The structure, function and evolution of proteins that bind DNA and RNA

Abstract

Proteins that bind both DNA and RNA epitomize the ability to perform multiple functions by a single gene product. Such DNA- and RNA-binding proteins (DRBPs) regulate many cellular processes, including transcription, translation, gene silencing, microRNA biogenesis and telomere maintenance. Proteins that bind RNA were typically considered as functionally distinct from proteins that bind DNA and studied independently. This practice is becoming outdated, in part due to the discovery of long non-coding RNAs (lncRNAs) that target DNA-binding proteins. DRBPs have unique functional characteristics that stem from their specific structural attributes; these have evolved early in evolution and are widely conserved.

Proteins that bind DNA or RNA are often considered and studied independently. Transcription factors for example are usually modeled relatively simply: they bind to genomic promoters and control target gene expression by activating or repressing RNA polymerases. Following transcription, RNA binding proteins modulate protein expression by regulating the stability and translation of mRNAs. However, the consideration of DNA- and RNA-binding functions within proteins as separate entities is becoming outdated. The unappreciated, dual DNA- and RNA-binding capacity of a growing body of proteins plays a key role in modulating gene expression, cell survival and homeostasis. Recent studies have demonstrated that many transcription factors are capable of binding diverse types of RNA, which enables them to bind to the mRNA products of transcription to regulate their turnover and to integrate other signals, such as responses to stress^1–7. Additionally, the prevalence and emerging functions of long non-coding RNAs (lncRNAs) have revealed that these RNAs target many types of proteins through direct interactions^1,8–11.

In this analysis, we attempt to enumerate these DNA and RNA binding proteins (DRBPs) and describe their function, structure, and evolution. We will first broadly discuss the prevalence of DRBPs within the human genome. We highlight known functions of DRBPs with specific examples of how the simultaneous and serial RNA and DNA interaction allows for better gene targeting, finer control of gene expression, and integration of metabolic state or stress to modulate protein activity. We discuss the structural features of DRBPs that enable dual nucleic acid specificity, focusing on the limited number of solved structures that allow for direct comparison of a DRBP complexed to either DNA or RNA. Finally, we discuss the evolution of dual DNA and RNA binding domains within DRBPs, including both ancient domains, for which dual DNA and RNA binding conferred a selective advantage, and more modern domains which have recently been targeted by rapidly evolving lincRNAs.

Defining DNA and RNA binding proteins

Defining the subset of human proteins that bind both DNA and RNA is a difficult task. Using gene ontology searches, only 64 human protein-coding genes in the QuickGO gene ontology database (European Bioinformatics Institute; ref. ¹²) are identified as having direct and specific experimental evidence for both RNA binding (GO:0003723) and DNA binding (GO:0003677) (Fig. 1a). The PROTEOME database (BioBase) returns 122 such proteins, although direct evidence is lacking for many of them.

Figure 1. Defining human DRBPs.

a. Venn diagram of DNA binding and of RNA binding proteins in the QuickGO database supported by low-throughput experimental evidence (as of July 2014)¹². The overlap of these two sets represents human DNA and RNA interacting proteins (DRBPs), consisting of 64 proteins. b. Venn diagram of DNA binding and of RNA binding proteins identified in high-throughput studies defining the human mRNA² and dsDNA³ interactomes^13,14. There are 407 proteins found in both studies, indicating that they may bind both mRNA and dsDNA. c. Molecular function gene ontology analysis reveals that RNA binding is a potentially major function of the dsDNA binding proteins identified in REF 2². d. Gene ontology analysis reveals that DNA binding is potentially a major function of the mRNA binding proteins identified in REF 3. In parts a and b, circles are drawn to scale. In parts c and d, only selected molecular function attributes are shown for brevity. p-values in parts c and d indicate the probability that the over-representation of the stated ontology term in the selected 407 genes compared to all human genes is due to chance. These were calculated in the TRANSFAC + PROTEOME database (BIOBASE) using the hypergeometric distribution; “very large” indicates a p-value of less than 1 x 10⁻⁴⁰ (−log(p) > 40).

An alternative approach involves combining evidence from studies that have separately attempted to catalog all human proteins that bind DNA or RNA (Fig. 1b). A study using protein microarrays and bioinformatics approaches identified over 4,000 human proteins that directly interact with double-stranded DNA (dsDNA) in vitro¹³. Gene ontology analysis of these proteins reveals that the term RNA binding is highly enriched for (p < 1x10⁻⁴⁰), indicating that RNA binding may be a common feature of DNA binding proteins (Fig. 1c). Among these dsDNA-binding proteins, the ontology term ‘dsRNA binding’ is much more represented than ssRNA binding.

Another study used a crosslinking and mass spectrometry-based approach to identify 860 mRNA-binding proteins from HeLa cells termed the “mRNA interactome”¹⁴. Functional analysis of these proteins again indicates that dual nucleic acid binding is a widespread phenomenon (Fig. 1d), with both the terms ssDNA- and dsDNA-binding being significantly enriched for (p = 8.9 x 10⁻²¹ and 5.5 x 10⁻⁸, respectively). Notably, of the 860 proteins identified as mRNA binding, 407 (47.3%) were independently characterized as dsDNA binding in REF 2. Together, the two studies indicate that DRBPs are a widespread phenomenon, perhaps comprising 2% of the human proteome (407 proteins, Fig. 1b). This number would likely increase if proteins that are expressed in other cell types that require ligand binding-dependent signals for nucleic acid binding, or that bind other types of DNA or RNA, are included.

We note that many of the proteins identified in REFS 2 and 3 as DNA and/or RNA binding lack corroborating evidence from other studies, and thus these findings should be interpreted with caution. For example, many identified proteins such as polymerase subunits may bind nucleic acid-bound proteins without binding DNA and/or RNA directly. Additionally, many proteins that bind DNA or RNA in vitro may not bind them in vivo. However, we believe the two studies provide a reasonable estimate of potential human DRBPs due to their wide coverage of the human proteome, and we will discuss examples where the demonstration of protein-nucleic acid binding in vitro has preceded the discovery of such binding in vivo, sometimes by decades.

