Paramagnetic relaxation enhancement‐assisted structural characterization of a partially disordered conformation of ubiquitin

Abstract

Nuclear magnetic resonance (NMR) is a powerful tool to study three‐dimensional structures as well as protein conformational fluctuations in solution, but it is compromised by increases in peak widths and missing signals. We previously reported that ubiquitin has two folded conformations, N₁ and N₂ and plus another folded conformation, I, in which some amide group signals of residues 33–41 almost disappeared above 3 kbar at pH 4.5 and 273 K. Thus, well‐converged structural models could not be obtained for this region owing to the absence of distance restraints. Here, we reexamine the problem using the ubiquitin Q41N variant as a model for this locally disordered conformation, I. We demonstrate that the variant shows pressure‐induced loss of backbone amide group signals at residues 28, 33, 36, and 39–41 like the wild‐type, with a similar but smaller effect on CαH and CβH signals. In order to characterize this I structure, we measured paramagnetic relaxation enhancement (PRE) under high pressure to obtain distance restraints, and calculated the structure assisted by Bayesian inference. We conclude that the more disordered I conformation observed at pH 4.0, 278 K, and 2.5 kbar largely retained the N₂ conformation, although the amide groups at residues 33–41 have more heterogeneous conformations and more contact with water, which differ from the N₁ and N₂ states. The PRE‐assisted strategy has the potential to improve structural characterization of proteins that lack NMR signals, especially for relatively more open and hydrated protein conformations.

1. INTRODUCTION

Proteins exist in a dynamic equilibrium among multiple conformations, which have marginal differences in Gibbs free energies.1, 2 High‐energy, typically low‐population protein states, that is, open conformations of enzymes and locally disordered conformations, can be functionally important and/or at risk of forming toxic aggregates.3, 4, 5 For instance, high‐energy conformations have been identified in Escherichia coli dihydrofolate reductase, which mediate substrate and cofactor exchange,3 while transient structural distortion of superoxide dismutase 1 (SOD1) triggers aberrant oligomerization, and the misfolding and oligomerization of SOD1 have been linked to amyotrophic lateral sclerosis.4

We have focused on the structural characterization of the high‐energy conformations of proteins using high‐pressure nuclear magnetic resonance (NMR) spectroscopy.6, 7, 8, 9 Elevated pressure can shift a population of protein conformers from the basic folded conformer to a fully unfolded conformer because the former has a larger partial molar volume.10 Although solution NMR, including high‐pressure NMR, is a useful technique to elucidate the dynamic equilibrium of proteins, structural determination remains difficult since conformational heterogeneity and the microsecond‐to‐millisecond time scale motions of the polypeptide chain lead to peak‐broadening and missing peaks in the NMR spectra.10, 11 In general, missing NMR signals result in a lack of structural information such as distance restraints derived from the nuclear Overhauser effect (NOE). To overcome this limitation, several NMR approaches that can provide structural restraints have been reported; for instance, paramagnetic relaxation enhancement (PRE),12, 13 which provides relatively longer distance restraints (e.g., ∼25 Å) than the NOE (e.g., ∼6 Å), and residual dipolar coupling,14, 15 which provides orientation restraints rather than distance restraints. However, to obtain the information, NMR signals must be observed in the NMR spectra. Therefore, it is still difficult to obtain structural information of regions with missing signals by conventional approaches.

Ubiquitin is a functionally important protein for many cellular activities, and thus its three‐dimensional structure and conformational fluctuations have been extensively studied using biophysical approaches.14, 16, 17, 18, 19, 20 However, the structural details of the transient high‐energy conformations of ubiquitin, for example, folding‐intermediate and partially unfolded conformations,19, 21, 22, 23 are severely limited because of their low population. They are nevertheless of considerable interest, because local minima in the energy landscape are usually functionally important.9 We previously reported that an alternatively folded conformation (N₂) accounted for ∼20% of the population of wild‐type ubiquitin at ambient pressure (298 K and pH 7.2), but ∼70% of the population of the ubiquitin Q41N variant (hereafter referred to as Q41N) at ambient pressure, and accounted for 97% of the population at 2500 bar (298 K and pH 7.2).24 Because there were only a couple of missing signals in N₂, we were able to determine the NOE‐based structure of the conformation using high‐pressure NMR.6, 7

More extreme solution conditions (low pH, high pressure, and/or low temperature) increase the population of another state, here denoted I. This state resembles N₂, but is more disordered, specifically at residues 33–42, for which many of the amide signals are broadened or missing, and at the C‐terminus. We showed that this state (pH 4.5 and 273 K) shares structural features with a proline‐trapped kinetic intermediate based on pulse‐labeling ¹H/²H‐exchange NMR.11, 23 It also has some features in common with an on‐pathway folding‐intermediate of a ubiquitin variant (V17A/V26A) studied by Charlier et al.25 using pressure‐jump NMR, which has a large chemical shift difference for the C‐terminal residues between the native and folded intermediate states. The state I is thus native‐like, but is equally clearly an unfolding intermediate. The missing amide NMR signals of residues within the region 33–42 in I prevented its NOE‐based structural determination.

