Decoding protein–RNA interactions using CLIP-based methodologies

Abstract

Protein–RNA interactions are central to all RNA processing events, with pivotal roles in the regulation of gene expression and cellular functions. Dysregulation of these interactions has been increasingly linked to the pathogenesis of human diseases. High-throughput approaches to identify RNA-binding proteins and their binding sites on RNA — in particular, ultraviolet crosslinking followed by immunoprecipitation (CLIP) — have helped to map the RNA interactome, yielding transcriptome-wide protein–RNA atlases that have contributed to key mechanistic insights into gene expression and gene-regulatory networks. Here, we review these recent advances, explore the effects of cellular context on RNA binding, and discuss how these insights are shaping our understanding of cellular biology. We also review the potential therapeutic applications arising from new knowledge of protein–RNA interactions.

Introduction

RNA-binding proteins (RBPs) have fundamental roles in controlling a wide range of biological processes, including differentiation, development and cellular stress responses. By finely regulating eukaryotic RNA metabolism at multiple levels (Box 1), including RNA splicing, polyadenylation, stability, localization and translation, RBPs influence all aspects of protein expression. Determining the binding preferences of RBPs for distinct regions of RNA molecules at various stages of maturation holds the key to deciphering the underlying mechanisms of RNA processing and gene expression.

Box 1 |. RNA-binding proteins and the processing of RNA.

The life cycle of a messenger RNA involves a series of processing steps that ensure the accurate and efficient transfer of genetic information from DNA to protein synthesis (see the figure). The process starts with transcription, during which RNA polymerase II transcribes a DNA sequence into a pre-mRNA molecule. This pre-mRNA undergoes capping, whereby a 5′ cap is added to be bound by the nuclear cap-binding complex, followed by splicing, whereby introns are removed and exons are joined together by the spliceosome. Splicing factors guide and facilitate these processes, ensuring the proper recognition of splice sites and maintaining structural integrity. Following splicing, the mature mRNA molecule is polyadenylated at its 3′ end, forming a poly(A) tail, a process that involves 3′ end cleavage and polyadenylation factors. This step not only helps to stabilize the mRNA but also aids in translation initiation. The exon junction complex binds stably to exon–exon junctions to coordinate splicing, mRNA export, translation and nonsense-mediated decay. The mature, polyadenylated mRNA, accompanied by nucleo-cytoplasmic shuttling proteins, is transported from the nucleus to the cytoplasm through nuclear pores. Export factors, such as the transcription and export complex (TREX), and other RNA-binding proteins (RBPs) ensure the proper recognition and transport of the mRNA molecule. During translation in the cytoplasm, RBPs, such as translation initiation and elongation factors, have crucial roles in facilitating ribosome binding to the mRNA and ensuring accurate translation of the coding sequence into a functional protein. Also in the cytoplasm, the mature mRNA is subject to quality control mechanisms such as nonsense-mediated decay, which involves RBPs that monitor and target mRNAs containing premature termination codons for degradation. Furthermore, mRNA decay is regulated by RBPs that ensure the timely recycling of cellular resources and dynamic responses to changes in cell state.

In addition to processing protein-coding mRNAs, RBPs are involved in the biogenesis of non-coding RNAs. MicroRNAs, which are transcribed as primary microRNAs, are processed by Drosha and Dicer to form mature microRNAs, which target mRNAs by complexing with the argonaute protein AGO2 in the RNA-induced silencing complex. Long non-coding RNAs (lncRNAs), which have roles in chromatin remodelling and transcriptional control, interact with RNA-binding transcriptional regulators. Transfer RNAs are essential for protein synthesis, and transfer-RNA-derived RNA fragments comprise another class of non-coding RNA with diverse functions. They undergo complex processing by RBPs, including RNA-modifying enzymes. RNA modifications, such as methylation, pseudouridylation and RNA editing, can affect mRNA stability, translation efficiency and splicing accuracy. RNA-modifying enzymes and RBPs function together to regulate these modifications, thereby expanding the repertoire of post-transcriptional regulatory mechanisms in cells.

RBPs are typically composed of sets of RNA-binding domains¹, with each domain having limited binding specificity for a particular RNA sequence or structure. Combinations of these RNA-binding domains have increased affinities for both sequence and structural motifs, but predicting these interactions de novo remains a formidable challenge. Moreover, the discovery of thousands of new RBPs through recent proteome-scale experimental and computational methods — such as RBDmap² and RNA-interactome capture³ (Table 1) — suggests that around half of these RBPs lack known RNA-binding domains, which poses further difficulties in identifying protein–RNA interactions.

Table 1 |.

CLIP and non-CLIP methodologies for profiling protein–RNA interactions

Aim	Strategy	Methods
Identify RNAs bound to protein	Ultraviolet crosslinking	HITS-CLIP or CLIP-seq4,5, PAR-CLIP6, iCLIP7, eCLIP8, irCLIP9
	RNA base editing	DART-seq188, STAMP184, TRIBE185, TRIBE-STAMP186, HYPERTRIBE191, PRINTER192, REMORA193
	Catalytically dead Cas variant plus guide RNA	CRUIS194, CARPID195
	Chemical and biotin labelling	RNA BioID196, Proximity-CLIP197, APEX-seq198,199
Capture interactions of specific RNAs	Ultraviolet crosslinking and mass spectrometry	ChIRP200, RAP-MS201, VIR-CLASP154, vIPR202, iDRIP203, CHART204
	Proximity labelling and mass spectrometry	RaPID205
	RNA–miRNA proximity ligation	CLASH206
Capture the transcriptome-wide interactome	Ultraviolet crosslinking and mass spectrometry	CLASP154, RIC207,208, XRNAX209, OOPS210
	Photoreactive nucleosides	Proximity-CLIP197
	Reactive nucleosides and chemical crosslinking	RICK211, CARIC212

APEX-seq, RNA sequencing based on direct proximity labelling of RNA using the peroxidase enzyme APEX2; CARIC, click chemistry-assisted RNA interactome capture; CARPID, CRISPR-assisted RNA–protein interaction detection method; CHART, capture hybridization analysis of RNA targets; ChIRP, comprehensive identification of RNA-binding proteins by mass spectrometry; CLASH, crosslinking, ligation and sequencing of hybrids; CLASP, crosslinking and solid-phase purification; CLIP, ultraviolet crosslinking followed by immunoprecipitation; CRUIS, CRISPR-based RNA-united interacting system; DART-seq, deamination adjacent to RNA modification targets followed by sequencing; eCLIP, enhanced CLIP; HITS-CLIP, high-throughput sequencing of RNA isolated by CLIP; iCLIP, individual-nucleotide resolution CLIP; iDRIP, identification of direct RNA-interacting proteins; irCLIP, infrared CLIP; OOPS, orthogonal organic phase separation; PAR-CLIP, photoactivatable ribonucleoside-enhanced CLIP; PRINTER, protein–RNA interaction-based triaging of enzymes that edit RNA; RaPID, RNA–protein interaction detection; RAP-MS, RNA antisense purification coupled with mass spectrometry; RIC, RNA interactome capture; RICK, capture of the newly transcribed RNA interactome using click chemistry; REMORA, RNA-encoded molecular recording in adenosines; STAMP, surveying targets by APOBEC-mediated profiling; TRIBE, targets of RNA-binding proteins identified by editing; vIPR, in vivo interactions by pulldown of RNA; VIR-CLASP, viral crosslinking and solid-phase purification; XRNAX, protein-crosslinked RNA extraction.

