Local unfolding of the HSP27 monomer regulates chaperone activity

The small heat-shock protein HSP27 is a redox-sensitive molecular chaperone that is expressed throughout the human body. Here, we describe redox-induced changes to the structure, dynamics, and function of HSP27 and its conserved α-crystallin domain (ACD). While HSP27 assembles into oligomers, we show that the monomers formed upon reduction are highly active chaperones in vitro, but are susceptible to self-aggregation. By using relaxation dispersion and high-pressure nuclear magnetic resonance (NMR) spectroscopy, we observe that the pair of β-strands that mediate dimerisation partially unfold in the monomer. We note that numerous HSP27 mutations associated with inherited neuropathies cluster to this dynamic region. High levels of sequence conservation in ACDs from mammalian sHSPs suggest that the exposed, disordered interface present in free monomers or oligomeric subunits may be a general, functional feature of sHSPs.


Introduction
Small heat-shock proteins (sHSPs) are a class of molecular chaperones present in all kingdoms of life and exhibit diverse functionality, from modulating protein aggregation to maintaining cytoskeletal integrity and regulating apoptosis 1 . Despite molecular masses in the range of 10-40 kDa, sHSPs assemble into large, heterogeneous oligomers 2 whose constituent monomers and dimers typically exchange between oligomers 3 . The chaperone activities of many sHSPs have been characterised in vitro 4 , but the active sHSP species remains unclear, with large oligomers 5,6 , small oligomers 7,8 , and dimers 9 all implicated.
The most abundant sHSP in humans, HSP27 (or HSPB1), is systemically expressed under basal conditions and upregulated by oxidative stress 10 , during aging 11 , and in cancers 12 and protein deposition diseases 13 . Numerous mutations in HSP27 have been linked to different neuropathies, including distal hereditary motor neuropathy (dHMN) and Charcot-Marie-Tooth (CMT) disease 14,15 , the most commonly inherited neuromuscular disorder. These maladies are themselves linked to oxidative stress 16,17 , and recent studies have indicated that the reducing environment of the cytosol progressively transitions to an oxidising environment over the lifetime of an organism 18,19 .
HSP27 is directly sensitive to the intracellular redox state via its lone cysteine residue (C137), which controls dimerization by forming an intermolecular disulphide bond in vivo 20 . This cysteine is highly conserved in HSP27 orthologs but not found in other mammalian sHSPs 21 , implying that it plays an important and specific role in regulating function. Accordingly, the presence of this disulphide bond impacts on the activity of HSP27 in vitro [22][23][24] , and on the resistance of cells to oxidative stress 20,21,25,26 . Intriguingly, variants of HSP27 that have an increased tendency to form monomers display hyperactivity both in vitro and in vivo 27,28 . More generally, sHSP monomers can mediate the subunit exchange between the oligomeric assemblies 3,29 . However, because they are typically present at low abundance in solution, no sHSP monomer has yet been characterised structurally.
Obtaining high-resolution information describing HSP27 is challenging, as it assembles into a polydisperse ensemble of inter-converting oligomers ranging from approximately 12 to 36 subunits [30][31][32] of average molecular mass of ca. 500 kDa. However, removal of the C-terminal region (CTR) and N-terminal domain (NTD) leaves a conserved ~80-residue, 'α-crystallin' domain (ACD) that does not assemble beyond a dimer (Fig. 1a). The subunits in the dimer adopt an immunoglobulin-like fold, and assemble through the formation of an extended β-sheet upon pairwise association of their β6+7 strands [33][34][35][36] . Under oxidising conditions, the dimer interface in HSP27 is reinforced by an intermolecular disulphide bond involving C137 from adjacent subunits centred on a two-fold axis [33][34][35] .
Based on evidence from the closely related sHSP, αB-crystallin 37 , the ACD is likely structurally similar in the context of the full-length oligomeric protein and in its isolated dimeric form. Moreover, as the excised ACD of αB-crystallin has been shown to be incorporated into full-length sHSP oligomers 38 and display potent chaperone activity in vitro 34,39 , it appears that important aspects of sHSP function are encoded within this domain.
Here, we have employed an integrative biophysical approach to interrogate the impact of redox-induced changes to the structural features of HSP27 and its excised ACD (cHSP27). We find that reduced HSP27 more effectively prevents protein aggregation than its oxidised counterpart.
However, neither the distribution of HSP27 oligomers, nor the conformations and fast dynamics of the CTR and cHSP27 dimer vary appreciably with oxidation state. Rather, upon cleavage of the disulphide bond, the release of monomers into solution is responsible for the observed differences in chaperone activity upon reduction. To interrogate the structure of the free, but sparsely populated, monomers we have used a combination of Carr-Purcell-Meiboom-Gill (CPMG) relaxation dispersion (RD) and high-pressure solution-state nuclear magnetic resonance (NMR) spectroscopy methods. Our data reveal that monomeric cHSP27 becomes highly dynamic and disordered in the region that previously constituted the dimer interface. While we find the monomer to be highly chaperone-active in vitro, we demonstrate that increasing the abundance of the monomer results in a heightened tendency for uncontrolled self-aggregation. The importance of the unstructured region in this delicate balance between function and malfunction can be linked by mutations in HSP27 that are associated with hereditary neuropathies, which mainly cluster to the disordered region of the monomer.

