Abundance and subcellular location of ceRNA components

Several studies18 ,19 using computational models have explored the molecular determinants of ceRNA activity at steady-state or dynamic conditions in response to molecular perturbations. In a study18 using a mathematical mass-action model and experimental validation, for example, the relative abundance and binding affinity of ceRNAs and miRNAs, number of MREs and interactions with other ceRNAs were shown to have influence on ceRNA activity. In another study19 using a minimal rate equation-based model to characterise ceRNA interactions at steady-state condition, similar results were yielded. Both studies18 ,19 revealed that ceRNA activity approaches optimal state when the abundances of miRNA and ceRNA are near equimolarity. One possible explanation is that competing RNAs would have little sequestrating effect on a highly abundant miRNA, and on the contrary, miRNAs with very low abundance were unlikely to mediate active ceRNA interplay because the number of miRNAs available for target repression is too limited.

Given the central role of Argonaute protein in miRNA-mediated gene expression regulation,2 its abundance may also influence ceRNA activity. In a study14 based on mathematical model, for example, small RNAs were shown to competitively bind to Argonaute protein, with lower abundance of Argonaute-inducing stronger competition.

Another possible influential factor is the location of ceRNA components. On the one hand, owing to the cell-type and tissue-specific manner of miRNA expression,20 some ceRNAs may be active only in specific cells where the shared miRNAs are available. On the other hand, the subcellular location of ceRNA components21–23 can affect the accessibility of ceRNAs to shared miRNAs and may thus impact ceRNA activity.

miRNA/ceRNA affinity

Given that each miRNA may be competitively bound by many transcripts, the competing effects of these transcripts are influenced by their respective binding affinities to miRNAs, with higher affinity enabling stronger competition ability in general.24 ,25 When the miRNA:target ratio is at a relatively low level, targets with low affinity are mostly unbound and thus have little competitive effects, whereas high-affinity targets can effectively sequester miRNAs and thus may serve as active ceRNAs. As the miRNA:target ratio increases, miRNA binding spreads gradually to lower-affinity sites, and therefore, the size of the total target pool becomes larger and larger. When the total target pool becomes large enough, it will be insusceptible to physiological levels of competing RNA even with high affinity.

The binding affinity of miRNA and its RNA target is mainly influenced by matching between MREs (on miRNA target) and seed regions (on miRNA).3 ,26 In general, 6-mer seed match has low affinity but high abundance, whereas 7-mer and 8-mer seed match has high affinity but low abundance10 ,26–28 (figure 2). The matching between MREs and seed regions, in turn, is subject to influences of multiple factors such as single-nucleotide polymorphism (SNP)29–31 and RNA alternative splicing.32–34 SNPs in miRNAs or their targets can affect their binding affinity and may even alter the target pool.29–31 It was estimated that alternative splicing occurs in 95% of multi-exon genes32 and there are an average of 6.3 alternatively spliced transcripts per gene locus.35 During tumorigenesis, many alternative splicing events are deregulated. In ovarian and breast cancers, for example, approximately half of alternative splicing events were found to be different from those in normal tissues.33 In another study,34 3′ untranslated regions (UTRs) shortening by alternative splicing was found to be widespread in multiple cancer cell lines. Given that the majority of MREs reside in the 3′UTRs, the deregulated alternative splicing in these cancer cell lines may change the MRE profile and thus the affinity of these transcripts to miRNAs.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Schematic of different types of seed matching. Binding affinity increases from the top to the bottom. Double-sided arrows indicate base pairing. This figure is partially based upon studies by Friedman et al,10 Grimson et al27 and Nielsen et al.28 miRNA, microRNA.

RNA editing and RBPs

RNA editing is a post-transcriptional process that leads to insertion, deletion or base substitution of nucleotides within the edited RNA, rendering the RNA sequence partially different from its DNA template.36 ,37 RNA editing occurs in both miRNAs38 and their RNA targets. In a study on miRNA editing in human brain, up to 16% of human primary miRNAs were predicted to be edited by adenosine deaminase, RNA-specific, that converts adenosine to inosine (A>I RNA editing).38 Moreover, studies showed that up to 85% of pre-mRNAs might be subject to A to I editing, and most of these editing sites were in introns and UTRs.39 ,40 Editing of miRNAs may alter their seed regions and thus their target spectrum,38 ,41 whereas editing of ceRNAs may create, modify or destroy miRNA binding sites. These alterations may change the binding spectrum and affinity of miRNAs and ceRNAs and add a new layer of complexity to ceRNA network.

