regulators controlling essential biological functions. The main epigenetic mechanisms are DNA methylation, histone modifications, chromatin remodeling, and non-coding RNA regulation. In this section, we describe relevant epigenetic alterations involved in metastasis, summarized in Table 1.

Chemical modifications deposited in RNA molecules shape another layer of biological complexity, the epitranscriptome. More than 170 modifications have been identified [216], most of them originally described in highly abundant non-coding RNAs, including ribosomal RNAs (rRNAs), transfer RNAs (tRNAs), and small nuclear RNA (snRNAs), but later also found in messenger RNAs (mRNAs). The functional significance of RNA modifications has begun to be uncovered in the last few years, and increasing evidences are supporting their role in diseases, including cancer [217]. In this section, we describe epitranscriptomic events particularly related to metastasis, summarized in Table 2 and depicted in Fig. 3. Moreover, considering the interplay between molecular mechanisms controlling gene expression, we highlight those targets impacted by both epitranscriptomic and epigenetic modifications. Also, epitranscriptomic modifications can regulate epigenetic modifiers like microRNAs or lncRNAs that will regulate downstream gene expression, a topic recently reviewed by Bove et al. for m⁶A modification [218].

The most known epitranscriptomic modification is certainly the N6-methyladenosine (m⁶A) deposition on mRNAs. The catalytic activity of m⁶A deposition is driven by METTL3 and assisted by METTL14, WTAP, RBM15, KIAA1429, and ZC3H13. The three described erasers are FTO, ALKBH3, and ALKBH5, and the known readers that bind to the m⁶A mark and mediate cellular outcomes are YTHDF1/2/3, YTHDC1/2, IGF2BP1/2/3, and HNRNPC/A2B1.

Starting with the m⁶A writer, METTL3 has been found upregulated in late stages of breast cancer and associated with worse prognosis [228]. Mechanistically, the authors have shown that METTL3-mediated m⁶A modification of SOX2, one of the regulators of Nanog, promotes stemness and malignant progression of breast cancer [228]. This example highlights that epitranscriptomic regulations of transcription factors (like SOX2) can control EMT effectors (like Nanog) potentially regulated by epigenetic events. In CRC, METTL3 also stabilizes SOX2 mRNA, and promotes drug resistance and lung metastasis [229]. Another METTL3 role in CRC is by downregulating SPRED2 and targeting the miR-1246/SPRED2/MAPK signaling pathway [219]. Moreover, Yin et al. have found that m⁶A methylation can regulate cancer growth and metastasis via macrophage reprogramming [220]. Strikingly, ablation of METTL3 in myeloid cells promoted tumor growth and metastasis, associated with decreased YTHDF1-mediated translation of SPRED2 and increased activation of NF-kB and STAT3 through the ERK pathway, leading to increased tumor growth and metastasis [220]. In a breast cancer lung metastasis model, upregulated METTL3 methylates KRT7-AS to enhance its expression, which then forms duplex with KRT7 mRNA to increase its stability and expression through recruitment of the IGF2BP1/HuR complex [221]. Moreover, YTHDF1/eEF-1 is involved in FTO-regulated translation elongation of KRT7 mRNA [221]. Of note, KRT7 has been described to promote EMT in ovarian cancer via the TGFβ/Smad2/3 signaling pathway [322]. Interestingly, it has been reported in breast cancer that BRCA1 promoter methylation is correlated strongly and specifically with both BRCA1 gene expression in ER − tumors and KRT7 promoter methylation with KRT7 gene expression in ER + tumors [323]. The same authors also reported differential KRT7 protein expression in luminal cell populations associates with survival [324]. This example underlines that some epigenetically regulated genes can co-participate in tumorigenesis in a way like BRCA1.

