HNF-1A protein sequences from various model organisms were extracted from the UniProt (UniProt, 2021) database under the following accession codes: Homo sapiens (human): P20823, Mus musculus (mouse): P22361, Gallus gallus (chicken): Q90867, Salmo salar (Atlantic salmon): Q91474, Xenopus laevis (African clawed frog): Q05041, Danio rerio (zebrafish): Q8UVH4.

Disorder predictions of full-length human HNF-1A (HsHNF-1A) were performed by using the IUPred3 (Erdős et al., 2021), ANCHOR2 (Mészáros et al., 2009) and the Fast Estimator of Latent Local Structure (FELLS) (Piovesan et al., 2017) webservers. The FELLS output was further utilized to investigate sequence composition, such as residue hydrophobicity, charge, and sequence compositional bias. AlphaFold (Jumper et al., 2021; Varadi et al., 2022) was employed for a tertiary structure prediction of full-length HsHNF-1A. The predicted structure (AF-P20823-F1-model_v4.pdb) with residue specific predicted local distance difference test (pLDDT) scores and the corresponding predicted aligned error plot were retrieved from the AlphaFold Protein Structure Database (https://alphafold.ebi.ac.uk/entry/P20823). The FuzDrop algorithm (Hatos et al., 2022) was used to predict the tendency of HsHNF-1A to undergo LLPS. A multiple sequence alignment of the HNF-1A TAD (HsHNF-1A, residues 280–631) was generated using the Clustal Omega alignment tool (Sievers and Higgins, 2021; Madeira et al., 2022) and visualized using ESPript3.0 (Robert and Gouet, 2014). Potential short linear motifs (SLiMs) within the TAD were retrieved from the Eukaryotic Linear Motif (ELM) resource database (Kumar et al., 2022). The probability of HNF-1A to contain acidic activation domains was predicted by the “Predictor of Activation Domains using Deep Learning in Eukaryotes” (PADDLE) algorithm (https://paddle.stanford.edu).

A mammalian pcDNA3.1/HisC expression vector harboring the cDNA sequence of full-length HsHNF-1A (residues 1–631, UniProt ID P20823, isoform A, Ensembl transcript ID: ENST00000257555.11) (Bjørkhaug et al., 2003) was used to generate a bacterial expression construct for DBD-TAD (residues 83–631) by Gateway^® cloning technology (Katzen, 2007). Two PCR reactions were conducted in order to introduce an N-terminal Tobacco-Etch Virus protease site (ENLYFQG) and attB1/2 sites required for recombination. A BP reaction was performed to transfer the gene sequence into the pDONR221 entry vector (Invitrogen). A subsequent LR reaction was conducted for DNA insert transfer into the pTH27 (Hammarström et al., 2006) destination vector, generating a bacterial expression plasmid encoding for a His₆-TEV-DBD-TAD construct (hereafter DBD-TAD). Sequence identity and integrity were verified by plasmid DNA sequencing. Primers used in cloning and sequencing are listed in Table 1.

DBD-TAD was recombinantly expressed in Escherichia coli Rosetta (DE3). Bacteria were cultured in LB medium at 37°C until an OD₆₀₀ value of 0.6–0.8 was reached. Protein expression was induced by the addition of 1 mM IPTG and conducted at 18°C for 20 h. Bacteria were harvested by centrifugation (6,000 × g, 10 min, 4°C) and resuspended in lysis buffer containing urea as denaturation reagent (50 mM Tris pH 8.5, 500 mM NaCl, 6 M urea, 10 mM imidazole, 1 mM DTT, 1 mM PMSF, 1× cOmplete EDTA-free protease inhibitor cocktail). The sample was ultrasonicated (7 min, 1 s on/off cycles, 25 W), and the lysate was clarified by centrifugation (16,000 × g, 30 min, 4°C).

The soluble fraction was loaded onto a pre-equilibrated Ni-NTA column with a 2-mL bed volume. The column was washed with 15 column volumes (CVs) of wash buffer (50 mM Tris pH 8.5, 500 mM NaCl, 6 M urea, 10 mM imidazole, 1 mM DTT, 1 mM PMSF), before elution was performed using a stepwise imidazole gradient (40 mM, 60 mM, 80 mM, 300 mM imidazole) and a fraction volume of 3 CVs. Fractions containing the protein of interest were pooled and dialyzed into a buffer lacking urea (20 mM Tris pH 8.5, 500 mM NaCl, 1 mM DTT), using dialysis tubing with a 12–14 kDa molecular weight cut-off (Spectra/Por). The dialyzed sample, containing the refolded His-tagged DBD-TAD protein, was concentrated and subjected to size exclusion chromatography (SEC). A Superdex 200 10/300 GL Increase column (GE Healthcare) was used at a flow rate of 0.5 mL/min with a running buffer containing 20 mM Tris pH 8.5, 500 mM NaCl, 1 mM TCEP. Pooled SEC fractions containing His-tagged DBD-TAD were concentrated, snap-frozen in liquid N₂, and stored at −80°C.

The DBD of HNF-1A (residues 83–279) was recombinantly expressed and purified as previously described (Kind et al., 2022) and used for control purposes.

