One of the challenges of systems biology is to understand how the interplay of individual cellular components gives rise to cellular behaviors such as cell division, apoptosis and differentiation. Generally, the expression of genes encoding proteins that are part of the same pathway is regulated in a coordinated manner, either at the level of transcription, or post-transcriptionally. Messenger RNAs (mRNAs) undergo multiple steps of processing, starting with transcription from the genome, splicing and maturation, export from the nucleus to the cytoplasm and finally degradation. Throughout, the mRNAs interact with RNA-binding proteins (RBPs) that protect them from degradation, participate in their processing, delay or accelerate their turnover. RBPs thereby regulate a wide range of processes, from developmental transitions to the response to stress.

In recent years, methods have been developed for transcriptome-wide identification of RBP binding sites. These studies found that RBPs typically have hundreds or thousands of targets and that an mRNA is frequently regulated by multiple RBPs in a context-dependent manner. Substantial efforts are therefore required to characterize the context-dependent biological function of individual RBPs.

The method of choice for transcriptome-wide and high-resolution identification of RBP binding sites, CLIP, involves crosslinking of proteins to target RNAs, immunoprecipitation of the protein of interest and its associated RNAs with specific antibodies and high-throughput sequencing of the RBP-bound RNA fragments. A few variants, known as single-end enhanced CLIP (seCLIP) [1], or HITS-CLIP (High-Throughput Sequencing-CLIP) [2], PAR-CLIP (photoactivatable-ribonucleoside-enhanced CLIP) [3, 4] and iCLIP (individual nucleotide resolution CLIP) [5], cross-linking analysis of cDNA (CRAC) [6] and cross-linking, ligation, and sequencing of hybrids (CLASH) have been described [7, 8].

Detailed protocols of the CLIP, and its variants, followed by high-throughput sequencing can be found in papers describing their application for the detection of the RNA-protein interactions in humans [9, 10] or viruses [11, 12]. These protocols can be easily extended to other applications. A variant of PAR-CLIP, the global photoactivatable-ribonucleoside-enhanced CLIP (gPAR-CLIP-seq), was developed for identifying regions of the transcriptome bound by RBPs in budding yeast [13]. In contrast to the standard PAR-CLIP method, the gPAR-CLIP-seq allows for the isolation and sequencing of all mRNA sites bound by the cellular "RBPome."

The main difference between PAR-CLIP and CLIP/ HITS-CLIP (in contrast to CLIP, HITS-CLIP uses high-throughput sequencing) is in the way the proteins are crosslinked to RNAs. In PAR-CLIP the crosslinking is carried out with 365 nm UV light after cells have been treated with a modified nucleotide, typically 4-thio-uridine, which is incorporated in mRNAs. In contrast, in CLIP/HITS-CLIP crosslinking is done with 254 nm UV light. In iCLIP the crosslinking is also carried out with 254 nm cross-linking, but the method aims to capture the reverse transcription products that stop at the site of crosslinking as opposed to read-through products. A detailed procedure of iCLIP is included in a review by Stork and Zheng [14]. A Fully Automated and Standardized iCLIP (FAST-iCLIP), with improved biochemistry and automated bioinformatic analysis became available in 2015 for the detection of RNAs not only from the protein-coding regions, but also spanning noncoding, repetitive, retroviral, and nonhuman transcriptomes [15].

It should be noted that essentially all the steps in these protocols are interchangeable. In particular, the type of lysis buffer in cell lysis depends on whether nuclear, cytoplasmic or whole cell lysates are needed, for the RNase treatments various RNases can be employed to fragment the RNA and adapters can be ligated in different ways [16].

Because CLIP is increasingly used to map regulatory RBP-RNA interactions, we have decided to critically compare and evaluate the various available protocols. Since our group was actively involved in the development of the PAR-CLIP protocol (Hafner et al 2010), the approach was to modify individual steps in this protocol according to other published protocols and evaluate the impact of these modifications on the set of target mRNAs. We chose for this study two well-known regulators of mRNA stability – HuR, that generally stabilizes mRNAs by binding AU-rich elements (AREs), and Argonaute 2 (Ago2), a protein that forms RNA-induced silencing complexes with microRNAs (miRNAs), inducing deadenylation and degradation of their targets as well as repression of protein translation [17].

