1.1. Biological functions of DBHS proteins

SFPQ and NONO are functionally diverse proteins that are ubiquitously present in mammalian nucleic acid-processing pathways, as reviewed by Knott et al. (2016). The known functions of SFPQ/NONO involve a role in almost every step of nucleic acid processing in mammalian cells, where they are involved in sequestration, co-repression/co-activation of transcription, RNA export, transport and retention, elongation/termination, co-transcriptional processing and the formation of subnuclear bodies (Knott et al., 2016). SFPQ is so involved in cellular function that studies have shown that it is a critical protein for life in mammals, with embryo-level knockouts causing organism death (Takeuchi et al., 2018). Interestingly, SFPQ and NONO are critical proteins for the assembly of the core region of paraspeckles: dynamic phase-separated nuclear condensates that are templated by approximately 50 molecules of the 23 kbp long noncoding RNA NEAT1 and are known to sequester, regulate and organize multiple types of RNA and proteins via liquid–liquid phase separation (LLPS) and extensive multivalency (Fox et al., 2018; West et al., 2016). SFPQ is also an emerging actor in neurodegenerative disease research due to its critical role in the development and regulation of neurons at multiple tiers of nucleic acid processing such as transcription, splicing, axonal RNA transport and stress-granule formation (Lim et al., 2020). The imbalanced nucleocytoplasmic distribution of SFPQ is reportedly a factor in the neurodegenerative diseases amyotrophic lateral sclerosis (ALS), frontotemporal lobar degeneration (FTLD) and Alzheimer's disease (AD) (Lim et al., 2020).

1.2. General architecture and behaviour of DBHS proteins

SFPQ and NONO, along with PSPC1, are the mammalian paralogs of the Drosophila behaviour/human splicing (DBHS) protein family and share 76% sequence identity in their conserved DBHS domain (Knott et al., 2016; Fig. 1a).

The DBHS domain is a structured 320-amino-acid conserved core region which contains two RNA-recognition motifs (RRMs), a NonA paraspeckle domain (NOPS) and a coiled-coil domain (Fig. 1a; Knott et al., 2016). SFPQ, NONO and PSPC1 form obligate dimers, with the core DBHS region being responsible for directing homodimerization and heterodimerization via an extensive network of stable interactions between monomers (Figs. 1b and 1c). The analysis by Passon et al. (2012) of the PSPC1–NONO dimer revealed that ∼25% of the solvent-accessible space of each monomer was buried as a result of dimerization, while analysis by Lee et al. (2015) of the interaction energy of the dimerization interface in a SFPQ homodimer revealed a very stable, high-affinity interaction (ΔⁱG = −42.7 kcal mol⁻¹, p-value = 0.05; Figs. 1b and 1c). The interaction is well preserved across the family, with crystal structures of all six dimeric permutations having been previously determined and analysed [Huang et al., 2018 (PDB entry 5wpa); Knott et al., 2022 (PDB entries 5ifn and 5ifm); Lee et al., 2015 (PDB entries 4wii, 4wik and 4wij); Passon et al., 2012 (PDB entry 3sde); Lee et al., 2022 (PDB entry 7lrq); Schell et al., 2022 (PDB entry 7pu5)]. An additional intermediate part of the DBHS conserved region is responsible for functional aggregation via a coiled-coil-forming interface, which plays a role in the cooperative binding of larger nucleic acids (Figs. 1b and 1c; Koning et al., 2025; Lee et al., 2015)

Despite strong evidence outlining the general functions of DBHS proteins and an apparent hierarchy of dimerization configurations (Huang et al., 2018, Knott et al., 2022; Lee et al., 2022), the role of the various dimers is poorly understood. The direct involvement of the DBHS region in nucleic acid interaction (Knott et al., 2022; Lee et al., 2015; Vickers & Crooke, 2016; Wang et al., 2022) suggests a potential biological role for this combinatorial expansion. Of note, the direct exchange of partners has been demonstrated in vitro (Lee et al., 2022) between SFPQ and NONO homodimers truncated to contain only the dimerization domain, resulting in a population of SFPQ heterodimers. Mechanistically, how partner swapping without cofactors occurs is currently unknown, and is remarkable considering the interaction energies of the various DBHS dimer interfaces. Lee et al. (2022) provided some clues through the identification of certain features such as a helix in the NOPS domain and the relative position of RRM1 in both molecules, which differed across various dimers, suggesting that instability or flexibility of certain stabilizing interactions may be involved in partner swapping or preferential dimerization. To date, direct partner exchange of full-length proteins has not been shown and has implications for understanding the roles of dimers within cells.

1.3. Intrinsically disordered regions in DBHS proteins and regulation of liquid–liquid phase separation

Outside of the DBHS domain, the three human paralogs are flanked by extensive regions which vary substantially in sequence and are predicted to be intrinsically disordered (IDR, intrinsically disordered region; Fig. 1c). Previously, this disorder had not been shown experimentally, but becomes apparent when using many sequence-structure prediction tools such as the AlphaFold pairwise local distance difference test (pLDDT; Fig. 1c, bottom; Mirdita et al., 2022) and the RIDAO (Rapid Prediction and analysis of Protein Disorder Online) suite of disorder-prediction tools (Dayhoff & Uversky, 2022). Together, these tools and others such as IUPred2A (Marshall et al., 2023) indicate that these regions are highly likely to be flexible and disordered. Interestingly, these regions have also been shown, in the case of SFPQ (Marshall et al., 2023), or predicted (Supplementary Figs. S1 and S2) to be capable of driving liquid–liquid phase separation (LLPS). Recently, Marshall et al. (2023) added further nuance to this idea by examining the contributions of the two predicted flanking IDRs of SFPQ towards LLPS. The predicted IDRs of SFPQ were shown experimentally to be directly involved in LLPS, with the C-terminal IDR driving phase separation and the N-terminal IDR attenuating phase separation (Marshall et al., 2023). Marshall et al. (2023) proposed a possible direct regulatory interaction between the IDRs of SFPQ in the context of individual dimers for the purpose of modulating condensate formation in the nucleus.