SRCD experiments were performed on the AU-CD beamline, ASTRID2 synchrotron (Aarhus, Denmark). DBD and DBD-TAD were dialyzed into SRCD buffer (20 mM Na phosphate pH 7.8, 150 mM NaF, 1 mM TCEP) prior to measurements. Samples with various 2,2,2-trifluoroethanol (TFE) concentrations were prepared by mixing with 100% TFE (Sigma-Aldrich T63002). All samples were measured at an HNF-1A concentration of 0.5 mg/mL, except for DBD-TAD in the presence of 70% TFE, where the HNF-1A concentration was 0.45 mg/mL. SRCD measurements were performed at 10°C, using 0.1 mm quartz cuvettes (Hellma Analytics). Three wavelength scans per sample were recorded, applying a scan range of 170–280 nm with a 1-nm step size. Data processing was done using Microsoft Excel and CDtoolX (Miles and Wallace, 2018). Spectrum deconvolution was performed by using the BeStSel algorithm (Micsonai et al., 2015; Micsonai et al., 2018) with a spectral range of 180–250 nm (https://bestsel.elte.hu/index.php) and by using the DichroIDP program with the IDP175 reference dataset (Miles et al., 2023). SRCD datasets can be accessed via zenodo.org (DOI: 10.5281/zenodo.8328701) and in the PCDDB (Ramalli et al., 2022) under the series codes CD0006461 (HNF-1A DBD-TAD) and CD0006462 (HNF-1A DBD).

Synchrotron SAXS experiments for DBD-TAD were performed on the CoSAXS beamline (Kahnt et al., 2021), MAX IV Laboratory (Lund, Sweden). DBD-TAD was measured at ∼2.5 mg/mL in batch mode (4 mM Tris pH 8.5, 100 mM NaCl, 1 mM TCEP). Measurements were performed at 10°C at a wavelength of 1 Å (12.4 keV). Data were collected over an angular range of 3.8 × 10⁻⁴–6.4 × 10⁻¹ Å⁻¹ with 300 frames at an exposure time of 20 ms per frame. Final sample and buffer scattering curves were obtained by averaging the frames that matched each other to avoid incorporation of radiation damage.

Data reduction and analysis were done using the ATSAS 3.2.1 package (Manalastas-Cantos et al., 2021). Analyses of the Guinier region, the particle distribution function p(r), and molecular weight estimates were calculated using PRIMUSqt. Additional molecular weight estimation was performed using the SAXS MoW 2.0 method (http://saxs.ifsc.usp.br, integration upper limit q_max = I(0)/I(q_max) = 10^2.25) as well as by using the Debye formalism, essentially as described in (Raasakka et al., 2019) and according to original methodology (Calmettes et al., 1994; Fitzkee and Rose, 2004; Bernadó and Blackledge, 2009).

The ensemble optimization method (EOM) algorithm was used to generate an ensemble of models satisfying the SAXS data (Bernadó et al., 2007; Tria et al., 2015). The crystal structure of the DBD [PDB: 1IC8, (Chi et al., 2002)] was used to provide rigid bodies for POU_S (residues 87–180) and POU_H (residues 209–276). The EOM algorithm was run for native-like behavior of disordered regions, with the POU_S domain being fixed in position. The EOM conformers were visualized using PyMOL (DeLano, 2002).

Parameters used in SAXS data collection, analysis, and modelling for both the DBD-TAD and the previously published DBD dataset (Kind et al., 2022) are listed in Supplementary Table S1. SAXS datasets were submitted to the SASBDB (Valentini et al., 2015) and are available under the accession codes SASDS29 (HNF-1A DBD-TAD) and SASDSZ8 (HNF-1A DBD).

Purified DBD-TAD and DBD were diluted in storage buffer (20 mM Tris pH 8.5, 500 mM NaCl, 1 mM TCEP), yielding a final concentration of 10–25 μM. PEG8000 was added to a final concentration of 10% w/v. For each protein concentration, a 10-µL sample was prepared on a microscopy slide and differential interference contrast (DIC) images were acquired using a Zeiss Axiovert 200M wide-field fluorescence microscope equipped with a LD Plan-NEOFLUAR 40×/0.6 Ph2 objective and an AxioCam HR camera. Image acquisition was performed using the AxioVision software (Carl Zeiss, version 4.5).

MIN6 cells (Miyazaki et al., 1990; Poitout et al., 1996), kindly provided by Professor Claes Wollheim (Lund University, Sweden), were cultured in DMEM growth medium (Gibco), supplemented with 15% fetal bovine serum and 1% v/v penicillin-streptomycin (Sigma-Aldrich P4333). Cells were incubated at 37°C and 5% atmospheric CO₂ in a humidity incubator. For immunofluorescence (IF), MIN6 cells were seeded on coverslips in 24-well plates. IF staining was performed at room temperature (RT). 24 h post-seeding, cells were washed twice with phosphate-buffered saline (PBS) and fixed by incubation in a phosphate buffer containing 3% paraformaldehyde (Sigma-Aldrich) for 25 min. Samples were washed three times with PBS, permeabilized in 0.1% Triton X-100 (Sigma-Aldrich) in PBS for 10 min, and washed three times with PBS. Binding epitopes were blocked by incubation with 2% normal goat serum (GS, Invitrogen)/8% bovine serum albumin (BSA, Sigma-Aldrich) in PBS for 1 h, followed by three PBS wash steps. Primary anti-HNF-1A antibody (Invitrogen, PA5-83263) was applied for 1 h, using a 1:100 dilution in 2% GS/8% BSA-PBS. Samples were washed three times and stored in PBS overnight at 4°C. Secondary antibody [Alexa Fluor^® 488 AffiniPure Goat Anti-Rabbit IgG (H + L), Jackson ImmunoResearch, 111-545-003] was applied for 45 min, using a 1:100 dilution in 2% GS/8% BSA-PBS. Samples were washed three times with PBS and mounted on glass slides using ProLong™ Diamond Antifade DAPI mountant (Invitrogen, P36962). Mounting solution was allowed to solidify overnight, and the slides were inspected on the following day. Samples were imaged using a Zeiss Axiovert 200M wide-field fluorescence microscope equipped with an AxioCam HR camera, a Plan-NEOFLUAR 100×/1.30 Ph3 oil immersion objective and filters appropriate for the detection of DAPI and GFP/Alexa-488 signals. Images were acquired by using the AxioVision software (Carl Zeiss, version 4.5) and superposed using the Fiji ImageJ software (Wayne Rasband, National Institutes of Health, United States, version 1.52a) (Schindelin et al., 2012).

