The Solution Ensemble of the C-Terminal Domain from the Transcription Factor Pdx1 Resembles an Excluded Volume Polymer

Abstract

The pancreatic and duodenal homeobox 1 (Pdx1) is an essential pancreatic transcription factor. The C-terminal intrinsically disordered domain of Pdx1 (Pdx1-C) has a heavily biased amino acid composition; most notably, 18 of 83 residues are proline, including a hexa-proline cluster near the middle of the chain. For these reasons, Pdx1-C is an attractive target for structure characterization, given the availability of suitable methods. To determine the solution ensembles of disordered proteins, we have developed a suite of ¹³C direct-detect NMR experiments that provide high spectral quality, even in the presence of strong proline enrichment. Here, we have extended our suite of NMR experiments to include four new pulse programs designed to record backbone residual dipolar couplings (RDCs) in a ¹³C,¹⁵N-CON detection format. Using our NMR strategy, in combination with small angle x-ray scattering (SAXS) measurements and Monte Carlo simulations, we have determined that Pdx1-C is extended in solution, with a radius of gyration and internal scaling similar to that of an excluded volume polymer, and a subtle tendency toward a collapsed structure to the N-terminal side of the hexa-proline sequence. This structure leaves Pdx1-C exposed for interaction with trans-regulatory co-factors that contribute with Pdx1 to transcription control in the cell.

Graphical Abstract

INTRODUCTION

Intrinsically disordered proteins (IDPs) have fascinated both the biological and physical chemistry communities due to their ability to mediate interactions that are critical for signal transduction in the cell and because of the heterogeneous conformational ensembles that ideally suit them to this function. IDPs do show some of the same structural characteristics as their folded counterparts, such as the propensity for secondary structure,¹ globular regions,² and long-range contacts;³ their defining feature is not necessarily the absence of such structures, but rather their transient nature in the overall solution ensemble. That said, not all IDPs display significant secondary structure elements or compaction. For example, polyelectrolytic regions and high total charge polyampholytes tend to exist as extended self-avoiding random flight chains.^{2, 4} Substantial effort has been invested in classifying the ensemble characteristics of IDPs,⁵ with common models from the polymer physics community having emerged as especially helpful tools for this effort. For example, detailed polymer description of size and shape in the ensemble of the disordered protein NUS helped to reconcile apparent discrepancies between SAXS and FRET data that have concerned the literature for several years.⁶

In addition to the powerful data available from SAXS and FRET, advances in NMR spectroscopy now make it possible to collect extensive, site resolved data that describe the solution behavior of IDPs.⁷ Having recognized this, the community has begun to call for development of advanced NMR techniques that provide constraints on statistical IDP ensembles.⁸ In our laboratory, we seek to fulfill this need through the development of ¹³C direct-detect NMR techniques that dramatically improve the completeness of NMR data sets for disordered systems.⁹ These efforts have allowed us to transform our early and qualitative identification of α-helical structure in the C-terminus of the phosphatase FCP1¹⁰ into a quantitative model for its solution ensemble.¹¹ In addition, ¹³C direct-detect techniques have allowed us to quantitatively demonstrate sequence-driven enrichment in the cis-proline content of phosphorylated RNA polymerase II C-terminal domain and connect this conformational switch to readout by the downstream enzyme Ssu72.¹²

Motivated by our work with RNA polymerase II, we have turned our attention broadly to quantitative description of proline conformational states and the impact proline-enrichment has on the conformational ensembles of IDPs. Proline has long confounded NMR spectroscopists, owing to its lack of an amide hydrogen when found in polypeptide chains, and traditional reliance on the amide ¹H-nucleus for NMR detection. Detection on the ¹³C-carbonyl nucleus removes this barrier and makes comprehensive NMR analysis of proline-rich IDPs practical. Proline-rich regions are both common in biology, where they often mediate protein-protein interactions, and of interest from a physical-chemical perspective. As proline residues lack an amide hydrogen, they possess no hydrogen bond donor with which to form stable secondary structures. Coupled to this, their reduced backbone torsional flexibility suits proline residues to serve as “extended spacers” that force disorder locally and enhance solvent exposure.¹³ Thus, proline paradoxically promotes disorder while simultaneously providing local stiffness that reduces the conformational entropy of the polypeptide chain – an important yet underappreciated concept that sculpts the ensembles of IDPs.

We have previously utilized the C-terminal disordered region of the transcription factor Pdx1, which we refer to as Pdx1-C throughout, to develop an efficient strategy for chemical shift assignment of IDPs.¹⁴ Recombinant human Pdx1-C is an 83 residue segment, containing 18 proline residues (22% of the sequence), with the most extreme cluster being a continuous 6-proline sequence located near the center of the construct. Pdx1 plays a central role in maintenance of the pancreatic β-cell phenotype,¹⁵ and is recognized for its significance in both diabetes¹⁶ and pancreatic cancer.¹⁷ Of further relevance for the current work, Pdx1-C is known to harbor a site of interaction with the E3-ubiquitin ligase substrate adaptor protein SPOP, which binds under low glucose conditions and promotes the proteolytic degradation of Pdx1.¹⁸ Thus, a better understanding of Pdx1-C structure and its relationship to function would be clinically significant.