Structural studies of LLPS proteins are often challenging as their high degrees of disorder, number of dynamic conformations, solubility and capacity for oligomerization and phase separation can make crystallization or studies with electron microscopy difficult or impossible (Martin, Hopkins et al., 2021). For this reason, despite the exhaustive characterization of the structured DBHS domain, the structural details of the predicted IDRs of SFPQ and whether a direct intradimer interaction between the IDRs is possible in solution have yet to be described experimentally. The potential for combinatorial dimerization and dimer partner exchange of full-length proteins under near-physiological conditions may impact on LLPS: it is possible that nature uses the divergent sequence features of each paralog through combinatorial dimerization to further control DBHS protein LLPS and the material properties of nuclear condensates.

Small-angle scattering using X-rays or neutrons (SAXS or SANS) has emerged as an effective method for studying disordered proteins structurally in solution. In this study, we employ both SAXS and SANS, in conjunction with lysine cross-linking mass spectrometry (XL-MS), to gain insights into the structure and dynamics of the predicted intrinsically disordered regions (IDRs) of SFPQ and to explore the potential for dimer partner exchange of full-length proteins in vitro. Firstly, we compared the scattering of a tractable truncate of SFPQ missing the C-terminal IDR (SFPQ1–598) and compared it with the data for the full-length protein (SFPQ1–707). Our solution scattering data demonstrate experimentally that the N- and C-terminal IDRs of SFPQ are long, disordered and flexible in solution. Ensembles of models generated with EOM 2.0 (Ensemble Optimization Method) suggest that a direct interaction between the IDRs as hypothesized by Marshall et al. (2023) is possible. Such an interaction may explain some degree of compaction seen in the ensemble that fits the scattering data relative to the initial pool of search models. The cross-linking mass-spectrometry data also encouragingly show that the distal ends of the C-terminal IDR can make points of contact with the folded domain and that both IDRs can come into close proximity to one another in solution. We additionally demonstrate that full-length protiated SFPQ is capable of swapping dimer partners in solution with other molecules of deuterated SFPQ and that it is possible to capture scattering data of the full-length protein as a monomer in place of a dimer using contrast-matching small-angle neutron scattering (SANS).

In this study, we show the first structural description of the IDRs of SFPQ, and their potential dynamics in solution, as well as the capability of full-length SFPQ dimers to exchange partners with each other in a stable manner in vitro. These findings are biologically relevant as the IDRs directly control the material state of SFPQ and are either directly or indirectly involved in all of the biological functions of the protein. Additionally, partner swapping between full-length DBHS proteins is likely to allow multiple possible interactions between IDRs of different dimers and the modulation of phase properties via unique combinations of dimers within condensates. Together, these factors are important for paraspeckle formation, disease pathology and the several functions that SFPQ carries out that are critical to mammalian life.

2.1. Protein expression of deuterated and protiated SFPQ

The plasmids (i) pET-mEGFP-SFPQ (full-length) and (ii) pET-mEGFP-SFPQ (1–598) were transformed into Invitrogen OneShot BL21 Star (DE3) cells separately. The proteins were expressed using RTF bioreactors according to the method of Duff et al. (2015). In all cases, the medium was composed of ModC1, 78.1% D₂O and 40 g l^{−1 1}H-glycerol. 78.1% D₂O was chosen, using empirical data on past protein deuteration runs, to achieve a neutron scattering length density match point equivalent to 95% D₂O. The proteins were induced with 0.5 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) at OD_600 nm values of (i) 12.44 and (ii) 12.18 for subsequent expression at 20°C. The cells were harvested directly after exhaustion of the carbon source, as shown by a small rise in pH above the setpoint of 6.2. Deuteration levels were determined by MS (partial trypsin digest MALDI-TOF). In some cases the MS spectra were low quality and a precise deuteration level was unable to be achieved; however, all results are consistent with a deuteration level of 61.5 ± 0.5%. A consistent deuteration level was expected due to the medium and growth characteristics being the same in both cases.

For the production of unlabelled biomass the steps were the same as above, but no D₂O was used to ensure the expression of protiated versions of full-length SFPQ and SFPQ1–598. The proteins were induced with 0.5 mM IPTG at OD_600 nm values of (i) 13.24 and (ii) 12.15 for subsequent expression at 20°C. The cells were harvested directly after exhaustion of the carbon source, as indicated by a small rise in pH above the setpoint of 6.2. In all cases the medium also contained 40 µg ml⁻¹ kanamycin to maintain plasmid selection.

2.2. Protein purification of deuterated and protiated full-length SFPQ and SFPQ1–598

For the purification of all of the variants of SFPQ used in this study, the purification buffers from Marshall et al. (2023) were used. Lysis was carried out in the buffer 1 M KCl, 5% glycerol, 10 mM imidazole, 50 mM Tris–HCl, 250 mM L-arginine, 1 mM PMSF; addition of PMSF to all steps was optional for the purification of SFPQ1–598 but was necessary for full-length SFPQ. Frozen biomass was chipped from the container into a sterile and clean Schott bottle. The Schott bottle was then filled to a final volume 24 times that of the biomass (i.e. 10 g of biomass resuspended in 240 ml solution). This was performed using a mixture of 50 ml BugBuster 10× Protein Extraction Reagent (Merck) at a 1/10 final volume and a 9/10 final volume of lysis buffer supplemented with DNase I (Merck) at 50 µg ml⁻¹, two cOmplete Mini EDTA-free protease-inhibitor cocktail tablets (for this volume) and lysozyme to a final concentration of 0.2 mg ml⁻¹. The sample mixture was stirred at room temperature using a magnetic stirrer for ∼1 h until adequate resuspension/dissolution of the biomass into solution.

Lysates were then clarified by centrifugation and filtered using Whatman 0.4 µm filters and a vacuum degassing setup to reduce sample viscosity. Following filtration, the lysate was loaded using a peristaltic pump (Bio-Rad) onto 5 ml nickel-affinity columns (GE Healthcare) pre-equilibrated with ten column volumes of water and ten column volumes of binding buffer (1 M KCl, 5% glycerol, 10 mM imidazole, 50 mM Tris–HCl, 250 mM L-arginine, 1 mM PMSF pH 7.4). The column was then washed with ten column volumes of binding buffer, followed by 5–10 column volumes of binding buffer spiked with 13% elution buffer (binding buffer with 250 mM imidazole) to remove further contaminants (five column volumes were sufficient for SFPQ1–598). His-tagged protein was then eluted in ∼1–1.5 column volumes of nickel elution buffer. To remove the GFP tag, the eluted protein was subjected to an overnight digest with Tobacco etch virus protease at a 1:25 mass ratio. This digest was dialysed overnight at room temperature with a magnetic stirrer in ∼1 l nickel binding buffer supplemented with DTT to a final concentration of 1 mM.

