This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.
MicroRNAs are small non-coding RNA molecules of 21-23 nucleotides that control gene expression at the post-transcriptional level and have been shown to play a vital role in a wide variety of biological processes (Ambros, 2004). Moreover, dysregulated expression of miRNAs is found in many pathologies, including cancer and cardiac disease (Lujambio & Lowe, 2012; Small & Olson, 2011). Understanding the mechanism of action and defining functional mRNA targets of a specific miRNA is essential to unravel its biological function and to develop therapeutic opportunities. This review summarizes the current understanding of the mechanistic aspects of miRNA-induced gene repression and focuses on the different approaches for miRNA target identification that have been proposed in recent years.
MiRNAs are a large family of ~22 nucleotides RNAs which regulate virtually every aspect of biology, including development, proliferation, differentiation or metabolism. It is then not surprising that disruption of miRNA function contributes to many human diseases, including cardiovascular disorders, cancer and neurological dysfunctions (Mendell & Olson, 2012).
MicroRNAs are processed from precursor molecules (pri-miRNAs) which are normally transcribed by RNA polymerase II (Lee et al, 2004). In animals, the majority (80%) of miRNA genes are located in introns of both protein-coding and non-coding genes (Rodriguez et al, 2004). Thus, expression of a large subset of mammalian miRNAs may be transcriptionally linked to the expression of other genes, allowing for coordinate regulation of miRNA and protein expression (Kim et al, 2009). The pri-miRNA hairpin is processed into the mature miRNA in a two-step process. Briefly, in the nucleous, the RNAse III enzyme DROSHA, in complex with other proteins such as DGCR8 in mammals, cleavage the pri-miRNA into a ~70 nucleotide precursor (pre-miRNA) which is transported to the cytoplasm by exportin 5. Once in the cytoplasm, the pre-miRNA is further processed by another RNAse III enzyme- DICER acting in conjunction with TRBP in mammals. As a result of this second cleavage a small RNA duplex (~20 nucleotides) is generated. In general, one of the strands (known as the guide) will be incorporated into a miRNA-induced silencing complex (miRISC) while the other one (passenger strand) is released and degraded (Review in (Yang & Lai, 2011). Further studies have confirmed that, in some cases, the passanger strand can also be loaded into miRISC, functioning as a mature miRNA (Chiang et al, 2010).
The core of miRISC is formed by the Argonaute proteins. Different members of this family have specific expression patterns, binding partners and biochemical capabilities (Czech & Hannon, 2011)). Indeed only Ago2, out of the four Ago proteins expressed in humans, is capable of mediating endoneoucleolitic cleavage of the target mRNA which occurs when complementarity between a miRNA and a target site is complete (Wang et al, 2009). Nevertheless, most metazoan miRNAs direct RISC to target mRNAs by interacting with sites of imperfect complementarity. As a result, the miRNA promotes the degradation and/or inhibit the translation of the target mRNA, resulting in repression of its expression (Pasquinelli, 2012).
In general, the most important region for target recognition comprises the nucleotides 2-8 of the miRNA -known as the "seed" region- and binding sites located in the 3'UTR of the cognate mRNAs are more common (Bartel, 2009). Despite of that, many examples have emerged recently showing that the miRNA binding sites can be located outside the 3'UTR (commonly in the coding region) or where lacking of perfect seed pairing is compensated by 3' complementarity or centered pairing (Hafner et al, 2010; Lal et al, 2009; Shin et al, 2010). Thus predicting whether a mRNA will be regulated by a given miRNA in the endogenous context is challenging and gets even more complicated when taking into account that different factors can also control the ability of a miRISC to bind and repress specific targets (Kedde et al, 2007; Kundu et al, 2012).
During the last decade, several bioinformatic tools have been developed aiming to predict miRNA targets (Witkos et al, 2011)). Most of them use algorithms based in the seed pairing and evolutionary conservation and typically predict hundreds to thousands of targets for each miRNA, including a high proportion of not bona-fide candidates (Alexiou et al, 2009).
