Extreme disorder in an ultra-high-affinity protein complex

An extremely unstructured protein complex

The binding of H1 to ProTa has been demonstrated both in vitro¹⁸ and in vivo¹¹. However, their high net charge, low hydrophobicity, and pronounced disorder in the free proteins raise the question of how much structure is formed when they interact. We used circular dichroism (CD) and nuclear magnetic resonance (NMR) spectroscopy to investigate the formation of secondary and tertiary structure. The CD spectra of unbound ProTα and H1 reflect the low secondary structure content of the individual IDPs, except for the small helix-turn-helix domain of H1^13,19,20 (Fig. 1c). Surprisingly, the CD spectrum of an equimolar mixture of the two proteins can be explained by the simple sum of the individual spectra, indicating that complex formation entails minimal changes in average secondary structure content.

To obtain residue-specific information, we employed NMR spectroscopy. ¹H,¹⁵N heteronuclear single quantum coherence (HSQC) spectra of the individual proteins exhibit low dispersion of the ¹H chemical shifts, as expected for IDPs^14,21-23 (Fig. 1e,f). Only the globular domain of H1, which is stably folded even in isolation (Extended Data Fig. 1), shows the large dispersion of resonances characteristic of tertiary structure^23,24 (Fig. 1g). Remarkably, the overall peak dispersion remains unchanged upon complex formation, confirming that no pronounced tertiary structure is formed upon binding. Nevertheless, small but clearly detectable peak shifts observed for ProTα and H1 indicate significant changes in the average chemical environment of the corresponding residues, as expected upon interaction with the large opposite charge of the other IDP. For ProTα, 95% of the amide backbone nuclei could be assigned (Extended Data Fig. 2), enabling a residue-specific analysis: The C^α secondary chemical shifts²⁵ of ProTα show no evidence for the induction of persistent or transiently populated secondary structure upon complex formation (Fig. 1d), in agreement with the CD data (Fig. 1c). The severe overlap in the NMR spectra of the unstructured parts of H1 precluded residue-specific assignments, but the clusters of H^α-C^α peaks in the ¹H,¹³C-HSQC spectrum from the lysine-rich disordered regions neither exhibit detectable chemical shift perturbations upon titration with ProTα, nor do additional resonances emerge (Extended Data Fig. 3e,f). We thus have no indications of changes in secondary structure content in H1 upon ProTα binding, even though we cannot exclude subtle structural adjustments within the uncertainty of the CD and NMR data.

The lower intensity of the resonances corresponding to the H1 globular domain (Fig. 1f,g, Extended Data Fig. 3) is likely to originate from the faster transverse (T2) relaxation of structured compared to unstructured regions; additionally, tumbling of the globular domain is decelerated by the drag of the unstructured regions it is embedded in²⁶. Upon complex formation, the intensity of many H1 (and ProTα) resonances decreases, and those of the globular domain drop below the noise (Extended Data Fig. 3b and Fig. 1f,g). The large hydrodynamic radii of H1 and the complex (Extended Data Fig. 4a,b) support a large effective rotational correlation time as the origin of peak broadening, but an additional contribution from chemical exchange cannot be excluded. Note, however, that the globular domain is dispensable for complex formation (Fig. 2b, cf. High-affinity binding in spite of disorder).

High-affinity binding in spite of disorder

To quantify the strength of the interaction between H1 and ProTα, we used single-molecule Forster resonance energy transfer (FRET), which enables measurements over a very broad range of affinities, down to the picomolar regime. By labeling two positions with a donor and an acceptor dye, site-specific distances and distance changes between or within the polypeptides can be determined by confocal fluorescence detection of molecules freely diffusing in solution^27,28. ProTα labeled at positions 56 and 110 exhibits a mean transfer efficiency, ⟨E⟩, of 0.33 at near-physiological ionic strength (Fig. 2a, Extended Data Table 2), as expected for this IDP, which is highly expanded owing to its large negative net charge^13,29,30. Upon addition of unlabeled H1, a population with higher ⟨E⟩ of 0.58 (i.e. shorter average distance) emerges: Evidently, binding the positively charged H1 leads to a marked compaction of ProTα by charge screening, analogous to that obtained upon addition of salt²⁹. The same behavior is observed for doubly labeled H1 (Extended Data Table 2), demonstrating a mutual adaptation of the conformational ensembles. The resulting dissociation constant in the low picomolar range reveals an extremely strong interaction with H1 (Fig. 2b, Extended Data Table 2), consistent with the physiological role of ProTα as a linker histone chaperone¹⁷ that needs to compete with the tight binding of H1 to chromatin³¹. Measurements with other FRET dyes and label positions resulted in very similar affinities (Extended Data Table 2), indicating that fluorescent labeling has only a small effect on binding. The dominant contribution to the interaction with ProTα stems from the unstructured C-terminal part of H1, which alone binds with picomolar affinity. The N-terminal half and the isolated globular domain of H1 also bind ProTα, but with much lower affinity (Fig. 2b). At least four isolated globular domains can bind to ProTα at the same time, with modest chemical shift changes (Extended Data Fig. 1), suggesting the absence of a specific binding interface.