Molecular dynamics simulations of the LLPS behavior of HNF-1A TAD (residues 280–631) were performed using the 1 bead per amino acid (1BPA) molecular dynamics model (Ghavami et al., 2013; Ghavami et al., 2014), with updated parameters described in (Driver et al., 2022). The initial cubic simulation box is populated with molecules (using a random initial conformation) with their center of mass placed upon a regular grid, with a small buffer region to avoid overlap between molecules. All simulations are carried out at a temperature of 300 K, 150 mM ion concentration (κ = 1.27 nm⁻¹), and use a timestep of 20 fs. For equilibration of the droplet, energy minimization on the initial configuration is used (energy tolerance of 1 kJ mol⁻¹ nm⁻¹), before 50 ns NVT Langevin dynamics simulations (Nosé-Hoover thermostat with τ_t = 100 ps), followed by 500 ns NPT Langevin dynamics (Nosé-Hoover thermostat with τ_t = 100 ps and a Berendsen barostat with τ_p = 10 ps, 1 bar reference pressure and a compressibility of 4.5 10⁻⁵ bar⁻¹). The end state of the NPT equilibration step is inserted into a new periodic box with a volume chosen to give a total residue density of 80 mM, after recentering on the center of mass and after the molecules have been unwrapped across the previous periodic boundary conditions. A second energy minimisation step is applied in the new simulation box to relax the molecules after the box expansion (energy tolerance of 1 kJ mol⁻¹ nm⁻¹). A final 3 μs NVT equilibration/production run (Nosé-Hoover thermostat with τ_t = 100 ps) is used for data collection. The trajectory is sampled every 5 ns to determine whether convergence was reached.

Bioinformatic sequence analyses provided insights into structural features and sequence characteristics of human HNF-1A (Figure 1). Disorder prediction using IUPred3 was performed to analyze the residue-dependent propensity of HNF-1A to fold into globular domains or remain disordered (Figure 1A) (Erdős et al., 2021). While the α-helical DD and DBD of HNF-1A were predicted to fold, the linker between DD and DBD showed a high tendency to remain disordered (Figure 1A). The flexibility of this linker region has previously been demonstrated by SAXS (Kind et al., 2022), illustrating the agreement between in vitro experiments and in silico prediction. The IUPred3 prediction indicated that the TAD likely contains long IDRs. Disorder was predicted for the residue ranges 280–420, 460–500, and 540–590, while the regions harboring residues 420–460, 500–540, and 590–631 may fold in a context-specific manner (Figure 1A). A complementary analysis using the ANCHOR2 algorithm was in agreement with the IUPred3 prediction (Figure 1A). The residue ranges 300–400 and 550–600 were predicted to contain disordered binding sites that can undergo disorder-to-order transitions upon binding to a partner protein (Figure 1A).

Disordered protein regions harbor intrinsic flexibility, which leads to a dynamic and non-globular protein conformation. We found that tertiary structure predictions of HNF-1A using AlphaFold (Jumper et al., 2021; Varadi et al., 2022) were in agreement with disorder predictions by IUPred3 and ANCHOR2 (Figure 1B). The per-residue estimate of prediction confidence (pLDDT score) was <50 for residues 300–631, indicating a disordered conformation for this protein region (Supplementary Figures S1A, B). Apart from a C-terminal extension of helix α8 in the DBD, the AlphaFold prediction suggested an entirely disordered TAD (Figure 1B; Supplementary Figure S1B). The model presented in Figure 1B depicts the TAD in one possible conformation which co-exists with numerous other conformations that are sampled by the TAD. This conformational flexibility is reflected in the predicted aligned error analysis (Supplementary Figure S1C), which indicated that the prediction confidence for the relative position of two given HNF-1A residues was high within the folded domains (DD, POU_S, POU_H), but low for residues within the TAD.

It should be noted that the AlphaFold model may not represent context-dependent structural changes of the TAD. According to the ANCHOR2 prediction (Figure 1A), specific segments of the TAD are likely to undergo disorder-to-order transitions when binding to other molecules. Depending on the interaction interface, these specific segments may undergo a transition from a coil-like secondary structure to an α-helical structure or a β-strand conformation to maximize intermolecular hydrophobic interactions when binding to the interaction partner (Dyson, 2013). Folding events, which follow such a coupled folding and binding mechanism, have been demonstrated for IDRs in other transcription factors (e.g., DREB2A, p53, HIF1α) when binding to specific protein partners (Dyson and Wright, 2005; Oldfield et al., 2008; O’Shea et al., 2017).