Here, we have used a low‐pH and low‐temperature (i.e., pH 4.0 and 278 K) condition, and show that I has a more hydrated and denatured conformation than N₂, and thus moves further toward a denatured state. Moreover, we successfully determined the structure of I using ubiquitin mutant Q41N with a simple strategy based on PRE‐assisted NMR at high pressure.

2. RESULTS AND DISCUSSION

2.1. NMR spectral changes under high pressure

We performed ¹H/¹⁵N and ¹H/¹³C HSQC measurements at a ¹H frequency of 800.34 MHz (AVANCE3‐800) using a pressure‐resistant NMR cell. HSQC cross‐peaks of Q41N at pH 4.0, 278 K, and 1 bar were assigned to individual amino acid residues based on temperature and pH titration experiments, referring to assignments of Q41N at pH 7.2, 298 K, and 1 bar (BMRB: 11505).24 These specific low‐pH and low‐temperature conditions were chosen after consideration of protein stability and spectral quality.

Figure 1a shows the ¹H/¹⁵N HSQC spectra of uniformly ¹⁵N‐labeled Q41N at pH 4.0 and 278 K recorded at 500‐bar intervals between 1 and 2,500 bar. These spectral changes were found to be reversible up to 2,500 bar. Many cross‐peaks showed substantial changes in ¹H and ¹⁵N chemical shifts with increasing pressure. In general, chemical shifts are sensitive parameters for monitoring the structural changes of proteins. Specifically, changes in the chemical shifts of amide ¹H and ¹⁵N are correlated to changes in the strengths of hydrogen bonds as well as the backbone ϕ and ψ torsion angles,9 accompanied by mechanical compression and/or a transition into a different conformational state with a different compressibility. In order to observe these structural changes by changes in chemical shifts, the time scale of the structural changes should be much faster than the NMR chemical shift timescale (i.e., approximately in the millisecond range). Figure S1 shows pressure‐induced chemical shifts of V70. We have previously shown that the N₂ conformation accounts for approximately 70% of the population of the ubiquitin Q41N variant at ambient pressure and 97% of the population at 2,500 bar (298 K and pH 7.2), based on pressure‐induced sigmoid‐shaped chemical shifts.24 The nonlinear shifts in the present data indicate a similar increase in N₂ conformation under high pressure. N₂ is 65% populated at 1 bar, but 98% populated at 2,500 bar (278 K and pH 4.0) according to the sigmoid‐shaped analysis (see Methods in Supporting Information)

Figure 1.

¹H/¹⁵N HSQC spectral analysis for backbone amide groups. (a) Overlay of ¹H/¹⁵N HSQC spectra of uniformly ¹³C/¹⁵N‐labeled ubiquitin Q41N at different pressures, from 1 bar to 2,500 bar, at pH 4.0 and 278 K. (b) Volumes of original cross‐peaks at different pressures relative to those at 1 bar (red, residues 28, 33, 36, 39–41; black, other residues). (c) Volumes of new cross‐peaks at different pressures (red, residues 28, 33, 36, 39–41; black, other residues). Peak volumes are depicted in arbitrary units. (d) ¹H/¹⁵N HSQC spectrum at 2,500 bar and signal assignments. Assignments of the original cross‐peaks are underlined and the others are new cross‐peaks

Pressure‐induced changes in the intensities (i.e., volumes) of the cross‐peaks present in the 1‐bar spectrum (hereafter referred to as the original cross‐peaks) are plotted in Figure 1b for Q41N. As the pressure increased, the cross‐peak intensities of residues 28, 33, 36, and 39–41 preferentially decreased while the remaining peaks decreased more gradually. Quite similar results were observed for wild‐type ubiquitin at pH 4.5 and 273 K but at much higher pressures, that is, ∼2,500–3,500 bar.11 Because the intensities of the cross‐peaks were restored at 298 K,26 we hypothesized that the loss of the ¹H/¹⁵N HSQC signals is likely due to increased relaxation originating from conformational fluctuations and changes in solvation operating on the microsecond‐to‐millisecond time scale. In addition, new cross‐peaks, corresponding to a disordered polypeptide chain, gradually appeared in the central part of the spectra recorded at pressures at and above 1,500 bar. Figure 1c shows the pressure‐induced incremental appearance of these new cross‐peaks. Assignments for the new cross‐peaks were obtained for uniformly ¹³C/¹⁵N‐doubly labeled Q41N at 278 K and 2,500 bar using triple‐resonance NMR experiments, that is, CBCA(CO)NH, HNCACB, HNCA, HN(CO)CA, HNCO, and HN(CA)CO, on a DRX‐600 (¹H, 600.23 MHz) spectrometer (Figure 1d). All new cross‐peaks, including those for residues 28, 33, 36, and 39–41, cooperatively increased their intensities with increasing pressure, indicating a growing population of the entirely disordered conformation of the protein. In contrast to the pressure‐induced loss of folded structure, which clearly occurred at different rates in different regions of the protein (Figure 1b), unfolded signals appeared at the same rate for the entire protein (Figure 1c).