Following this, the sample was recovered from the tubing, filtered using a 0.4 µm syringe filter and then flowed over a 5 ml nickel-affinity column (pre-equilibrated in binding buffer) to remove the TEV protease and residual GFP. The sample was further pushed through the column with binding buffer containing 5% elution buffer to remove any nonspecific interactions between SFPQ and the nickel resin (i.e. 5 ml was loaded onto the column and ∼10 ml was recovered). The sample was then purified by loading the eluate onto a Superdex 200 16/60 size-exclusion column pre-equilibrated in storage buffer (0.5 M KCl, 5% glycerol, 20 mM HEPES, 1 mM DTT pH 7.4). Elution peaks were monitored using the absorbance at 280 nm. The eluted protein was analysed with SDS–PAGE, flash-frozen with liquid nitrogen and stored at −80°C until further use.

2.3. Sequence analysis and structure prediction

Sequence analysis of SFPQ was performed using AlphaFold pLDDT scores retrieved from ColabFold (Mirdita et al., 2022) to test for regions of predicted structure outside of the DBHS domain. Phase-separating regions for the DBHS proteins were predicted using the FuzDrop tool (Hatos et al., 2022).

Protein amino-acid composition analyses were performed using custom R scripts available at https://github.com/acmarshall88/AA-CounteR. To calculate the `enrichment' of each of the 20 naturally occurring amino acids in a protein sequence of interest, the proportion of each amino acid was calculated and divided by its proportion within the entire human proteome (UniProt ID UP000005640_9606; contains one protein sequence per gene). Grey bars indicate infinite depletion (i.e. that amino-acid type is absent from the sequence). To visualize the occurrence of amino acids that are particularly enriched at any point along each DBHS protein sequence, the proportion of each amino acid was calculated in a sliding window of pre-defined width (i.e. 30 amino acids) across each protein sequence.

2.4. Small-angle X-ray scattering

2.5. Small-angle neutron scattering

2.6. 3D modelling

To model the conformers of full-length SFPQ and SFPQ1–598, a model of an SFPQ homodimer (residues 276–598) was generated using ColabFold (Mirdita et al., 2022). In order to generate flexible ensembles, EOM 2.0 (Petoukhov et al., 2012; Tria et al., 2015) was used to build residues 1–277 and 601–709 as disordered for full-length SFPQ and just residues 1–277 as disordered for SFPQ1–598. To assess the sampling of conformational space by our structures in solution, the distributions of the selected pool that fit the data and the random initial RanCh distribution were compared visually and also numerically using values such as the geometric mean R_g, R_flex and R_sigma. Reduced χ² values were used to assess the agreement of each ensemble with the experimental data, as well as normalized error-weighted residual plots. To model the SANS data, DAMMIF was run with ten repetitions on fast mode. The subsequent averaged DAMAVER envelope was compared with the atomic structure of a monomer of SFPQ without the IDRs attached.

2.7. Lysine cross-linking mass spectrometry (XL-MS)

For cross-linking mass spectrometry (XL-MS) the methodology was essentially the same as the method used by Sethi et al. (2023); purified full-length SFPQ and SFPQ1–598 protein samples were diluted to 10 and 20 µM for both proteins using storage buffer and mixed with a 100-fold excess of DSSO cross-linker (Kao et al., 2012) dissolved in dimenthyl sulfoxide (DMSO). Following the termination of the cross-linking reaction, the cross-linked proteins were digested with trypsin. LC-MS/MS was performed using a Fusion Lumos Orbitrap mass spectrometer with a FAIMS Pro source (Thermo Fisher, USA). To find the cross-linked peptides, the MS2CID–MS3HCD (MS2–MS3) workflow was used. Cross-linked peptides were then analysed using the XlinkX (Liu et al., 2017) node-implemented Proteome Discoverer 2.3 (Thermo Fisher Scientific). The results and subsequent data were then visualized in xiVIEW (Combe et al., 2024).

3.1. Full-length SFPQ in solution revealed by SEC-SAXS

To investigate the structure of the flanking IDRs of SFPQ in the context of the full-length protein, small-angle X-ray and neutron scattering (SAXS/SANS) experiments were performed on full-length SFPQ (707 residues; includes both IDRs) and on a truncation containing only the N-terminal IDR and the core folded DBHS region (residues 1–598; Fig. 2a). Scattering data from previous studies (Hewage et al., 2019; Koning et al., 2025; SASBDB entries SASDFK3 and SASDMG8; Kikhney et al., 2020) and some unpublished data (Supplementary Fig. S3) of protein variants lacking IDRs were used as a reference for comparison with IDR-containing data sets (Figs. 2a, 2f and 2g). The full set of scattering parameters according to the guidelines set out by Trewhella et al. (2023) are reported in Table 1.

Initial SEC-SAXS experiments were conducted in high-salt buffers due to the ability of this condition to prevent phase separation, a phenomenon which is not typically useful if one is interested in monodisperse structural information of proteins (Marshall et al., 2023). However, increasing concentrations of KCl also contain greater amounts of material that can burn onto the capillary under irradiation, potentially causing drifting/sloped baselines in the data. Early attempts at SAXS experiments with full-length DBHS proteins suffered from these problems and were unsuccessful (Yee Seng Chong, personal communication, unpublished data). Both SEC-SAXS experiments on SFPQ variants were performed in potassium nitrate buffers due to the ability of nitrate to function as a powerful radioprotectant through free-radical scavenging (Stachowski et al., 2021). Additionally, nitrate contributes to a lower scattering background compared with chloride, owing to the lower atomic scattering factors of nitrogen and oxygen compared with chloride.