To address these challenges, the field has relied heavily on experimental methodologies involving ultraviolet crosslinking followed by immunoprecipitation (CLIP), a powerful tool with which to gain information as to where RBPs interact with RNAs at nucleotide-level resolution^4,5. Following RNA fragmentation, ultraviolet-crosslinked RBP–RNA complexes are immunoprecipitated and protein-linked RNA fragments are isolated and subjected to RNA adaptor ligation and complementary DNA library preparation, followed by deep sequencing. Advances in CLIP methodologies, such as photoactivatable ribonucleoside-enhanced CLIP (PAR-CLIP)⁶, individual-nucleotide resolution CLIP (iCLIP)⁷, enhanced CLIP (eCLIP)⁸ and infrared CLIP (irCLIP)⁹ (Table 1), and the computational tools needed to process these datasets^10–13 have facilitated the generation of comprehensive protein–RNA atlases at the transcriptome-wide level.

In this Review, we focus on the mechanistic insights gained from more than a decade of using CLIP-based methodologies to map protein–RNA interactions in mammalian systems. Large-scale CLIP-based datasets, when combined with other analyses such as differential expression, subcellular localization and functional reporter assays¹⁴, offer invaluable information for understanding the regulatory effects of protein–RNA interactions. However, deciphering these complex datasets to identify key functions of these interactions remains challenging. Thus, we provide a summary of reference features and profiles of protein–RNA interactions, shedding light on how RBPs modulate post-transcriptional gene regulation. In addition, we highlight how protein–RNA maps are useful tools for studying changes in cellular states in both healthy and disease settings. Finally, we explore the exciting potential of using protein–RNA maps to investigate the structure–function relationships of small molecules that perturb RBP biology.

Pre-mRNA processing

RBPs have integral roles in all aspects of mRNA processing, and their precise interactions and coordination ensure the fidelity of gene expression (Box 1). Here, we describe how comprehensive protein–RNA interactomes have provided insights into multiple stages of RNA processing.

Co-transcriptional processing

During RNA polymerase II (RNAPII)-mediated transcription, the nascent RNA undergoes a series of co-transcriptional processing steps, including 5′ capping, splicing, RNA modification, 3′ end cleavage and polyadenylation. The nuclear cap-binding complex (CBC; a heterodimer of NCBP1 and NCBP2) has a crucial role in these processes by binding to N⁷-methylguanosine (m⁷G)-capped RNAs^15–18 and initiating subsequent processing events specific to different RNA families, such as mRNAs, small nuclear RNAs (snRNAs), small nucleolar RNAs (snoRNAs), promoter upstream transcripts (PROMPTs) and transfer RNAs (tRNAs). To understand how the CBC identifies and directs different RNA families to the appropriate processing machinery, iCLIP was carried out on NCBP2 (also known as CBP20) and the CBC effectors ARS2, PHAX and ZC3H18. These effector proteins compete for CBC binding and if bound have different effects on the levels of their RNA substrates. The iCLIP data showed that they interact with cap-proximal regions of targeted RNAs, indicating that this interaction is primarily due to binding with the m⁷G cap (Fig. 1a). Surprisingly, despite their previously published roles in snRNA export, binding of ARS2 and PHAX was only moderately enriched on snRNAs compared to binding of NCBP2, and these effectors were not specific for a particular transcript type, with their RNA interactions being solely dependent on cap binding¹⁷. This implies that RNA–CBC interactions are not sufficient to define RNA maturation pathways, which may depend on other interacting RBPs or ribonucleoprotein complexes. Indeed, a recent characterization of heterogeneous nuclear ribonucleoprotein C (HNRNPC) showed that it competes with PHAX for binding to transcripts of >200–300 nucleotides. This competition with HNRNPC probably constrains PHAX binding to the caps of shorter snRNAs and snoRNAs¹⁹, thus providing a model for understanding substrate recognition by PHAX despite the apparently promiscuous binding of this effector. Owing to the highly transient nature of CBC interactions with RNAs, temporal iCLIP was developed — using inhibitors of RNAPII-mediated transcriptional elongation followed by ultraviolet crosslinking — to capture the binding of RBPs to sample RNA targets 0–60 minutes after transcription reactivation²⁰. Intriguingly, the study found that whereas NCBP2 primarily binds to the 5′ cap, NCBP1 (also known as CBP80) binds the elongating RNA with increasing distance from the transcription start site at a rate that matches the RNAPII elongation speed²⁰. This provides evidence for a co-transcriptional role of NCBP1 in recruiting additional factors to direct the fate of the elongating RNA.

Fig. 1 |. Insights into pre-mRNA processing from protein–RNA interaction profiles.

a–h, Schematics show the read density of RNA-binding protein (RBP)–RNA mapping assays distributed across salient features in nuclear RNA. All metaplots are idealized observations of ultraviolet crosslinking followed by immunoprecipitation (CLIP) read densities. Blue boxes represent exons; black lines represent introns. a, Cap-binding proteins such as the cap-binding complex (CBC) and its component NCBP2 enrich reads near the 5′ untranslated region (UTR) in unspliced exons. b, The RBP FUS has a higher read density at the beginning of long introns that gradually decreases towards the 3′ end, creating the ‘sawtooth’ pattern observed for the co-transcriptional deposition of FUS. c, CLIP studies of the spliceosome component SNRPB define regions of interaction with RNA near splice sites and branchpoints. d, psiCLIP studies of C, C* and P spliceosome complexes via immunoprecipitation of the helicases PRP16 and PRP22 show their read density profiles on in-vitro-transcribed UBC4 RNA (for the C complex) and ACT1 RNA (for C* and P complexes). The ATPase defective mutants PRP16-G378A and PRP22-K512A have similar profiles as their respective wild-type (WT) proteins, indicating that the function of both proteins does not involve translocation. Exon ligation shifts PRP22 binding (wild-type and mutant) from both intronic regions and the 3′ exon of RNA in the C* complex to only the 3′ exon in the P complex. e, The cleavage and polyadenylation factors (CPSFs), cleavage stimulation factors (CSTFs) and poly(A) polymerase (PAP), which carry out 3′ end-processing of RNA, have well-defined positions of interaction with RNA that determine the sites for cleavage and polyadenylation. f, The nuclear export factors ALYREF and CHTOP interact with the 5′ and 3′ ends of the transcript, respectively, whereas NXF1 lacks sequence specificity. g, SRSF proteins, which are involved in nucleocytoplasmic shuttling, map to exonic regions around splice sites, similar to the binding pattern of NXF1. h, The exon junction complex protein eIF4A3 binds to RNA about 24 nucleotides (nt) upstream of the exon 3′ ends.

In addition to CBC proteins, the co-transcriptional deposition of the RBP FUS (also known as TLS) has been captured by CLIP-seq. On long intronic regions, a higher read density of FUS is observed at the beginning of the region, gradually decreasing towards the 3′ end. This sawtooth-shaped binding pattern of FUS on long introns provides supporting evidence for its involvement in transcriptional elongation^21,22 (Fig. 1b). As confirmation of the co-transcriptional function of FUS, further investigations showed an essential role for FUS in the co-transcriptional recruitment of U1 snRNP by mediating its interaction with RNAPII, which is a requirement for splicing to occur²³. Thus, close examination of the patterns of CLIP-seq profiles can shed light on the co-transcriptional dynamics of RNA-processing factors such as the CBC and FUS, providing insights into the regulatory mechanisms of RNA processing during transcriptional elongation.