To investigate the effects of pressure on the CH, CH₂, and CH₃ groups of Q41N, ¹H/¹³C HSQC spectra were recorded at 500‐bar intervals between 1 and 2,500 bar at 278 K. As shown in Figure S2a, ¹H and ¹³C chemical shifts of several CH, CH₂, and CH₃ groups (e.g., K6Cβ, T9CβH, L43βH, I61CγH, E64CαH) changed significantly as the pressure increased. We note in particular the ring‐current shifted signal from I61Hγ, which becomes even more strongly shifted with an increase in pressure, indicating a compression of the hydrophobic core. These changes were almost linear with pressure, indicating that the local structural or dynamic changes responsible for the linear shifts were constant over the pressure range studied.27 Therefore, the structural change of N₁–N₂ to I is occurring within the same folded ensemble. Figure 2a,b shows plots of the cross‐peak intensities of the CαH and CβH groups, respectively, at different pressures relative to those observed at 1 bar. Many of these signals gradually lost their intensities with increasing pressure. Figure S2b shows the HSQC spectra of the CαH and CβH groups at 1 and 2,500 bar. Figure S2c shows the pressure‐induced incremental appearance of some well‐isolated new cross‐peaks of the CαH groups (depicted by green arrows in Figure S2b). New cross‐peaks, corresponding to a disordered polypeptide chain, gradually appeared in the spectra recorded at pressures at and above 1,500 bar, similar to the data of the amide groups (Figure 1c). These results also indicate a growing population of completely disordered conformation of the protein. However, the CβH groups of residues 23, 31, and 40 showed preferential decrease in their intensities, similar to the data of the backbone amide groups.

Figure 2.

Changes in peak volumes of CH groups versus pressure. Peak volumes of the (a) CαH and (b) CβH groups of uniformly ¹³C/¹⁵N‐labeled ubiquitin Q41N at different pressures relative to those at 1 bar (pH 4.0 and 278 K; broken line, residues 23, 31, and 40 preferentially losing intensity)

Pressure‐induced changes in the cross‐peak intensities of the NH, CαH, and CβH groups are summarized for each residue in Figure S3. Most of these signals lose their intensities with the same pressure dependence, except some of the amide group signals (i.e., residues 28, 33, 36, and 39–41) and the CβH group signals of residues 23, 31, and 40, showing initiation of global unfolding above 1,500 bar, but with localized differences in unfolding, specifically formation of the partially unfolded intermediate state I.

2.2. Pressure‐induced broadening and disappearance of NMR signals

To further examine the pressure‐induced changes in the backbone dynamics of Q41N, we performed ¹⁵N‐transverse relaxation experiments at elevated pressure using a DRX‐600 (¹H, 600.23 MHz) spectrometer. ¹⁵N‐transverse relaxation of individual amide groups is affected by chemical exchange processes taking place at the microsecond‐to‐millisecond time scale as well as internal motions at the picosecond‐to‐nanosecond timescale. Figure 3 shows the ¹⁵N‐transverse relaxation rate constants, R ₂, estimated for the original cross‐peaks. Larger R ₂ values than average were clearly observed for residues 16, 23, 25, 33, 39, and 70–71 at 1 bar, indicating the presence of chemical exchange processes.6, 28, 29 Note that cross‐peaks of E24 and G53 were missing in the spectrum obtained under the experimental solution condition. Specifically, a backbone flip motion of E51‐R54 coupled with a deformation of the hydrogen bond between I23H^N‐R54CO and a side‐chain rearrangement of Ile23 affect R ₂ values of residues 23–25 and 53.28, 29 In addition, reorientation of the C‐terminus of α₁‐helix and β₅‐strand coupled with the N₁–N₂ conformational transition affects the R ₂ value of residues 70–71.6, 7

Figure 3.

Pressure dependence of ¹⁵N‐transverse relaxation rate constants. ¹⁵N‐transverse relaxation rate constants, R ₂, determined for uniformly ¹⁵N‐labeled ubiquitin Q41N at different pressures (bars, 1 bar; open circles, 1,000 bar; closed circles, 2,500 bar) at pH 4.0 and 278 K. Error bars show the root‐mean‐square‐deviation obtained from the single exponential fit. Secondary structure elements are indicated at the top of the panel by bars (α‐helices) and arrows (β‐strands)

With increasing pressure up to 2,500 bar, remarkable increases in the R ₂ values were observed specifically for the amide groups of residues 28–39. This can be explained by an increasing contribution of chemical exchange processes (i.e., R _ex) to the R ₂ at elevated pressures. The N₂ conformation accounts for approximately 70% of the population of the ubiquitin Q41N variant at ambient pressure, and the population of N₂ increases with increasing pressure (Figure S1). Therefore, N₂ is more highly populated at higher pressures, resulting in a sharp decline in the contribution to R _ex from the N₁–N₂ exchange. Although the N₁–N₂ exchange rate constant is likely to decrease with increasing pressure as shown in WT (i.e., 4.8 × 10⁵ s⁻¹ at 30 bar, 1.7 × 10⁵ s⁻¹ at 3 kbar, pH 4.6 and 293 K),6 the contribution of the population shift to R _ex will dominate that of the rate constants and reduce R _ex rates arising from the N₁–N₂ exchange. We therefore conclude that the increases seen in R ₂ in the 28–39 region at high pressure are due an increased population of the locally disordered conformation I.

