Autoinhibition of untimely DNA binding by structural disorder in the full-length peroxisome proliferator-activated receptors (PPARs)

Abstract

Structural disorder in nuclear receptors is essential for gene transcription, but mechanisms are lacking. Using experiment and computation, including residue-resolution NMR, we demonstrate how structural disorder in full-length peroxisome proliferator-activated receptors (PPARs) contributes to function. We show that the disordered AB domain of PPARγ and PPARα interact dominantly and dynamically with the DNA binding domain (DBD) via two regions that, when tested in cells, constitute active transactivation domains. For PPARγ, we provide an integrative structural model of the full-length protein, and show that upon DNA binding to the DBD, but not by ligand binding to the LBD, the AB domains are released to an open, accessible state. We demonstrate that the interactions between the AB domain and the DBD lead to autoinhibition affecting DNA binding affinity negatively with little effect on DNA binding specificity. We show that autoinhibition is also present in the heterodimer with RXRα, but is less pronounced due to higher DNA binding affinity. Thus, our work suggests a role for the disordered AB domain in preventing untimely DNA binding to the PPAR monomers.

Interdomain interactions by the disordered N-terminal region in PPAR transcription factors lead to autoinhibition of DNA binding. The effect is strongest for the monomer suggesting inhibition of untimely DNA binding prior to heterodimer formation.

Introduction

In humans, around 1600 transcription factors (TFs) regulate and maintain a broad range of biological processes and their malfunctions lead to diseases including diabetes and cancer^1,2. TFs regulate gene transcription by binding to specific genomic cis-regulatory elements in the vicinity of their regulated target genes and by engaging in complex protein networks of the transcriptional machinery^1–3. Typically, such networks rely on intrinsically disordered transcriptional regulatory domains present in most TFs as well as in their co-regulators⁴. The human nuclear receptor (NR) superfamily of transcription factors contains large, disordered regions, and especially the disordered N-terminal domains have been reported to play functional roles including regulation of transcriptional activity by driving transcription of selected set of genes^5,6. The NR family consists of 48 TFs, including the three members of the peroxisome proliferator-activated receptor (PPAR) family, PPARɑ (NR1C1), PPARβ/δ (NR1C2) and PPARγ (NR1C3). Of these, PPARγ2 and PPARɑ are the most studied.

All PPARs have a large promiscuous hydrophobic ligand binding pocket and are activated by binding to a broad range of fatty acids and fatty acid derivatives^7,8. Once activated, they bind as obligate heterodimers with members of the retinoid X receptor (RXR) family to PPAR responsive elements (PPREs) in the genome. The different PPAR family members bind to overlapping sets of target enhancers and regulate many of the same target genes; however, they also display subtype specific functions and expression patterns^9,10. PPARα is a key regulator of fatty acid oxidation ketogenesis and is highly expressed in cell types such as hepatocytes and brown adipocytes, which have a high level of fatty acid catabolism^11,12. PPARγ instead activates fatty acid anabolism and storage and is distinctive among the PPARs by being the master regulator of adipogenesis^13,14. PPARγ is expressed in two isoforms, PPARγ1 and PPARγ2, differing in length by 28–30 residues depending on species. The PPARγ2 isoform is specifically expressed in adipocytes and has the strongest adipogenic potential^15,16. Consistent with roles in lipid metabolism, synthetic PPARα-specific agonists act as hypolipidemic drugs, whereas PPARγ-specific agonists act as insulin sensitizers by increasing lipid storage and insulin sensitivity of adipocytes¹⁷.

Like most NRs, the modular structure of PPARs consists of four domains, an N-terminal AB domain, a DNA binding domain (DBD), a hinge region, and a ligand binding domain (LBD) (Fig. 1A). The AB domain is the least conserved between species and subtypes and is disordered^18,19, and in PPARγ and -α, it encompasses ligand independent activation function (AF)²⁰. The DBD is highly conserved and is composed of two zinc-finger motifs, stabilizing heterodimerization with RXR when bound to DNA. The DBD is connected to the LBD through a shorter disordered hinge region, and the LBD constitutes the main interaction surface for heterodimerization¹⁸. Binding of an agonist to the LBD induces a conformational change that leads to the exposure of a hydrophobic patch on the surface of the LBD allowing co-regulators of transcription to bind via a conserved LXXLL short linear motif (SLiM)^21–23—‘X’ being any residue. High-affinity synthetic PPARγ agonists, including thiazolidinediones such as rosiglitazone are efficient insulin-sensitizing drugs and have been used for the treatment of type II diabetes mellitus^24,25; however, more recent reported side effects including liver injuries and heart failures have led to their discontinuation¹⁷. The consensus PPRE of PPARs is an imperfect direct repeat of 5’-AGGTCA-3’ spaced by one nucleotide²⁶. The PPARs bind to the 5’ half site and sequences at the 5’ flanking region interact with a region of the hinge region^18,27 termed the C-terminal extension, located immediately after the DBD^18,28.

Fig. 1. The AB domain of PPARγ and PPARɑ are disordered with little local secondary structure.

A The canonical NR domain organization of PPARγ (dark green) and PPARɑ (light green) with an AB domain, a DBD, a hinge region, and an LBD. Domain borders are highlighted. B Sequence alignment of the AB domains of PPARγ (dark green) and PPARɑ (light green). Negatively charged residues (D/E) are red, positively charged residues (R/K) blue, and transient helices are in bold. C pLDDT scores from AlphaFold 2 and D IDDomainSpotter profiles^38,115 (window size 15 residues) of the AB domains of PPARγ (left) and PPARɑ (right) illustrating the scores of positive charges ( + RK), and the overall charge ( + RK-DE). E ¹H,¹⁵N HSQCs of the isolated AB domains of PPARγ (dark green) and PPARɑ (light green). The C^ɑ secondary chemical shifts (SCS) of (F) PPARγ^AB and (G) PPARɑ^AB. Dimensionless Kratky plot representation of the SAXS data of (H) PPARγ^AB and (I) PPARɑ^AB. For both (H) and (I), error bars represent measurement error. Overlay of 200 models from CALVADOS simulations of (J) PPARγ^AB and (K) PPARɑ^AB selected by randomly subsampling the trajectories using weights from reweighting against SAXS data. Data in (F, G) reported in Source Data file.

The disparity in gene activation between PPAR subtypes suggests that PPAR-mediated gene regulation and the subsequent modalities of PPAR specificity are determined by a combination of residue-specific differences within the DBD and LBDs²⁴, but also from differences manifested through the diverse intrinsically disordered AB and hinge domains, distinct to each subtype¹⁰. A role for the AB domain in transcriptional selectivity is supported by the observation that regulation of a subset of target genes is distinctly AB domain-dependent¹⁰, but also from experiments showing that swapping AB domains between PPAR subtypes exchanges the subtype-specific expression patterns of these target genes^10,29. Furthermore, it has been proposed that the AB domains play roles in targeting of NRs, including PPARs to transcriptional condensates^30–32. Finally, interdomain interactions in other NRs have been shown to exist, such as direct interactions between the AB domain and the LBD in the androgen receptor (AR)³³, in the glucocorticoid receptor (GR)³⁴, and in PPARγ³⁵ (preprint), just as allostery enabled by disorder³⁴ and direct interactions between the AB domain and the DNA³⁶ have been reported. Collectively, these observations highlight the significance of the AB domains for gene transcriptional specificity; however, information on the mechanism(s) through which the AB domains regulate gene transcription and confer specificity remains scarce.

Here, we decipher roles of structural disorder within full-length PPARγ2 (referred to here as PPARγ) and PPARɑ, using PPARγ as the main model. We apply nuclear magnetic resonance (NMR) spectroscopy, small-angle X-ray scattering (SAXS), and molecular dynamics (MD) simulations to obtain a model of full-length PPARγ. We relate observations from assays reporting on transcriptional activity and degradation in cells to quantitative data on the structure, dynamics and DNA binding affinity and specificity of PPARs to understand how structural disorder contributes to function in PPARs. We show that the AB domains are disordered but engage dynamically and dominantly with their DBD whereby they negatively regulate DNA binding affinity but not specificity. The AB domains compete with DNA binding and gives rise to autoinhibition, equipping them with key functional roles.

Results

The AB domains are disordered with little residual structure

The AB domains of PPARγ and PPARɑ have little sequence identity. They have low isoelectric points (4.2 and 3.0, respectively), conferring them with a predicted overall nominal negative charge of −18 (PPARγ) and −17 (PPARɑ) at physiological pH (Fig. 1B–D, Supplementary Fig. 1). Previous hydrogen-to-deuterium exchange mass spectrometry data has shown low amide protection in the AB domain of PPARγ^18,19 and AlphaFold2 pLDDT scores, reliable in predicting disorder in proteins^37,38, also suggest the AB domains (PPARγ, amino acid residues 1–137; PPARɑ, 1–100) to be disordered (Fig. 1C; pLDDT scores <50). Similarly, the AB domains of PPAR isoforms β/δ, and of the heterodimeric partner RXR are predicted to be disordered (Supplementary Fig. 2). To examine this disorder in detail, the isolated AB domains of PPARγ and PPARɑ were recombinantly expressed, isotope-labeled, and investigated using NMR, SAXS, and coarse-grained (CG) MD simulations to obtain residue-specific information, overall dimensions and a reliable conformational ensemble, respectively.

At relevant physiological conditions (i.e., pH 7.3, 135 mM KCl), the ¹H,¹⁵N HSQC NMR spectra of the AB domains (PPARγ^AB and PPARɑ^AB) revealed sharp sets of peaks with low chemical shift dispersion in the proton dimension demonstrating disorder (Fig. 1E). The assigned NMR backbone resonances of PPARγ^AB and PPARɑ^AB (99% completeness; Supplementary Fig. 2B) revealed low amplitude secondary C^ɑ chemical shifts (SCS C^ɑ), indicative of an ensemble mostly devoid of transiently formed local helical and strand structure. However, consecutive positive C^ɑ SCS-values were observed from Y93–I106 (PPARγ) and E25–M31 (PPARɑ), originating from transient helicity (∼10–15%) in these regions. Notably, in the two subtypes, the helix is positioned differently relative to the DBD (Fig. 1F, G).

The overall dimensions of the AB domains were evaluated by SAXS. The dimensionless Kratky plot, which qualitatively evaluates the presence of structural flexibility showed both PPARγ^AB and PPARɑ^AB to be highly disordered (Fig. 1H, I). We ran CG-MD simulations of the AB domains using the CALVADOS 2 model^39,40, a one-bead-per-residue model for intrinsically disordered proteins (IDPs) trained to predict global dimensions and transient long-range interactions. We calculated SAXS intensities from the ensembles and found them to be in good agreement with the experiments (Supplementary Fig. 3). We improved the agreement further by refining the ensembles using Bayesian maximum entropy (BME) reweighting against the experimental SAXS data⁴¹, resulting in excellent agreement with the SAXS data (Supplementary Fig. 3) and showing the SAXS data to be consistent with disordered ensembles for PPARγ^AB and PPARɑ^AB (Fig. 1J, K). From the reweighted ensembles, we estimated the radius of gyration, R_g of PPARγ^AB and PPARɑ^AB to be 3.59 ± 0.02 nm and 3.04 ± 0.01 nm, respectively, consistent with the R_g determined from the SAXS data (Supplementary Tables 1 and 2) (Supplementary Fig. 4).