Both SFPQ experiments in potassium nitrate produced CHROMIXS and UV chromatograms indicating the presence of a cleanly separated single peak for both constructs (Supplementary Fig. S4), with UV absorbance ratios at 260:280 nm that were consistent with that of a pure protein solution without any bound nucleic acid (0.62 for full-length SFPQ and 0.688 for SFPQ1–598). A small drift in R_g across eluted frames can be indicative of flexibility in the sample (Koenigsberg & Heldwein, 2018) as larger conformers typically elute first. The chromatogram for full-length SFPQ indicated a slight reduction in R_g across the peak from ∼83 to ∼78 Å (Supplementary Fig. S4). A total of 20 frames across the peak were averaged, where 18 frames varied between 83.4 and 77.1 Å and an additional two frames with R_g values 74.2 and 75.7 Å were also included. As examined later in the analysis using EOM, full-length SFPQ can be modelled by populations of conformers with R_g values between ∼70 and 100 Å, so a slight variation in R_g values upon elution is to be expected. This was not the case for the chromatogram of SFPQ1–598, which had some variation in R_g within the main elution peak, with a small peak surrounded by two more stable plateaus (Supplementary Fig. S4). The first plateau was chosen for averaging, which contained six frames which varied in R_g from 72.2 to 73.7 Å. Regions prior to protein elution were chosen as the frames for buffer subtraction in CHROMIXS (Supplementary Fig. S4).

Each experiment successfully produced monodisperse scattering with log(I) versus log(q) data sets (Figs. 2b and 2c) having linear fits to the Guinier region (Figs. 2d and 2e) when constrained to qR_g max = ∼1.1, which can be a necessary limit in accurately determining R_g for disordered proteins (Zheng & Best, 2018). The normalized residual plots of each Guinier fit indicated a reasonable degree of variation around the fit and a lack of any curvature (Figs. 2d and 2e, bottom). Comparative P(r) functions of the different variants from this study indicate the difference in size between an SFPQ dimer containing the structured region and additional variants containing the C- and N-terminal IDRs, which have much larger distributions (Fig. 2f). Interestingly, the distribution for SFPQ1–707 contains a small peak between 300 and 434 Å in the distribution which was initially assumed to be spurious. However, varying D_max around the tabulated values, fitting different q-ranges with GNOM and increasing the value of the GNOM regularization parameter (α) all allow the peak to persist (Supplementary Fig. S5). Typically, the stability of each distance distribution function with varying α is an additional criterion that can be used to assess the correctness of the solution (Svergun, 1992). Additionally, as small errors in D_max can change P(r) slightly (Grant et al., 2015), the persistence of features in the face of varying D_max serves as an additional indicator of the robustness of the solution. Upon further consideration, it is likely that this peak corresponds to vectors contributed by the C-terminal IDRs, which are anchored ∼270 Å apart and should naturally contribute many pairwise distances at r > 270 Å (Supplementary Fig. S5). As expected, this peak is not present in the data for SFPQ1–598 (the variant lacking the C-terminal IDRs), which reaches r = 0 roughly where the peak begins (Fig. 2f), supporting this point.

For both SFPQ1–707 and SFPQ1–598 the Guinier- and P(r)-derived R_g estimations were in agreement and within error margins of each other (Table 1). For SFPQ1–707 the lowest q point measured was 0.00506 Å⁻¹ and for SFPQ1–598 it was 0.0047 Å⁻¹. Considering the important limit of q_min < π/D_max (Kikhney & Svergun, 2015) and the D_max values of 434 and 344 Å from our P(r) analysis of SFPQ1–707 and SFPQ1–598, respectively, the data are within the appropriate q-range to resolve species of this size. In theory, this limit and our data range should allow proteins with a maximum size of 620 and 668 Å to be analysed, which is well beyond the size of the longest conformers that have been used to fit the data in our ensemble modelling (474.65 Å for SFPQ1–707 and 362.98 Å for SFPQ1–598; see next section).

The dimensionless Kratky plots for these experiments indicate a progression from globular to partly rod-like to flexible upon the addition of either the N-terminal or both IDRs (Fig. 2g), corroborating the predictions that both the C- and N-terminal IDRs are long, flexible and disordered.

Given the predicted disordered regions in these proteins and their dimensionless Kratky plots in this study, we used the ensemble-modelling program EOM (Tria et al., 2015), which creates an initial pool of realistically flexible models using a subprogram called RanCh (Random Chains). The scattering profiles of these models are calculated using FFMAKER and a genetic algorithm (GAJOE) searches the initial random pool of models for ensembles which together fit the data. The statistics of the best ensembles (usually 50–100 different ensembles) are then pooled and the distribution of their R_g and D_max values are compared with the initial random starting pool of models for insight into compaction, flexibility and conformational state. In our analysis, we have termed the ensembles that fit the data simply as `ensemble' and the random initial pool as `starting pool'. The resulting models generated by EOM had excellent fits to the data, as indicated by near-ideal reduced χ² values for both SFPQ and SFPQ1–598 of 1.04 and 1.012, respectively (Figs. 3a and 3f). The error-weighted residual plots for both experiments also reflected agreement with the experimental data (Figs. 3b and 3g), with no syste­matic variation observed. For full-length SFPQ the filtered selected ensemble of models that fit the data (`ensemble') had an average R<sub>g</sub> that was more compact than that of the initial random ensemble of models (`starting pool') that was generated by EOM. This can be seen by the R_g ensemble distribution that fits the data shifting left relative to the RanCh pool (Fig. 3c). The geometric average of the R_g distribution for the selected ensemble was 82.58 Å compared with 91.08 Å for the random starting pool, again indicating compaction.

Comparison of the D_max distributions for the starting pool and the ensemble indicate that they are very similar (Fig. 3d), possibly because of a high number of degrees of freedom in the model. Full-length SFPQ has four disordered domains, perhaps meaning that larger distances can still be reached whilst the average R_g can simultaneously become smaller. R_flex of the system is calculated to be ∼74.29%, compared with that of the random pool which is ∼81.12%, with an R_σ of 0.83, i.e. below unity. These results taken together indicate a degree of compaction in the models that fit the data, compared with the initial random pool of conformers, supporting the notion that full-length SFPQ experiences some compaction in solution (Fig. 3e).

