DNA binding and dimerization of the SOG1 NAC domain are functionally linked with its ability to undergo liquid–liquid phase separation

Abstract

Liquid–liquid phase separation is a key phenomenon in the regulation of transcription in eukaryotes, leading to the formation of so-called membraneless organelles. While transcription factors take part in several types of membraneless organelles, it remains unclear how specific DNA binding, multivalent interactions with DNA/RNA, and condensation are interlinked. Here, we show that the NAC domain of suppressor of gamma response 1 (SOG1) (SOG1^NAC), a transcription factor that is central to the DNA damage response in plants, can undergo liquid–liquid phase separation in vitro in the presence of both RNA or double-stranded DNA. This behaviour and the ability of SOG1^NAC to bind DNA in a sequence-specific manner are dependent on its potential to form homodimers and the presence of a cluster of positive charges in its DNA-binding site. Short double-stranded DNA fragments containing the sequence motif that is specifically recognized by SOG1^NAC inhibit RNA-mediated phase separation, suggesting overlapping binding sites for DNA and RNA. This may reflect a complex interplay between DNA and RNA binding that could control the formation of condensates at transcription sites.

Graphical Abstract

Graphical Abstract.

Introduction

Many cellular functions such as RNA metabolism and processing, ribosome biosynthesis, gene regulation, and stress responses involve the formation of membraneless organelles (MLOs) [1–5]. The latter are biomolecular condensates of specific macromolecular components that are formed via a process of liquid–liquid phase separation (LLPS). MLOs are typically dynamic in composition with fast exchange of components with their surroundings. The entropic cost associated with the formation of MLOs is overcome by favourable multivalent interactions between the biomolecules they consist of [1, 6].

MLOs are implicated in transcription regulation during which many macromolecules, including transcription factors and RNA species, cluster at specific genomic locations (for a review, see [7]). In general, eukaryotic transcription factors recognize and bind specific DNA motifs of 5–15 base pairs (bp) in promoter or enhancer sequences with their folded DNA-binding domains [8]. Other components of the transcription machinery like coactivators, regulatory cofactors, and RNA polymerase II are recruited by multivalent protein–protein interactions with the intrinsically disordered transactivation domains of Eukaryote transcription factors (eFTs) [9, 10]. Henninger et al. proposed that the formation of transcription condensates is controlled by an RNA-mediated feedback mechanism [11]. Low concentrations of RNA produced during transcription initiation promote condensation, while the burst production of messenger RNA (mRNA) during elongation leads to dissolution of the condensates [8, 10, 11]. Research on LLPS of transcription factors has mostly focused on their intrinsically disordered domains. The latter often harbour low complexity regions that create multivalency and are more than folded domains prone to post-translational modifications. Both these properties haven proven to be key in LLPS (for reviews, see [12, 13]. Also folded domains, and especially RNA- and DNA-binding domains of LLPS-prone proteins, have been observed to influence the formation of MLOs [14–16]. However, the link between LLPS and the ability of folded nucleic acid binding domains to specifically or non-specifically bind DNA or RNA is poorly understood.

Suppressor of gamma response 1 (SOG1) belongs to a large plant-specific family of NAC [no apical meristem (NAM), Arabidopsis transcription activation factor (ATAF), cup-shaped cotyledon (CUC)] transcription factors that function in growth and development or respond to abiotic and biotic stresses [17, 18]. Among these, SOG1 activates more than 300 genes involved in DNA repair, cell cycle arrest, or apoptosis in response to DNA damage in the model plant Arabidopsis thaliana [19, 20]. The protein consists of a conserved, folded DNA-binding domain of 155 amino acids (AAs) (called the NAC domain) followed by a highly divergent C-terminal domain that is intrinsically disordered [21, 22]. SOG1 contains an additional 57 AA extension of unknown function that is lacking in most other NAC transcription factors [19, 23]. The transcriptional activation function of SOG1 is controlled by the phosphorylation of five Serine-Glutamine (SQ) sites by the DNA damage sensing kinases Ataxia-Telangiectasia Mutated (ATM) and Ataxia and RAD3-related (ATR) and by phosphorylation of a single casein kinase 2-dependent site [24–27].

In this study, we investigate the in vitro DNA binding of the folded NAC domain of Arabidopsis thaliana SOG1 (which will be called SOG1^NAC from now on). Next, we link its specific double-stranded DNA (dsDNA) binding to its potential to form functional homodimers and its ability to undergo DNA- and RNA-driven LLPS. In summary, we investigate the putative correlation between SOG1^NAC dimerization, specific and non-specific DNA/RNA binding, and the potential for phase separating behaviour. We conclude that the NAC domain of SOG1 undergoes LLPS in the presence of RNA and dsDNA, and that this potential requires an intact DNA-binding site and the formation of dimers.

Materials and methods

Proteins and nucleic acids

The SOG1 AA sequence used in this paper corresponds to UniProt accension code Q6NQK2. The SOG1^NAC domain ranges from L58 to Q212. The structure of SOG1^NAC in complex with the pBRCA1 31 bp oligo was predicted using AlphaFold3 [28].

An overview of the DNA sequences that were used in the various experiments of this paper can be found in Table 1. The single-stranded DNA (ssDNA) oligos of these sequences (and their reverse complements) and the used mRNA proxy called polyadenylic acid potassium salt (poly-A, CAS number: 27416-86-0) were obtained from Sigma–Aldrich. Arabidopsis thaliana mRNA was kindly provided by the Center for Plant Systems Biology (VIB, Technologiepark 71, 9052 Zwijnaarde, Belgium).

Table 1.