In summary, PPARγ^AB and PPARɑ^AB exhibit exclusive as well as common features. Both AB domains have low sequence conservation with an overall large negative charge and are predominantly disordered with a single transiently formed helix, located at different distances to the DBD (32 and 71 residues in PPARγ and PPARɑ, respectively).

The AB domains interact dynamically with the DBD via two main regions

We next asked if and how the properties of the AB domains would be affected by context⁴² and included the other PPAR domains expressing six additional protein variants, the full-length PPARs (PPAR^FL), PPARs lacking the AB domain (PPAR^ΔAB), and the isolated AB-DBDs (PPAR^AB-DBD), (Fig. 2A). The C-terminal extension was included in the AB-DBDs. We first used PPARγ to understand the effect on the overall dimensions and recorded SAXS data on PPARγ^AB, PPARγ^ΔAB, and PPARγ^FL. While the R_g and maximum particle dimension, D_max were 2.44 ± 0.01 and 8.4 nm, respectively for PPARγ^ΔAB, the values for full-length PPARγ were 3.65 ± 0.01 nm and 16.5 nm, similar to those of its isolated AB domain (Fig. 2B and Supplementary Table 1). This result suggests some compaction of the AB domain in the full-length context, originating from interactions with the remainder of the protein.

Fig. 2. The AB domains interact dynamically with the DBD via two main regions.

A Modular structure of PPARγ and PPARɑ and color-code of variants, PPARγ^AB, dark green; PPARγ^AB-DBD, dark red; PPARγ^ΔAB, beige; PPARγ^FL, black, PPARɑ^AB, light green; PPARɑ^AB-DBD, brown; PPARɑ^FL, gray. B Dimensionless Kratky plot of the SAXS data (left) of PPARγ^AB (green), PPARγ^ΔAB (beige), and PPARɑ^FL (black). Error bars represent measurement errors. The scattering P(r) (right) of PPARγ^AB (green), PPARγ^ΔAB (beige), and PPARɑ^FL (black). C ¹H,¹⁵N HSQCs of the AB domain, AB-DBD, and FL variants of PPARγ (left) and PPARɑ (right) and zooms in (D) PPARγ and E PPARɑ. ¹⁵N R₂ relaxation rates of residues 1–137 (AB domain) in the AB domain, the AB-DBD, and the FL variant of (F) PPARγ and (G) PPARɑ. Data recorded at 750 MHz, error bars are standard errors calculated from the fits. Black star; unassigned residue in one or more variants, gray circle; overlapping residues, orange triangle; prolines. H Sequences of the main DBD interaction sites SIγ/ɑ and SIIγ/ɑ with negatively charged residues highlight in red and positively charged residues in blue. Aromatics are underlined. I Average contact number to residues 1−127 (AB domain excluding the last 10 residues) calculated from reweighted simulations of PPARγ^FL based on pairwise C^ɑ-distances within a cut-off of 1.1 nm and plotted on a crystal structure of PPARγ (PDB 3E00)¹⁸. J R_g distributions of the PPARγ AB domain from reweighted simulations of PPARγ^AB (green) and PPARγ^FL (gray). Data in (F, G) reported in Source Data file.

To pinpoint the interactions responsible for compaction, we assigned the backbone NMR resonances of the AB domains in the ¹⁵N,¹³C-labeled PPARγ^AB-DBD and PPARɑ^AB-DBD variants. The data were recorded at 5 °C to exclude signals from folded regions and reduce amide exchange⁴³, observing solely the disordered regions. Comparing the ¹H,¹⁵N HSQC spectra of PPARγ^AB-DBD and PPARɑ^AB-DBD with those of their isolated AB domains, we observed large chemical shift perturbations (CSPs) within the AB domain in both PPARs in context with the folded DBD (Fig. 2C–E, Supplementary Fig. 5). The most N-terminal regions of the AB domains remained unperturbed (e.g., PPARγ^T4, PPARγ^S8, and PPARɑ^D3) (Fig. 2D, E). However, changes in both chemical shifts and NMR peak intensities were seen distributed along the AB domains, as exemplified by PPARγ^K63, PPARγ^T75, PPARɑ^A16, PPARɑ^G42, and PPARɑ^L72 (Fig. 2D, E). Thus, for both PPARs, interdomain interactions exist between the AB domain and the DBD.

We probed backbone dynamics on the pico- to nano-second timescales via ¹⁵N R₁ and R₂ relaxation NMR measurements and steady-state ¹H,¹⁵N heteronuclear NOEs (hetNOEs)^43,44 (Fig. 2 and Supplementary Fig. 5). The transverse relaxation rate, R₂, varied between 1 s⁻¹ and 10 s⁻¹ for the isolated AB domains (Fig. 2F, G). Two predictors of R₂s of unstructured regions^45–47 demonstrate agreement with the experimental data (Supplementary Fig. 6), supporting the dynamic and random nature of the isolated AB domains. In context of the DBD, R₂ increased throughout most of the AB domains with two regions more affected (Fig. 2F, G). These were E35–F65 (termed SIγ) and V89–K107 (SIIγ) in PPARγ (Fig. 2F, gray shade) and S24–E43 (SIɑ) and S46–S59 (SIIɑ) in PPARɑ (Fig. 2G, gray shade). Only few peaks fully disappeared suggesting that the interactions remain dynamic. Notably, the regions found to form transient helical structure appeared to be involved in the interaction. No obvious properties united these interaction regions apart from their content of negative charges and aromatics, with SIIγ having a net charge closer to neutral (Fig. 2H).

We next addressed the interaction between the AB domain and the DBD using several approaches but with the caveat that we were unable to produce the isolated DBD. First, we investigated the properties of the interaction by monitoring the R₂ values of the AB domain in the context of the DBD by increasing the ionic strength to 500 mM KCl. Although the R₂ values decreased, suggesting more access to an open (dynamic) state at high salt, they never returned to the values of the isolated AB domain (Supplementary Fig. 7A). Thus, the interaction between the AB domain and the DBD is not solely electrostatic, in agreement with sequence properties of the interaction regions. Second, we asked whether the contacts between the AB domain and DBD are inherent to the AB domains. For this, we designed a chimeric PPAR consisting of the AB domain from PPARα (residues 1–94) and the DBD from PPARγ (residues 132–220), named PPAR^ɑAB-γDBD, and recorded NMR data on this variant (Supplementary Fig. 8). If the contacts are solely encoded by the AB domain, we would expect the peaks to overlay those of the PPARɑ^AB-DBD. However, we generally observed that the chemical shifts of the chimera were positioned between the peaks of the isolated AB domain and the peaks of those in the PPARɑ^AB-DBD (Supplementary Fig. 8A and B). This suggests that the AB domain in the chimera is on average less in contact with the DBD and hence the ensemble changed to a more accessible state compared to the AB domain in PPARɑ^AB-DBD. The CSP- and R₂-analyses support this conclusion, showing values positioned between those of the isolated AB domain and the AB-DBD (Supplementary Fig. 8C and D). Finally, we monitored the release of the AB domain from the ¹⁵N- PPARγ^AB-DBD in cis as a function of added PPARγ^AB or RXRɑ^AB in trans (Supplementary Fig. 7B). We did not observe full release of the AB domain, even at 10 molar excess of the AB domain in both cases; however, the CSPs were slightly larger for PPARγ, suggesting a specific interaction between the AB domain and the DBD. To further address specificity, we individually titrated in trans the ΔAB of PPARγ, RXRα or PPARα to either ¹⁵N-PPARγ^AB or ¹⁵N-PPARα^AB (Supplementary Fig. 7C, D). In PPARγ^AB the CSPs were small with no specific interactions observed at 4× molar excess of PPARγ^ΔAB or RXRα^ΔAB (Supplementary Fig. 7C). In PPARα^AB the CSPs were slightly larger for PPARγ^ΔAB or RXRα^ΔAB compared to PPARα^ΔAB, originating from higher molar ratio added (4× vs 1×, respectively) (Supplementary Fig. 7D). However, when adding PPARα^ΔAB, the CSPs increased for residues 30–40 and 50–60 in the AB domain, overlapping with SIα and SIIα and indicating more specific contacts for PPARα^ΔAB to PPARα^AB. Taken together, our results suggest that the AB domains make pronounced interdomain interactions with the DBDs existing in a dynamic equilibrium between states, representing partially exposed AB domains, and AB domains with pronounced contacts with the DBD. This dynamic interdomain interaction involves two different regions in each PPAR, one of which includes the single transient helical region. The contacts involved are context dependent and thus encoded in the sequences of both the AB domain and the DBD.

The AB-DBD communication dominates in intact PPARs

We next asked if the AB-DBD interactions would be preserved in the full-length proteins. To remove any ligand occupancy in the ligand binding pocket of the recombinantly produced proteins, an intense dialysis step was included. As confirmed by NMR analyses of the ¹⁵N-LBD with and without the addition of saturating amounts rosiglitazone, the LBD is dominantly in the ligand free form (Supplementary Fig. 9)⁴⁸. Despite their high molecular weight (PPARγ, 58 kDa; PPARɑ, 52 kDa), which can hamper observations by NMR, the ¹H,¹⁵N HSQCs of the full-length proteins resolved 64 (PPARγ) and 68 (PPARɑ) peaks from the disordered AB domains (Fig. 2C). Since the peaks either directly overlapped with or were very close to those measured in the AB-DBD, and peaks from the folded domains were suppressed by low temperature, transfer of assignments was straightforward. Again, the chemical shifts of e.g., T4 and S8 of PPARγ^FL and D3 of PPARɑ^FL were unperturbed. For e.g., K63 and T75 of PPARγ^FL and A16, G42 and L72 of PPARɑ^FL the peaks were positioned between the signals from the isolated AB domains and those in the AB-DBD context (Fig. 2D, E). Such population-averaged positions suggest that the LBD in both PPARs competes for the interactions with the DBD. Notably, a nearly identical R₂ relaxation profile was observed for the AB domains in their full-length context compared to the profiles of PPARγ^AB-DBD and PPARɑ^AB-DBD, indicating similar dynamics and likely reflecting a compensating effect from increased molecular size and changes in the population of DBD interacting states (Fig. 2F, G). For the region between SIγ and SIIγ, we observed slightly higher R₂ rates in full-length PPARγ, whereas in PPARɑ, the AB domain consistently exhibited higher R₂ rates throughout most of its sequence, suggesting stabilizing contacts or an overall reduced tumbling rate or both.

Compared to the spectra of the AB-DBD, resonances from PPARγ^F38, PPARγ^W39, and PPARγ^T41 were not observed in the spectrum of the full-length proteins. Therefore, it is possible that a direct interaction between the LBD and the AB domain would lead to the observed effect, and we therefore recorded a ¹H,¹⁵N HSQC spectrum of ¹⁵N-PPARγ^AB either alone or upon addition of PPARγ^LBD in trans (Supplemental Fig. 10A). Most peaks were unperturbed by the addition of the LBD, supporting the main interactions of the AB domain to be with the DBD (Supplementary Fig. 10A). As observed previously³⁵ (preprint), three residues, F38, W39 and T41 underwent a large decrease in intensity, pinpointing this region as the primary LBD interaction site (Supplementary Fig. 10B). Using ¹⁵N-PPARγ^LBD, we mapped the binding site for ₃₈FWPT₄₁ on LBD to be positioned close to, but not within, the ligand binding pocket and the AF2 site (Supplementary Fig. 10C–E). Addition of ligand (rosiglitazone) to full-length PPARγ had no effect on the R₂ profile or interactions with the DBD but did shift some peaks of the AB domain very slightly back towards the positions of the AB-DBD (Supplemental Fig. 11), highlighting a communication between the ligand binding pocket and the AB domain. The ₃₈FWPT₄₁ remained bound to LBD in the presence of ligand. Importantly, ligand binding to the LBD did not lead to release of the AB domain.