Our understanding of RBPs has also expanded to reveal their involvement in direct interactions with chromatin to control transcriptional processes, including gene activation, chromosome three-dimensional organization, long-range enhancer–promoter interactions and phase transitions^24–29 (reviewed in ref. 30). DNA-binding transcription factors such as SOX2 (ref. 31), CTCF³² and YY1 (ref. 33) are increasingly recognized as also being RBPs that interact directly with nascent transcripts. With recent studies indicating that nearly half of the transcription factors in human K562 cells can bind RNA, primarily through arginine-rich motifs³⁴, we anticipate that future protein–RNA mapping studies will enable a deeper understanding of the coordination between DNA-binding and RNA-binding interactions in the control of transcriptional programmes.

Spliceosome regulation

Protein–RNA interaction mapping strategies to investigate spliceosome processes have elucidated complex interactions beyond individual proteins^4,5,7,35,36. Spliceosome profiling studies in yeast and mammalian cells have identified new RNA regions near splice sites and branchpoints that correspond to key points of interaction with the spliceosome^37,38 (Fig. 1c). These studies have also uncovered non-canonical splicing events and determinants of intron retention that trigger degradation³⁹. However, the stage-specific mechanisms of the core spliceosome remain obscure because the protein–RNA binding profiles that have been elucidated are aggregates across various splicing stages, possibly favouring rate-limiting steps in this process. In particular, the transient nature of helicase interactions with pre-mRNA substrates makes it difficult to study their specific RNA interaction sites. The presumptive model of helicase function involves processive translocation to remodel the spliceosome⁴⁰. Using purified spliceosome C, C* and P complexes crosslinked with in-vitro-transcribed RNA (psiCLIP), a recent study delineated the interaction of PRP16 and PRP22 helicases with pre-mRNA before and after exon ligation (the second stage of splicing)⁴¹. Both wild-type PRP16 and an ATPase mutant that lacks translocation activity broadly bind the region of RNA between the branchpoint adenosine and the 3′ splice site. This pattern suggests that multiple rounds of binding and dissociation by one or more helicase molecules can occur in the absence of translocation (Fig. 1d). After exon ligation, PRP22 shifts its pattern of binding from both intronic regions and the 3′ exon of RNA in the C* spliceosome complex to only the 3′ exon in the P complex. This led the authors to propose a ‘winch model’ of helicase function involving anchoring of PRP22 on the 3′ exon and pulling on the RNA substrate to release the mRNA, followed by PRP22 dissociation⁴² (Fig. 1d). This example highlights the potential for CLIP-based studies to map RNA-binding sites for even transient interactions such as are involved in helicase activity.

Safeguarding against repeat elements

By showing that RBPs bind to repetitive elements, several studies have highlighted that a subset of RBPs may safeguard the integrity of the transcriptome. The repetitive nature of numerous retrotransponson-derived elements — such as Alu elements, long interspersed nuclear elements (LINEs) and short interspersed nuclear elements (SINEs) — has historically posed challenges for determining their precise location in the genome. However, efforts to re-examine the mapping of iCLIP and eCLIP reads to repeat elements, irrespective of their genomic locations, have provided new perspectives. For example, HNRNPC has been observed to interact directly with cryptic splice sites in Alu elements, effectively preventing aberrant exonization (the acquisition of new exons) facilitated by U2AF65 (ref. 43). Furthermore, extensive analysis of iCLIP and eCLIP datasets has revealed the binding patterns of repressive RBPs such as MATR3 and PTBP1 on LINE1 (L1) elements⁴⁴. Multivalent binding of these RBPs insulates L1 elements and their surrounding regions from RNA processing. This results in the preservation of unspliced, evolutionarily young L1 elements in deep intronic regions; by contrast, tissue-specific alternatively spliced exons include evolutionarily older L1 elements that have lost motifs for the multivalent binding of repressive RBPs. Knockdown of RBPs that bind to L1 elements results in the inappropriate inclusion of cryptic poly(A) sites and splice sites within LINEs^45,46. Thus, protein–RNA interaction mapping has revealed the roles of RBPs in preventing repeat element dysregulation and preserving transcriptome integrity.

3′ end processing and alternative polyadenylation

As transcription nears its termination, 3′ end processing of the nascent RNA is carried out by cleavage and polyadenylation factors (CPSFs), cleavage stimulation factors (CSTFs) and poly(A) polymerase (PAP). The selection of cleavage sites by these proteins determines the sites of alternative polyadenylation and the length of the 3′ untranslated region (UTR), thereby affecting mRNA stability, localization and/or translation. However, the mechanisms involved in the selection of cleavage sites and their consequences had remained elusive. Using CLIP-seq, these processing factors were mapped to the 3′ end processing sites of RNA, revealing distinct binding patterns and rules for alternative polyadenylation^47,48 (Fig. 1e). The binding maps of these cleavage and polyadenylation factors can be compared with the interaction profiles of other 3′ UTR-binding RBPs to identify co-regulatory processes. For example, the adenosine deaminase ADAR1 was found to interact with 3′ UTRs depleted of interaction sites for CSTF64, CSTF64τ and CPSF6 (also known as CFIM68), leading to changes in 3′ UTR lengths⁴⁹. In addition, the multifunctional splicing regulator FUS is recruited by U1 snRNP to interact with the region upstream of poly(A) sites in nascent transcripts, which affects their alternative polyadenylation⁵⁰. Thus, CLIP studies continue to unravel the interplay between RBPs and alternative polyadenylation.

mRNA export mechanisms

mRNA export from the nucleus to the cytoplasm is orchestrated by nuclear export factor 1 (NXF1) and various adaptor proteins such as ALYREF and CHTOP. Whereas the RNA-binding domain of NXF1 is known to lack sequence specificity⁵¹, iCLIP studies have revealed binding preferences of ALYREF and CHTOP (Fig. 1f). ALYREF has a 5′ bias for binding mRNA within the first exon, which is linked to its transient interaction with the CBC⁵². ALYREF also binds upstream of the exon junction complex (EJC; see below), which suggests that it continues to depend on the EJC for deposition after recruitment by the CBC⁵². Independently of the CBC, ALYREF is anchored upstream of the exon–exon junction after the first transesterification step of pre-mRNA splicing but before the second transesterification step in a splicing-dependent manner²⁰. Using the binding profiles of known export proteins as a reference has enabled annotation of other proteins with export regulatory functions. For example, iCLIP of the serine/arginine-rich proteins SRSF1 to SRSF7 showed that they bind to RNA adjacent to NXF1, with SRSF3 and SRSF7 being involved in coupling alternative splicing and polyadenylation to NXF1-mediated mRNA export⁵³ (Fig. 1g). Thus, the protein–RNA interaction profiles of nuclear export factors can help us to understand the regulation of mRNA export mechanisms and the specificity of adaptor proteins.