Sequences of the DNA oligos used in this paper 

Name	Sequence (5′--3′)
*pBRCA1 99 bp oligo	GGTATATAGATAGATAGATGACGAAGCGTGTCCTTCAAAGAGGCTTCC-
	CTTGAAAGCTTTACAGACTAAATAAAAAGCGCCAAATTTTCATCTACTTCA
*pBRCA1 49 bp oligo	GCGTGTCCTTCAAAGAGGCTTCCCTTGAAAGCTTTACAGACTAAATAAA
*pBRCA1 31 bp oligo	TCAAAGAGGCTTCCCTTGAAAGCTTTACAGA
*pBRCA1 27 bp oligo	AAAGAGGCTTCCCTTGAAAGCTTTACA
*pBRCA1 23 bp oligo	AGAGGCTTCCCTTGAAAGCTTTA
*pBRCA1 17 bp oligo	GGCTTCCCTTGAAAGCT
*pBRCA1 mutated	TCTGTAAAGACGTCAAGGGCGACCTCTTTGA
Thirty-one base pairs oligo
*Randomly generated, non-target 49 bp oligo	TAGTGCGCTAATCTGAGTCGAAATAAAATCACCAGTACCCAAAACCAGGG
*Randomly generated, non-target 31 bp oligo	GTCGGCGATGGTGGTAACTAATTATGTTCCT
*Designed oligo 1	TCTGTAAAG CTT TCAAGGG AAG CCTCTTTGACGTAACAGAAACCCCG-
	GAGTAAGCTCCCGTGGGTCTGTAAAGCTTTCAAGGGAAGCCTCTTTGA
*Designed oligo 2	TCTGTAAAG CTT TCAAGGG AAG CCTCTTTGACGTAACAGAAACCCCG-
	GAGTAAGCTCCCGTGGGGTCGGCGATGGTGGTAACTAATTATGTTCCT
*pRAD51 31 bp oligo	TAATCGAGACTTGTTGAAGAAGCCTTTGCCC
*pRAD51 mutated	TAATCGAGATCCGTTGAAGGGACCTTTGCCC
Thirty-one base pairs oligo
*pSMR5 97 bp oligo	TACCTAGCAACTTCCAGAACAAGCTACGGTCACTTCCAAAACTAGTGT-
	AATAAAG AAG CAATCTAATTAAGAACCTTATTAGGAAAGTGCGAGAGAG
*pSMR5 site 1(RC)	TGACCGTAGCTTGTTCTGGAAGTTGCTAGGT
*pSMR5 site 2 (RC)	TTAGATTGCTTCTTTATTACACTAGTTTTGG
*pSMR5 site 3	ATTAAGAACCTTATTAGGAAAGTGCGAGAGA
*pXRI1 31 bp oligo	ATAAAAAGCTTCTCTTGCAATTCTCTCTTCC
*pTKa1 31 bp oligo	TGACAAATCTTCTCCAGCAATTTTCTTCCTT
pBRCA1 310 bp FW	AGTAATGGGTTCAAATCTTGC
pBRCA1 310 bp RV	TTCTCTTCCCATCCTCTCC
pUBQ10 293 bp FW	CTTTCTCTCAATTCTCTCTACC
pUBQ10 293 bp RV	CACAAACTAGAAACTAACACC

*ssDNA oligo of this sequence was annealed with its reverse complement to obtain a dsDNA oligo of a specific length in base pairs (bp) mentioned in the name.

The six CTT-AAG nucleotides of the CTT(N7)AAG ‘Ogita motif’ or variations of these nucleotides are highlighted in bold text. Mutated nucleotides are written in italic. The reoccurring 31 bp oligos present in the two ‘Designed oligos’ are underlined. The three 23 bp potential binding sites within the pSMR5 97 bp oligo are also underlined. RC = reverse complement, to represent the binding motifs more clearly.

Cloning of the SOG1^NAC construct and its mutants

The gene sequence of SOG1^NAC preceded by a 6x His-tag was synthetically produced and cloned into the pET-11d vector via restriction–ligation (NcoI and BamHI) by Genscript. Site-specific mutagenesis was performed on this clone to obtain the three SOG1^NAC mutant clones: SOG1^NACΔ1–6, SOG1^NAC-RKRR, and SOG1^{NAC-DNAstrand}. In contrast to wild type (WT) SOG1^NAC, the SOG1^NACΔ1–6 mutant protein lacks the first six AAs (ΔL58-K63). The SOG1^NAC-RKRR mutant contains following four mutations: R79A, K80A, R81A, and R82A. The SOG1^{NAC-DNAstrand} mutant harbours following five mutations: R93S, H95S, T97A, R99S, and T100S. To obtain these three mutant clones, specific mutant primer sets were designed (Supplementary Table S1). The mutated reverse primers were used in combination with a forward primer containing an XbaI restrictive site (‘XbaI FW’) to amplify the sequence of the SOG1^NAC clone upstream of the mutations with homologous regions of the pET-11d vector. The mutated forward primer was used in combination with a reverse primer with a mutated BamHI restrictive site to Acc651 [‘BamHI (to Acc651 mutated) RV’] to amplify the sequence of the SOG1^NAC clone downstream of the mutations with homologous regions of the pET11d vector. Via overlap PCR using primers XbaI FW and BamHI (to Acc651 mutated) RV, the whole coding sequences of the mutants with a 6x His-tag were obtained. Subsequently, the construct was cloned into the pET-11d vector at the XbaI and BamH1 site via Gibson Assembly. Positive colonies were selected with restriction enzyme Acc651 and correct insertion was confirmed by Sanger sequencing.

Expression and purification of the SOG1^NAC domain and its mutants

Escherichia coli BL21(DE3) cells, transformed with a pET-11d vector carrying the gene sequence of SOG1^NAC or its mutants with an N-terminal 6x His-tag, were inoculated in 1 l Luria Broth (LB) medium with ampicillin (100 µg/ml) and grown at 37°C until an OD₆₀₀ of 0.6–0.8 was reached. Protein expression was induced by adding 0.5 mM isopropyl β-d-1-thiogalactopyranoside and the cultures were incubated at a temperature of 16°C for WT SOG1^NAC and SOG1^NACΔ1–6 mutant and 23°C for the SOG1^NAC-RKRR and SOG1^{NAC-DNAstrand} mutants. Cells were collected by centrifugation (5000 × g, 30 min) the next morning, except for the SOG1^{NAC-DNAstrand} mutant for which 2 h of incubation was sufficient for optimal expression. Pellets of 2 l bacterial culture were resuspended in 30 ml lysis buffer (50 mM Tris pH 7.5, 500 mM NaCl, 10% glycerol, 2 mM imidazole, 2 mM β-mercaptoethanol) supplemented with ½ protease inhibitor tablet [ethylenediaminetetraacetic acid (EDTA) free cOmplete™ ULTRA tablets, Roche]. Resuspended pellets were used immediately for purification or were flash frozen and stored at −80°C until use.

After thawing, the pellets were incubated for 20 min rotating at room temperature after adding 2 mM MgCl₂ and 50 µg/ml DNase. Hereafter, the cells were lysed by sonication for 5 min with 5 s ON/5 s OFF cycles at 80% amplitude using a VCX-70 Vibra Cell (Sonics). Next, samples were centrifuged at 35 000 × g for 45 min at 4°C. The supernatant was filtered (0.45 µm) and loaded on a 1 ml HisTrap™ HP column (Cytiva) that was equilibrated prior with the lysis buffer. After loading the sample, the column was washed with lysis buffer. When the UV₂₈₀ stabilized, a step gradient over 6 column volumes (CV) of 2%, 6 CV of 50%, and 6 CV of 100% elution buffer (50 mM Tris, pH 7.5, 500 mM NaCl, 10% glycerol, 1 M imidazole, 2 mM beta ME) was applied to elute the bound proteins. The eluted fractions containing the protein of interest (POI) were loaded on a Superdex 75 16/90 column (GE Healthcare) equilibrated with gel filtration buffer (20 mM NaPi, pH 7.5, 150 mM NaCl, 1 mM Tris(2-carboxyethyl)phosphine (TCEP)). Peak fractions of the POI were pooled, flash frozen, and stored at −80°C.

Circular dichroism

Protein samples were buffer exchanged to 10 mM NaPi (pH 7.5) by dialysis. Circular dichroism (CD) spectroscopy of the POI samples was performed on a Biologic MOS-500 spectropolarimeter using a 1-mm quartz cuvette (Hellma) at a POI concentration of ~0.25 µg/µl. Spectra were collected at 25°C and at wavelengths ranging from 190 to 260 nm in steps of 1 nm. The buffer spectrum was subtracted from the raw CD data (both θ in mdeg) and the resulting CD spectra were normalized to molar ellipticity ([θ] in deg cm² dmol⁻¹ res⁻¹) using following formula:  with MW the molecular weight in Da, C the concentration in mg/ml, l the cuvette path length in cm, and N the number of AAs.

