The DNA-binding High Mobility Group Box domain of Sox family proteins directly interacts with RNA in vitro

Abstract

There is growing evidence that a substantial number of protein domains ascribed as DNA-binding also interact with RNA to regulate biological processes. Several recent studies have revealed that the Sox2 transcription factor binds RNA through its high mobility group box (HMGB) domain in vitro and in vivo. High conservation of this domain amongst members of the Sox family of transcription factors suggests that RNA-binding activity may be a general feature of these proteins. To address this hypothesis, we examined a subset of HMGB domains from human Sox family proteins for their ability to bind both DNA and RNA in vitro. We observed selective, high affinity interactions between Sox family HMGB domains and various model RNA elements including a four-way junction RNA, a hairpin RNA with an internal bulge, G-quadruplex RNA, and a fragment of the long non-coding RNA ES2, which is known to directly interact with Sox2. Importantly, the HMGB domains bind these RNA ligands significantly tighter than non-consensus dsDNA and in some cases with affinities rivaling their consensus dsDNA sequences. These data suggest that RNA-binding is a conserved feature of the Sox family of transcription factors with the potential to modulate unappreciated biological functions.

Graphical Abstract

Increasingly, proteins previously recognized as only binding DNA have begun to be appreciated for their RNA-binding capabilities which are essential for biological processes.^{1, 2} Several large-scale surveys seeking to identify proteins that indirectly or directly interact with RNA have putatively found a number of DNA-binding proteins not classified as RNA-binding.^3–5 Two classes of DNA-binding proteins frequently observed to be associated with RNA are chromatin remodeling proteins and transcription factors (TFs). In these surveys, high mobility group box (HMGB) DNA-binding domains have been detected across multiple independent studies,^{2, 6–8} leading to the hypothesis that a subset of HMGB domain-containing proteins known to site-specifically bind DNA also recognize and interact with RNA in the cell.

An important HMGB-containing TF found to interact with RNA is Sox2; these interactions are mediated directly by the HMGB domain.^9–18 Sox2 is essential for embryonic development, maintenance of stem cell populations, induction of pluripotent stem cells and is considered an oncogene.^19–22 The interaction of Sox2 with RNA is important for a variety of Sox2-driven biological processes.^{13, 15, 16, 18} Sox2 modulates alternative splicing decisions and also associates with a subset of these alternatively spliced transcripts.^{11, 18} Sox2-RNA interactions are also essential for Sox2 genomic localization; knockdown of Sox2’s interacting RNAs modifies it genomic localization with deleterious impacts on normal cellular processes.^{10, 15} Sox2-RNA interactions within cancer cells impact cellular decisions such as differentiation or by promoting tumorigenesis.^{12, 17} In vitro, Sox2 uses distinct residues when interacting with DNA or RNA and RNA binding is competitive with DNA.⁹ Together, these studies demonstrate Sox2 is an RNA-binding protein (RBP) and these interactions are an essential component of Sox2’s function.

Sox2 is a member of the larger family of Sox proteins, all of which contain the highly conserved HMGB DNA-binding domain, raising the question of whether other Sox protein family members directly bind RNA. In humans, 20 Sox family proteins have been identified, many of which are conserved throughout eukarya.^23–25 While compelling evidence exists for RNA associations by Sox4, Sox9, Sox10, and Sox21^26–29, their ability to directly interact with RNA has not yet been assessed. To address our question, we selected a representative set of HMGB domains from each Sox group (see below) within the Sox family and tested the affinity for consensus dsDNA, non-consensus dsDNA and four model RNAs. We find that most Sox HMGB domains examined directly interact with RNA in vitro with similar or slightly weaker affinities than measured for the consensus dsDNA but much higher than a non-consensus dsDNA ligand. These data strongly suggest that, like Sox2, a significant subset of human Sox proteins may interact with RNA in vivo.

All Sox family proteins possess the ~80 amino acid HMGB DNA-binding domain containing a core DNA recognition element, RPMNAFMVW, that is conserved with nearly 100% identity (cyan box, Figure 1).^{23, 30, 31} Alignment of Sox family HMGB domains shows that all members share at least ~50% sequence identity with the parent Sox protein, sex-determining region Y protein (SRY).³¹ Further, structures of Sox HMGB domains across the family, either in the unbound or DNA-bound states are highly similar, with all comprising three alpha helices forming an L-shaped wedge that inserts residues into the minor groove of B-form DNA to induce bends of ~70°.^{30, 32, 33} Recognition of DNA targets in different contexts is made possible by the flexible and malleable structure of the HMGB domain³⁴—a property also attributed to many RBPs.³⁵

Figure 1.