3.2. Removal of the C-terminal IDR of SFPQ abolishes the preference for chain compaction

For comparison with the EOM data on full-length SFPQ, EOM was additionally run on the data for SFPQ1–598. However, for SFPQ1–598 the ensemble that fits the data reproduces much of the middle region of the random starting pool in terms of R_g (Fig. 3h), with the geometric average R_g values of the selected ensembles and the starting random pools being 70.92 and 73.66 Å, respectively. The distribution of D_max values selected to fit the data also appears to reproduce the dimensions of the random pool (Fig. 3i). The highest frequency distance is ∼270 Å, as this is the arm-to-arm distance between the long helices in the structured parts of SFPQ. R_flex and R_σ reveal that the selected ensembles are not accessing the full conformational space of the random starting pool [R_flex = 62.04% (∼83.61%) and R_σ = 0.51]. This is likely to be because the tail ends of the RanCh R_g distribution do not overlap with the selected pool (Fig. 3h), perhaps because the tail ends of the distribution represent more extreme cases of compaction or extension, which do not occur in solution. However, what is obvious is that much of the distribution of ensemble R_g values appears to overlap with the middle of the starting-pool distribution for SFPQ1–598 but is significantly shifted to the left for full-length SFPQ (Figs. 3c and 3h). This is also reflected in the geometric average R_g values of the respective pools.

A possible explanation for the compaction of the chosen ensembles compared with the starting random pool in the full-length SFPQ experiment is that, as hypothesized by Marshall et al. (2023), the N- and C-terminal IDRs directly interact. This would explain why SFPQ1–598, a variant which lacks the C-terminal IDR, reproduces elements of the random pool more closely and shows less of a preference for more compact conformations. Inspecting the EOM models reveals that some models place the C- and N-terminal IDRs in relatively close proximity to one another or show compaction of the C-terminal IDRs back onto the dimer (Fig. 3e). Given that EOM models disordered chains realistically using a C^α–C^α Ramachandran distribution in line with that of disordered proteins as well as the user-supplied amino-acid sequence (Tria et al., 2015), this suggests that it is physically and sterically possible for the IDRs to interact directly. The notion of a direct interaction between the IDRs is also a possible explanation for an additional SANS measurement of full-length SFPQ in a low-salt buffer (150 mM KCl) that produced an interesting P(r) function that appears to be contracted, with a shoulder, compared with full-length SFPQ in high-salt conditions (see Section 3.4; Fig. 5g).

3.3. The N-terminal IDR of SFPQ collapses at a physiological salt concentration

Given the drastic effect of salt concentration observed by Marshall et al. (2023), some experiments on these proteins under different salt conditions were carried out to probe whether charge screening had any measurable effect on the structure of SFPQ. A SEC-SAXS experiment performed on SFPQ1–598 in a lower salt buffer yielded results which differed from the high-salt buffer. SFPQ1–598 is evidently capable of running over a size-exclusion column in a low-salt buffer to some extent, as indicated by the SEC-SAXS data in Fig. 4(a) and Supplementary Fig. S6. EOM models indicated good agreement with the experimental data, with a reduced χ² of 0.891 and an error-weighted residuals plot indicating no systematic variability of the fit against the data (Figs. 4a and 4b). These data had a lower Guinier R_g than that of SFPQ1–598 in the 500 mM KNO₃ buffer (Fig. 4c, Table 1), with acceptable fit parameters and overall distribution of residuals (Fig. 4d). A comparison of the P(r) functions between SFPQ1–598 in high- and low-salt conditions reveals a difference in the size of their distributions, whilst maintaining the same approximate shape (Fig. 4e). This difference between salt conditions is also echoed in the dimensionless Kratky plots of both experiments, which reveal that the low-salt condition appears to be more folded compared with the high-salt condition (Fig. 4f). The EOM analysis of this condition also supports this, with the R_g distribution of the selected ensembles shifting significantly to the left compared with that of the random starting pool, with average R_g values for the selected and random pools of 64.92 and 73.74 Å, respectively. For these data R_flex(random)/R_σ = ∼70.17 (∼82.68%)/0.80, indicating that the selected ensemble pool does not equally sample the entire conformational space of the random starting pool. Visual inspection of the models also appears to show bunching of the N-terminal IDRs around the dimer core (Fig. 4h). Taken together, the results indicate that changing the buffer from 500 mM KNO₃ to 150 mM KCl induces some form of compaction. However, we cannot exclude that the change of counterion could contribute to differences, alongside the change in ion concentration.

Analysis of the sequence of the N-terminal IDR indicates a multitude of basic and acidic amino acids (Fig. 4i). Additionally, an electrostatic potential map of an SFPQ dimer indicates several pockets of positive and negative charge on the surface of the dimerization region as well as the coiled-coil domain (Fig. 4j). Given that the protein noticeably becomes more compact, likely due to reduced charge screening, it may be that these pockets of charge on the surface of the DBHS domain or the alternating regions of charge on the N-terminal IDR allows the collapse of the IDR into a more compact state. This could occur through self-interaction either with the folded DBHS domain or inter-residue contacts within the N-terminal IDR.

3.4. Small-angle neutron scattering demonstrates that full-length SFPQ can exchange dimeric partners in vitro

Contrast-matching SANS experiments were performed in an attempt to measure the shape of SFPQ in the condensed phase. In our match-out experiment, a significant amount of deuterated protein (95% by ratio) was mixed with a small amount of protiated protein (5% by ratio). This experiment was carried out at the match-point of the deuterated protein (95% D₂O) such that buffer subtraction should in theory eliminate any contributions to the data from the deuterated protein. Observing SFPQ as a monomer in this instance would indicate dynamic partner exchange between the deuterated proteins, as at this concentration and much lower (see Fig. 5) SFPQ typically exists as a dimer. Whilst our experiments attempting to study SFPQ inside droplets ultimately failed to produce monodisperse scattering, the assumption that full-length protiated SFPQ and deuterated SFPQ could exchange partners with each other was shown to be correct via a match-out experiment (Figs. 5a, 5g and 5h).

In addition, a data set was collected on a mixture of deuterated and protiated dimers of SFPQ without any match-out conditions, yielding a P(r) function that could be directly compared with that of full-length SFPQ collected via SEC-SAXS (Figs. 5d and 5g). For the monomer experiment, the data indicated a linear Guinier fit that passed through the error bars of all chosen data points with an R_g of 61.06 ± 3.55 Å (Fig. 5b) and had an acceptable amount of variation around the fit (Fig. 5c). For the dimer experiment the data also indicated a linear Guinier fit passing through all of the error bars, except for the presence of one point which seemed to deviate from linearity and was likely to be an experimental outlier (Figs. 5e and 5f). This data produced a Guinier R_g of 86.55 ± 4.29 Å (Fig. 5e), which more or less agrees with the Guinier-derived R_g of full-length SFPQ from SEC-SAXS (Table 1 and Fig. 2). The Guinier fit of these data had an acceptable amount of variation around the fit (Fig. 5f).