Back to METTL3, its oncogenic activity has also been linked to tumor formation in CRC where m⁶A methylation of lncRP11 increases its nuclear accumulation, drives Siah1 and Fbxo45 mRNA degradation, and prevents ZEB1 mRNA decay, thus, regulating EMT and liver metastasis [222]. In HCC, METTL3 decreases SOCS2 mRNA stability (through an m⁶A-YTHDF2-dependent mechanism) and promotes lung metastasis [232]. Interestingly, epigenetic downregulation of SOCS2 has been described in AML cells carrying NRAS mutation, driven by a reduction of the active enhancer marker H3K27ac at the SOCS2 locus [325]. In gastric cancer, METTL3 stabilizes hepatoma-derived growth factor (HDGF) that activates GLUT4 and ENO2 expression and subsequent angiogenesis and glycolysis that promote liver metastasis [223]. Of note, concomitant upregulation of HDGF and PRKCA has been associated with poor prognosis in lung adenocarcinoma [326]. Conversely, in a study in papillary thyroid cancer (PTC), low level of STEAP2 correlated with aggressiveness, and METTL3-mediated STEAP2 mRNA stabilization decreased PTC progression [224]. In cervical cancer, METTL3 is upregulated, promoting tumorigenesis and Warburg effect, and correlates with lymph node metastasis and poor prognosis [230]. Mechanistically, METTL3 targets the 3′UTR of the hexokinase HK2 mRNA, and recruits YTHDF1 to enhance its stability [230]. In hepatoblastoma, a rare liver cancer usually diagnosed during the first 3 years of life, the upregulation of METTL3, YTHDF2, and FTO has been correlated with poor clinical outcomes. METTL3 knockdown in hepatoblastoma cells could suppress proliferation, invasion, and migration [327]. In melanoma, METTL3-mediated stabilization of the uridine cytidine kinase 2 (UCK2) by m⁶A modification plays a role in melanoma metastasis by enhancing the Wnt/β-catenin pathway [225]. Of note, LINC00470 recruits METTL3 to drive m⁶A methylation of PTEN mRNA, leading to YTHDF2-dependent PTEN mRNA decay and finally promoting metastasis in gastric cancer [233]. In NSCLC, Pan et al. found that the METTL3/YTHDF2 m⁶A axis also downregulates miR-1915-3p that has tumor suppressor function by targeting SET nuclear proto-oncogene (SET) [234]. In a more functional study, Li et al. suggested that METTL3 regulates EMT by modulating β-catenin expression and subcellular localization [328]. Moreover, by studying TGFβ-mediated EMT in cancer cell lines, a significant increase of m⁶A levels in cells undergoing EMT has been described. Particularly, m⁶A deposition in SNAI1 mRNA triggers its YTHDF1-mediated translation to promote EMT [231]. Interestingly, the microbiota can also exert a strong influence on the host m⁶A epitranscriptome. A recent study has shown that gut microbes Fusobacterium nucleatum reduce the m⁶A levels of CRC cells through the YAP/FOXD3/METTL3 axis, resulting in upregulation of KIF26B, enhanced aggressiveness, and metastasis [329]. From a therapeutic standpoint, the screening of a library of small molecules identified Elvitegravir, originally developed to treat human immunodeficiency virus (HIV) infection, as a drug able to repress the invasion and metastasis of ESCC cells by enhancing the proteasomal degradation of METTL3 mediated by STUB1. The authors have shown that METTL3-mediated m⁶A modified EGR1 mRNA is stabilized by YTHDF3, leading to increased activation of EGR1/Snail signaling [241]. In breast cancer cells, METTL3 mediates methylation of the lncRNA MALAT1 that promotes metastasis. MALAT1 acts as a decoy for miR-26b, which targets and downregulates HMGA2. Thus, m⁶A methylation of MALAT1 indirectly increases HMGA2 expression, with an effect on EMT. Notably, high expression of METTL3 and MALAT1 in breast cancer has been associated with poor prognosis [226]. In lung cancer, Wang et al. have shown that METTL3 targets miR-143-3p to enhance its biogenesis, leading to the silencing of vasohibin-1 (VASH1) and subsequent increased angiogenesis and blood–brain barrier invasion capability, finally promoting brain metastasis [227]. Conversely, METTL14 acts as a metastasis suppressor by mediating the decay of its targets via recognition by m⁶A readers. Examples include m⁶A-dependent degradation of NEAT1 lncRNA [242] and ITGB4 mRNA [243] in RCC; or SOX4 mRNA [244] and XIST lncRNA [245] in CRC, all of them via recognition by YTHDF2. Moreover, METTL14 was found downregulated in CRC, and decreased expression correlates with poor overall survival [244].

Regarding m⁶A readers, YTHDF1 expression correlates with lymph node metastasis and worse clinical stages in CRC. The oncogene c-MYC promotes the YTHDF1 expression in CRC cells [330]. YTHDF1 has been directly implicated in EMT by controlling Snail mRNA modification and translation rate [231]. YTHDF1 promoted tumorigenesis and metastasis in ovarian cancer by augmenting the translation of m⁶A-modified eIF3C mRNAs [246]. eIF3C upregulation has been associated with tumorigenesis in RCC [331] and HCC, in the later by increasing the secretion of extracellular exosomes to promote angiogenesis [332]. Moreover, simultaneous high expression of EIF3C and S100A11 is associated with poor HCC patient survival [332]. In HCC, a clinically relevant m⁶A-YTHDF1-EGFR axis has been described. Radiofrequency ablation (RFA) is recommended as a minimally invasive curative therapy in this tumor type. However, a recurrent rate of 50% is observed within 3 years, mainly attributed to metastasis due to sublethal heat stress from insufficient RFA. A recent study has shown that out of the ablation center, sublethal heat treatment in the surrounding transitional zone results in cellular stress that promotes YTHDF1 expression and could contribute to HCC metastasis. Mechanistically, YTHDF1-mediated recognition of m⁶A-modified EGFR mRNA enhanced its translation, supporting the rationale for targeting m6A machinery in combination with EGFR inhibitors to prevent metastasis in HCC after RFA [247]. Also in HCC, HIF1α-induced expression of YTHDF1 drives ATG2A and ATG14 translation, two autophagy-related genes, promoting tumor progression [248]. Other examples of the pro-metastatic role of YTHDF1 by promoting gene translation have been described, including YTHDF1-mediated increased translation of USP14 in gastric cancer [249], ARHGEF2 in CRC [250], FOXM1 and PKM2 in breast cancer [251, 252], PLK1 in prostate cancer [253], SLP2 and ANLN in HCC [254, 255], DDX23 in pancreatic ductal adenocarcinoma [256], and HRAS in various cancers [257]. In breast cancer, YTHDF1 and YTHDF3 are frequently amplified and consequently overexpressed, and significant correlations with intrinsic subclasses and nodal metastasis have been described, suggesting their use for prognosis stratification and therapeutic intervention in breast cancer [333]. Regarding regulation, YTHDF1 can be targeted by miRNAs, including miR-136-5p in CRC [334] and also miR-1285-3p in lung cancer, where Linc00337 is upregulated and acts an miR-1285-3p sponge enhancing YTHDF1 expression [335]. The study in lung cancer also has shown that Linc00337 silencing represses tumor progression in vitro and in vivo [335].