The large and opposite net charges of ProTα (−44) and H1 (+53) imply a strong electrostatic contribution to binding. Indeed, a mere doubling of the ionic strength from the physiological 165 mM to 340 mM reduces the affinity by six orders of magnitude (Fig. 2c). By extrapolation, a reduction of ionic strength to ~140 mM would take this interaction into the femtomolar range. From low picomolar to 100 μM protein concentrations, the stoichiometries obtained from intermolecular FRET (Extended Data Fig. 4c) and NMR chemical shift titrations (Extended Data Figs. 2 and 3), as well as the hydrodynamic radii measured with pulsed-field gradient NMR and two-focus fluorescence correlation spectroscopy (2f-FCS) (Extended Data Fig. 4a,b) indicate the predominant formation of one-to-one dimers and the absence of large oligomers or coacervates³². However, in the presence of a large excess of one of the binding partners, a decrease in FRET efficiencies is indicative of the weak association of additional molecules with a K_D in the 10 to 100 μM range (Extended Data Fig. 4d,e). A weak propensity for trimer formation is also observed in the simulations described below (Extended Data Fig. 6).

A highly dynamic complex

The lack of structure formation in the H1-ProTα complex implies great flexibility and a highly dynamic interconversion within a large ensemble of configurations and relative arrangements of the two IDPs. The presence of a broad, rapidly sampled distance distribution is also supported by the analysis of fluorescence lifetimes^28,33,34 (Extended Data Fig. 5). Since fluctuations in distance cause fluctuations in the fluorescence intensity of donor and acceptor, the timescale of these long-range distance dynamics can be measured by single-molecule FRET combined with nanosecond fluorescence correlation spectroscopy (nsFCS)^34,35. For individual unfolded or disordered proteins, reconfiguration times (inter-dye distance relaxation times) between ~20 ns and ~200 ns have been observed²⁷. ProTα alone, with its highly expanded chain^13,29 and corresponding lack of impeding intramolecular interactions³⁶, is a particularly dynamic IDP and yields reconfiguration times, τ_r, between ${29}_{-2}^{+2}$ ns and ${78}_{-9}^{+15}$ ⁵ ns, depending on the chain segment probed^34,36 (Extended Data Table 2). H1 (labeled at positions 113 and 194) reconfigures more slowly, with ${\tau }_r={118}_{-14}^{+24}$ ns, but within the range previously observed for unfolded and intrinsically disordered proteins^27,34.

Strikingly, these pronounced and rapid long-range dynamics are retained in the complex, with values of τ_r between ${66}_{-2}^{+2}$ ns and ${191}_{-19}^{+22}$ ns for 13 different labeling pairs throughout the dimer (Fig. 3a-d, Extended Data Table 2). The similarity of τ_r for the two proteins in the complex suggests a coupling of the dynamics of the two intertwining chains. The highly dynamic nature of the complex is further supported by NMR: The longitudinal (T₁) and transverse (T₂) ¹⁵N relaxation times reflect rapid backbone dynamics in the pico- to nanosecond range, both for free ProTα and in the complex (Fig. 3h, Extended Data Fig. 2). The increase in T₁/T₂ (Fig. 3h) and R_H (Extended Data Fig. 4), and the reduced peak intensities (Fig. 3f) are consistent with the increase in τ_r for ProTα observed by nsFCS in the complex (Fig. 3a), where chain-chain interactions are expected to moderate both local and long-range dynamics.

Architecture of an unstructured protein complex

To develop a structural representation of the conformational ensemble of the H1-ProTα complex, we combine single-molecule FRET, NMR, and molecular simulations. We first map the complex with singlemolecule FRET by probing a total of 28 intra- and intermolecular distances with donor and acceptor dyes in specific positions (Figs. 3i, 4a). The resulting intermolecular transfer efficiencies lack pronounced patterns that would be expected for persistent site-specific interactions or an alignment of the chains in a preferred register. The intermolecular transfer efficiencies are most sensitive to the labeling position on ProTα, with the highest efficiencies (i.e. shortest average distances) for the central position ProTα 56, intermediate efficiencies for ProTα 110, and lowest efficiencies (i.e. longest distances) for ProTα 2. These results indicate that the region of highest charge density of ProTα (Fig. 1b) most strongly attracts the oppositely charged polypeptide chain of H1. The charge density along H1 is more uniform (Fig. 1a), as are the transfer efficiencies to ProTα, with only a slight decrease towards the termini (Fig. 3i).