Reducing HSP27 affects chaperone activity by increasing the amount of free monomer
We first examined full-length HSP27 to analyse redox-dependent changes to its oligomeric distribution. Native mass spectra of reduced and oxidised HSP27 were highly similar, with overlapping signals in the 5,000-15,000 m/z region (Fig. 1a), consistent with previous data 30,31 This reveals that HSP27 assembles into large, polydisperse oligomers with similar distributions under both conditions ( Supplementary Fig. 1). We also observed monomeric and dimeric HSP27 in the spectra of both oxidised and reduced forms, with a significant increase in the population of free monomer upon reduction (Fig. 1a).
To confirm that dissociation of the dimers, rather than modulation of the oligomers, is the major consequence of reduction, we used solution-state NMR to examine the CTR, which can mediate the assembly of sHSPs 40,41 . As HSP27 oligomers have an average mass of ca. 500 kDa, only the disordered CTR from 15 N labelled HSP27 can be observed in a 2D 1 H-15 N heteronuclear single quantum coherence (HSQC) NMR spectrum [42][43][44] . To probe the local dynamics in this region quantitatively, we recorded NMR spin relaxation experiments that characterize motions on the ps-ns timescale ( Supplementary Fig. 1). No significant differences in the conformation or fast backbone motions were detected between oxidised and reduced forms of HSP27. Our combined native MS and NMR data on the polydisperse ensemble populated by HSP27 demonstrate that the primary impact of reduction is the release of free monomers.
To ascertain whether the presence of monomers impacts on chaperone function, we used the model substrate citrate synthase (CS) 4,45 to probe the activity of HSP27 in vitro. The aggregation of CS alone was suppressed upon the addition of reduced HSP27 at low concentrations (0.5 M) (Fig.   1b). The reduced sample contains a mixture of monomers and dimers, whereas the in the oxidised sample the sub-oligomeric species are predominantly dimeric 31 . These data therefore suggest that monomerisation regulates the chaperone activity of HSP27, rendering it more effective at suppressing aggregation in vitro. Fig. 1: Reduction of HSP27 releases monomers from the polydisperse oligomers. (a) Domain architecture of the human molecular chaperone HSP27, which forms polydisperse oligomers that reach >500 kDa. Native mass spectra collected at 25 M total monomer concentration for both oxidised (red) and reduced (blue) HSP27 reveal the formation of polydisperse oligomers. More monomers are present in the reduced sample, but the oligomeric distributions are highly similar ( Supplementary Fig. 1). (b) The chaperone activity of HSP27 was assayed by monitoring the increase in light scattering at 340 nm of 10 M of a thermo-sensitive substrate (CS) in the presence or absence of 0.5 M reduced (blue, 5 mM 2-mercaptoethanol, BME) or oxidised (red) HSP27. The average and ± standard deviation of two replicates are indicated.

HSP27 monomers are potent chaperones in vitro that readily self-aggregate
To examine these redox effects on the monomer:dimer equilibrium in the absence of oligomeric forms of HSP27, we turned to the truncated form, cHSP27 (Fig. 2a, Supplementary Fig.   2), which forms dimers whose structures are essentially independent of oxidation state [33][34][35] . In addition to the wild-type sequence, we produced two disulphide-incompetent variants, C137S and H124K/C137S 46 . Native mass spectra of these constructs at 5 M revealed pure dimers (oxidised), monomers (H124K/C137S), or mixtures of the two (C137S, reduced) (Fig. 2b). This redox-dependent monomerisation is consistent with the major difference we observed in the oligomeric distributions of the full-length HSP27 (Fig. 1a). To compare the functional activity of the cHSP27 dimer and monomer, we used the model protein -lactalbumin (Lac) 34 . At 70 M, where C137S exists predominantly as a dimer and H124K/C137S a monomer, the double mutant revealed enhanced protection against Lac aggregation ( Fig. 2c), further indicating that the monomers are particularly active chaperones. Interestingly, at neutral pH, the H124K/C137S monomer showed a greater propensity than C137S to self-aggregate, forming large amorphous aggregates (Fig. 2d, Supplementary Fig. 3e, 3f) that did not display the Thioflavin-T binding characteristic of amyloid fibrils ( Supplementary Fig. 3d). In addition, the thermal stabilities of both reduced cHSP27 and C137S were markedly reduced at low concentrations that favour monomer release ( Supplementary Fig. 3). Taken together, these results suggest that, in addition to being more active, the cHSP27 monomer is also kinetically unstable and aggregation prone.