About YTHDF2, the METTL3/YTHDF2 m⁶A axis promotes tumorigenesis in bladder cancer by mediating degradation of RRAS [235], SETD7, and KLF4 [236]. Also, METTL3 depletion suppressed metastasis [236]. In HCC, YTHDF2 expression has been correlated with poor survival, and it drives the translation of OCT4 [258]. Moreover, the loss of YTHDF2 led to reduction of tumor burden and lung metastasis in mice [258]. Also in HCC, m⁶A-modified FYN mRNA bound YTHDF2 and the proliferation associated protein 2G4 (PA2G4). Thus, PA2G4 upregulation in HCC promotes EMT, and plays a pro-metastatic role by increasing FYN expression through binding with YTHDF2 [259]. In lung adenocarcinoma, YTHDF2 is upregulated and drives AXIN1 mRNA decay, activating the Wnt/β-catenin signaling pathway that finally promotes tumorigenesis [260]. In cervical cancer, YTHDF2-mediated degradation of the tumor suppressor GAS5 enhances growth and metastasis, and inversely, the lncRNA GAS5-AS1 has the opposite effects [261]. In breast cancer, YTHDF2 downregulates FGF14-AS2, promoting osteolytic metastasis by enhancing RUNX2 mRNA translation [262], and in NSCLC, it downregulates circ_SFMBT2 to promote metastasis [263]. Other examples supporting the role of the METTL3/YTHDF2 m⁶A axis in metastasis include the decay of targets with tumor suppressor functions, as YPEL5 (associated with decrease of CCNB1 and PCNA expression) in CRC [237], USP4 in prostate cancer [238], DUSP5 in gallbladder cancer [239], and IFFO1 in different types of cancer [240]. Nevertheless, there are also examples associating a decrease in YTHDF2 expression with poor prognosis, tumor progression, and metastasis in HCC [264]. YTHDF2 degrades m⁶A-containing IL-11 and SERPINE2 mRNAs, two mediators of hypoxia-induced cancer cell survival and vascular reconstruction. Thus, HIF-2α-mediated inhibition of YTHDF2 in HCC promotes cancer-associated inflammation. Importantly, the use of HIF-2α antagonist restored YTHDF2-programed epigenetic machinery and repressed liver cancer [264]. Also, in lung adenocarcinoma, YTHDF2 was found to inhibit migration and invasion, and high expression correlated with better overall survival [336].

The third member of the YTH family, YTHDF3, has also been implicated in metastasis. In HCC, ZEB1 mRNA is upregulated by circ_KIAA1429 in a YTHDF3 m⁶A-dependent manner, leading to enhanced invasion and metastasis [265]. Moreover, in TNBC, increased expression of YTHDF3 correlated with worse prognosis, and YTHDF3 positively regulated cell migration, invasion, and EMT by also stabilizing ZEB1 mRNA [266]. As previously commented, YHTDF1 and YTHDF3 are amplified and overexpressed in breast cancer, correlating with poor overall survival and nodal metastasis [333]. In breast cancer brain metastasis, YTHDF3 overexpression due to increased copy number has been also described [53]. YTHDF3 induces the translation of ST6GALNAC5, GJA1, EGFR, and VEGFA mRNAs in an m⁶A-dependent manner [53]. In pancreatic cancer, YTHDF3 increased expression correlates with poor overall survival, and YTHDF3 regulates the lncRNA DICER1-AS1 that promotes glycolysis through inhibition of miR-5586-5p [267]. The pro-metastatic effect of YTHDF3 was also observed in melanoma, where it increases the translation of LOXL3 mRNA [268]. In HCC, YTHDF3 expression was found upregulated and related to worse prognosis, and it promotes aerobic glycolysis by increasing the translation of PFKL [269].