Analytical size-exclusion chromatography

Analytical size-exclusion chromatography (SEC) was performed on a Superdex 75 Increase 10/300 column (GE Healthcare) equilibrated with 20 mM NaPi (pH 7.5), 150 mM NaCl, and 1 mM TCEP. For each POI, the column was loaded with 500 µl of protein at a concentration of ~2 mg/ml. In a separate run, 200 µl of the Bio-Rad Gel Filtration Standard was loaded under the same conditions as those of the protein samples under investigation. Based on the elution volumes of the proteins present in the standard, the apparent molecular weights (MWs) of the proteins in solution were determined. The apparent MW has the following linear relationship with the K_av retention factor [29]:

The latter is determined by following formula: with V_e the elution volume, V₀ the void volume, and V_t the total column volume. For each SEC set-up, the exact linear relationship is defined by plotting the logarithm of the MW values of the proteins in the standard (bovine thyroglobulin = 670 kDa, bovine γ-globulin = 158 kDa, chicken ovalbumin = 44 kDa, horse myoglobin = 17 kDa, and vitamin B12 = 1.35 kDa) as a function of its K_av value and fitting a linear regression curve (using the least square method) through these data points. Finally, using the equation of the regression curve, the apparent MWs of the proteins in solution could be determined based on their elution volumes.

Dynamic light scattering

Dynamic light scattering (DLS) measurements were performed using a DynaPro NanoStar (Wyatt) device and the data were analysed using the DYNAMICS 7.10.1.21 software. Growth of droplets was followed by adding poly-A to spin-filtered SOG1^NAC to induce LLPS. A reaction with following reaction conditions was set up in triplicate: 20 µM protein and 25 µg/ml poly-A in 9.1 mM phosphate buffer, 68 mM NaCl, and 0.45 mM TCEP. Every 5 min, five acquisitions of 8 s were measured for 4 h at room temperature.

Electrophoretic mobility shift assay: regular and ³²P-labelled

Unlabelled dsDNA fragments used during electrophoretic mobility shift assay (EMSA) experiments were obtained by hybridizing two complementary ssDNA oligos for 10–15 min at 99°C and subsequently letting the sample cool down slowly to room temperature (duration ~1.5 h). Varying concentrations of POI and hybridized dsDNA at 2 µM were mixed in 20 mM sodium phosphate (pH 7.5), 150 mM NaCl, and 1 mM TCEP. Binding reactions were carried out in a total volume of 10 µl at 25°C for 25 min. Hereafter, 2 µl of loading dye (25% ficoll, 0.1% xylenexyanol, and 0.1% bromophenol) was added to each sample. Samples were run on a 13% native polyacrylamide gel prepared with 1× TBE (89 mM Tris Base, 89 mM boric acid, 2.5 mM EDTA) that was pre-ran at 100 V for 30 min. Electrophoresis was performed at 120 V until the fastest migrating dye reached two-third of the length of the gel. Gels were stained for 15 min with ethidium bromide or GelRed^® for DNA visualization.

For radioactive EMSAs with SOG1^NAC, a (5′-³²P) single-end-labelled 99 bp DNA fragment of pBRCA1 was used. The ssDNA oligo of this fragment was first labelled with [γ ³²P]-ATP (PerkinElmer) by T4 polynucleotide kinase (Fermentas). Next, dsDNA was obtained by hybridizing the labelled ssDNA oligo with the complementary oligo using the method described previously. Subsequently, the hybridized, labelled, dsDNA fragment was purified from gel (6% polyacrylamide gel electrophoresis) as described [30]. Varying concentrations of SOG1^NAC and labelled DNA at 15 000 counts per minute were mixed in 20 mM sodium phosphate (pH 7.5), 1 M TCEP, and 50, 150, 300, or 500 mM NaCl. Binding reactions were carried out in a total volume of 20 µl at 25°C for 25 min. After the addition of 3 µl loading dye (25% ficoll, 0.1% xylenexyanol, and 0.1% bromophenol), samples were run on a 10% polyacrylamide gel prepared with 1× TBE (89 mM Tris Base, 89 mM boric acid, 2.5 mM EDTA) that was pre-ran at 100 V for 30 min. Electrophoresis was performed at 120 V until the fastest migrating dye reached the bottom of the gel. X-ray sensitive films were used for visualization.

Isothermal titration calorimetry

Isothermal titration calorimetry (ITC) measurements were carried out on a MicroCal PEAQ-ITC system (Malvern) at 25°C and in 20 mM NaPi pH 7.5, 150 mM NaCl, and 1 mM TCEP. The DNA samples (obtained in the same way as for the unlabelled EMSA experiments) were loaded into the sample cell at concentrations ranging between 6 and 9 µM. Protein samples were loaded into the syringe at concentration ranging between 155 and 190 µM. For the measurements, a reference power of 10 µcal/s and a stirring speed of 750 rpm were used. Each measurement entailed 20 injections of 2 µl with 150 s intervals, preceded by a first test injection of 0.4 µl. Prior to the start of each run, an initial delay of 300 s was included. For data integration and determining the best fitted binding model for the resulting binding isotherms, the MicroCal Peaq-ITC Analysis Software (version 1.41, Malvern Panalytical) was used. In all cases, the ‘one set of sites’ model was used for fitting the binding isotherms, assuming the presence of a single binding event DNA(duplex) + SOG1(dimer) → DNA(duplex)-SOG1(dimer). The equations describing this model are provided in the Malvern Microcal PEAQ-ITC user manual (https://www.malvernpanalytical.com/en/learn/knowledge-center/user-manuals/man0573en). Stoichiometry and thermodynamic parameters were derived as the average of triplicate measurements, with errors determined as the mean deviation from the respective means.

Turbidity assay

Absorbance of mixed protein and poly-A samples in total volume of 30 µl was measured in triplicate at 390 and 600 nm in non-binding 384 black well plates with transparent bottom (Greiner bio-one, µClear^®). Although 600 nm is commonly used to monitor LLPS, a shorter wavelength of 390 nm was selected to enhance sensitivity for detecting smaller condensates. Measurements were performed on a SynergyTM Mx plate reader at room temperature in 9.1 mM phosphate buffer, 68 mM NaCl, and 0.45 mM TCEP. Phase separation was induced by addition of poly-A right before starting the measurement for minimum 30 min.