The structure predictions by IUPred3, ANCHOR2, and AlphaFold indicated that the HNF-1A TAD is predominantly disordered. As intrinsic protein disorder is critical for eukaryotic gene transcriptional regulation (Liu et al., 2006; Mar et al., 2023), the predictions of TAD disorder underline its potential role in the gene transcriptional activity of HNF-1A. A disordered conformation of the TAD may be important for HNF-1A to engage in the formation of macromolecular complexes, as it may allow HNF-1A to sample a large physical space and to simultaneously bind several other transcription factors and co-activators. The ability of the TAD to fold in a context-specific manner could provide versatility in protein-protein interactions of HNF-1A, which may be important for the response to different signals in different cellular states. Besides mediating the binding to other proteins, the intrinsically disordered TAD may contribute to DNA recognition and binding, thereby modulating the gene transcriptional activity of HNF-1A (Mar et al., 2023).

A general feature of intrinsically disordered proteins (IDPs) is the absence of hydrophobic clusters and an enrichment of charged and polar residues (Uversky, 2002). Indeed, a hydrophobicity analysis using FELLS illustrated the lack of hydrophobic residues in the TAD compared to the globular DD and DBD in the N-terminal half of HNF-1A (Figure 1C). Interestingly, charge distribution analysis of the HNF-1A sequence revealed that the TAD contains very few charged residues (Figure 1C). This observation was surprising, as charged amino acids generally promote solubility and are frequently found in IDPs (Uversky, 2002). However, the sequence analysis also revealed that the TAD is highly enriched in Ser and Thr residues. Such a polar tract may facilitate the interaction with the polar environment and promote a disordered state of the TAD (Figure 1C).

In addition to modulating protein solubility, the side chains of Ser and Thr residues can be phosphorylated, leading to changes in protein function (Day et al., 2016). The high prevalence of Ser and Thr residues in the TAD of HNF-1A may indicate that it is a target of post-translational modification by kinases and phosphatases. As a transcription factor, HNF-1A may be phosphorylated by kinases within the basal transcriptional machinery, e.g., cdk7-9, TAFII250, or TFIIF (Riedl and Egly, 2000). Cdk7 phosphorylates the activation domains of the transcription factors ERα, E2F-1, and p53, which results in changes in their transactivating properties (Riedl and Egly, 2000). In line with the above, phosphorylation of TAD residues may dynamically affect the ability of HNF-1A to interact with transcriptional co-activators/repressors and chromatin remodeling proteins by altering the protein interaction interface, ultimately leading to changes in the transactivation potential of HNF-1A. Alternatively, side chain phosphorylation of the TAD may modulate the recognition sites for specific adapter proteins, which could lead to changes in HNF-1A localization (Day et al., 2016). Finally, TAD phosphorylation may create or disturb binding sites for the components of the ubiquitination machinery, which could lead to changes in HNF-1A protein turnover due to altered proteasomal degradation (Hunter, 2007).

Transcription factors contain specific activation domains, which are crucial for the recruitment of co-activators and the formation of the pre-initiation complex at specific promoter target sites (Frietze and Farnham, 2011). The most reported activation domains in transcription factors are acidic activation domains (AADs), which require a balance between acidic and hydrophobic residues (Sanborn et al., 2021; Staller et al., 2022). The suggested “acidic exposure model” of AAD function describes that hydrophobic residues are interspersed with acidic residues, thereby preventing a hydrophobic collapse of the protein region and promoting exposure of the hydrophobic residues for interaction with co-activator protein surfaces. Based on these observations, the “Predictor of Activation Domains using Deep Learning in Eukaryotes” (PADDLE) algorithm was developed to identify such segments in protein sequences (Sanborn et al., 2021). When applying PADDLE to the HNF-1A sequence, we found that the protein does not contain any regions with an amino acid composition indicative for an AAD (Supplementary Figure S2). The transactivation potential of the TAD may thus unfold via a different mechanism. Considering the sequence composition of the TAD (Figure 1C), we speculate that activation domains rich in Gln, Ser, and Pro may play a role in the function of the HNF-1A TAD. Only a limited number of such alternative activation domains have been annotated thus far, such as the Glu and Pro rich activation domains in the POU homeodomain protein Oct-2 (Tanaka et al., 1994) and a Ser rich activation domain in the v-Rel protein of the Rel/NF-kappaB transcription factor family (Chen et al., 1999). Expanding on the current knowledge of activation domain characteristics, a recently reported large-scale mapping approach identified novel activation domains in human transcriptional effector domains by testing transcription factor fragments (tiles) in a cell-based reporter assay (DelRosso et al., 2023). Several identified activation domains were composed of hydrophobic residues interspersed with acidic, Glu, Pro, and/or Ser residues (DelRosso et al., 2023). Employing a quantitative cell-based reporter screen with defined HNF-1A TAD fragments, as demonstrated by Sanborn et. al (2021), will aid in the identification of HNF-1A activation domains and contribute to the understanding of alternative activation mechanisms besides the AAD model.