However, a comparison of our SAXS and SANS data sets on full-length SFPQ revealed the P(r) functions from SANS to be shorter than that of the full-length protein as observed with SAXS and to also represent different asymmetric shapes (Fig. 5g). The monomer scattering data set produced a P(r) function much smaller than both other data sets, as shown by the smallest maxima in P(r) at ∼32.5 Å and a D_max of 228 Å (Fig. 5g), which is close to half of the D_max of SFPQ as seen with SAXS (Fig. 5g). Additionally, an atomistic model of an SFPQ monomer missing the IDRs conformed reasonably well to the shape of the DAMAVER envelope derived from the monomer scattering data (Fig. 5h). These data indicate that protiated and deuterated full-length SFPQ are capable of swapping partners dynamically in solution to reach a population of protiated monomers of SFPQ as the predominant scatterer in solution, surrounded by an excess of matched-out deuterated SFPQ. To determine whether deuterated SFPQ was appropriately matched-out in the context of these experiments, measurements were taken of dSFPQ at its match-out point of 95% D₂O, which yielded scattering consistent with that of the background (Supplementary Fig. S7). Interestingly, a comparison between full-length SFPQ dimers as observed with SEC-SAXS or SANS yields different P(r) functions. The function for SFPQ from SEC-SAXS has a maximum at ∼55 Å and a D_max of 434 Å, whereas the function from SANS has a maximum at ∼71 Å and a D_max of 307 Å.

Given that both functions are of a reasonable quality (Table 1) and we have demonstrated that the peak in the function for full-length SFPQ between 300 and 434 Å is likely to be a real structural feature, this could be another case of compaction of the protein due to differing experimental conditions. The reduction in D_max in the low-salt SANS condition and the broadening of the main peak compared with the function derived from SEC-SAXS in high salt is likely to represent contraction or interaction of the IDRs due to reduced electrostatic screening.

3.5. Lysine cross-linking mass spectrometry (XL-MS) shows that the N- and C-terminal IDRs both contact the core DBHS region

In order to obtain information on the intramolecular interactions in play in the context of a dimer of SFPQ, lysine cross-linking mass-spectrometry experiments were performed using full-length SFPQ and SFPQ1–598. The results showed a large number of cross-links forming between the different parts of both protein variants (Figs. 6a and 6b). The significant number of cross-links between regions 276–598 in both data sets is consistent with the large number of lysines close to each other in the structured DBHS region of an SFPQ homodimer (Lee et al., 2015; Figs. 6a and 6b). To minimize the possibility of self-interaction of SFPQ via the coiled-coil domain (Koning et al., 2025; Lee et al., 2015), the experiments were performed at a low concentration (a 10 and 20 µM experiment for each protein variant), which could still provide an interpretable signal for XL-MS. Our additional static SANS measurement supports the notion that at around this concentration dimers are likely to be the only species in solution (Fig. 5g). The only difference is that the initial dilution step for SFPQ in XL-MS was performed in a 500 mM KCl buffer, which we would expect to further inhibit self-interaction of the protein, rather than the 150 mM KCl buffer for SANS.

Outside of the DBHS domain, cross-links were made between the distal lysines in the C-terminal IDR, the coiled-coil domain and parts of the folded domain, even a partially buried lysine in the dimerization domain (Figs. 6a and 6c). Additional cross-links were also made between the N-terminal IDR and the DBHS domain (Fig. 6a). Both the C-terminal and N-terminal IDRs appear to contact positions on the folded domain in close proximity to one another (Fig. 6a). Despite cross-links not being detected between the two IDRs directly, this demonstrates that there is some overlap in the conformational space that both IDRs can access. Unexpectedly, the N-terminal IDR cross-links with the folded domain disappear in the SFPQ1–598 experiment (Fig. 6b). This could be due to the disruption of an interaction between the C- and N-terminal IDRs which in the full-length protein causes the N-terminal IDR to sample more conformational space near the DBHS domain. A caveat of this experiment is that parts of the N-terminal IDR of SFPQ are highly enriched in proline, which may have resulted in the poor tryptic digest of the region at the distal end of the N-terminus, which did not form any cross-links in either experiment. In theory, this could have led to the subsequent lack of detection of some peptides involving the N-terminal IDR due to a larger undigested mass.

3.6. Human DBHS protein sequence bias, enrichment and depletion analysis

To further explore the possible role of combinatorial dimerization and the different DBHS IDRs in the control of LLPS, we analysed the sequence bias and enrichment/depletion of certain amino acids in the different regions of the three human DBHS paralogs (Fig. 7). The analysis reveals some striking differences in the compositional bias across the IDRs of all of the paralogs. Comparative plots indicate relatively conserved enrichment of amino acids in the DBHS domain across all the paralogs, but highly variable composition across the N- and C-terminal IDRs of the different paralogs (Fig. 7). Currently, the sequence contribution of each human DBHS paralog to LLPS is poorly understood. These differences and their potential relevance to phase separation and the material properties of different dimeric combinations are discussed in Section 4.

4.1. The structure of the N- and C-terminal IDRs and their biological relevance

Our data indicate that as per the predictions, the N- and C-terminal regions outside of the DBHS domain of SFPQ are highly flexible in solution and intrinsically disordered. The realistic modelling of the disordered IDRs by EOM, combined with the large number of interconverting states that disordered proteins can naturally sample (Holehouse & Kragelund, 2024), means that it is likely to be physically possible for the IDRs to come into close proximity to one another and interact (Fig. 8a). An interaction between the N- and C-terminal IDRs in SFPQ is a possible explanation for the relative compaction of full-length SFPQ as seen with EOM. This interaction, which might be more pronounced at a physiologically relevant salt concentration due to reduced charge screening of the IDRs, might also explain the SANS P(r) function, which has a larger peak maxima of ∼71 Å, a broader shoulder in the distribution at ∼150–200 Å and a far shorter D_max of 307 Å compared with the function derived from the SAXS data (Fig. 5g). An interaction between the IDRs is in line with the physical model proposed by Marshall et al. (2023), where the C- and N-terminal IDRs were proposed to interact directly to modulate LLPS (Figs. 8a and 8b). However, it is not necessarily possible to delineate between the compaction of the IDRs or a direct interaction between the two IDRs within the context of this study. Compaction of the C-terminal IDR is also a possibility, given that many points of contact were made between the C-terminal IDR and the DBHS domain in the XL-MS experiments. Perhaps both compaction of the respective IDRs and a direct interaction between them are effects that can occur simultaneously, and these become more exaggerated at a physiological salt concentration due to reduced charge screening. Given the high proline content of the N-terminal IDR (Figs. 7a and 7b), it is also possible that just through the inclusion of the N-terminal IDR on the same structure, and not through direct interaction with the C-terminal IDR, phase separation is hindered due to the well known role of proline as a solubilizing amino acid, which promotes solvation rather than intra-chain interactions (Borcherds et al., 2021). However, our modelling suggests perhaps otherwise and we have additionally observed the collapse of the N-terminal IDR in low-salt conditions, likely because of intra-chain interactions or interactions with the folded domain.