An integrative structural model of full-length PPARγ

To combine the experimental data into an integrative structural model we focused on PPARγ^FL where a complete SAXS data set was obtained. First, we ran CG-MD simulations and calculated SAXS scattering intensities. The simulations were in a reasonable but evidently imperfect agreement with the SAXS data on PPARγ^FL (Supplementary Fig. 12). To improve the accuracy of the ensemble, we integrated information from several experiments: (i) SAXS data on PPARγ^FL, to inform the overall ensemble dimensions and placement of domains, (ii) SAXS data on PPARγ^ΔAB, to inform on the extent of DBD-LBD interactions and compaction, and (iii) the difference between AB domain R₂ relaxation rates in PPARγ^FL and PPARγ^AB (ΔR₂), to inform the interactions between the AB domain and the rest of the protein. To accomplish this integration, we used a two-step BME reweighting approach in which we (i) reweighted the ensemble against SAXS data on PPARγ^ΔAB (Supplementary Fig. 13), and (ii) used the resulting weights as a Bayesian prior for simultaneously reweighting against ΔR₂ values and SAXS data on PPARγ^FL (Supplementary Fig. 13). This analysis made it possible to integrate the SAXS data on PPARγ^ΔAB, while allowing the final ensemble to deviate from this data if it was not consistent with the experiments on PPARγ^FL. Using this approach, we generated an ensemble model of full-length PPARγ consistent with all the experimental data (NMR and SAXS). Reweighting increased the population of conformations with DBD-LBD interactions and decreased the population of conformations where the hinge region is extended and the DBD and LBD are dissociated (Supplementary Fig. 14). To shed light on the molecular interactions between the different domains, we calculated contacts between PPARγ^AB and the DBD and LBD in the ensemble. The analysis revealed that the AB domain interacts with both folded domains, with most of the interactions concentrated on the DBD and hinge region (Fig. 2I), consistent with the NMR data. These contacts result in a relative compaction of the AB domain in the context of the full-length protein (Fig. 2J). The observation of dominant AB-DBD interactions is consistent with the R₂ relaxation data. These show that the R₂ rates in the AB domain are substantially increased in the AB-DBD context, while the AB-DBD and full-length variants have similar R₂ rates in the AB domain, except in the F38–T41 and T67–S72 regions. To determine whether the strength of AB–DBD interactions in our ensemble model was driven by the inclusion of NMR relaxation data, we also generated an ensemble targeting only the SAXS data. When compared with the full model, this resulted in only minor differences in AB contacts on the DBD and LBD, showing that our model predicts stronger interactions with the DBD independently of the NMR relaxation data (Supplementary Fig. 15). Thus, our results consistently show that the AB domain dominantly and dynamically interacts with the DBD, with fewer interactions to the LBD.

The interdomain contact regions in the AB domain are potent activation domains

Previous work had shown the AB domain of both PPARγ and PPARɑ to carry transactivating function, with the activity of the AB domain of PPARɑ being the highest⁴⁹. We therefore sought to investigate if the segments of the disordered AB domain that in PPARγ interact with the DBD possess activity as transactivation domains (ADs). We generated synthetic transcription factors consisting of the AB fragments from PPARγ, which made contacts to the DBD—or the full AB domain—fused to a halotag followed by the estrogen binding domain, a synthetic zinc finger, a T2A site and mScarlet (see methods), all expressed from a strong promoter (Fig. 3A). We transiently transfected these constructs into HEK293 cells bearing a synthetic reporter locus leading to the transcription of GFP, the fluorescence of which was taken as proxy for transcriptional activity^50–52. We observed a complex pattern of activity (Fig. 3B). First, we found that the chimeras covering parts of SIγ or the entire SIIγ, respectively, were individually very active (Fig. 3B, C). Complicating the pattern, the fragment bearing SIγ, the intervening sequence and SIIγ was less active than the minimal fragments of SIγ and SIIγ, and the full AB domain even less so, suggesting some inhibitory properties (Fig. 3B, C). Addition of residues N-terminal of SIγ demonstrated transcriptional activity like SIγ alone, indicating that the first 30 residues in PPARγ do not in HEK cells influence transcriptional activity, at least out of full-length context. The SIγ region resides in the middle of a large region predicted to be an AD (Fig. 3D). SIIγ is not classified as an AD using the PADDLE⁵³ and ADpred⁵⁴ prediction models, possibly because it is not sufficiently acidic. SIIγ therefore likely belongs to a different class of AD⁵⁵. Analogously, for PPARɑ, regions from E20 to S40, overlapping SIɑ, were predicted to be ADs (Supplementary Fig. 16). We note that in line with our observation, a larger fragment of PPARγ covering S22–E101 was in a previous high-throughput AD-screen categorized as a repressor domain⁵⁶. Also, effects of phosphorylation are not apparent in large-scale tile-based screens^55,57, likely due to limited modification of the short fragments. Thus, we cannot exclude that effects of phosphorylation or other PTMs that regulate the full-length protein could remain undetected in this assay.

Fig. 3. The interdomain communication centers in the AB domain are potent activation domains.

A Fragments (1–5) from the AB domain in PPARγ were fused to a synthetic DBD and B their activities were measured based on their GFP signal. The gating strategy is described in the supplementary materials (Supplementary Fig. 23). A minimum of 10,000 cells were analyzed (Table S3). Error-bars indicate the standard deviation across three transfections each performed on a different day. C The R₂ relaxation rates (left y-axis) of the AB-DBD (red) and full-length (gray) PPARγ are depicted together with fragments 1–5 (right y-axis). Bars represent one value and error bars the standard error of the fit. D Prediction of AD in PPARγ and PPARɑ using PADDLE and ADpred. A threshold at 0.6 (PADDLE) and 0.8 (ADpred) are marked with dotted lines^87,88. E Experimental setup for measuring degron activity. PPARγ was tiled into 33 different 30-residue fragments each overlapping by 15 residues and fused to the C-terminus of GFP. The GFP-tile fusions were expressed from a genomic locus in HEK293T cells. For normalization, mCherry was produced from an internal ribosomal entry site (IRES). Cells were analyzed by flow cytometry. (F) The mean GFP/mCherry ratio reporting on the abundance of the GFP-tile fusions plotted for each of the 33 different tiles. A minimum of 10,000 cells were analyzed. The data are presented as mean values. The error-bars indicate the standard deviation (n = 10,000). The experiment was performed twice using independent biological replicates (separate transfections). No statistical tests were performed. The gating strategy is including in the supplementary materials (Supplementary Fig. 17). AB domain highlighted in gray. Data in (F) reported in Source Data file.

To better understand the complex transcriptional activity pattern, we explored whether any of the fragments would carry a degron motif⁵⁸ that would lead to their degradation and hence explain a lower transcriptional activity. PPARγ was tiled into 30-residue fragments, overlapping with 15 residues, and introduced into a mammalian expression vector fused to the C-terminus of GFP (Fig. 3E)^57,59,60. The expression vector was integrated into HEK293T cells containing a landing-pad for site-specific recombination. mCherry was produced from an internal ribosomal entry site (IRES) in the same mRNA to enable normalization of the GFP signal. The cells were analyzed by flow cytometry and the GFP/mCherry ratios reflecting the abundance of the GFP-fused tiles determined (Supplementary Fig. 17). Overall, only slight variations in the abundance of the AB domain fragments were observed. Thus, compared to the stronger degrons found in both the DBD (tiles 11 and 12) and the LBD (e.g. tiles 23 and 27), no degron activity could be observed in the AB domain (Fig. 3F). The lower transcriptional activity of some fragments, including the full AB domain, is therefore unlikely to be due to degradation.

The AB domains are released from the DBD upon DNA binding

To investigate whether the interdomain interactions between the AB domains and the DBD affect, or are themselves affected by DNA binding, we recorded ¹H,¹⁵N HSQC spectra of the AB-DBD and full-length variants of PPARγ and PPARɑ bound to a consensus PPRE sequence or non-target DNA (Fig. 4A, Supplementary Fig. 18). When we added consensus PPRE DNA to PPARγ^AB-DBD and PPARɑ^AB-DBD or to PPARγ^FL and PPARɑ^FL, we observed substantial CSPs, where the chemical shifts—with few exceptions—fully returned to those of the isolated AB domains (Fig. 4B-E). We did not observe any peaks originating from F38 and W39, indicating that contacts remained to the LBD in the DNA bound state. However, this data suggests that the AB-DBD interactions observed in the DNA-free protein are abolished upon DNA binding, indicating that the AB domain and DNA has overlapping interaction sites on the DBD. CSPs were also observed upon addition of non-target DNA, but not to the same extent, suggesting weaker affinities—a scenario observed for many other transcription factors such as p53⁶¹ and Sox2⁶² (Fig. 4B-E). The backbone dynamics of the AB domains in the AB-DBD and full-length variants were measured in the presence of consensus PPRE DNA or non-target DNA by ¹⁵N R₁, R₂ and steady-state hetNOEs (Fig. 4F–I, Supplementary Fig. 19). The R₂s decreased upon addition of consensus PPRE DNA in both PPARγ^AB-DBD and PPARɑ^AB-DBD (Fig. 4F, G) and in both full-length proteins (Fig. 4H, I) and returned to levels similar to those observed for the isolated AB domains. As expected, the C-terminal residues of the AB domains retained increased R₂ rates in the DNA bound state due to the slower tumbling rates.

Fig. 4. The AB domains are released from the DBD upon DNA binding.

A Modular structure of PPARγ and PPARɑ and the color of variants, PPARγ^AB, dark green; PPARγ^AB-DBD, dark red; PPARγ^FL, black, PPARɑ^AB, light green; PPARɑ^AB-DBD, brown; PPARɑ^FL, gray. ¹H,¹⁵N HSQC spectrum zooms on K63 and T75 when adding DNA to (B) PPARγ^AB-DBD and C PPARγ^FL and zooms on A16 and L72 when adding DNA to (D) PPARɑ^AB-DBD and E PPARɑ^FL. ¹⁵N R₂ relaxation rates of the AB domain (dark green) and the AB-DBD (gray bars) in (F) PPARγ and G PPARɑ in the presence and absence of a consensus (pink) and non-target (blue) 15-bp DNA fragment. F ¹⁵N R₂ relaxation rates of the AB domain (dark green) and the full length (gray bars) in (H) PPARγ and I PPARɑ in the presence and absence of a consensus PPRE (pink) and non-target (blue) 15-bp DNA fragment. All relaxation data was recorded at 750 MHz. Error bars represent standard errors from the fits. Black star; unassigned residue in one or more of the variants, gray circle; overlapping residues, orange triangle; prolines. Data in (F, G, H, I) reported in Source Data file.

Collectively, these results suggest that DNA binding is associated with a release of the AB domains from the DBDs, causing the AB domains to behave more like AB in isolation. For PPARγ^FL bound to DNA, the amide chemical shifts of F38 and W39 remained undetectable, indicating that, while the AB domain is released from the DBD, ₃₈FWPT₄₁ remain in contact with the LBD (Fig. 4H). At similar concentrations, consensus PPRE DNA was more efficient in releasing the AB domain compared to non-target DNA, pointing towards affinity differences.