Based on this information, we sought to establish a molecular model of the H1-ProTα complex. Given the lack of specific structure formation and residue-specific interactions, the dominance of electrostatics in the complex, and the size of the system, we used a simplified model in which each residue is coarse grained into a single bead. Coulombic interactions between all charged residues are explicitly included, with a screening factor to account for buffer ions, representing an ionic strength of 165 mM. Other attractive interactions as well as excluded volume repulsion are captured via a short-range potential, with the radius of the residues determined from their volumes³⁷. A structure-based potential³⁸ is used to describe the folded globular domain of H1. The transfer efficiencies computed from Langevin dynamics simulations can be matched to the measured values (Fig. 4a) via the single adjustable parameter in our model, namely the contact energy of the short-range potential, which is the same for all residues (see Methods); explicitly including a representation of the chromophores in the simulations yielded very similar results (Fig. 4a). The resulting intra- and intermolecular distance distributions (Extended Data Fig. 6d) are smooth and unimodal, in accord with the absence of site-specific interactions and lack of structure formation observed experimentally, and attesting to the convergence of the simulations. The good agreement between the transfer efficiency values from experiment and simulation indicates that this simple model, in which the only consequential difference between residues is their charge, captures the essential properties of the structural ensemble. Considering its simplicity, the femtomolar affinity estimated from the model (Extended Data Fig. 5b) is remarkably consistent with the affinities observed experimentally near this ionic strength. Furthermore, the affinity for a second molecule of H1 or ProTα to the complex is predicted to be orders of magnitude weaker, consistent with experiment (Extended Data Figs. 4d,e and 6b).

The resulting intra- and intermolecular distance maps (Fig. 4b) indicate that the interactions between ProTα and H1 are broadly distributed along their sequences, but they also reflect the asymmetry in electrostatic attraction owing to the higher charge density of ProTα in its central and C-terminal regions, as revealed by the single-molecule experiments (Fig. 4a). The NMR results provide an independent experimental test of the model: Indeed, the distribution of the average number of contacts made by the residues of ProTα based on the simulation (Fig. 3e) is strikingly similar to the distribution of changes in chemical shifts, peak intensities, and T₁/T₂ ratios observed upon binding (Fig. 3f-h). These changes occur across the same broad region between residues 46 and 106, encompassing the most acidic tracts of ProTα. Even though peak overlap in clusters of Glu residues prevents a quantitative analysis for some residues in this region, these resonances are not broadened beyond detection, as might be expected for persistent interactions.

Further analysis of the simulated structural ensemble shows a lack of identifiable distinct clusters of configurations (Extended Data Fig. 6a), implying a continuous, unimodal structure distribution. A projection of the simulation onto the first three principal components of the inter-residue distances (Extended Data Fig. 6c) reveals a highly heterogeneous ensemble of arrangements with a wide variety of configurations of the two entwining chains (Fig. 4c). Given the rapid intramolecular dynamics and lack of structure in the complex, the activation barrier for binding is likely to be close to zero. Indeed, association of H1 and ProTα occurs at the diffusion limit, with a binding rate coefficient of (3.1 ± 0.1)· 10⁹ M⁻¹s⁻¹ (Extended Data Fig. 7). The simulations support this mechanism, with a downhill free energy surface for binding, and attractive fly-casting³⁹ interactions enhanced by electrostatics⁴⁰ emerging already at a distance of ~22 nm, much greater than the sum of the hydrodynamic radii (Extended Data Fig. 6b).

Conclusions

Our results suggest that high-affinity complex formation between two oppositely charged IDPs is possible without the formation of structure or the need for folded domains. In contrast to the prevalent paradigm of molecular recognition in biomolecular interactions, this type of highly dynamic complex does not require structurally defined binding sites or specific persistent interactions between individual residues. Rather, the results are well described by long-range electrostatic attraction between the two interpenetrating polypeptide chains, especially between their charge-rich regions. The exceedingly rapid interconversion of many different arrangements and configurations on the 100-ns timescale results in efficient averaging and thus a mean-field-type interaction^41,42 between all charges. This type of complex further expands the known spectrum of disorder in protein-protein interactions³. Although the complex of H1 and ProTα is extreme in its extent of disorder retained for both binding partners, the possibility of this interaction mechanism may not be entirely unexpected, given the prevalence of charged amino acids in many IDPs², the previous observation of disorder in IDPs interacting with folded proteins^4-7, and the role of electrostatics in the formation of dynamic binding interfaces between folded proteins⁴³. Moreover, the H1-ProTα interaction resembles polyelectrolyte complexes formed by charged synthetic polymers⁴², even though the latter usually phase-separate into coacervates. The absence of coacervation^32,42 or liquid-liquid phase separation⁹ for ProTα and H1 at concentrations from picomolar to high micromolar may be due to the complementarity⁴⁴ of the two proteins in terms of effective length and opposite net charge, leading to optimal, mutually saturating electrostatic interactions, or the lack of hydrophobic and aromatic side chains and cation-π interactions, which have been suggested to favor phase separation mediated by proteins^32,45,46.