Microscopy

Droplet formation was visualized using a Leica DMi8 microscope connected to a Leica DFC7000 GT camera. Reactions of 25 µl in volume (25 mM phosphate, pH 7.5, 75 mM NaCl, 0.5 mM TCEP or 9.1 mM phosphate, pH 7.5, 68 mM NaCl, 0.45 mM TCEP) with fluorescently labelled proteins and/or nucleic acids and unlabelled molecules in excess were prepared in non-binding 384 black well plates with transparent bottom (Greiner bio-one, µClear^®). Prior to visualization with an 100× immersion oil objective, the plates were incubated on ice for 10 min. The SOG1^NAC protein and its mutants were labelled with Dylight 488 (Thermofisher scientific™) according to the manufacturer’s protocol. Poly-A was labelled with Cy5 using pCp-Cy5 (Jena Bioscience) and T4 RNA ligase (Thermofisher scientific™). Cy5-labelled DNA oligos were obtained from Integrated DNA Technologies. Green fluorescence was detected by using a fluorescein isothiocyanate filter. Red fluorescence was detected by a rhodamine filter. All experiments were performed in duplicate.

Fluorescence recovery after photobleaching

Sample preparation is identical to the one described above for the microscopy experiments. Firstly, 10 pre-bleaching images were taken every 50 ms. Next, bleaching was performed for 50 ms with a laser at 488 nm with 100% intensity. Finally, post-bleaching images were recorded each second for 2–3 min. Fluorescence recovery after photobleaching (FRAP) was conducted on five droplets containing SOG1^NAC with either poly-A, mRNA, or BRCA1 promotor sequence (310 bp).

Biolayer interferometry

Biolayer interferometry (BLI) competition assays using nucleic acids, DNA and RNA, and POI bait were performed using an Octet 96 Red instrument in 10 mM phosphate (pH 7.5) 75 mM NaCl, and 0.5 mM TCEP at room temperature while shaking at 1000 rpm. Ni-NTA biosensors (Satorius) were loaded with 5 µg/ml of 6x His-tagged SOG1^NAC to attain a shift between 4 and 4.5 nm, followed by a baseline in buffer for 120 s. Next, association of 0.5 µM poly-A for 240 s and a second association with different concentrations BRCA1 or random 31 non-target oligo was performed. Fitting of k_off values was done using the Octet^® Data Analysis software. The partial fit option was selected to fit dissociation for a window going from 10 to 700 s by the 2:1 heterogeneous ligand model.

Results

The SOG1^NAC dimer binds specifically to DNA in vitro

The in vivo DNA-binding specificity of Arabidopsis thaliana SOG1 has previously been investigated [19, 20]. Using ChiP-seq and microarray experiments, Ogita et al. proposed the palindromic consensus sequence CTT(N)₇AAG as the essential binding motif of SOG1 [19]. This ‘Ogita motif’, however, occurs only in 51% of the identified SOG1 targets, and targets where the ‘Ogita motif’ is not present or imperfect are assumed to be indirect targets by these authors. In a different study, Bourbousse et al. identified enriched sequence motifs present in the promoters of gene groups that are upregulated by SOG1 in response to genotoxic stress, as potential SOG1 binding sites (Supplementary Fig. S1) [20]. These ‘Bourbousse motifs’, which are similar to each other, are less restrictive than the pure CTT(N)₇AAG consensus sequence, allowing variations thereof and imposing some further weaker preferences outside the CTT and AAG triplets.

To confirm these in vivo results in vitro, EMSA experiments were carried out using SOG1^NAC and a 49 bp, ds fragment of the BRCA1 gene promoter (pBRCA1), an established in planta target of SOG1 (Fig. 1A and Table 1) [19]. The pBRCA1 fragment contains the strict CTT(N)₇AAG Ogita motif and also matches with the Bourbousse motifs (Supplementary Fig. S1). SOG1^NAC shows specific binding to the pBRCA1 fragment, but not to a randomly generated dsDNA fragment of the same length lacking the Ogita and Bourbousse motifs (Table 1)—from now called non-target DNA (Fig. 1A). For the latter, only a transient protein–DNA complex is visible, which is likely formed via non-specific electrostatic interactions between the negatively charged DNA and the overall positively charged NAC domain (Fig. 2D). This complex is also visible with pBRCA1 DNA in presence of excess SOG1^NAC (Fig. 1A).

Figure 1.

SOG1^NAC binds specifically to DNA (2 μM) in vitro. (A) Binding motif located on pBRCA1 and EMSA of SOG1^NAC with a 49 bp dsDNA fragment of pBRCA1 DNA on which this motif is centrally located and a 50 bp fragment of randomly generated, non-target dsDNA. (B) EMSA of SOG1^NAC with pBRCA1 dsDNA oligos of decreasing length, all centrally containing the binding motif. (C) EMSAs of SOG1^NAC with a WT and mutated 31 bp ds oligo of pBRCA1. (D) EMSAs of SOG1^NAC with a WT and mutated 31 bp ds oligo of pRAD51.

Figure 2.

The SOG1^NAC dimer binds with nanomolar affinity to pBRCA1 DNA. (A, B) ITC of SOG1^NAC with a 31 bp pBRCA1 ds oligo (containing the complying binding site) and a randomly generated, non-target dsDNA fragment of 31 bp, respectively. Upper graphs show the raw heat plot; lower graphs show the integrated heat plot. Indicated stoichiometry (of protein dimer) N and K_D values are averaged from triplicate measurements (two biological and one technical replicate). (C) Signature plot of SOG1^NAC binding to target 31 bp pBRCA1 DNA, showing the average ΔG, ΔH, and −TΔS values from triplicate measurements (two biological and one technical replicate). (D) Surface charge map of the SOG1^NAC dimer in complex with dsDNA, as predicted by AlphaFold3.

The specificity of SOG1^NAC for the binding site on the pBRCA1 fragment was further confirmed by probing two designed 95 bp dsDNA oligos: one containing two 31 bp pBRCA1 binding sites connected by a random linker and one containing only one such 31 bp site connected by the same linker to a 31 bp non-target DNA fragment (Table 1 and Supplementary Fig. S2). The data obtained indicate that SOG1^NAC specifically targets the binding site on the pBRAC1 fragment as two bands of specifically bound protein–DNA complex are present for the first DNA oligo, whereas only one band is observed with the second oligo.

The binding site of SOG1^NAC on pBRCA1 was further delineated using fragments of decreasing lengths. The minimal length of the SOG1^NAC binding site lays between 17 and 23 bp (Fig. 1B and Table 1). The six CTT-AAG nucleotides that are separated by the N₇-linker in the CTT(N)₇AAG Ogita motif appear to be critical for binding of pBRCA1 as mutations therein result in weak non-specific binding only (Fig. 1C). The same result is obtained when this motif is mutated in the context of a RAD51 promoter fragment. The latter is also a confirmed SOG1 target and contains a binding motif complying to the Bourbousse motifs, including the CTT(N)₇AAG sequence (Fig. 1D, Supplementary Fig. S1, and Table 1). The loss of specific DNA binding likely explains why the same mutations in the RAD51 promoter in planta result in the ∆RAD51 phenotype of A. thaliana transgenic plants upon genotoxic stress [19].