The AB domains negatively regulate DNA binding affinity but not specificity

To quantify the effect of the AB-DBD interdomain interaction on DNA binding, we determined the dissociation constants (K_D) of monomeric PPARɑ and PPARγ for DNA in two ways; first, through fluorescence polarization (FP) assays using 6-carboxyfluorescein (FAM)-fluorescently labeled consensus PPRE DNA and non-target DNA as probes and second, using isothermal titration calorimetry (ITC) with unlabeled DNA. From monitoring the change in the FP signal as a function of increasing PPAR concentration, we obtained K_Ds of DNA interactions for all six variants (Table 1, Supplementary Fig. 20). For PPARγ^AB-DBD and PPARγ^ΔAB, we also obtained K_Ds for consensus PPRE DNA from ITC (Table 1, Supplementary Fig. 21). We observed a single transition in FP, but note that at high protein concentration, a continued increase in the anisotropy was seen (Supplementary Fig. 20). A similar effect was not observed with ITC given the PPARs were in the cell, and we contribute the increase in FP to nonspecific binding or protein aggregation at high concentration.

Table 1.

Binding affinities of consensus PPRE and non-target DNA to PPAR variants at 298 K^#

	Consensus PPRE DNAa	Non-target DNAa	Specificity factorb
FP	KD (μM)	KD (μM)	KDnon-target/KDconsensus
PPARγFL	1.0 ± 0.1	53 ± 5	53 ± 7
PPARγΔAB	0.14 ± 0.02	10 ± 0.5	71 ± 11
PPARγAB-DBD	0.5 ± 0.1	2.9 ± 0.1	6 ± 1
PPARαFL	0.11 ± 0.01	2.0 ± 0.1	18 ± 2
PPARαΔAB	0.03 ± 0.004	0.8 ± 0.04	27 ± 4
PPARαAB-DBD	0.18 ± 0.04	1.1 ± 0.1	6 ± 1
PPARɑAB-γDBD	2.0 ± 0.3	5.7 ± 0.5	3 ± 0.5
PPARγFL:RXRαFL PPARγΔAB:RXRαFL RXRα:RXRαFL	0.003 ± 0.001c 0.005 ± 0.001c 0.05 ± 0.02c
ITC	Consensus PPRE DNAd KD (µM)
PPARγΔAB PPARγAB-DBD	0.27 ± 0.04 0.5 ± 0.1

^# Data in reported in Source Data file. For FP: Errors in K_D represent standard errors from the fits. For ITC: Errors in K_D represent propagated standard deviations calculated from independent measurements.

^aDNA used was 14 base pairs (Consensus: 5′-CAAACTAGGTCAAG-3′, non-target: 5′-CGGGTGACTATAAT-3′).

^bSpecificity factor error calculated as $\sigma \frac{K_{D-ran}}{K_{D-con}}=\sqrt{{\left(\frac{\sigma K_{D-ran}}{K_{D-ran}}\right)}^2+{\left(\frac{\sigma K_{D-con}}{K_{D-con}}\right)}^2}\times \frac{K_{D-ran}}{K_{D-con}}.$

^cDNA used was 21 base pairs (Consensus: 5′-GCAAACTAGGTCAAAGGTCAG-3′).

^dDNA used was 15 base pairs (Consensus: 5′-GCAAACTAGGTCAAG-3′).

The two methods provided comparable affinities, showing that the fluorophore did not interfere with the main interaction. Additionally, we find that PPARɑ^FL binds ninefold stronger to consensus PPRE DNA than PPARγ^FL (0.11 µM versus 1.0 µM K_D). Furthermore, the PPARs only had a ∼20-fold (PPARɑ) and ∼50-fold (PPARγ) preference for consensus PPRE DNA relative to the non-target DNA sequence that we used, supporting the observed difference in the propensity of the two oligonucleotides to release the AB domains from their DBDs. Finally, and remarkably, the presence of the AB domain negatively affected the affinity for DNA in both proteins, weakening the affinity 4- to 7-fold compared to the full-length proteins. Thus, the AB domains carry autoinhibitory function. Likewise, and only in PPARγ, the presence of the LBD has a negative effect on affinity, in accordance with the interaction we observe between ₃₈FWPT₄₁ in the AB domain and the LBD (Supplementary Fig. 10). The specificity constants (defined here as K_D^non-target/K_D^consensus) highlight that the AB domain does not impact DNA binding specificity. We note that the specificity constant is lower in the absence of the LBD, suggesting that the LBD contributes to DNA binding, despite negatively affecting affinity. Finally, we measured the affinity of the chimeric PPAR^ɑAB-γDBD for DNA. The chimera bound weaker to consensus PPRE DNA compared to PPARγ^AB-DBD (2.0 µM versus 0.5 µM) and had a weaker affinity for non-target DNA, resulting in a specificity factor similar to the wild-type PPARs. Together, these results show that all PPAR domains affect DNA binding affinity, with the DBD as the main determinant of DNA binding specificity and the AB domain exerting autoinhibitory function in DNA binding.

Autoinhibition remains in the heterodimer of PPARγ:RXRα

To address whether the effect we observe from the AB domain on the monomer remains in the more biologically relevant heterodimer with RXRα, we recombinantly produced the full-length RXRα (RXRα^FL). As RXRα^FL also forms homodimers of high affinity (40–60 nM)⁶³, we generated the heterocomplex in two ways to minimize any observed effect originating from the RXRα homodimer, depending on the experiment. In the first approach, where we apply NMR, we added a ∼2-molar excess of RXRα to rosiglitazone-bound ¹⁵N-PPARγ^FL and recorded R₂-values and compared to those of the PPARγ^FL monomer and to the AB domain alone (Fig. 5A). Although complicated by the size of the complex, we were able to obtain R₂-values for many residues, showing a large increase in the R₂-values consistent with the formation of a complex of twice the size, confirming the formation of the heterodimer. From the elevated R₂-values and the positions of the peaks in the ¹⁵N-HSQC NMR spectrum, which generally resemble those of the PPARγ^FL monomer (Fig. 5B and Supplementary Fig. 22B), we conclude that the interdomain contacts between the AB-domain and the DBD are maintained in the heterodimer. We note that in the heterodimer, additional CSPs were seen for the N-terminal region of the AB domain, suggesting some additional RXR-dependent contacts for this region (Fig. 5B). We next added consensus DNA to the NMR sample in an excess to saturate both the RXRα^FL homodimer and the PPARγ^FL:RXRα^FL heterodimer and recorded R₂-values. Adding DNA lowered the R₂s to levels of the monomer (Fig. 5A), suggesting either the AB domain to remain associated with the DBD or that the AB domain is released. From an analysis of the chemical shift (Fig. 5B, Supplementary Fig. 22A), we observed a return of the chemical environment to that of the isolated AB domain as well as sharpening of the peaks, supporting AB release from the DBD. In the DNA bound heterodimer, we now observed the NMR peaks from the LBD-contacting residues F38–T41, suggesting that these contacts are weakened, but still present. Thus, in summary the data on the biologically relevant complex highlight that also in the heterodimer with RXRα^FL, the AB domain of PPARγ^FL form interdomain interactions, which are released upon DNA binding.

Fig. 5. The AB domain acts as a gatekeeper for untimely DNA binding.

A ¹⁵N R₂ relaxation rates of the AB domain in PPARγ^AB (green), PPARγ^FL (black), in the heterodimer of rosiglitazone bound PPARγ^FL and RXRɑ^FL (brown), PPARγ^FL with consensus DNA (pink), and the heterodimer of rosiglitazone bound PPARγ^FL and RXRɑ^FL with consensus DNA (blue). Error bars represent standard errors from the fits. Black star; unassigned residue in one or more of the variants, gray circle; overlapping residues, orange triangle; prolines. B ¹H,¹⁵N HSQC spectrum zooms on D11, T20, I45, and A101 of PPARγ in PPARγ^AB (green), PPARγ^FL (black), in the heterodimer of rosiglitazone bound PPARγ^FL and RXRɑ^FL (brown), PPARγ^FL with consensus DNA (pink), and the heterodimer of rosiglitazone bound PPARγ^FL and RXRɑ^FL with consensus DNA (blue). C DNA binding affinities measured by fluorescence polarization of isolated PPARγ^FL (lower left) and PPARγ^ΔAB (upper left) and in the heterodimer complex of rosiglitazone-bound PPARγ^FL:RXRɑ^FL (lower right), rosiglitazone-bound PPARγ^ΔAB:RXRɑ^FL (upper right). The K_Ds are in nM. Error bars represent standard deviation of n = 3 biological replicas. D Suggested model of the interdomain interaction with the AB domain acting as a gatekeeper for untimely DNA binding to monomeric PPARs. At the top left, the AB domain (green) exists in a dynamic open-close equilibrium forming contacts mainly with the DBD and less so with the LBD and the AB domain is to some extent is accessible. The AB domain provides auto-inhibition of DNA binding. In the DNA bound state, the AB domain is released and is as accessible as in the isolated state. Once the heterodimer of PPARγ (grey) with RXRɑ (blue) forms, the AB domain (green) still contacts the DBD, and not fully released upon DNA binding, The heterodimer binds DNA much stronger overcoming the interaction barrier of the AB domain. The AB domain is shown in green, ligand binding pocket as a ring, RXRɑ in blue and a coregulator in yellow. Data in (A, C) reported in Source Data file.

In the second approach, to quantify the effect of the interdomain interactions on DNA binding to the heterodimer by FP, we isolated heterodimeric complexes between PPARγ^FL:RXRα^FL and PPARγ^ΔAB:RXRα^FL, respectively, using SEC. Here we mixed a 3:1 molar excess of PPARγ^FL or PPARγ^ΔAB with RXRα^FL to minimize contributions from the RXRα^FL homodimer. We further added rosiglitazone to stabilize the heterodimer. Using FP, the DNA binding affinity of the heterodimer was two orders of magnitudes stronger than of monomeric PPARγ^FL with a K_D of 5 nM for consensus DNA (Fig. 5C, Table 1). Removing the AB domain increased the affinity by a factor of two to 3 nM (Fig. 5C, Table 1), highlighting that also in the heterodimer with RXRα^FL, the AB domain exert an autoinhibitory function in DNA binding. In this case, however, the effect is less pronounced. Notably, the RXRα^FL homodimer has a 10x lower affinity for the same DNA (Table 1).

Discussion

Here, we have studied the diverse functional properties of the disordered AB domains in PPARγ and PPARɑ by combining experiments and computation. Our results show that the AB domains of the two proteins share common features; they are disordered, highly dynamic with a single transient helical region, and they make dynamic interactions with the DBD—interactions that dominate interdomain communication. For PPARγ, as shown previously³⁵ (preprint), the AB domain also makes contacts with the LBD, but in our work, the interactions with the DBD are more prevalent. The interactions with the DBD are mediated mostly by two regions in the AB domain, SI and SII. These regions have confirmed and predicted AD function. Upon DNA binding to the DBD, but not upon ligand binding to the LBD, the AB domain is released from the DBD, fully exposing SI and SII making them available for AB domain-dependent coregulator binding, e.g. for p300, Tip60 and CBP reported to be involved in AF1-regulated responses¹⁰. For both PPARs, the AB domain negatively regulates DNA binding affinity but not DNA binding specificity, showing that the AB domains carry autoinhibitory function (Fig. 5D).