RBPs participate in multiple post-transcriptional regulations, such as RNA splicing, stability and degradation.42 ,43 Accordingly, they might serve as potential mediators of ceRNA activity in several ways. Previous studies in cancer showed that RBPs can hamper miRNA-target binding by occupying MREs16 ,17 or conversely promote miRNA-target binding via recruiting miRNA to the target,44 indicating their possible involvement in ceRNA network. However, the exact roles of RBPs in regulating ceRNA activities remain largely unknown and are interesting areas for future investigation.

Pseudogenic RNAs as ceRNAs

Pseudogenes are conventionally defined as genomic loci that derive form and thus resemble real genes but are unable to encode functional proteins due to mutations.6 Although once regarded as non-functional junk DNA,46 recent studies revealed that some pseudogenes are aberrantly expressed in cancer47 ,48 and that pseudogenic RNAs may function in tumorigenicity as antisense RNAs, endogenous small-interference RNAs or ceRNAs.49 ,50

Owing to the high homology of pseudogenes and their parental genes, there are many identical MREs in pseudogenic RNAs and their parental RNAs, which therefore may form ideal ceRNAs pairs. In a recent bioinformatics study49 in breast cancer, >100 pseudogenic RNAs were predicted to hold ceRNA potential.

PTENP1 is a processed pseudogene (derived from mature mRNAs that are retrotranscribed and integrated into new loci) of PTEN, a famous oncosuppressor. Within the 3′UTR of PTENP1, conserved binding sites for PTEN-targeting miR-17, miR-21, miR-214, miR-19 and miR-26 families were found.6 The interactions between PTENP1 transcripts and these miRNAs were subsequently validated by luciferase assay. In DU145 prostate cancer cells, these miRNAs suppressed both PTEN and PTENP1 levels and inhibition of these miRNAs showed a de-repressive effect.6 Overexpression of PTENP1 3′UTR increased PTEN expression, leading to inhibited cancer cell growth and colony formation, whereas PTENP1 knockdown decreased PTEN expression. Moreover, the de-repressive effect of PTENP1 3′UTR on PTEN was blunted in DICER-null colon cancer cells, indicating that this effect is miRNA-dependent given the indispensable role of DICER enzyme in the maturation of the vast majority of miRNAs.6 Similar phenomenon was observed regarding oncogene KRAS and its psueudogenes KRAS1P, in which KRAS1P 3′UTR overexpression in prostate cancer cells elevated KRAS mRNA level and accelerated cell growth,6 indicating that pseudogenes may mirror the functions of their parental genes via ceRNA cross-talk. However, it is worth noting that it remains unclear whether these functional consequences observed in overexpression experiments can faithfully reflect those of spontaneous ceRNA interplay during tumorigenesis.

A recent study by Karreth et al51 showed that pseudogene overexpression was sufficient to induce malignant transformation in mice. In this study, in vivo overexpression of Braf-rs1, the murine pseudogene of B-Raf, increased the abundance of B-Raf and its downstream effector pERK, and resulted in the formation of diffuse large B cell lymphoma (DLBCL) in mice. Subsequent mechanism analysis showed that in human and murine cancer cells pseudogene BRAFP1 (or B-Raf) mRNA and its parental gene BRAF (or Braf-rs1) mRNA regulate each other possibly through ceRNA interplay.51 To our knowledge, this is the first to show the malignant-transforming ability of pseudogene overexpression in vivo. However, it also raised some concerns. In mouse, the physiological expression of Braf-rs1 was 6-fold to 115-fold lower than that of B-Raf, and in human DLBCL tumours, the abundance of BRAFP1 RNA was found to be much lower than that of BRAF mRNA.51 Whereas in this study,51 Braf-rs1 was overexpressed to a level comparable to that of B-Raf. Therefore, the transforming effect of Braf-rs1observed in this study may not faithfully reflect the real function of Braf-rs1in spontaneous tumorigenicity.

