Hsp70 biases the folding pathways of client proteins

Abstract

The 70-kDa heat shock protein (Hsp70) family of chaperones bind cognate substrates to perform a variety of different processes that are integral to cellular homeostasis. Although detailed structural information is available on the chaperone, the structural features of folding competent substrates in the bound form have not been well characterized. Here we use paramagnetic relaxation enhancement (PRE) NMR spectroscopy to probe the existence of long-range interactions in one such folding competent substrate, human telomere repeat binding factor (hTRF1), which is bound to DnaK in a globally unfolded conformation. We show that DnaK binding modifies the energy landscape of the substrate by removing long-range interactions that are otherwise present in the unbound, unfolded conformation of hTRF1. Because the unfolded state of hTRF1 is only marginally populated and transiently formed, it is inaccessible to standard NMR approaches. We therefore developed a (1)H-based CEST experiment that allows measurement of PREs in sparse states, reporting on transiently sampled conformations. Our results suggest that DnaK binding can significantly bias the folding pathway of client substrates such that secondary structure forms first, followed by the development of longer-range contacts between more distal parts of the protein.

Keywords: Hsp70; PRE; excited states; molecular chaperones; protein folding.

Fig. 1.

Using PREs to compare long-range interactions in DnaK-bound and uw-hTRF1. (A) Cartoon representation of the structure of hTRF1 [Protein Data Bank (PDB) ID code 1BA5] (16). The DnaK binding site is highlighted in blue, and the sidechains of the ILL residues at its core are shown as sticks. The site of attachment of the nitroxide spin label (residue 52) is shown as a gray sphere. (B) PREs can be measured directly on DnaK-bound hTRF1 (Right), but not on uw-hTRF1 because it is a sparsely populated and transiently formed state that exchanges slowly with native (ground) hTRF1.

Fig. 2.

¹H CEST-based method for measuring PREs in excited protein states in slow conformational exchange with a ground state. (A) ¹H-CEST pulse scheme that has been used. Narrow and wide filled bars correspond to 90° and 180° pulses, respectively. Open shapes are water selective 90° pulses, whereas striped 180° pulses are of the composite variety (57), 90°_x-180°_y-90°_x. Central to the experiment is the delay T_EX during which a weak ¹H radio frequency field is applied to longitudinal order, ultimately generating the CEST profiles. Further details are given in SI Text. (B) Representative CEST profiles of L7 (Left) and E10 (Right) of uw-hTRF1 with the nitroxide spin label at position 52 in reduced (Upper) and oxidized (Lower) forms. Ratios of intensities of cross-peaks in the presence (I) and absence (I_o) of delay T_EX are plotted along the y axis as a function of the position of the weak ¹H B₁ field along x. (C) Correlation between ¹H CEST-derived and directly measured R₂ values of amide protons in the native (ground) state of hTRF1 when the spin-label is reduced (Left) or oxidized (Right). The equation for the best-fit line (solid line) is indicated on the plot.

Fig. S1.

Effect of cross-relaxation on CEST profiles acquired using longitudinal order, 2I_zN_z (A) and longitudinal magnetization, I_z (B). CEST profiles were simulated as described in SI Text including cross-correlated relaxation, with η_xy = 4.5 s^-1 and η_z = 0.1 s⁻¹. The ground (excited) state peak resonates at 0 (1.3) ppm, whereas the chemical shift of the proximal ¹H spin S is at −1.5 ppm (position of NOE dips). Cross-relaxation in the simulation originates from a single proton placed 2.75 Å away from the probe proton. Cross-relaxation values were determined for protein rotational correlation times varying from 4 to 24 ns and CEST profiles simulated for a B₁ field of 25 Hz and T_EX = 125 ms. Note that the simulations for I_z (B) assumed equilibrium values of longitudinal magnetization for spins I and S at the start of the CEST element, as would be expected for an experiment in which T_EX precedes t₁ or where ¹⁵N chemical shifts of the one-bond coupled nitrogens are degenerate, for the case where the CEST delay follows t₁. Baselines in A decrease with increasing σ (or correlation time, τ_C) because the auto-relaxation rate of 2I_zN_z increases with correlation time: Eqs. S5 and S7. In contrast, baselines do not change in B because the initial conditions include equilibrium values of both I_z and S_z (that are assumed to have equal relaxation rates). Thus, magnetization lost from I to S due to dipolar exchange is replenished exactly from magnetization gained from the transfer from S to I.

Fig. S2.

¹H-¹⁵N HSQC spectrum of ²H/¹⁵N K52C-tempol hTRF1 with the nitroxide spin label in the reduced form, acquired at 600 MHz, 35 °C, pH 6. Resonance assignments are indicated alongside the peaks.

Fig. S3.

¹H CEST profiles for (A) reduced and (B) oxidized K52C-tempol hTRF1 acquired at four different B₁ field strengths using the pulse scheme of Fig. 2A.

Fig. S4.

(A) Populations of the excited state (p_E) and exchange rates (k_ex) for reduced (green) and oxidized (red) K52C-tempol hTRF1 obtained by bootstrapping fits of ¹H CEST data to a two-state model as described in SI Text. Differences in exchange parameters are likely due to slight differences in buffer conditions because it is known that the exchange parameters for hTRF1 are very sensitive to even slight variations in the composition of the buffer. (B) Differences between ground and excited state chemical shifts (Δϖ_GE) correlate very well between oxidized and reduced samples showing that the excited states in both samples are identical. (C) ¹⁵N chemical shifts of the excited state of WT hTRF1 (x axis) correlate well with the corresponding values for K52C-tempol hTRF1 (y axis). The solid line in B and C is y = x. The outlier corresponding to H32 may be the result of the differences in pHs of the two samples (K52C: pH 6; WT: pH 6.8).