Beyond the shared features of the two proteins, we also found distinct traits. Full-length PPARɑ binds with a 9-fold higher affinity to consensus PPRE DNA compared to full-length PPARγ. Earlier studies have found that PPARγ in complex with RXRɑ generally binds DNA stronger than PPARɑ, although this does depend on the DNA sequence tested²⁷. In line with this, recent work has decomposed half-site sequence requirements for a series of NRs, revealing a preference for PPARγ for thymine at the 5’-side of the PPRE⁶⁴. Our DNA sequences are low in thymine at the 5’ flanking region, which may affect the PPARγ affinity. PPARγ may also be more dependent on RXRɑ binding to induce an optimal DBD conformation prior to DNA binding⁶⁵, but quantitative affinities for DNA binding of the heterodimer have to our knowledge not been reported. We were unfortunately not able to produce the isolated DBD, and further experiments are needed to relieve this caveat.

To evaluate the effects of the disordered region in the biologically relevant heterodimer of PPAR with RXR, we generated the heterodimer between the full-length versions of RXRα and rosiglitazone-bound PPARγ, both in the presence and absence of the AB domain of PPARγ. Remarkably, the interaction of the AB domain with the DBD remains in the heterocomplex and is not in competition with binding to the RXRα DBD, consistent with a specific interaction. In the heterodimer, we also observed that the AB domain increases K_D, albeit to a smaller extent than in monomeric PPARγ. Biologically this makes sense, as binding of DNA to the monomer before it reaches chromatin can be prevented by the interdomain interactions. However, once the heterodimer forms, the affinity for DNA increases, allowing binding to bypass the gatekeeper function of the AB domain. Thus, we suggest that the interdomain interaction between the AB domain and the DBD in PPARs serves as gatekeeper for untimely DNA binding (Fig. 5D).

Surprisingly, DNA binding affinities have been quantified only for a relatively limited number of NRs. In these studies, different effects of the AB domain on DNA binding affinity have been observed. In the AR, the AB domain contains co-regulator-like SLiMs that interact with the LBD to stabilize the lifetime of the hormone-bound state, likely by competing with binding at the coregulator site³³. Still, removing the AB domain from the AB-DBD context of the AR lowers the K_D by a factor of 2, very similar to what we observed here for the PPARγ:RXRα complex and with a similar range of affinities⁶⁶. For HNF4ɑ, DNA binding by the DBD and hinge alone gave a K_D of 6 µM, whereas adding the LBD increased affinity 75-fold to 80 nM. For this NR, removing the AB domain—only 51 residues long—had no effect on DNA binding affinity⁶⁷. Affinities for DNA have been quantified for isolated DBDs of REV-Erb (K_D = 0.13 µM), RXR (K_D = 0.2 µM), RORγ1 (K_D = 0.4 µM for the monomer; 0.12 µM for the dimer (0.12–0.19 µM depending on DNA sequence)), and of NR4A2 (K_D = 0.75 µM)⁶⁸. These affinities are comparable to what we measure here. However, much stronger in vitro affinities were quantified for the GR (0.3–0.7 nM)⁶⁹, whereas in living cells, K_Ds were estimated to be 0.7– 0.9 µM for GR and 0.1 µM for AR⁷⁰. These differences suggest that chromatin context matters. Importantly, the presence of RXR will impact these affinities and could do so differentially for the two PPARs. This remains to be quantified.

The autoinhibitory effect on DNA binding originating from the AB domains is different for PPARγ and PPARɑ. The change in affinity upon removal of the AB domain is more pronounced in PPARγ (sevenfold) compared to PPARɑ (fourfold); nonetheless, in both cases, deletion of the AB domain did not impair DNA binding, consistent with previous findings¹⁰. This autoinhibitory effect on DNA binding can originate from the AB domain acting as a stochastic chain, sterically hindering access to the DBD, with a larger effect from the longer AB domain in PPARγ, or it may originate from direct interactions of the AB domain with the DNA binding sites on the DBD. Most likely, the effects we observe originate from a combination of the two. Of possible importance, the AB domain is host for many potential and different posttranslational modifications, including phosphorylation, which will act to increase the overall negative charge of the AB domain. Depending on the site of phosphorylation, and type of modification, these changes to the AB domain can be expected to shift the ensemble distribution, increasing or decreasing the autoinhibitory function further. This suggestion aligns with the phosphorylation-dependent increase and decrease in transactivation by PPARγ^71,72, feeding into the AB-driven recruitment of co-regulators as well.

Comparing the observed autoinhibitory function of the disordered AB domain on DNA binding in PPARs to other TFs, we see that for Sox2, whose IDR also contacts the DBD, DNA binding affinity is unaffected by its presence, independently of whether the DNA is non-target or consensus DNA⁶². For p53, the disordered N-terminal domain has been shown to similarly regulate DNA binding affinity negatively (both non-target and consensus DNA) but here, the disordered region increases DNA binding specificity by some yet to be understood mechanism⁶¹. For both PPARs, we did not observe any effects from the AB domains on DNA binding specificity, indicating that the highly negatively charged AB domains do not make contact to DNA or promote DNA-binding-favorable conformations of the DBD. Earlier cellular data have shown that the subset of target genes that are affected by the removal of the AB domain, are inversely regulated when the AB domains are swapped between subtypes¹⁰. Combined with the biophyicsal data presented here, these data suggest that for certain genes, transcriptional specificity can be somewhat decoupled from the DNA binding affinity and specificity and rather be linked to the properties of the AB domain. Thus, the AB domains in PPARs may carry both similar and different functions compared to disordered regions in other TFs.

For PPARγ, we combined NMR, SAXS and MD simulations to construct an experimentally supported model of the full-length receptor. The data supports previous models of PPARγ:RXRɑ:DNA, which lacked the AB domain, in which the DBD and LBD in PPARγ are in contact^18,73. Such contacts may allow for allosteric effects, as suggested to be present in the GR⁷³ and other NRs^67,74–76 and here supported by our DNA binding data on PPARγ. For PPARγ, removal of the LBD leads to stronger binding of both consensus PPRE DNA and non-target DNA. The effect was largest for non-target DNA, lowering the specificity factor, and making non-target DNA sequences competitive to consensus PPRE DNA. The LBD did not exert similar negative regulation of DNA binding to PPARɑ. Thus, the negative effect originating from the LBD could depend on contacts between the AB and the LBD, which in PPARγ involves ₃₈FWPT₄₁, a sequence motif that is absent in PPARɑ. It is known that the C-terminal extension of the DBD into the hinge region is important for DNA binding affinity and selectivity, and this region is maintained in both AB-DBD constructs used here. In other NRs, synthetic ligands for the hydrophobic binding pocket were demonstrated to cause significant changes in the DNA binding affinity, although they bind only to the LBD, supporting further domain-domain communication^77,78. Similar communication was observed in PPARγ where addition of rosiglitazone redistributed the AB ensemble slightly towards DBD interactions. The differences we observe between PPARγ and PPARɑ point towards different disorder-based regulatory mechanisms provided by the properties of the AB domains. That such disorder-based mechanisms are specific is supported by the chimeric PPAR^ɑAB-γDBD, leading to a larger population of open states, as supported by the observation that regulation of a subset of isotype-dependent genes is sensitive to which AB-domain they are carrying.

In conjunction with the results from AB domain-swap and AB domain-deletion studies performed in cells^29,79, the low discrimination of the DBD between consensus PPRE DNA and non-target DNA we observe here, suggests that transcriptional specificity for a subset of genes to some extent resides in the AB domain, but it is not clear how this specificity arises. The AB domain may enhance the efficiency of promoter search by scanning extensive stretches of DNA, as suggested⁸⁰. It is also possible that the dynamic interactions with the DBD may allow not only for AB-release in the presence of DNA, but also AB-release in the presence of co-regulators, suggesting that genome specificity for some target genes is encoded in the AB domains. Similar mechanisms lending support to disorder-embedded specificity were suggested from recent high-throughput TF-studies in budding yeast⁸¹ and combined these results challenge current views on gene regulation⁸². If the AB domain sequences carry so-called barcodes for binding to co-regulators and hence only bind discrete sets of co-regulators and other TFs, the AB domains may help determine transcriptional specificity. Our results on the chimera suggest that swapping the AB domains also directly impacts the interdomain interactions, altering the population between accessible and inaccessible states. Depending on the DBD used in activity-based screening assays, any interactions between a disordered region and the unrelated DBD can have potential impact on the readout of these methods. Whether interdomain interactions can explain the complex activity data we obtain for the longer tiles tested remains to be addressed. The chimera binds DNA weaker, hinting at combined effects from the contacts between the AB domain and the DBD which were also not disclosed here. Still, our results combined with those of others suggest that the sequences of the AB domains encode a complex interplay between recruitment of co-regulators, stabilizing interactions with the DBD, and autoinhibitory effect on DNA binding. How these effects are encoded in their sequence should be studied further. With disorder as a common denominator for the AB domain of the NRs, it is possible that their very different sequence features encode different functions. The AB domains of the estrogen receptor and the androgen receptor behave differently to the PPARs, forming internal hydrophobic clusters within their AB domains and polyglutamine α-helical structures, respectively^83,84. More directed studies towards understanding the role of disorder in nuclear receptor are needed, just as mapping of interdomain interactions and to what extent such communication affects function.

In summary, we find that structural disorder in PPARs leads to dynamic interdomain communication between the AB domain and the DBD and LBD, dominated by the DBD interactions and with the interacting regions carrying AD activity. The interdomain communication negatively affects DNA binding affinity, whereas the DNA binding specificity of the PPAR DBDs is inherently low and unaffected by the AB domain. The disordered AB domain in PPARs contributes with autoinhibitory function through dynamic interaction with the DBD, an interaction that remain in the heterodimer with RXRα. Once the heterodimer forms, affinity for DNA increases, bypassing the inhibitory interactions between the AB domain and the DNA. Thus, DNA binding releases the AB domain, and—once released —the AB domain becomes more accessible to co-regulators, the selection of which may be encoded in the AB domain sequence. Disorder in PPARs thus carry key functional roles and we suggest that similar roles yet to be uncovered exist in the many NRs that so far has gone under the radar.