There have been instances of IDRs which seem extended in high-salt conditions then interacting with pockets of charge on folded RRMs as a result of reduced charge screening (Martin, Thomasen et al., 2021). It is possible that both the N- and C-terminal IDRs make electrostatic interactions with the pockets of charge on the DBHS domain, creating a complicated balance of direct interactions between the N- and C-terminal IDRs and also the DBHS domain. Given that a symmetry exists between intra-chain interactions in LLPS and inter-chain interactions (Martin, Thomasen et al., 2021), it is possible that sticker regions (Borcherds et al., 2021) in the N-terminal IDR cause its collapse and so are also important residues which also may interact with the C-terminal IDR (Figs. 8c and 8d). In this study, we have not attempted to delineate between the interaction of the N-terminal IDR with itself or the folded domain as a cause for its collapse. However, this is worth examining in the future.

Parts of the N-terminal IDR have been shown to be necessary for binding dsDNA (Lee et al., 2015; Song et al., 2005; Wang et al., 2022). This was initially investigated by Urban et al. (2002), who attempted to probe DNA binding through truncations of the N-terminus of SFPQ and concluded that the entire N-terminal IDR could bind DNA. Later studies (Lee et al., 2015; Wang et al., 2022) have concentrated the DNA-binding ability of SFPQ to a smaller region within the N-terminal IDR between residues 214 and 298, putatively dubbing it the `DNA-binding' domain. This notion was strengthened by the presence of RGG/RG motifs within the DNA-binding domain (DBD), which are commonly observed in nucleic acid binding (Chong et al., 2018). However, the distal N-terminal part of the IDR outside the DBD also contains RGG/RG motifs and it is currently unclear whether these are also involved in nucleic acid binding. It is a possibility that RGG tracts outside of the putative DBD in SFPQ are also involved in nucleic acid binding, as in the protein FUS the inclusion of additional disordered RGG motifs to restore mutants of FUS to wild-type FUS enhanced the affinity of the protein for RNA (Ozdilek et al., 2017). The collapse of the N-terminal IDR that was observed in our modelling is perhaps also relevant to nucleic acid binding. Disordered DNA-binding domains can fold into a more structured conformation when interacting with DNA either via the large-scale folding of entire domains or of more local loops and motifs (Dyson & Wright, 2005). This may be the case for the interaction of the DBD of SFPQ with certain dsDNA targets (Figs. 8d and 8e) and may be how a low-affinity interaction might stabilize into an interaction with more specificity.

Presuming that the N- and C-terminal IDRs interact directly, it is possible that phase separation might be modulated through further direct interaction of a larger part of the N-terminal IDR (in place of just the DBD) with nucleic acids (Fig. 8e). The binding of a nucleic acid, such as DNA, or larger structured RNA might sequester the N-terminal IDR and leave the C-terminal IDR, the main driver of LLPS (Marshall et al., 2023) free for interactions with other components (Fig. 8e). Nucleic acids in this sense might act as a further driver or an `on-switch' for phase separation through steric sequestration of the N-terminal IDR, as was also hypothesized by Marshall et al. (2023). This is potentially relevant for the assembly of the initial NEAT1–SFPQ RNP, where SFPQ initially binds core parts of NEAT1 following transcription (West et al., 2016; Yamazaki et al., 2018), possibly using both the N-terminal IDR and the RRMs. In theory, this would then free the C-terminal IDR to promote phase separation with other unbound DBHS proteins and seed paraspeckle formation. The competition of nucleic acid targets for SFPQ, which has been demonstrated by Song et al. (2005) and Wang et al. (2022) between VL30 RNA and the GAGE6 oligonucleotide, may also play a role in regulating LLPS. Substituting one target for another might influence the occupancy of the N-terminal IDR and therefore additionally modulate LLPS. Further experimental work is required to assess whether the direct inclusion of nucleic acid targets in LLPS assays can influence the saturation concentration of SFPQ (Marshall et al., 2023).

4.2. Dimer swapping and relevance to phase behaviour

Our study is the first instance in which partner exchange has been demonstrated between full-length homodimers of a DBHS protein, which is remarkable considering the intimate nature of the dimerization core and the extensive set of interactions that make up the dimerization interface (Huang et al., 2018; Lee et al., 2022; Passon et al., 2012) seen in all prior DBHS protein crystal structures. The data shows that full-length SFPQ is capable of swapping partners with itself without the need for cofactors in vitro. Structurally, this may occur due to the inherent flexibility/disorder observed in parts of DBHS structures such as the NOPS domain, which could indicate that the coiled-coil domain unravels first and the rest of the structure naturally unfolds and refolds when it finds another partner (Knott et al., 2022; Lee et al., 2022). Combinatorial dimerization is likely to serve many purposes such as the differential recognition of nucleic acid targets and protein partners, or as a compensatory mechanism (Lee et al., 2022; Huang et al., 2018).