Fig. S5.

Residue-specific ¹H R₂ values for ground (A and C) and excited (B and D) states of K52C-tempol hTRF1 in the reduced (A and B) and oxidized (C and D) forms of the spin label, derived from fitting ¹H CEST data globally to a two-state model of chemical exchange. (E) PREs measured in ground and (F) excited states of hTRF1 are independent of concentration. Correlation between R₂^G and R₂^E in concentrated (800 μM) and dilute (400 μM) hTRF1 obtained from ¹H CEST. The solid line in E and F is y = x.

Fig. S6.

Sensitivity of ground (A) and excited (B) state dips to changes in R₂^G and R₂^E, respectively. (A) R₂^G is varied from 0 (blue) to 100 (light green) s⁻¹ keeping R₂^E constant (11 s⁻¹). (B) R₂^E is varied from 0 to 100 s⁻¹ keeping R₂^G fixed at 11 s⁻¹.

Fig. S7.

Sensitivity of CEST-derived R₂^G and R₂^E rates to errors in B₁ calibration. (A and C) Ground and (B and D) excited state R₂ values obtained with correct B₁ in the fit (x axis) plotted against corresponding values on the y axis derived by missetting the B₁ field 3% higher (A and B; red), 3% lower (A and B; blue), 6% higher (C and D; red), or 6% lower (C and D; blue) than the correct value. The black line is y = x in all panels, whereas colored lines are best fits of the data (circles) to a straight line whose equation is indicated.

Fig. S8.

(A) Ground and (B) excited state ¹H R₂ rates and (C) Δϖ_GE values obtained by fitting simulated ¹H CEST profiles correlate well with input values, showing that R₂ can be obtained robustly via ¹H CEST. Differences between ground state output and input R₂ values plotted as percentage of input rates show that the deviations in ¹H R₂ values roughly correlate inversely to Δϖ_GE (D). In A–E, the circles denote averages over the 500 simulations performed, and error bars are 1 SD from the mean. Correlation between input (x axis) and fit (y axis) values of R₂^G (F) and R₂^E (G) where data simulated using nonzero values of ${\eta }_{H,xy}^i$(from 0 to 4.5 s⁻¹) and ${\eta }_{H,z}^i$(0 to 0.1 s⁻¹) are fit to a model in which ${\eta }_{H,xy}^i$ and ${\eta }_{H,z}^i$ are fixed to 0.

Fig. 3.

Residue-specific PRE values of native (ground, A) and water unfolded (excited, B) hTRF1. The secondary structure of hTRF1 is shown above each plot and the DnaK binding site is indicated in magenta.

Fig. 4.

Comparisons of long-range interactions within DnaK-bound hTRF1, uw-hTRF1, and uu-hTRF1. (A) PRE values in the DnaK-bound (blue), uw- (green), and uu- (black) states of hTRF1. (B) Secondary structural propensities (58) of DnaK-bound (blue) and uu- (black) hTRF1. In both panels, the secondary structure of hTRF1 is indicated above the plots with the DnaK-binding site denoted in magenta.

Fig. S9.

The DnaK-bound conformations of WT and K52C-tempol hTRF1 are identical. (A) ¹H-¹⁵N HSQC spectrum of 150 μM ²H/¹⁵N K52C-tempol hTRF1 with the nitroxide spin label in the reduced form, containing 300 μM ²H ADP-DnaK, acquired at 600 MHz, 35 °C, pH 6. Resonance assignments are indicated alongside the peaks. Green-colored peaks are those that have been aliased in F₁. Comparison of ¹⁵N (B) and ¹H (C) chemical shifts of ADP-DnaK–bound WT (13) and K52C-tempol hTRF1.

Fig. S10.

¹H-¹⁵N HSQC spectrum of 670 μM ²H/¹⁵N K52C-tempol hTRF1 with the nitroxide spin label in the reduced form, unfolded in 3.5 M urea, acquired at 600 MHz, 35 °C, pH 6. Resonance assignments are indicated alongside the peaks. Green-colored peaks are those that have been aliased in F₁.

Fig. 5.

A model for DnaK-dependent folding and disaggregation. (A) DnaK-bound hTRF1 has local secondary structure but lacks long-range tertiary interactions, unlike uw-hTRF1, which has contacts between helix 3 and helices 1 and 2. Consequently, folding pathways of client proteins from the DnaK-bound state in vivo may be different from in vitro refolding of the unfolded state present in water, as DnaK binding biases the energy landscape of the bound substrate. (B) The presence of multiple DnaK molecules on the client protein and their asynchronous release from the client allows local secondary and tertiary structure to develop before the establishment of long-range contacts. DnaK binding thus appears to favor a framework model for client folding, rather than a collapse-dependent nucleation-condensation mechanism. (C) The ability of DnaK to disrupt long-range contacts in the client substrate is of potential importance in the role that it plays in refolding, by converting misfolded substrates into a folding-competent conformation, and in disaggregation where DnaK unfolds the substrates and presents them to Hsp104/ClpB.