Methods

Protein expression and purification

All hPPARγ2, hPPARα, and hRXRα protein variants were produced with an N-terminal hexahistidine small ubiquitin-like modifier (H₆-SUMO) tag, which can be cleaved off with ubiquitin-like protein protease 1 (ULP1). The coding regions were inserted into a modified pET24a vector (Twist Bioscience, US) and expressed in Escherichia coli (E. coli) BL21(DE3) cells. All hPPARγ2, hPPARα and hRXRα variants were cultured in Luria-Bertani (LB) broth medium, or M9 minimal medium containing ¹⁵N-NH₄Cl or ¹⁵N-NH₄Cl and ¹³C₆-glucose. Expression of all variants was induced at high OD₆₀₀ ~ 1–1.2 with 0.5 mM isopropyl-β-D-1-thiogalactopyranoside (IPTG). Cells expressing hPPARγ2^AB and hPPARα^AB were grown for 4 h at 37 °C after induction, while all other protein variants were grown overnight at 16 °C before being harvested by centrifugation (5000 x g, 15 min, 4 °C). The cell pellet was resuspended in Buffer A (20 mM Hepes-HCl pH 7.4, 500 mM NaCl, 10 mM Imidazole, 5 mM β-mercaptoethanol (β-ME)), and the cells lysed by sonication. The soluble fraction was collected by centrifugation (20,000 x g, 45 min, 4 °C) and applied to 3 mL of pre-equilibrated Nickel Sepharose Fast Flow resin (Cytiva, Uppsala, Sweden), pre-washed with 10 x column volume (CV) of Buffer B (20 mM Hepes-HCl, pH 7.4, 1 M NaCl, 10 mM Imidazole, 5 mM β-ME), and then with 10 x CV of Buffer A. The proteins were eluted with 20 mL of Buffer C (20 mM Hepes-HCl, pH 7.4, 500 mM NaCl, 250 mM Imidazole, 5 mM β-ME). All protein variants, except hPPARγ2^AB, hPPARα^AB, hRXRα^AB and hRXRα^FL were dialyzed in a three-step manner at 4 °C; first the eluted fraction was dialyzed against 20 mM Hepes-HCl, pH 7.4, 400 mM NaCl, 200 mM Imidazole, 5 mM β-ME, then against 20 mM Hepes-HCl, pH 7.4, 300 mM NaCl, 100 mM Imidazole, 5 mM β-ME, and finally against 20 mM Hepes-HCl, pH 7.4, 200 mM NaCl, 5 mM β-ME. ULP1 was added to the protein solution prior to the final dialysis step. For hRXRα, just one dialysis step was used, identical to the last step for PPAR (low salt, no imidazole and added ULP1). Subsequently, the samples were applied to a 5 mL HiTrap Heparin HP column (Cytiva, Uppsala, Sweden) and eluted with a linear gradient of 200 mM to 2 M NaCl over 20 x CV. Relevant fractions were pooled and concentrated using a 10-kDa cutoff Amicon® centrifugal filter device (Merck) before purified further on a Superdex 200 increase 10/300 GL (Cytiva, Uppsala, Sweden) in 20 mM Hepes-HCl (pH 7.4), 200 mM NaCl, and 2 mM dithiothreitol (DTT). ULP1 was directly added to the eluted fractions of hPPARγ2^AB, hPPARα^AB and hRXRα^AB and the solutions dialyzed in a one-step manner against 20 mM Hepes-HCl, pH 7.4, 200 mM NaCl and 5 mM β-ME at 4 °C. The cleaved sample was loaded onto a 3 mL pre-equilibrated Nickel Sepharose Fast Flow resin (Cytiva, Uppsala, Sweden) and the flowthrough concentrated using a 10-kDa cutoff Amicon® centrifugal filter device (Merck, Darmstadt, Germany). The hPPARγ2^AB and hPPARα^AB samples were applied to a Superdex 200 increase 10/300 GL column for final purification in 20 mM Hepes-HCl (pH 7.4), 200 mM NaCl, and 2 mM DTT. The hRXRα^AB sample was purified by reversed-phase high-performance liquid chromatography on a Zorbax 300SB-C18 column (Agilent) and eluted with a gradient from 25% (v/v) acetonitrile and 0.1% (v/v) trifluoroacetic acid in aqueous solution to 50% (v/v) acetonitrile before the relevant fractions were lyophilized. The purification of the PPARγ^LBD followed a previous protocol^85,86 and the PPAR^αAB-γDBD chimera followed the purification of the AB-DBDs.

DNA The specific 14 base pairs (5’-CAAACTAGGTCAAG-3’) and 15 base pairs DNA (5´-GCAAACTAGGTCAAG-3´) PPRE sequence duplex, and the non-target 14 base pairs (5’-CGGGTGACTATAAT-3’) DNA sequence duplex were purchased from Integrated DNA Technologies, Inc. (Coralville, IA, USA).

Sequence alignments and prediction of activation domains

Multiple sequence alignments were performed in Jalview⁸⁷ using Clustal Omega. Prediction of activation domains was done using ADpred and PADDLE^53,54.

NMR

All NMR samples were made in 20 mM Hepes-HCl, pH 7.3 (adjusted at 5 °C), 135 mM KCl, 2 mM DTT, 0.25 mM 4,4-dimethyl-4-silapen-tane-1-sulfonic acid (DSS), 10% (v/v) D₂O (v/v) at 5 °C. High-salt conditions for R₂ relaxation measurement of PPARγ AB-DBD was achieved by adding KCl to a final concentration of 500 mM in a pre-existing NMR sample. Unless specified otherwise, all NMR data were acquired on a Bruker Avance III HD 750 MHz (¹H) spectrometer equipped with a 5 mm TCI Cryoprobe H-C/N-D. Raw free induction decays were processed and visualized using NMRPipe⁸⁸ or Bruker Topspin 3.7.0 and analyzed in CcpNmr Analysis 2.5.2⁸⁹. Chemical shifts were referenced according to the DSS signal (¹H)⁴³ or indirectly using gyromagnetic ratios (¹³C,¹⁵N). Spectra for backbone assignment of hPPARγ2^AB (800 µM) and hPPARα^AB-DBD (800 µM) were acquired on an 800 MHz (¹H) Bruker Avance Neo spectrometer equipped with a 5 mm CPTXO cryoprobe C/N-H-D. The assignments were done manually based on the analysis of ¹H,¹⁵N-HSQC, HNCACB, HNCOCACB, HNCO, HNCACO, and HN(CA)NNH experiments obtained with non-uniform sampling⁹⁰ and standard Bruker BioPack pulse sequences. Backbone assignment of hPPARɑ^AB-DBD (115 µM) and hPPARγ2^AB-DBD (330 µM) were performed analogously relying solely on HN(CA)NNH experiments. CSPs were calculated using Eq.( 1) (Eq. 1)⁴³:

$$CSP=\sqrt{{\left(\mathrm{\Delta }{\delta }_H\right)}^2+{\left(0.154\cdot \mathrm{\Delta }{\delta }_N\right)}^2}$$ 1

and the percentage helicity from the SCS Cɑ, with the value for a fully formed helix set to 3.09 ± 1.00 ppm⁹¹.

Transverse and longitudinal relaxation rates were recorded on ¹⁵N-labeled PPAR^AB samples (PPARγ, 175 µM; PPARɑ, 100 µM) through multiple ¹H,¹⁵N HSQC experiments each with different relaxation delays; 20, 60, 100, 200, 400, 600, 800, and 1200 ms (T₁) and 0, 33.9, 67.8, 101.8, 135.7, 169.6, 203.5 and 271.4 ms (T₂). The relaxation delay was set to 1.0 s. Each measurement was performed in triplicate, and the spectra were processed and analyzed using NMRpipe⁸⁸. The relaxation decays were fitted to a single exponential function to obtain T₁ (or R₁) and T₂ (or R₂), with non-linear regression used to approximate the best-fit parameters and their corresponding standard errors (SE). The hetNOEs were recorded with and without ¹H saturation with a 5 s relaxation delay and measured as triplicates. The hetNOEs values were calculated as the ratio of the peak heights in the spectra recorded with and without ¹H saturation⁹² and the error the standard error of the mean (SEM). T₁, T₂, and hetNOEs were analyzed using CcpNmr Analysis software version 2.5.2⁸⁹. Identically, T₁ (or R₁), T₂ (or R₂) and hetNOEs were measured of PPAR^AB-DBD (PPARγ, 130 µM; PPARɑ, 100 µM; PPARɑγ, 90 µM), PPAR^FL (PPARγ, 80 µM; rosiglitazone-bound PPARγ, 80 µM; PPARɑ, 100 µM) in the presence of a 1:1 molar ratio of protein:DNA with PPRE-specific (PPARγ^AB-DBD, 70 µM; PPARγ^FL, 80 µM; PPARɑ^AB-DBD, 70 µM; PPARɑ^FL, 100 µM) or scrambled (PPARγ^AB-DBD, 110 µM; PPARγ^FL, 80 µM; PPARɑ^AB-DBD, 100 µM; PPARɑ^FL, 100 µM) and heterodimer complex of 70 µM PPARγ^FL with rosiglitazone: 120 µM RXRɑ^FL, in the absence and presence of 200 µM PPRE-specific DNA. Specific duplex DNA (5’-GCAAACTAGGTCAAAGGTCAG-3’) and scrambled duplex DNA (5’-TTCGCAAGTTGCTATGTTCTC-3’) were purchased from Integrated DNA Technologies, Inc. (Coralville, IA, USA).

For in-trans titrations of PPARs, ¹H,¹⁵N-HSQC spectra were recorded of ¹⁵N-PPARγ^AB-DBD (50 µM) in the absence and presence of either 250 µM PPARγ^AB or 500 µM PPARγ^AB or 450 µM RXRɑ^AB. Additionally, ¹H,¹⁵N-HSQC spectra were recorded of ¹⁵N-PPARγ^AB (50 µM) in the absence and presence of either 200 µM PPARγ^ΔAB or 200 µM RXRɑ^ΔAB. Further, ¹H,¹⁵N-HSQC spectra recorded of ¹⁵N-PPARɑ^AB (50 µM) in the absence and presence of either 55 µM PPARɑ^ΔAB or 200 µM PPARγ^ΔAB or 200 µM RXRɑ^ΔAB. All samples were prepared in the above-mentioned NMR buffer, and all spectra were recorded at 5 °C.

For titration of ¹⁵N-PPARγ^LBD with PPARγ^AB, NMR samples containing 40 μM rosiglitazone-bound ¹⁵N PPARγ LBD were changed into PBS buffer (pH 7.3, 137 mM NaCl, 10 % D2O, 0.7 mM 4,4-dimethyl-4-silapentane-1-sulfonic acid). 200 µM unlabeled AB domain was added to measure the interaction between the proteins at 298 K on a Bruker AVANCE III 750-MHz (1H) spectrometer equipped with a cryogenic probe. Assignments of PPARγ LBD were exported from BMRB entry 17975⁹³, and transferred to the spectra without ambiguities. To predict the interaction of the AB domain on the ligand- and coactivator-bound LBD, AF3⁹⁴ was provided with the sequence of the PPARγ LBD (residues 231 to 505), the sequence of the interacting AB domain (PPARγ residues 35 – 44), and the sequence of a Mediator subunit 1 (MED1) peptide (NTKNHPMLMNLLKDNPAQD). In place of rosiglitazone, palmitic acid was chosen as a ligand from the AF3-provided ligand selection. All provided models were overlaid for an overview of the AF3-predicted ensemble of possible AB interactions on the LBD.