Regulation of RNA processing by the EJC

The EJC is composed of four core proteins (eIF4A3, BTZ, MAGOH and RBM8A) and several peripheral proteins (for example, UPF3B, RNPS1 and Acinus), with effects on the splicing, cellular localization, translation and nonsense-mediated decay (NMD; see next section) of RNAs. The canonical binding site of the EJC was initially identified as being approximately 24 nucleotides before exon–exon junctions^54,55 (Fig. 1h). However, subsequent studies showed that eIF4A3, which was initially thought to be exclusively part of the EJC, has only 40–50% of its target sites located in this region. To identify a central EJC protein, iCLIP analysis was carried out on eIF4A3, BTZ, UPF3B and RNPS1. BTZ was found to bind exclusively at sites about 24 nucleotides upstream of exon–exon junctions (around 90% of target sites), whereas eIF4A3, UPF3B and RNPS1 had a greater percentage of binding to non-EJC sites. Further investigations revealed that RNPS1 and BTZ bind the EJC in the nucleus and cytoplasm, respectively, in a mutually exclusive manner. RNA–protein immunoprecipitation in tandem (RIPiT) analysis showed that the RNPS1-containing EJC is the main EJC involved in binding to targets of NMD, whereas the BTZ-containing EJC is required for NMD of select RNAs only, which suggests that different EJCs have distinct roles in NMD⁵⁶.

mRNA stability and translation regulation

After pre-mRNA processing, mature mRNA molecules undergo further regulation of their stability and translational efficiency. Here, we describe examples of the complex and dynamic protein–RNA interactions involved in maintaining mRNA integrity as revealed by CLIP-based studies.

Nonsense-mediated decay

NMD is a crucial quality control process that eliminates mRNA molecules containing premature stop codons. If an EJC is present more than about 50–55 nucleotides downstream of a stop codon⁵⁷, proteins UPF2 and UPF3, associated with the EJC, recruit UPF1, the essential mediator of NMD, which initiates mRNA degradation. CLIP-seq studies have provided evidence that UPF1 preferentially binds GC-rich⁵⁸ and long 3′ UTRs^59,60 in a translation-dependent manner (Fig. 2a). An early model suggested that UPF1 is recruited to mRNA by ribosomes stalling at stop codons where signals for the proper termination of translation are absent⁶¹. However, an iCLIP study found that UPF1 binds to a small percentage of long non-coding RNAs (lncRNAs), which suggests it does not have an intrinsic preference for binding to translated mRNA, and that translation inhibition redistributes UPF1 binding from the 3′ UTR to coding regions. This suggests that ribosomes mediating translation displace UPF1 molecules that are already interacting with mRNA transcripts^59,62. The complexity of NMD will be further delineated through advances in mapping the protein–RNA interactions of NMD factors.

Fig. 2 |. Insights into cytoplasmic mRNA regulation from protein–RNA interaction profiles.

a–e, Schematics show the read density of RNA-binding protein (RBP)–RNA mapping assays distributed across salient features in cytoplasmic RNA. All metaplots are idealized observations of ultraviolet crosslinking followed by immunoprecipitation (CLIP) read densities. a, The nonsense-mediated decay protein UPF1 binds downstream of the termination codon; binding is disrupted by cycloheximide (CHX) treatment, which blocks protein synthesis. b, RBPs involved in translation regulation bind to different regions in the mRNA substrate that are characteristic of their specific mechanisms of regulation. c, Interaction of the La-related protein LARP1, which is involved in translation inhibition, with mRNA changes depending on its phosphorylation state; phosphorylated LARP1 no longer binds to 5′ terminal oligopyrimidine (5′ TOP) motif-containing mRNAs. d, AATF binds 45S precursor ribosomal RNA (rRNA), particularly in spacer sequences outside 18S and 28S rRNA, to facilitate rRNA biogenesis, whereas DDX3 binds to 18S rRNA to regulate translation. e, Interactions with the 3′ untranslated region (UTR) of mRNAs are most frequently associated with RBPs regulating mRNA stability. Pumilio (PUM) proteins and argonaute 2 (AGO2) bind to regions centred around the PUM responsive element (PRE) motif and microRNA seed, respectively. Sequence logos show the signature PRE motif bound by PUM proteins, and AU-rich elements (AREs) bound by ZFP36 and ELAVL1. Also shown is the positional distribution of read densities for the 3′ UTR-binding proteins AGO2, ZFP36 and ELAVL1. CDS, coding sequence; ETS, external transcribed spacer; ITS, internal transcribed spacer.

Fine-tuning translation

In eukaryotic cells, translation is a tightly regulated process coordinated by multiple RBPs that interact with mRNAs and with ribosomes. CLIP-based studies have identified patterns of binding of these RBPs to different regions of mRNAs that are characteristic of the specific mechanisms of each RBP (Fig. 2b). For example, translation initiation is reflected in the 5′ UTR binding of factors, such as eIF3 and DDX3, that are involved in the assembly and scanning function of the pre-initiation complex^45,63,64. CNBP, a CCHC-type zinc finger protein, binds downstream of the start codon at G-rich elements, probably resolving stable structures to enhance ribosome elongation⁶⁵. Binding of an RBP to the mRNA coding sequence indicates co-translational regulation of elongation. For example, Fragile X messenger ribonucleoprotein 1 (FMR1) represses translation by inducing ribosome stalling by directly binding to the ribosome including during elongation^66,67. Cytoplasmic polyadenylation element-binding (CPEB) proteins bind to the 3′ UTR and regulate translation by recruiting TACC3 (also known as Maskin) to compete with eIF4G for binding to eIF4E in the assembly of the 48S translation initiation complex^68–70. These reference profiles correlating with different properties of translation-regulating RBPs allow for hypotheses about the function of other RBPs to be tested. For example, the La-related protein LARP1, which is known to be involved in translation inhibition, has sequence-specific crosslinking to the 5′ terminal oligopyrimidine tract (5′ TOP) at the 5′ UTR. Phosphorylation of LARP1 by mTORC1 relieves this inhibition, enabling translation initiation and ribosome biogenesis, as evident by the disappearance of binding enrichment in the 5′ UTR for the phosphorylated form of LARP1 (ref. 71) (Fig. 2c). In summary, distinct patterns of protein–mRNA interactions provide clues to the regulation of translation initiation, elongation and repression.

Ribosomal RNA dynamics

Studies using eCLIP to show protein interactions with ribosomal RNA (rRNA) have provided insights into ribosomal biogenesis and co-translational regulation⁴⁵. A comprehensive analysis of eCLIP data for 150 RBPs made the surprising observation that about 70% of these predominantly bind non-mRNA elements, including rRNAs, prompting a re-evaluation of their functions⁴⁵. Enrichment patterns of binding to the 45S precursor and the 28S and 18S rRNAs vary among rRNA-enriched RBPs, revealing differences in their mechanisms of regulation. For example, eCLIP of the RBP AATF revealed strong enrichment of binding to snoRNAs and the 45S precursor rRNA, particularly in spacer sequences outside 18S and 28S rRNA, lending support to a role for AATF in ribosomal biogenesis⁷² (Fig. 2d). By contrast, human DDX3 binds to helix 16 of 18S rRNA near the mRNA entry channel, which affects the translation of specific transcripts with structured 5′ UTRs⁷³ (Fig. 2d). With the recent discovery of numerous ribosome-associated proteins that contribute to ribosomal heterogeneity⁷⁴, investigating their interactions with rRNAs and mRNAs promises to uncover diverse regulatory effects on translation.