Here we describe the solution ensemble of Pdx1-C, which we find to be highly extended and devoid of secondary structure outside the polyproline-II helix formed by the hexa-proline cluster. Our work relies on a combination of SAXS data with measurements of backbone ¹⁵N spin relaxation, residual dipolar couplings (RDCs), and NMR paramagnetic relaxation enhancement (PRE). Significantly, we find that the extent of non-local contacts reported through ¹³C direct-detect PRE depends on whether the excitation nucleus is also ¹³C (“protonless” ¹³C,¹⁵N-CON detection¹⁹), or ¹H (¹³C,¹⁵N-(HACA)-CON detection²⁰). The experimental PRE profiles documented here are consistent with simulated PRE profiles derived from Monte Carlo simulations using the Absinth force field,²¹ as implemented in the CAMPARI software package,²² and indicate a high similarity to a self-avoiding polymer chain. These findings illustrate how Pdx1-C is able to maintain an accessible conformation in solution, required for interaction with SPOP and potentially other co-regulators of gene expression in the β-cell.

MATERIALS AND METHODS

Construct generation.

The DNA insert coding for the C-terminal residues of Pdx1 (Pdx1-C; residues 205–283 of the human sequence) were synthesized by Geneart (Invitrogen) and subcloned into pET-49b. Expression of recombinant Pdx1-C from the resulting plasmid yields an N-terminal glutathione-S-transferase (GST) and a 6x histidine tag, separated from Pdx1-C by a 3C Protease cleavage site. The mutant Pdx1-C C227S, S273C, required for PRE studies, was generated using site-directed mutagenesis, per manufacturer protocols for the Q5 site-directed mutagenesis kit (New England Biolabs). Plasmids encoding wild-type or mutant Pdx1-C were transformed into BL-21 (DE3) Escherichia coli for protein expression.

Protein expression and purification.

To make protein with naturally abundant isotopes, as needed for SAXS measurements, BL-21 (DE3) E. coli were grown in Luria Broth (LB). For all NMR experiments, the E. coli were grown in M9 minimal media supplemented with ¹⁵N-ammonium chloride and U-13C glucose. In either case, E. coli were incubated at 37 °C under agitation until an OD₆₀₀ of 0.6–0.8 was reached. Pdx1-C expression was induced by adding IPTG to a final concentration of 0.5 mM, after which cells were incubated for 4 hours at 37 °C under agitation. Cells were harvested by centrifugation and lysed by sonication. Cell debris was cleared through centrifugation and cell lysate applied to a Ni-NTA column pre-equilibrated with 50 mM Tris (ph 7.5), 500 mM sodium chloride, and 20 mM imidazole. The Ni-NTA column was then washed with 10 column volumes of 50 mM Tris (pH 7.5), 500 mM sodium chloride, 20 mM imidazole, and 0.1 % Triton X-100. Proteins were eluted from the Ni-NTA column with 50 mM Tris (ph 7.5), 500 mM sodium chloride, and 200 mM imidazole. The affinity tags on Pdx1-C were cleaved with 3C-Protease and buffer exchanged into 50 mM Tris (pH 7.5), 500 mM sodium chloride. Subsequently, Pdx1-C was separated from the cleaved affinity tags by passing the protein solution over a Ni-NTA column pre-equilibrated with the same buffer. An additional size exclusion chromatography purification was achieved through passage over a Sephacryl-200 column, which efficiently removed high molecular weight contaminants and any cleaved affinity tags that may have co-eluted from the Ni-NTA column with Pdx1-C.

Small angle x-ray scattering (SAXS).

Pdx1-C samples for SAXS were buffer exchanged into 50 mM HEPES (pH 7.0), 100 mM NaCl, and 5% glycerol (to protect against radiation damage). Samples were concentrated to > 5 mg mL⁻¹ as determined using a Direct Detect FTIR spectrometer (EMD Millipore). SAXS data were collected at the Cornell High Energy Synchrotron Source (CHESS) on the G1 beamline. Incident radiation was estimated to be 9.9 keV, producing a flux of 8 × 10¹¹ photons s⁻¹, providing a q-space range of 0.01– 0.7 Å⁻¹. Scattering from a silver behenate standard was used for q-axis mapping. Data collection was performed using dual Pilatus 100K-S detectors. Reduction of the 2D images to 1D scattering profiles was performed using BioXTAS Raw.²³ For each experimental run, 50 μL of Pdx1-C solution was injected onto a Superdex 200 Increase (5×150 mm) column at a flow rate of 0.2 mL min⁻¹ for in-line SEC-SAXS data acquisition. Final reported SAXS data sets represent the average of three SEC-SAXS runs, each of which produced a total of >100 exposures stored as 20 s frames. Solvent subtraction was performed using equivalent numbers of frames of the Pdx1-C elution peak and surrounding baseline data. No signs of aggregation, inter-particle effects, or radiation damage were observed. Guinier fitting was performed using the method of non-linear least squares in BioXTAS RAW.²³ Data fitting of the Guinier region was restricted to the data points satisfying qR_g < 0.8 as has been recommended for highly flexible proteins.²⁴

General NMR spectroscopy.

Chemical shift assignments of Pdx1-C have been published previously and are deposited in the BMRB (access code 19596).¹⁴ Additional chemical shift assignments of MTSL-tagged Pdx1-C, used for the PRE experiments, were determined using (HACA)N(CA)CON, HNCO, and HN(CA)CO triple resonance experiments (see representative 2D spectra in Figure S1). All NMR data were collected on a Bruker Avance III 500 MHz spectrometer, equipped with a TCI Cryoprobe. NMR spectra were processed using Bruker TopSpin (Billerica, MA), followed by analysis in NMRFAM SPARKY.²⁵

NMR spin relaxation.