Complexity of identifying endogenous miRNA targets has resulted in the generation of a wide variety of experimental approaches, from identification of mRNAs/proteins missregulated by overexpressing or antagonizing a given miRNA to genome-wide approaches including immunoprecipitation of miRISC (Long & Lahiri, 2012). In this review, we will first briefly present the molecular mechanisms that drive miRNA silencing. Next we will summarize the bioinformatic tools currently available for miRNA target prediction and finally we will discuss in detail the different experimental approaches that have been undertaken during the last years, paying special attention to their limitations for identifying the biologically relevant targets.
Mechanisms of miRNA repression
The mechanisms used by miRNAs to regulate gene expression remain unclear and have been a controversial subject over the past few years. Evidence for translational repression, mRNA destabilization and even activation of gene expression has been shown (Huntzinger & Izaurralde, 2011).
First evidences in animals pointed to regulation at the level of mRNA translation. Translation in mammals requires numerous factors that allow the recruitment of the ribosomal subunits to the mRNA initiation codon (initiation), the elongation of the nascent polipeptyde chain (elongation) and the release of the mature protein (termination). Some of these factors recognize the 5'Cap or the 3'UTR tail of the mature mRNAs allowing its efficient translation (Gebauer & Hentze, 2004). Although most of the studies pointed to miRNAs as inhibitors of mRNA translation initiation (Humphreys et al, 2005; Pillai et al, 2005), inhibition of elongation, premature termination of translation and co-translational protein degradation have also been described (Filipowicz et al, 2008; Huntzinger & Izaurralde, 2011; Mathonnet et al, 2007; Nottrott et al, 2006; Petersen et al, 2006).
Whereas perfect pairing of a miRNA with its target resulting in endonucleolitic cleavage is a rare mechanism in animals, exonucleolitic degradation often occurs. Thus miRNAs promote the recruiment of deadenilation factors to the target mRNA which remove its poly(A) tail making the mRNA susceptible to exonucleolitic degradation . Indeed, recent works show that binding of some of these factors, such as CCR4-NOT promotes deadenilation to repress translational initiation (Braun et al, 2011; Fabian et al, 2011)
Recently, several groups have taken advantage of the advances in mass spectrometry and RNA sequencing to study on a genome-wide scale the global effects of miRNA action on mRNA and protein levels after miRNA ectopic expression or removal in animals (Baek et al, 2008; Hendrickson et al, 2009). Although in general these studies conclude that degradation of miRNA targets is a widespread effect of miRNA action and provides a major contribution to silencing by animal miRNAs (Guo et al, 2010), whether target degradation occurs as a consequence of an initial block in translation remains as an unresolved question. The fact that some authors have found hundreds of miRNA targets for which protein levels were decreased far more than mRNA levels (Selbach et al, 2008) supports the idea that the precise mechanism of action is highly cell type and/or specific target-dependant. The effect of miRNAs in both plants and animal mRNA translation and decay have been comprehensively reviewed recently elsewhere (Fabian & Sonenberg, 2012; Huntzinger & Izaurralde, 2011; Pasquinelli, 2012).
Computational microRNA target identification
The best characterized features determining miRNA-target recognition are six-nucleotide (nt) seed sites, which perfectly complement the 5' end of the miRNA (positions 2-7)(Bartel, 2009). Also, a match with miRNA nucleotide 8, an A across from nucleotide 1 or both augment the seed pairing and enforce the miRNA-mediated repression (Grimson et al, 2007). These seed-pairing rules are widely used to predict functional miRNA target-sites, normally in combination with the secondary structure of the 3'UTR, the neighboring context information or/and the evolutionary conservation ((Bartel, 2009; Lewis et al, 2005; Long et al, 2007). It has been recently showed that miRNAs can also effectively repressed the expression of their targets through "G-bulged sites", which include a G-bulged in the position 5-6 of the seed-complementary region (Chi et al, 2012). Based on that, in the last years several miRNA target prediction programs have been published (Alexiou et al, 2009; Sethupathy et al, 2006) which characteristic features we summarize in table 1.