Under the lower pH and lower temperature condition (i.e., pH 4.0, 278 K, and 2,500 bar), Q41N is expected to exhibit greater destabilization of the hydrogen bonds in the backbone and partial exposure of amide groups to the solvent. Indeed, pressure‐induced increments of water–amide proton interactions were detected spatially at the C‐terminal side of wild‐type ubiquitin (e.g., residues 32–35, 40–41, and 71) at pH 7.4 and 298 K by CLEANEX‐PM NMR experiments.30 We therefore attempted to measure water–amide proton exchange in our low‐pH/low‐temperature conditions, to determine whether signal loss could be due to solvent exchange. Rate constants of water–amide proton exchange were too small to detect using CLEANEX‐PM experiments under acidic conditions (pH 4.0). Indeed, no signals were observed using the CLEANEX‐PM experiments with a 100‐ms mixing time at 1 and 2,500 bar (data not shown). According to a 1‐ms molecular dynamics simulation for BPTI, two water molecules must be directly coordinated to backbone amides in an open state, to allow amide protons to exchange with water.31 Therefore, the pressure‐induced broadening and loss of cross‐peak intensities of the amide and CβH groups (Figures 1a and 2b) in the regions from the α₁‐helix to the β₃‐strand are concluded to originate from heterogeneity of atomic packing, including protein and hydration water, and/or fluctuation among the conformations within the NMR chemical shift timescale (i.e., the millisecond scale).

In contrast to pressure‐induced increments of R ₂, a decline of R ₂ was observed for E16 (Figure 3) of the Q41N variant unlike WT. Actually, the value of E16 was significantly larger than that of WT at 1 bar,24 indicating that the substitution of Q41 to N41 increased the local dynamics or conformational heterogeneity at the opposite side of the protein, namely at and around E16. The decline of R ₂ implies that pressure reduced the effects of the substitution, although the cause is still unknown.

2.3. PRE‐assisted NMR structure determination at high pressure

When the NMR spectral quality is poor due to peak broadening and missing peaks, it is often difficult to obtain a sufficient number of distance restraints for protein structural determination. A series of triple‐ and double‐resonance NMR experiments for signal assignments were performed on uniformly ¹³C/¹⁵N‐doubly labeled Q41N using a DRX‐600 spectrometer at pH 4.0, 278 K and 2,500 bar (see Section 4). Using an AVANCE3‐800 spectrometer, we simultaneously acquired the ¹³C/¹⁵N‐edited NOESY spectrum for uniformly ¹³C/¹⁵N‐doubly labeled Q41N at pH 4.0, 278 K, and 2,500 bar to collect NOE data. Barely any inter‐residue NOEs were observed between residues 35–41 and the rest of the protein (Figure S4a). Structural calculation of Q41N was performed with CYANA version 3.97 using 1,238 NOE‐based distance restraints and 74 torsion‐angle restraints derived using TALOS+32 (Figure S5a). Details of the restraints and geometrical statistics are shown in Table S2. However, structural convergence was rather poor between the end of the α₁‐helix and the β₃‐strand region as well as in the β₅‐strand at the C‐terminus. The root‐mean‐square‐deviation (RMSD) to the mean coordinates for the backbone atoms of residues 1–70 was 0.55 Å.

To overcome the limitation of the low number of distance restraints between the NMR‐invisible amide groups and ‐visible sites on the convergence of the ensemble, we performed PRE‐assisted structure determination. PRE effects from the paramagnetic agent S‐[(1‐oxyl‐2,2,5,5‐tetramethyl‐2,5‐dihydro‐1H‐pyrrol‐3‐yl)methyl] methanesulfonothioate (MTSL) tagged in the loop region where pressure‐induced broadening and NMR signal disappearance occurred were observed on NMR‐visible amide groups. Cysteine residues are usually used to add MTSL. To minimize the effects of residue‐substitution to cysteine on protein stability and dynamics, we chose to label K33 and Q40, which are exposed to the solvent. E34, I36, P37, and P38 were excluded as candidates for the substitution because the side‐chains of E34 and I36 face the inside of the protein and thus the residue‐substitution and ‐modification may significantly change the stability and dynamics of the region. A cis–trans isomerization of P37 and/or P38 is related to a population of the cis‐trapped kinetic intermediate23 and thus the substitution may change structural properties of the equilibrium intermediate. Ubiquitin G35C/Q41N variant was not sufficiently produced by the current E. coli expression (see Section 4). Note that wild‐type ubiquitin has no cysteine residues. We therefore prepared uniformly ¹⁵N‐labeled ubiquitin K33C/Q41N and Q40C/Q41N double variants. When the spin‐labeled (oxidized) MTSL was covalently bound to the substituted cysteine side chain, the protein possesses a paramagnetic property resulting from an unpaired electron present in MTSL. The paramagnetic effects from the unpaired electron are very strong, which affect the spin‐relaxation of nuclei up to an approximate 25‐Å distance. When the spin‐label is reduced with twofold excess ascorbic acid, the protein becomes diamagnetic. ¹H/¹⁵N HSQC spectra of ubiquitin K33C/Q41N and ubiquitin Q40C/Q41N were measured with the oxidized and reduced forms of MTSL at pH 4.0, 278 K, and 2,500 bar using a DRX‐600 spectrometer. HSQC cross‐peaks in the double‐variant proteins were assigned to individual amino acid residues using ¹⁵N‐edited TOCSY and NOESY experiments based on assignments of ubiquitin Q41N.