RNA stability and translation

RNA stability and translation are finely modulated by RBPs, often through interactions with the 3′ UTR of transcripts (Fig. 2e). A comprehensive investigation in the yeast Saccharomyces cerevisiae of 30 RNA degradation proteins, including XRN1 exonuclease and factors from decay complexes, showed patterns of co-localization and shared targeting sites within these complexes, which suggests that they have coordinated effects in target selection and involvement of different degradation processes⁷⁵. Complex, counteracting protein–RNA networks involve RBPs that bind to the 3′ UTR in a manner dependent on cell type and disease state. These RBPs include AU-rich element-binding protein families, such as ELAV⁷⁶, CELF⁷⁷, ZFP36 (ref. 78), HNRNPD (also known as AUF1), KH-type splicing regulatory protein (KHSRP; also known as FUBP2)⁷⁹ and the splicing factor TIA1 (refs. 36,80), and the pumilio (PUM) proteins^6,8. Most of these RBPs destabilize specific target transcripts by recruiting mRNA degradation proteins to the substrate RNA⁸¹. By contrast, IMP family proteins function to stabilize target mRNAs^82–84. IMP2 is highly expressed in glioblastoma cells. PAR-CLIP of IMP2 in human pluripotent stem cells showed that IMP2 binds to let-7 microRNA (miRNA)-recognition elements and prevents let-7-mediated silencing of target genes^6,85. The interplay between ELAV proteins and ZFP36 also reveals their competing effects on mRNA stability in response to various stimuli⁸⁶. Intriguingly, plant pentatricopeptide repeat protein 10 (PPR10) outcompetes PUM proteins with its high affinity and long binding interface, thereby preventing RNA degradation⁸⁷. This exemplifies how the counteracting regulatory effects of RBP–RNA interactions at the 3′ UTR are an important mechanism for fine-tuning mRNA stability.

Profiling the interactions of RBPs with mRNA 3′ UTRs uncovers their multifaceted roles. For example, the DNA-binding transcription factor SP1 binds upstream of mRNA 3′ end distal cleavage sites to prevent their use as polyadenylation sites and, in turn, represses mRNA transcript levels⁸⁸. In addition, the splicing factors CELF1, MBNL1, ELAV, HNRNPC and PTBP1 have marked 3′ UTR binding preferences, which reveals their functions in regulating co-splicing alternative polyadenylation^89–94. In fact, the coordinated actions of 3′ UTR-associated RBPs are not restricted to effects on mRNA stability, because the interaction of many RBPs with the 3′ UTR also influences translation^95–97. An early CLIP-seq study of the RBP LIN28A revealed extensive transcriptome binding, as well as binding to let-7 miRNA precursors, and integrated analysis with polysome profiling suggests that LIN28A has a global role in regulating translation⁹⁸. However, more recent insights suggest that the effect of LIN28B on translation involves competition with argonaute proteins (part of the RNA-induced silencing complex), which alters miRNA dynamics and target gene regulation⁹⁹. The overlapping interactions of 3′ UTRs with various factors underscore their complex role in regulating RNA stability and translation.

Long and small non-coding RNAs

Although CLIP-based studies have focused mainly on protein-coding genes, their potential extends to uncovering interactions between RBPs and non-coding RNAs¹⁰⁰. Extensive CLIP datasets¹⁰¹ and analyses of more than 22,000 tumour-associated RBP–lncRNA interactions from 50 studies¹⁰⁰ hint at a large reservoir of untapped knowledge. For example, database resources such as ENCORI/starBase predicted the interaction of the lncRNA linc00668 with the RBP SND1. This interaction was validated and found to regulate transcriptional activation of SND1 target genes in breast cancer cells¹⁰². CLIP datasets have also been used to investigate interactions between RBPs and miRNAs. A combination of eCLIP datasets for 126 RBPs in HepG2 and K562 cell lines were cross-referenced with annotations for pre-miRNA loci, which led to the validation of ten RBPs newly implicated in the regulation of miRNA biogenesis¹⁰³ (Supplementary Fig. 1a). Notably, many of these RBPs exhibited binding to fewer than 25 distinct miRNA loci, which distinguishes them from broad regulators of miRNAs such as DROSHA. These RBP–miRNA interactions were cell-type specific, and hold promise for further exploration in disease models. These examples showcase how database resources can be used as a starting point to identify candidate RBP–non-coding RNA interactions.

The use of CLIP has offered insights into RBP–snoRNA interactions. For example, eCLIP analysis of RBPs in the ENCODE database revealed interactions between the RBP LARP7 and C/D box snoRNAs¹⁰⁴ (Supplementary Fig. 1b). C/D box snoRNAs guide enzymatic modifications of target RNAs within the snoRNP complex. Specifically, seven associations between LARP7 and C/D box snoRNAs that guide 2′-O-methylation of U6 snRNA were identified, marked by a conserved snoRNA motif. This interaction between LARP7 and snoRNAs revealed a novel mechanism for the developmental disorder Alazami syndrome, a disease characterized by loss-of-function LARP7 mutations. In cells from patients with Alazami syndrome, 2′-O-methylation of U6 snRNA was reduced, which suggests that disruption of LARP7–snoRNA interactions may contribute to the altered splicing patterns observed in this disease.

CLIP data have also shown that the snoRNA binding profiles of core snoRNA-regulatory RBPs are more expansive than previously predicted. PAR-CLIP of fibrillarin (FBL), NOP56 and NOP58, which are core C/D box snoRNA-regulatory proteins, showed that they have similar binding profiles and a preference for binding C/D box snoRNAs¹⁰⁵. Surprisingly, however, these RBPs were found also to bind H/ACA box snoRNAs, and the H/ACA box snoRNA-binding protein dyskerin was shown also to interact with some C/D box snoRNAs. This suggests that core RBPs regulating C/D box and H/ACA box snoRNAs can also bind different classes of snoRNAs, which may be a compensation mechanism in which loss of a C/D box-binding core protein could be somewhat compensated by an H/ACA box-binding protein. Conversely, these processes may be dependent on each other, in which case reduction of C/D box-binding proteins could dysregulate the functions of H/ACA box snoRNAs. A recent study indicates the latter, as knockdown of the C/D box-binding protein FBL was found to reduce pseudouridylation. This was surprising as C/D box snoRNAs are implicated in 2′-O-methylation whereas H/ACA box snoRNAs are involved in guiding pseudouridylation¹⁰⁶. This indicates that further knockdown studies of core snoRNA-binding proteins and subsequent CLIP analyses are needed to determine the relevance of binding of core snoRNA RBPs to the alternative class of snoRNAs.

CLIP methodologies have enabled the discovery of novel genomic origins of tRNAs. The pre-tRNA processing protein SSB (also known as La) attaches to pre-tRNAs after RNAPII-mediated transcription. PAR-CLIP of SSB not only revealed its binding to the 3′ oligo-U terminus of known pre-tRNAs, but also led to the discovery of seven novel pre-tRNAs that had not been identified by sequencing of mature tRNAs¹⁰⁷ (Supplementary Fig. 1c). As a final example, a meta-analysis of CLIP datasets of Argonaute family proteins (except AGO2) uncovered their affinity for tRNA-derived RNA fragments, which hints at a novel mechanism of RNA silencing mediated by tRNA-derived RNA fragments, resembling the silencing mediated by miRNAs¹⁰⁸ (Supplementary Fig. 1d). Collectively, these results show that CLIP-based studies not only validate known RBP–non-coding RNA interactions but can also uncover novel interactions¹⁰⁹.

Modification of RNAs

More than 100 types of RNA modification are present on coding and non-coding RNAs, although most of their functions are unclear. ‘Writer’ and ‘eraser’ proteins add and remove these modifications, and ‘reader’ proteins translate them into regulatory effects (Supplementary Fig. 2a). tRNAs, in particular, are extensively modified at specific sites by various enzymes^110–112 (Supplementary Fig. 2b). Modification-specific antibodies can be used to enrich modified RNAs, and sequencing methods harnessing modification chemistry can be used to map their distribution across the transcriptome^{111,113–116}. Integrating CLIP-seq for ‘reader’ proteins with modification sequencing has been pivotal for understanding the consequences of RNA modifications (Supplementary Fig. 2c).

