The regulatory landscape of interacting RNA and protein pools in cellular homeostasis and cancer

Abstract

Biological systems encompass intricate networks governed by RNA-protein interactions that play pivotal roles in cellular functions. RNA and proteins constituting 1.1% and 18% of the mammalian cell weight, respectively, orchestrate vital processes from genome organization to translation. To date, disentangling the functional fraction of the human genome has presented a major challenge, particularly for noncoding regions, yet recent discoveries have started to unveil a host of regulatory functions for noncoding RNAs (ncRNAs). While ncRNAs exist at different sizes, structures, degrees of evolutionary conservation and abundances within the cell, they partake in diverse roles either alone or in combination. However, certain ncRNA subtypes, including those that have been described or remain to be discovered, are poorly characterized given their heterogeneous nature. RNA activity is in most cases coordinated through interactions with RNA-binding proteins (RBPs). Extensive efforts are being made to accurately reconstruct RNA-RBP regulatory networks, which have provided unprecedented insight into cellular physiology and human disease. In this review, we provide a comprehensive view of RNAs and RBPs, focusing on how their interactions generate functional signals in living cells, particularly in the context of post-transcriptional regulatory processes and cancer.

Background

Biological systems are entities of high complexity, with genomically encoded proteins and RNAs serving as key building blocks to give rise to functional cues. Both RNA and protein pools are major contributors to the assortment of biomolecules constituting the cell (1.1% and 18% of the total weight of mammalian cells, respectively) [1], and act in concert to govern virtually all cellular processes. Albeit different, RNAs and proteins assume essential structural, catalytic, and regulatory roles within the cell, modulating central processes all the way from genome organization and transcription to post-transcriptional regulation, subcellular localization, and translation [2, 3]. These events are often mediated by cooperative interactions between RNAs and RNA-binding proteins (RBPs), which in turn enable the assembly of effector ribonucleoprotein (RNP) complexes. Depending on the expression and location of RNAs and RBPs, RNPs can have shared or tissue-specific functions. Most importantly, the integrity of these intricate RNA-RBP networks is of utmost importance for maintaining cellular homeostasis, where perturbations affecting individual components can have profound effects on human pathophysiology, including the development of neurodegenerative disorders and cancer [4, 5].

For over a decade, the fraction of the nearly 3.1 billion base pairs in the human genome deemed functional has remained controversial. Notably, the concept of biological function is often understood as the capacity to contribute to cellular or organismal fitness. However, the attribution of function does not necessarily imply a reproductive advantage in all cases or may confer this advantage only under certain conditions. Estimates of the functional fraction of the human genome have ranged from as little as the protein-coding regions (~ 1–2%) to the full repertoire of genomic DNA covered with functional features determined by the Encyclopedia of DNA Elements (ENCODE) project (80%) [6,7,8,9]. Moreover, evolutionary-based efforts have attempted to estimate this fraction using sequence divergence or conservation information (3–15%) [10,11,12,13]. However, mapping the complete effective component of the human genome has proven challenging, especially in regions that are not actively transcribed or translated into functional protein products. Many of these genomic regions exhibit activity without clear patterns of evolutionary constraint at the sequence level. This has led to the investigation of additional dimensions to uncover conserved relationships of noncoding elements that can be indicative of function, including structure, spatiotemporal expression, and regulatory conservation (i.e., the maintenance of interactions and regulatory circuitry between distantly related species) [14]. In particular, noncoding RNAs (ncRNAs) are a diverse class of RNA molecules that originate from both genes (i.e., sense, antisense, intronic or gene overlapping) and intergenic regions of the genome and harbor unique potential to regulate central biological processes. The variable lengths of ncRNAs are accompanied by differences in sequence conservation. Shorter ncRNA genes tend to be more conserved, similar to protein-coding genes, whereas longer ncRNAs often exhibit evidence of rapid sequence evolution [15,16,17]. Nonetheless, there have been ever-growing reports of functional ncRNAs independent of their length and sequence conservation status [18,19,20,21,22,23,24,25,26]. A prominent feature is their ability to form complex RNA structures that facilitate interactions with RBPs. These structural elements are particularly predictable and measurable for small ncRNAs but have proven difficult to identify in long ncRNAs in which interspersed covarying sites are separated by longer stretches of nucleotide sequences. Moreover, it is even harder to define the cases where these structural elements are relevant in a biological context [27].