Additional information resources record nomenclature, sequence annotation such as genomic organization, precursor sequences, literature citations and links to target prediction sites ( miRBase, http://www.mirbase.org/ (Kozomara & Griffiths-Jones, 2011); miRGen, http://www.diana.pcbi.upenn.edu/miRGen.html (Megraw et al, 2007)), experimentally validated microRNA targets (TarBase, http://diana.cslab.ece.ntua.gr/tarbase/ (Vergoulis et al, 2012) or microRNA expression profiles (miRNAMap, http://mirnamap.mbc.nctu.edu.tw/index.php (Hsu et al, 2008); smiRNAdb, http://www.mirz.unibas.ch/cloningprofiles/).
The fact that animal miRNA target sites have only limited complementarity to their target sites makes that even small differences in the algorithms used produce a great diversity in target predictions. Although the normal thought would be that targets predicted by more than one program are more accurate than others, this does not seem to be the case. Few recent papers have performed very interesting studies where miRNA binding sites predicted by different algorithms were tested against genes proposed as targets experimentally (Alexiou et al, 2009; Rajewsky, 2006). The major conclusion of this kind of studies is that, in general, programs that rely on the evolutionary conservation of the seed tent to have a high precision but a low sensitivity. The same occurs when a combination of algorithms is used: better specificity is achieved by a higher price for the sensitivity. In the practice, this is translated in a very good performance of the bioinformatics tools for prediction of highly conserved and consensus binding sites (a precision of ~50% in many cases (Alexiou et al, 2009) but a very low efficiency for the detection of non-conserved binding sites, as well as for those bindings sites with poor pairing in the seed sequence. Similarly, one of the biggest limitations of the current algorithms is that the predictions are in most cases restricted to the 3'UTRs, when recent experimental data indicate that a big proportion of the miRNA/mRNA interactions may occur through the CDS or even the 5'UTR (Hafner et al, 2010). In addition, the binding rules for miRNA/target interaction through the 3'UTR could be different than through other mRNA regions, reducing even more the capacity of the current logarithms to predict these interactions. Finally, none of these methods takes into account the possibility of tissue specific interaction. Thus, experimental approaches for miRNA target determination are not only indispensable to confirm the biologically relevant targets of a given miRNA but are also essential for significantly improve the capacity of the current tools for predicting miRNA-target interactions.
Experimental miRNA target identification
Although many miRNAs and their binding sites are highly conserved, suggesting an important function, a typical miRNA-target interaction produces only subtel reduction (<2 fold) in protein level and many miRNAs can be deleted without generating any obvious phenotype. Accordingly the emerging view is that miRNAs may act to confer robustness to biological processes, for example reinforcing transcriptional programs to sharpen developmental transitions and entrench cellular identities or buffering fluctuations in gene expression sharpening the cell response to stress signals or certain regulatory networks (Ebert & Sharp, 2012). This does not make any easy the identification of the relevant targets of a given miRNA. Indeed, when bioinformatic tools have been used to pre-select the putative targets of a given miRNA, the vast number of predicted targets-usually with very disparate functions and in most of the cases including a high percentage of false positives and missing many genes in others-challenges scientists to choose which one is worth it to validate experimentally and will have a major impact in any biological process. Although methods for target validation will be also discussed in this section, we will focus first in those experimental approaches that allow the researcher to directly determine the actual targets in a specific biological system avoiding the biases inherent in the use of computational predictions.
High-throughput methods for identifying miRNA targets have been developed using a wide variety of experimental technologies and context normally in combination with overexpression/inhibition of the miRNA or after biochemical isolation of target mRNAs that are bound to miRNAs (Ørom & Lund, 2010)
Expression profiling following miRNA overexpression/inhibition.
Because microRNAs act by inhibiting the translation and/or promoting the degradation of their targets, the most straight-forward approaches rely on transfection of specific microRNA mimics or inhibitors into the cells followed of high-through put analysis of mRNA expression (by microarray or high-throughput sequencing (HT-seq), or proteomics (Thomas et al, 2010).