N⁶-methyladenosine (m⁶A) is the best characterized RNA modification and is highly abundant throughout the transcriptome¹¹⁷. CLIP-seq showed that YTHDF ‘reader’ proteins bind to regions near stop codons and 3′ UTRs that are enriched with m⁶A modifications to redundantly promote mRNA degradation (Supplementary Fig. 2d)^118,119, whereas IGF2BP proteins compete with YTHDF proteins to stabilize m⁶A-modified mRNAs¹²⁰. Given that m⁶A methylation states change dynamically across developmental stages¹²¹ and disease states^122–124, mapping of protein interactions with m⁶A-modified RNAs will help to elucidate their function in diverse cellular contexts.

In addition to m⁶A, CLIP has also been used to study 5-methylcytosine (m⁵C), uridylation and pseudouridine modifications. PAR-CLIP of the m⁵C ‘writer’ protein NSUN2 uncovered the role of m⁵C in translationally repressing GC-rich coding RNA targets¹²⁵ and in processing the small RNA svRNA4 from vault RNAs to silence CACNG7 and CACNG8 calcium transporters¹¹¹. In addition, PAR-CLIP studies of the m⁵C ‘reader’ proteins YBX1 (ref. 126) and ALYREF¹²⁷ have uncovered their m⁵C-modification-dependent effects on stabilizing target mRNAs and promoting mRNA export, respectively.

Uridylation of RNA regulates mRNA stability, miRNA biogenesis and quality control¹²⁸. RNA molecules can be uridylated by terminal uridylyl transferases (TUTases) for degradation by the closely associated exoribonuclease DIS3L2. CLIP-seq of a catalytically inactive mutant of DIS3L2 mapped to the uridylated 3′ UTR of histone RNAs, the uridylated 5′ UTR of mRNA fragments and in the proximity of transcription start sites, which suggests a specific role of uridylation in the quality control of transcription start site-associated short RNAs^129,130 (Supplementary Fig. 2e). Future studies mapping interactions of DIS3L2 with uridylated RNAs in relevant models could reveal stress-dependent effects¹²⁸ of the TUTase–DIS3L2 pathway in diseases such as Perlman syndrome, a rare kidney condition associated with germline mutations in DIS3L2 (ref. 131).

Pseudouridine modification is directed primarily by the pseudouridine synthesome (PUS) family of proteins. In a recent study, CLIP-seq of PUS10 identified binding to primary miRNAs to regulate miRNA biogenesis in the nucleus¹¹². Interestingly, unlike other PUS proteins, PUS10 binds tRNAs in the cytoplasm instead of nucleus to promote site-specific pseudouridylation (Supplementary Fig. 2f). Thus, profiling protein–RNA interactions in different subcellular compartments can reveal subcellular-specific consequences of RNA modifications.

Dynamic RBP interaction networks

Large-scale protein–RNA interactome studies have focused mainly on cancer cell lines, providing fundamental insights into the functions of RBPs. However, understanding the relevance of RBPs in other disease states requires expanding the CLIP dataset to encompass diverse cell types and conditions. Here, we provide examples that highlight the importance of evaluating protein–RNA interactions in a range of cell types and cell states, as well as in health and disease.

Cell type- and state-specific interactions

Recent CLIP-based studies have shed light on cell-type-specific functions of RBPs. For example, the RBP RBFOX3 (also known as NeuN) is expressed highly in neurons and binds RNAs that have crucial roles in regulating cellular plasticity. Mapping of RBFOX3 binding sites uncovered links to genes that shape brain development⁵, and in Rbfox3^−/− mice, dendritic spine anomalies were observed⁶. External factors such as inflammation can also shape the binding of RBPs to RNAs. The inflammation-associated RBPs FXR1 and ELAVL1 have opposing effects on stability of the mRNA that encodes tumour necrosis factor (TNF) through binding to AU-rich elements in the 3′ UTR¹³². FXR1 induction by the anti-inflammatory cytokine IL-19 counters the stability-enhancing effects of ELAVL1 on TNF mRNA. In a human monocyte cell line, ELAVL1 binding to mRNA is augmented by treatment with cGAMP, a signalling intermediate of the innate immune response (Fig. 3a); this correlates with decreased stability of immune-related transcripts upon ELAVL1 knockout. Marked alterations in RBP–RNA binding patterns can also be induced by cellular stress. For example, the RNA interactions of G3BP1, TDP43 (also known as TARDBP) and TIA1, key RBPs associated with stress granules, are modified in the context of puromycin stress in human motor neurons (Fig. 3b). Under stress conditions, TDP43 binding at cryptic exon 2a of STMN2 mRNA is no longer observed, leading to decreased mRNA and protein levels of STMN2 (ref. 133), a hallmark of amyotrophic lateral sclerosis. Decreased levels of STMN2 are observed in the spinal cords of patients with amyotrophic lateral sclerosis, and studies in induced pluripotent stem cell (iPSC)-derived motor neurons indicate that loss of STMN2 protein is caused by loss of nuclear TDP43 and altered splicing of STMN2 (refs. 134,135). Although puromycin is an artificial stressor, this CLIP study suggests that RBP–RNA interactions observed in neurodegeneration can be recapitulated in stressed cells, thus highlighting the need to profile RBP–RNA interactions in different cellular states.

Fig. 3 |. Dynamics of protein–RNA interactions depending on cellular state.

a–c, Schematics show the read density of RNA-binding protein (RBP)–RNA mapping assays distributed across salient RNA features. All metaplots are idealized observations of ultraviolet crosslinking followed by immunoprecipitation (CLIP) read densities. a, Stimulation of human monocyte THP-1 cells with the innate immune mediator cGAMP (cyclic GMP–AMP) results in an increase in the number of transcripts bound by ELAVL1 at the 3′ untranslated region (UTR). b, Puromycin stimulation of human induced pluripotent stem cell (iPSC)-derived motor neurons results in a shift in the RNA-binding patterns of the stress granule proteins G3BP1 (from coding sequence (CDS) to 5′ UTR (left)) and TDP43 (from 3′ UTR to CDS (right)). c, Wild-type RBM20 (nuclear) primarily binds introns in human iPSC-derived cardiomyocytes, whereas R636S-mutated RBM20 (cytoplasmic) preferentially binds the 3′ UTR.

Disease-associated interactions

RNA metabolism is highly dysregulated in cancer and many RBPs, such as MSI2 and STAU2, have been implicated in its processes¹³⁶. STAU2 was identified in a myeloid leukaemia screen, and CLIP studies subsequently identified that STAU2 primarily binds the 3′ UTR of transcripts involved in RAS signalling and chromatin binding. This implicates STAU2 as a potential upstream modulator of chromatin modifications, which are a therapeutic target of interest in cancer. CLIP studies of cardiovascular-disease-associated RBPs in cardiac cells have highlighted their role in cardiac function. In both human HEK293T kidney epithelial cells and rat cardiomyocytes, the RBP RBM20 has similar binding preferences to intronic regions of mRNA¹³⁷. However, CLIP studies of RBM20 in rat cardiomyocytes but not in HEK293T cells identified many transcripts implicated in cardiac function and disease, such as those encoding RYR2 and TTN. Recent CLIP assays in human stem-cell-derived cardiomyocytes found cardiac-specific targets of RBM20 that undergo alternative splicing¹³⁸, which was not detected in CLIP analysis of rat cardiomyocytes, again highlighting the importance of profiling disease-relevant RBPs in relevant disease models. The study of human stem-cell-derived cardiomyocytes carried out eCLIP assays for both wild-type RBM20 and R636S-mutated RBM20 — a mutation that is observed in patients with dilated cardiomyopathy. Intriguingly, wild-type RBM20 predominantly bound introns in both human and rat cells, whereas the R636S mutation shifted binding of RBM20 towards 3′ UTRs, such that the binding of mutant RBM20 only overlapped with the binding of wild-type RBM20 for 27 target genes (Fig. 3c). This mutation is correlated with the localization of RBM20 in the cytoplasm rather than nucleus, which may explain the shift in binding from introns to 3′ UTR. Together, these studies highlight the influence of cellular, subcellular and disease contexts for RBP–RNA interactions. Not only did profiling of RBP–RNA interactions uncover cardiac-specific binding events, but also CLIP of R636S-mutant RBM20 suggests that the binding spectrum of an RBP can markedly change in disease.