What are the functional implications of such a high-affinity yet unstructured dynamic complex between two IDPs? Histone H1 is a key factor in chromatin condensation and transcriptional regulation¹¹, and ProTα acts as a chaperone of H1 that facilitates its displacement from and deposition onto chromatin¹⁷. ProTα thus needs to be able to compete with the very high affinity of the histone to chromatin³¹. However, high affinities between structured biomolecules are usually linked to exceedingly slow dissociation⁴⁰, incompatible with fast regulatory response. By contrast, the high affinity of the H1-ProTα complex is facilitated by its ultra-fast association, which enables dissociation on a biologically useful timescale in spite of the high affinity required for function. Another consequence of polyelectrolyte interactions is the possibility of ternary complex formation⁴⁷, signs of which are detected here with a large excess of ProTα or H1 (Extended Data Figs. 4d,e and 6b), resulting in largely unexplored kinetic mechanisms that cannot be explained by competition via simple dissociation and re-association⁴⁸. Finally, the flexibility within such unstructured complexes may facilitate access for enzymes adding posttranslational modifications, which play key roles in the regulation of cellular processes, including those of H1. Another example of this mechanism may be the interaction of the acidic domain of the oncogene SET with the lysine-rich C-terminal tail of p53, which is regulated by acetylation⁴⁹.

The behavior we observe for ProTα and H1 might be surprisingly widespread, since highly charged protein sequences that could form such complexes are abundant in eukaryotes. In the human proteome alone, several hundred proteins that are predicted to be intrinsically disordered⁵⁰ contain contiguous stretches of at least 50 residues with a fractional net charge similar to that of H1 or ProTα. Since the interaction of highly oppositely charged IDPs is unlikely to be very sequence-specific¹⁸, achieving binding selectivity may be linked to other regulatory mechanisms, e.g. cellular localization or synchronized expression during relevant stages of development or the cell cycle.

Extended Data

Extended Data Table 1.

Sequences of protein constructs and fluorescently labeled variants of H1 and ProTα.(top) Sequences of H1 and ProTα wildtype and variants used. Bold yellow-shaded residues are positions mutated to Cys for fluorophore conjugation. Residues in red are part of protease recognition sites used to cleave the HisTag with thrombin (GGPR or GC) or HRV-3C (GP). (Note that the wt sequence of H1 starts with "T"; the preceding Cys residue (−1) was added for labeling.) The underlined H1 sequence indicates the globular domain (GD), identified based on a sequence alignment with the G. gallus homolog²⁰ (PDB access code 1HST, 82% sequence identity). Surface-exposed residues (as shown in Fig. 1a and 5b) are shaded in light blue. The net charge of each variant is indicated in parentheses. ^aC-terminal disordered region. ^bN-terminal disordered region including GD. (bottom) Labeled variants of H1 and ProTα. ^cFörster radius of the corresponding dye pair.

Open in a new tab

Extended Data Table 2.

Binding affinities, molecular dimensions, and reconfiguration times of fluorescently labeled H1 and ProTα.(top left) Affinities of labeled ProTα for H1 at different ionic strength (IS) and for H1 fragments for 165 mM IS (^bsee Extended Data Table 1). Uncertainties for the IS dependence are standard errors estimated from two independent titrations (^auncertainty at 165 mM: see Methods), for fragment binding from dilution errors (see Methods). ^cApparent K_D from fraction of all bound species. (top center) Binding affinities of ProTα and H1 labeled with different dye pairs for the respective unlabeled partner. dUncertainties based on dilution errors. (top right) Transfer efficiencies and average distances of ProTα and H1 labeled with different dye pairs in the bound (R_bound) and unbound state (R_unbound). Uncertainties in distance are based on an estimated systematic error of ± 0.05 in the transfer efficiency from instrument calibration for the different dye pairs. (bottom left) Intermolecular reconfiguration times for the complex of donor-labeled H1 and acceptor-labeled ProTα and vice versa. (bottom right) Reconfiguration times of doubly labeled ProTα and H1 (unbound and bound). Uncertainties estimated by propagating the error on the transfer efficiency (± 0.05).

Open in a new tab

Supplementary Material