Initial studies transiently transfected miR-1 (muscle specific) and miR-124a (brain specific) into Hela cells, where they are not normally expressed, and used microarray analysis to identify those mRNAs downregulated as a consequence (Lim et al, 2005). Similar studies where soon performed (Grimson et al, 2007; Linsley et al, 2007) showing that mRNA transcripts downregulated after miRNA overexpression were significantly enriched for matches to the seed sequence. To avoid the bias that could be caused by the delivery of supraphysiological levels of a exogenous miRNA, the same kind of approach has been applied using antisense oligonucleotides that inhibit the miRNA action (Krützfeldt et al, 2005), observing for example that a high number of transcripts were increased upon the inhibition of miR-122 in liver (Elmén et al, 2008). Despite the beautiful piece of work presented by these groups, introduction of miRNA inhibitors in primary cells or tissues is challenging - so it is the inhibition of a highly expressed miRNA-, and most of the results have been obtained by overexpression of miRNAs in cell lines. Interestingly, when knockdown of miR-15 and -16 was performed at the same time as overexpression (Linsley et al, 2007), the increase in mRNA expression after knockdown was small compared to the downregulation observed after overexpression. Then, the effect of ectopic miRNAs might be easier to detect and therefore a more powerful tool, although more prone to lead to artifacts. Technical advances in next-generation sequencing technology have now enable the use of RNA-seq as an alternative to microarray gene expression analysis, providing a larger list of inferred miRNA targets in over-expression studies (Xu et al, 2010) . The higher sensitivity of this technology may be specially useful when miRNA inhibition is performed, since small changes in RNA levels are more prone to be detected.
The transcriptome profiling approach has two clear limitations: first, this method cannot distinguish direct from indirect targets (which can be partially minimized by harvesting the cells soon after the transfection) and second, those targets that are only regulated at the level of translation, occurring without significant alteration of the transcript levels, will not be detected. This second handicap can be overtaken by the use of proteomics.
Proteomic approaches have the inherent advantage of assaying the ultimate effect of miRNAs. Newly synthetize proteins can be metabolically labelled by growing the cells in medium containing heavy isotopes of essential aminoacids (SILAC: Stable isotope labelling by aminoacids in cell culture). Then, mass spectrometry can be used to determine the ratio of peptide peak intensities from the light and heavy isotopes as a measurement of protein synthesis and the differences can be assessed after overexpression/inhibition of a given miRNA (Selbach et al, 2008; Vinther et al, 2006). This approach was for example successfully used to determine targets of overexpressed miR-1, miR-124 and miR-181 in Hela cells and after deleting miR-223 in mouse neutrophils (Baek et al, 2008), finding that in most of the targets, changes in transcripts levels were also detected. This kind of correlation has been detected in several other studies as well as an enrichment in seed regions in the detected targets (Kaller et al, 2011; Selbach et al, 2008).
Polysome profiling following miRNA overexpression/inhibition.
An mRNA that is being actively translated is going to be found bound to a high density of ribosomes. Accordingly, ribosome-profiling strategy is based in the recovery of the mRNA fragments that are bound by ribosomes (and then protected from the action of RNAses) and identification of those fragments by deep sequencing. The abundance of different fragments corresponding to an mRNA is a direct indication of the amount of translation of that gene. (Ingolia et al, 2009). This approach was used in 2010 by Bartel's group in combination with overexpression of miR-1 or miR-155 in HeLa cells, which do not normally express those miRNAs (Guo et al, 2010). Basically, they transfected HeLa cells with miR-155/miR-1 mimics and treated with cycloheximide to arrest the translating ribosomes. Then, they digested the non-protected RNA with RNAseI and purified the resultant monosomes by a sucrose gradient. Those monosomes contained RNA fragments of ~30 nucleotides, which were released and identified by high-throughput sequencing (Ribosomal protected RNA fragments, or RPF). In parallel, total mRNA after miRNA overexpression was also sequenced to assess the contribution of mRNA stability (RNA-seq) versus the translation to the miRNA effect . The same experiment was performed with neutrophils from a miR-223 knockout mouse versus a wild-type mouse, which express high levels of miR-223. What they found is that detection of genes with at least one predicted miR-155 or miR-1 binding site in their 3'UTR was lower after miR-155 or miR-1 transfection in both RNA-seq and RPF samples. Conversely, genes with at least one binding site for miR-223 were over-represented in RNA-seq and RPF from miR-223-/- neutrophils. By using this approach, the main finding of these authors was that at least 84% of the miRNA-mediated repression was due to mRNA destabilization, since only minor differences were found between the RPF (measure of translation) and the RNA-seq (measure of RNA degradation) after miRNA overexpression/depletion. Nevertheless, using a similar technique, Bazzani and coworkers (Bazzini et al, 2012) found that the effects of miR-430 in zebrafish occur at the level of translation preceding RNA decay, so this disparity may result from the steady-state conditions used, the biological system or the miRNA to study.