¹⁵N R₁ and R₂ spin relaxation parameters were determined using ¹H_N-start CON experiments previously developed in our laboratory.²⁶ The R₁ and R₂ spin relaxation experiments were collected as pseudo-3D spectra with a data matrix of 1024 × 512 points for each of the 8 interleaved acquisitions. Spectra were collected with 64-scans of signal averaging per FID. The R₁ data were collected as a randomized series of ¹³C,¹⁵N-CONs with relaxation delays set to 50, 150, 250, 350, 450, 550, 650, and 750 ms. For the R₂ data set, the relaxation delays were 16, 48, 80, 112, 144, 176, 208, 240, 272, and 304 ms. The sweep widths for the relaxation experiments were set to 19 ppm for the direct ¹³C-dimension, centered on 173 ppm, and 28 ppm for the indirect ¹⁵N-dimension, centered on 120 ppm.

Residual dipolar coupling (RDC) measurements.

NMR samples were prepared by dissolving Pdx1-C at a concentration of 1.1 mM in dilute buffer (50 mM sodium cacodylate, pH 6.5, 50 mM potassium chloride, 5 mM TCEP, 0.01% sodium azide, and 10% (v/v) D₂O) prior to transfer into a Shigemi tube. The aligned sample was generated by swelling a neutral 7% polyacrylamide gel in the tube, with the plunger height set to achieve a compression ratio of 1.6:1.²⁷ All spectra were collected at 298K. Spectra were acquired as in phase/antiphase pairs for improved spectral presentation of the data, using in-house pulse sequences for measurement of ¹J_HN (Figure S2), ¹J_CαHα (Figure S3), ¹J _NĆ (Figure S4), and ¹J_CαĆ (Figure S5); see Supporting Information for further description of how these experiments can be implemented. For spectral clarity, ¹J_CαHα and ¹J_CαĆ were rescaled by a factor of 0.5 using the accordion principle.²⁸ All spectra for RDC measurements were collected with 64-scans of signal averaging and a 1024 × 512 data matrix in both the in-phase and antiphase spectra. The sweep width for the direct ¹³C-dimension was set to 14 ppm, centered at 172 ppm, while the indirect ¹⁵N-dimension had a 24.2 ppm sweep width centered at 118 ppm.

All pulse programs developed for RDC measurements, as well as those we have developed and reported previously, are available in Bruker Topspin-compatible format for open-access download through the Penn State ScholarSphere (https://scholarsphere.psu.edu/collections/5712mv15x), or by searching for the Corresponding Author’s surname on the ScholarSphere homepage. All ¹³C direct-detect pulse programs utilized in this study, but previously published, are also available in the same collection.

Paramagnetic relaxation enhancement (PRE) measurements.

Following purification, Pdx1-C was incubated in 50 mM Tris (pH 7.5), 150 mM NaCl, and 5 mM DTT for a minimum of 1 hour at 4 °C in the dark. Pdx1-C samples were then buffer exchanged into 50 mM Tris (pH 7.5) and 150 mM NaCl using PD-10 buffer exchange resin (GE Healthcare Life Sciences of Pittsburgh, PA). A 5x molar excess of MTSL, relative to the Pdx1-C molarity, was then added and the sample incubated overnight at 4 °C. Unconjugated MTSL was separated from conjugated Pdx1-C using PD-10 resin. MTSL-labeled proteins were then buffer exchanged into 50 mM sodium phosphate (pH 6.5), 50 mM potassium chloride, 0.01% sodium azide, and 10% (v/v) D₂O for spectroscopy. ¹³C,¹⁵N-CON and ¹³C,¹⁵N-(HACA)CON spectra were acquired on this paramagnetic sample as described below. Following data acquisition, the MTSL radical was quenched using a 10x molar excess of sodium ascorbate. Identical NMR spectra were acquired under diamagnetic conditions. All spectra for the PRE measurements were collected with 32-scans of signal averaging and a 1024 × 512 data matrix. The sweep width for the direct ¹³C-dimension was set to 18 ppm, centered at 173 ppm, while the indirect ¹⁵N-dimension had a 40 ppm sweep width centered at 124 ppm.

CAMPARI simulations.

Monte Carlo simulations were performed using the CAMPARI software package.²² All simulations included Pdx1-C in a random starting conformation, represented using the OPLSS-AA/L force field, and embedded in a spherical droplet with a 250 Å radius. Sodium chloride pairs were modeled explicitly to give a final concentration of ~10 mM, with excess sodium to preserve neutrality. The ABSINTH implicit solvation model was used for all trajectories.²¹ Replica exchange trajectories were run with twenty replicas at temperatures evenly spanning the range from 230 – 420 K for 5.1 × 10⁷ steps, with the first 1 × 10⁶ steps discarded as equilibration. For the unbiased simulation, the move set was composed as follows: rigid body moves, 10 % (50 %, 2 Å, 20°); side chain moves, 20% (40 %, 30°); omega moves, 10 % (10 %, 4°); pucker moves, 10 %; backbone moves, 56.7% (30 %, 10°). Percentages in parentheses correspond to the fraction of fully randomizing moves, while distances and angles represent the maximum displacement and rotation, respectively. Two additional control simulations were run in which the proline ω-bonds were constrained to be either all-cis or all-trans, with up to a 5° rotation allowed per move. For the ω-constrained simulation, the move set was composed as follows: rigid body moves, 10 % (50 %, 2 Å, 20°); side chain moves, 20% (40 %, 30°); omega moves, 10 % (0 % randomizing moves); pucker moves, 10.0 %; backbone moves, 56.7% (30 %, 10°).