Integration of spatial information

RNA mislocalization is implicated in the pathology of many diseases, including cancer and neurodegenerative diseases such as Alzheimer disease and amyotrophic lateral sclerosis (reviewed in ref. 139). Mutations in RBPs such as TDP43 have been shown to affect RNA localization¹⁴⁰, yet many of these effects remain unknown owing to the lack of high-throughput techniques to examine RBP–RNA interactions in the spatial context. Spatial transcriptomic approaches such as MERFISH, FISHSEQ and STARmap have been developed to map thousands of transcripts in both cultured cells and tissue contexts (reviewed in ref. 141). Most recently, RIBOMap was developed to study the localization of actively translated RNAs¹⁴². RIBOMap was found to identify only those transcripts that were actively translated, whereas STARmap also identified non-coding RNAs and mRNAs localized in granules in which translation is suppressed. Intriguingly, RIBOmap detected differentially expressed genes across the cell cycle, which was not observed in STAR-map. This suggests that although total RNA levels may be similar across the cell cycle, marked changes can occur in the actively translated RNAs.

As spatial approaches improve, we can begin to examine how RBP–RNA interactions change based on location. For example, APEX-based approaches have been applied to identify organelle- and complex-specific transcripts¹⁴³, as well as RBPs in G3BP1⁺ stress granules¹⁴⁴. Although current technologies cannot examine the spatial context of RBP–RNA interactions across the whole transcriptome, there are several interesting biological questions to which spatial technology could be applied. For example, the RBP G3BP1 is known to be involved in stress granules, but also to tether the tuberous sclerosis complex to the lysosome^145,146 and to function as a DNA helicase in the nucleus¹⁴⁷. G3BP1 also transitions from a diffuse cytoplasmic state to discrete puncta in response to cellular stress. As described above, the differential localization of an RBP can alter its binding profiles. For example, in stress granules, which sequester RNAs to prevent translation, G3BP1 preferentially binds the 5′ UTR rather than the coding sequence. It is likely that G3BP1 has a different binding profile when it is tethered with the lysosome. In addition, nuclear G3BP1 may interact with intronic or coding sequence regions. Many RBPs are known to change their subcellular localization in different diseases and contexts; further study of the spatial localization of RBPs will help to resolve and delineate these processes.

Viral protein and RNA interactions

Unbiased host protein–viral RNA interactome studies have discovered hundreds of host RBPs that are associated with viral RNAs^148–155. CLIP studies of the proviral splicing factors HNRNPK, HNRNPM and HNRNPI (also known as PTBP1) have identified interaction sites on the viral genome that are important for the infectivity of Sindbis virus virions¹⁵⁶ (Fig. 4a). CLIP-seq of zinc-finger antiviral protein (ZAP; also known as ZC3HAV1) showed that its binding preference for CG dinucleotides has probably had a role in the evolution of HIV-1 to avoid using this motif in its genome to evade host defence¹⁵⁷. By comparing CLIP reads of cytidine deaminase APOBEC proteins in HIV-1 virions versus infected cells, these host proteins were found to mimic the binding of the viral nucleocapsid protein in virions, indicating that they can mutate viral RNA after packaging into virions¹⁵⁸ (Fig. 4b). Intriguingly, some viruses produce short RNAs that interfere with host miRNAs. Argonaute CLIP (AGO-CLIP) has been used to elucidate the effect of these viruses on host miRNA levels and the redistribution of silenced targets^159–162. CLIP-seq assays also provide insights into how modifications of viral RNA affect the virus life cycle¹⁶³. PAR-CLIP with photo-assisted m⁶A sequencing (PA-m⁶A-seq) uncovered overlapping YTHDF binding and m⁶A signals in the 3′ UTR of HIV-1 that augment HIV-1 replication^164,165. Similarly, PAR-CLIP of NSUN2 combined with PA-m⁵C-seq uncovered NSUN2-specific, m⁵C-modified sites on the HIV-1 genome that are important for translation of the viral protein Gag¹⁶⁶.

Fig. 4 |. Protein–RNA interactions implicated in host–virus interactions.

a, Ultraviolet crosslinking followed by immunoprecipitation (CLIP) read densities for the host proviral splicing factors HNRNPI (also known as PTBP1), HNRNPK and HNRNPM map to three distinct sites on the Sindbis virus (SINV) genome. b, Schematic representation of interactions between host APOBEC3 proteins and the HIV-1 genome in host cells and in HIV-1 virions. c, Rift Valley fever virus N protein binds to different RNA targets (host and viral) at different stages of the infection cycle, with effects on infectivity. d, In host cells, the matrix (MA) domain of HIV-1 Gag binds to host transfer RNAs (shown for tRNA-Lys (anticodon TTT)), which prevents Gag localization to the cell membrane. T-to-C substitutions at sites of interaction with the tRNA are shown. Env RRE, the conserved region in the env gene to which the HIV accessory protein Rev binds; UTR, untranslated region.

Mapping interactions between viral proteins and viral RNA has uncovered both essential^167,168 and redundant^169,170 interactions in virus assembly. Furthermore, CLIP assays at different stages of infection can reveal differential RNA binding patterns. For example, the Rift Valley fever virus N protein predominantly binds to host RNAs early in infection, resulting in nascent virus particles with reduced infectivity. As the viral RNA concentration increases later in infection, N protein binds more efficiently to viral RNA to produce more infectious particles¹⁷¹ (Fig. 4c).

Viral protein–host RNA interactions have been less extensively studied but also have a role in virus–host dynamics. Binding of the amino-terminal matrix (MA) domain of HIV-1 Gag to host tRNA disrupted Gag localization to the cell membrane and virion assembly¹⁷² (Fig. 4d). Mapping interactions between tagged SARS-CoV-2 proteins and host RNAs showed interactions of viral NSP16 with host U1/U2 snRNA, viral NSP1 with host rRNA, and viral NSP8 and NSP9 with host 7SL RNA. These interactions inhibited host RNA splicing, translation and protein trafficking, thereby suppressing the expression of host antiviral genes¹⁷³. A recent preprint reports eCLIP in SARS-CoV-2-infected cells, confirming the interaction of viral NSP8 and RNA-dependent RNA polymerase (RdRp) with host 7SL RNA, as well as the interaction of viral RdRp with numerous host mRNAs, many of which were differentially regulated during infection¹⁷⁴. Although the biological relevance of these observations is unclear, these viral protein–host RNA interactions could shape the host transcriptome, thereby contributing to the complex interactions between a virus and its host cell.