High conservation of amino acids across the Sox family HMGB domain supports their RNA-binding potential. Sequence alignment of Sox family and Lef1 HMGB domains and their categorization into distinct groups. Sox family core residues are within the cyan box (top left). The asterisk denotes the HMGB domains and sequences used in this study which include two non-native N-terminal GP residues. Red highlighted residues are involved in Sox2 RNA-binding.

Previously, we conducted a thorough alanine scan of the Sox2 HMGB domain, revealing that the DNA- and RNA-binding surfaces of the Sox2 HMGB domain significantly overlap, but residues that are only important for binding either DNA or RNA were also identified.⁹ Residues found to be important for Sox2 binding of RNA are highly conserved throughout the Sox family (red bars, Figure 1), suggesting RNA-binding potential for other Sox HMGB domains.

To support our hypothesis, we surveyed the literature and databases for biological properties that suggest Sox family proteins interact with RNA in vivo (Figure S1). First, evidence that uses experimental approaches that determine direct binding of proteins to RNA was considered, such as UV-crosslinking RNA immunoprecipitation (RIP) in vivo and/or in vitro biochemical binding assays. Evidence for Sox-RNA interactions using these approaches are separate from methods such as formaldehyde crosslinking RIP in vivo or non-crosslinking pull down assays that cannot distinguish between direct and indirect interactions with RNA. This analysis reveals that while Sox1 and Sox2 are the only Sox proteins with evidence for directly interacting with RNA,^{9, 18} Sox4, Sox9, Sox10 and Sox21 at a minimum, indirectly interact with RNA (Figure S1A and B).^26–29

Sox family proteins interact with protein partners on DNA,³⁶ and it is likely they also use protein partners when binding RNA, as evidenced by interactions between Sox2 and the DLX5-SMARCA4-EVF2 lncRNA RNP complex.¹⁰ Like Sox2, six other Sox proteins are components of RNP complexes, RNP condensates, or localize to sites where RNA processing occurs including the nucleolus (Sox15), nuclear speckles (SRY, Sox6, Sox7), paraspeckles (Sox9), lamp brush chromosome lateral loops (Sox9), and stress granules (Sox3) (Figure S1C).^37–42 Sox2’s protein partners are highly enriched with proteins involved in RNA metabolism such as splicing.⁴³ Examination of the protein-protein interaction database APID⁴⁴ supports that Sox2-RNP complexes are an important component of Sox2 biology as ~30% (human) and ~20% (mouse) of Sox2’s known interacting protein partners are labeled as RBPs (human shown in Figure S1D). Using APID, thirteen other Sox proteins interact with proteins known to bind RNA (Figure S1D), suggesting Sox family proteins are components of RNP assemblies. Collectively, this supports our hypothesis that Sox family proteins bind RNA, although, these data do not demonstrate direct Sox-RNA interactions.

To determine whether human Sox family proteins directly bind RNA in vitro and thereby have the potential to directly interact with RNA in vivo, a subset of Sox-family HMGB domains were selected for examination. Human Sox family proteins are divided into nine distinct groups based upon sequence conservation within the HMGB domain and within each group the HMGB domain residues are at least ~70% conserved (Figure 1).^{23, 31} For this study, we chose nine human Sox HMGB domains—one HMGB domain from each Sox group (denoted by asterisks, Figure 1). To ensure that each protein in this analysis comprised the full HMGB domain, the sequence of each domain was defined by UniProt⁴⁵ and at least three additional residues on the N- and C-termini were added (sequences of proteins and UniProt ID given in Table S1).

To determine if a non-Sox family sequence specific DNA-binding HMGB domain can bind RNA, we included Lef1 in this study.⁴⁶ Lef1’s HMGB domain is similar in sequence and structure to the Sox family members HMGB domain (Figure 1), but is part of the TCF/Lef1 family of sequence specific TFs and their HMGB domains are nearly 100% sequence identical (Figure S2).⁴⁷ Previously, TCF7 has been shown to directly interact with RNA in vitro through its HMGB domain⁴⁸, suggesting Lef1 and the other TCF proteins may also directly interact with RNA through their HMGB domain.