Our amino-acid composition analysis across the human DBHS paralogs indicates a significant difference between the C- and N-terminal IDRs (Fig. 7). Strikingly, the N-terminal IDR of SFPQ has significant tracts which vary between being 40% and 55% proline (Figs. 7a and 7b). In comparison, the much shorter N-terminal IDRs of NONO and PSPC1 have proline tracts which are closer to ∼25% proline (Figs. 7d and 7f). Interestingly, both the N-terminal IDRs of NONO and PSPC1 are depleted in glycine, which is enriched in SFPQ, which contains glycine-rich tracts (Fig. 7b). Given the roles of proline in chain expansion and solubility (Borcherds et al., 2021; Lotthammer et al., 2024), the N-terminal IDR of SFPQ is perhaps more expanded and soluble than the other N-terminal IDRs. This could be further enhanced by its enrichment in glycine, which is known to contribute to IDR flexibility due to the conformational flexibility of the peptide bond (Wang et al., 2018). Combined with the ∼275 amino-acid length of the N-terminal IDR of SFPQ, the end result is a relatively long, expanded, disordered chain that samples many conformations in space, which is reflected in our modelling data. This is likely relevant to the role of the N-terminal IDR as a nucleic acid-binding domain as well as for interactions with the C-terminal IDR, as increased flexibility and expansion may allow a wider sampling of conformational space, plasticity in the selection of nucleic acid targets and increased contact in solution with the C-terminal IDR to regulate LLPS. Conversely, the lower enrichment of proline, and the depletion of glycine in the N-terminal IDRs of the other paralogs, combined with their significantly shorter length, would contribute to less flexible, compact, IDRs. Intradimer interactions between the N- and C-terminal IDRs may be entirely absent from the other paralogs due to diminished flexibility, shorter domain length and the presence of hydrophobic residues such as alanine (Fig. 7f), which may instead promote self-interaction (Holehouse & Kragelund, 2024) or work to hinder phase separation. Another striking difference is the enrichment of histidine in the N-terminal IDRs of NONO and SFPQ, which is depleted in the N-terminal IDR of PSPC1 (Fig. 7). Histidine is capable of π–π stacking and cation–π interactions; in the right context (Liao et al., 2013) this may be important for interaction with tyrosines, which are enriched in the C-terminal IDR of SFPQ. Recently, King et al. (2024) identified pH-gradient differences across nuclear condensates, with the nucleolus reportedly containing regions of pH 6.5. It is possible that such an effect is more drastic in paraspeckles or other DBHS condensates, and so histidines may contribute to pH sensing in the DBHS IDRs because of their variable protonation state in response to pH.

4.3. Compositional differences in the C-terminal IDR;relevance to LLPS

Comparing the C-terminal IDRs of the paralogs also reveals some striking differences, which are of interest given that the C-terminal IDR may be the driver of phase separation for all of the paralogs (Marshall et al., 2023). SFPQ is enriched in tyrosine (Fig. 7a), which is known to act as a sticker via π–π and cation–π interactions (Bremer et al., 2022). This, in theory, could contribute to a more compact C-terminal IDR via π–π and cation–π mediated collapse of the chain (Holehouse & Kragelund, 2024) or an interaction with the histidines or arginines in the N-terminal IDR. Strikingly, tyrosine is depleted in the other paralogs, but phenylalanine, which is also capable of π–π and cation–π interactions (Bremer et al., 2022), is slightly enriched only in NONO (Fig. 7d). Glycine and proline are enriched in all of the C-terminal IDRs, suggesting chain expansion and flexibility (Lotthammer et al., 2024). Rather strikingly, alanine tracts feature in both NONO and PSPC1 (Figs. 7d and 7f), but are entirely absent from SFPQ, in which the amino acid is depleted. The departure from tyrosine enrichment in SFPQ to alanine enrichment in NONO and PSPC1 may indicate some reliance on hydrophobic inter­actions for LLPS in NONO and PSPC1 and on π–π and cation–π interactions in SFPQ. Alternatively, alanine tracts may act to hamper phase separation due to their relatively chemically inert nature. An additional interesting point is the conservation and significant enrichment of methionine tracts in the C-terminal IDRs of all of the DBHS proteins (Fig. 7). Methionine has documented roles in LLPS via its conversion to methionine sulfoxide in response to reactive oxygen species (Aledo, 2021; Kato et al., 2019). Given the documented roles of paraspeckles in stress response (McCluggage & Fox, 2021), methionine sulfoxidation in the DBHS protein IDRs might present a chemical mechanism by which paraspeckles can respond to oxidative stress via post-translational oxidative changes to solvent-exposed methionine tracts. This is particularly interesting given that the methionine enrichment is localized to all of the DBHS C-terminal IDRs, which in SFPQ is the region considered to be the main driver of phase separation. Methionine sulfoxidation would likely alter the chemical properties of the IDR, and as a response alter the phase-separating abilities of all of the DBHS proteins, either promoting or hindering paraspeckle formation. The contributions of the DBHS IDRs to phase separation are complicated and involve many interrelated principles and effects. However, it is likely that dynamic homodimerization and heterodimerization with partner swapping serves to contribute different IDRs to the mixture of interactions that can trigger phase separation and so modulate the material properties of condensates or their occurrence in vivo (Figs. 9a and 9b). Further experiments are required to decode the relationship between the sequence composition of IDRs, folded domain behaviour and phase separation.

4.4. Disease-associated cysteine mutants in the C-terminal IDRs

As examined in our previous study (Koning et al., 2025) cysteine mutations in the coiled-coil domain of SFPQ have been shown to cause disulfide oligomerization of the protein. We deduced that due to their structure and flexibility, it might be possible for a variety of cysteine mutations in the C-terminal IDRs of DBHS proteins to also cause disulfide-bound aggregates, which could, in theory, contribute to disease (Fig. 9c).

We have identified numerous cysteine mutants in the C-terminal IDRs of SFPQ, NONO and PSPC1, which we propose may contribute to disease (Supplementary Table S1). Our XL-MS experiments indicate that the C-terminal IDR of SFPQ makes points of contact with the folded DBHS domain, the lysines in which are very close to the solvent-exposed reactive cysteine in NONO C145 (Kathman et al., 2023). Given the approximate length conservation of the C-terminal IDR between SFPQ and NONO and the longer IDR of PSPC1, it is likely to be possible that all of the paralog IDRs are capable of contact with the DBHS domain. Combined with a capacity for dimer exchange, an SFPQ cysteine IDR mutant might contact the solvent-exposed cysteine in NONO, for example (see the cartoon in Fig. 9c). A cysteine mutation near the middle of the C-terminal IDR of NONO (Reinstein et al., 2016) has a reported causative role in intellectual disability, presumably through disulfide-bridge formation (Fig. 9c). These ideas may be relevant for other disease states associated with cysteine mutants in the C-terminal IDRs of human DBHS proteins, given the involvement of NONO in certain cancers (Feng et al., 2020) and the role of SFPQ as a tumour suppressor (Song et al., 2005).