In Supplementary Table 1 we provide a detailed list of 149 human DRBPs, with comments on their nucleic acid binding properties, structure, and function. These proteins, drawn form the 271 proteins mentioned above, were selected based on experimental evidence demonstrating their ability to bind directly to both DNA and RNA, generally obtained from studies using more traditional experimental approaches than the high-throughput studies referenced above. While many of the proteins in Supplementary Table 1 have only been shown to bind DNA and/or RNA in vitro, the remainder of this review will focus on selected human DRBPs with known cellular roles.

Functions of DRBPs

We performed gene ontology and domain enrichment analyses (Fig. 2) to illuminate the main biological functions of our list of human DRBPs (Supplementary Table 1). The gene ontology analysis revealed expected biological processes such as transcriptional regulation, mRNA processing, and DNA replication. However, several surprising functions are also implicated, including the DNA damage response, apoptosis, and responses to extreme temperatures (Fig. 2a).

Figure 2. Functional and structural properties of DRBPs.

The 149 DRBPs (Supplementary Table 1) were subjected to gene ontology enrichment of biological process (PROTEOME database, Biobase) and to INTERPRO domain enrichment (DAVID ontology^77,78), to explore the biological functions of and protein domains commonly found in DRBPs. a. Gene ontology analysis. Biological processes such as transcriptional regulation and mRNA processing are expectedly prominent terms found enriched for DRBPs. However, unexpected functions are also enriched, including response to many cellular stresses (heat, viral, radiation, etc.). For brevity, only selected functions are shown. b. Domain enrichment analysis. All domains enriched in the set of 149 DRBPs that have p-values equal or smaller than p = 10⁻³ are shown. p-values in parts a and b indicate the probability that the over-representation of the stated term in the 149 DRBPs compared to all human genes is due to chance.

Ultimately, DRBP function is governed by their inherent structural and biochemical properties. One can envision DRBPs capitalizing on both RNA and DNA binding in a number of ways; for example, a transcription factor that binds DNA and RNA may interact orthogonally with RNAs that compete with DNA binding to repress transcription, or simultaneously with a promoter and an RNA coactivator to upregulate transcription. The following section will focus on DRBPs that bind DNA and RNA competitively (Fig. 3a).

Figure 3. Three archetypes of DRBP function.

a, RNA can compete with DNA for binding to DRBPs, typically at the same protein interface. In the case of transcription factors, this can reduce promoter occupancy and the transcription of target genes. b, DRBPs can regulate gene expression at multiple levels. In addition to binding to the promoters of genes to regulate their transcription, DRBPs can also affect miRNA processing and mRNA stability and translation. c, DRBPs, such as SF-1, can bind DNA and RNA simultaneously, whereby the RNA functions as scaffold to recruit other proteins to a specific DNA locus. Shown here is the DRBP, SF-1, binding to the lncRNA SRA to recruit the steroid receptor coactivator 1 (SRC-1) transcriptional complex in a ligand independent manner.

Binding DNA or decoy RNAs

The role of certain lncRNAs as decoys of genomic DNA is illustrated by the reduction in promoter occupancy by transcription factors, typically measured by chromatin immunoprecipitation (ChIP), in response to the overexpression of competing decoy RNAs. The glucocorticoid receptor (GCR), a steroid hormone receptor, is a classic example of a ligand-activated transcription factor (reviewed in ¹⁵). In its inactivated state, GCR is kept in the cytoplasm by chaperone proteins. Upon ligand binding, GCR translocates to the nucleus where it can bind to the promoters and regulate the transcription of hundreds of genes¹⁶. Given the anti-inflammatory role of the GCR, much effort has been put into developing modulators of GCR-driven transcription¹⁷. The lncRNA Growth Arrest Specific 5 (Gas5) was found to inhibit the transcriptional activity of GCR by competing directly with DNA for protein binding in vitro and in cells¹; overexpression of Gas5 leads to a decrease in ChIP-detected GCR occupancy at its target promoters as well as decrease in the mRNA levels of glucocorticoid-activated genes^1,18. Since cellular Gas5 levels are regulated by nonsense mediated decay¹⁸ in response to serum starvation and other stressors¹, the transcriptional activity of the GCR is tuned by titrating the levels of Gas5 against the fixed number of genomic GCR binding sites in response to cellular stress. Three closely related steroid receptors that share the DNA specificity of the GCR, the androgen, progesterone, and mineralocorticoid receptors, are also susceptible to Gas5-mediated transcriptional repression¹. Although steroid receptors have traditionally been thought of as DNA binding proteins, the affinity of the GCR for RNA and DNA is similar, as measured in vitro by GST pulldowns and fluorescence-based competition assays¹. The most distantly related member of the steroid receptor family, the estrogen receptor, does not share the DNA specificity of the GCR and is not susceptible to Gas5-mediated transcriptional repression, indicating that the binding of steroid receptors to RNA is sequence specific¹.

Additional examples of pairs of transcription factors and decoy RNAs are Nuclear Factor-Y, which binds also the lncRNA PANDA⁸, and NF-κB, which binds also a mouse pseudogene-derived RNA, Lethe⁹. The dual nucleic acid binding activity of NF-κB had been demonstrated in vitro many years prior to the discovery of an endogenous RNA target¹⁰, suggesting that transcription factors that are DRBPs such as AML1/RUNX1, whose RNA binding capacity has only been shown in vitro¹⁹, may have also endogenous RNA targets awaiting discovery. Although structural information on the interaction of human proteins with their RNA decoys is lacking, a recent study demonstrates an elaborate mechanism of an analogous bacterial system, the sequestration of RsmE by the ncRNA RsmZ¹¹. Competitive DNA and RNA binding is not only a feature of transcription factors, but also of nucleic acid-modifying enzymes such as DNA methyltransferases. In humans, DNA methylation is initiated by DNMT3A and 3B; DNMT1 maintains this methylation by binding to hemimethylated DNA after replication (reviewed in ²⁰). RNAs binding can inhibit the DNA binding and methylation activity of both DNMT3A²¹ and DNMT1²². In vitro, DNMT1 binds RNA with a higher affinity than DNA as shown in electrophoretic mobility experiments²¹. In the cases of DNMT1 and likely DNMT3A, RNAs bind to the catalytic domain of the methyltransferase to inhibit DNA methylation^21,22.