Simulation analysis.

All molecular graphics images were generated using UCSF Chimera.²⁹ All other analysis was performed using custom Python scripts, with graphs generated in the Matplotlib (https://matplotlib.org) prior to export into Adobe Illustrator. Calculation of R_g from the simulated ensembles was performed using CRYSOL, which is a program used to evaluate solution scattering from known structures by fitting to experimental SAXS data³⁰ as previously described.¹¹

Simulated PRE data were generated using the Battiste-Wagner equation to estimate reduction in peak intensity through previously described methods,^{3, 31} which will be summarized here as this approach remains uncommon. Here, we have collected PRE using two pulse programs that differ in the gyromagnetic ratio of the excitation nucleus and so we have generated theoretical PRE curves separately for the ¹³C-start and ¹H-start CON and (HACA)CON, respectively. Simulated curves were created to model MTSL attachment at either residue 227 or 273, producing a total of four simulated curves. All calculations were performed using in-house Python (v3.6, NumPv v1.15) scripts. The predicted ratio of ¹⁵N,¹³C cross-peak intensities under paramagnetic (oxidized; I_ox) and diamagnetic (reduced; I_red) conditions is determined as:

$$\frac{I_{ox}}{I_{red}}=\frac{R2\text{exp}\left(-R{2}^{sp}t\right)}{R2+R{2}^{sp}}$$ Equation 1

where R2^sp is the paramagnetic relaxation enhancement due to the spin probe in the paramagnetic form, and R2 is the intrinsic transverse relaxation rate of the protein in the absence of the spin probe, and τ is the total INEPT transfer time in the preparation element of the pulse sequence. To estimate R2, we used the average full-width at half-height for all analyzed peaks in the spectrum. For each site, the distance separating the excitation ¹H^α or ¹³Ć and the unpaired electron is calculated as:

$$r={\left[\frac{K}{R{2}^{sp}}\left(4{\tau }_c+\frac{3{\tau }_c}{1+{{\omega }_{1Hor13C}}^2{{\tau }_c}^2}\right)\right]}^{\frac{1}{6}}$$ Equation 2

where ω is the Larmor frequency of the excitation nuclear spin and τ_c is the rotational correlation time, estimated from the molecular weight of Pdx1-C using previously described methods that account for the internal flexibility of intrinsically disordered proteins.³² The parameter K is composed of fundamental constants:

$$K=\frac{1}{15}S\left(S+1\right){\gamma }^2{g}^2{\beta }^2$$ Equation 3

where γ is the gyromagnetic ratio of the nuclear spin, g the electronic g factor, and β the Bohr magneton. The parameters and constants used in our calculation were set to R_2,H = 13 s⁻¹, R_2,C = 8.1 s⁻¹, τ = 9.0 × 10⁻³ s, τ_c = 1.2 × 10⁻⁹ s, ω_1H = 2π ∙ 500 × 10⁶ s⁻¹, ω_1H = 2π ∙ 126 × 10⁶ s⁻¹, K_1H = 1.23 × 10⁻³² cm⁶ s⁻², and K_13C = 7.778 × 10⁻³⁴ cm⁶ s⁻².

The sulfur atom in the side chain of C227 or C273 was used as an estimate for the location of the unpaired electron in MTSL. Distances between the sulfur atom and each Ć or H_α were used to estimate the reduction in the peak intensity using previously described methods.^{3, 31} The scaling profiles were calculated as the average distance between two residues in 3-dimensional space over all simulation frames, <R_ij>, as a function of the sequence separation in the primary structure, |i-j|, using Equation 1:

$$\langle R_{ij}\rangle =\langle \frac{1}{Z_{ij}}{\sum }_{m\in i}{\sum }_{n\in j}|r_m^i-r_n^j\rangle$$ Equation 4

where i and j are residue indexes; m and n are atoms in residues i and j, respectively; Z_ij is equal to the number of unique pairwise distances; and r indicates the position vector for an atom in a particular residue.

RESULTS AND DISCUSSION

SAXS of Pdx1-C.

Intrinsically disordered regions by definition adopt multiple conformations in space and/or time, necessitating characterization of both their average ensemble characteristics and the details of individual high-probability structural features. To begin this study of Pdx1-C, we performed SAXS measurements with the aim of determining what features are present in its solution ensemble at the nm length scale. In order to control for aggregation of other inter-particle interactions in the beamline, we made measurements at three starting concentrations (3.8, 7.6, and 9.8 mg/mL) prior to loading on a Superdex 200 Increase (5×150 mm) column from which the eluate solution flowed directly into the x-ray beamline for SAXS analysis. Data were averaged over the set of frames spanning the SEC peak determined to have approximately the same radius of gyration (R_g). For each run, buffer subtraction was achieved by utilizing a matched number of frames from the SEC chromatogram baseline. The raw scattering data are consistent with preservation of the same assembly state in solution across the full concentration range (Figure 1A) and the low-q region in particular suggests Pdx1-C is not aggregating under our conditions. Analysis of the data in the Guinier region of the scattering data (Figure 1B) is consistent with a solution R_g of 27.4 ± 0.8 Å for the 3.8 mg/mL data set, which will be the source of all discussed statistics and fitting parameters going forward (see Table S1 for a full summary). The fitted R_g value is not consistent with the expectation for a cooperatively folded 83-residue monomeric protein, but is reasonable for a monomeric yet disordered chain (and with the molecular weight determined by in-line MALS following exit from the SEC column; data not shown). Similarly, Kratky analysis suggests that Pdx1-C lacks a closed volume in solution, as evidenced by the divergence of the dimensionless Kratky data at high-q (Figure 1C). In summary, the SAXS profile of Pdx1-C is consistent with the presence of an ensemble of unfolded, but monomeric, conformers in solution.

