Microsecond Backbone Motions Modulate the Oligomerization of the DNAJB6 Chaperone

Abstract DNAJB6 is a prime example of an anti‐aggregation chaperone that functions as an oligomer. DNAJB6 oligomers are dynamic and subunit exchange is critical for inhibiting client protein aggregation. The T193A mutation in the C‐terminal domain (CTD) of DNAJB6 reduces both chaperone self‐oligomerization and anti‐aggregation of client proteins, and has recently been linked to Parkinson's disease. Here, we show by NMR, including relaxation‐based methods, that the T193A mutation has minimal effects on the structure of the β‐stranded CTD but increases the population and rate of formation of a partially folded state. The results can be rationalized in terms of β‐strand peptide plane flips that occur on a timescale of ≈100 μs and lead to global changes in the overall pleat/flatness of the CTD, thereby altering its ability to oligomerize. These findings help forge a link between chaperone dynamics, oligomerization and anti‐aggregation activity which may possibly lead to new therapeutic avenues tuned to target specific substrates.

rates and their errors were extracted by fitting single exponentials to the resulting data for each spin lock field strength. 3 JHN-Hα couplings were measured using the ARTSY-based method as described previously. [11] The value of 3 JHN-Hα is calculated as c[cos -1 (IA/IB)ptd) where IA and IB are the intensities in reference and attenuated (where the coupling is active for td = 30 ms) spectra, respectively, and c is a scale factor that depends upon R1 and td, and was estimated to have a value of 1.04.

Measurement of 3 JHN-Hα couplings.
Measurement of kZZ rates. kZZ rates were measured using the DÉCOR experiment of Skrynnikov and Ernst [12] at 298 K. Briefly, after generation of HzNz two-spin order using a standard INEPT transfer, a delay is inserted (τDECOR) during which exchange of the amide proton with solvent serves to break up the correlation with 15 N, resulting in signal loss. T1 relaxation also takes place during τDECOR but R1,HzNz rates are reasonably assumed to be the same for the WT and T193A CTDs, and therefore the differences in kZZ rates shown in Figure 4D reflect differences in water exchange rates between the two proteins. As a control, we also performed hydrogen exchange experiments in which the water resonance was inverted selectively that also revealed elevated hydrogen exchange rates for residues in the β2 strand.
Structure calculation and refinement. Simulated annealing calculations were carried out in XPLOR-NIH [13] to refine the structure of the T193A CTD mutant, starting from the refined WT CTD structure (PDB ID: 7JSQ). A total of 222 unambiguously assigned NOEs were used, with 8 of these involving protons of A193. NOEs were classified as strong (<2.5 Å), medium (<3.5 Å) or weak (<4.5 Å), with an extra 1 Å added to the restraint if one of the two protons involved originated from a methyl group. The target function included terms for the experimental NOE-derived interproton distance restraints, 3 JHN-Ha couplings, and Talos-derived [14] backbone torsion angle restraints. Two statistical potentials of mean force were also included in the target function: torsionDB, [15] to ensure stereochemically reasonable backbone and sidechain torsion angles, and HBPot, [16] to optimize hydrogen bonding interactions. Control calculations showed that HBPot does not bias the resulting structures but does improve structure quality. All simulated annealing calculations were carried out in torsion angle space. The first stage in the structure calculation consisted of 5000 steps of energy minimization, followed by simulated annealing torsion angle dynamics with all the potential terms active. During the high temperature phase (T = 3000 K), the NOE terms were underweighted to allow sampling of a large region of conformational space, and then geometrically increased during the cooling phase (3000 to 25 K).

TwistPot.
To calculate the backbone twist from atomic coordinates, we use four consecutive Cα atoms and define the midpoints between Ca(i)-Ca(i+1), Ca(i+1)-Ca(i+2), and Ca(i+2)-Ca(i+3) as M1, M2, and M3, respectively. P is the midpoint between M1-M2, and Q is the midpoint between M2-M3. The calculated twist angle (Dcalc) is then defined as the dihedral angle between Ca(i+1), P, Q, and Ca(i+2) (see Figure S6A), making sure that the twist angle of the trans state is 0°. The pseudo-potential that was incorporated into XPLOR-NIH [13] was therefore defined as: Dtarget is the desired twist angle in degrees, N the number of restraints, and ktwist a force constant (set to a high value of 1000 kcal -1 mol -1 , since N = 1 here). Dtarget of the central β2 strand window was set to a series of values ranging from 4° to 64° (the value observed in the refined T193A CTD structure is ~20°). TwistPot was used in combination with all the experimental (NOE, 3 JHN-Ha, φ, ψ torsion angles), and geometrical and statistical (torsionDB, HBpot) terms, in an effort to introduce backbone twists and investigate the conformational changes needed to accommodate them, while preserving the overall architecture of the βsheet. TwistPot and associated TwistPotTools are freely available at karamanoslab.com/resources.
Global fitting of relaxation-based data to a 4-state kinetic model. Concentration-independent 15 N-CPMG, off-/on-resonance 15 N-R1ρ, and on-resonance 1 HN-R1ρ relaxation dispersion data were globally fit to 4-state model shown in Figure 4A and Scheme S1.
, ∆ * D (the difference in chemical shifts between states A and B, and B and D, respectively) were fixed to their optimised values as described previously [1] , leaving ! , *! , ! A , and ∆ * C as fitting parameters.
Since the A « C interconversion is fast on the chemical shift timescale, ∆ * C (the difference in chemical shifts between states A and C) and $ are correlated and cannot be deconvoluted using R1ρ data alone. The latter, however, provide a very good estimate of ! A . To overcome this issue, we incorporated 15 N-CPMG relaxation dispersion data, which, for fast exchange rates (>5000 s -1 ), decrease monotonically as a function of the pulse frequency ( Figure S5B). Numerical simulations and fits to well-digitised R1ρ and CPMG relaxation dispersion data generated using a 3% population for the excited state ( * ), with ∆ * * = 6 ppm and ! A = 7 s -1 , are shown in Figure S5. Off-resonance R1ρ data at various spinlock field strengths can be fit within error, even if * is fixed at a high value of 25% ( Figure S5A). The fit to the CPMG relaxation dispersion curves is also reasonable, although the optimized value of !,,-./ A is unrealistically low (~1 s -1 ) in comparison with !,0&1 A , which is much closer to the correct value of 7 s -1 ( Figure S5A). Restraining results in a much poorer fit to the CPMG relaxation dispersion data and thus provides a useful upper limit on * . Thus, when fitting actual experimental data, !,,-./ A was set to be within ± 2 s -1 of !,0&1 A . In the case of the T193A mutant, very low values for $ were excluded based on unrealistically high values for the chemical shifts of amide protons, which exceed their chemical shift dispersion in the Biological Magnetic Resonance Bank (BMRB). Using this strategy, values of $ and ∆ * C are relatively well-defined for both WT CTD and the T193A mutant ( Figure S5C). The 15 N-CPMG, 15 N-R1ρ, and 1 HN-R1ρ relaxation dispersion data were fit simultaneously by minimizing the sum of square differences between observed (obs) and calculated (calc) values, using an in-house Python script that employs the lmfit module. [17] Data avaliability. The 10 lowested energy structural models of T193A CTD were deposited in the ProteinDataBank, PDB ID 7QBY. T193A CTD backbone chemical shifts were deposised in the BMRB, ID 34686. NMR relaxation data and processing scripts were deposited in digital format on figshare (doi:       Figure 4D. kZZ rates were obtained using the DÉCOR experiment [10] at 298 K and 600 MHz as a function of residue number for WT (grey) and T193A (red) CTD.