Other examples of ceRNA pairs comprising pseudogenic and parental RNAs include PTENP1 and PTEN via miR-21 in renal cancer,52 OCT4-pg4 and OCT4 via miR-145 in hepatocellular carcinoma,53 and CYP4Z2P and CYP4Z1 in breast cancer via multiple miRNAs.54

LncRNAs as ceRNAs

LncRNAs are transcripts of >200 nt in length without protein-coding function that arise from intergenic, antisense or promoter-proximal regions.55 ,56 LncRNAs can play multifaceted roles in both health and diseases including cancer.54 Recent evidence shows that their functionalities in tumorigenesis may be partially mediated by ceRNA cross-talk.57–59

LncRNA highly upregulated in liver cancer (HULC), as its name implies, is a lncRNA with upregulated expression in liver cancer. In liver cancer cells, the binding of miR-372 to HULC and to PRKACB mRNA was predicted by bioinformatics and validated by luciferase assays.7 Knockdown and overexpression experiments revealed a positive correlation between HULC and PRKACB expression, leading to the speculation that HULC and PRKACB transcripts may regulate each other via competing for miR-372 in liver cancer.7 Moreover, this study demonstrated an interesting positive regulatory loop, in which HULC upregulated PRKACB expression, which promoted phosphorylation of CREB. The latter in turn bound to the promoter region of HULC and promoted its expression.7

Hox transcript antisense intergenic RNA (HOTAIR) is an oncogenic lncRNA in many cancers.60 A recent study57 revealed that HOTAIR upregulation was correlated with unfavourable prognosis of gastric cancer. The oncogenic role of HOTAIR in gastric cancer was confirmed by overexpression and knockdown experiments both in vitro and in vivo. Luciferase assay showed that HOTAIR expression was suppressed by miR-331-3p and this suppression was abrogated by mutation of miR-331-3p binding site in HOTAIR sequence. HER2, a gene encoding a transmembrane protein functioning in gastric carcinogenesis and resistance to trastuzumab-based therapy,61 was also found to be a target of miR-331-3p by luciferase assay. The expressions of HER2 and HOTAIR were positively correlated and were both inhibited upon ectopic expression of miR-331-3p.57 These results indicated that regulating HER2 level via ceRNA interplay might be one of the mechanisms underlying the oncogenic roles of HOTAIR in gastric cancer. Using similar experimental protocols, Liang et al.62 reported that the oncogenic functions of lncRNA H19 in colorectal cancer (CRC) might be attributable to its ceRNA activity to sequester miR-138 and miR-200a and therefore upregulate expressions of VIM, ZEB1 and ZEB2, the critical genes involved in epithelial mesenchymal transition (EMT).

ceRNA potentials of circRNAs

First reported in mouse testis >20 years ago,63 circRNAs are a kind of transcripts whose 3′- and 5′-ends are joined covalently to form a closed continuous loop.64 circRNAs have multiple origins. The majority of them are originated from exons of coding regions, the rest minority from 3′UTR, 5′UTR, introns, intergenic regions or antisense RNAs.65 As stated above, they are inneglectable constituents of transcriptome in that they were estimated to account for at least 10% of the transcripts derived from their genomic loci.45

Many circRNAs shows a tissue-specific and developmental stage-specific expression pattern.66 A recent study67 revealed that the global abundance of circRNAs was lower in CRC relative to normal tissues. In gastric cancer, the abundance of a circRNA called Hsa_circ_002059 was also found to be lower in cancerous versus para-cancerous tissues.68 Another recent study reported the presence of circRNAs in exosomes (small membrane vesicles containing endocytic origins secreted by cells) and found that the profiles of seral exosomal circRNAs can distinguish patients with CRC from healthy controls.65

CircRNAs are highly homologous to but generally more stable than their linear counterparts because they lack accessible ends and thus are resistant to exonucleases.66 These natures of circRNA lead to the speculation that they may form active ceRNA pairs with their linear counterparts. A circRNA called circular RNA sponge for miR-7 (ciRS-7), for example, was found to contain >70 MREs for miR-7 and can effectively suppress miR-7 activity and therefore affect its target gene expression.66 ,69 As miR-7 regulates the expression of several oncogenes, whether ciRS-7 can modulate oncogene expression through miR-7-mediated ceRNA interplay is an interesting area for future study.70 However, the ceRNA potentials of circRNAs should not be overestimated from the ciRS-7 example because a recent bioinformatics study45 showed that in the 7112 annotated human circRNAs, only two (one of them is ciRS-7) harbour more MREs than expected by chance. This means that the strong sponging effect of ciRS-7 on miR-7 may be an individual case rather than a general scenario for circRNAs.