SOG1^NAC binds to pBCRA1 as a dimer with nanomolar affinity

Generally, NAC transcription factors form homodimers and bind to DNA as a dimer. Using analytical SEC, the apparent MW of the SOG^NAC in solution was determined to be 55.7 kDa (Supplementary Fig. S3A). This value is larger than the theoretical value of 39.4 kDa for a homodimer and may result from the non-globular shape of the SOG1^NAC dimer. To assess the stoichiometry and affinity of SOG1^NAC interacting with DNA, ITC was performed with the 31 bp pBRCA1 ds oligo containing the binding site and the non-target dsDNA fragment of the same length (Fig. 2A and B). The ITC results confirm the specificity of SOG1^NAC for the pBRCA1 DNA observed in EMSA experiments. Furthermore, the results show that the binding of the NAC domain to the DNA is an exothermic reaction and that SOG1^NAC binds as a dimer to the DNA, which is in line with the behaviour of other NAC transcription factors [31, 32]. The dissociation constant (K_D) was determined to be 139 nM and the reaction is enthalpy driven (Fig. 2A and C). AlphaFold3 structure predictions of SOG1^NAC bound to pBRCA1 DNA suggest a cluster of positive charges on the NAC surface to be involved in DNA binding (Fig. 2D), similar to what was previously observed for ORE1 [32]. These positively charged regions are presumed to give rise to the non-specific, transient DNA binding observed in the EMSA experiments. In consequence, altering the buffer salt content and thus the ionic strength influences the affinity of the interaction between SOG1^NAC and pBRCA1 (Supplementary Fig. S4).

SOG1^NAC specifically targets the Bourbousse binding motifs

The strict CTT(N)₇AAG Ogita motif covers only half of the in vivo identified targets of SOG1, suggesting that the presence of this motif alone does not suffice as sole SOG1 binding prerequisite. In contrast, the Bourbousse motifs are more elaborate.

The pBRCA1 site that was extensively used in this research thus far matches the Ogita motif as well as all three Bourbousse motifs. The promoter region of the SMR5 gene (pSMR5), another in vivo target of SOG1, contains three potential SOG1 binding sites (Table 1 and Fig. 3A) [19, 33]. Of these, site 1 contains the CTT(N)₇AAG Ogita motif and is also compatible with the Bourbousse motifs. Site 2 does not contain CTT(N)₇AAG, but is still compatible with a Bourbousse motif. Finally, site 3 does again contains the CTT(N)₇AAG Ogita motif, but is otherwise less compatible with any of the Bourbousse motifs in the context of the (weaker) sequence preferences for the seven nucleotides in between the CTT and AAG triplets (Supplementary Fig. S1). An EMSA with pSMR5 dsDNA shows two bands of protein–DNA complex, suggesting the presence of only two functional binding sites (Fig. 3A). To detect to which two sites SOG1^NAC binds, EMSAs were performed with shorter dsDNA fragments that each contain one of the three putative binding sites (Fig. 3B). These show that SOG1^NAC only binds to the two sites that contain an identified Bourbousse. No specific binding is observed, for site 3, even if the CTT(N)₇AAG sequence is present, indicating a role for adjacent nucleotides in SOG1 specificity. Similarly, site 2 does interact with SOG1^NAC, even if the Ogita consensus sequence is imperfect.

Figure 3.

The Bourbousse identified motifs best describe the SOG1^NAC binding site. (A) Putative binding motifs present on pSMR5: sites 1 and 2 contain Bourbousse motifs, while site 1 also confirms to the Ogita motif. Site 3 only contains an Ogita motif. EMSA of SOG1^NAC with a 95 bp dsDNA fragment of pSMR5 containing all three sites. (B) EMSA of SOG1^NAC with 31 bp dsDNA fragments of the three putative binding sites present on pSMR5, showing only affinity for the Bourbousse motifs. (C) Binding motifs present on pXRI1 and pTKa1, both containing a Bourbousse motif without including a strict CTT(N)₇AAG Ogita motif. EMSA of SOG1^NAC with 31 bp dsDNA fragments of pXRI1 and pTKa1.

Further confirmation of the Bourbousse motifs as the true interaction sites for SOG1^NAC comes from EMSA experiments utilizing XR1 and TKA1 promoter fragments (pXRI1 and pTKa1). These in vivo SOG1 targets both encompass Bourbousse motifs (Supplementary Fig. S1) without including the strict CTT(N)₇AAG motif (G to T substitution in final position of the motif; Fig. 3C) [19]. SOG1^NAC is able to bind both ds promoter fragments with a similar affinity as for the pBRCA1 binding site. This confirms that the more general motifs defined by Bourbousse and co-workers better describe the target binding site of SOG1 and that the strict CTT(N)₇AAG consensus site is too restrictive. Additional base pairs surrounding the CTT(N)₇AAG motif likely also play a role and variations on this motif are also possible within this context.

SOG1^NAC undergoes nucleic acids-driven phase separation in vitro

To better understand our observation that mixtures of purified SOG1^NAC and nucleic acids become turbid over time, the optical density at 390 nm (OD₃₉₀) was measured after mixing SOG1^NAC with poly-A RNA. SOG1^NAC evolves a turbidity profile that is typical for many phase separating proteins: an increase of OD₃₉₀ followed by a decrease that stabilizes to a baseline (Fig. 4A) [34]. From these data, a re-entrant phase behaviour is observed. The OD_390nm maxima that are reached in the OD₃₉₀ versus time profiles increase with rising concentrations of poly-A, but decline starting at concentrations poly-A between 100 and 250μg/ml until the signal eventually no longer detectable is (Fig. 4B). Formation of spherical droplets for varying protein and poly-A concentrations was confirmed using fluorescence microscopy (Fig. 4C and Supplementary Fig. S5). Through dual colour labelling, SOG1^NAC and poly-A were demonstrated to cointegrate into the droplets (Fig. 4D). Interestingly, both droplets with a hollow shell-like fluorescence pattern and droplets with an even solid fluorescence pattern are observed (Fig. 4E). Variation of the ionic strength does influence turbidity profiles of SOG1^NAC–poly-A mixtures, indicating that electrostatic interactions are involved in the phase behaviour (Fig. 4F), as was also seen for the DNA-binding experiments. In contrast to salt concentrations between 68 and 100 mM NaCl, a substantial drop in turbidity is observed for higher salt concentrations containing 150 mM NaCl. Droplet growth in function of time was mapped by DLS for 4 h (Fig. 4G). The hydrodynamic radius R_h in function of time follows Ostwald ripening for the first 2 h [35], highlighting the dynamic and liquid-like nature of the droplets that are initially formed [36]. After 2 h, the droplets size stabilizes. FRAP of a 10 min old reaction shows limited recovery of ~10%, 2 min post bleaching (Fig. 4H). To rule out the possibility that insufficient amounts of non-bleached protein in the surroundings could limit recovery, droplets were only partially bleached. The recovery profiles of the bleached regions were similar to fully bleached droplets, with a recovery percentage averaging up to ~10% in both cases (Supplementary Fig. S6A). It should be noted that for a subset of droplets (~25%) recovery was observed primarily in their outer layers (Supplementary Fig. S6B), reminiscent of the previously mentioned droplets with shell-like fluorescence pattern. Recoveries reach mean values of 28.8%, 20.6%, and 17.3% for the shell, the core, and the full droplet, respectively. This indicates that while there is a significant mobile fraction present in the condensates, there is also a less mobile fraction present. In some cases, however, the fluorescence of the inner part and the shell becomes more uniform over time.