At the RNA-RBP interface, the protein’s composition and its ability to recognize RNA sequence and structural motifs are key aspects. Great progress has been made over the last decades in the RBP field, from initial discoveries revealing the existence of conserved RNA-binding domains (RBDs) to more recent observations that RBPs interact with RNA molecules via intrinsically disordered regions [28, 29]. In addition, the advent of high-throughput technologies has not only enabled the deep characterization of both protein and RNA pools within the cell but also allowed to capture RNA-RBP interactions at an unprecedented scale. Currently, the estimated number of RBPs extends well beyond those that contain canonical RBDs (up to 5,000 RBPs in humans) [30,31,32,33]. However, the corresponding RNA interactomes have been described for only a fraction of these proteins, and even fewer have well-defined functional roles. The ENCODE project has produced the most comprehensive resource to date, establishing RNA-RBP dependencies for a total of 150 RBPs in human cell lines [3]. Their approach combined experimentally derived interaction maps for RBPs with in vitro detection of binding specificities and functional assays, which highlighted the importance of using integrative approaches for elucidating functional RNA-RBP interactions. Besides ENCODE, additional databases have become valuable resources for exploring and analyzing RNA-RBP interactions. For instance, POSTAR3 [34] collects RBP annotations and interactions together with additional information layers from public resources to facilitate the exploration of post-transcriptional regulation mechanisms. Similarly, EuRBPDB [35] provides a multidimensional view of a comprehensive set of eukaryotic RBPs by combining data from various experimental observations. Databases such as starBase/ENCORI [36] or RNAInter [37] are specialized for RBP interactions with different RNA types, while others, including RBR-ID [38] and RBPDB [39], focus on RNA-binding regions within RBPs. Other repositories collecting RNA-RBP interaction data include RBPbase [5] and RBP2GO [40]. Alongside these efforts toward the systematic profiling of RNA-RBP associations, novel computational methods are being developed to facilitate the integration of orthogonal data types [24,25,26, 34, 41, 42]. Notably, reconstructing RNA-RBP regulatory networks is an essential step in the quest of understanding cellular homeostasis and treating disease.

In this review, we broadly present the functional landscape of interacting RNA and RBP pools in eukaryotic cells, focusing on the diversity of RNA species, their biogenesis, and central regulatory principles dictating RNA life. For an RBP-centric perspective, we refer readers to the following reviews [5, 28, 43, 44]. We highlight the emerging roles of ncRNAs as drivers of cellular phenotypes, emphasizing their implications in modulating protein synthesis in healthy and cancer cells. Lastly, we exemplify methodological advances empowering the study of RNA-RBP regulatory networks and their clinical relevance.

RNA diversity and riboregulatory roles

High-throughput technologies have revolutionized the world of RNA biology, changing the view of RNA molecules from just carriers of genetic information to a diverse functional class orchestrating a wide array of processes in living cells [45, 46]. Current understanding separates messenger RNAs (mRNAs), which have protein-coding potential, from ncRNA species that exert a variety of roles independent of the ability to generate a functional protein product (Fig. 1). Several attempts have been made at categorizing ncRNA types, but the most evident grouping distinguishes only between long ncRNAs, for those greater than 200 nucleotides in length, and small ncRNA, for those below this arbitrary cutoff [47]. However, there are still certain ncRNA species that are not clearly classified by this criterion, with examples both above and below the 200-nucleotide threshold. Alternatively, ncRNAs can be grouped by function into constitutive and regulatory types [48] but also stratified based on their biogenesis pathways, including differences in transcription caused by dedicated DNA-dependent RNA polymerases (Pols) [49, 50].

Fig. 1

Diversity of the eukaryotic transcriptome. RNA species are separated into coding (red box) and noncoding (green box) classes. The approximate length of each individual RNA type is indicated.

During and after transcription, binding events between RNAs and RBPs are fundamental for regulating most cellular processes (Fig. 2), shaping the fate of not only the individual molecules but also that of assembled RNP complexes and the cell itself. Nascent RNAs undergo a series of co-transcriptional processing events. These steps involve the RBP-mediated removal of sequences from the nascent RNA molecule, which is often referred to as primary or precursor RNA (pre-RNA). Additionally, RBPs can chemically modify nucleotides within RNA molecules. Further RBP-controlled RNA processing occurs within the nucleus or in the cytoplasm after export through nuclear pore complexes (NPCs), resulting in mature transcripts. These and subsequent steps are collectively referred to as post-transcriptional regulation of gene expression. At this stage, RBPs continue to interact with RNAs within the cellular environment, defining core processes such as translation and degradation, and ultimately dictating turnover rates and cell functioning (Fig. 2). Moreover, newly synthesized proteins and RNAs can shuttle back and forth between the nucleus and cytoplasm to exert their different roles. For instance, RBPs can translocate to the nucleus to guide chromatin remodeling through RNA-binding and change gene expression programs. An example of this involves the ncRNA HOTTIP, which interacts with the WDR5/MLL remodeling complex to promote histone 3 lysine 4 trimethylation (H3K4me3) of genes at the HOXA locus, enhancing their expression [51]. However, other examples are highly discussed, such as the case of X chromosome inactivation in mammals, where the ncRNA X-inactive specific transcript (XIST) recruits, among other factors, the Polycomb repressive complex 2 (PRC2) to establish a H3K27me3-repressive chromatin state that silences one copy of the X chromosome in females [52,53,54,55].