Dynamics at the dimer interface are redox sensitive
To obtain insight into the structural rearrangements that trigger the enhanced chaperone activity and aggregation propensity of the cHSP27 monomer, we turned to solution-state NMR. 2D 1 H-15 N HSQC NMR spectra of oxidised and reduced cHSP27 were recorded at 1 mM, a concentration that favours dimer formation. The spectra were highly similar, a finding that is consistent with the 2.5-Å backbone RMSD between the two forms ( Fig. 3a). A quantitative analysis confirmed that both the secondary structure and hydrogen-bonding network ( Supplementary Fig. 2) were consistent with published structures [33][34][35] . Moreover, 15 N spin relaxation experiments that probe backbone motions revealed that the ps-ns dynamics in cHSP27 were essentially unaltered by changes in redox state ( Supplementary Fig. 4). Similarly, we confirmed that C137S effectively mimics the reduced form, as their NMR spectra revealed very similar CSPs to the oxidised form (Fig. 3b), apart for the residues immediately adjacent to the mutation.
Although the structure of the underlying dimer 34,35 and fast backbone dynamics were redoxindependent, when comparing either reduced or C137S to the oxidised form of cHSP27, the signal intensities for residues in the vicinity of the dimer interface were substantially reduced ( Supplementary Fig. 2), with elevated 15 N transverse relaxation rates ( Supplementary Fig. 4). These observations demonstrate that reduction of the disulphide bond leads to dynamics on the s-ms timescale near the dimer interface of cHSP27. Conformational fluctuations between multiple states on this timescale can be characterised using CPMG RD NMR spectroscopy experiments 47,48 , which employ a variable pulse frequency, ν CPMG , to measure effective 15 N transverse (R 2 ) relaxation rates [49][50][51] . The relative populations of the states that are interconverting (p G, p E ), their rate of interconversion (k ex ), and the chemical shift differences (|Δω|) associated with the structural changes can be obtained through quantitative analysis of CPMG RD data. H-15 N HSQC spectra of oxidised (red), reduced (blue), and C137S (green) cHSP27 under identical conditions reveal their similarity. Significant CSPs are indicated with arrows. (b) CSPs for individual residues between reduced (i) or C137S (ii) and the oxidised state reveal differences that localize to the 5 and 6+7 strands. The resonance from C137 was broadened beyond detection in reduced cHSP27 (cyan). Proline residues that do not contribute to this spectrum and unassigned residues are indicated.
(c) Where R ex , approximated here as the difference in R 2,eff at low and high CPMG pulse frequency in 15 N CPMG experiments, is greater than zero, there is motion on the s-ms timescale. Residues in the vicinity of L 5,6+7 show large R ex values in oxidised, reduced, and C137S cHSP27. The motion extends into the β5 and β6+7 strands for reduced and C137S cHSP27. (d) R ex values > 2 s -1 are mapped onto the structure of cHSP27 (PDB 4mjh), indicating that residues with slow dynamics cluster near the dimer interface.
We observed motions on the s-ms timescale in oxidised, reduced, and C137S cHSP27 (Table S1). H124K/C137S was insufficiently stable to perform the CPMG RD measurements (Fig   2d). In reduced cHSP27 and C137S, the dynamics encompassed β5, L 5,6+7 , and β6+7 (  Table   S1). From the CPMG RD data, we calculated a K d for the C137S monomer-dimer equilibrium of 0.5 M (Supplementary Figure 5, Table S1), a result qualitatively consistent with results obtained by native MS (Fig. 2 b). Activation and thermodynamic parameters of the monomer-dimer interconversion were obtained by analysing the variation in CPMG RD data with temperature.
Dissociation of the dimer was endothermic, consistent with a disruption of stabilising interactions, and entropically favoured, consistent with an increase in structural disorder (Table S1). The transition state for dissociation was more disordered than the dimer as evidenced by a positive activation enthalpy and entropy ( Supplementary Fig. 5, Table S1).
Similar to reduced cHSP27 and C137S, oxidised cHSP27 exhibited dynamics in L 5,6+7 ( Fig.   4a iii, Supplementary Fig. 5), but these motions were not observed in β5 or β6+7. Detailed analysis revealed that the motions in L 5,6+7 involve unfolding of the loop, thereby disrupting the intermolecular salt bridge between D129 in L 5,6+7 and R140 in β6+7 from the adjacent monomer (Fig. 4b), an orderto-disorder transition on the s-ms timescale that is independent of oxidation state. In addition to the local unfolding of L 5,6+7 , the oxidized form of cHSP27 showed s-ms motions in the vicinity of residue C137 on a faster timescale, consistent with isomerisation of the disulphide bond 47 ( Supplementary Fig. 5).  15 N CPMG RD experiments quantify s-ms motions in oxidised (red), reduced (blue), and C137S (green) cHSP27. Fitted curves from a global analysis are shown as solid lines. Significant CPMG RD curves were observed in the β5 strand (a), L 5,6+7 (b), and the β6+7 strand (c). Redox-independent motions were observed in L 5,6+7 , which arise from unfolding of the loop, whereas only the noncovalent dimers in C137S and reduced cHSP27 show motions throughout β5 and β6+7. (d i) CPMG RD-derived 15 N chemical shift changes in C137S (||) plotted onto the structure (PDB: 4mjh) revealing significant changes that are localised to L 5,6+7 and the β5 and β6+7 strands. (d ii) The 15 N || values in L 5,6+7 are similar in both C137S and oxidised cHSP27, and correlate with those expected for a transition to a random coil, indicating unfolding of L 5,6+7 that is independent of oxidation state. In L 5,6+7 , D129 forms an intermolecular salt bridge with R140 from an adjacent subunit, and the amide nitrogen from E130 forms a hydrogen bond with the carbonyl of D129 within the same subunit. (d iii) 15 N || values in L 5,6+7 and β6+7 upon monomerisation are compared to the changes expected for random coil formation. The agreement is reasonable, indicating the monomer is substantially disordered in these regions. Outliers are indicated and mainly residue in the end of β6+7.