In recent years, researchers have uncovered the ability of transcription factors to activate gene transcription by undergoing LLPS. The formation of transcriptional condensates is often mediated via activation domains and leads to the dynamic compartmentalization of the gene transcriptional machinery, including RNA polymerase II and its co-factors (Boija et al., 2018; Sabari, 2020). We hypothesized that the HNF-1A TAD may have the potential to activate gene transcription in the same way. We utilized the FuzDrop algorithm to compute the probability of HNF-1A to undergo spontaneous LLPS (Hatos et al., 2022), and the obtained pLLPS score of 0.9932 suggested a strong LLPS tendency. The computed droplet-promoting regions (DPRs) were found across the entire HNF-1A sequence (Figure 1D), whereby long DPRs were localized to the intrinsically disordered DD-DBD linker (residues 27–84) and the HNF-1A TAD (residue ranges 278–366, 390–432, and 545–617). The prediction is in line with the observation that LLPS often involves weak, multivalent interactions between IDRs (Dignon et al., 2020). Interestingly, some of the predicted DPRs overlapped with two annotated activation domains in the TAD (residue ranges 398–470 and 544–631), which may hint towards a transactivation mechanism involving LLPS (Toniatti et al., 1993; Bjørkhaug et al., 2005). In cases of LLPS dysregulation, protein condensates can develop into a transient gel and ultimately evolve into an amyloid state of protein aggregates, which is usually driven by characteristic cross-β-structures (Vendruscolo and Fuxreiter, 2022). Such irreversible amyloid states can lead to cytotoxicity and consequently cause various diseases, such as cancer, neurodegeneration, and infectious diseases (Alberti and Dormann, 2019; Mathieu et al., 2020). The switch between a droplet and amyloid state involves protein regions which are able to adopt both a disordered droplet-promoting and an ordered aggregate-promoting binding mode. Computational predictions indicated that in 87% of the human proteome, DPRs and amyloid-promoting regions co-occur in the same sequence region (Vendruscolo and Fuxreiter, 2022). In accordance with this observation, the FuzDrop algorithm predicted that some of the DPRs in HNF-1A contain aggregation hotspots, which can initiate the irreversible maturation of droplets and thus lead to protein aggregation (Figure 1D). Future studies may reveal the relevance of the predicted DPRs in HNF-1A and the potential impact of droplet aggregation upon dysregulation of β-cell function, e.g., in type 2 diabetes, or alteration of the HNF-1A protein sequence, e.g., in HNF1A-MODY.

We generated a multiple sequence alignment to investigate the sequence conservation of the TAD (Figure 1E). Some TAD regions are highly conserved across species, such as residue ranges 374–387, 457–469, and 487–496, while conservation in other regions is mainly restricted to mammals, e.g., in residue ranges 285–311, 330–350, and 571–589.

IDRs frequently harbor SLiMs, which are sequence elements of 6–12 residues that mediate the binding to folded interaction partners or promote LLPS by partaking in multivalent interactions (Bugge et al., 2020). In order to identify potential SLiMs in the TAD of HNF-1A, we performed a search in the ELM resource (Kumar et al., 2022). High-confidence SLiMs are marked in the multiple sequence alignment (Figure 1E) and include interaction sites for WW domains (residues 287–308, residues 555–560), SH3 domains (residues 287–308), FHA domains (residues 525–530), as well as a recognition site for USP7 ubiquitination enzymes (residues 626–631). Interestingly, many of these SLiMs overlapped with recognition sites for different kinases, such as MAPK, Cks1, and NEK2, which may promote Ser or Thr phosphorylation within the motifs and thereby alter their ability to interact with the respective protein domains. The predicted SLiMs were localized in both highly conserved and less conserved regions (Figure 1E).

In conclusion, our sequence analyses provided a useful starting point for the functional dissection of the HNF-1A TAD. Experimental studies will be required in the future to investigate the biological roles of specific TAD regions and their dynamic regulation. Biochemical and structural approaches will shed light on the structure-function relationships of HNF-1A in solution.

In order to provide a basis for functional studies, we set out to biophysically and structurally characterize the TAD of HNF-1A. As full-length HNF-1A had not been characterized in vitro prior to this study, we designed several TAD-containing expression constructs in order to cover all defined HNF-1A domains, and, at the same time, provide flexibility in case of expression/solubility problems caused by certain domains. We thus generated three N-terminally His₆-tagged HNF-1A expression constructs harboring the TAD (DD-DBD-TAD, residues 1–631; DBD-TAD, residues 83–631; TAD, residues 280–631), whereby the information on the DBD and TAD boundaries was acquired from a published X-ray crystal structure of DNA-bound DBD [PDB ID: 1IC8, (Chi et al., 2002)]. The N-terminal His₆-tag was chosen as a simple strategy to allow for Ni-NTA affinity chromatography-based purification of recombinant HNF-1A.

Initial overexpression in Rosetta (DE3) resulted in insoluble protein, which may be functionally connected to the predicted disordered nature of the TAD (Figures 1A, B) or the tendency to form condensates or amyloid-like protein aggregates (Figure 1D). In order to enhance the solubility of the proteins, we screened for optimal expression and lysis conditions. Following established strategies to improve protein folding and to reduce protein aggregation (Sørensen and Mortensen, 2005; Leibly et al., 2012), we varied expression systems [Lemo21 (DE3) E. coli, Sf9 Spodoptera frugiperda], expression conditions (1–72 h, 20/37°C), and lysis buffer compositions (screening for NaCl concentrations, pH value, and stabilizing additives). However, none of these adjustments yielded soluble protein (data not shown). Future attempts to improve the soluble expression of HNF-1A may involve a fusion to the Bombyx mori 30Kc19 protein, which has been proven useful for the expression of otherwise unstable transcription factors, e.g., Oct-4, Sox2, and Klf4 (Ryu et al., 2016).