Fig. 2

Overview of transcriptional and post-transcriptional processes influenced by RNA-RBP interactions. In the nucleus (blue sphere), nascent RNAs are actively processed and modified before being exported to the cytoplasm (white), where additional processing follows. In the cytoplasm, messenger RNAs (red lines) are translated into proteins (purple shapes). Noncoding RNAs (green lines) interface with the protein pool to shape post-transcriptional regulatory processes in the cytoplasm or translocate to the nucleus as ribonucleoprotein complexes to remodel gene expression programs. Decay of molecules occurs simultaneously, with degradation determining turnover rates in the cell.

There is a great diversity of RNA species, each of which participates in different cell functions and has specific RBP interaction partners. In the following sections, we briefly introduce the major RNA types together with their roles and discuss the core processes for cellular homeostasis that are driven by RNA-RBP specificities.

Transcriptome heterogeneity

The cell harbors an ample array of RNA molecules that come in different shapes and sizes (Fig. 1). These differences are also reflected in terms of their biogenesis pathways, abundance, and roles within the cell [45, 48, 50, 56].

Messenger RNA (mRNA)

Among all RNA types, mRNAs (~ 8,900 molecules per million) carry the information needed for protein production. Transcribed by eukaryotic Pol II, mRNA molecules are subjected to multiple processing steps including capping of 5’ ends, splicing, cleavage, and polyadenylation at the 3’ end [57]. Adequate processing of these RNAs into mature mRNAs is essential for translation in eukaryotes, as this enables the recognition by translation initiation factors, ribosome assembly, and protein synthesis [58]. The size of mRNAs varies greatly from as little as 81 to approximately 110,000 nucleotides for the longest protein annotated in humans, which is encoded by the TTN gene [59].

Ribosomal RNA (rRNA)

The ribosome is a remarkable RNP complex that translates mRNAs into proteins. It is largely composed of highly abundant rRNAs (~ 89,000 molecules per million), which form the large (60 S: 28 S, 5 S and 5.8 S rRNAs) and small (40 S: 18 S rRNA) subunits together with ribosomal proteins [60]. The transcription of rRNA genes is driven by Pol I (45 S rRNA) and Pol III (5 S rRNA) and localizes to the nucleolus and its vicinity, a highly active region in the nucleus that acts as a hub for the cleavage and modification of ribosomal components [61,62,63]. For instance, the 45 S pre-rRNA is further processed into mature 28 S, 18 S and 5.8 S rRNAs for ribosome assembly [64]. Their mature length can range from around one hundred to several thousand nucleotides.

Transfer RNA (tRNA)

Like rRNAs, tRNAs (~ 890,000 molecules per million) are highly abundant, modified and structured ncRNA molecules that enable protein synthesis. They function by complementarily binding nucleotide triplets (codons) at the ribosome-mRNA interface and adding their charged amino acids to the growing polypeptide chain [65]. There are a total of 429 highly confident tRNA genes annotated in the human genome, which are transcribed by Pol III and can be classified into 47 isoacceptor groups according to their anticodon sequence [66,67,68]. After processing, their length is approximately 70 to 90 nucleotides. Notably, mature tRNAs can be further processed by RBPs into tRNA-derived halves or tRNA-derived fragments (tRFs), which exert roles beyond mRNA translation, including the impairment of reverse transcription and silencing of endogenous retroviruses, or the displacement of stabilizing RBPs [69,70,71].

Small nuclear RNA (snRNA)

A characteristic of eukaryotic organisms is the presence of introns in pre-RNAs. These are accurately removed co- or post-transcriptionally by the spliceosome, a nuclear-localized RNP complex containing snRNAs (~ 4,100 molecules per million). Except for selected cases like the nuclear-processed U6 snRNA that is Pol III-transcribed [67], snRNAs are transcribed by Pol II and partake in the major (U1, U2, U4, U5 and U6 snRNAs) and minor (U5, U11, U12, U4atac and U6atac snRNAs) spliceosomal complexes that target divergent sets of intron sequences [72, 73]. Briefly, the processing of snRNAs involves 5’-end capping and 3’-end cleavage, followed by export to the cytoplasm for snRNP assembly and translocation back to the nucleus [72]. Moreover, snRNPs concentrate in nuclear substructures termed Cajal bodies, which are considered hubs for snRNP processing and remodeling. Their length varies in the range of 100 to 300 nucleotides.

Small nucleolar RNA (snoRNA)

The established role of snoRNAs (~ 3,400 molecules per million) is to guide RNA editing proteins to their target sites within other RNA sequences, for example rRNAs, snRNAs, and tRNAs. In addition, other roles have been defined for this class including the modulation of alternative splicing events or 3’ end processing of RNAs influencing transcript stability and abundance [74]. There are over 2,000 snoRNAs annotated in humans [75] that are transcribed by Pol II and separated into two groups according to their conserved sequence elements and interaction partners (i.e., C/D- and H/ACA-box snoRNAs). Box C/D snoRNPs classically mediate 2’-O-methylation, while box H/ACA snoRNPs mediate pseudouridylation of targets [74, 76]. In addition, certain snoRNAs from both C/D- and H/ACA-box types, which contain distinct features, such as CAB boxes (UGAG) and GU repeats, localize to Cajal bodies, where they facilitate snRNA processing. These are hence termed small Cajal body-specific RNAs. After transcription, snoRNAs are primarily released from intronic sequences of host genes in humans and are subsequently processed by cleavage at the 5’ and 3’ ends. This process results in a mature length of approximately 60 to 300 nucleotides.