Partially disordered monomers of cHSP27 characterized by RD NMR
Given the increased chaperone activity of the monomeric ACD (Fig. 2b), we pursued a structural characterisation of the C137S monomer. The 15 N chemical shift differences from the CPMG RD experiments (|Δω|) on C137S report on the structure of the monomeric state. The values that we obtained indicate that residues at the dimer interface (β6+7, L 5,6+7 ) adopt 'random-coil-like' disordered conformations 48 (Fig. 4c, Supplementary Fig. 7d). Consistent with this finding, we observed that H124K/C137S displayed characteristics of a partially disordered protein, as evidenced by its 2D 1 H-15 N HSQC spectrum ( Supplementary Fig. S3a,3b), circular dichroism spectra, and bis-ANS (4,4'-dianilino-1,1'-binaphthyl-5,5'-disulfonic acid) fluorescence. While the majority of the molecule retains its fold upon monomer release, as evidenced by the small |Δω| values, residues in L 5,6+7 and 6+7, the region responsible for the dimer interface, become disordered.

High pressures and acidic conditions enable direct detection of the cHSP27 monomer
While CPMG RD enables a direct characterisation of the HSP27 monomer under near physiological conditions, its low population (ca. 1.5% at 1 mM) renders further high-resolution analysis challenging. We sought to stabilise the monomeric fold. A well-resolved resonance (G116) provided a straightforward marker for distinguishing between the monomeric and dimeric states. Two resonances from this residue were observed in slow exchange at low concentrations for reduced cHSP27 and C137S and the |Δω| between the two resonances (1 ppm) matched the value obtained from the CPMG analysis. Following the intensities of these resonances allowed us to determine a K d for C137S of 0.7 M, a value consistent with the CPMG analysis (Table S1). Similarly, the concentration dependence of the intensity of the G116 monomer and dimer resonances for H124K/C137S revealed an increase in the K d by 3 orders of magnitude to ca. 1.1 mM.
To preserve the monomeric form at sufficiently high concentrations to render it amenable for atomic-resolution characterization by NMR spectroscopy, increased hydrostatic pressure 52 was employed. NMR spectra of 200 M C137S were recorded in a baroresistant buffer 53 at pH 7 as a function of pressure from 1 bar to 2500 bar, revealing a shift in the equilibrium from folded dimer at low pressure to entirely unfolded monomeric C137S at high pressure (Fig. 5a, Supplementary Fig. 6) via an intermediate species that was maximally populated at 1600 bar ( Fig. 5b, 5c, Supplementary   Fig. 6). The variation in population of dimer, monomer, and unfolded monomer as a function of pressure was explained quantitatively by a three-state linear equilibrium model (Fig. 5b, Table S1).
Volumetric changes on application of pressure were obtained, together with the equilibrium constant of monomer unfolding, K u , at 1 bar (Table S1), revealing a free energy difference (Δ u G) of 5 ± 0.4 kJ mol -1 between the monomer and unfolded species at 1 bar and pH 7. The K d for dimerisation increased ten-fold at 1600 bar (Fig. 5c, Supplementary Fig. 6). The K d was further increased by three orders of magnitude by moving from pH 7 to 5 at 1 bar (Fig. 5a, Supplementary Fig. 6, 7). We were able to combine the effects in a phosphate buffer whose pH decreases with pressure, to maximally stabilise the monomeric form (Fig. 5b ii).
While C137S monomer aggregated under acidic conditions at elevated protein concentrations, it remained stable up to 100 M at pH 4.1 at 1 bar, a finding supported by NMR translational diffusion measurements ( Supplementary Fig. 7). Triple resonance spectra for high-resolution analysis of the monomer were acquired under these conditions (Fig. 6a). All observable H N , N, C, and CO nuclei were assigned (Fig 6a; Supplementary Fig. 7) and, similar to observations by CPMG RD, the largest CSPs fell in L 5,6+7 and 6+7 (Fig. 6b i, Supplementary Fig. 8). A reasonable correlation was observed (RMSD 1.2 ppm, Supplementary Fig. 7) when the 15 N chemical shifts from CPMG RD acquired at pH 7 were compared to those measured directly at pH 4.1, indicating that the monomer conformation is similar in both cases. Further confirming their similarity, minor resonances from the monomeric protein could be observed in a sample of C137S at 20 M at pH 7, whose chemical shifts were close to the values obtained directly under acidic conditions (Fig. 6a, Supplementary Fig. 7). (a) NMR spectra of C137S, focusing on residue G116 as a function hydrostatic pressure, pH, and concentration. Shown here are (i) increasing pressures at pH 7 in a baroresistant buffer (i), increasing pressures in phosphate buffer at pH 6.8 at 1 bar, with the pH of phosphate buffer varying with pressure and decreasing by ca. 1 unit over 2 kbar, (ii), decreasing pH at 1 bar (iii), and decreasing concentration at pH 7 at 1 bar (iv). Resonances from the dimer (green), monomer (purple), and unfolded (black) state are readily distinguishable. Decreasing concentration and increasing either pH or pressure favour the monomer over the dimer. At pressures greater than approximately 1.5 kbar, the unfolded form becomes the principally populated state. (b) Variations in NMR signal intensities with pressure from four residues that were unambiguously assigned in all three conformations were well explained (solid lines) by the quantitative 3-state equilibrium model shown. (c) The K d for dimerisation is shown as a function of pressure (black), pH (purple), and a combination of the two, acquired in phosphate buffer, where the pH of the buffer varies with pressure (grey). (d) The three-state mechanism of C137S dimer dissociation, where pH and pressure favour the partially disordered monomer, before ultimate the monomers completely unfold.