We proceeded with protein purification from the insoluble lysate fraction, whereby we focused on the DBD-TAD construct due to relatively high expression levels compared to the other two tested constructs (DD-DBD-TAD, TAD). We solubilized the protein under denaturing conditions during bacterial lysis and performed an initial purification step using Ni-NTA affinity chromatography. The protein did not have a strong affinity towards the Ni-NTA material, as relatively low imidazole concentrations (40–80 mM) were sufficient for elution. Moreover, the elution fraction contained three strong bands, indicating a C-terminal truncation of overexpressed DBD-TAD. In order to refold the denatured protein into its native form, we dialyzed the elution fractions into buffer lacking denaturant. We performed a final SEC step to separate monomeric DBD-TAD from aggregates and suspected degradation products (Figures 2A, B). Mass spectrometry confirmed that both the purified protein and the suspected degradation products were derived from human HNF-1A. The degradation from the C-terminus may be explained by the expected disordered nature of the TAD (Figures 1A, B), as degradation motifs might be exposed and can be recognized by bacterial proteases. The degradation of DBD-TAD may already take place during the protein expression step; however, additional degradation might occur during protein purification. In the future, this problem could potentially be solved by the introduction of a C-terminal protein tag acting as protective chaperone, e.g., maltose-binding protein (MBP), N-utilizing substance A (NusA), small ubiquitin related modifier (SUMO), or glutathione S-transferase (GST). In particular, MBP and SUMO exhibit the useful property of promoting the translocation of the recombinantly expressed fusion protein from the cytosol to cellular locations with lower protease contents, such as the cell membrane and the nucleus, respectively (Butt et al., 2005; Costa et al., 2014).

Despite the observed protein degradation, we were able to isolate a pure protein that exhibited a migratory behavior expected for the theoretical molecular weight of DBD-TAD (62 kDa, Figure 2B). As our goal was to biophysically characterize a recombinant protein that contained the entire TAD, it was important to retain a homogenous sample without degradation products. Hence, to avoid the gradual formation of DBD-TAD degradation products over time, we aimed to conduct all characterization experiments immediately after purification.

SRCD experiments were performed to assess the secondary structure content of the purified protein. The SRCD spectra indicated that DBD-TAD contained both α-helical structures and random-coil regions (Figure 2C). In order to quantify the secondary structure elements within the protein, we performed a deconvolution of the SRCD spectra (Table 2). Calculations with the BeStSel algorithm (Micsonai et al., 2015; Micsonai et al., 2018) revealed that DBD-TAD contained 12% α-helical structures, 26.9% β-sheet, 17.5% turns, and 43.6% other structures, e.g., bends, loops, and irregular or invisible structure elements (Table 2). As we expected DBD-TAD to contain a large fraction of disordered regions, we also employed the recently developed DichroIDP program, which uses a novel protein reference dataset (IDP175) that is suitable for analyses of proteins containing significant amounts of disordered structure (Miles et al., 2023). DichroIDP calculated a DBD-TAD composition of 4% α-helical structures, 39% β-sheet, 30% turns, and 27% disordered regions (Table 2). The deconvolution with both algorithms indicates that a major part of DBD-TAD is intrinsically disordered, which confirms the secondary structure predictions from IUPred3, ANCHOR2, and AlphaFold (Figures 1A, B). The α-helical CD signals likely arise from the folded POU_S and POU_H domains in the DBD (Chi et al., 2002; Kind et al., 2022). The considerable amount of β-sheet content could correspond to amyloid-promoting disordered regions which may adopt transient β-like formations, as predicted by the FuzDrop algorithm (Figure 1D). We compared DBD-TAD with a TAD-deficient DBD protein (Figure 2C). The SRCD spectrum of DBD indicated that the protein predominantly contained α-helices (Figure 2C), which was confirmed by the spectrum deconvolution (Table 2) and in agreement with previous studies (Narayana et al., 2001; Kind et al., 2022). Since DBD-TAD contained both α-helical and disordered regions, and the isolated DBD produced an SRCD spectrum indicative of a high α-helical structure content (Figure 2C), we concluded that the TAD of HNF-1A likely adopts a random-coil structure in solution. We performed a subtraction of the DBD-TAD and DBD spectra, yielding a theoretical TAD spectrum, which indicated a disordered protein conformation and confirmed our hypothesis (Figure 2C). A deconvolution of this theoretical TAD spectrum confirmed that the TAD predominantly consisted of turns, disordered regions, and β-sheet structures (Table 2). As noted above, the β-sheet content may correspond to transient β-sheet structures adopted by the predicted amyloid-promoting regions in the TAD (Figure 1D). To test the propensity of the TAD to fold into an α-helical structure, we performed a TFE titration experiment, in which α-helical structure of the DBD-TAD construct was induced by the addition of ≥50% TFE (Figure 2D; Table 2). Notably, intermediate TFE concentrations (10%–30%) decreased the SRCD signal across the entire wavelength range (Figure 2D), which may be due to secondary processes causing light scattering, such as LLPS or aggregation.