In general, it has been assumed that changes in mRNA levels closely reflect the impact of miRNAs on gene expression (Guo et al, 2010) and, given that proteomics approaches are more expensive, less sensitive and technically more complex -even not accessible to many researchers- than mRNA profiling techniques, it is not surprising that most of the publications generated are based in transcriptome analysis. Ribosomal profiling on the other hand is technically very challenging and the most recently developed technique, but the quick reduction in the cost of the high-throughput sequencing makes highly probable that this approach will be often used in combination with mRNA profiling in the future.
Nevertheless, and as mentioned before, none of these techniques allows distinction between direct and indirect targets and during the last past few years efforts have been directed to the development of techniques for an efficient and straightforward determination of the truly miRNA targets, normally those ones found in association with miRISC.
Pull-down assays with members of miRISC
The mammalian miRISC contains a mature miRNA and several proteins, including an AGO protein and GW182 (Ding & Han, 2007; Landthaler et al, 2008). Different methods have been developed to recover those mRNAs bound by miRISC (and therefore direct miRNA targets) using pulldown or immunoprecipitation of a component of miRISC (both a native member or an epitope-tagged one), normally in combination of overexpression/inhibition of the miRNA of interest. Then, the recovered mRNAs can be detected individually (normally by RT-qPCR) or more commonly by high-throughput techniques such as microarrays or deep-sequencing.
Tagged miRNA pull-down
Orom and colleagues developed a direct affinity purification method for experimental identification of miRNA targets based in the transfection into the cells of synthetic miRNA duplexes carrying a biotin group attached to the 3'-end of the miRNA sense strand. These sense strands are incorporated into miRISC and, after cell lysis, the miRNA-mRNA complexes are captured on streptavidin beads from which the mRNA species can be purified and analysed (Ørom & Lund, 2010). This technique led to the discovery of miR-10a interaction with the 5'UTR of ribosomal protein transcripts which surprisingly enhanced their translation and independently demonstrated previously detected targets of miRNAs bantam and miR-124 in cell lines (Ørom et al, 2008). Nevertheless this study is rather controversial for several reasons: first, the pulled down mRNAs were not enriched for miR-10a seed matches, second, the fact that they were mostly abundant ribosomal mRNAs suggest they might have associated with the biotinylated mRNA non-specifically (It is not known what is the effect that the biotin tag may have in miRNA binding) and finally, as already mentioned, most identified genes were translationally upregulated, rather than downregulated, which the authors attribute to the presence of binding sites in the 5'UTRs. Thus, the ability of this technique to comprehensively identify true miRNA targets has yet to be fully demonstrated.
An in vitro variation of this technique -called LAMP (Labeled microRNA pull-down assay)- utilizes digoxigenin (DIG)-labeled pre-miRNA oligonocleotides that are mixed with cell extracts, submitted afterwards to immunoprecipitation with anti-DIG antibodies and the analysis of the co-IPed mRNAs (Hsu & Tsai, 2011).