Effects of residue‐substitution and MTSL‐tag were evaluated in terms of structure and dynamics of the protein. Chemical shift perturbations (CSPs) by residue‐substitution and MTSL‐tag at 1 and 2,500 bar are shown in Figures S6 and S7, respectively. At 1 bar, CSPs (i.e., [(Δδ H)² + (Δδ N/5)²]^0.5) of greater than 0.2 ppm were observed at residues 40, 43–44, and 71 in the MTSL‐tagged K33C/Q41N variant (Figure S6a) and at residues 45 and 72–74 in the MTSL‐tagged Q40N/Q41N variant (Figure S6b). At 2,500 bar, CSPs were smaller particularly at residues 42–44. Because CSPs (e.g., greater than 0.2 ppm) were observed only at residues 71–73 spatially proximal to the substituted residue at 2,500 bar, the effects of residue‐substitution and MTSL‐tag on the average conformation of the intermediate are presumed to be limited to the C‐terminal strand. The effects on the backbone dynamics were evaluated by analysis of line‐broadening and ¹⁵N spin‐relaxation. In the case of the MTSL‐tagged K33C/Q41N variant, more cross‐peaks of residues 27–42 disappeared from the spectrum (Figures S6a and S7a), while results of the MTSL‐tagged Q40C/Q41N variant are almost identical to the Q41N variant (Figures S6b and S7b). In addition, a similar decline in the cross‐peak volumes of residues 33, 39, and 41 was observed in the MTSL (reduced form)‐tagged Q40C/Q41N variant with increasing pressure (Figure S8), indicating that the MTSL‐tagged variant also shares similar disordered conformation with that of Q41N and WT. The missing signals with the addition of the MTSL‐tag prevented the same analysis of the K33C/Q41N variant. We also compared the rotational correlation time, τ _r, derived from ¹⁵N‐T ₁/T ₂ of each backbone amide group at 2,500 bar and 278 K, among the variants (Figure S9). τ _r values were not altered by the residue‐substitution and modification. Taken together, the effects of the residue‐substitution and MTSL‐tag on the protein conformation are limited, although backbone dynamics of the regions specifically from the α₁‐helix to the β₃‐strand seems to be sensitive to residue‐substitution and ‐modification as well as pressure perturbation. Thus, it is meaningful to supply additional distance information between the partially disordered region (i.e., α₁‐helix to the β₃‐strand) and other sites using the paramagnetic tag. As discussed below, PRE‐based distance information was used as gentle distance restraints for structure determination (see Section 4) to reduce the risk of overfitting the structural models. Moreover, PRE‐based distance restraints for residues 71–76, whose chemical shifts are sensitive to residue‐substitution and ‐modification, were not used for structural determination.

Rather than directly measuring the relaxation rates from a series of two‐dimensional NMR experiments, relaxation rates can be indirectly obtained from the ratios of the peak heights of a paramagnetic sample to those of a diamagnetic sample, that is, I _para/I _dia.33 Figure 4a,b shows the residue‐specific I _para/I _dia ratios for ubiquitin K33C/Q41N and ubiquitin Q40C/Q41N, respectively, measured at 1 and 2,500 bar. A lower I _para/I _dia value indicates a greater PRE effect. At 1 bar, the ratios were almost null for the residues spatially close to the spin‐labeled MTSL (i.e., residues 7–17, 39, and 69–72 in ubiquitin K33C/Q41N; residues 7–11, 35, 42, 52, and 69–76 in ubiquitin Q40C/Q41N). At 2,500 bar, the I _para/I _dia ratio showed both increases and decreases in value for several sites. For instance, increases in the ratio for residues 5–6 and 11–12 in ubiquitin Q40C/Q41N indicated that the PRE effects were weakened, and thus the residues moved away from residue 40. Decreases in the ratio for residues 25–29, 49–51, and 54–55 in ubiquitin Q40C/Q41N indicated that these residues are located closer to residue 40. Note that pressure‐induced changes in the orientation of the MTSL side‐chain may also affect changes in the I _para/I _dia ratio. Based on the results of the PRE experiments, a total of 140 PRE‐based distance restraints were used for the structural determination of Q41N at 2500 bar. Intensity ratios were converted into distances according to a previously described method33 (see Section 4). Figure S5b shows 20 structural models calculated with 140 PRE‐ and 1,246 NOE‐based distance restraints, and 74 torsion angle restraints. The number of inter‐residue NOEs for each residue is shown in Figure S4b. Note that a consequence of addition of the PRE‐based restraints is that the total number of NOEs accepted in the automated NOE assignments by CYANA is slightly increased. In particular, the number of inter‐residue NOEs was increased for residues 40 and 41. The RMSD for the backbone atoms of residues 1–70 decreased from 0.55 to 0.45 Å, and the improvement of structural convergence was particularly noticeable at residues 33–41 (Figure S5b).

Figure 4.