MicroRNA (miRNA), PIWI-interacting RNA (piRNA) and small interfering endogenous RNA (endo-siRNA)

Silencing of gene expression mediated by miRNAs (~ 2,700 molecules per million) has been studied exhaustively since the original discovery of lin-4 in C. elegans [77]. Like the other small ncRNA types described above, miRNAs act by guiding proteins to target RNA sequences through antisense complementarity. In this case, Argonaute (AGO)-bound miRNAs commonly enable the recognition of target sites toward the 3’ end of transcripts, leading to RNA degradation or translational repression [78]. Newly Pol II-transcribed miRNAs, termed pri-miRNAs, are processed by RNase III-type endonucleases of the Microprocessor complex (Drosha and DGCR8) into individual stem-loops, which are hairpin-like RNA structures that are essential intermediates in miRNA biogenesis and are also referred to as pre-miRNAs. Pre-miRNAs are then exported from the nucleus by exportin 5, followed by stem-loop cleavage by Dicer and AGO coupling as mature single-stranded miRNAs of 21 nucleotides in length [79]. Complementing miRNAs, small interfering RNAs that are produced endogenously (i.e., endo-siRNAs), such as those found in mouse stem cells, can be loaded onto AGO proteins to serve as guides for gene silencing. These endo-siRNAs are approximately 21 nucleotides in length [79, 80]. Moreover, piRNAs, which are primarily expressed in germline cells, interact with PIWI-AGO proteins to repress transposable elements, regulate gene expression, and fight viral infection [81]. These small ncRNA molecules are between 21 and 35 nucleotides in length.

Long noncoding RNA (lncRNA), circular RNA (circRNA) and enhancer RNA (eRNA)

Large-scale profiling of the human transcriptome has resulted in the identification of a substantial catalog of lncRNAs (~ 2,100 molecules per million), with recent annotations of the human genome setting their number on par with that of protein-coding genes at over 19,000 [82]. However, due to factors such as their low abundance and sequence conservation as well as restricted expression in specific cell types and under certain conditions, the detection and functional characterization of lncRNAs has remained a challenge [17]. Among all ncRNA types, lncRNAs exhibit the highest functional versatility, covering roles ranging from chromatin remodeling to transcriptional and post-transcriptional regulation [83, 84]. Similar to mRNAs, the transcription and processing of lncRNAs occurs via Pol II, including 5’-end capping, splicing, and 3’-end polyadenylation. A distinctive feature of mRNAs and lncRNAs is that the splicing process appears more inefficient in the latter, which affects their alternative splicing landscape [83, 85]. Moreover, there are other ncRNA molecules that are generally grouped with lncRNAs. For instance, circRNAs, which are covalently closed RNA molecules generated by back-splicing that are enriched in brain tissue, have been shown to modulate cellular processes by acting as miRNA sponges (i.e., molecules that sequester and inhibit miRNAs by binding to them, thereby preventing miRNAs from interacting with their target mRNAs), transporters, or scaffolds for RBPs [86, 87]. In addition, eRNAs, which are transcribed from active enhancer regions in the genome, have been shown to sustain enhancer activation and the expression of target genes [88, 89]. lncRNAs are usually defined as RNA molecules greater than 200 nucleotides in length although certain circRNAs and eRNAs may be shorter than this cutoff.

RNA-RBP specificities guiding cellular processes

Many functional outcomes in the cell are governed by the specific recognition and coupling between RNA molecules and RBPs. These interactions rely on distinct sequence motifs, secondary structures, and post-transcriptional modifications at the RNA level that collectively orchestrate various cellular processes (Fig. 3). On the protein side, RBPs that utilize established RBDs to interact with RNAs are termed canonical RBPs. There is a wide variety of RBDs, including the more abundant RNA recognition motifs and K homology domains, but also less prevalent ones, such as those found in ribosomal proteins [31]. RBDs often repeat or co-occur with other RBDs, and the presence of multiple RBDs within a single RBP can greatly enhance the versatility and robustness of RNA interactions. In turn, RBPs lacking such domains are referred to as noncanonical, unconventional, or moonlighting RBPs given their alternative classical roles independent of RNA-binding [28]. The binding of noncanonical RBPs to RNA is facilitated via intrinsically disordered regions. Moreover, post-translational modifications such as phosphorylation or acetylation play crucial roles in modulating RNA-binding affinity and specificity. These modifications can alter the conformation of RBPs, influence their interaction with other proteins and cofactors, and regulate their localization and stability, thereby fine-tuning the cellular functions they govern. For instance, allosteric regulation via interdomain communication of tandem RNA recognition motifs has been shown to influence the ability of hnRNPA1 to interact with protein partners and bind RNA [90].