It is notable that several metabolic enzymes are DRBPs with competitive DNA and RNA binding capacity, such as the glycolytic enzymes lactate dehydrogenase (LDH)^23–25, glyceraldehyde-3-phosphate dehydrogenase (GAPDH)^26–28, and α-enolase (ENO1)^14,29,30. In the case of GAPDH, both DNA and RNA compete for binding of the cofactor NAD+ to the enzyme^26,28, suggesting that Rossmann fold-containing proteins such as GAPDH may be sensitive to cellular DNA and/or RNA levels. ENO1 binds RNA as a monomer³⁰, which inhibits the formation of the catalytically active protein dimer^31,32. NAD+ specific isocitrate dehydrogenase, which converts isocitrate to alpha-ketoglutarate, is allosterically inhibited by the 5′untranslated regions of yeast mitochondrial mRNA³³. Binding of RNA and DNA to metabolic enzymes indicates that nucleic acids can modulate the function of proteins other than transcription factors to modulate cellular metabolism³⁴.

DRBPs that regulate gene expression at multiple levels

Approximately half of the DRBPs we identified in our analysis are transcription factors. As discussed above, some such proteins have been shown to be the targets of RNA decoys. In contrast, several others bind both the DNA and the mRNA of their target genes (Fig. 3b). Regulating genes at both the DNA and RNA levels enables powerful, combinatorial control over protein expression, and may allow DRBPs to generate both immediate effects (through regulating RNA turnover) as well as long-lasting effects (through regulating transcription).

For the first example of such a DRBP, we return to GCR, which when activated can promote the transcription of anti-inflammatory genes³⁵ as well as repress the transcription of pro-inflammatory genes^36,37. Agonist-bound GCR destabilizes the mRNA of pro-inflammatory genes such as MCP-1 through direct RNA binding, perhaps by the recruitment of ribonucleases³⁸. The identification, using RNA immunoprecipitation (RNA-IP), of a GCR binding motif in many immunogenic mRNAs suggests that GCR can accelerate the decay of a large number of mRNAs, broadening its role in the anti-inflammatory reponse². Given that GCR also binds directly to pro-inflammatory transcription factors such as AP-1 and NF-κB^39,40, it would appear that GCR takes advantage of its diverse DNA, RNA, and protein binding capacities to regulate inflammatory genes at the transcriptional and post-transcriptional level.

Transcription factors can also regulate gene expression post-transcriptionally through the regulation of microRNA (miRNA) biogenesis. miRNAs are small RNAs that facilitate gene silencing through sequence-specific pairing to and recruitment of the target mRNAs to the RNA induced silencing complex (RISC; reviewed in ref ⁴¹). Several transcription factors have been shown to regulate Drosha-mediated primary (pri-)miRNA processing, a key step in the biogenesis of functional miRNAs⁴². Smad proteins, the transducers of TGF-β signaling, activate transcription by forming a DNA-binding heterodimer (reviewed in ⁴³). Smad proteins also increase the levels of several miRNAs, including miR-21 (ref. ⁴⁴), which plays important roles and development and immunity⁴⁵. Surprisingly, the increase in miR-21 levels is due not to increased transcription of pri-miR-21 transcripts, but due to increased Drosha-mediated processing of pri-miR-21 to precursor (pre-)miR-21 (ref. ⁴⁴). Bioinformatics analysis identified a conserved RNA motif in TGF-β-regulated miRNAs that was shown by RNA-IP and electrophoretic mobility shift assays (EMSA) to bind directly to the Smad MH1 domain to mediate Drosha processing³. Interestingly, the RNA sequence motif that is bound by SMAD4 and mediates the regulation of miRNA expression post-transcriptionally, is identical to the DNA sequences that are bound by SMAD4 and mediate regulation of gene expression transcriptionally³.

NF90 (ILF3) is a particularly versatile DRBP that, along with its partner NF45, plays important roles in T cell activation⁴⁶. Through the direct binding of DNA, mRNAs and miRNAs, NF90 controls transcription^4,5, regulates mRNA turnover and translation^47,48, and affects miRNA processing ⁴⁹, respectively. All of these functions assist in its role in T cell activation: NF90 upregulates the mRNA levels of IL-2, a critical cytokine in T cell development⁵⁰, by both binding its promoter and activating its transcription and by stabilizing the IL-2 mRNA through direct binding to its 3′ UTR, as was found by using EMSA and ribonucleoprotein immunoprecipitation analysis^5,48. Additionally, using in vitro pri-miRNA processing assays and RNA-IP, NF90, when in complex with NF45, was shown to inhibit the processing of the pri-miRNA pri-let-7a binding it directly⁴⁹. let-7a represses IL-6, a cytokine critical for T cell survival and proliferation⁵¹, which may link inflammation to cancer⁵², and let-7 downregulation following NF90 upregulation reduces survival in several cancer types^53,54. In summary, these examples illustrate that DRBPs can utilize both transcriptional and post-transcriptional mechanisms to serve as potent controllers of gene expression.