Figure 1.

SAXS characterization of Pdx1-C reveals an extended and disordered structure in solution. (A) Raw scattering data for Pdx1-C at concentrations ranging from 3.8 – 9.8 mg/mL. (B) Raw data in the Guinier region (red circles) and Guinier fits (black lines) for Pdx1-C. Data were fit to the Guinier approximation in Python using the method of non-linear least squares. (C) Kratky plots generated from the data in panel (A) show a linear increase as a function of q in the high-q region, which is suggestive of a disordered structure in solution. (D) Pairwise distance distribution plot, which demonstrates few features and suggests an extended, disordered structure for Pdx1-C in solution.

Fourier transformation of SAXS data reveal additional details of the conformational ensemble adopted by proteins in solution. Pairwise distance distribution analysis of Pdx1-C reveals a smooth distribution, best described by R_g = 26.2 Å (in excellent agreement with the Guinier analysis) and a maximum intramolecular distance of 95.7 Å. The pairwise distance distribution of Pdx1-C is otherwise surprisingly smooth and devoid of identifiable features (Figure 1D). For comparison, the experimental R_g compares well to the prediction for an excluded volume random chain (R_g,EV = 27 Å) and is only slightly extended compared with a re-calibration of the Flory scaling law intended to model intrinsically disordered protein behavior (R_g,IDP = 24.6 Å).³³ These findings are unsurprising, given the high proline enrichment of Pdx1-C and, nearly complete lack of aromatic or large-chain aliphatic residues, modest fraction of charged residues (FCR = 0.20), and very low net charge per-residue (NCPR = −0.012); all of which are expected to favor a disordered conformation in solution.

Absence of secondary structure indicated by chemical shift and RDC measurements.

The absence of stable tertiary structure and the adoption of a relatively extended average conformation are not necessarily predictive of the secondary structure present at individual motifs within a polypeptide chain. Our previous work has produced complete Pdx1-C chemical shift assignments that are consistent with the absence of significant α-helix or β-strand conformation at any position in Pdx1-C (reproduced in Figure 2A).¹⁴ While our chemical shift data suggest the absence of secondary structure, they do not rule out the possibility of other persisting and/or highly populated local structural features at variance with respect to random-coil structure, which might otherwise be well-described as enriched tertiary contacts.

Figure 2.

NMR characterization of Pdx1-C reveals an absence of detectable secondary structure. (A) Secondary chemical shift analysis from data deposited in the BMRB (accession number 19596). Residual dipolar couplings are displayed as a function of residue in the Pdx1-C sequence for (B) ¹D_NH, (C) ¹D_CαHα, (D) ¹D_NĆ, and (E) ¹D_CαĆ.

As an orthogonal measure of local order, we collected residual dipolar coupling (RDC) data for Pdx1-C in the presence of compressed polyacrylamide gels. RDCs are frequently relied on in contemporary investigations of IDP structure because they allow comparison of the relative orientations sampled by individual bond vectors in the polypeptide chain. An additional strength of RDC measurements in the context of protein disorder is their sensitivity to transient long-range interactions, such as those observed in the solution ensemble of α-synuclein.³ Consistent with our general strategy of reliance on ¹³C direct-detect spectroscopy, we have taken inspiration from the prior literature on scalar coupling measurement^{20, 34–36} and developed variants of the ¹³C,¹⁵N-CON for this study. In this way, we have generated pulse programs that are fully described in the Supporting Information, which enabled measurement of ¹J_NH, ¹J_CαHα, ¹J_NĆ, and ¹J_CαĆ (see Figures S6–S9 for representative spectra). For a random flight chain, the ¹D_NH RDC profile takes on a bell-shape, with the depth in the center determined by the overall length of the polypeptide.³⁷ The ¹D_NH RDCs for Pdx1-C shown in Figure 2B are consistent with an inverted bell-shaped profile. Similarly, the ¹D_CαHα, ¹D_NĆ, and ¹D_CαĆ feature a bell-like character that is also consistent with a coil configuration for Pdx1-C in solution. Taken together, our SAXS, chemical shift, and RDC data suggest that Pdx1-C adopts an extended and highly random ensemble in solution, with little to no evidence for globularity.

Backbone NMR spin relaxation of Pdx1-C.