The results showed that both PAR-CLIP and HITS-CLIP recover binding sites with similar accuracies for both proteins - HuR, which binds low-complexity U-rich sequences, and Ago2, which has a complex binding specificity. Importantly, we found that the step of RNA fragmentation, which is performed during the CLIP procedure to generate the short, protein-bound RNA fragments, can introduce a substantial bias in the sets of target sites that are identified. The sequence specificity and amount of ribonuclease (RNase) used during this step can substantially bias the recovered RNA fragments, resulting in the identification of only a subset of sites that were bound by the RBP of interest. To obtain the most representative binding sites, mild RNase digestion is recommended. Furthermore, we found that reverse transcription of the fragments through crosslinked sites induces mutations that are specific to each crosslinking strategy (T to C substitutions in PAR-CLIP and T deletions or mutations in CLIP/HITS-CLIP). These crosslink induced mutations can be used to pinpoint the binding site with very high resolution, down to a single nucleotide level.

At present, all CLIP methods require rather large amounts of cells or tissue to yield libraries that cover the transcriptome. The main obstacle that needs to be overcome in this process is the low efficiency of the various, rather numerous steps in the protocol. A few of these steps are discussed in more detail below.

In the PAR-CLIP method [3], 4-thio-uracil (4SU) at 100 micromolar concentration in the medium is fed to the cells 14-16 hrs before the crosslink. 4SU is usually dissolved in DMSO but is soluble in H2O. Although it is assumed that the rate of incorporation in vivo is high [3], a rather large range of rates has been reported, between 10 and 90% (Meisenheimer and Koch, 1997, and references therein). 4SU as a photolabile nucleotide analog with an absorption spectrum around 325 nm, that is usually crosslinked with UV light of 365 nm (Hafner et al 2010). Although it has initially been reported that the crosslinking is more efficient in PAR-CLIP compared to HITS-CLIP [3] we did not observe substantial differences [17].

In some circumstances, it may be very difficult to apply PAR-CLIP. This is the case when one has to work with tissues or organs, or when resting cells, with low RNA turnover and low efficiency of 4SU incorporation, are used. In these cases, UV crosslinking has to be performed with 254 nm UV light. Nevertheless, in a publication 4SU was fed to C. elegans to then perform “in vivo PAR-CLIP” or iPAR-CLIP [18].

We have in the past performed CLIP experiments with several FLAG-tagged proteins in stably transformed HEK293 cell lines and also a number of experiments using direct antibodies against the target proteins. FLAG fusion proteins are immunoprecipitated with monoclonal anti FLAG antibodies that are very efficient and commercially available. The drawback is that fusion proteins are often overexpressed to various degrees and, because a specific transcript form is used for cloning the expression construct, binding sites of different splice variants that the protein may have are not simultaneously captured. The main problem with specific antibodies is their variable suitability for immunoprecipitation. Nonetheless, a vast selection of antibodies that have already been tested for IP is commercially available. Ultimately, antibodies can be tested for IP before starting the CLIP experiment under the same conditions (buffers) as used in the particular CLIP procedure.

For the preparation of cDNA libraries RNA can be fragmented enzymatically with RNases, chemically by alkaline hydrolysis or mechanically by sonication in a Covaris or similar device. The goal is to obtain rather short RNA fragments at sizes between about 20 and 50 nucleotides. Smaller fragments would allow the mapping of the crosslink site at higher resolution, but too small fragments would not map unambiguously to the genome and would not allow identification of target sites. When enzymatic digestion is performed, care needs to be taken to prevent biases caused by nucleotide-specific cleavage. For example, RNase T1 that cleaves after G cannot produce RNA fragments of optimal sizes from targets that are very rich or very poor in G content. Partial digestion with RNase T1 or micrococcal nuclease can reduce this bias [17]. Less sequence bias is expected with an RNase T1/A mix or with RNase I which cleaves after every ribonucleotide [16]. However, care has to be taken with RNase fragmentation because it is difficult to monitor digestion and over- or under-digestion happens fast if the time or temperature is not precisely controlled.