ceRNA pairs composed of mRNA-mRNA

Besides the aforementioned non-coding RNAs, protein-coding mRNAs that contain MREs on their 3′UTR are also potential participants in the ceRNA cross-talk. ceRNA network confers non-coding functions to protein-coding mRNAs and provides a mechanism in which coding and non-coding RNAs cross-regulate each other.6 ,12 ,71–80

PTEN mRNA, for example, can form ceRNA pairs not only with its pseudogenic RNA (PTENP1)6 but also with other mRNAs.12 ,71 ,72 In human DU145 (prostate cancer) and human HCT116 (colon cancer) cell lines, PTEN, VAPA and CNOT6L mRNAs were targeted by common pools of miRNAs.12 PTEN and VAPA/CNOT6L reciprocally regulated each other, and this regulation was abolished in DICER-mutant HCT116 cells, indicating its miRNA dependence. Functional investigation revealed that knockdown of VAPA or CNOT6L downregulated PTEN level, activated the PI3K/AKT pathway (downstream effector of PTEN) and resulted in increased malignancy, which were again failed to be observed in DICER-mutant HCT116 cells.12 Interestingly, ectopic expression of 3′UTRs of PTEN, VAPA or CNOT6L also led to growth inhibition of DU145 cells,12 indicating that these oncosuppressive effects were at least partially coding independent. In melanoma cells,71 ZEB2 mRNA, whose protein product is an established activator of the EMT, was found to be targeted by a set of PTEN-targeting miRNAs and to regulate PTEN expression in a manner of miRNA-dependence (validated in DICER-mutant HCT116 cells) yet coding-independence (evidenced by the upregulation of PTEN by ZEB2 3′UTR ectopic expression), indicating the ceRNA interaction between PTEN and ZEB2 mRNAs. A study72 in glioblastoma cells identified 13 PTEN ceRNAs or miR-mediated post-transcriptional regulators using transcriptome analysis. Locus deletions of these ceRNAs or regulators downregulated PTEN expression and promoted glioblastoma cell growth.72 Moreover, the authors analysed post-transcriptional regulations of PTEN beyond binary ceRNA pairs and revealed that PTEN ceRNA interplays were not standalone but rather part of complicated regulatory networks mediated by >248 000 miRNAs.

Another reported PTEN ceRNA is the VCAN 3′UTR, which sequester miR-144 and miR-136 (two shared miRNAs) to de-repress PTEN mRNA translation in murine breast cancer.73 VCAN 3′UTR was also found to regulate RB1 expression possibly via competing for miR-199a-3p and miR-144.73 Overexpression of VCAN 3′UTR led to simultaneous upregulation of Rb1 and PTEN and reduced malignancy.73 In HepG2 cells,74 3′UTRs of VCAN, CD34 and FN1 were found to be targeted by same pools of miRNAs, which were confirmed by MREs mutation and luciferase assay experiments. Ectopic VCAN 3′UTR expression sequestered CD34- and FN1-targeting miRNAs (miR-133a, miR-144, miR-431 and miR-199a-3p) and upregulated expression levels of CD34 and FN1,74 pointing to possible ceRNA interplays among the three. Interestingly, overexpression of VCAN 3′UTR in transgenic mouse was sufficient to induce hepatocarcinogenesis.74 To our knowledge, this is the first study to demonstrate the carcinogenic effect of ceRNA perturbation in vivo. And once again we should be aware that the phenotypic effects of ceRNA ectopic overexpression should not be taken as bona fide effects of spontaneous ceRNA interplay in tumorigenicity.

CD44 mRNA is another one having multiple reported ceRNAs in cancer.75 ,76 However, its phenotypic effects in tumorigenicity seem to some extent controversial. In a human breast cancer cell line MT-1,75 overexpression of CD44 3′UTR retarded proliferation, colony formation and tumour growth, at least partially owing to its de-repressive effect on CD44 and CDC42 via sequestering miR-216a, miR-330 and miR-608. In another human breast cancer cell line MDA-MB-231, however, its overexpression promoted cell motility, invasion and adhesion in vitro and metastasis in vivo.76 Subsequent mechanism exploration revealed that CD44 3′UTR shares miR-328 with COL1A1 and shares miR-491, miR-512-3p and miR-671 with FN1. Overexpression and knockdown experiments showed that CD44 3′UTR can regulate the expressions of FN1 and COL1A1 through competing for these shared miRNAs.76 The discrepancies in these results may root in different cell types used in this two study75 ,76 or may reflect the distinct effects of CD44 3′UTR on cancer cell growth and on invasion.