PRE effects for ubiquitin K33C/Q41N and ubiquitin Q40C/Q41N. Ratios of peak heights between paramagnetic and diamagnetic samples (I _para/I _dia) of ubiquitin K33C/Q41N (a) and ubiquitin Q40C/Q41N (b) at 1 bar (bars) and 2,500 bar (closed circles). Secondary structure elements are indicated at the top of the panel by bars (α‐helices) and arrows (β‐strands). Asterisks show the sites of MTSL covalent attachment. The error bars for I _para/I _dia were estimated from the noise considering error propagation, that is, $\Delta \left(\raisebox{1ex}{${\mathrm{I}}_{\text{para}}$}\!\left/ \!\raisebox{-1ex}{$I_{\mathrm{dia}}$}\right.\right)=\left(\raisebox{1ex}{$I_{\text{para}}$}\!\left/ \!\raisebox{-1ex}{$I_{\mathrm{dia}}$}\right.\right)\sqrt{{\left(\frac{\Delta I_{\text{para}}}{I_{\text{para}}}\right)}^2+{\left(\frac{\Delta I_{\mathrm{dia}}}{I_{\mathrm{dia}}}\right)}^2}$. MTSL, S‐[(1‐oxyl‐2,2,5,5‐tetramethyl‐2,5‐dihydro‐1H‐pyrrol‐3‐yl)methyl] methanesulfonothioate; PRE, paramagnetic relaxation enhancement

The line‐broadening observed for residues 33–41 implies exchange between multiple conformations in this region. The structures calculated therefore represent r ⁻⁶‐weighted averages. It is worth noting that this statement is true for all NMR structures obtained using NOE distance restraints: the difference here is that we can be confident that there is conformational exchange occurring, and it is slower (and therefore possibly more extensive) than normal.

2.4. NMR structure refinement based on Bayesian inference

We used an NMR structure refinement method based on Bayesian inference34 implemented in CYANA.35 The method involves a molecular dynamics simulation to quickly obtain global structures with automatic NOE assignment and subsequent structure refinement by Bayesian inference, providing a useful method when only sparse experimental data are available. Figure 5a,b shows the Bayesian‐refined structures determined without and with PRE‐based distance restraints, respectively. A total of 1,900 structures are presented in each top panel. Principal components analysis (PCA) was performed for the Cα position of residues 1–70 to visualize the distributions of the calculated structural ensembles. The bottom panels show the first and second principal components, PC1 and PC2, in the PCA. Without PRE restraints, the 1,900 structures were widely distributed into three main regions in the PC1–PC2 map, which is likely attributed to the limited distance restraints between the NMR‐visible and ‐invisible regions. In contrast, the structures were more tightly grouped in the map with PRE restraints included in the Bayesian‐refined structure calculation. Twenty of the lower energy structures from the 1,900 models were selected, showing an RMSD for residues 1–70 of the PRE‐assisted and Bayesian‐refined structures of 0.27 Å for backbone atoms and 0.48 Å for heavy atoms, representing a remarkable improvement of structural convergence. The correlation between the distance calculated from the PRE and the distances determined from 20 structural models calculated with PRE or without PRE is shown in Figure S10. When PRE data were used in the structural calculation, the correlation was generally improved and most expected distances determined from three‐dimensional structures appeared within 5 Å boundaries.

Figure 5.

NMR structure refinement based on Bayesian inference. (a and b) (top) The 1,900 structures yielded by the Bayesian‐refined method without and with PRE‐based structural restraints, respectively. (Bottom) Distributions of the first and second principal components (PC1, PC2) for the structural ensembles shown in the top panel. PCA was performed for the Cα atoms of residues 1–70. Twenty energy‐minimized structural models were deposited in the Protein Data Bank (ID: 6K4I). NMR, nuclear magnetic resonance; PCA, principal components analysis; PRE, paramagnetic relaxation enhancement

2.5. Pressure‐induced unfolding of ubiquitin

Comparing the Bayesian‐refined structure of Q41N at pH 4.0, 278 K, and 2,500 bar (I) with the wild‐type conformation (i.e., N₁ conformation, PDB: 1D3Z) and the alternatively folded conformation (i.e., N₂ conformation, PDB: 2RU6; Figure S11), this structure shows reorientation of the loop region and the C‐terminal β5‐strand, resulting in a more open conformation at the C‐terminal side of the protein. This phenomenon is similar to the structural characteristics of the protein observed at pH 7.2, 298 K, and 2,500 bar, indicating that the protein under the present solution conditions largely retained the N₂ conformation, the main difference being that the amide groups at residues 33–41 seemed to have more heterogeneous conformations and more contact with water. According to kinetic NMR experiments,23 the intermediately folded species was trapped with a cis conformer of P37 and/or P38 peptide bonds. However, the P37 and P38 peptide bonds entirely adopted trans conformations in the equilibrium I structure. The ψ torsion angles were closely similar to those of N₂, specifically those of P37 and P38 (PDB: 6K4I). In addition, pressure‐induced changes in chemical shifts (e.g., ¹H and ¹³C) and cross‐peak volumes of P37 and P38 were not unusual, and inter‐residue NOEs were also consistent with the trans conformation of P37 and/or P38 peptide bonds (Figures S3 and S12). In other words, we can characterize four states whose populations can be controlled using temperature, pH and pressure: N₁↔N₂↔I↔U, plus a further kinetically trapped cis‐proline conformation.

Overall, the more disordered conformation I observed at pH 4.0, 278 K, and 2,500 bar represents a further unravelling at the same 33–41 region, that differs between N₁ and N₂. Based on hydrogen/deuterium exchange NMR experiments, Englander and coworkers have suggested the presence of cooperative unfolding units in protein structures.36 Residues 33–41, which are involved in the conformational transition from N₁ to N₂ and then have more heterogeneous conformations and more contact with water in I, could be the primary cooperative unfolding unit of ubiquitin.