Simultaneous binding of DNA and RNA

In contrast to DRBPs that target DNA or RNA serially to perform different or related functions, another class of DRBPs binds RNA and DNA simultaneously to perform a single function (Fig. 3c). Generally, transcription factors not only require DNA binding to target promoters but also bind to corepressors or coactivators to affect transcriptional regulation. There are several examples of RNA molecules acting as coactivators, by simultaneously binding DNA and various transcription factors. The lncRNA RMST (rhabdomyosarcoma 2-associated transcript), in particular, is required for binding of neurogenic gene promoters and subsequent upregulation by SOX2⁵⁵, a transcription factor with important roles in development, pluripotency, and cell fate⁵⁶. RNA-IP and RNA pulldown experiments demonstrated that RMST interacts directly with SOX2^55,57, and DNA occupancy of SOX2 measured by ChIP-seq was reduced following RMST depletion^55,57. The lncRNA Evf-2 serves as a transcriptional coactivator for Dlx-2⁵⁸ and recruits MECP2 to intergenic enhancers⁵⁹. A direct interaction between Dlx-2 and Evf-2 has been demonstrated by the immunoprecipitation of Dlx-2 followed by RT-PCR of the Evf-2 lncRNA⁵⁸, and MECP2 also has previously been shown to bind RNA⁶⁰. It should be noted that RNA-mediated recruitment of a protein to a particular DNA locus might not require direct binding of both DNA and RNA by the protein, as lncRNAs could recruit transcription factors to a particular DNA locus to which the lncRNA is bound. Dual nucleic acid recognition also facilitates targeted gene repression, through RNA-guided DNA methylation. This phenomenon was first discovered in plants⁶¹, and some mammalian RNA guides of DNA methylation have since been found^62,63, although their mechanisms of action are less clear. In mice, DNMT3A forms a complex with Tsix RNA to promote methylation of the Xist promoter⁶⁴.

Several nuclear receptors, including SF-1, DAX-1, and the thyroid receptor α (TRα), bind directly to both gene promoters^65,66 as well as to the RNA co-activator SRA in order to modulate the transcriptional activation^6,7 (Fig. 3c). Using pulldown experiments, SF-1 and TRα have been shown to bind SRA through their hinges, which are flexible, disordered regions that connect their DNA and ligand binding domains^6,67. Knockdown of SRA decreases the interaction of SF-1 with protein transcriptional activators as well as the transcription of SF-1-regulated genes⁶. Several other nuclear receptors associate with, but lack direct evidence for direct binding to, SRA, including the androgen, progesterone, and estrogen receptors as well as the retinoic acid receptor α (RARα), which may bind SRA and its target gene promoters simultaneously^7,68–70. Crosslinking immunoprecipitation has demonstrated that RARα can bind to and regulate the translation of target mRNAs⁷¹ through a unique RNA binding motif at its C-terminus⁷¹.

Another example of simultaneous DNA and RNA binding that is required for DRBP function is the role of TRF2 (Telomeric Repeat binding Factor-2) at telomeres. Deletion of TRF2 leads to an arrest in cell division caused by the formation of chromosome end-fusions⁷². Crystal structures have revealed that TRF2 binds to telomeric DNA in a sequence-specific manner through a C-terminal DNA binding domain that resembles a homeodomain⁷³. Part of the role of TRF2 at the telomere includes the recruitment, through its positively charged N-terminal GAR domain, of the origin recognition complex (ORC; a collection of proteins that serves as a scaffold for DNA replication factors, among other functions⁷⁴), where it can assist in the maintenance of telomere structure⁷⁵. Using biotinylated RNA pulldown experiments, RNA-IP, and EMSAs, the GAR domain responsible for ORC recruitment was later shown to bind telomere repeat-encoding RNA (TERRA)⁷⁶. Depletion of TERRA hampers ORC recruitment to the telomere without affecting TRF2 binding to the telomere itself, suggesting a model in which TRF2 serves as a mediator between telomere DNA and TERRA, which in turn recruits factors required for telomere maintenance⁷⁶.

Structural characteristics of DRBPs

For a protein such as TRF2 to coordinate telomeric DNA binding and recruit protein complexes by binding of RNA, it must have multiple nucleic acid binding motifs. Some DRBPs such as GCR and NF-κB have maintained domains capable of binding both DNA and RNA, allowing decoy RNAs to evolve and compete with DNA for protein binding. In this section, we will analyze the prevalence of structural domains in DRBPs and discuss examples of DRBPs that bind both single-stranded and double-stranded nucleic acids.

DRBP domains that enable DNA and RNA interactions

We performed INTERPRO domain enrichment analysis by DAVID^77,78 on our 149 DRBPs in order to identify domains enriched in proteins that bind both DNA and RNA (Fig. 2b). The RNA recognition motif (RRM; also known as ribonucleoprotein domain {RNP} or RNA binding domain {RBD}) was the most highly enriched domain in DRBPs (p = 2 x 10⁻²⁶). The RRM is an abundant, short (~100 amino acids-long) domain that generally recognizes single-stranded RNA (ssRNA) and is often present in proteins with other domains, such as zinc finger domains, WW domains, or additional RRMs⁷⁹. Such multi-domain DRBPs may bind RNA and DNA simultaneously through separate domains, such as the two RRM-containing hnRNP A1 (ref. ⁸⁰). Single RRM-containing proteins are also capable of binding both DNA and RNA, a function present for example in RBM3, TAF15, and TDP-43 (refs. 81–83) (Supplementary Table 1). Such bivalent domains may not have the same sequence specificity when binding DNA and RNA, highlighting the complexity of recognizing nucleotide bases in a sequence-dependent manner.

Nuclear receptor domains are also enriched in DRBPs: RARα binds mRNA through a unique C-terminal domain⁷¹, SF-1 binds the RNA coactivator SRA through its hinge and a unique Ftz-F1 domain, and TRα binds SRA through its hinge^6,67 (Fig. 3c). The majority of nuclear receptors have two highly-conserved Cys₄ zinc fingers through which they bind DNA, and some nuclear receptors, such as the GCR, can also bind RNA through these domains¹. Other types of zinc fingers are also enriched in DRBPs, such as the RanBP2-type. Other notably enriched domains in DRBPs are the K homology (KH) domain, the double-stranded RNA binding domain (dsRBD), the cold shock domain (CSD), and various helicase domains. Each of these domains is capable of binding DNA and RNA, and here we will focus on the structural mechanisms underlying this dual specificity.