Structural and dynamical characterisation of the partially disordered cHSP27 monomer
To characterise the monomeric state of C137S structurally, we used the observed chemical shifts to determine -strand formation in the C137S dimer and monomer using the chemical shift 54 and random coil 55 indices (CSI, RCI, Fig. 6b ii, Supplementary Fig. 8). This analysis confirmed that, while the disordered L 5,6+7 spans from Q128 to Q132 in the dimer, it is substantially elongated in the monomer, running from K123 to S137, thereby shortening the 5 and 6+7 strands.
Finally, { 1 H}-15 N heteronuclear nuclear Overhauser effects (hetNOEs) for the monomer and the dimer were recorded, allowing direct comparison of fast backbone motions on the ps-ns timescale 56 . In the monomer, residues including L 5,6+7 , the C-terminal portion of 5, and the Nterminal portion of 6+7 were highly dynamic ( Fig. 6b iii, Supplementary Fig. 8), consistent with the random-coil-like chemical shifts observed in this area by CPMG at pH 7 and directly at pH 4.1. The rigid dimer interface in cHSP27 partially unfolds and becomes highly dynamic in the monomeric state. The mutations cluster to the regions that become solvent exposed upon monomer formation and tend to lower the charge density in the region. (e) Overlaid dimer structures of human HSP27 (PDB 4mjh), human B-crystallin (PDB 4m5s), bovine A-crystallin (PDB 3l1f), and rat HSP20 (PDB 2wj5) are shown in ribbon format for one subunit of each dimer, and indicate the highly conserved fold of vertebrate ACDs. The second subunit of HSP27 is shown in cartoon format. Highly conserved residues among human sHSPs (HSPB1-HSPB6) are shown as spheres with the same color format as (f) The combined results from the CPMG RD and high pressure NMR experiments allow us to propose a hierarchical mechanism for monomer formation. The oxidised, reduced, and C137S forms of cHSP27 exhibit similar dynamics in L 5,6+7 and form a disordered loop. In the absence of a disulphide bond, this motion in L 5,6+7 propagates, resulting in the eventual unfolding of the β5 and β6+7 strands in the free monomer.

HSP27 monomers partially unfold and become highly active
The function and monomerisation of the molecular chaperone HSP27 is regulated by its redox state through an inter-dimer disulphide bond (Fig. 2) 21,23,24 . Here, we determined the structural basis for this regulation. The oligomeric distribution of HSP27 and ps-ns dynamics within the flexible CTR in the full length protein, and the structure and ps-ns dynamics of isolated dimers from excised ACDs were largely invariant to changes in oxidation state. Reduction of full-length HSP27, however, leads to the release of the free monomers (Fig. 1c). Under conditions that favour the free monomer in the context of both the full-length sequences and core domains, we observe enhanced chaperone activity in vitro (Fig. 1e). Using a combination of CPMG RD and high pressure NMR, we established that the monomeric protein partly unfolds upon dissociation (Fig. 6), such that the region responsible for the rigid interface in the dimer becomes highly dynamic.
These results suggest that the partly disordered monomer of HSP27 is a particularly active chaperone. This observation is similar to two recent findings, where both the acid-induced unfolding of HdeA and HdeB 57 and the oxidation dependent unfolding of HSP33 58 result in relatively potent structural forms for aggregation inhibition. Likewise, the homodimeric chaperone CesAB exists in a molten globule-like state with residual helical structure 59 , but undergoes a disorder-to-order transition upon binding to its substrate 60 . More generally, the plastic nature of intrinsically disordered proteins (IDPs) is thought to aid their ability to bind a wide variety of partners 61-64 via specific, yet transient interactions 65 . It is interesting to speculate that the same mechanism for rapid, promiscuous recognition of binding partners by IDPs is responsible for the heightened activity of partially unfolded chaperones. Interestingly, many of the residues that are unfolded in the HSP27 monomer are charged or polar (Fig. 6d), suggesting that electrostatics may play a role in substrate-recognition, as suggested by bioinformatic analyses 66 . Electrostatically mediated binding has been described for the molecular chaperone Spy 61,62 , but contrasts with the recognition of exposed hydrophobic residues by other molecular chaperones including SecB 67 , trigger factor 68 , DnaK/HSP70 69 , and ClpB/HSP100 70 .

Mechanism of HSP27 monomer release is hierarchical
Our CPMG RD data inform on a specific mechanism for monomer release. The loop L 5,6+7 located at the dimer interface of HSP27 undergoes redox-independent motions on the ms timescale, leading it to unfold (Fig. 3, Table S1). When the disulphide bond is present, the local unfolding does not propagate further. However, in the reduced form and C137S variant, the disordering process extends, on the same timescale, into both the end of the β5 and beginning of the β6+7 strands, effectively destabilizing the interface and facilitating monomer release (Fig. 6). Given the conservation of sHSP residues in L 5,6+7 ( Figure 6, Supplementary Fig. 9) and the previously observed millisecond motions in this region of cABC 71,72 , transient unfolding of L 5,6+7 and the adjacent strands upon monomerisation is likely a common property of mammalian sHSPs.
The process of monomer release and unfolding in HSP27 mirrors the 'docking and locking' behaviour observed in a range of protein-protein molecular recognition processes 73 , in which a relatively rapid encounter complex is formed prior to a slower step that provides additional stabilising interactions, which effectively 'lock' the complex into place 74 . In the case of HSP27, our data suggest that the β6+7 interface forms relatively quickly, before being 'locked' down by the acquisition of structure in L 5,6+7 and the formation of an inter-molecular salt bridge between D129 and R140 (Fig.   4b).