About m⁶A erasers, the FTO demethylase is upregulated in breast cancer and associated with poor overall survival [272]. The study demonstrated that FTO targets m⁶A in the 3′UTR region of the pro-apoptotic BNIP3 mRNA, suggesting FTO as a potential therapeutic target to enhance BNIP3 expression and decrease tumor growth and metastasis [272]. In RCC, BNIP3 is epigenetically silenced by histone deacetylation [337], and treatment with the HDAC inhibitor (HDACi) trichostatin A restores BNIP3 expression and led to cell growth inhibition and apoptosis [337], suggesting a potential therapeutic strategy in that setting. This strategy was also proposed in TNBC, where the combination of the HDACi YCW1 with radiation induced autophagic cell death by downregulation of BNIP3 [338]. This combination has also been proposed to tackle breast cancer lung metastasis [339]. Also in breast cancer, FTO upregulates ARL5B by inhibiting miR‐181b‐3p [270]. In endometrial cancer, FTO demethylates HOXB13 mRNA, promoting metastasis by activating Wnt signaling pathway [273]. In this cancer type, FTO was upregulated in metastases, and demethylation of HOXB13 mRNA led to a decreased decay by lack of YTHDF2 recognition [273]. In multiple myeloma (MM), FTO has also a pro-metastatic role by targeting HSF1, enhancing HSF1 mRNA stability and translation [274]. Importantly, FTO inhibition combined with bortezomib treatment synergistically inhibited myeloma bone tumor formation [274]. Nevertheless, a tumor suppressor role of FTO has been described in intrahepatic cholangiocarcinoma (ICC) by controlling different pathways, including EGFR, and by decreasing the stability of the oncogene TEAD2 [271].

Regarding the ALKBH5 m⁶A eraser, exposure to hypoxia induced the HIF-1α and HIF-2α-dependent expression of ALKBH5 that demethylates and stabilizes Nanog mRNA, leading to the acquisition of a cancer stem cell phenotype in breast cancer cells [279]. Moreover, inhibitions of either ALKBH5 or HIF factors were effective strategies to decrease Nanog expression. The same authors observed that hypoxia also leads to the overexpression of Nanog and another pluripotency factor, KLF4, in a ZNF217-dependent and HIF-dependent manner in breast cancer cells [280]. Interestingly, ALKBH5 knockdown in a breast cancer cell line decreased metastasis from breast to lung in vivo [280]. In laryngeal squamous cell carcinoma (LSCC), increased expression of ALKBH5 can promote tumorigenesis in an m⁶A‐YTHDF2‐dependent manner by promoting the lncRNA KCNQ1OT1‐HOXA9 signaling axis [281]. In HCC, the lncRNA CASC11 is upregulated and associated with poor survival and metastasis [282]. CASC11 stabilizes UBE2T mRNA by recruiting ALKBH5 m⁶A activity and also by inhibiting UBE2T mRNA association with YTHDF2 [282]. Also in HCC, YAP activation via the ALKBH5-mediated m⁶A demethylation of circCPSF6 has been associated with malignancy, as circCPSF6 sustains the stability of YAP1 mRNA [283]. m⁶A-modified circCPSF6 is recognized and destabilized by YTHDF2, but circCPSF6 m⁶A demethylation by ALKBH5 facilitates the proliferation and metastasis of HCC cells via competitive interaction of circCPSF6 with PCBP2, activating the YAP1 signaling [283]. In CRC, another circRNA involved in progression and metastasis is circ3823. The authors have shown that circ3823 could be targeted by YTHDF3 and ALKBH5 to promote its degradation; meanwhile, these modifiers are downregulated in CRC, leading to increased expression of circ3823. Circ3823 repressed miR-30c-5p that normally impedes transcription factor 7 (TCF7) mRNA translation, leading to increased TCF7 protein level and promoting proliferation, angiogenesis, and metastasis [284]. However, in NSCLC, ALKBH5 inhibited growth and metastasis by reducing YTHDF-mediated YAP expression [275]. In pancreatic cancer cells, ALKBH5 also inhibits cell motility and invasion by demethylation-dependent increase of KCNK15-AS1 expression [276, 277]. Mechanistically, KCNK15-AS1 is able to recruit MDM2 to promote REST ubiquitination, thus transcriptionally upregulating PTEN to inactivate AKT pathway [277]. In NSCLC, ALKBH5 overexpression decreases the mRNA stability and expression of TGFβR2 and SMAD3 but enhances those of SMAD6, leading to a reduction of metastasis in vivo [278]. About the ALKBH3 eraser, one study linked its overexpression to metastasis and poor progression-free and overall survival in HCC [340].

Another epitranscriptomic modification that has been implicated in metastasis is 5-methylcytosine (m⁵C). This modification is written by the NOP2/Sun RNA methyltransferase family members NSUN1 to NSUN7, the DNA methyltransferase DNMT2, or the TRNA aspartic acid methyltransferase 1 TRDMT1, depending on the RNA target. Thus, for example, NSUN1 and NSUN5 modify rRNAs; NSUN2, NSUN6, and DNMT2 methylate tRNAs; NSUN3 and NSUN4 modify mitochondrial RNAs (mtRNAs), whereas NSUN2, NSUN6, and DNMT2 methylate mRNAs [341‐343]. In addition, our group has recently shown that NSUN7 is also able to deposit the m⁵C mark in mRNA, characterizing the transcript of coiled-coil domain containing 9B (CCDC9B) gene as a relevant target in HCC. Interestingly, we have shown that the epigenetic-mediated inactivation of NSUN7 in liver cancer decreases CCDC9B mRNA half-life and is associated with poor overall survival [290]. Few years ago, we also described the epigenetic silencing of NSUN5 by gene-promoter methylation in glioma. NSUN5 methylates cytosine at position C3782 of 28S rRNA and its loss drives an adaptive translational program. In contrast to NSUN7, NSUN5 silencing is a hallmark of glioma patients with long-term survival [344]. These examples further support the crosstalk between epigenomic and epitranscriptomic mechanisms as a key player in cancer. Regarding m⁵C readers, the first identified was the Aly/REF export factor (ALYREF) [345]. Also, Y-box binding protein 1 (YBX1) recognizes and enhances the stability of m⁵C-modified mRNA by recruiting ELAVL1 [286].