We first validated the ability of the selected HMGB domains to bind both site-specific and non-site-specific dsDNAs by determining apparent dissociation constants (K_{D, app}) for each protein-nucleic acid complex using fluorescence anisotropy (FA) (Figure 2). We further used protein-binding stoichiometric electrophoretic mobility shift assay (EMSA) (Figure 2), which are completed with nucleic acid ligand concentrations greater than the K_{D, app}. The HMGB domains of nine Sox-family proteins and Lef1 were expressed as a maltose binding protein-HMGB domain fusion, which were cleaved with 3C protease and purified to homogeneity (Figure S3). Sox family proteins bind the same core consensus DNA motif (5’-TTGT) while allowing variations in recognition of the flanking base pairs immediately 5’- and 3’- to the core.^{30, 49} This core DNA motif is confirmed for multiple Sox proteins in vivo.^50–52 However, many of these Sox HMBG domains have not been assessed using in vitro methods to determine their affinity for the core consensus dsDNA motif.

Figure 2.

Sox17 binds dsDNA. A. FA assay with dsDNA containing the Sox consensus motif, 5’ TTGT dsDNA. B. EMSA to measure protein:DNA stoichiometry with dsDNA containing the Sox consensus motif. C. FA assay with dsDNA without Sox consensus sequence. D. EMSA to measure protein:DNA stoichiometry with dsDNA without the Sox consensus motif. Fit FA binding curves of Sox17 HMGB domain against each ligand with K_{D, app} obtained from first transition, N=at least 3, standard deviation reported.

The Sox2 FGF4 enhancer sequence containing the core TTGT sequence was used for all Sox proteins^{53, 54} and a separate consensus dsDNA sequence of 5’-CCTTTGAA was used for Lef1.^{46, 55}

Representative binding curves for the Sox17 HMGB domain are shown in Figure 2A. All the HMGB domains bound the consensus dsDNA ligand with affinities ranging from 0.6 to 72 nM (Table 1), consistent with reports for those previously tested (Table S2).^{9, 49, 54, 56–59} Almost all were in the 1 nM range, with only two family members (Sox6 and Sox30) being significantly weaker. In the consensus dsDNA FA titrations, two distinct transitions were observed that correspond to 1:1 and 2:1 protein:dsDNA complexes. This was established by correlating each transition observed by FA to a protein-induced shift in an EMSA (Figure 2B and Figure S4); this behavior has been previously observed.^{9, 59} The first transition corresponds to high affinity binding to the core consensus sequence and is the K_{D, app} reported in Table 1 for all HMGB domains. The second transition likely results from a lower affinity binding event corresponding to a second protein binding with K_{D2, app} values given in Table S3. The total anisotropy change in our FA experiments are consistent with previous reports of proteins binding FAM labeled nucleic acids.^{60, 61} Importantly, these data demonstrate that most Sox family HMGB domains bind the core 5’-TTGT sequence with highly similar in vitro affinities.

Table 1.

Binding affinities of Sox- and Lef1-HMGB domains against the nucleic acid ligands.

HMGB domain	Nucleic Acid Ligands, KD, app, nM
	Consensus dsDNA	Non-consensus dsDNA	ES2 lncRNA fragment	Stem loop w/ internal bulge	RNA 4-way junction	TERRA G-quad
SRY	0.6 ± 0.3	840 ± 130	6.9 ± 0.6	20 ± 1	4.6 ± 0.5	130 ± 80
Sox2	0.6 ± 0.1	600 ± 20	2.8 ± 1.2	16 ± 0.8	5.2 ± 0.3	0.7 ± 0.5
Sox6	31 ± 15	1200 ± 300	72 ± 7	290 ± 15	96 ± 4	34 ± 5
Sox9	0.6 ± 0.2	970 ± 390	4.2 ± 2.0	28 ± 7	8.4 ± 0.1	1.2 ± 0.5
Sox11	4.4 ± 1.9	1200 ± 600	100 ± 16	360 ± 30	68 ± 9	1.9 ± 1.3
Sox15	1.2 ± 0.8	1200 ± 800	12 ± 6	260 ± 7	70 ± 11	2.1 ± 2.5
Sox17	0.5 ± 0.2	310 ± 40	1.5 ± 0.7	6.9 ± 3.8	1.6 ± 0.2	0.2 ± 0.2
Sox21	0.5 ± 0.2	1200 ± 300	3.2 ± 0.5	79 ± 17	23 ± 7	3.1 ± 2.5
Sox30	72 ± 12	≥5000	910 ± 130	3100 ± 500	990 ± 120	460 ± 100
Lef1	1.6 ± 0.3	1000 ± 300	61 ± 8	630 ± 120	89 ± 8	4.1 ± 3.2

Note: N=at least 3 with standard deviation reported.