3. CONCLUSIONS

We demonstrated that the Q41N variant of ubiquitin, like the WT, shows pressure‐induced loss of backbone amide group signals at several residues in the 33–41 region, while most of the CαH and CβH groups showed no preferential decline in cross‐peak intensities, showing more heterogeneous conformations and more contact with water at those backbone amide sites under high pressures. Because residues 33–41 are also involved in the N₁–N₂ transition, they could be the primary cooperative unfolding unit of the protein. In addition, we have observed PRE under high pressures for the first time, and by adopting PRE‐based distance restraints, we determined the structure of the I conformation of ubiquitin. Peak‐broadening and missing peaks resulting from conformational exchange on the microsecond‐to‐millisecond time scale often reduce the quality of NMR spectra, making NOE‐based structural determination difficult. The PRE‐assisted strategy has the potential to improve structural characterization of proteins that have missing NMR signals, especially for relatively more open and hydrated protein conformations.

4. METHODS

4.1.

4.1.1. Sample preparation

Uniformly ¹⁵N‐labeled and ¹³C/¹⁵N‐doubly labeled Q41N ubiquitin and uniformly ¹⁵N‐labeled double‐variant ubiquitin K33C/Q41N or Q40C/Q41N were produced by conventional Escherichia coli expression (e.g., pGEX‐6P‐1 vector and E. coli competent cells BL21 [DE3]).24 The cysteine variants were dissolved in 30 mM d‐acetate buffer at pH 4.0, mixed with MTSL (Toronto Research Chemicals, Ontario, Canada), and incubated for approximately 12 hr at 277 K. The protein solution was filtered and concentrated using Microcon (Merck Millipore, Burlington, Massachusetts). The final protein solution was adjusted to a concentration of 0.3 mM for PRE experiments and to 1.0 mM for NOESY experiments in 30 mM d‐acetate buffer (pH 4.0) containing 12% ²H₂O. The spin‐labeled MTSL was reduced with twofold excess ascorbic acid relative to the protein concentration.

4.1.2. NMR measurements and analyses

Nuclear magnetic resonance experiments were performed on an AVANCE3‐800 (¹H, 800.34 MHz) or DRX‐600 (¹H, 600.23 MHz) spectrometer (Bruker, Coventry, UK and Bruker, Billerica, Massachusetts, respectively) using a pressure‐resistant NMR cell (Daedalus Innovations, Aston, Pennsylvania) and a TXI probe (Bruker and Bruker, respectively). Triple‐ and double‐resonance NMR experiments, that is, CBCA(CO)NH, HNCACB, HNCA, HN(CO)CA, HNCO, HN(CA)CO, HBHA(CO)NH, HCC(CO)NH, CC(CO)NH, HCCH‐TOCSY, CCH‐TOCSY, and ¹⁵N edited TOCSY‐HSQC for signal assignments were performed on uniformly ¹³C/¹⁵N‐doubly labeled Q41N using a DRX‐600 spectrometer at 278 K and 2,500 bar. ¹H/¹⁵N HSQC, ¹H/¹³C HSQC, and simultaneous ¹³C/¹⁵N‐edited NOESY spectra were recorded on uniformly ¹³C/¹⁵N‐doubly labeled Q41N using an AVANCE3‐800 spectrometer, while ¹⁵N‐longitudinal and transverse relaxation experiments were recorded on uniformly ¹⁵N‐labeled Q41N using a DRX‐600 spectrometer. ¹H/¹⁵N HSQC and ¹H/¹³C HSQC spectra at different pressure were collected with a delay of 1 s. The simultaneous ¹³C/¹⁵N‐edited NOESY consisting of 308 complex t ₁ points, 64 complex t ₂ points, and 2048 real t ₃ points used a 100 ms mixing time. The spectral widths were 13.95 ppm (¹H, F ₁), 26.0 ppm/23.0 ppm (¹⁵N/¹³C, F ₂), and 13.95 ppm (¹H, F ₃). A 256 complex × 2,048 real data matrix was acquired for 10 different T ₁ delays: 10, 20, 40, 70, 100, 200, 400, 600, 900, and 1,200 ms. A CPMG type pulse sequence was used to measure ¹⁵N transverse relaxation time constants, T ₂. A 256 complex × 2,048 real data matrix was acquired for 10 different T ₂ delays: 20, 35, 50, 70, 100, 130, 160, 200, 250, and 280 ms. For the ¹⁵N‐T ₁ and ‐T ₂ experiments, the spectral widths were 29.0 ppm (¹⁵N, F ₁) and 14.56 ppm (¹H, F ₂). Schemes for the ¹⁵N‐T ₁ and ‐T ₂ experiments are described in the literature.37 To measure PRE, ¹H/¹⁵N HSQC spectra of the MTSL‐bound ubiquitin K33C/Q41N and Q40C/Q41N were collected for the oxidized and reduced forms of MTSL at 278 K and 1 or 2,500 bar using a DRX‐600 spectrometer. A 128 complex × 2,048 real data matrix was acquired with spectral widths 29.0 ppm (¹⁵N, F ₁) and 14.56 ppm (¹H, F ₂). NMR data were processed using Topspin version 3.2 (Bruker), NMRPipe,38 NMRViewJ,39 and KUJIRA.40 The signal assignments of Q41N at pH 7.2, 298 K, and 2,500 bar (BMRB: 11505) were used as reference. Temperature and pH titration experiments were performed to obtain resonance assignments of Q41N at pH 4.0, 278 K, and 2,500 bar.