In breast cancer, NSUN2 is overexpressed by gene promoter hypomethylation and correlates with clinical stages [346]. In gastric cancer, NSUN2 overexpression has been shown associated with metastasis and regulated by SUMO-2/3 that stabilized and mediated its transport into the nucleus. PIK3R1 and PCYT1A expression correlated with prognosis and NSUN2 expression, and they could be the target genes that participate in gastric cancer progression [285]. In bladder cancer, NSUN2 methylates and stabilizes HDGF mRNA 3′UTR (also stabilized by METTL3 in gastric cancer) promoting cancer metastasis [286]. Moreover, a high co-expression of NUSN2, YBX1, and HDGF predicts poor survival in this tumor type [286]. NSUN2 overexpression and NSUN6 downregulation were associated with poor prognosis in TNBC by upregulating monocytes/macrophages and Tregs in the tumor immune microenvironment [347]. Recently in cholangiocarcinoma, Zheng et al. have shown that NSUN2 targets the NF-kB interacting lncRNA NKILA enabling the recognition by YBX1. They also showed that NKILA targets YAP1 via miR-582-3p and the upregulation of NKILA was correlated with lymph node and distant metastasis [287]. LRRC8A is another lncRNA that has been demonstrated targeted by NSUN2 and stabilized by YBX1 in cervical cancer. In this context, NSUN2 expression was associated with metastasis [288]. Regarding NSUN3, considering its function as mtRNA modifier, and the metabolic flexibility and plasticity during cancer progression [348], NSUN3 could play a role in the metabolic adaptability required for metastasis formation. Indeed, a pivotal study in the field has recently shown that m⁵C in mtRNAs is essential for the dynamic regulation of mitochondrial translation rates, and thereby shapes metabolic reprogramming during metastasis [289]. The authors have shown that the translation of mitochondrially encoded subunits of the oxidative phosphorylation complex depends on the formation of m⁵C at position 34 in mitochondrial tRNA^Met. Importantly, this modification was required for efficient tumor metastasis, but not for primary tumor development and growth [289]. This study also identified an NSUN3-driven gene signature that predicts metastasis in patients with HNSCC. Remarkably, these findings support the inhibition of m⁵C formation in mitochondria as a therapeutic opportunity to prevent the dissemination of tumor cells from primary tumors [289]. Also in HNSCC, NSUN5 and ALYREF expression correlated with TNM stages [349], underlying again the different routes cancer cells can exploit to increase the metastatic competence. Moreover, the tRNA m⁵C methyltransferase TRDMT1 has been positively associated with tumor size, histological grade, invasion depth, lymph node metastasis, and TNM stage in gastric cancer [350].

Regarding other RNA modifiers, TRIT1 (tRNA isopentenyltransferase 1) is the enzyme responsible for the hypermodification of the A37 position in the anticodon region of human tRNAs containing serine and selenocysteine to N6-isopentenyladenosine (i⁶A). A TRIT1 polymorphism and a haplotype were associated with lymph node metastasis in gastric cancer [351]. Moreover, our group has found that TRIT1 undergoes gene amplification-associated overexpression in small cell lung cancer (SCLC) that promotes tumor growth [352]. SCLC is a disease characterized by aggressive dissemination and poor prognosis. In most of the cases, the cancer has already metastasized by the time of diagnosis. Another modification affecting adenine is N1-methyladenosine (m¹A). Like other modifications, its deposition is dynamically regulated by several proteins: the tRNA methyltransferases TRMT61A, TRMT61B, and TRMT10C assisted by TRMT6 that introduce m¹A, and the demethylases ALKBH1 and ALKBH3 that remove the modification. Moreover, YTHDF1, YTHDF2, YTHDF3, and YTHDC1 recognize m¹A sites and induce downstream effects. In CRC, it has been recently shown that MFAP2 mRNA was overexpressed and highly m¹A modified, and its expression was related to lymph node metastasis and distant metastasis, leading to poor prognosis. Nevertheless, the authors did not elucidate the TRMT writer implicated in this process [353]. TRMT6 non-catalytic subunit is overexpressed and associated with the TNM stage in HCC, through the PI3K/AKT/mTOR axis, and it has been suggested as potential therapeutic target for this tumor type [354]. Moreover, a recent study has found that the demethylase ALKBH1 is overexpressed in CRC and associated with metastasis and poor prognosis. The study showed that ALKBH1-catalyzed m¹A demethylation is essential for CRC cell migration and invasion, suggesting that ALKBH1 accelerates CRC metastasis by downregulating SMAD7 expression. Mechanistically, the authors described that ALKBH1-mediated m¹A demethylation of METTL3 mRNA inhibits SMAD7 expression by METTL3-mediated m⁶A modification [291]. YTHDC2 is an m¹A reader that has been related to regulation of EMT effectors and metastatic ability of breast cancer cells [355].