Neuropathy-related HSP27 mutations cluster in and near the unfolded region of the monomer
We analysed the positions of 28 mutations in HSP27 that cause either CMT or dHMN (Fig.   6a), including the 17 missense mutations that reside in the ACD 55 . Our structural and dynamical analysis of the cHSP27 monomer reveals that 16 of the 17 mutations in the ACD are located in the disordered region adjacent to or within L 5,6+7 and the β5 and β6+7 strands. As previously noted 28 , a number of these mutations cluster to the ACD dimer interface. Our NMR data further reveals that the mutations that occur in regions beyond the dimer interface, predominantly fall in regions that are highly disordered in the monomer, suggesting that the behaviour of the monomer is important for understanding the molecular bases of CMT disease and dHMN 75 perhaps in terms of altered activity, abundance or through causing uncontrolled self-aggregation (Fig. 1g).
While certain mutations decrease chaperone activity, some disease-related HSP27 variants that are more monomeric (e.g. R127W, S135F) and exhibit significantly elevated chaperone activity both in vitro and in vivo 27,28 . Conversely, disease-related mutants that did not impact monomerisation have shown either no change (T151I) or a decrease (P182L) in activity. These observations suggest that the position of the monomer:dimer equilibrium is an important factor in neuropathies associated with variants of HSP27.

Disordered monomers are likely a common property of mammalian sHSPs
In light of our findings, we hypothesize that partial unfolding of sHSP ACDs upon monomer release may be a general feature of this class of chaperone. A recent study of cABC showed that the chemical shift changes upon monomer formation are larger for residues located at the dimer interface 71 . We analysed the data and found a strong correlation between 15 N chemical shift changes in cABC upon monomer formation with those expected for the formation of a random coil ( Supplementary Fig. 9). More generally, the dimeric structure and sequence composition of residues at the dimer interface are highly conserved in the mammalian sHSPs HSP27, αB-crystallin, αAcrystallin and HSP20 (Fig. 6e, Supplementary Fig. 9). These results suggest that partial unfolding of monomers upon dissociation is a common property of human sHSPs and that the dimeric building block of sHSP oligomers 36 is assembled first through partly unfolded monomers. Odd-numbered sHSP oligomers 22 , as encountered in both human sHSPs αB-crystallin and HSP27, will have at least one monomer without a complete dimer interface, indicating that the unstructured monomer can also exist within larger oligomers. Interestingly, for the related αB-crystallin we observed that the dimeric form was more chaperone-active than the monomer, particularly for an amyloidogenic substrate 34 .
These differences between the two sHSPs could reflect their contrasting substrate profiles 4 , or multiple binding modes 76 .

Disordered sHSP monomers as flexible sensors for misfolded proteins
In the context of the isolated ACD, our data suggests that increased disorder in the HSP27 monomer renders it a more potent chaperone in vitro. In addition to the partially unfolded ACD, fulllength HSP27 contains a disordered 80-residue NTD 77 and a 28-residue highly flexible CTR 78,79 . The exposed residues from the HSP27 NTD and CTR might also contribute to its mechanism of aggregation inhibition. The inherent plasticity in disordered regions would, in principle, allow for the monomer to sample a wide range of conformational space and thereby facilitate its ability to interact with a diverse set of misfolded target proteins. As the monomer is itself prone to aggregation (Fig.   2d), we speculate that the aggregation-prone contacts in HSP27 are largely responsible for detecting misfolded proteins [80][81][82][83] . In the context of the cell, it is would seem undesirable to have high concentrations of aggregation-prone monomers, making it advantageous to store them in oligomers.
By holding monomers in this 'storage' form, the population of the active but unstable monomeric form is kept both transiently low and highly available 9,41 .
In conclusion, our analysis combining CPMG RD and high-pressure NMR with chaperone and aggregation assays provides the first structural characterisation of the sparsely populated and experimentally elusive monomeric form of HSP27. Although populated at a relatively low abundance in the context of oligomers, the monomeric state is nevertheless particularly active in vitro. Part of the region that forms the rigid interface in the dimer unfolds and becomes highly dynamic in the monomer, with the additional structural plasticity in the monomer rendering it a more effective chaperone. However, the monomer is itself more aggregation-prone, whereby off-pathway selfassembly culminates in uncontrolled aggregation. With most sHSPs forming large oligomers, it appears likely that regulation of their self-assembly is a finely tuned proteostatic mechanism inherent to this class of chaperones.

cHSP27
DNA encoding the residues 84-171 of HSP27 (cHSP27) was inserted into kanamycin-resistant (Kan R ) pET28-b plasmids, which contained an N-terminal hexahistidine (His 6 )-tag followed by a TEV protease recognition site 34 . The residual glycine that remains after TEV protease cleavage corresponds to G84 in the HSP27 amino acid sequence. All media for growth of cells containing pET28b(cHSP27) plasmids contained 30 g/mL of Kan. cHSP27 cultures were grown as described for full length HSP27. Upon inoculation of the 1 L (LB medium) or 500 mL (M9 minimal medium) cultures, A600 was allowed to reach between 0.6 and 0. 8  and concentrated into NMR buffer for the oxidised state at pH 7 unless otherwise specified. 5 mM BME was added when studying the reduced state ( Fig. 2-7, Supplementary Fig. 2-8).