A recently developed alternative to the biotin-labelling is the miR-TRAP (miRNA target affinity purification) in which basically the miRNA is conjugated to psoralen to produce a highly photo reactive probe. These probes function similarly to endogenous miRNAs and, when the cells are expose to UVA radiation (360 nM-less harmful than the 254nM using in other crosslinking experiments, which is relevant for in vivo experiments) the Pso moiety of the miRNA reacts with uridin on target mRNAs, enabling the bound complex to be stringently purified by biotin-streptavidin affinity purification. The biotin is incorporated in the 3'UTR of the miRNA as an affinity tag (Baigude et al, 2012). The authors have successfully used this approach to detect two novel targets of miR-15b and are currently applying these methods to asses miRNA targets in various disease models. Although susceptible of the same handicaps than the simply biotinylated-miRNA based technique, the covalent link between the Psoralen-tagged miRNAs and target mRNAs allows the use of much stringer purification conditions, which could putatively greatly diminish the recovery of unspecific targets.
Interestingly, all of these methods could be modified to identify miRNAs targeting a mRNA of interest by replacing labelled miRNA with labelled transcript. In this same direction Yoon and colleagues propose a systematic approach termed MS2-TRAP (tagged RNA affinity purification) for identifying miRNAs associated with a target transcript in the cellular context. Briefly, they tagged the mouse linRNA-p21 with MS2 hairpins and co-expressed it in MEFs along with the chimeric protein MS2-GST. Then they affinity-purified the miRNAs present in the RNP complexes using glutathione-SH beads and those were detected by qPCR. Out of the 5 miRNA analysed (predicted to target linRNA-p21), 4 were enriched in the pulldown and two of them functionally validated (Yoon et al, 2012). This approach could be widely used if coupling the pulldown with high-throughput approaches for miRNA detection. Nevertheless one of its biggest limitations is that both the tagged RNA and the MS2-protein have to be exogenously introduced, which restricts its use in cell lines with low efficiency of transfection, primary cells or in vivo.
Immunoprecipitation of miRISC proteins
Different approaches have been developed in order to recover and detect those mRNAs co-immunoprecipitated with proteins from RISC. To identify targets of specific miRNAs, cells are transfected with a synthetic miRNA or a miRNA inhibitor, versus a control. After incubation, cells are mildly lysed, miRISC complexes immunoprecipitated using antibodies against Ago2 or other members of miRISC and captured mRNA is isolated and analysed by microarray analysis or RNA-seq. These methods are known as RIP-ChIP (Ribonucleoprotein Immunoprecipitation followed by microarray chip analysis) or RIP-Seq (Ribonucleoprotein Immunoprecipitation followed by High-throughput sequencing) (Baroni et al, 2008; Keene et al, 2006). Karginov and colleagues used microarrays to identify mRNA co-immunoprecipitated with c-myc-tagged-Ago2 that was stably transfected in 293S cells in combination with miR-124a mimics, detecting several targets that were affected by this miRNA only at the translational level (Karginov et al, 2007). Artifacts introduced by the used of epitope-tagged Agos (chances of unspecific association with RNA can not be obviated) can be eliminated by the precipitation of endogenous components of RISC. As an example, Tan and coworkers simultaneously inhibited four miRNAs (miR-17/20/93/106) in two different cell lines and performed immonuprecipitation of endogenous Ago2. The Ago2-associated (IP) transcripts were differentially detected by microarray as well as the transcripts present in total RNA fractions (T) of the extracts used for IP (Tan et al, 2009). They identified more than two thousand mRNAs depleted in the IP fraction versus the total RNA (comparing IP/T ratio), and identified more than 100 candidates with a high enrichment in seed target sites in their 3'UTRs, supporting the suitability of this method to detect miRNA specific targets. In addition, they validated by luciferase assays all the targets assayed (nine that contained at least a 6-mer site for miR-17 from the list of top regulated genes), while the two potent targets assayed without a 6-mer site in the 3'UTR did not increased luciferase activity upon miR-17 inhibition. Although the present of those transcripts in the IPs could be due to non-specific binding, the authors do not rule out the possibility that those transcripts may be targeted through the CDS. Immunoprecipitation with Ago1 has also been used to isolate targets in mammalian cells (Beitzinger et al, 2007; Landthaler et al, 2008).