SAXS

Small angle X-ray scattering coupled to continuous-flow size exclusion chromatography (SEC-SAXS) was performed at the EMBL P12-bioSAXS beam line at PETRAIII (DESY, Hamburg, Germany⁹⁵). Detailed descriptions of the samples, sample environment and experimental setup/parameters, e.g., SEC running buffer, SEC column, flowrates etc., are reported in Supplementary Table 1 (PPARγ; ΔAB-PPARγ; PPARγ-AB) and Supplementary Table 2 (PPARɑ-AB). Briefly, the scattering intensities, I, measured as I(s) vs s, where s = 4πsinθ/λ, 2θ is the scattering angle and λ the X-ray wavelength, were recorded as sets of multiple 2D images spanning the entire SEC elution for each respective sample encompassing both running buffer and sample elution-peak scattering. Each individual data frame underwent 2D-to-1D azimuthal averaging using the SASFLOW pipeline⁹⁶, incorporating normalization of the scattering intensities to beam transmission (to consider buffer/sample absorption) and the calibration of I(s) to an absolute scale, cm⁻¹ (relative to the scattering from water at 20 ^oC). The resulting azimuthally averaged 1D scattering profiles for each SEC-SAXS dataset underwent additional processing in CHROMIXS⁹⁷ where the representative buffer- and sample-scattering data frames were selected, followed by buffer scattering subtraction, scaling and averaging, to generate the final reduced and background subtracted SAXS profiles for each PPARγ/ɑ variant. Additional multi-angle laser light scattering and refractive index measurements were coupled to the SAXS measurements either in parallel or sequentially using a Wyatt (Dernbach, Germany) miniDAWN® TREOS® (3-angle MALLS) instrument and an Optilab T-rEX refractometer (SEC-SAXS-MALLS-RI; λ MALLS/RI = 658 nm, 25 ^oC⁹⁸). Molecular weight correlation estimates of the eluting species across the SEC peaks for each sample were obtained from concentration estimates measured from RI in combination with the MALLS intensities. The MALLS/RI data were processed and analyzed using Wyatt ASTRA 7.0.1 software using refractive index increments, dn/dc, at 25 ^oC of 0.185 or 0.187.

Fluorescence polarization

Fluorescence polarization measurements of 5’−56-FAM labeled duplex DNA fragment in the absence or presence of increasing concentration of PPARγ^FL, PPARγ^ΔAB, PPARγ^AB-DBD, PPARɑ^FL, PPARɑ^ΔAB, PPARɑ^AB-DBD, and the chimera were done at a constant concentration of 20 nM 5’−56-FAM labeled duplex DNA fragment: either using the specific PPRE sequence: 56-FAM-CAAACTAGGTCAAG or a non-target DNA sequence: 56-FAM-CGGGTGACTATAAT generated based on a specified length of base-pairs (https://www.bioinformatics.org/sms2/random_dna.html). The concentration of protein varied in serial dilutions, starting at 60 µM in a total volume of 100 µL reaching 114 pM. The heterodimer complexes for affinity measurements (PPARγ^ΔAB:RXRɑ^FL and PPARγ^FL:RXRɑ^FL) were prepared by mixing the proteins in a 3:1 ratio with PPARγ in excess, as well as adding the ligand rosiglitazone, leaving the sample to incubate overnight at 4 °C. The heterodimer complexes were isolated from a Superdex 200 increase 10/300 GL (GE Healthcare). The complex concentration was varied in serial dilution, starting at 200 nM in a total volume of 100 µL reaching 137 pM. The DNA concentration was held constant at 20 nM 5’−56-FAM labeled duplex DNA fragment specific PPRE sequence: GCAAACTAGGTCAAGGTCAG. The affinity for this DNA was also quantified for the RXRɑ^FL homodimer, isolated from a Superdex 200 increase 10/300 GL (GE Healthcare).

The assay was performed in a 96-well non-binding surface black X40 Microplate (Greiner, Austria) at 25 °C in 20 mM Hepes-HCl, pH 7.4 (at 25 °C), 150 mM KCl, 2 mM DTT. Fluorescence anisotropy was measured on an Agilent BioTek Synergy H1 plate reader (excitation filter: 485/20 nm, emission filter 528/20 nm). The data were fitted in GraphPad Prism version 10.6.1 for MacOS (GraphPad Software, San Diego, California USA, www.graphpad.com) using the equation (Eq. 2):

$$Y=A+B*\frac{Dna+X+K_D-\sqrt{{\left(Dna+X+K_D\right)}^2-4*Dna*X}}{2}$$ 2

where A represents the signal amplitude normalized to probe concentration, B is the plateau value for unbound probe, DNA is the constant probe concentration, X is the varying protein concentration, and Y is the observed signal. Standard errors of the mean (SEM) were calculated from triplicates, and SEM for K_D was obtained directly from the non-linear regression analysis.

Isothermal titration calorimetry

All ITC experiments were recorded on a MicroCal PEAQ-ITC microcalorimeter (Malvern Panalytical, Malvern, UK) at 25 °C. Prior to recording, all samples were dialyzed extensively against 20 mM HEPES-HCl, pH 7.4, 150 mM KCl, 2 mM TCEP and spun down at 20,000 g for 10 min at 25 °C. Both PPARγ^ΔAB (10 µM) and PPARγ^AB-DBD (10 µM) were placed in the cell and specific PPRE DNA (100 µM) in the syringe. The first injection was 0.4 µL followed by 18 injections of 2 µL at an interval of 180 s between each. ITC data were processed using in-house software for ITC data-modeling (https://doi.org/10.5281/zenodo.14832178). The last 18 injections of each experiment were fitted to a one set of sites binding model assuming one or more binding sites that all behave identically and independently of one another. Triplicates were recorded for each interaction. The error propagation for N = 3 was done using (Eq. 3):

$$mean\sigma =\sqrt{{\left(\frac{\sigma 1}{N}\right)}^2+{\left(\frac{\sigma 2}{N}\right)}^2+{\left(\frac{\sigma 3}{N}\right)}^2}$$ 3

HEK293T cell culture and transfection

HEK293T cells (human, female) clonally integrated with an GFP reporter at the AAVS1 locus similar to previously described synthetic biological tools⁵¹ were kept at 37°C in a humidified atmosphere at 5% (v/v) CO₂. The reporter contains 4 binding sites for the synthetic TF directly upstream of a minimal heat shock (HSP68) promoter that drives transcription of a fast-degrading PEST tagged mNeongreen fluorophore. The construct also had an SV40 poly A tail. Cells were cultured in DMEM containing 4.5 g/L glucose and GlutaMAX supplement (Thermo 10566016), supplemented with 10% (v/v) Fetal Bovine Serum, 1% (v/v) non-essential amino acids, and 1% (v/v) Penicillin-streptomycin (Thermo, Waltham, MA, USA). Cells were passaged every 3 days at 1:10, using Accutase (StemPro; Thermo, Waltham, MA, USA) to disassociate. This cell line tested negative for mycoplasma as assessed through imaging by the UC Berkeley Cell Culture Facility. For investigating reporter activation by constructs expressed from plasmids, reporter cells were transfected with Mirus-293 lipid-based transfection reagent (Mirus). Cells were seeded onto a 24 well plate and transfected when 80% confluency was reached. Reporter cells transfected with PPAR gamma constructs were treated with 10 μM B-estradiol 24 h post transfection to induce reporter expression. Cells were assayed for transcriptional activity 24 h post induction by flow cytometry.

Design and testing of synthetic transcription factor constructs

The PPARγ^AB was ordered as a geneblock (IDT). Synthetic transcription factor chimeras were cloned by Gibson assembly with the AB domain fragment of interest and a synthetic transcription factor in the piggybac transposase backbone. This backbone has an EF1a promoter, HaloTag, Estrogen binding domain, a synthetic zinc finger DNA binding domain, and T2A mScarlet. This co-translationally cleaved mScarlet allows us to quantify transfection efficiency. The tested AB domain fragments are inserted on the N-terminus, before the HaloTag. All constructs were verified by whole-plasmid oxford nanopore sequencing through the UC Berkeley sequencing core.

Cells were disassociated using Accutase and resuspended into single cells with complete media as described above. Flow cytometry was performed on a Beckman Coulter CytoFlex S containing 3 lasers (488 nm, 561 nm, and 638 nm) and 9 detectors. GFP was detected with the 488 nm laser and mScarlet with the 561 nm laser. Three biological replicates were each performed on different days. The machine was calibrated each day, and the laser detector voltages were kept the same. Data were analyzed using CytExpert software utilizing a consistent gating strategy to compute activity. Briefly, we first gated for live cells by plotting Forward Scatter Area against Side Scatter Area (Supplementary Fig. 23). Then we took live cells and gated Forward Scatter Height against Forward Scatter Area to isolate single cells. Within single cells, we gated on mScarlet positive cells to isolate cells that were successfully transfected with the fragment of interest. Next, within the Scarlet positive population, we gated GFP positive cells to quantify the fraction of cells that activated the reporter. We computed activity by dividing the number of cells that were GFP positive and mScarlet positive by those that were only mScarlet positive. This method ensured that activity was solely determined by reporter expression in successfully transfected cells, or that activity would not be skewed by the variable transfection efficiency of certain constructs. This method of quantification yielded reproducible activity measurements across the three biological replicates for each construct where transfections occurred on different days.

Mapping degrons

For mapping degrons, the human (female) HEK293T (ATCC, CRL-3216) TetBxb1BFPiCasp9 Clone 12 cell line⁹⁹ was cultured in DMEM (Sigma-Aldrich) containing 10% (v/v) fetal bovine serum (FBS) (Sigma-Aldrich, F7524), supplemented with 0.24 mg/mL streptomycin sulfate (BioChemica, Espoo, Finland), 0.29 mg/mL penicillin G potassium salt (BioChemica, Espoo, Finland), 0.32 mg/mL glutamine (Sigma-Aldrich) and 2 µg/mL doxycycline (Sigma-Aldrich, Saint-Louis, MA, USA). Cells were passaged at ~80% confluency using 0.25% trypsin (Gibco) and tested negative for mycoplasma (Mycostrip, InvivoGen, San Diego, CA, USA). Cell line authenticity was confirmed by selection of recombinants with 10 nM of AP1903 (MedChemExpress, HY-16046) (see below).

Tiles encoding 30 amino acid residues of PPARγ (sequences provided in Source Data file 1), overlapping by 15 residues, each followed by a stop codon, were inserted into the attB-EGFP-PTEN-IRES-mCherry-562bgl (p2127) vector¹⁰⁰ fused to the C-terminus of GFP (Genscript). Each of the constructs was individually transfected along with pCAG-NLS-Bxb1 (Addgene, plasmid no. 51271) into the HEK293T TetBxb1BFPiCasp9 Clone 12 cells using Fugene HD (Promega, Madison, WI, USA) and OptiMEM (Thermo Fisher Scientific, Waltham, MA, USA). After 2 days, 2 μg/mL doxycycline and 10 nM AP1903 were added to induce expression and select for recombinants. After another 4 days of culturing, the cells were washed with PBS and resuspended in PBS with 2% (v/v) FBS. Flow cytometry was performed using a BD FACSJazz cell sorter (BD Biosciences). The cells were gated for live cells, singlets, BFP-negative and mCherry-positive as described in the supplementary material (Supplementary Fig. 17), and the GFP/mCherry ratios were determined using a minimum of 10,000 cells. The lasers were: 405 nm for BFP, 488 nm for GFP and 561 nm for mCherry. The filters were: 450/50 for BFP, 530/40 for GFP and 610/20 for mCherry. The experiment was performed twice (biological replicates, separate transfections) with similar results. FlowJo (v.10) (BD Biosciences, San Jose, CA, USA) was used to analyze the data.

Martini simulations of full-length PPARγ

We ran coarse-grained MD simulations of full-length PPARγ with the Martini 3 force field¹⁰¹, with protein-water interactions rescaled by a factor of 1.06¹⁰², using Gromacs (2021.1)¹⁰³. As the starting structure, we used chain B in PDB 3E00¹⁸ [https://doi.org/10.2210/pdb3E00/pdb] with missing residues added with Modeler¹⁰⁴. We generated the coarse-grained structure and parameterization with Martinize2 and used DSSP to assign secondary structure-specific potentials; the AB domain (residue 1–134) was manually assigned to coil. Within the DBD (residue 138–209) and LBD (residue 234–505, except loop region residues 290–304) we added an elastic network model with a force constant of 700 kJ mol^-1 nm^-2 between all backbone beads inside a 1.2 nm cut-off. The structure was placed in a dodecahedral box with the box size set to have a minimum distance of 0.1 nm between the protein structure and the box edge. We solvated the system and added 150 mM NaCl with additional Na⁺ or Cl^- ions to neutralize the system using Insane.