The binding affinity for Sox17 (Figure 2C) and the other HMGB domains to dsDNA that does not contain the consensus core motif was measured to assess the specificity for core consensus dsDNA and an EMSA confirms this qualitatively weaker affinity (Figure 2D and Figure S5). All HMGB domains bound the non-consensus dsDNA with measured K_D, _apps of 300 – 1200 nM (Table 1), reflecting a >1000-fold greater affinity of consensus DNA for most domains. The exception was Sox6 which only discriminates by ~40-fold; this lower specificity could be due to lower affinity for the FGF4 enhancer sequence compared to the other Sox HMGB domains. Differences in affinity when bases at the 5’ and/or 3’ ends of the core TTGT motif are altered, were observed for SRY,⁵⁷ which may explain Sox6’s observed lower affinity for FGF4.

As evidence for Sox proteins acting as RNA binders in vivo continues to grow, it is important to investigate if there are any family-wide RNA preferences. Since potential high affinity RNA-binding sites are unknown for Sox family HMGB domains, we developed a set of four model RNAs representing various structural and sequence motifs to assess binding (sequence and secondary structures shown in Figure S6 and Table S4). We selected a fragment of the ES2 lncRNA previously shown to bind the Sox2 HMGB domain with high affinity in vitro⁹ while the full length ES2 lncRNA associates with Sox2 in vivo.¹⁴ In addition, we previously found that the Sox2 HMGB domain bound small duplex and hairpin RNAs.⁹ Another study identified stem-loop RNA targets of Sox2 using SELEX.¹⁸ Other HMGB proteins directly bind similar RNAs, including TCF7’s interaction with double stranded RNA with an internal bulge⁴⁸, HMGB1’s interaction with hairpin RNA⁶², and HMGD binding dsRNA stem loops containing an internal bulge.⁶³ Thus, we included a small hairpin RNA with an internal bulge in this survey. Other HMGB proteins directly interact with RNA 4-way junctions^{64, 65}, and therefore an RNA 4-way junction substrate was included. Finally, Sox2, Sox21, and other HMGB proteins directly and/or indirectly interact with G-quadruplex RNA.^{29, 66} Thus, the well-characterized TERRA G-quadruplex RNA was used in this study.⁶⁷ The RNAs used in this study were 3’ fluorophore labeled and purified to homogeneity (Figure S7).

These RNA ligands were each assessed for binding to the panel of Sox family and Lef1 HMGB domains using the FA binding assay. As a representative, the Sox17 HMGB domain binding with the four RNAs is shown in Figure 3. The K_{D, app} was measured by FA for each RNA under the same conditions as the previously described dsDNAs (Figure 3A, C, E, G) and EMSAs were used to assess whether multiple Sox17 HMGB domains bind the RNA ligands (Figure 3B, D, F, H). Sox17 binds these four RNAs with affinities ranging from 0.2 – 6.9 nM and multiple transitions are observed in the FA assays for all RNA ligands (Figure 3A, C, E, G). This is likely due to multiple Sox17 binding sites with a variety of affinities (Figure 3B, D, F, H). For example, at a 2:1 Sox17:lncRNA ES2 fragment ratio, three bound bands are present (Figure 3B and Figure S8). The RNA 4-way junction appears to accommodate up to four Sox17 molecules at a 3.25:1 ratio (Figure S9) and at a 0.25:1 ratio a second band is already present, suggesting these binding sites may have similar affinities (Figure 3F, S9). In the case of the smaller stem loop with internal bulge RNA and TERRA G-quadruplex, at 2:1 (Figure 3D, H) and 3:1 (Figure S10, S11) ratios respectively, the RNAs still only form two predominant species suggesting the RNAs can accommodate two HMGB domains or that the other potential sites are very low affinity. While the TERRA RNA is homogenous under denaturing conditions (Figure S7), under native conditions there is one predominate species and a lesser populated species (Figure 3H, Figure S11). This could be a result of unfolded and folded states, TERRA’s ability to bind either K⁺ and Na⁺ ions,^{69, 70} or to dimerize.⁷¹ Interestingly, Sox17 appears to prefer the predominate TERRA species, further suggesting Sox17’s selectivity of certain RNA features. Thus, Sox17 appears to be able to bind to a variety of RNA features with high affinity and to distinct sites within the same RNA.

Figure 3.

Sox17 binds various RNA ligands with high affinity. A. FA assay with lncRNA ES2 fragment. B. EMSA to determine protein:RNA stoichiometry with lncRNA ES2 fragment. C. FA assay with stem loop with internal bulge. D. EMSA with stem loop with internal bulge. E. FA assay with RNA 4-way junction. F. EMSA with RNA 4-way junction. G. FA assay with TERRA G-quadruplex RNA H. EMSA with TERRA G-quadruplex. Sfold⁶⁸ predicted secondary structures of nucleic acid ligands used in this study are overlayed into each binding curve. Fit FA binding curves of Sox17 HMGB domain against each ligand with K_{D, app} obtained from first transition, N= at least 3, standard deviation reported.