One obvious limitation of these procedures is that, in many cases, the use of antibodies against only one of the several Ago proteins (four in humans (Wang et al, 2009)) expressed in an organism. Hypothetically, mRNAs and miRNAs pulled down with different Agos may not be identical, although so far a high extent of overlapping functions have been observed for the different Agos (Su et al, 2009). Antibodies that recognize all the Agos present in the cells or other members of RISC could be then preferentially used (Nelson et al, 2007; Zhang et al, 2007).
Although these approaches potentially identify miRNA direct targets, they do not give information about the precise binding site and they are normally combined with bioinformatic tools and experimental validations. On the other hand, these techniques require that the binding miRISC-mRNA is stable enough to survive the IP conditions and the quantities of recovered RNA are normally very low (especially when using primary cells as starting material).
In order to stabilize the RNA-protein binding -allowing the capture of more transient interactions- the most recently developed techniques include the covalent UV-mediated cross-linking of the RNA to the proteins in situ preceding the lysis of the cells and the immunoprecipitation. The first group in developing this technique for the identification of miRNA binding sites was the Darnell lab (Chi et al, 2009) and they termed it HITS-CLIP (High-throughput sequencing of RNAs isolated by crosslinking immunoprecipitation). The same group had previously used this technique for the identification of protein-RNA interactions in living tissues in a genome-wide manner (Licatalosi et al, 2008). Briefly, the method consists in the use of unltraviolet radiation to crosslink RNA-protein complexes that are in-theoretically- direct contact. The immunoprecipitation of miRISC can then be performed in much more stringent conditions (possibly decreasing the unspecifically-bound material recovered). In addition, the unbound RNA is digested to leave miRISC-protected RNA fragments, which are analysed by HT-seq. Those fragments correspond to miRISC binding sites, so this technique allows the straight-forward mapping of the interaction. This first study not only elegantly proved that the CLIP can be successfully used for the mapping of miRISC positioning along the target mRNAs, but also made an important contribution to the mechanism governing the miRNA-target recognition: a high percentage (25%) of the readings were mapped to ORFs, although the majority corresponded to 3'UTRs and only a 1% to the 5'UTR, confirming that miRNAs preferentially bind their targets through the 3'UTR but underlying the importance of the CDS-mediated interaction. They also found that a high percentage of the mapped sites did not contain seed sequences for any of the most abundant miRNAs in the studied tissue (brain), indicative of seedless interactions. Indeed, a posterior study from the same group proved an alternative binding mode used by miRNAs: mRNAs containing a G-bulge site after the five consecutive nucleotides in positions 2-6, comprised >15% of all Ago-miR-124-mRNA interactions in mouse brain, and more than 75% of the non-seed detected sites (Chi et al, 2012). The authors also found that these sites are present both in the 3'UTR and in the coding region of the transcripts and the degree of miRNA-mediated repression is similar to the one observed for canonical seed binding sites.
Up to our knowledge, this approach has only been used so far to identify genome-wide miRNA-RISC-mRNA ternary complexes from a cell line or tissue but, its combination with over-expression/inhibition of a given miRNA could potentially be used to determine the targets of that specific miRNA.
HITS-CLIP allows mapping miRISC binding sites with a resolution of ~50 nucleotides (Chi et al, 2009) while a modification of this method, the PAR-CLIP (Photoactivatable-Ribonucleoside-Enhanced Crosslinking and Immunoprecipitation) allows the precise location of the crosslink (Hafner et al, 2010). This method uses 365nm-UV light to induce efficient cross-linking of RNA-binding proteins with photoreactive ribonucleoside analogs (such as 4-thiouridine) that are incorporated into nascent mRNAs by living cells. When this is used in combination with Ago IP, miRNA-targets are isolated, fragmented and identified by high-throughput sequencing. When the RNA contains 4-thiouridine, the crosslinking causes thymidine to cytidine transitions, which are detected as mutations during the deep-sequencing making then possible to separate them from the background and giving a much more accurate mapping (Hafner et al, 2012). Also, the 365nm UV crosslinking proteins-photoreactive ribonucleoside analogs is more efficient than the 245nm protein-RNA thus improving the precipitated-RNA yields. Similarly to Chi and colleagues, Hafner and coworkers found abundant miRNA binding sites in the coding regions.