We employed SAXS to study the molecular dimensions and shape of DBD-TAD (Figures 2E–G). We compared the scattering data to a published SAXS dataset for the isolated DBD (Kind et al., 2022), allowing us to study the contributions of the HNF-1A TAD. The DBD-TAD sample was free of aggregation (Figure 2E). Molecular weight analysis resulted in differing molecular weight estimates depending on the algorithm used (Supplementary Table S1). The molecular weight estimates based on the V_c and the SAXSMoW methods, which are suitable approaches for the analysis of intrinsically disordered proteins (Rambo and Tainer, 2013; Piiadov et al., 2019), indicated that DBD-TAD was present in monomeric state (Supplementary Table S1). The radius of gyration (R _g ) of DBD-TAD was determined using Guinier analysis (R _g = 4.9 nm) and the Debye formalism (R _g = 5.1 nm). The molecular dimensions of DBD-TAD were also analyzed using the paired distance distribution function (R _g = 4.8 nm, D _max = 17 nm). DBD-TAD presented larger R _g and D _max compared to the isolated DBD [R _g = 2.8 nm, D _max = 10 nm, Supplementary Table S1, (Kind et al., 2022)], which was expected due to the higher molecular weight of DBD-TAD. The dimensionless Kratky plot indicated that both DBD and DBD-TAD were highly flexible and had a non-globular shape (Figure 2G), which is likely due to the linker region between the POU_S and POU_H domain and the disordered TAD. Based on the scattering data, we generated a 3-dimensional model for DBD-TAD to visualize its molecular shape (Figure 2H). We utilized EOM to generate an ensemble of conformers that together satisfy the SAXS data (Supplementary Table S1; Figure 2H). A published crystal structure of a DBD:DNA complex [PDB: 1IC8 (Chi et al., 2002)] was used to model POU_S and POU_H as globular domains. Missing residues were modeled as flexible regions. The data were satisfied by an ensemble containing seven conformers with approximately equal abundance, exhibiting an average R _g of 64 Å and a D _max of 198 Å. An analysis of the ensemble properties revealed a bi modal distribution of the R _g and D _max parameters (Figure 2H), with one population exhibiting a peak R _g of ∼45 Å and a peak D _max of ∼140 Å, and another population having larger molecular dimensions with a peak R _g of ∼90 Å and a peak D _max of ∼280 Å. This is reflected in the generated conformer models, in which the TAD is present in compact, intermediate, and extended conformations (Figure 2I).

The SRCD and SAXS analyses confirmed the predicted intrinsic disorder of the HNF-1A TAD. The employed methods are, however, limited in their information content. In order to obtain data at atomic resolution, the biophysical analysis could be extended by the application of nuclear magnetic resonance (NMR) spectroscopy. This technique has been employed for numerous transcription factors, e.g., p53, human glucocorticoid receptor, and radical-induced cell death 1, and helped deconvolute the structure of TADs as well as their role in DNA recognition and IDR-mediated protein binding (Arai et al., 2012; Kim et al., 2017; Krois et al., 2018; Staby et al., 2021).

Based on the FuzDrop predictions (Figure 1D), we suspected that the disordered TAD may have the potential to promote LLPS of HNF-1A. We observed that the purified DBD-TAD protein could not be concentrated above ca. 2.5 mg/mL. The protein solution increased in viscosity at higher concentrations, which may be due to the predicted formation of protein droplets, hydrogel or fibrous aggregates (Figure 1D) (Vendruscolo and Fuxreiter, 2022; Wang et al., 2022). To investigate the potential cause, we performed DIC microscopy using purified DBD-TAD and DBD (Figures 3A, B). As LLPS is a concentration-dependent process and HNF-1A condensation in the cell was expected to be modulated by the dense molecular environment of the nucleus, we used the crowding agent PEG8000 to mimic the native environment in vitro. We observed droplet formation for the DBD-TAD construct, with a concentration-dependent increase in droplet number (Figure 3A). Most droplets had a round shape, suggesting that they were liquid condensates, as opposed to aggregates. The DBD protein also produced droplets; however, higher protein concentrations were required, and the droplet size differed compared to the TAD-containing protein (Figure 3B). The latter indicates that the TAD may be a major driver of the LLPS behavior of HNF-1A. However, it is important to note that the comparison between the two proteins and their contribution to LLPS at a given molar concentration is not straightforward due to their different molecular size. Based on the FuzDrop prediction (Figure 1D), it may be possible that certain regions of the DBD contribute to LLPS of HNF-1A, such as residues 179–200 in the intrinsically disordered POU_S-POU_H linker (Kind et al., 2022). Additional experimental evidence for the presence of HNF-1A-containing droplets in vitro may be acquired by performing fluorescence microscopy with a fluorescently labeled (DBD-) TAD, observing the mixing dynamics and potential fusion events between the droplets. The characteristic liquid-like behavior of biomolecular condensates could be tested using fluorescence recovery after photobleaching (FRAP) microscopy (Alberti et al., 2019; Ganser and Myong, 2020).