Simulations were performed using the leap-frog integrator. We used a Verlet list cut-off scheme for non-bonded interactions, a cut-off of 1.1 nm for van der Waals interactions, and a dielectric constant of 15 and a cut-off of 1.1 nm for Coulomb interactions. We used the velocity rescaling thermostat with a temperature of 298 K and a 1 ps time coupling constant and the Parrinello-Rahman barostat with a pressure of 1 bar, a 12 ps time coupling constant, and an isothermal compressibility of 3×10⁻⁴bar⁻¹. For equilibration simulations, we instead used the Berendsen thermostat with a temperature of 298 K and a 2 ps time coupling constant and the Berendsen barostat with a pressure of 1 bar, a 12 ps time coupling constant, and an isothermal compressibility of 3 × 10⁻⁴bar⁻¹.

We performed steepest descent energy minimization with a step size of 0.01 nm for 10,000 steps. We then generated velocities sampled from the Maxwell-Boltzmann distribution at 298 K and ran a 50 ns equilibration simulation with a 10 fs time step. We ran two replicate production simulations of 40 μs each with a 20 fs time step and saved protein coordinates every ns. We backmapped our trajectories to all-atom using a shortened version¹⁰⁵ of Backward¹⁰⁶. We added two coordinated Zn²⁺ ions to the DBD in each frame by superposing to the DBD in PDB 3E00 (residue 139-199) and copying the Zn²⁺ coordinates.

CALVADOS simulations of AB domains from PPARα and PPARγ

We ran coarse-grained simulations of the AB domains of PPARα (residue 1–100) and PPARγ (residue 1–137) with the CALVADOS 2 force field⁴⁰ using HOOMD-blue 2.9.3¹⁰⁷. The simulations were performed using a Langevin integrator at 298 K with an ionic strength of 150 mM and the partial charge of His side chains set corresponding to pH 7.2. We used a 2 nm cut-off for Ashbaugh-Hatch potentials and a 4 nm cut-off for Yukawa potentials. Simulations were started from an Archimedean spiral arrangement of the protein chain. We ran equilibrations simulations for 10,000 steps with a 5 fs time-step and production simulations for 350 ns with a 10 fs time step. We saved 5010 evenly spaced frames and back-mapped to all-atom using Pulchra 3.06¹⁰⁸.

Analysing MD simulations

For SAXS calculations from MD, we calculated SAXS intensities from all-atom back-mapped trajectories using Pepsi-SAXS¹⁰⁹ with fixed values for the contrast of the hydration layer (δρ = 3.34 e/nm³) and the displaced solvent (r₀/r_m = 1.025)¹¹⁰.

For analysis of trajectory, we calculated contacts from the Martini trajectories of full-length PPARγ based on backbone-bead distances within a 1.1 nm cut-off using compute_contacts in MDTraj (1.9.9). For contacts calculations, regions were defined as residue 1–127 (AB domain), 138–205 (DBD), 235–505 (LBD), and 138–505 (DBD-hinge-LBD). We calculated Rg from coarse-grained trajectories using gyrate in Gromacs (2019.4). The PPARγ AB domain was defined as residue 1–137 for Rg calculations.

For Bayesian Maximum Entropy reweighting for AB domains, we reweighted the ensembles of the AB domains of PPARα and PPARγ against the SAXS data using BME reweighting⁴¹. We iterated between reweighting and fitting the scale and offset of the simulated SAXS intensities using linear least-squares regression for 30 iterations or until the χ²_r improved by <0.001. θ is a scaling parameter in BME reweighting that is set to balance the fit to the experimental data with the deviation from the prior ensemble. We scanned 20 values of θ evenly spaced on a log scale from 0.1 to 10,000 for PPARα and 0.01 to 10,000 for PPARγ and selected θ = 78.48 for PPARα and θ = 14.38 for PPARγ (Supplementary Fig. 2).

For Bayesian Maximum Entropy reweighting for full-length PPARγ, we reweighted our simulated ensemble of full-length PPARγ using a two-step BME reweighting approach: (1) We removed all atoms outside residue 133–505 (corresponding to the ΔAB construct) from our ensemble and calculated SAXS intensities. We reweighted this ensemble against the SAXS data on ΔAB using BME as described above and selected θ = 545.56 (Supplementary Fig. 9). (2) We used the weights from step 1 as a prior for reweighting the full ensemble against SAXS data on full-length PPARγ and differences in ¹⁵N R2 (ΔR2) (Supplementary Fig. 10). ΔR2 is defined as:

$$\mathrm{\Delta }{R}_2=R_2^{\mathrm{FL}}-R_2^{\mathrm{AB}}$$ 4

where $R_2^{\mathrm{FL}}$ and $R_2^{\mathrm{AB}}$ are the 15 N R2 relaxation rates in the full-length and the AB domain construct respectively. For reweighting against $\mathrm{\Delta }{R}_2$, we assumed that ΔR2 for a reside in the AB domain is correlated with the ensemble-averaged number of contacts, $⟨n_c⟩$, to the DBD-hinge-LBD region^111–113:

$$\mathrm{\Delta }{R}_2=a⟨n_c⟩+b$$ 5

where a and b are an unknown scale and offset relating $⟨n_c⟩$ to $\mathrm{\Delta }{R}_2$ and were treated as fitting parameters. To avoid effects on $\mathrm{\Delta }{R}_2$ from proximity to the DBD-hinge-LDB region, we only reweighted against $\mathrm{\Delta }{R}_2$ for residue 1–109. As error for $\mathrm{\Delta }{R}_2$ we combined the experimental error, ${\sigma }_{\mathrm{\Delta }{R}_2}$, (from error propagation from $R_2^{\mathrm{FL}}$ and $R_2^{\mathrm{AB}}$) and an additional relative error, ${\eta }_{\mathrm{FM}}$, resulting from the approximation in Eq. 2 to obtain the final error, ${\sigma }_{\mathrm{BME}}$, used for reweighting:

$${\sigma }_{\mathrm{BME}}=\sqrt{{\sigma }_{\mathrm{\Delta }{R}_2}^2+{\left({\eta }_{\mathrm{FM}}\mathrm{\Delta }{R}_2\right)}^2}$$ 6

We scanned values of ${\eta }_{\mathrm{FM}}$ from 0 to 0.5 in steps of 0.05 and 10 values of θ evenly spaced on a log scale from 10 to 10,000. We iterated between reweighting and fitting the scale and offset of the simulated SAXS intensities and $\mathrm{\Delta }{R}_2$ using linear least-squares regression for 15 iterations. φ_eff is an estimate of the fraction of effective frames retained after reweighting. As the profiles of φ_eff versus χ²_r had a similar shape for different values of ${\eta }_{\mathrm{FM}}$, we first selected θ = 1000 and then, keeping θ fixed, selected ${\eta }_{\mathrm{FM}}$ = 0.3 as a value that balances the fit to the $\mathrm{\Delta }{R}_2$ (χ²_r = 0.98) and SAXS (χ²_r = 1.44) data to avoid overfitting to $\mathrm{\Delta }{R}_2$.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

E.G.K.T., M.R.L., and B.B.K. conceived the study, with input on function from S.M. L.W.B. performed initial NMR assignment of the chemical shifts of the AB domains. E.G.K.T. and M.R.L. purified all PPAR variants, designed and performed all further NMR experiments, and reassigned the NMR chemical shifts with assistance from A.P. and analyzed all NMR data with input from B.B.K. M.R.L., E.S.A., and S.K. purified all RXRα variants and E.S.A. isolated the heterocomplex. D.S. purified and recorded data on the isolated LBD. E.G.K.T., M.R.L., J.G.O., and C.M.J. performed and analyzed the SAXS experiments. E.G.K.T., M.R.L, E.S.A., and C.G.C. conducted and analyzed FP assays, and M.R.L. performed the ITC experiments. C.G.C. conducted experiments and analyzed data on the chimera. M.R.L., E.S.A., and S.K. performed and analyzed the NMR data on the heterocomplex. F.E.T. and K.L.L. designed, performed, and analyzed the molecular simulations and integrative modeling of full-length PPARγ, PPARγ^AB, and PPARɑ^AB with input from E.G.K.T. and B.B.K. A.U. and M.V.S. performed transcriptional activity assays with assistance from E.G.K.T. J.M. generated the plasmid and cell line for activity screens. E.H.O. and R.H.P. performed and analyzed the degron screen. E.G.K.T., M.R.L., F.E.T., and A.U. made the figures. E.G.K.T., M.R.L., and B.B.K. wrote the manuscript with input from all authors. B.B.K. was responsible for the overall project management and supervised the research.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was made possible by the Novo Nordisk Foundation Challenge grant REPIN – rethinking protein interactions (#NNF18OC0033926 to B.B.K. and R.H.P.), by HALOS Cross Border Research Project funding from the EU ÖKS Interreg program (to B.B.K., C.M.J., and E.K.T.), by the PRISM (Protein Interactions and Stability in Medicine and Genomics) center funded by the Novo Nordisk Foundation (#NNF18OC0033950 to K.L.-L.), a Novo Nordisk Foundation project grant (#NNF21OC0071057 to R.H.P.) and the European Union’s Horizon Europe research and innovation program under the Marie Skłodowska-Curie Doctoral Network IDPro (grant agreement No 101119633). We thank Villum Fonden for supporting the NMR infrastructure. NMR data were in part recorded at cOpenNMR funded by the Novo Nordisk Foundation (#NNF18OC0032996). We acknowledge access to computational resources from the ROBUST Resource for Biomolecular Simulations (supported by the Novo Nordisk Foundation grant no. NNF18OC0032608 to K.L.L.), Computerome 2.0, and the Biocomputing Core Facility at the Department of Biology, University of Copenhagen. SAXS measurements were done at EMBL P12-bioSAXS beam line at PETRAIII (DESY, Hamburg, Germany DESY⁹⁴). M.V.S. is a Biohub, San Francisco, Investigator.

Data availability

Chemical shifts of PPARγ^AB and PPARɑ^AB have been deposited in the BioMagResBank under accession codes 53393 and 52294, respectively. SEC-SAXS and SEC-MALLS data are made available in the Small Angle Scattering Biological Data Bank (SASBDB¹¹⁴) under the accession codes: PPARγ SASDWX3; ΔAB-PPARγ SASDWY3; PPARγ-AB SASDWZ3 and PPARɑ-AB SASDW34. Simulation data are available at Github https://github.com/KULL-Centre/_2025_thomsen_ppar and on Zenodo at https://doi.org/10.5281/zenodo.21517744. Accession codes of previously published structural data used in this study are PDB 3E00 and BMRB entry 17975. Data for the degron screen and data from ITC, FP, NMR relaxation and CSPs are all available in Source data file. Source Data are available as a Source Data file. Source data are provided with this paper.

Code availability

Code and scripts used for this paper are available via https://github.com/KULL-Centre/_2025_thomsen_ppar and a permanent version of this repository is available via https://doi.org/10.5281/zenodo.21517744.

Competing interests

One author, K.L.L., holds stock options in and is a consultant for Peptone Ltd. All other authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76962-x.