Coinciding with the switch from transcription initiation to elongation of Pol II-transcribed mRNAs and lncRNAs, the carboxy-terminal domain of Pol II is phosphorylated and recruits capping factors [91]. Subsequent enzymatic triphosphatase, guanylyltransferase, and methyltransferase activities result in the formation of a 7-methylguanosine cap (m⁷G-cap) structure at the 5’ end of the nascent RNA molecule, which is further bound by the cap-binding complex (CBC), mediating, among other processes, export from the nucleus [92]. Notably, the Pol III-transcribed 7SK snRNA is known to modulate the kinase activity needed for transcription elongation by sequestering the positive transcription elongation factor b (P-TEFb) [93]. Here, a combination of structural elements and sequence motifs enables the assembly of the 7SK RNP complex, which contains the HEXIM, LARP7 and MePCE proteins, and controls the repression of P-TEFb activity [94].

Fig. 3

Determinants of RNA-RBP specificities and their functional implications. The RNA-RBP interface (central sphere) is facilitated by sequence elements, structural arrangements, and chemical modifications of RNAs (green lines) that enable recognition by selected proteins (purple shapes). RNA-RBP interactions intervene in multiple cellular processes. Anticlockwise: capping to add a m⁷G-cap protecting the 5’ end of RNAs, splicing to excise sequences from RNA molecules, polyadenylation to extend the 3’ end of RNAs with adenine stretches, editing to introduce chemical modifications (orange lollipops) in RNAs, decay to degrade RNA molecules, cleavage to cut RNAs, transport to localize RNAs and other molecules at different cellular locations, and translation to decode messenger RNA to produce proteins.

During the elongation phase of transcription by Pol II, intron sequences within the growing pre-RNA are removed by splicing. The spliceosome is built through the coordinated assembly of U-rich snRNP complexes that enable the recognition of key sequence elements defining intron boundaries. These are the GU 5’ (donor) and AG 3’ (acceptor) splice sites and the branchpoint sequence neighboring the splice acceptor (~ 18–40 nucleotides upstream) [95]. Successful splicing is accomplished after two subsequent transesterification reactions that occur co-transcriptionally as the transcript emerges from Pol II [95, 96]. Moreover, splicing regulatory elements (SREs) can influence the splicing process by recruiting trans-acting RNPs [97, 98]. Simultaneously during transcription, RNA editing factors can introduce (writers) or remove (erasers) chemical modifications at individual nucleotides harboring great potential for regulation [99, 100]. For instance, N⁶-methyladenosine (m⁶A) modifications can be recognized by the canonical RBP hnRNPG (reader), which has been shown to promote alternative splicing when they occur close to splice sites [101].

Such as with splicing, there are further processing steps of the pre-RNA that take place before transcription termination. This includes initiating polyadenylation (poly(A)) at the 3’ end by a poly(A) polymerase (PAP). The choice of poly(A) site is mediated by a core set of factors (CPSF, CSTF, CFI and CFII) that recognize the AAUAAA poly(A) signal together with auxiliary sequence elements and cleave the RNA molecule in the presence of PAP, resulting in the addition of a 3’ poly(A) tail of approximately 250 nucleotides [102,103,104]. Poly(A)-binding proteins (PABPs) that are bound during PAP synthesis have been shown to control poly(A) tail length [105]. Moreover, the differential usage of regulatory elements leads to alternative cleavage and poly(A) events that change 3’ untranslated regions (UTRs), influencing future steps of RNA life and adding to the diversity of the transcriptome [102, 104].

After these co-transcriptional processes, which are primarily but not exclusively linked to protein-coding RNAs, transcripts are exposed to additional post-transcriptional regulatory cues, including those aforementioned, that affect their stability, subcellular localization, or translation. Mature mRNAs are composed of both protein-coding and noncoding exonic regions that are retained after splicing. In fact, only a fraction of exonic sequences are annotated as protein-coding in eukaryotic genomes (~ 23% in humans), although noncoding regions of protein-coding RNAs have important regulatory roles mediated via interactions with RBPs and are often altered in disease [106]. For instance, RNA sequences and structures at the 5’ UTRs of mRNAs modulate ribosome recruitment controlling translation efficiency, and the use of alternative 5’ UTR isoforms has been shown to correlate with enhanced protein synthesis rates in diseases such as squamous cell carcinoma [107, 108]. Moreover, RNA elements at 3’ UTRs and poly(A)-tail length have been shown to influence mRNA stability, translation efficiency, and protein localization at different developmental stages [109, 110]. In the case of ncRNAs, they are assembled from noncoding exons (i.e., intergenic ncRNA genes) but can also arise from exons and introns of protein-coding genes through alternative transcription or pre-RNA processing. For example, the lncRNA FAST partially overlaps and is transcribed in the antisense direction of the FOXD3 gene, and its expression has been shown to maintain pluripotency in human embryonic stem cells through the activation of WNT signaling [111]. In addition, the lncRNA-PNUTS is an alternatively spliced isoform generated from the protein-coding gene PNUTS that has been associated with breast cancer progression [112]. Importantly, the generalizability of regulatory cues exerted by RNA-RBP interactions is still debated since these are often context-dependent. Some level of contradiction in the literature can be explained by contrasting cellular states at different developmental stages (i.e., differentiating and differentiated cells) or in diseases such as cancer (i.e., dedifferentiating cells), where there are differences in the subcellular localization of RNA and RBP partners or changes in their binding affinities.