NAT10, the writer of N4-acetylcytidine (ac⁴C), is another epitranscriptomic player that has been implicated in various types of cancer. As previously commented, a role of NAT10 as metastasis suppressor has been found in CRC, where miR-6716-5p-dependent downregulation of NAT10 enhances metastasis [356]. Nevertheless, several studies have described an opposite role. For instance, in gastric cancer, NAT10 promotes metastasis and EMT by targeting COL5A1 mRNA and increasing its stability [292]. Also, NAT10 is upregulated in colon cancer and associated with poor overall survival. In this setting, NAT10 inhibits ferroptosis through N4-acetylation and stabilization of the ferroptosis suppressor protein 1 (FSP1) mRNA [293]. Additional roles of NAT10 include ac⁴C modification of targets as the HSP90AA1 mRNA in ER stress-mediated metastasis and lenvatinib resistance in HCC [295], the CTC-490G23.2 lncRNA that interacts with PTBP1 to increase CD44 alternative splicing in primary ESCC [296], and the KIF23 mRNA to regulate the Wnt/β-catenin signaling pathway in colorectal cancer [294].

N7-methylguanosine (m⁷G) is another epitranscriptomic modification with critical roles in the regulation of gene expression that is dysregulated in several diseases, including cancer. m⁷G was originally identified in mRNAs, but later found at defined internal positions within multiple types of RNAs, including tRNAs, rRNAs, and miRNAs [357]. The most common m⁷G writer is METTL1, in complex with the non-catalytic subunit WDR4. In nasopharyngeal carcinoma (NPC), METTL1/WDR4 promotes tumorigenesis via regulation of Wnt/β-catenin, and their upregulation has been associated with poor prognosis [297]. Moreover, as previously commented for YTHDF1 [247], the same team found that insufficient radiofrequency ablation of HCC also results in a significant upregulation of METTL1/WDR4 and m⁷G levels in tRNAs that could promote metastasis by enhancing the translation of SLUG/SNAIL EMT regulators [298]. In addition, METTL1/WDR4-mediated m⁷G modification of tRNAs could promotes tumor progression and metastasis by enhancing the translation of a subset of oncogenic transcripts in HNSCC, including genes related to the PI3K/AKT/mTOR signaling pathway [299]; or by promoting the processing of miR-760 in an m⁷G modification-dependent way and indirectly degrading the tumor suppressor ATF3 mRNA via miR-760 in bladder cancer [300]. Besides, m⁷G modifications in 18S rRNA are introduced by the WBSCR22/TRMT112 methyltransferase complex. Decreased expression of WBSCR22 has been observed in inflammatory and neoplastic lung pathologies [358]. WBSCR22 enhances glucocorticoid receptor (GR) function through binding to the GR co-activator GRIP1. Its expression is decreased by cytokines TNFα and IFNγ by driving ubiquitination of two conserved lysine residues [358]. In pancreatic cancer, WBSCR22/TRMT112 negatively regulates the transcription of ISG15, diminishing proliferation, migration, invasion, and globally tumorigenesis [301]. Nevertheless, WBSCR22 overexpression has been described in glioma and correlates with poor prognosis [359].

Another methylation event that provides tRNA stability and enhance translation occurs at position 10 in tRNA, the N2-methylguanosine (m²G), introduced by TRMT11. Interestingly, TRMT11 gene fusion with GRIK2 (TRMT11-GRIK2) has been identified as a frequent event in several tumor types, including primary and lymph node metastatic cancer from breast, colon, and ovary [360]. GRIK2 encodes a glutamate receptor with tumor suppressor features. Chromosomal recombination between TRMT11 and GRIK2 destroys the open-reading frames of both genes and produces functional knockouts of these two proteins. Consequently, lack of TRMT11 may produce less efficient and unstable translation due to tRNA defects, or protein translation repression [360]; plus the loss of GRIK2 tumor suppressor activity. Further investigations are guaranteed to unravel the functional consequences of this genetic lesion.

Regarding modifications in uridine, the wobble uridine of specific tRNAs is transformed to 5-methoxycarbonylmethyl-2-thiouridine (mcm⁵s²U), which is critical for proper mRNA decoding and protein translation. ELP3 is the enzymatic core of the elongator complex that promotes this modification [361]. DNA hypermethylation-mediated decrease of ELP3 has been detected in invasive ductal breast cancer [362]. In contrast, examples of pro-metastatic roles of ELP3 have also been described. ELP3 expression has been correlated with lymph node metastasis in endometrioid adenocarcinoma [363]. In breast cancer, ELP3 together with its partners CTU1/2 are upregulated and sustain metastasis by driving the translation of the oncoprotein DEK in a mcm⁵s²-tRNA-dependent manner. Furthermore, DEK promotes the IRES-dependent translation of the pro-invasive transcription factor LEF1 [302]. Of note, DEK has been shown regulated by gene promoter hypomethylation in HCC [364]. Also, CTU1 was one of the nine genes used to generate a risk score model in prostate cancer that recognize patients with worse prognosis and higher proportions of lymph node metastasis [365]. CTU1 has also been included in two analogous risk score models of bladder cancer [366, 367].