We also observed that almost all the HMGB domains directly interact with RNA ligands with a wide range of nM affinities (Table 1). These affinities are consistent with previous reports of Sox2’s affinity for RNA ligands, 10–100 nM,⁹ and are also within the range observed for other HMGB proteins, such as TFAM’s affinity to G-quadruplex RNA of ~1 nM.⁶⁶ Importantly, most of the affinities for RNA were significantly tighter than those observed to non-consensus dsDNA and, in many cases, approximately the same or comparable as to the consensus dsDNA containing the Sox core motif (Table 1). The one exception to our observed general pattern of binding was Sox30, which displayed lower affinities for all RNAs tested in this study. It is not clear why Sox30 does not bind RNA with high affinity, but it should be noted that it is one of the least conserved members of the Sox family and is missing some of the conserved positively charged residues within the Sox HMGB domain that contribute to nucleic acid binding (Figure 1).

In addition to high affinity RNA binding, the HMGB domains displayed varying degrees of selectivity between the RNA substrates. Binding affinities of all RNA ligands for each HMGB domain from Table 1 were compared to each other and ranked from highest affinity (1) to lowest affinity (4) (Figure 4). This representation of data from Table 1 allows a direct and straightforward visualization to identify HMGB domains that exhibit divergent preferences amongst the RNA substrates. Most of the HMGB domains bind the TERRA G-quadruplex RNA substrate with the highest affinity. Selective recognition of this motif by Sox HMGB domains suggest interactions could occur in vivo. In contrast, the majority of the HMGB domains bound the stem-loop with an internal bulge the weakest except for SRY and Sox15. However, most Sox HMGB domains showed a strong preference for the longer ES2 lncRNA hairpin which contains several internal loops, suggesting that more complex RNA structures might be targeted by these proteins. As Sox proteins have diverse, yet distinct and highly specific tissue expression with separate genomic occupancies to regulate gene expression, differences in Sox-HMGB RNA selectivity may be important to promote RNA-binding in each Sox protein’s native cellular environment with their distinct transcriptomes, which merits further exploration.

Figure 4.

HMGB domains show selectivity for RNA ligands. Each protein’s affinity for the RNA ligand is ranked from highest affinity ligand to lowest affinity ligand using data from Table 1. SL w/ IB is a Stem Loop with an Internal Bulge.

The driving factors for selectivity are not yet known but likely arise from differences in each Sox protein’s HMGB domain that favor RNA structural features and/or sequence preferences. As our small RNA library was not normalized by size and electrostatics, known to be important factors in Sox HMGB-nucleic acid interactions, it is possible the size of the RNA impacts affinity by presenting multiple binding sites. For example, many of the HMGBs bound the smaller stem loop with internal bulge RNA (35 bases) with weaker affinities than to the more complex ES2 lncRNA (89 bases). However, the highest affinity RNA ligand overall was the TERRA G-quadruplex (27 bases). Together, these data suggest that RNA size is not the primary factor driving high affinities and RNA structure and/or sequence preferences dominate.

This survey of Sox-family and Lef1 HMGB domains’ ability to bind RNA in vitro serves to advance a developing understanding of DNA-binding domains that directly interact with RNA. Clear evidence for the Sox family as an RNA-binding family has emerged from several studies that indicate multiple Sox proteins directly and/or indirectly with RNA in vivo, localize to RNP complexes, and interact with RBPs. This study furthers this understanding by providing compelling evidence that nine Sox HMGB domains and the Lef1 HMGB domain directly interact with a variety of RNA substrates with affinities comparable to that of their consensus dsDNA targets. Notably, all the HMGB domains exhibit selectivity for distinct RNA ligands which could play a role in binding specific RNAs in the context of each protein’s cellular environment. While this study does not demonstrate RNA interactions in vivo or that these direct interactions impact Sox/Lef1 family driven biology, the collective evidence presented herein provides motivation for further exploration of unappreciated aspects of the biological functions of HMGB proteins.

ABBREVIATIONS

FA

Fluorescence anisotropy

FGF4

Fibroblast growth factor 4

HMGB

High mobility group box

RIP

RNA immunoprecipitation

RNP

Ribonucleoprotein particle

Sox

Sex-determining region Y-Box

TF

Transcription factor