For export from the nucleus, mature transcripts interact with adaptor molecules and transport factors that facilitate their transit through the NPC. For example, transcription-export complexes (TREXs) bind mRNAs co-transcriptionally and are further recognized by the NXF1-p15 heterodimer, which mediates the export of mature molecules across the NPC [113]. Nonetheless, a detailed picture of the underlying mechanisms and RBP complements enabling the selective export of mature mRNAs and the array of other processed RNA species is still lacking.

In the cytoplasm, the CBC-bound m⁷G-cap structure of mRNAs binds to eukaryotic translation initiation factor 4E (eIF4E), which promotes protein production by recruiting additional translation initiation factors and circularizing RNA through interactions with PABPs anchored at the poly(A) tail [92, 114]. Among the different factors influencing translation, the availability of correctly processed, modified, and structured core molecules such as rRNAs and tRNAs is essential. For instance, codon bias and tRNA supply have been shown to ensure translation efficiency within the cell [115,116,117].

Furthermore, RNA decay pathways compromise the stability of mRNA molecules, leading to degradation and limiting the rate of translation. The processes contributing to degradation include deadenylation-dependent, deadenylation-independent, and endonuclease-mediated mRNA decay pathways [118]. Aside from their role in translational repression, miRNAs or siRNAs targeting the 3’ UTR of mRNAs can unfold mRNA decay by recruiting deadenylation (CCR4-NOT and PAN2-PAN3) as well as decapping (DCP1 and DCP2) factors, which enables degradation by the exosome complex (3’ to 5’) and exoribonuclease XRN1 (5’ to 3’), respectively [119, 120]. Moreover, both miRNAs and siRNAs can promote decay via cleavage at the target site when the sequence complementarity is high [121]. Surveillance mechanisms also occur in both the nucleus and the cytoplasm to ensure the integrity of transcripts [118]. For instance, nonsense-mediated decay recognizes and degrades mRNAs with premature termination codons that could result in truncated proteins. In addition, alternative translation-dependent mRNA decay mechanisms include the degradation of transcripts lacking a stop codon (non-stop decay) or with stalled ribosomes (no-go decay). Aberrant transcription or processing of mRNAs in the nucleus can lead to rapid degradation through homologous exosome and exoribonuclease activities to that of the cytoplasm.

Lastly, RNA transport is another key process facilitating the localization and formation of dense RNP complexes in different cellular compartments. Examples include canonical RBPs containing a YTH domain that selectively bind m⁶A-modified transcripts to localize at processing bodies (P-bodies) together with proteins associated with RNA decay [122] and zipcode-binding proteins (ZBPs), such as ZBP1, which localizes Actb mRNAs to the leading edge of migrating fibroblasts for translation [123]. Besides P-bodies, which function in mRNA degradation, storage and translational repression, there are other interesting subcellular condensates of phase-separated RNAs and proteins such as germ granules, neuronal granules, and stress granules [124, 125]. Germ granules are unique RNA-protein aggregates present in germline cells that contain piRNAs and AGO family proteins. They play a crucial role in modulating transcript stability and translation during gametogenesis and ensuring genome integrity through the silencing of transposable elements. Neuronal granules are composed of RBPs (e.g., FMRP, STAU2, TDP-43) that facilitate long-distance RNA transport in neurons and local translation at synapses. Stress granules are temporary assemblies of mRNAs and associated proteins, including G3BP1/2, TIA-1 and TIAR, that stall translation in response to cellular stress, helping to regulate gene expression under adverse conditions.

Overall, understanding the specificity and complexity of RNA-RBP interactions at the molecular scale provides unique insight into the elaborate regulatory networks that sustain cellular homeostasis. In the next section, we cover examples of RNA-RBP interactions influencing cellular phenotypes with a primary focus on the divergent roles of different ncRNA species in cancer.