HMGA2, which encodes a non-histone chromosomal high-mobility-group protein, was recently reported to function as ceRNA to sequester miRNA let-7 family members and upregulate TGFBR3, the transforming growth factor (TGF)-β co-receptor, thereby activating the TGF-β signal pathways and promoting lung carcinogenesis.77 These examples indicate that besides functioning in cancers through their protein products mRNAs can also function in a coding-independent manner as ceRNAs.

Controversies on current ceRNA theory and possible directions

Despite the increasing number of studies that report the phenotypic effects of ceRNA deregulation in cancer, the current ceRNA theory remains controversial regarding to whether they are really active in the endogenous cellular context.13 ,15

Although mathematical models18 ,19 predicted that ceRNA activity approaches optimal when the miRNA and targets are near equimolar concentration, it is still unclear whether endogenous miRNA and targets concentrations fall into these susceptible regimes. A recent quantitative study15 showed that the ceRNA activity of ALDOA mRNA (a validated miR-122 target) was detectable in liver cells only when ALDOA transcripts abundance increased from 3.3×10³ (physical levels) to 1.5–2.7×10⁵ per cell (45-fold to 82-fold increase), which is beyond the abundance range of any endogenous transcripts. Moreover, in vivo experiments showed that up to 23-fold increase of ALDOA abundance did not result in detectable miR-122 inhibition or de-repression of miR-122 targets.15 This is in agreement with the result of another study,13 revealing that active ceRNA interplay was observed only when a large number of ectopic ceRNA transcripts were introduced. This is possibly due to that ceRNA need to compete with the whole target pool of the miRNA species to titrate the latter. To win this ‘one versus multiple’ battle, the ceRNA abundance must be upregulated to an aberrantly high level comparable to the sum of whole target pool of the miRNA species. Therefore, although ceRNA interplays are possibly widespread,72 their activity might be rather week in the real cellular context.15

On the other hand, there are also evidences that support the activity of ceRNA interplay in endogenous cellular context. As mentioned above, miRNA and target binding is influenced not only by their relevant abundance18 ,19 but also by the characteristics of MREs.3 ,10 In general, MREs with high affinity were favourably bound and thus have stronger competition potency,24 ,25 indicating that for high-affinity ceRNAs meaningful competition may occur without approaching the sum abundance of all its competitors in the transcriptome. Moreover, during tumorigenesis, the abundance of some transcripts may increase remarkably84 and thus contribute substantially to the total target pool. These highly abundant transcripts during tumorigenesis might serve as effective ceRNAs.

Technological advances in epigenetic modulation may help to settle this controversial issue in the future. For example, CRISPR-Cas9-mediatd epigenetic editing85 ,86 can be used in the future to regulate ceRNA expression to levels similar to those observed in spontaneous tumorigenicity so as to observe their functional consequences. Unlike complementary DNA-mediated overexpression that often approaches an aberrantly high level, catalytically dead Cas9-mediated transcriptional activation regulates gene expression from endogenous loci and thus can provide a more faithful recapitulation of the ceRNA deregulation in cancer.

Delineating ceRNA network in the context of post-transcriptional regulation

Increasing evidence shows that ceRNA interplays are part of the entire post-transcriptional regulation networks rather than standing alone.14 ,72 ,83 ,87