Conformational rearrangements of proteins often occur on the ns-μs timescales;³⁸ unsurprisingly, the backbone dynamics of IDPs on these timescales are particularly extensive and have been heavily studied by NMR.^39–40 Therefore, in order to probe the backbone dynamics of Pdx1-C, we utilized our previously published ¹³C direct-detect, ¹⁵N spin relaxation experiments.²⁶ In the presence of uniform ¹³C-enrichment, passive scalar couplings can introduce sinewave distortions onto the exponential decays used to measure ¹⁵N spin-relaxation. As previously described,²⁶ our pulse sequences were designed to minimize the influence of passive coupling; the overall quality of the recorded decays is evident in representative decay curves that span the observed range of systematic distortions (Figure S10). T₁ and T₂ relaxation times are sensitive to both global reorientation and local fluctuations of the chain, as has been extensively modeled in studies of cooperatively folded proteins. For IDPs, where separation of global and local structure can lose its meaning, a variety of analysis techniques have been adopted.^41–44 In the case of Pdx1-C, the T₁ and T₂ profiles are largely featureless (Figure 3), yet the depressed transverse relaxation times, compared to a folded protein with an equivalent number of amino acids, are again consistent with overall disorder and a lack of collapsed structure. Interestingly, it does appear that the dynamics in the N-terminal half of Pdx1-C are non-uniform. Residues 210–225 appear to be more flexible than their neighbors, consistent with its inclusion of seven glycine residues. Similar variance in relaxation times in correlation with biased amino acid representation has been observed in previous studies of IDP dynamics.^45–46 In contrast, a reduction of T₁ and T₂ relative to baseline can suggest high contact density or the presence of secondary structure.⁴⁷ While there is a slight depression in the T₁ profile near residues 205–210, there is little other evidence for such contacts throughout the chain. In summary, the spin relaxation data suggest local structure near residues 205–225 that is distinct from the remainder of the Pdx1-C chain, but overall these data are consistent with the picture of an extended and highly dynamic backbone in solution.

Figure 3.

Backbone ¹⁵N spin relaxation for Pdx1-C. (A) T₁ and (B) T₂ relaxation profiles as a function of residue position indicate the highly disordered nature of the Pdx1-C backbone. The resonance intensity of each well-resolved residue in ¹³C,¹⁵N-CON spectra was recorded and the resulting decay curves fitted to a single-exponential function. Error bars represent uncertainty in fitting.

PRE data for Pdx1-C.

Paramagnetic relaxation enhancement (PRE) is a powerful measurement for constraint of IDP ensembles because of its ability to detect minor populations that feature long range contacts or locally collapsed states. For example, sensitivity to minor states enabled detection of pathological but low-population conformations in the solution ensemble of α-synuclein.⁴⁸ An additional feature that enhances the value of PRE as a structure constraint for expanded systems is that the high gyromagnetic ratio of the electron, compared with those of nuclear spins, leads PRE to produce distance constraints for contacts out to a range of 20 Å. Here, PRE was measured through attachment of the MTSL spin-label to the native cysteine residue C227, or to an engineered C227S/S273C double mutant.

Consistent with our preference for ¹³C direct-detection displayed throughout this study, PRE measurements were recorded in the ¹³C,¹⁵N-CON format. We note that CON-based detection allows measurement of a PRE intensity ratio for most proline residues, but due to spectral overlap the hexa-proline sequence at residues 239–244 was excluded from analysis. Unique to this phase of the study, spectra were acquired in both the “protonless” format and through the proton-start ¹³C,¹⁵N-(HACA)CON. The transverse relaxation rate contribution from proximity to an unpaired electron scales with both the electron g-factor and the gyromagnetic ratio of the observed nucleus.³ In the case of steady-state PRE, as detected here, this suggests that sensitivity to longer range or shorter lived contacts should be enhanced by selection of a high gyromagnetic ratio excitation nucleus (¹H) and reduced through excitation on a low gyromagnetic ratio nucleus (¹³C). Indeed, this effect is observed in the PRE profiles displayed in Figure 4, where in all data panels the dashed line reports the expectation value of the PRE under the null hypothesis that Pdx1-C behaves as a self-avoiding random flight chain.^49–50 For both MTSL attachment sites the expected Gaussian-shaped well surrounding the site of attachment, representing PRE intensity loss due to obligatory proximity within the covalent network, is substantially broader in the ¹H-start spectra (Figure 4A, C) than in the ¹³C-start spectra (Figure 4B, D), consistent with the expectation that a higher gyromagnetic ratio excitation nucleus will impart sensitivity to the PRE at greater distances.

Figure 4.

PRE data for Pdx1-C demonstrate non-uniform local chain collapse and the absence of long-range contacts. (A,B) PRE profiles for MTSL conjugation to C227. (C,D) PRE profiles for MTSL conjugation to C273, in the C227S/S273C double-mutant. The profiles in panels A and C report data collected using the proton-start ¹³C,¹⁵N-(HACA)-CON, whereas panels B and D report data collected using the carbon start ¹³C,¹⁵N-CON. All data bars represent the average intensity ratio over duplicate measurements. The dashed line indicates the predicted profile for a self-avoiding random flight chain as described in the Materials and Methods. The solid line represents the back-calculated PRE profile from CAMPARI simulation trajectories. (E) The net charge per residue (NCPR) is plotted to reveal local clusters of positive (blue) and negative (red) charge along the primary structure of Pdx1-C. NCPR is calculated as the average charge over a 5-residue window and assigned to the central residue of the window.

As stated above, Pdx1-C possesses a native C227 residue that was exploited for our initial PRE measurements. The data for ¹H-start (Fig. 4A) and ¹³C-start (Fig. 4B) PRE from C227-MTSL reveal two intriguing features. Recall that perturbations in the spin-relaxation profile near C227 suggested the possibility of globular contacts in the N-terminal half of Pdx1-C. Consistent with this observation, C227-MTSL showed short and medium range contacts consistent with a globule or compaction bias in the ensemble of Pdx1-C (Figure 4A and B), while the S273C mutant PRE profile (Figure 4C and D) was more similar to a random coil. This trend is hinted through somewhat reduced intensity ratios reaching all the way to the N-terminus, but is more pronounced in the region spanning residues 233–239. Further analysis of this trend required molecular simulations that will be discussed below.