Obtaining reproducible RNA sizes could be easier with chemical or mechanical fragmentation than using RNases although we have not tested these methods with CLIP. These techniques are frequently used in mRNA-seq library preparation. We found the rate to be relatively linear in the case of alkaline hydrolysis.

During IP, abundant RNAs like ribosomal RNAs and tRNAs bind unspecifically to the Dynabead-antibody-protein complexes and must be removed. The PAR-CLIP protocol uses cartridges to electro-elute RNA from gel slices inside dialysis membranes of small molecular weight cutoff [3]. In the CLIP and iCLIP protocols of Ule et al 2003 [19] and König et al 2010 [5], respectively, the protein-RNA complexes are transferred from the SDS gel to nitrocellulose membranes. This way, only RNA that is crosslinked to protein is retained whereas contaminating abundant RNAs migrate through the filter.

UV crosslinking during the CLIP procedure results in the formation of a covalent bond between the base of 4-thio-uracil and a nearby amino acid moiety where alcohols and amines are preferred [20]. The crosslinked proteins are then removed from the RNA by proteinase K digestion. In the subsequent reverse transcription reaction, 4SU nucleotides within the RNA base-pair with G, later resulting apparent T to C substitutions in the sequenced reads. These mutations are diagnostic and useful for mapping the crosslinks at very high resolution [3]. In addition, because of incomplete protease digestion, peptide fragments of various sizes may remain attached to the crosslinked RNA. This in turn can cause the reverse transcriptase enzyme to leave the template, resulting in truncated transcription products that are missing the 5’ adapter sequences. These molecules are in a later step not amplified and are therefore lost from the PAR-CLIP/CLIP library. It is estimated that around 80% of the total library constructs are lost this way [21]. A possible reason for this could be the degree of proteinase K digestion. An extended or a second round of proteinase K digestion could be the remedy for removing these traces of protein from the RNA.

The main reason to amplify the fragment cDNA library is to render the constructs competent for sequencing by diluting out nonproductive molecules that lack 5’ or 3’ adapter. These stretches of DNA are necessary for the cluster formation and the subsequent sequencing reaction on Illumina sequencers. Otherwise many of these non-productive molecules would occupy binding sites on the surface of the flow-lanes. On the other hand, over-amplification of libraries must be avoided which can lead to clonal over-representation. Since identical sequences are normally removed during filtering or annotation of the data sets, the total number of unique sequences can dramatically drop. To determine the optimal number of PCR cycles we first do a pilot PCR reaction with between 14 to 25 cycles. Typically this can be a 50 ul PCR reaction with 9 ul samples removed at 14, 17, 20, 23 and 26 cycles. These are run on 2.0 to 2.5% agarose gel. For a large scale PCR we choose a number of cycles in the middle of a linear range of amplification.

CLIP involves ligation of RNA fragments to two different adapters. This is necessary to amplify the library by PCR and also for the cluster formation and both the sequencing and barcoding reactions on the Illumina workstation. Yet, it is difficult to prevent direct ligation of the 5’ to the 3’ adapters. Particularly when the starting material is limiting, adapter dimers end up being very abundant at the end of the procedure. These dimers are also often strongly over-amplified in final PCR reactions. Methods to suppress adapter-adapter ligation to a certain degree have been proposed [22]. However, most of the adapter-adapter products can be separated from library products by gel electrophoresis and size specific elution after final PCR amplification.

There exist several published methods for high-throughput strand-specific cDNA sequencing (RNA-Seq) with small amounts of material (Levin et al Nat Meth 2010). Some of these strategies could also be useful for the CLIP procedure.