General properties of DRBPs

There are only two chemical differences between RNA and DNA. First, RNA, but not DNA, has a 2′OH group on the ribose sugar, which allows for an additional hydrogen bond and a greater diversity of secondary structure than is possible in DNA. Second, RNA contains uracil rather than thymine in DNA, which differs by the presence of a methyl group at the C5 position. A comparative analysis of known protein-nucleic acid structures revealed that the recognition of DNA occurs largely through electrostatics and direct base–protein interactions. RNA recognition by proteins, on the other hand, is dependent largely on shape complementarity and interaction with the 2′OH⁸¹. Given these general differences, one could expect that during evolution, highly selective protein interfaces would be generated that are optimized for either RNA or DNA, with minimal cross-binding. However, the most energetically favourable associations between proteins and nucleic acids relay on hydrophobic and charge–charge interactions. These interactions are less constrained than interactions with the sugar backbone or with the nucleotide base edge, which is capable of highly-specific Watson–Crick base pairing. Thus, DRBP domains that competitively bind DNA and RNA probably rely on the less specific hydrophobic and charge–charge interactions. For example, ssRNA-binding proteins are more likely to form hydrogen bonds with bases rather than with the phosphate–sugar backbone, compared to those that recognize folded RNA, such as ribosomal proteins and tRNA synthetases⁸². Because ssRNA-binding proteins do not rely heavily on sugar recognition, they are more likely to also bind DNA. This may explain why the RRM is the most enriched domain in our DRBPs (Fig. 2b)⁸².

The RNA recognition motif

The RRM is an extremely versatile domain that is capable of binding (mainly single-stranded) RNA and DNA, as well as proteins⁷⁹. As stated above, RRMs preferentially interact with nucleic acid bases rather than with the phosphate-sugar backbone. The structural nature of ssRNA and ssDNA allows for much easier access to the exposed aromatic base faces, as opposed to hydrogen bonding to the base edges that occur frequently with double-stranded nucleic acid binding DRBPs. Additionally, stacking interactions with the faces of bases are more energetically favorable than recognition of the nucleotide edge. Therefore, stacking interactions between aromatic protein side chains and nucleic acid bases are often observed in single-stranded nucleic acid binding proteins.

TDP-43 is a DRBP that plays important roles in mRNA splicing and miRNA biogenesis^83,84. It contains two RRMs that are separated by a short loop, both capable of binding DNA and RNA. Crystal structures of both RRMs in complex with DNA and RNA have been reported^85,86 (see also the currently unpublished structure of TDP43 RRMs: PDB4 4IUF), making TDP-43 an excellent case study for dual DNA and RNA recognition by RRMs. The DNA- and RNA-bound structures of TDP-43 reveal nearly identical modes of nucleic acid recognition. Aromatic side chains, such as Phe149 within the first RRM (RRM1), form stacking interactions with DNA or RNA bases (Fig 4a). Trp113, part of the more flexible loop 1, is able to shift conformations and base-stack slightly differently when bound to differing nucleic acid sequences (Fig. 4b), whereas Phe149, part of the rigid β₃ sheet of the RRM1 fold, makes similar interactions with DNA and RNA (Fig. 4a). Relying on the more energetically favorable π-stacking interactions through the planar face of the DNA and RNA bases, results in less specificity than would be gained from hydrogen bonding with the base edge. Uracil and thymine interact with Phe194 of the second RRM in the RNA- and DNA-bound structures, respectively (Fig. 4c, d). Despite the additional methyl moiety at position C5 in the DNA, no TDP-43 residues recognize the edge of the nucleotide in order to interact or clash with the additional carbon (Fig. 4d). Thus, one of the chemical differences between DNA and RNA, the use of uracil in RNA, plays no role in nucleic acid discrimination in this example.

Figure 4. The structural basis for dual DNA and RNA recognition by TDP-43 and by the NF-κB subunit p50.

Protein-RNA structures are shown in blue and protein-DNA structures in green, with protein in the darker shade. π-stacking interactions play a prominent role in both the ssDNA and ssRNA binding activities of TDP-43: a–d. Phe149 (a) and Trp113 (b) within the first RRM of TDP43 stack with both RNA and DNA bases. c, d. In the second RRM of TDP-43, Phe194 is capable of recognizing both uracil in RNA and thymine in DNA; the additional methyl group at C5 in thymine does not contribute to nucleic acid specificity. e. When bound to RNA, both the terminal amine and ε nitrogen of Arg227 in the second RRM of TDP-43 contact a 2′OH on the RNA backbone. f. In contrast, these same groups can also make contacts with the DNA backbone, both directly and through water mediated hydrogen bonding. g, h. The p50 subunit of NF-κB makes strikingly similar base-specific contacts when bound to double-stranded DNA (g) or an RNA aptamer (h). This is due in large part to the similar secondary structure and chemical moieties presented by the RNA and DNA. Major groove width was calculated by 3DNA using phosphate-phosphate distances⁹².

In contrast, the second RRM of TDP-43 does make RNA-specific contacts with the 2′OH group. The majority of protein – 2′OH interactions are mediated through protein side chains⁸¹, and both the RRM2 Lys263 and Arg227 (Fig. 4e) side chains contact a 2′OH when bound to RNA. When it is bound to DNA, however, these same protein side chains contact the DNA backbone phosphates, demonstrating that amino acids are capable of reorienting to enable distinct types of interactions to support RNA and DNA binding (Fig. 4f). DNA binding is not a general property of all RRMs, however. For example, the RRM of PABP relies on a large number of RNA-specific 2′OH contacts for RNA interaction, and binding to DNA may be at low-affinity, if detectable at all⁸⁷.