Strikingly, there were no observable long-range contacts between the N-terminal and C-terminal half of Pdx1-C for either construct; all structural perturbations or deviations from random coil behavior are limited to relatively local contacts. This observation was surprising, given that the net charge per-residue (NCPR) indicates a tendency to negative charge near C227 and positive charge near the C-terminus of the protein that provided our initial motivation for selecting C227S/S273C as a second target for PRE studies. On a more local scale, although the N-terminal half of Pdx1-C displays a higher local NCPR, it is noteworthy that this arises mostly from reduced charge mixing compared to the C-terminal half. The degree of segregation in charge has been quantified by the Pappu laboratory in the form of a parameter (κ) that represents charge asymmetry over a sliding window.² The N-terminal and C-terminal halves of Pdx1-C have a κ of 0.331 and 0.132. Thus, one possible explanation for the more collapsed structure of the N-terminal region in Pdx1-C is that it is driven by electrostatic interaction between the strongly negative patch near C227 and the positively charged patch consisting of residues 207–213.

CAMPARI simulations of Pdx1-C.

The analysis of SAXS and NMR data presented above left us with several molecular hypotheses that can best be tested through computational modelling of the Pdx1-C solution ensemble. Recently, Monte Carlo (MC) simulations performed in the CAMPARI software package, utilizing the ABSINTH implicit solvation model have been used to characterize the effects of phosphorylation on the ensemble conformations of Ash1 and the physical properties that contribute to liquid-liquid phase separation of NCID.^51–52 Similarly, we have recently succeeded in modelling the C-terminal region of the phosphatase FCP1 using this methodology, yielding results that were consistent with both SAXS and NMR measurements.¹¹ Therefore, we chose to conduct MC simulations in ABSINTH to generate an ensemble of Pdx1-C for analysis in the context of the current data, with particular emphasis on reproduction of the PRE data sets described above.

Both our SAXS and PRE data indicated that Pdx1-C adopts a coil-like ensemble in solution. The radius of gyration was determined from the ensemble of CAMPARI-generated structures and was determined to be 25.0 ± 3.4 Å; while slightly smaller than expected, this average is indistinguishable within experimental uncertainty from the SAXS-derived Rg of 27.4 ± 0.77 Å. The simulated PRE data, derived from our MC calculation, is plotted as a solid black line in Figure 4A–D. The MC data show extensive short-range contacts between residue 227 and the proximal N-terminal basic patch, but no long-range interaction with residues C-terminal to the central polyproline stretch (Figure 4A, B). Interestingly, PRE experiments performed in either a carbon-start or proton-start format were in excellent agreement with their respective simulated PRE data. This suggests that adjusting the gyromagnetic ratio term in the Battiste-Wagner equation and solving for the peak intensity change was enough to account for the differences in the width of the relaxation enhancement zone surrounding the site of MTSL attachment in carbon-start and proton-start experiments. Similarly, simulated PRE data of Pdx1-C C227S, S273C were in agreement with experimental PRE data (Figure 4C, D). Taken together, these data preclude any persistent long-range intramolecular interaction between the N- and C-terminal halves of Pdx1-C.

The coil-like character of our ensemble is intriguing, given that Pdx1-C features a low fraction of charged residues and net charge per residue, both which are sequence features that often correlate with collapse to a disordered globule state.² We note that this general tendency was established for sequences depleted in proline residues, an abundance of which will tend to disfavor chain compaction in the disordered state. Given Pdx1-C’s high enrichment in proline, we next turned our attention to the potential impact of these residues on the chain’s dimensions. Proline is the only residue which appreciably populates the cis-peptide isomerization state in solvated polypeptides (~5% prolines are cis);⁵³ because Pdx1-C is 22% proline there is a high probability of a non-trivial amount of cis-proline in any single conformer that contributes to the overall ensemble. Improved Monte Carlo methods described by the Radhakrishnan et al., which were utilized here, can reproduce experimental proline isomer distributions.⁵⁴ Our MC data showed significant cis-proline in Pdx1-C, with the highest probabilities localized in the center of the protein (Figure 5A). To study the effects of cis-proline on the conformational ensemble of Pdx1-C, we computed two additional MC ensembles that describe the (experimentally hypothetical) states in which all prolines are either completely in the trans or completely in the cis configuration (though small fluctuations in the ω-torsion angles were allowed; see Materials and Methods). Theses all-trans or all-cis MC ensembles were used to compare with the unbiased ensemble, where the ω-bond was allowed to flip between cis and trans isomers for all proline residues. Aspherecity vs. R_g was determined for each model of the unbiased, all-cis, and all-trans Pdx1-C ensembles, as shown in Figure 5B, C, D (left panel). In the unbiased trajectory the mean R_g of the distribution was 25.0 ± 3.4 Å, as described above, and the mean asphericity A = 0.378 ± 0.165 (Fig. 5B). Locking all the prolines into a trans-peptide configuration does not change the mean R_g or asphericity, which were 24.8 ± 3.5 Å and 0.370 ± 0.164, respectively. The all-cis proline Pdx1-C trajectory yields a slight decrease in the R_g to 22.5 ± 3.1 Å, but this decrease is not significantly different within error from the mean in the unbiased trajectory; the same was true for the asphericity, which produced a mean value of 0.36 ± 0.16. Thus, even in an all cis-Pro context, the global chain dimensions are only marginally different from those in the state with native isomer distribution. Fitzkee and Rose have elegantly demonstrated that, consistent with expectations from polymer theory, local non-random structure generally has no impact on global coil-like chain dimensions.⁵⁵ Consistent with this polymer view, our data demonstrate that proline isomerization cannot significantly alter the global dimensions of the Pdx1-C chain.