Site-directed mutagenesis in cHSP27
Primers were designed to encode the single-point mutations H124K and C137S in cHSP27. Sitedirected mutagenesis was performed using the QuikChange II Mutagenesis Kit (Agilent), and the correct mutations were verified by DNA sequencing. Protein expression and purification was carried out as above. The final yield of the double mutant, cHSP27(H124K/C137S) (~10 mg/L of E. coli) was significantly lower than either WT cHSP27 or C137S (30-40 mg/L of E. coli).

Chaperone activity assays
Porcine heart citrate synthase (CS; Sigma-Aldrich) was buffer exchanged into NMR buffer M, a concentration where C137S is predominantly a dimer and H104K/C137S a monomer (Fig. 6,   Supplementary Fig. 7).

Native MS
For native MS data acquisition, nanoelectrospray ionization (nESI) experiments were executed according to previously published protocols using instrumental settings optimised to transmit intact protein complexes 85 . A 25 M sample of HSP27 (purified in the absence of reducing agent) was prepared in 200 mM ammonium acetate (pH 6.9) in the presence and absence of 250 M DTT (Fig.  1c, Supplementary Fig. 1). cHSP27 samples (5 M) were prepared in the same buffer with and without 250 M DTT (Fig. 1d).

Thermal denaturation of cHSP27
A nanoDSF instrument (NanoTemper) was used to monitor the intrinsic fluorescence of cHSP27 and C137S as a function of temperature. Capillaries contained ~10 L of cHSP27 that had been prepared in NMR buffer without (oxidised) or with 5 mM BME (reduced). C137S was prepared in NMR buffer without BME. The concentrations of cHSP27 and C137S were 1000, 100, 10, and 1 M. The initial temperature was 20 °C and was set to increase by 1 °C per minute. Fluorescence readings were recorded at 330 and 350 nm, and the melting temperature (T m ) recorded ( Supplementary Fig. 7).

NMR spectroscopy Backbone and side-chain resonance assignments for dimeric cHSP27
All NMR spectroscopy experiments for resonance assignments of cHSP27 and HSP27 at ambient pressure were recorded on a 14.1 T Varian Inova spectrometer equipped with a 5 mm z-axis gradient triple resonance room temperature probe. 2D 1 H-15 N sensitivity-enhanced HSQC spectra 86  When NUS was employed, an exponentially-weighted sampling scheme was employed in the indirect dimensions and time-domain data were reconstructed with MddNMR 88 . All 3D NMR spectra at ambient pressure were acquired at 25 °C, processed with NMRPipe 89 , and visualized with Sparky 90 .
The resultant 1 H N , 15 N, 13 CO, 13 C, and 13 C chemical shifts were analysed with TALOS-N 91 and RCI 92 to respectively estimate the secondary structure and NH order parameters ( Supplementary Fig.   2).

Backbone resonance assignments for monomeric C137S
Resonance assignments for monomeric C137S were obtained on a 14.  Fig. 2). d 1 H/dT values that are more negative than -4.6 ppb/K are more likely to be solvent exposed and hydrogen bonded to water 94 . Residues with temperature coefficients more positive than -4.6 ppb/K are more likely to be involved in intra-or inter-protein hydrogen bonds. However, it should be noted that residues that are near aromatic rings can yield false positives with values less than -4.6 ppb/K 94 .
To provide an independent NMR dataset that also indirectly probes hydrogen bonds, we

N spin relaxation experiments (T 1 , T 2 , NOE) on cHSP27 and full-length HSP27
Standard pulse sequences to measure 15 Fig. 3). Upon completion of these measurements, 5 mM β-mercaptoethanol (BME) was added to the NMR tube, and 15 N R 1 , R 2 , and hetNOE values were recorded on reduced cHSP27 at 600 MHz only ( Supplementary   Fig. 3). Similarly, 15 N R 1 , R 2 , and hetNOE experiments at 14.1 T were recorded on oxidised [U-13 C, 15 N]-labelled full-length HSP27, after which 5 mM BME was added to obtain relaxation measurements on the reduced and oxidised species (Supplementary Fig. 2). For data acquired exclusively at 14.1 T, reduced spectral density functions were calculated as described previously 96 .
For data acquired at multiple magnetic field strengths (oxidised cHSP27), a model-free analysis was conducted (see below). All data sets mentioned here were processed with NMRPipe 89 , visualized with