In this procedure (Clontech Laboratories), an M-MLV reverse transcriptase is first used to synthesize a DNA complementary strand of the RNA fragments. When the polymerase reaches the end, it extends the product by about three untemplated C residues. A DNA oligo with the sequence of a 5’ adapter ending with 3 G residues at the 3’end is included in the reaction. The GGG stretch of the oligo anneals with CCC at the end of the cDNA and acts as a continuation of the template (template switching) such that the polymerase continues to complete the cDNA by addition of a 5’ adapter sequence [Chenchik et al (1998), In RT-PCR Methods for Gene Cloning and Analysis. BioTechniques Books, MA, 305-319. The method is used in SMARTer™ Ultra Low RNA Kit (Clontech) for the preparation of cDNA samples from as low as 100 pg of RNA for sequencing with the Illumina HiSeq and Genome analyzer instruments. One disadvantage with the SMART technology is that when the final products are sequenced on an Illumina sequencer in the sense direction, all clusters first read a short sequence consisting of only Cs. This “mono-template issue” is a problem for the Illumina Genome Analyzer software because it uses the first four cycles to define the coordinates for the clusters. This, however can be circumvented by either operating at low cluster density, mixing the smart-CLIP libraries with other libraries (that have no CCC at the beginning), using specific sequencing primers which include CCC at the 3’ end or sequencing in the other direction by inverting the adapters.

It is not known whether the reverse transcriptase stop that is caused by crosslinked peptides still allow the addition of the 3 Cs. We are presently testing this possibility.

Another powerful method to amplify cDNAs from low amounts of starting material is T7 RNA polymerase linear amplification of cDNA libraries, resulting in aRNA (Eberwine method [23] ). This requires adapters that contain a T7 RNA polymerase promoter sequence preceding the regular 5’ adapter sequence that is used to prime the sequencing reaction. After reverse transcription of the library, products are amplified up to 1000-fold with T7 RNA polymerase. For this, the cDNA first has to be made double-stranded or a DNA oligo complementary to the T7 promoter has to be included in the polymerase reaction. The resulting RNA then is again reverse transcribed and PCR amplified for the final library.

The other half of a successful CLIP-seq experiment is the computational analysis, which, as the experimental part, is not without pitfalls. Mapping the reads to the genome or transcriptome of interest is one of the most crucial tasks and can often only be done in reasonable time with specialized computing hardware. Of the seemingly endless number of parameters that mapping programs usually have to offer, control of the mapping error rate is crucial, especially for PAR-CLIP experiments since an increased frequency of T to C conversions is expected. In order to facilitate consistent analysis of data sets of binding sites of RBPs generated across various laboratories and protocols our group developed an analysis and database server (www.clipz.unibas.ch) that provides easy access and visualization of the data sets, as well as de novo analysis of user-provided data. A similar service, but without the functionality of analyzing one own’s data, is the doRiNA database (dorina.mdc-berlin.de). For bioinformatics gurus that are familiar with navigating in a Linux environment, software packages like PARalyzer (www.genome.duke.edu/labs/OhlerLab/research/PARalyzer/) or the R package “wavClusteR” to extract binding sites from CLIP sequence data are available.

Kloetgen et al [24], describe other available bioinformatics tools commonly used in PAR-CLIP data analysis, including tools for the identification and removal adaptor sequences, tools to map the reads against reference genomes and to connect the reads to functional annotations of genes. Additionally, new statistical and computational frameworks for PAR-CLIP data analysis became available. For example, one such analysis workflow, implemented in the R package wavClusteR 2.0, became available at: http://www.bioconductor.org/packages/devel/bioc/html/wavClusteR.html [25].

CLIP is a very powerful technique for the study of RNA-protein interactions in all biological systems. Several variations of the method have already been described and used in different studies. Further improvements in the experimental procedure and data analysis are expected to increase the number of studies based on such a technique. Ramanathan M et al reviewed CLIP methodology, including HITS-CLIP/PAR-CLIP, iCLIP, eCLIP, irCLIP, GoldCLIP, fCLIP, BrdU-CLIP, TRIBE, and RNA tagging [7]. AGO-CLIP (argonaute-crosslinking and immunoprecipitation ) or CLEAR-CLIP (Covalent ligation of endogenous Argonaute-bound RNAs) have been used to identify the target genes of miRNAs [26-28]. GEO database is a useful resource, and data can often be directly analyzed. For example, H Seok et al analyzed Ago HITS-CLIP dataset GSE83410 [29] from human left ventricle tissues [27].

The article was updated by Labome on August 12, 2016 and later. For an earlier version, please see here.