Figure 4.

RNA-dependent phase behaviour of SOG1^NAC. (A) Turbidity profiles of NAC in function of time at 390 nm (left panel, n = 2 biological replicates each with n = 3 technical replicates) and phase diagram of NAC at different concentrations of poly-A based on turbidity profiles (right panel, n = 3 technical replicates). (B) OD₃₉₀ versus poly-A concentrations: re-entrant phase behaviour with increasing OD₃₉₀ signals for increasing poly-A concentrations that drop again for even higher concentrations (n = 2 biological replicates each with n = 3 technical replicates). (C) Fluorescence microscopy image of DyLight488-labelled protein (SOG1^NAC, SOG1^{NAC-DNAstrand} mutant, SOG1^NAC-RKRR mutant, and SOG1^NACΔ1–6) in presence of poly-A RNA for different concentrations (scale bar = 10 µm; similar results observed in n = 2 biological replicates). (D) Fluorescence microscopy image of cointegration of DyLight488-labelled NAC and Cy5-labelled poly-A RNA in spherical droplets (scale bar = 10 µm; similar results observed in n = 3 biological replicates). (E) Shell-like and fully coloured droplets of DyLight488 labelled NAC in presence of unlabelled poly-A. (F) OD₃₉₀ versus NaCl concentrations for reactions with 20 µM SOG1^NAC and 25 µg/ml poly-A [n = 2 technical replicates; similar results observed in n = 3 biological samples with different concentrations (5, 20, and 40 µM) of SOG1^NAC]. (G) Droplet growth of a reaction with 20 µM SOG1^NAC and 25 µg/ml poly-A, followed by DLS (n = 3 technical replicates; similar results observed in n = 2 biological samples with slightly different concentrations) (H) FRAP of droplets with 20 µM SOG1^NAC and 25 µg/ml poly-A (scale bar = 2 µm, n = 2 biological replicates, each with n = 2–3 technical replicates).

To mimic more biologically relevant conditions, buffer conditions were adjusted and experiments were conducted using plant-derived mRNA instead of poly-A. The buffer was supplemented with bivalent cations (Mg²⁺) in addition to monovalent salt as magnesium in particular plays critical roles in nucleic acid condensation [37]. Magnesium modulates condensate formation involving SOG1^NAC and poly-A RNA, with elevated concentrations leading to the generation of less and smaller condensates. At the physiological relevant condition of 150 mM NaCl and 5 mM MgCl₂, condensations is clearly reduced (Fig. 5A). When mRNA is used instead, interestingly, condensation is less perturbed in this physiological condition (Fig. 5B). Furthermore, FRAP reveals that these mRNA-driven condensates exhibit high dynamicity, with recovery percentages reaching ~80% (Fig. 5C).

Figure 5.

(A) Fluorescence microscopy image of 20 µM DyLight488-labelled SOG1^NAC in presence of 25 µg/ml poly-A RNA for different concentrations NaCl and MgCl₂ (scale bar = 10 µm; similar results observed in n = 2 technical replicates). (B) Fluorescence microscopy image of 20 µM DyLight488-labelled SOG1^NAC in presence of 25 µg/ml mRNA for different concentrations NaCl and MgCl₂ (scale bar = 10 µm; similar results observed in n = 2 technical replicates). (C) FRAP of droplets with 20 µM SOG1^NAC and 25 µg/ml mRNA (n = 5 technical replicates).

Next, we investigated whether addition of dsDNA to SOG1^NAC similarly could induce phase separation. DNA fragments used include the full 310 bp promoter sequence of BRCA1, the 31 bp pBRCA1 ds oligo containing the Ogita/Bourbousse binding motif, and the 293 bp promoter of the non-target house-keeping gene UBQ10 (Fig. 6A). Small condensates were formed in SOG1^NAC mixtures with the short 31 bp pBRCA1 fragment, while shell-like droplets and irregular non-spherical structures are formed in presence of the longer 310 bp fragment or the 293 bp non-target sequence. The higher the concentration of dsDNA, the more non-spherical condensates are present (Fig. 6B). Also, in more relevant physiological conditions (150 mM NaCl and 5 mM MgCl₂), SOG1^NAC does phase separate with DNA (BRCA1 31 bp) (Fig. 6C). Interestingly, FRAP experiments confirm a dynamic behaviour of those irregular formed condensates with a recovery of around 80% after 2 min in some cases (Fig. 6D and E). Not all condensates seem to be that dynamic as recovery rates of only 20% are also observed. Maturation of the condensates was mapped by fluorescence microscopy and FRAP experiments after different incubation times. After 5 and 10 min, the recovery percentages show a broad distribution, indicating heterogeneity in early condensate maturation. In contrast, for 30 and 60 min, the recovery values converge into a narrower range (Fig. 6F). Morphologically, the irregularly shaped condensates seen at earlier time points gradually transition into large spherical droplets as incubation progresses (Fig. 6G). This suggests that condensate maturation leads to increased structural and dynamic uniformity after 30 min of incubation.

Figure 6.

DNA-driven phase separation of SOG1^NAC and its mutants (A) with pBRCA1 31 bp oligo, and target versus non-target DNA promoter sequences (B) with different concentrations of the promoter sequence of BRCA1 target (scale bar = 10 µm; similar results observed in n = 2 biological replicate). Protein concentration is 20 μM in all cases. (C) Fluorescence microscopy image of 20 µM DyLight488-labelled SOG1^NAC in presence of 12.5 µg/ml BRCA1 31 bp DNA oligo for different concentrations NaCl and MgCl₂ (scale bar = 10 µm) (D, E) FRAP on condensates of 20 µM NAC and 12.5 µg/ml BRCA1 promoter sequence (310 bp) after 10 min (scale bar = 2 µm, n = 2 biological replicates, each with n = 2–3 technical replicates). (F) Summary of FRAP recovery percentages of condensates of 20µM NAC and 12.5 µg/ml BRCA1 promoter sequence (310 bp) at different time points (n = 4 technical replicates). (G) Fluorescence microscopy image of 20 µM DyLight488-labelled SOG1^NAC in presence of 12.5 µg/ml BRCA1 promotor sequence (310 bp) over time, evolving from small and irregular formed condensates to large spherical droplets (scale bar = 10 µm; similar results observed in n = 2 technical replicates).

This discrepancy in recovery rate between dsDNA-, poly-A-, and mRNA-driven LLPS may indicate a different internal structure. Possibly this is related to differences in affinity of SOG1 for the different RNA and DNA partners. Indeed, SOG1_NAC exhibits a higher binding affinity for poly-A than for the short DNA fragment used as an apparent affinity of <10 picomolar is obtained from BLI data fitted with a 2:1 (heterogeneous ligand) binding model (data not shown). Elevated binding affinities generally correlate with reduced conformational dynamics due to increased complex stability. However, as the binding model is not known the obtained affinities should be considered with care. One should also take into account that, in multivalent systems, factors such as avidity, cooperativity (positive or negative), local concentration effects, and spatial proximity between binding sites can significantly influence the measured affinities.