Various factors such as RBPs16 ,17 ,44 and transcription factors (TFs)18 ,87 might interact with ceRNAs to add to the complexity. For example, computational analysis87 revealed that miRNAs and ceRNAs regulatory patterns might be altered by their interactions with TFs. Besides the competition among miRNA targets for shared miRNAs, miRNAs that share common targets can also compete with each other for target binding.14 ,83 In this way, individual ceRNA pairs might be cross-linked to each other to form a whole regulatory network. Given this, we recommend to extend the term ceRNA to include both miRNA targets and miRNA themselves. A recent study83 based on bioinformatics analyses of confirmed miRNA-mRNA interactions followed by experimental validation demonstrated that expression perturbations of one gene can potentially propagate through a cross-linked ceRNA and miRNA network with a marginal deceasing effect, leading to mutual regulation between distant ceRNAs that do not share common miRNAs (figure 3). A study72 in glioblastoma revealed that PTEN ceRNA interplays were part of complicated regulatory networks involving >248 000 miRNAs.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Schematic presentation of competing endogenous (ceRNA) propagation. Expression perturbation of one gene can propagate through a cross-linked ceRNA and microRNA (miRNA) network with a marginal deceasing effect, leading to mutual regulation between distant ceRNAs that do not share common miRNAs. (A) miRNAx, miRNAy and miRNAz targets ceRNA₁ and ceRNA₂, ceRNA₂ and ceRNA₃, ceRNA₃ and ceRNA₄, respectively. (B) Overexpression of ceRNA₁ sequesters miRNAx and hence de-represses ceRNA₂. (C) The increased expression of ceRNA₂ binds more miRNAy and therefore free ceRNA₃ from suppression. (D) Similarly, de-repressed ceRNA₃ leads to increased expression level of ceRNA₄. This figure is partially based on study by Nitzan et al.83

Therefore, how to predict, validate and functionalise ceRNA interplay in the context of entire post-transcriptional regulatory network in a dynamic way is an area of future investigation with great challenge and interest.

Potentials of ceRNA as diagnostic markers or therapeutic targets in cancer

As cancer transcriptome differs significantly from that of its normal counterpart,84 ceRNA profile in cancer may also differ from that in normal state. Increasing evidence shows that ceRNA abundance and activity are deregulated or reprogrammed in cancer.7 ,45 ,46 ,57 ,72 ,88 A recent bioinformatics study87 revealed a rewiring in the ceRNA profile in breast cancer tissue compared with matched normal tissue, with some ceRNAs active in caner but not in normal cells and vice versa. For example, lncRNA PVT1, whose overexpression shows antiapoptotic effect in breast cancer,89 was found to act as ceRNA in the normal but not in breast cancer cells.87 These discriminating ceRNA profiles in cancer and in normal cells hold the potential to be developed as diagnostic markers in the future.

MiRNAs are involved in the pathogenesis of a diversity of human diseases including cancer90 and have been proposed as promising anticancer therapeutic targets.91 Given that miRNAs lie in the central of ceRNA interplay, it is reasonable for us to envision the potentials of cancer-related ceRNAs as therapeutic targets. For example, artificial miRNA sponges containing multiple copies of MREs in tandem have been used to inhibit oncogenic miRNAs and showed tumour suppressive effects in multiple cancer cell lines or xenograft animal models such as Ewing sarcoma,92 renal cancer,93 liver cancer94 and lung cancer.95 ,96

However, there are several challenges lying in the way of turning the idea into reality. First, ceRNAs with potential as therapeutic targets should be ceRNA hubs or essential nodes that have a high number of links to post-transcriptional regulation elements, because targeting non-essential nodes might be circumvented by cancer cells through resorting to alternative pathways.97 However, how to identify these hubs or essential nodes? Second, how to specifically manipulate cancer-causing ceRNAs while sparing the others? This is a rather challenging task given that ceRNAs are cross-linked with each other to form a multilayered yet holistic regulatory network.72 ,83 Non-specific manipulation of ceRNA network is dangerous due to the possibility of altering otherwise normal gene expression in an unpredictable manner. Third, methods to deliver the ceRNA therapeutic agents into cancer cells effectively and specifically are still to be developed. Vehicles that have been developed for gene therapy delivery98 ,99 can be adapted in the future for the delivery of ceRNA therapeutic agents.

In conclusion, ceRNA cross-talk is a kind of post-transcriptional regulation that is mediated by miRNAs and links the functions of coding and non-coding RNAs. ceRNA activity is influenced by multiple factors such as the abundance and subcellular location of ceRNA components, miRNA/ceRNA affinity, RNA editing and interactions with RBPs and TFs. Some ceRNAs are deregulated or show a cancer-specific profile in cancer and might play a role in cancer initiation and progression, lending support to the speculation that they might be explored as diagnostic markers or therapeutic target in the future. However, many of the phenotypic effects of ceRNAs deregulation observed in current studies were induced by overexpression and/or knockdown experiments, raising concern about whether they can faithfully reflect real functions of ceRNAs in spontaneous tumorigenicity. Future efforts should be made to faithfully mimic endogenous ceRNA deregulation and to functionalise ceRNA interplay in the context of the whole post-transcriptional regulation.