DRBPs that recognize double stranded nucleic acids

Crystal structures of protein – dsRNA complexes are less common than their single-stranded counterparts, but there are some examples that are instructive for dual nucleic acid recognition. NF-κB is a central transcription factor of the immune signaling, and is formed of homo- or heterodimers of Rel family proteins such as p50 (NFKB1) or p65 (RELA)⁸⁸. High-affinity aptamers have been developed for both the p50 and p65 subunits of NF-κB, with affinity of RNA binding that approaches that of the transcription factor’s affinity for native DNA response elements^10,89. DNA with an identical sequence to the p50-targeting aptamer will not bind p50 (ref. ¹⁰); p50 is therefore another DRBP that binds RNA and DNA is a sequence-specific manner, with different sequence specificities for each.

Crystal structures of p50 bound to DNA and to RNA have been solved, revealing that both bind at the same surface of the p50 Ig-like domain^90,91. Although p50 binds to RNA as a monomer and to DNA as a dimer, similar networks of base-specific interactions occur in each structure between protein and nucleic acid (Fig. 4g, h). Not only do the DNA and RNA contacting residues of p50 maintain an equivalent position, but both DNA and RNA present similar interfaces for p50 recognition in charge distribution and in secondary structure (Fig. 4g, h)⁹². This is a seminal, structurally-confirmed example of “DNA mimicry” by RNA in order to bind to a transcription factor, and although the RNA in this case was artificial, DNA mimicry has been hypothesized to play a role in the endogenous regulation of several transcription factors^1,93.

Structures have also been solved of the DRBP ADAR1, which binds both double stranded Z-DNA and double-stranded Z-RNA through its unique Zα domain^94,95. The ability of the Zα domain to make sequence-independent interactions with the Z-form phosphate backbone of both DNA and RNA enables ADAR1 to sense nucleic acid secondary structure conformations. Thus, double-stranded nucleotide-binding DRBPs can recognize their DNA and RNA targets through sequence-specific interactions (in the case of NF-κB) or through non-specific interactions with the DNA and RNA backbone (in the case of ADAR1).

The evolution of DRBPs

The evolutionary forces driving the structure and function of DRBPs are complex, yet understanding them will help us identify new DRBPs and perhaps predict their susceptibility to interactions with lncRNAs. While the DRBPs identified in our analysis are members of many different structural classes, each with their own evolutionary history, we will focus on members of two very dissimilar DRBP families: cold-shock domain (CSD)-containing proteins and eukaryotic DNA methyltransferases. CSD-containing proteins, required to protect cells from low temperatures, are members of an ancient DRBP family that utilize weak selection criteria to interact with nucleic acids and therefore intrinsically bind to both DNA and RNA. Members of the eukaryotic DNA methyltransferase family represent DRBPs that have more recently evolved the ability to recognize both DNA and RNA: they prefer interactions with DNA, and only one family member (DNMT2) acquired the ability to bind and methylate tRNAs⁹⁶. The discovery of a eukaryotic DNA methyltransferase with exquisite tRNA methylation activity and only modest DNA methylation activity showcases how evolution can rewire protein surfaces to create new functions of DRBPs.

The ancient cold shock domain DRBPs

The CSD is one of the most ancient nucleic acid binding domains, found in both prokaryotes and eukaryotes. All CSD-containing proteins bind DNA and RNA (Supplementary Table 1). In humans, several CSD-containing proteins exist, such as the three Y-box binding proteins, the lin-28 homologs LIN28A and LIN28B, and CSDE1 (also known as UNR). The Y-box protein 1 (YB-1) was originally named for its ability to bind and repress the “Y-box” of MHC Class II promoters⁹⁷. YB-1 also binds RNA, with roles in alternative splicing⁹⁸, translational control⁹⁹, and RNA stabilization¹⁰⁰. In addition, YB-1 binds to damaged DNA and is involved in the DNA damage response^101–103 - it translocates to the nucleus following stresses such as UV radiation^104,105.

In bacteria, CSDs exist in short proteins that contain one CSD with little flanking sequences. In Escherichia coli, there are nine such proteins (cspA-cspI), which are likely products of multiple gene duplication events¹⁰⁶. Of these, cspA, cspB, cspG, and cspI are induced by cold stress, with cspA briefly comprising over 10% of all protein synthesized during cold shock^107–110. Simultaneous deletion of the four genes results in lack of E. coli colony formation at 25°C or lower¹¹¹. cspD is induced by nutrient stress¹¹², but cspC and cspE are constitutively expressed at normal growth temperature¹¹³. Many (if not all) of the csp genes bind DNA and RNA^114,115 and have similar roles as the human CSD-containing proteins: in maintaining RNA stability¹¹⁴, in translational regulation¹¹⁶, transcriptional control^116,117, DNA replication and repair^118,119, and chromosome folding¹²⁰.

CSD-containing proteins are widespread in plants¹²¹, where they perform similar cellular functions. The first csp-like protein found in plants was the wheat WCSP1, which is upregulated specifically by cold stress and binds ssDNA, dsDNA, and RNA homopolymers¹²². WCSP1 was found to complement the cold-sensitive phenotype of the E. Coli cspA/B/G/I knockouts mentioned above¹²³, exhibiting remarkable functional conservation. In addition, WCSP1 showed in E. Coli nucleic acid melting activity, critical to preventing inappropriate nucleic acid secondary structures that disrupt and terminate transcription. This activity is similar to the endogenous E. coli cspA, which also has transcription antitermination activity¹²³. In Arabidopsis thaliana four CSD-containing proteins are found, AtCSP1-4, all of which can also complement the quadruple csp knockout in E. coli to varying degrees, suggesting that their DNA and RNA interactions are well conserved during evolution^124–126.