These approaches may be useful not only to precisely map a direct interaction between a given miRNA and its targets, but also to detect those targets only affected at the level of translation or the interactions mediated by atypical binding sites. On the minus side, both PAR-CLIP and HITS-CLIP can present a selection bias for strong RNA-protein interactions, and highly expressed transcripts are generally over-represented (Zhang & Darnell, 2011). In adition, they are incredibly challenging technically, require the use of high-advance techonology not always available and a very complex post-experimental analysis (bioinformatic analysis) is also essential. Furthermore, the long multistep protocols normally require a huge amount of initial material, which is not accessible in many cases-i.e. primary cells, and possibly difficulties the combination of these approaches with the use of miRNA mimics/inhibitors. It is not surprising then that only few research groups are currently applying those techniques, although further optimization of the protocols that will make them more accessible to other groups and decrease the bias is (hopefully) anticipated.
Experimental validation of miRNA targets
No matter how a miRNA-target interaction has been detected (bioinformatically or experimentally), it is essential to validate its functionality in the biological model of interest.
Since an inverse relationship between the levels of expression of a miRNA and its target is anticipated, the most direct and straightforward method for validation consists in mimicking/inhibiting the miRNA of interest (as previously described, generally by transient transfection) and assess the effect in the target gene expression. Since overexpression can lead to interactions not normally occurring in the cells, inhibition should give more relevant conclusions. Nonetheless it is not always easy to inhibit the action of a miRNA to a significant extent so overexpression is more broadly used (Martinez-Sanchez et al, 2012).
The analysis of the effect in the target gene expression should be preferentially done at the protein level (normally by Western blotting, ELISA, immunostaining, etc.), although the impossibility of protein detection in some cases (and the higher cost) drives the researchers to often analyse changes in the target mRNA instead. In that case and as stated multiple times before, an effect will only be observed if the miRNA affects the stability of the target, which occurs in many cases (Guo et al, 2010). mRNA changes are normally quantified by RT-qPCR, which requires low amounts of starting material and is quick and cheap. Alternatively, Northern blotting can be useful, although time-consuming, especially in those cases where different isoforms of the target are expressed.
Finally, reporter assays are universally used to confirm a direct regulation of the gene expression by the miRNA (Long & Lahiri, 2012). Generally the 3'UTR of the transcript of interest is cloned downstream of a luciferase ORF. When this vector is transfected into cells expressing the targeting miRNA, luciferase activity should be lower than for the empty vector. Inhibition of the miRNA action by co-transfection of antagonists should then result in the recovery of the luciferase expression levels, confirming that the repression is specifically mediated by that miRNA. Nevertheless, in many cases, transcription from the promoter contained in the vector is too strong and the endogenous miRNA is not enough to mediate a detectable repression. In those cases, co-overexpression of the miRNA is widely used. To corroborate the direct interaction, mutation of the anticipated miRNA-binding site/s is introduced in the reporter. This mutation is expected to drop any miRNA-mediated effect.
In those cases where the miRNA-target interaction is not mediated by the 3'UTR, but through the coding region, validations using reporter experiments can also be used and the predicted binding site -together with the surrounded sequences- may be cloned downstream the luciferase coding region. Generally, sites in the coding region have a weaker effect on gene expression, thus their validation is more challenging.
Since discovered in 1993 (Lee et al, 1993), miRNAs have capture the attention of a high number of researchers. Nevertheless their specific functions in the cells have barely started to be unravelled. The quick development of the bioinformatics tools and the high-throughput sequencing techniques during the last years have led to both the discovery of many new miRNA classes in multitude of different organisms and to the generation of a huge amount of data indicating their possible targets and mechanism of action. Nevertheless only a small proportion of miRNA-target interactions have been functionally validated. Although each method presented here has important limitations, combining them wisely can lead to the identification of many biologically important targets which is essential for understanding miRNA function and for their consideration as therapeutic targets in the medium/long-term future.