Mechanisms in cellular homeostasis and cancer

Given the central role of RNA-RBP interactions in cell function, it is not surprising that there is mounting evidence that perturbations in RNA-RBP regulatory networks lead to disrupted cellular physiology and human diseases, such as cancer. Aberrant gene expression and transcript formation are common in cancerous cells and disrupt self-renewal, proliferation, and differentiation processes, which are essential for maintaining cell homeostasis. Efforts to profile the expression and mutation rate of RBPs across human cancers have revealed substantial alterations that are potentially implicated in their cellular phenotypes [126, 127]. In fact, RBP signatures have been proposed as prognostic markers for hepatocellular carcinoma (HCC), underscoring the relevance of these genes [128]. Developmental studies have also highlighted the importance of RBPs during embryogenesis, whereby resuming RBP activity may result in oncogenic effects that enhance cancer traits [100, 129, 130]. Notably, these alterations are tightly associated with changes in RNA modifications, structure, and abundance, which in turn contribute to influencing the cellular processes driving the cancer phenotype.

The equilibrium of core components of the translation machinery is a fundamental aspect that defines cellular states. Consequently, disturbances in the rRNA pool and tRNA supply in the cell can perturb translation dynamics and promote cancer (Fig. 4A). Elevated proliferation in tumorigenesis requires efficient protein synthesis; therefore, the translation process is frequently dysregulated in cancer cells [131]. Several studies investigating the coupling of mRNA codons and tRNA anticodons in homeostasis have found that the translation efficiency landscape is stable overall across most mammalian cell types and during development [16, 132, 133]. The balance of codons and anticodons is further known to mediate translation elongation rates and mRNA stability [134]. However, changes to selected tRNAs have been shown to increase proliferative phenotypes and metastasis, such as the overexpression of tRNA^Glu (UUC) and tRNA^Arg (CCG) in breast cancer [135, 136]. In addition, increased rRNA levels have been associated with prostate and cervical cancer but are not always linked to promoter hypomethylation [137, 138]. Both rRNAs and tRNAs are extensively modified to fulfill their canonical roles, and changes in these modification patterns can also lead to the development of cancer. For instance, defective rRNA pseudouridylation (Ψ) mediated by snoRNA-guided Dyskerin (DCK1) alters rRNA and ribosomal structures, thus limiting the fidelity of translation and increasing cancer susceptibility [139]. Similarly, defects in rRNA 2’-O-methylation (Nm) by snoRNA-guided Fibrillarin (FBL) can compromise translation influencing cancer progression [140]. Indeed, recent work suggests that higher levels of SNORD97/133 may facilitate methionine-rich proliferation-related gene expression programs by increasing 2-O-methylation of target methionine tRNAs in cancer cells [141]. Besides this, tRFs, which were originally thought of as byproducts of tRNA degradation have been described as functional entities. Certain tRFs derived from tRNA^Glu, tRNA^Asp, tRNA^Gly, and tRNA^Tyr have been shown to post-transcriptionally suppress oncogenic factors in breast cancer by displacing YBX1 binding at 3’ UTRs [69]. Conversely, tRNA^Leu-derived tRFs can exert oncogenic roles by promoting the translation of selected ribosomal mRNAs [142].

Fig. 4

Cellular processes with oncogenic and tumor-suppressive roles involving RNA-RBP interactions. A) Alterations affecting the central components of the translation machinery. B) Regulatory processes mediated by noncoding RNAs that influence gene silencing and genome stability. C) Changes perturbing RNA modifications, the splicing process, and protein products.

An alternative regulatory layer in cancer biology encompasses the role of miRNAs and lncRNAs. The primary functions of miRNAs include mRNA destabilization and translational repression of oncogenes (oncomiRs) and tumor suppressors, whereas lncRNAs function through various modes of action including miRNA or RBP decoy activity (or sponging) and scaffolding to guide RNP complex assembly (Fig. 4B). Studying the roles of miRNAs in cancer is an area of intense research. For example, the oncomiR miR-21 is overexpressed in HCC, where it targets PTEN mRNA and inhibits its tumor suppressor activity, enhancing cell proliferation, migration, and invasion [143]. High levels of miR-21 are known to be associated with additional cancer types and downregulation of target genes, such as BCL2 and PDCD4, which mediate cellular apoptosis [144]. Moreover, miRNAs can also target oncogenes to exert tumor-suppressive effects. This applies to let-7a, which has been shown to reduce MYC levels, limiting cell growth in Burkitt lymphoma [145]. In the case of lncRNAs, they remain an enigmatic class of ncRNA, though a versatile set of functions impacting cancer progression has been described to date. Reports have shown that the lncRNA H19 competitively binds to miR-17-5p and miR-152, acting as an endogenous sponge that diverts these miRNAs from their respective targets, YES1 and DNMT1, thus modulating cellular responses in thyroid and breast cancer, respectively [146, 147]. The lncRNA PCAT1 exhibits an oncogenic role in prostate cancer by post-transcriptionally inducing MYC expression and safeguarding MYC levels from miR-34a-induced repression [148]. In addition, certain lncRNAs display protective effects against cancer transformation. For instance, NORAD can sequester PUM proteins and impede their nuclear function promoting genome instability [149]. Also, in the nucleus, HOTAIR acts as a lncRNA scaffold bridging PRC2 and LSD1 protein complexes to repress gene transcription and promote metastasis [150, 151]. Furthermore, the interaction between the chaperone and the noncanonical RBP CCT3 and LINC00326 has been shown to regulate cellular metabolism, whereby downregulation of CCT3 enhances LINC00326 expression to repress lipid accumulation and cell proliferation in HCC [23].