Uridines are also target of pseudouridylation. Pseudouridine (Ψ) is the most abundant RNA modification and the first to be discovered, more than 70 years ago. DKC1 gene codes for Dyskerin, the catalytic subunit of H/ACA small nucleolar ribonucleoprotein (H/ACA snoRNP) complex, which catalyzes pseudouridylation of rRNA. Dyskerin also plays a role in telomerase stabilization. DKC1 expression has been related to unfavorable TNM stages in ccRCC [368]. In CRC, Dyskerin facilitates angiogenesis and metastasis via increased expression of HIF-1α and VEGF by a mechanism unrelated to its RNA-modifier role. Dyskerin directly bound to the HIF-1α promoter and enhance its transcription and promote CRC progression [303]. Moreover, a recent meta-analysis of nine studies confirmed the link between Dyskerin overexpression and metastasis [369]. In pancreatic ductal adenocarcinoma, SNORA23, an H/ACA-box type small nucleolar non-coding RNA participating on pseudouridylation, was found upregulated and associated with metastasis formation by increasing the expression of spectrin repeat-containing nuclear envelope 2 (SYNE2) [370]. In addition, in gastric cancer, Liu et al. found that the overexpression of SNORA21 was associated with increased lymph node metastasis and distant metastasis [371]. However, these studies did not determine if SNORA23 pseudouridylation activity was involved in SYNE2 expression or the potential targets of SNORA21. In ovarian cancer, SNORA70E promotes tumorigenesis through pseudouridylation of RAP1B but direct correlation with metastasis was not established [372]. Thus, further studies are needed to uncover a potential deregulation of pseudouridylation in metastasis.

About 2′-O-methylation (Nm, where N stands for any nucleotide), it is an abundant and highly conserved modification found at multiple locations in tRNAs, rRNAs, mRNAs, and small non-coding RNAs [373]. Fibrillarin (FBL) is a 2′-O-methyltransferase that binds to box C/D snoRNAs, like SNORD89, to form a snoRNP complex and targets rRNA, tRNA, and other RNAs. For instance, a recent study has shown that SNORD89 combined with FBL increased the level of 2′-O-methylation of BIM mRNA, affecting the complementary base pairing of BIM mRNA, thus reducing the pro-apoptotic BIM protein [304]. Moreover, SNORD89 is overexpressed in endometrial carcinoma with lymph node metastasis [304]. Also, FBL is highly expressed in HCC and correlates with lung metastasis and poor prognosis [374]. In NSCLC, SNORD88C was observed upregulated and associated with metastasis. The authors have shown that SNORD88C guides 2′-O-methylation of 28S rRNA that increases SCD1 translation and consequently upregulates MUFA expression, leading to autophagy inhibition and promoting growth and metastasis [375].

RNA editing is another epitranscriptomic event that affects RNA molecules. RNA editing changes the RNA sequence, and consequently it can dramatically alter the amino acid composition and properties of the mRNA-derived protein. Recent studies indicate that many thousands of transcripts are affected by RNA editing, including not only mRNAs but also non-coding RNAs. Two families of proteins carry out editing by deamination: the adenosine deaminases acting on RNA (ADARs), which convert adenosine to inosine (A-to-I); and the apolipoprotein B mRNA editing (APOBECs), which convert cytosine to uracil (C-to-U) [376].

Regarding the A-to-I editing, it is catalyzed by ADAR1, ADAR2, and ADAR3. Inosines are read as guanosines during translation, which has important functional implications. Pro-metastatic roles for ADAR1 have been described in several studies. In HCC, ADAR1 upregulates integrin ITGA2 and drives metastasis by enhancing adhesion of tumor cells to the extracellular matrix [305]. Moreover, aberrant overexpression of ADAR1 in gastric cancer promotes metastasis via mTOR/p70S6K/RPS6 axis activation [306]. In TNBC, ADAR1 high expression and low tumor-infiltrating lymphocytes define worse disease-free survival in patients with lymph node metastasis [377]. In CRC, high ADAR1 expression increases editing of AZIN1. Importantly, elevated level of AZIN1 editing has been identified as an independent risk factor for lymph node and distant metastasis, and as a prognostic factor for overall survival and disease-free survival in CRC patients [307], later confirmed in gastric cancer by the same group [308]. Also, it has been shown that CPEB3 interacts with the 3′UTR of ADAR1 mRNA and inhibits its translation by localizing it to processing bodies (P bodies). CPEB3 is downregulated in gastric cancer. Remarkably, CPEB3 re-expression inhibited gastric cancer growth and metastasis by decreasing ADAR1 expression, opening a therapeutic opportunity [378]. In intrahepatic cholangiocarcinoma, ADAR1 overexpression and concomitant KPC1 overediting lead to distant metastasis [309]. ADAR1 has also been linked to circular RNAs. In gastric cancer, hsa_circ_0004872 circRNA was downregulated by ADAR1 and associated with tumor size and local lymph node metastasis [379]. Next, the circNEIL3, upregulated in PDAC, enhances ADAR1-mediated editing of GLI1 by sponging the miR-432-5p that targets ADAR1 [310]. This mechanism promoted cell cycle progression and EMT. Moreover, high circNEIL3 levels were associated with poor prognosis or TNM stage [310]. ADAR1 has also been suggested as a therapeutic target against gastric cancer metastasis, since ADAR1 knockdown suppressed peritoneal metastasis of gastric cancer in mouse models partially by inhibiting Wnt/β-catenin pathway and EMT [380]. Very recently, Ding et al. designed a nanovesicle as RNA interference tool to overcome resistance to immune checkpoint blockade therapies [381]. The vesicle with siADAR1 silenced tumor cell ADAR1 expression and increased type I/II interferon production, making cells more sensitive to secreted effector cytokines such as IFN-γ with significant cell growth arrest [381].