Figure 5.

Analysis of CAMPARI trajectories to determine the influence of proline trans-to-cis isomerization on Pdx1-C compaction. (A) The cis-Pro isomer content is shown for the control CAMPARI trajectory at 310K. The (A) control, (B) all cis-Pro, and (d) all trans-Pro trajectories yield the displayed asphericity vs. R_g distributions (left) and internal scaling profiles (right). Analyzed structure ensembles were from the 310 K temperature replica. For the internal scaling plots, the excluded volume (EV) and Flory random coil (FRC) reference curves are shown in blue and red respectively.

While proline trans-to-cis isomerization will not collapse Pdx1-C on average, it remains likely that the persistence length of a polypeptide enriched in proline will be greater than that expected for a random polypeptide sequence. The scaling of internal distances for each of the three trajectories were calculated and are shown in Figure 5B, C, D (right panel). Internal distances are calculated as the average distance between all atoms in a pair of residues vs. the sequence separation of those two particular residues. Scaling profiles for excluded volume (EV) chains and a Flory random coil (FRC) are shown as references. The internal scaling profile of the unbiased and all-trans proline trajectories were not observably different, but the all-cis trajectory showed a significant overall reduction in the internal scaling parameter.

The analysis above demonstrates that on the global scale, Pdx1-C adopts an open and extended ensemble, inconsistent with globular collapse. On the other hand, our MC ensemble provides us with the opportunity to investigate whether Pdx1-C features segments of its primary structure that collapse into more globular states. Indeed, the PRE data recorded with MTSL attached to C227 suggests the existence of just such a cluster of local interactions. Figure 6 shows a contact map of the 310K replica from the trajectory of Pdx1-C with unbiased proline isomerization. To construct this map, distances were measured for each pairwise Cα-Cα in a model and averaged over all the models in the trajectory. Small distances are depicted in red, medium distances in yellow, and large distances in blue (quantitative scale bar shown in Figure 6). One of the most surprising elements of the Pdx1-C primary structure is a 6-residue polyproline segment encompassing residues 239–244. This polyproline region appears to separate Pdx1-C into two structurally isolated segments; few contacts are indicated between residues bridging this segment. Consistent with the PRE data, region 200–237 appears more globular than region 245–283. We note that the charged residues in the 200–237 region are poorly mixed, leading us to hypothesize that electrostatic attraction between clusters of charged residues in this region drives the chain toward the (slight) compaction observed (Figure 4E). In contrast, relative charge depletion and much higher proline content (~26%) in the C-terminal half of Pdx1-C yields an unusually extended ensemble that is depleted in local residue-residue contacts.

Figure 6.

Heatmap of contacts observed in the Pdx1-C ensemble, constructed from the 310 K temperature replica in the REMC calculation. Colors depict distances from Cα to Cα as quantified in the color key. In order to generate a representative set of conformations, five structures were extracted at random from the computed ensemble and aligned by the central six proline residues shown in gray (residues 239–244)

CONCLUSIONS

NMR is well-suited to characterize the conformational ensembles adopted by IDPs in solution, even in the absence of stable secondary structure. Here, we have presented extensive NMR characterization of the highly proline-enriched IDP Pdx1-C, relying exclusively on the ¹³C direct-detect methodology our laboratory has developed. Toward this broader objective, we have introduced four new RDC experiments, which are available for download from our website. Further, PRE measurements are reported in both the ¹⁵N,¹³C-CON and ¹⁵N,¹³C-(HACA)CON detection format, yielding additional structure constraints. We note that it appears the ¹³C,¹⁵N-(HACA)CON is more sensitive to long-range contacts, which will merit further investigation with an IDP that features persistent long-range contacts in its solution ensemble. Here, our PRE data show that Pdx1-C is highly extended overall, with a slight tendency to globular structure at the N-terminus.

We set out to describe the solution ensemble of Pdx1-C, with the purpose of building necessary information regarding the unbound structure of this critical regulatory region of the full transcription factor. Pdx1-C is known to be the target of the E3-ubiquitin ligase substrate adaptor protein SPOP, which promotes Pdx1-C degradation under low blood glucose conditions.¹⁸ Often in the IDP community, there is a tendency to seek confirmation of secondary structure, such as partially stable α-helix motifs, in the unbound ensemble of IDPs, as these are presumed to be the site of interaction with biological partners. In the case of SPOP, available crystal structures of the protein’s MATH recruitment domain bound to client protein-derived peptides demonstrate that those peptides bind in an extended format;⁵⁶ there is no secondary structure for one to seek in the unbound ensemble. However, the SPOP binding motif within Pdx1-C embedded within the region we find to possess the most local compaction, suggesting that SPOP must either capture a rare Pdx1-C conformer or induce a more extended structure in conjunction with binding. Overall, Pdx1-C serves as a valuable counterexample to the broad trends in the IDP literature: molecular interaction motifs are not always found within secondary structural element; yet, even in their absence, a detailed understanding of the unbound-state structure is necessary to develop full mechanistic understanding. The new pulse programs and extended analysis strategy reported here should be of general benefit to future efforts toward determination of IDP ensemble properties.