Decoding potential therapeutic mechanisms

Despite the challenges posed by the complex structures of RBPs, including multiple disordered protein–RNA interfaces^175–177, the targeting of RBPs by small molecules and natural products has been demonstrated¹⁷⁸. RBPs such as the Musashi (MSI) family, ELAVL1, LIN28A, LIN28B, SF3B1 and eIF4A have been targeted by small-molecule inhibitors (Fig. 5a). Notably, the FDA-approved small molecules risdiplam and branaplam, which modulate RNA splicing to treat spinal muscular atrophy, mediate their target RNA specificity through stabilization of the U1 splicing machinery^179,180.

Fig. 5 |. Protein–RNA interaction maps as drug discovery assays.

a, Schematic showing different RNA-binding protein (RBP) targets of small-molecule inhibitors (in parentheses) for modifying protein–RNA interactions involved in the regulation of mRNAs. b, Schematic showing changes in protein–RNA interaction profiles upon small-molecule or drug treatment. Presence of a small-molecule ligand for the RBP NONO leads to increased binding of NONO at mRNA 5′ ends, which correlates with increased intron retention and degradation of the mRNA (left). A small-molecule ligand locks SF3B1 in an open state and increases the affinity for DDX42 at splice sites, which triggers marked changes in alternative splicing patterns leading to increased cell death (right). All metaplots are idealized observations of ultraviolet crosslinking followed by immunoprecipitation (CLIP) read densities. UTR, untranslated region.

A recent ‘function-first’ approach used size-exclusion chromatography to detect electrophilic compounds that affect protein complex formation¹⁸¹. The study identified tryptoline acrylamide as disrupting spliceosome function, with anti-proliferative effects. eCLIP data, coupled with immunoprecipitation–mass spectrometry, have highlighted how an RNA ligand locks SF3B1 in an open state and increases the affinity for DDX42 at splice sites, which triggers marked changes in alternative splicing patterns. Covalent ligands such as those targeting NONO (also known as NRB54) can also affect RBP–RNA interactions¹⁸², as exemplified by eCLIP validation of covalent ligand-mediated antitumour effects. The presence of the covalent ligand led to increased binding of NONO at mRNA 5′ ends, which correlated with degradation of the mRNA (Fig. 5b).

Further research is warranted to discern the RNA targets that are affected by modulation of RBPs. Protein–RNA interactome analysis can be used to elucidate altered binding sites and affinities, which offers insights into the precise subset of RNA targets influenced by RBP modulation and promises to expand our knowledge of potential therapeutic mechanisms across diverse contexts.

Limitations of CLIP-based methods

Although CLIP is an accepted standard methodology for identifying the RNA targets of RBPs, it is not without limitations. Notably, several types of ‘peak calling’ software used to identify regions of protein binding, such as CLIPper and Piranha, have high rates of false negatives¹³. The sensitivity of CLIP depends on the choice of peak caller and associated thresholds, which can introduce variability in the results, a challenge that is faced by most genome-wide methods. Conversely, false positives in CLIP arise from the incorporation of small, highly expressed RNAs, such as snoRNAs and small Cajal body-specific RNAs, rRNAs and intronic reads. The incidence of false positives increases as the signal from real RBP targets decreases, attributed either to the RBP having a small number of targets or to the inefficiencies of the antibody used to immunoprecipitate the RBPs (H.-L. Her & G.W.Y., unpublished data).

Studying protein–RNA interactions in relevant models is crucial but the large amount of sample material and the many purification steps required for each CLIP assay can be prohibitive. To address the limitations of scalability, a recent development involving barcoding of antibodies with oligonucleotide conjugation enabled the multiplexing of several proteins in a pooled sample to streamline sample preparation¹⁸³. Another limitation of CLIP is the lack of information on sensitivity for RNA isoforms, as well as the inability to study protein–RNA interactions at single-cell resolution. Non-CLIP methods, such as fusing RBPs with RNA base editors, have enabled sequence-based identification of mutations caused by an RBP–base editor, allowing for single-cell resolution of protein–RNA interactions without the need for immunoprecipitation^184–188 (Table 1). Furthermore, these methods eliminate the need for RNase digestion to generate a ‘footprint’ protected by the RNA-bound protein, thus enabling long read sequencing and isoform-specific interrogation of protein–RNA interactions. However, there is still a need for methods requiring less starting material and with improved sensitivity to study protein–RNA interactions in complex models, particularly those not amenable to genetic manipulation.

Notably, CLIP binding data alone cannot discern the functional binding targets of an RBP. An RBP may engage in many binding interactions but not all of these have a discernible functional outcome^189,190. To address this limitation, CLIP assays should be combined with complementary techniques such as ribosome profiling, SLAM-seq (for RNA half-life determination) or differential splicing analysis during knockdown and/or overexpression of the RBP. Knockdown or knockout of the RBP helps to identify the RNA targets that are affected by the loss of RBP binding. An integrative approach will be essential for refining the identification of regulatory targets influenced by RBP binding.

Future directions

Large-scale protein–RNA interaction profiling studies in reference cell lines have enhanced our understanding of how RBPs recognize substrate RNAs and their involvement in the life cycle of an mRNA (Box 1). Protein interaction profiles with non-coding RNAs have provided insights into their biogenesis and the repression of repeat elements and have expanded the set of known tRNA genes. Furthermore, profiling of the RNA modifications associated with ‘reader’, ‘writer’ or ‘eraser’ proteins has provided valuable insights into the regulatory consequences of RNA modifications.

Protein–RNA interactions have been shown to have crucial roles in various contexts, including disease states, viral replication, host responses to infection and the response to small-molecule drugs. As methodologies for protein–RNA interaction profiling develop further and noncanonical RBPs are discovered, we anticipate continued insights into previously uncharacterized interactions between RBPs and their substrate RNAs.

Despite recent advances, many questions remain regarding the effects of RNA sequence, structure, localization, modification and species on protein–RNA interactions, as well as the effects of cell type and disease context. No generalizable method for the spatial analysis of protein–RNA interactions currently exists, but if and when available, will no doubt provide important insights into the effects of subcellular localization, cell type and tissue structure on RBP–RNA interactions. As demonstrated by psiCLIP and temporal iCLIP, CLIP methodologies that ‘freeze’ protein–RNA interactions by halting key processing steps of splicing and RNA transcription, respectively, enable greater time-resolution of these processes. Future methods that extend to protein–RNA interactions beyond splicing and RNA transcription will lead to a greater understanding of the temporal dimension of protein–RNA interactions. Combining this with future advances in spatial sequencing could reveal insights into the complex spatiotemporal control of post-transcriptional gene regulation.

Improved methods to inhibit RBP binding and quickly measure the effects on RNA targets should prove crucial to understanding RBP function. Using antisense oligonucleotides or small molecules to obstruct RBP–RNA interactions, coupled with techniques that directly measure differential splicing, RNA stability or translation rate, could offer a refined approach to measuring RBP dynamics¹⁷⁸. Implementing these approaches in a high-throughput manner would offer unprecedented insights into RBP functions and network interactions.

In conclusion, our current knowledge of protein–RNA interactions provides a strong foundation for further research. Understanding these interactions is essential for drug discovery and therapeutic development. Advances in the sensitivity and scalability of methods will pave the way for more comprehensive investigations of protein–RNA interactions, potentially leading to novel therapeutic targets and more effective treatments for a wide range of diseases.

Supplementary Material

The online version contains supplementary material available at https://doi.org/10.1038/s41576-024-00749-3.