Unlike their counterparts in bacteria, most plant and animal CSD-containing proteins have additional functional domains (including more CSDs), which expand their functions, protein-protein interactions, and/or nucleic acid binding specificities. For example, the human protein UNR has five CSDs, which serve to increase the protein’s affinity for target RNA sequences¹²⁷. YB-1 has both an N- and a C-terminal domain flanking its CSD, which can support homomultimerization and interactions with many other protein partners (reviewed in ¹²⁸). In addition to its CSD, WCSP1 has three CCHC zinc fingers through which most of its dsDNA binding is mediated¹²². Nevertheless, the exceptional sequence and functional conservation between eukaryotic CSD-containing proteins and bacterial csps proteins demonstrate a conserved, ancient role and origin of the domain. Most likely, a CSD fold capable of binding DNA and RNA was present in the last common ancestor of bacteria, archaea, and eukaryotes¹²⁹.

The curious case of DNMT2

DNA methylation plays important roles in gene expression and in repressing transposable elements in eukaryotic cells. There are three eukaryotic proteins in the cytosine-C5 DNA methyltransferase family, DNMT1, DNMT2, and DNMT3. Whereas DNMT1 and DNMT3 play important roles in maintaining genome-wide methylation, DNMT2 has diminutive DNA methylation activity¹³⁰ and instead is capable of methylating tRNAs⁹⁶ (Fig. 5a). When this activity was discovered, it was speculated that the three eukaryotic DNMTs might have evolved from an RNA methyltransferase⁹⁶. However, there is no evidence that DNMT2 is more closely related to the ancestral protein of the family members. In fact, the three eukaryotic DNMTs may not be monophyletic and may have evolved from separate prokaryotic DNA methylation restriction-modification enzymes¹³¹. Thus, it seems likely that DNMT2 shifted its nucleic acid specificity from DNA to RNA in the last common eukaryotic ancestor¹³¹ (Fig. 5b).

Figure 5. DNA methyltransferases target both DNA and RNA.

a, Best known for their role in gene silencing, all DNA methyltransferase family members are able to interact with both RNA and DNA^21,22,96. DMNT1 and DMNT3 play a role in initiating and maintaining DNA methylation while DNMT2 methylates tRNAs. This tRNA modification is critical for maintaining tRNA stability and cell viability. b, Cladeogram depicting the evolution of the three major families of DNMTs¹³¹. DNMT2 likely diverged from its ancestral DNA methyltransferase activity to perform a critical role in methylating tRNAs, a function which it performs redundantly with NUSN2¹³². This radical change in DNMT’s substrate specificity highlights the ability of evolution to reshape a DNA-binding interface into one that preferentially recognizes RNA.

Despite the relatively narrow substrate specificity of DNMT2 compared to its family members, it is highly conserved and is the only extant DNMT in some species such as Saccharomyces pombe and Drosophila melanogaster¹³¹. This seems to indicate that DNMT2 has an important physiological role; however, DNMT2^−/− mice are viable, fertile, and yield no obvious phenotype⁹⁶. This apparent contradiction was resolved with the recent report that deleting DNMT2 in addition to the deletion of another tRNA methyltransferase, NSUN2, is lethal¹³². These mice exhibit defects in tRNA stability, protein synthesis, and differentiation¹³², implying that the DNA methylation activity of DNMT2 is dispensable whereas its tRNA methylation activity is not.

NSUN2 is a member of the NCL1 family of eukaryotic RNA cytosine-C5 methyltransferases, which are broadly distributed among eukaryotes¹³³. Interestingly, NSUN2 itself is a DRBP, able to bind and methylate both tRNA and hemi-methylated DNA¹³⁴. CLIP-based analyses showed that NSUN2 methylates also ncRNAs and mRNAs¹³⁵. Given the distant evolutionary relationship between DNA and RNA cytosine-C5 methyltransferases¹³¹, NSUN2 and DNMT2 have most likely undergone convergent evolution from an RNA- and a DNA-binding protein family, respectively, to ensure proper tRNA modification. Furthermore, these evolutionary trajectories have bestowed on both proteins the ability (if residual) to bind and modify both DNA and RNA. Not only does this indicate that proteins with evolutionary conserved DNA-binding activities are capable of binding RNA (and vice-versa), but also that some nucleic acid substrates may be similar enough in sequence and structure to foment protein promiscuity. As mentioned above, this phenomenon is exploited by RNAs, both endogenous and artificial, that function as decoys in order to modulate DRBP function^1,19,136.

Conclusion and perspectives

In this analysis, we have demonstrated that DRBPs comprise a significant fraction of cellular proteins – perhaps 2% of the human proteome – and play important cellular roles. Their functions include the control of transcription and translation, DNA repair, mediating responses to stress, splicing, apoptosis and more. These functions are intimately linked to their structure: orthogonal binding of DNA and RNA provides an opportunity for competitive regulation of transcription by decoy RNAs whereas simultaneous binding of DNA and RNA permits transcriptional activation by RNA coactivators or allows the recruitment of RNA-containing complexes to specific DNA loci. In turn, the structures underlying DRBP functions are linked to their evolution. Some DRBPs contain ancient domains that have long bound DNA or RNA; others contain multiple domains that separately confer DNA and RNA binding and mediate their functional roles.