Notably, perturbations in the alternative splicing landscape can significantly impact the development of cancer. This can be influenced by changes in the splicing machinery, such as the availability of snRNAs and other splicing factors, as well as by RNA modifications altering key regulatory sequences in pre-RNA molecules (Fig. 4C). The U1 snRNA silences proximal poly(A) signals at the end of transcripts; hence, its downregulation has been shown to favor the shortening of 3’ UTRs as well as to alter the splicing and expression of cancer-related genes increasing cell migration and invasiveness [152]. Moreover, the lncRNA MALAT1 has been associated with diverse roles in tumorigenesis and is known to regulate alternative splicing by influencing the levels of trans-acting splicing factors, such as serine/arginine-rich (SR) proteins [153]. Regarding RNA editing, an adenosine-to-inosine (A-to-I) modification of AZIN1 transcripts mediated by ADAR proteins leads to a serine-to-glycine substitution at residue 367 (S367G), altering the AZIN1 structure and enhancing its activity, which drives pathogenesis in HCC [154]. Lastly, recent reports have implicated m⁶A modification readers, including METTL3 and YTHDC1, in the recruitment of splicing factors leading to different alternative splicing programs that can potentially explain tumor-associated phenotypes via the production of malignant proteoforms [155, 156].

Conclusion

The ever-growing collection of dependencies between RNA and protein pools in living cells has contributed greatly to our understanding of molecular processes orchestrating cellular physiology in homeostasis and disease. Despite recent advances, our knowledge of RNA-RBP interactions remains limited and is constrained by the current catalog of experimental methods and computational tools. This is especially true for functionally diverse classes such as lncRNAs, where the mechanistic basis for recognition is highly heterogeneous. Given the explosive progress in the topic and the identification of new functional types, old classifications have become outdated, raising the need for novel strategies for categorizing RBPs, RNA species, and their modes of interaction. In fact, recent efforts have been made to separate ncRNA classes [83]. Nonetheless, accommodating emerging functional types, such as short open reading frames (micropeptides), which can exist within ncRNA molecules, is still a challenge [157]. Moreover, interactions between RNA molecules and RBPs have increasingly been recognized as key mechanisms for controlling gene expression, encompassing a wide array of ncRNA molecules. These ncRNAs are crucial for various regulatory processes, including transcriptional and post-transcriptional regulation, chromatin remodeling, and DNA modifications. Notably, recent studies have highlighted the profound impact of ncRNAs on epigenetics, where they modulate gene expression by altering chromatin states and DNA methylation patterns, thus influencing cellular identity and function under both physiological and disease conditions.

On the experimental side, classical approaches to establish RNA-RBP binding events have relied on crosslinking and immunoprecipitation (CLIP) followed by sequencing or mass spectrometry to probe the transcriptome-bound (RBP-centric) or proteome-bound (RNA-centric) fractions, respectively. However, these protocols are time-consuming and scale poorly to large sets of factors, thus limiting the ability to capture the complexity of RNA-RBP regulatory networks. To circumvent some of these limitations, the field is moving toward developing novel antibody-free methods that are versatile for identifying structure- or modification-dependent interactions, highly multiplexable for profiling several factors in parallel, and extendable for reaching single-cell resolution [158,159,160].

In the case of computational approaches, current tools leverage sequence motifs and conservation to define structures and predict functional elements, although their performance is far from optimal. Validation experiments are still of paramount importance for determining the functional relevance of newly discovered structures and interactions. Furthermore, establishing sophisticated methods that can accurately identify RNA-RBP interactions will be essential for reconstructing complex regulatory networks. Recently, deep learning approaches have shown remarkable results for modeling protein structures [161]. However, the performance of AI-based approaches is still limited when comprehensive reference datasets, such as RNA structure and RNA-protein interaction data, are lacking [162, 163]. These approaches may also enable further stratification of elements into functional groups. As we continue to unveil RNA-RBP regulatory networks experimentally with novel methods, enough reference quality data will be generated to train the next generation of prediction tools. Here, large datasets covering multiple layers of biological information combined with integrative strategies will prove instrumental in refining the reconstructed networks and their dynamics across different conditions.

Owing to their specificity for certain cell types or disease states, the ability to robustly define regulatory networks could pave the way for novel diagnostic and therapeutic strategies. Ultimately, deciphering the complexities of RNA-RBP interactions will shed light on fundamental cellular processes while offering unique insight for clinical intervention in cancer and other diseases.