Dimerization of the SOG1 NAC domain is required for spherical droplet formation and proper DNA binding

We next investigated whether dimerization is required for SOG1^NAC DNA binding and phase separating behaviour. Based on AlphaFold3 structural predictions of the SOG1^NAC dimer, a truncation mutant lacking the N-terminal β-strand (ΔL58-K63) was designed from now called SOG1^NACΔ1–6 (Supplementary Fig. S7A). The SOG1^NACΔ1–6 protein has an estimated MW of 19.8 kDa, as measured using analytical SEC, confirming a (globular) monomeric state in contrast to the non-truncated SOG1^NAC that eluted as a dimer (Supplementary Fig. S3A). The CD profiles of SOG1^NAC and SOG1^NACΔ1–6 are similar, with a negative peak around 220 nm indicating the presence of predominantly β-sheets in both proteins (Supplementary Fig. S3B).

Despite being monomeric, SOG1^NACΔ1–6 still binds to the 31 bp pBRCA1 ds oligo, but with lower affinity compared to WT SOG1^NAC (Fig. 7A and B). The affinity and stoichiometry could not be further assessed by ITC, because the SOG1^NACΔ1–6 protein precipitates at higher concentrations used for this. This implies that dimerization is needed for the intrinsic stability of the isolated NAC domain. The decrease in affinity is likely due to the independent binding of the SOG1^NACΔ1–6 monomers compared to the avidity generated via a stable dimer, although a lower thermodynamic stability of SOG1^NACΔ1–6 may also play a role.

Figure 7.

DNA binding of SOG1^NAC and its mutants. (A)–(D) EMSA of WT SOG1^NAC, SOG1^NACΔ1–6 (=monomer), the SOG1^{NAC-DNAstrand} mutant, and the SOG1^NAC-RKRR mutant, respectively, with a 31 bp ds oligo of pBRCA1 containing the SOG1^NAC binding site.

SOG1^NACΔ1–6 shows a changed LLPS behaviour as the formation of spherical droplets in presence of poly-A is no longer observed. Instead, irregular structures are formed that resemble gel-like condensates or fibrils [38] (Fig. 4C and Supplementary Fig. S5). As observed in ITC experiments, the protein may begin to aggregate in the presence of high protein concentrations. This may occur immediately after the start of condensation when the protein concentration increases. In presence of DNA, phase separation is again defective. No spherical droplets are observed, but occasionally non-spherical structures appear. This suggests that dimerization is required for spherical droplet formation by both RNA- or/and DNA-driven LLPS.

Two positively charged clusters of SOG1^NAC link DNA binding with RNA- and DNA-driven LLPS

AlphaFold3 structure predictions of SOG1^NAC bound to the 31 bp pBRCA1 fragment reveal two segments rich in positively charged AAs that likely determine DNA specificity due to their proximity to the nucleotides making up the target binding site. These are a loop encompassing residues R136–R139 and a β-strand encompassing residues R150–T157 (residues positioned on the full length protein) (Supplementary Fig. S7B and C). The corresponding segments in the NAC domains of ANAC019 and ORE-1 differ in AA sequence, but also constitute the DNA-binding site [31, 32]. Two SOG1^NAC mutants harbouring mutations in these two putative DNA-binding segments were designed. For both mutant proteins, the positive charges were mutated to non-charged AAs. The first mutant referred to in this paper is the SOG1^NAC-RKRR mutant that has following mutations: R79A, K80A, R81A, and R82A. The second is called the SOG1^{NAC-DNAstrand} mutant with mutations R93S, H95S, T97A, R99S, and T100S, which are located in the β-strand that docks into the major DNA groove. CD spectra of both mutants are very similar to that of the WT SOG1^NAC protein, indicating proper folding (Supplementary Fig. S3B). Both elute at approximately the same volume as WT SOG1^NAC in analytical SEC: apparent MWs of 67.4 and 58.2 kDa for SOG1^NAC-RKRR and SOG1^{NAC-DNAstrand} mutant, respectively. We therefore assume that, like SOG1^NAC, the mutants maintain their ability to form homodimers in solutions (Supplementary Fig. S3A).

EMSA experiments performed with both mutants and the 31 bp pBRCA1 fragment show a complete loss of DNA-binding capabilities (Fig. 7C and D). This indicates that both structural elements that were mutated in these NAC domain variants are essential parts of the DNA-binding site. Furthermore, it implies that SOG1^NAC employs a similar binding mechanism as ANAC019 and ORE-1 that is presumably conserved within the entire family of NAC transcription factors [31, 32].

We next investigated whether the cluster of positive charges that is essential for DNA binding is also DNA- or RNA-mediated phase separation (Fig. 6B). DNA-mediated phase separation is abolished for both mutants, confirming that an intact DNA-binding site, or at least the presence of positive charges, is essential for multivalent protein–DNA interactions.

RNA-driven phase behaviour of the two mutants is altered compared to WT SOG1^NAC. However, RNA-induced phase behaviour is not completely abolished. For the SOG1^{NAC-DNAstrand} mutant, small droplets are still formed in most of the conditions tested (Fig. 4C and Supplementary Fig. S5). The intact RKRR loop is suggested to be essential to support the LLPS potential of the SOG1^{NAC-DNAstrand} mutant with DNA. Altogether, these results establish a functional link between the DNA-binding potential of SOG1^NAC and its ability to undergo phase separation both with RNA and DNA.

The phase-separating behaviour of SOG1^NAC is controlled by a tight interplay between DNA and RNA binding

The results above suggest that both RNA and DNA bind to overlapping regions on the surface of SOG1^NAC. Therefore, the question follows whether DNA and RNA can cointegrate in the SOG1^NAC condensates or whether one would exclude the other. We first challenged poly-A–SOG1^NAC condensates with the addition of pBRCA1 31 bp oligo. Below equimolar concentrations of SOG1^NAC dimer and DNA, the DNA fragment can integrate in the RNA-induced condensates (Fig. 8A). However, increasing pBRCA1 31 bp oligo concentrations weakens the RNA-driven phase behaviour. At equimolar concentrations of SOG1^NAC dimer and the pBRCA1 31 bp target sequence, poly-A-induced condensates of SOG1^NAC are fully dissolved. The pBRCA1 31 bp DNA target competes with RNA and prevents poly-A-driven condensation by excluding poly-A from the droplets. This agrees with the hypothesis that poly-A and the pBRCA1 31 bp oligo bind to the same or overlapping sites on SOG1^NAC. Binding of the pBRCA1 31 bp oligo to SOG1^NAC presumably breaks the multivalent dynamic SOG1^NAC–RNA interactions, which stabilize the condensates, although few small condensates still form.

Figure 8.

Competition assays RNA versus DNA binding of SOG1^NAC: fluorescence microscopy to visualize RNA and DNA cointegration in poly-A-induced SOG1^NAC condensates. Ten micromolars SOG1^NACdimer (green fluorescence) in presence of 25 ug/ml poly-A (left: unlabelled, right: red fluorescence) was titrated with various amount DNA (left: red fluorescence, right: unlabelled), either pBRCA1 31 bp DNA oligo (A) or random 31 bp DNA oligo (B) (scale bar = 10 µm; similar results observed in n = 2 biological replicates for left panels and n = 1 replicate for right panels). Bio-layer interferometry of immobilized SOG1^NAC and poly-A RNA versus various concentrations of pBRCA1 31 bp DNA oligo (n = 2 technical replicates) (C) and random non-target DNA oligo compared to pBRCA1 31 bp DNA oligo and no DNA (n = 1 replicate) (D).