The majority of RNA binding proteins have had remarkably similar motifs during evolution¹³⁷, although individual members of protein families, such as the forkhead box transcription factors, can have diverse nucleic acid-sequence specificities arising from independent evolutionary events¹³⁸. It is also worth noting that intrinsically disordered protein domains that do not fold into defined secondary structures may also play important roles in mediating nucleic acid binding¹⁴, as was found for RNA chaperones¹³⁹. In addition to protein evolution, nucleic acid sequence evolution also plays important roles in the evolution of DRBP function. The emergence of lncRNAs has illuminated new cellular binding-targets for proteins previously thought of as DNA-specific binding proteins. Tens of thousands of human lncRNAs have been catalogued¹⁴⁰, and it is likely that many of them have yet-undiscovered functions requiring binding to proteins that are currently considered as DNA-specific binding proteins or that have so far only been shown to bind RNA in vitro. For example, GCR and the estrogen receptor were shown to bind DNA and RNA competitively over 20 years before a physiological role for RNA-steroid receptor interactions was established^141–143. Experimental selection techniques such as Systematic Evolution of Ligands by Exponential Enrichment (SELEX) have been used to develop inhibitory RNA aptamers for DNA binding proteins such as NF-κB. If such inhibitory RNA binding is functionally advantageous, the rapidly evolving sequences of lncRNAs¹⁴⁴ could provide a platform for the evolution of an analogous, endogenous function, and many DRBPs may have species-specific RNA targets. For example, the RNA Lethe, which binds NF-κB, exists only in mice, and is not present even in the closely related rat genome¹⁴⁵.

Proteins that bind both DNA and RNA could have several obvious functional advantages. By binding to both mRNAs and their encoding promoters, DRBPs can exert a powerful, amplified effect on gene expression. This also allows for greater flexibility in generating cellular responses, since these DRBPs could generate rapid effects on protein synthesis as well as impart long-acting changes on gene expression. At a cellular level, using one DRBP rather than two independent DNA- and RNA-binding proteins is more efficient, as it requires the transcription and translation of only one gene product. Finally, competitive RNA and DNA binding by some DRBPs allows for an additional level of transcription factor regulation through RNA decoys. These functional advantages, in addition to the rapid pace at which lncRNAs and their functions are being discovered, strongly indicate that more DRBPs and DRBP-mediated functions will be discovered in the coming years.

Glossary

Rossmann Fold

A common protein folding pattern that contains the topology of a β-α-β fold. It is found in many nucleotide binding proteins

Drosha

Drosha is a nuclease of the RNaseIII family, which as part of a complex cleaves primary (pri-)miRNA transcripts into precursor (pre-)miRNAs in the nucleus

Primary (pri-)miRNA

The primary transcripts of microRNA genes are stem-loop structures pr ocessed by the Drosha complex into precursor (pre-)miRNAs. Typical pri-miRNAs are hundreds of bp-long and can co ntain several pre-miRNAs

Precursor (pre-)miRNA

The product of Drosha-mediated cleavage of primary microRNAs. The stem-loop structured pre-miRNA are exported to the cytoplasm and further processed by Dicer to fully mature miRNAs

MAD Homology 1 (MH1) domain

The MH1 domain is evolutionarily conserved and found in SMAD proteins. It contains four alpha-helices, six short beta-strands, five loops, and recognizes specific DNA sequences

Homeodomain fold

A protein structural domain that binds DNA and RNA and is found most commonly in transcription factors. It consists of a 60 amino acids-long helix-turn-helix structure

GAR domain

A glycine- and arginine-rich motif that adopts a repeated beta-turn structure. Found most commonly in proteins that bind RNA

Zinc Finger domains

A structural motif that is characterized by the coordination of one or more zinc ions. There are a number of different zinc finger motifs and each display different binding modes and structures

WW domains

A small protein motif of 40 amino acids that mediates specific protein-protein interactions with short proline-rich or proline-containing motifs

Ftz-F1 domain

A protein domain first identified in the Fushi tarazu factor 1 (FTZ-F1) nuclear receptor. The domain contains an evolutionarily conserved LXXLL motif that recognizes other LXXLL-related motifs

K Homology domain

A conserved protein domain that interacts with both RNA and DNA through a binding cleft formed between two alpha helices, two beta sheets, and the GXXG loop

Cold shock domain (CSD)

A small, ~70 kDa domain with high similarity to the RNP-1 RNA-binding motif. This domain is found in DNA and RNA binding proteins in prokaryotes and eukaryotes

Interface

The solvent accessible portion of a protein that is capable of binding a ligand – including DNA and RNA – in a competitive or non-competitive manner

π-stacking interactions

A non-covalent interaction between two aromatic molecules due to the attractive force originating from the opposing electrostatic potentials between two adjacent aromatic amino acid residues

Aptamers

A single stranded DNA or RNA that selectively binds small molecules, proteins, and peptides with high affinity. Aptamers have dynamic tertiary structures, which contributes to their diversity of binding targets

Z-DNA

DNA in the conformation of a left-handed double-helix

Z-RNA

RNA in the conformation of a left-handed double helix

Y-box

A DNA target sequence with the consensus CTGATTG, which is recognized and bound by certain proteins

RNA homopolymers

A sequence of ribonucleotides consisting of a single base; for exampl.e, CCCCCCCC

Systematic Evolution of Ligands by Exponential Enrichment (SELEX)

A technique to identify ligand-binding sequence specificity, based sequential rounds of binding of an oligonucleotides library to the ligand, followed by PCR amplification of the bound sequences

Intrinsically disordered proteins

Proteins that do not have a well-ordered, three dimensional structure, such as proteins containing random coils or proteins containing multiple domains connected with flexible linkers

Biographies

William H. Hudson is a Ph. D. candidate in the Molecular Systems Pharmacology Program in the lab of Eric Ortlund at Emory University, Atlanta, GA. Will obtained his B.S. in Biomedical Engineering at the Georgia Institute of Technology. His thesis work aims to characterize how the glucocorticoid receptor, a central transcription factor controlling metabolism and inflammation, interacts with both newly discovered repressive DNA elements and with long intergenic noncoding RNAs,

Eric Ortlund received his Ph. D. from the University of South Carolina, USA, in 2002. Following a postdoctoral fellowship at the University of North Carolina in Chapel Hill, USA, he joined the faculty within the department of Biochemistry at Emory University as an Assistant Professor in 2008. Dr. Ortlund’s lab pursues structural, evolutionary, and biochemical studies of human nuclear receptors, which are ligand regulated transcription factors that play central roles in development, cancer, stress and metabolism. http://www.biochem.emory.edu/ortlund/