As transcription factors are normally present in the nM concentration range (copy numbers between 10³-10⁶ per mammalian cell) (Biggin, 2011), our experimental conditions do not fully recapitulate the protein concentration at endogenous levels. However, even with lower HNF-1A concentrations in the cell, phase separation may take place in the presence of other IDR-containing proteins promoting multivalent interactions. LLPS-dependent cellular compartments typically concentrate ten to several hundred different proteins (and often RNA molecules), whereby many of them contribute to the formation of the biomolecular condensate (Banani et al., 2017). We next investigated whether the suspected LLPS behavior of HNF-1A may lead to the formation of biological condensates in cells. We employed fluorescence microscopy to image the localization of endogenous full-length HNF-1A in fixed MIN6 β-like cells, revealing that HNF-1A was predominantly located in the nucleus (Figure 3C). Moreover, HNF-1A produced strong fluorescent punctae, which co-localized with the nuclear stain and may thus represent nuclear condensates (Figure 3C). It is important to note that the cell fixation using paraformaldehyde can significantly alter the appearance or presence of biomolecular condensates, which is why the presented results should be interpreted with care and verified with alternative methods (Irgen-Gioro et al., 2022). Live-cell imaging techniques will be valuable for verifying the existence of HNF-1A-containing nuclear condensates in pancreatic β-cells. The HNF-1A signal in living cells may be observed over the course of time to investigate their potential dynamic behavior in different cellular states. In addition, liquid-like properties of the putative condensates may be probed by FRAP microscopy (Alberti et al., 2019). In order to prevent artefacts due to overexpression of exogenous HNF-1A, the cells may be genetically engineered to endogenously express fluorescently labeled HNF-1A. In this case, a cell line expressing the fluorescent protein alone will serve as an important control (Alberti et al., 2019).

Following the verification of HNF-1A condensate formation, the biological function of the structures may be determined by the identification of other condensate components. HNF-1A may phase separate with other proteins, DNA, or RNA (Banani et al., 2017). Studies of HNF-1A DNA interactions may be initiated by an antibody-based identification of HNF-1A bound DNA fragments, such as in chromatin immuno-precipitation (ChIP) or the Cleavage Under Targets & Release Using Nuclease (CUT&RUN) technique (Park, 2009; Skene and Henikoff, 2017). The detected DNA fragments can point towards genomic locations, at which HNF-1A condensates nucleate and initiate gene transcription. Proteins interacting with HNF-1A may be identified using proteomic approaches (Roux et al., 2018; DeCaprio and Kohl, 2020), either investigating stable interaction partners by immuno-precipitation (IP) or assessing transient interactions by performing biotin-labelling of interaction partners based on proximity in the cellular environment (BioID). Co-localization studies with general components of the transcriptional machinery (e.g., RNA polymerase II, TFIID/H, Mediator, DSIF) could elucidate whether HNF-1A partakes in transcriptional condensates (Schier and Taatjes, 2020). This approach is exemplified by a study from Boija et al. (2018), which demonstrated that the transcription factors OCT4, GCN4, and ER recruit Mediator into phase-separated droplets in vitro and in cells and that their ability to do so is linked to gene activation potential.

To understand the LLPS propensity of the HNF-1A TAD, we undertook computational droplet formation simulations (Figure 4; Supplementary Figure S3; Supplementary Video S1). We assessed the internal structure of the droplets by using the 1 bead per amino acid (1BPA) molecular dynamics model (Driver et al., 2022), developed for the study of intrinsically disordered proteins in the nuclear pore complex (Ghavami et al., 2013; Ghavami et al., 2014). Self-assembly and clustering of individual monomers into phase-separated condensates can be a slow process to observe. To speed up this process, a condensed phase droplet is formed at the start of the simulation, which is then inserted into an empty dilute phase. If LLPS is favored, the droplet structure should remain stable throughout the subsequent simulation; if non-favored, the droplet would break up into a dilute phase of monomers.

We simulated the behavior of 120 HNF-1A TAD molecules and observed LLPS (Supplementary Video S1). In the beginning of the simulation, two to three small clusters with one to six HNF-1A TAD molecules co-existed with a large cluster containing approximately 115 HNF-1A TAD molecules. The smaller clusters rapidly fused with the big cluster, yielding a droplet of 120 HNF-1A TAD molecules that remained stably associated throughout the simulation period (Supplementary Video S1). By analyzing the interactions between the TAD molecules in detail, we found that intermolecular interactions along the whole length of the TAD contribute to LLPS (Figure 4A). The LLPS behavior of the TAD is driven by hydrophobic aliphatic-aliphatic contacts, with minimal aliphatic-aromatic interactions due to the low abundance of aromatic residues (Figure 4C). This is driven by contacts between Ala, Leu, Pro, and Val (Figure 4B). We also observed many interactions between hydrophilic residues (Gly, Ser, His, Thr, and Gln) driven by their high abundance (Figure 4B), but whose effect is reduced when interactions are normalized by amino acid abundance (Supplementary Figure S3). The predicted LLPS-promoting regions (Figure 1D) are highlighted on the contact map in Figure 4A. These regions partially align with the regions we found to form the most contacts, with only the FuzDrop DPR 466–488 showing full alignment (box marked with “*” in Figure 4A). The more extensive interactions discovered by the simulations indicate that the even distribution of aliphatic residues throughout the TAD contributes to LLPS, whereas an absence of intermolecular interactions among charged and polar residues, which make up 90% of the TAD, was evident (white stripes in Figure 4A). In summary, the results of our simulations corroborate with our in vitro data (Figure 3A) and suggest that several residues along the entire TAD collectively contribute to the droplet formation.