The ratio of the peak heights of a paramagnetic sample to those of a diamagnetic sample, that is, I _para/I _dia is equal to

$$\frac{I_{\text{para}}}{I_{\mathrm{dia}}}=\frac{R_{2\mathrm{int}}\exp \left(-R_{2\mathrm{PRE}}t\right)}{R_{2\mathrm{int}}+R_{2\mathrm{PRE}}}$$ (1)

where R _2int and R _2PRE are the intrinsic transverse relaxation rate and the PRE effects on transverse relaxation rate, respectively, for each amide proton, and t is the total INEPT evolution time of the HSQC (11.5 ms). The R _2int for each amide proton was estimated from the half‐height line‐widths of peaks of the diamagnetic sample. The I _para/I _dia was used to estimate R _2PRE. The R _2PRE was converted into distance using the following equation,

$$r={\left[\frac{K}{R_{2\mathrm{PRE}}}\left(4{\tau }_c+\frac{3{\tau }_c}{1+{\omega }_H^2{\tau }_c^2}\right)\right]}^{1/6}$$ (2)

where r is the distance between the unpaired electron and nucleus, ω _H is the Larmor frequency of proton, K is 1.23 × 10⁻³² cm⁶ s⁻², and τ _c is the correlation time given as 1/τ _c = 1/τ _r + 1/τ _e (τ _r the rotational correlation time of the electron‐nucleus vector, τ _e the electron spin‐relaxation time).33 Because τ _e (>10⁻⁷ s) of the unpaired electron is much longer than τ _r, τ _c equals approximately τ _r. τ _c (≈τ _r) was estimated assuming no anisotropy of motion using the equation:

$${\tau }_c\approx {\tau }_r=\left(1/2{\omega }_N\right)\sqrt{\left(6T_1/T_2-7\right)}$$ (3)

where ω _N is the ¹⁵N frequency in radians s⁻¹.41 In the case of Q41N ubiquitin at the current conditions (i.e., 278 K and 2,500 bar), τ _r values were distributed within 5–20 ns, and the mean of τ _r in the secondary‐structured regions was 8.4 ns. The half‐height line‐widths (Δν[hertz] = R _2int/π) were 15–35 Hz for ¹H, and the mean was 20 Hz. Estimated distances from the I _para/I _dia with different τ _r and R _2int are plotted in Figure S13. In the current case, taking into account the diverse values of τ _r and R _2int and the uncertainty of the MTSL position due to the side‐chain flexibility, three classes of distance restraints between Cβ of K33 (or Q40) and other amide protons were used for structure calculations as described below.

Structure calculations for the ¹³C/¹⁵N‐doubly labeled Q41N were performed with CYANA ver. 3.97. Automated NOE peak assignments were performed by the CYANA program. Technical details and reliability of the program are discussed in the literature.36 Distance restraints were obtained from a simultaneous acquisition ¹³C/¹⁵N‐edited NOESY spectrum with a 100 ms mixing time. Backbone ϕ and ψ angle restraints were derived using TALOS+32: –120° < ϕ < −20° and − 100° < ψ < 0° for an α‐helix, and −200° < ϕ < −80° and 40° < ψ < 220° for a β‐sheet. Distance restraints were established based on the PRE experiments: if I _para/I _dia < 0.15 or 0.85 < I _para/I _dia, it was assumed that the distance between the spin‐label and the nucleus was less than 14.6 Å or more than 18.1 Å, respectively. If 0.15 ≤ I _para/I _dia ≤ 0.85, a distance restraint of 11.7 Å‐23.1 Å was imposed (the lower limit at 0.15‐the upper limit at 0.85; Figure S13).

We used the NMR structure refinement method based on Bayesian inference33 implemented in CYANA ver. 3.97.35 The 20 lowest energy structures from the 1,900 models were selected, and energy minimization was performed with distance and angle restraints using OPALp.42 In order to support the validity of this approach, the distance calculated from the PRE and distances determined from 20 structural models were compared.

AUTHOR CONTRIBUTIONS

The manuscript was written by R.K and M. W. NMR measurements and structural determination were performed by T.W., T.I., S.K., N.B., M.W., and R.K. All authors reviewed the manuscript.

CONFLICT OF INTEREST

The authors declare no potential conflict of interest.

DATA DEPOSITION

Twenty energy‐minimized models were deposited in the Protein Data Bank (ID: 6K4I). Chemical shift data and structural restraints are available in the Biological Magnetic Resonance Bank (ID: 36241).

Supporting information

Appendix S1: Supporting Information

Wakamoto T, Ikeya T, Kitazawa S, Baxter NJ, Williamson MP, Kitahara R. Paramagnetic relaxation enhancement‐assisted structural characterization of a partially disordered conformation of ubiquitin. Protein Science. 2019;28:1993–2003. 10.1002/pro.3734