When this competition experiment is repeated with a non-specific DNA fragment of the same length (Fig. 8B), again poly-A is excluded from the droplets at a 1:1 molar ratio. Differently, more droplets (for quantification, see Supplementary Fig. S8) are assembled with SOG1^NAC and non-target DNA at this equimolar ratio. Probably, no stable soluble SOG1^NAC–non-cognate DNA complexes are formed and transient and weak multivalent protein–DNA interactions involving a non-cognate DNA sequence enable DNA-driven condensation.

The finding that DNA with a specific SOG1-binding motif can outcompete RNA is supported by a BLI competition assay (Fig. 8C). After association of poly-A to immobilized SOG1^NAC, subsequent dipping in pBRCA1 31 bp oligo results in a dissociation profile for which the observed k_off rate increases with rising concentrations DNA, ranging from 10⁻⁵ to 10⁻³ s⁻¹. For non-target DNA, no such dissociation is observed (Fig. 8D), thus in line with ITC results where no binding is detected for non-target DNA. However, as the binding model for the SOG1^NAC–poly-A interaction is not characterized, the obtained k_off rates should be considered with care.

Discussion

SOG1 is a plant transcription factor central to the DNA damage response in which it takes on the role of mammalian p53 [23]. Upon genotoxic stress, it controls the transcription of around 300 genes, either directly or indirectly. The target binding sites of SOG1 have been previously studied in vivo using ChIP-sequencing and RNA sequencing by two groups [19, 20]. Here, we show that SOG1^NAC targets sequences compatible with the three sequence logos described by the Bourbousse, i.e. less strict than the CTT(N)₇AAG consensus motif identified by Ogita et al. [19]. This motif can be perturbed by two nucleotides without loss of SOG1^NAC binding and residues outside of the CTT(N)₇AAG consensus also affect in vitro binding.

We further show that SOG1^NAC displays phase DNA- and RNA-driven separation in vitro. Somewhat surprising is that SOG1^NAC is an isolated fully folded globular. The majority of the available literature as of yet typically focuses on the importance of intrinsically disordered and low complexity domains in LLPS. [16]. Folded domains typically drive LLPS when present as repeats in a beads-on-a-string format that creates multivalency or when interactions with LLPS partners induce unfolding (for a review, see [16]).

Functional LLPS of fully folded proteins is rare. While LLPS can typically be induced at high concentrations using crowding agents, as initially observed for lysozyme [39], only few do so under physiological conditions, a rare example being E. coli Lipoate-protein ligase A [40]. However, different studies suggest that globular domains that bind multivalent biomolecular strands—such as nucleic acids or intrinsically disordered regions (IDRs)—may represent a generalizable mechanism for regulating LLPS. This first principle is exemplified by SH3–proline-rich motif (PRM) and polySUMO–polySIM systems, where repeats of folded domains (SH3 or SUMO) drive LLPS through multivalent interactions with short peptide sequences (like proline-rich motifs) in IDRs [1, 12]. Supporting this, Su et al. [41] showed that T cell receptor signalling relies on LLPS mediated by two SH3 domains within adaptor protein Grb2, which cluster signalling molecules into liquid-like condensates. These structures enhance signalling by concentrating kinases and excluding phosphatases, illustrating how folded domains embedded in multivalent networks can drive physiologically relevant LLPS. The required multivalency necessarily comes from weak protein–protein interactions and or weak protein–nucleic acid interactions, in addition to nucleic acid binding via the DNA-binding site of SOG1^NAC.

Given the similar role of SOG1 as p53 in animals, it is worthwhile comparing their physicochemical properties and how they link DNA binding to condensate formation. p53 can both form liquid condensates and aggregates. However, the core DNA-binding domain on its own is not known to undergo phase separation, only aggregation due to misfolding (for a recent review, see [42]). Full length p53 forms liquid condensates that mature into solid aggregates [43, 44], and mutations in its tetramerization domain, especially those that affect multivalency, reduce LLPS [45], similar to what we observe for the SOG1^NAC dimer. In addition, condensation of p53 is affected by post-translational modifications, and in particular phosphorylation [44–46]. It will therefore be of interest to also study full length SOG1 and its phospho-variants as both SOG1 and p53 become activated upon DNA damage through phosphorylation by ATM or ATR [26, 27, 47–50]).

LLPS is a key component of regulation of transcription and translation, involving a complex interplay between the components involved: transcription factors, DNA, RNA polymerase, RNA, the mediator complex, chromatin, and a variety of other regulatory proteins [7, 51]. The RNA-dependent phase behaviour of the conserved SOG1^NAC implies an RNA-mediated feedback control mechanism for the formation of condensates at transcription start sites, corresponding to the model proposed by Henninger et al. [11]. According to this model, condensate formation at transcription sites is abolished in the presence of too high concentrations of mRNA. The latter is mimicked by the re-entrant phase behaviour of SOG1^NACseen in presence of poly-A. A similar re-entrant RNA-dependent phase behaviour was observed for the human androgen receptor where its folded DNA-binding domain serves as the minimal region driving LLPS [14].

A tight and complex interplay between DNA (target versus non-target) and RNA binding controls the formation of condensates at transcription sites. For SOG1^NAC, RNA and dsDNA compete with each other for LLPS. Mutations that abolish specific DNA binding of SOG1^NAC also induce an altered DNA- and RNA-dependent phase separation behaviour compared to the WT protein. This suggests that the binding site for specific DNA binding overlaps with that for both non-specific DNA and RNA recognition that drives LLPS, allowing DNA specificity to influence the LLPS behaviour of SOG1^NAC. This agrees with similar differential behaviour of other transcription factors where both RNA- and DNA-mediated condensation benefit from sequence-specific interactions (for a review, see [52]).

The interplay between RNA- and DNA-induced LLPS is an essential part of transcription; typically, transcription factors first undergo DNA-driven LLPS to initiate transcription, while the resulting RNA accumulation subsequently transits the transcription factors into RNA-driven LLPS. The behaviour of SOG1 with respect to the competition between RNA and DNA reported on in this manuscript reflects this switch in LLPS driving force. Competition of DNA and RNA for a shared or overlapping binding site may form the structural basis for the transition between two types of transcription-related condensates.

Notes

Present address: Biozentrum, University of Basel, Spitastrasse 41, 4056 Basel,Switzerland

Present address: Department of Neurosciences, VIB-KUL Laboratory for Neurobiology,ON5 Herestraat 49, bus 602, 3000 Leuven,Belgium

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

This work was supported by an FWO-Vlaanderen grant to R.L. and L.D.V. (G011420N), by a research infrastructure grant from VUB to R.L. (OZR4292) and by FWO PhD fellowships to K.M. (1103622N) and M.D. (1S18116N).

Data availability

The data underlying this article are available in the article and in its online supplementary material.