Model-free analysis of 15 N spin relaxation data from oxidised cHSP27
We analysed the 15  Using the five aforementioned values of 15 N relaxation rates from two magnetic fields, we fit these data using ModelFree4.15 102 to four models: model 1 (optimized S 2 ), model 2 (S 2 and τ e ), model 3 (S 2 and R ex ), and model 4 (S 2 , τ e , and R ex ). No further benefit was obtained when S 2 f was included in the fitting, and thus was not required for this analysis. 15  fit with FuDA 97 to extract peak intensities, which were then converted into R 2,eff values using the following relation: R 2,eff ( CPMG ) = ln(I( CPMG )/I(0) * 1/T relax ), where I( CPMG ) is the intensity of a peak at  CPMG , T relax is the constant relaxation delay of 39 ms that was absent in the reference spectrum, and I(0) is the intensity of a peak in the reference spectrum. Two duplicate  CPMG points were recorded in each dispersion data set for error analysis, and uncertainties in R 2,eff were calculated using the standard deviation of peak intensities from such duplicate measurements. From plots of R 2,eff as a function of CPMG , R ex was estimated by taking the difference of R 2 (54 Hz) and R 2 (950 Hz). The program CATIA 97 was used for analysis (Table S1), which accounts for all relevant spin physics including imperfect 180 o pulses and differential relaxation between spin states 97 , factors that are not accounted for when using closed form analytical solutions 104 . For each combination of concentration and temperature, data were analysed over multiple field strengths for all residues where R ex > 2 s -1 . All residues were assumed to experience a 'two-state' equilibrium between a majorly populated 'ground'

Analysis of 15 N CPMG RD data
state and a sparsely populated 'excited' conformational state, and so the same interconversion rates were applied to all residues during analysis. Uncertainties in fitted parameters were estimated by a boot-strapping procedure where synthetic datasets were created by sampling residues with replacement, re-analysing the new dataset and storing the result. The distribution of parameters achieved from 1,000 such operations provided a measure of experimental uncertainty.
Twenty-six residues with R ex > 2 s -1 at 600 MHz were selected for further analysis (Supplementary Table 2 (Supplementary Fig. 6).
For oxidised cHSP27, residues D129, E130, R136, C137 and F138 showed significant variation in R 2,eff with CPMG (Table S3, S4) (R ex > 2s -1 ). Analysis assuming each residue had independent exchange rates revealed that D129 and E130 clustered with a k ex ~1500s -1 (Supplementary Table 3), whereas R136, C137 and F138 formed a cluster with k ex > 3000s -1 (Table   S4). This distinction was apparent from the raw data ( Supplementary Fig. 4) as CPMG RD curves from the latter group showed variation in R 2eff with ν CPMG at high pulse frequencies, whereas those from the first group were effectively in the fast exchange limit at much low CPMG frequencies. The reduced chi squared value, χ 2 red of the first group was 1.01, and that of the second group was 1.02.

Determination of the dimerization dissociation constant (K d ) from 15 N CPMG RD data
The CPMG data were analysed according to a two-state equilibrium scheme where G k ge k eg ¾ ® ¾ ¬ ¾ ¾ E. As described in the text, k eg was found to be concentration dependent, allowing identification of the minor state E to be monomeric, suggesting the equilibrium has the form A 2 k off k on ¾ ® ¾ ¬ ¾ ¾ 2A. The twostate rate constants derived from CPMG measurements, k eg and k ge , were converted to K d , k on and k off measurements as described in Supplementary Table 1. The K d values obtained from CPMG analysis from C137S using data acquired at 1 mM and 0.3 mM were highly similar ( Supplementary Fig. 4), supporting the identification of the equilibrium to be monomer/dimer exchange. 15 N CPMG RD data were recorded on 1 mM 2 H, [U- 15 Table 1). Justifying this, locally obtained values of k eg , k ge and K eq vary with temperature in a manner consistent with these equations ( Supplementary Fig. 4).

pH-induced dissociation and unfolding of C137S dimers
A 2D 1 H-15 N HSQC spectrum was recorded on a sample of [U-15 N]-C137S at 250 M in NMR buffer at pH 7 at 25 °C. Separate samples were independently prepared in NMR buffer at pH 6.5, 6.0, and 5.0 and NMR spectra were recorded to assess the effect of pH on dimerization. The well-resolved peak from G116 was used to calculate the dimerization K d as a function of pH ( Supplementary Fig. 5, inset). At pH 5, no dimer was observed, and thus the K d at pH 5 (~5 mM) is four orders of magnitude larger than that at pH 7 (0.4 μM). Below pH 6.5, the sample was highly unstable and white precipitant was evident by the end of the NMR experiments (20-40 minutes).  Table 1).

CD spectroscopy and bis-ANS fluorescence of C137S and cHSP27(H124K/C137S)
Samples for CD spectroscopy were prepared at 20 M in NMR buffer and placed into cuvettes with a 1 cm path length. Data were recorded using a Jasco Model J720 CD spectrophotometer at wavelengths between 200 and 260 nm. The fluorescent probe bis-ANS (Sigma Aldrich) was dissolved in NMR buffer and added to a final concentration of 10 M to wells containing a final volume of 100 L of 40 M C137S or cHSP27(H124K/C137S) in NMR buffer. Under these conditions, C137S is primarily a dimer and cHSP27(H124K/C137S) is primarily a monomer. Fluorescence at 500 nm was recorded following excitation at 350 nm.

Aggregation of cHSP27(H124K/C137S)
The aggregation cHSP27(H124K/C137S) was monitored by following the absorbance at 340 nm at 37 °C in NMR buffer. The total volume was 100 L and the protein concentration was either 200 M or 800 M. For comparison, C137S was prepared at 800 M and subjected to identical treatment.
Notably, the sequence-based aggregation propensity predictors Tango 107 and Zyggregator 108 do not predict any notable change owning to the choice of mutation. All concentrations had either three or six replicates, and the mean ± SD is reported in Fig. 5 and Supplementary Fig. 7.

Random Coil Index
To identify  strands in the C137S monomer and dimer, we utilized the software Random Coil