Although most of the studies have described pro-tumorigenic and metastatic roles for ADAR1, a low expression of ADAR1 has been observed in melanoma metastasis. In this setting, ADAR1-mediated editing of miR-455-5p inhibits metastasis, while unedited miR-455-5p form promotes melanoma metastasis through inhibition of the CPEB1 tumor suppressor gene [311]. Also in melanoma, ADAR1-mediated editing of miR-378a-3p prevents melanoma progression by targeting the PARVA oncogene [312]. Moreover, the same group described a role of ADAR1 in melanoma cell invasion by controlling ITGB3 expression independently of RNA editing [313], regulated by miR-30a and miR-30d [382]. In breast cancer, it was shown that GABAA receptor alpha3 (GABRA3), normally expressed in brain, is upregulated and drives cancer progression and metastasis. Interestingly, the A-to-I edited form of GABRA3 had opposite properties [314]. ADAR1 could also downregulate circRBMS3, a circRNA that promotes osteosarcoma progression by sponging miR-424-5p and protecting YRDC/eIF4B [315]. Thus, ADAR1 downregulation in osteosarcoma could be at least partially responsible for circRBMS3 overexpression, enhancing proliferation and metastasis [315].

Regarding ADAR2, in liver metastasis of CRC, ADAR2 has been identified as a direct substrate of the tumor suppressor protein kinase PKCζ [316]. The authors showed that PKCζ-mediated ADAR2 phosphorylation activates its gene editing activity, which is required to maintain miR-200 steady-state levels. Remarkably, loss of PKCζ in metastasis is associated with lower ADAR2 activity, which promotes the loading of miR-200s into extracellular vesicles, thereby decreasing the intracellular steady-state levels of miR-200s promoting an EMT/stemness phenotype and metastasis [316]. ADAR2 has also been described as driver of invasion and metastasis in esophageal cancer by editing and downregulating the tumor suppressor SLC22A3 [317]. Moreover, the role of androgen receptor (AR) in prostate and bladder cancer invasion linked to ADAR2 has been identified [318, 319]. In prostate cancer, R-2HG accumulation in IDH1 mutated cells increases cell invasion via enhancing TGFβ1/p-Smad2/3 signaling, and AR can reverse this mechanism. AR expression increases ADAR2 expression and negatively regulates circRNA-51217 production and TGFβ1/p-Smad2/3. Restoring AR in cancer cells increases ADAR2 expression and diminishes circRNA generation, which finally resulted in interruption of R-2HG effect [318]. Moreover, this study highlights a therapeutic potential of bipolar androgen therapy (BAT) for patients with IDH1 mutation and low AR expression [318]. In bladder cancer, AR increases the levels of circFNTA to promote cancer cell invasion and cisplatin chemoresistance [319]. In this case, AR strikingly downregulates ADAR2 by binding its 5′ promoter region to increase circFNTA levels, which sponges miR‐370‐3p and increased expression of its host gene FNTA. High level of FNTA activates KRAS to promote cell invasion and chemo‐resistance. Thus, depletion of circFNTA suppressed cancer cell invasion and increased cisplatin chemo‐sensitivity [319].

The C-to-U editing process, driven by APOBEC enzymes, is also relevant in metastasis. APOBEC3G promotes CRC hepatic metastasis through inhibition of miR-29-mediated suppression of MMP2 [320]. Overexpression of APOBEC3G also correlates with worse prognosis and increased risk of hepatic metastasis in patients with CRC [383], and might be a risk factor for HCC progression [384]. Interestingly, it has been shown that APOBEC3G increased expression in cancer cells repressed the expression of the tumor suppressor KLF4 by binding to its mRNA [321]. In parallel, SP1 was seen overexpressed and controlling the increased expression of c-myc, Bmi-1, BCL-2, and MDM2, and decreased of p53, supporting its oncogenic role [321]. Moreover, Zhou and Guo observed that APOBEC3G is increased in CRC metastasis by copy number variation [385]. In melanoma, APOBEC3G was recently identified within a signature of genes modulated by combined inhibition of PLK1 and NOTCH [386], providing a potential therapeutic opportunity. Finally, the Apobec-1 complementation factor (A1CF) promoted progression of RCC through DKK1-MEK/ERK signaling pathway [387], without assessing RNA edition. The same group also described that A1CF promotes RCC cell migration by promoting nuclear translocation of SMAD3 [388]. A1CF overexpression correlated positively with SMAD3, Snail1, and N-cadherin expression [388]. These two studies investigated the role of a co-factor of APOBEC1, suggesting that complementary studies are needed to determine if edition is altered in RCC.