A DNA condensation code for linker histones

Significance

Here, we deploy an in vitro model system to study DNA compaction by linker histone (H1) proteins in chromatin. With it, we show how natural sequence variation between H1 C-terminal tails gives control over both the degree of condensation and higher-order structuring of the DNA within condensates, which is key to regulating its inherent activity in a chromatin context. Canonical H1 tails comprise mainly lysine, alanine, and proline, but we demonstrate how a raised arginine:lysine ratio and overall positive charge density can contribute to the unusual levels of compaction and organization seen in some variants. Further, we show how proline-free regions found in several germ-line H1s permit contraction of the tail into helical structures that have particularly potent condensing properties.

Abstract

Linker histones play an essential role in chromatin packaging by facilitating compaction of the 11-nm fiber of nucleosomal “beads on a string.” The result is a heterogeneous condensed state with local properties that range from dynamic, irregular, and liquid-like to stable and regular structures (the 30-nm fiber), which in turn impact chromatin-dependent activities at a fundamental level. The properties of the condensed state depend on the type of linker histone, particularly on the highly disordered C-terminal tail, which is the most variable region of the protein, both between species, and within the various subtypes and cell-type specific variants of a given organism. We have developed an in vitro model system comprising linker histone tail and linker DNA, which although very minimal, displays surprisingly complex behavior, and is sufficient to model the known states of linker histone-condensed chromatin: disordered “fuzzy” complexes (“open” chromatin), dense liquid-like assemblies (dynamic condensates), and higher-order structures (organized 30-nm fibers). A crucial advantage of such a simple model is that it allows the study of the various condensed states by NMR, circular dichroism, and scattering methods. Moreover, it allows capture of the thermodynamics underpinning the transitions between states through calorimetry. We have leveraged this to rationalize the distinct condensing properties of linker histone subtypes and variants across species that are encoded by the amino acid content of their C-terminal tails. Three properties emerge as key to defining the condensed state: charge density, lysine/arginine ratio, and proline-free regions, and we evaluate each separately using a strategic mutagenesis approach.

Chromatin—the complex of genomic DNA and proteins in eukaryotes—is packaged in stages (1). The basic unit is the nucleosome (2), in which the DNA performs almost two full turns around an octamer of core histones. Nucleosomes form arrays of “beads on a string” that make up an 11 nm fiber. The next stage, in which linker histones bring about further condensation, is more enigmatic, despite it spanning a length scale that is critical for the control of key transactions such as gene expression, replication, and repair (3). The orderly 30-nm fiber originally proposed (4) has been detected in vivo (5, 6), but only in avian erythrocytes, that are terminally differentiated and transcriptionally silent. More recently, cryo-electron tomography (cryo-ET) and live-cell imaging have revealed that the 11 nm fiber compacts via irregular folding (7) and is dynamic (8, 9), specifically that clusters of nucleosomes show coherent motions suggestive of liquid flow (10, 11). It is likely that, at the local scale, linker histone condensed chromatin can adopt a virtual continuum of states between solid and liquid, and that these impact the accessibility of the underlying DNA: a liquid-like state is a compelling means by which chromatin could respond quickly to environmental stimuli, allowing transactions such as transcription, replication, and repair. Conversely, unfolding and refolding a fiber-like architecture would be much less efficient.

The growing appreciation of dynamics in chromatin packaging by “top–down” imaging has paralleled developments by us and others in our “bottom–up” understanding of the linker histone proteins at the molecular level. Linker histones consist of a winged-helix domain (~80 residues) flanked by short N-terminal and long C-terminal disordered regions (25 to 30 and ~100 residues, respectively). The globular domain locates H1 at the nucleosome dyad (12), and along with the first 13 residues of the C-terminal domain, stabilizes the entering and exiting DNA strands in a stem arrangement (13). The rest of the C-terminal domain binds the linker DNA between nucleosomes to facilitate their closer approach, but is not well defined by cryo-electron microscopy (cryo-EM) (14) and invisible by cryo-ET (7) due to dynamics (15). Previously, we studied the isolated C-terminal domain (“CH1”) in order to reveal its structure and condensation behavior with DNA by NMR spectroscopy and isothermal titration calorimetry (ITC) (16). This model system of CH1 and linker DNA, although very minimal, displays surprisingly complex behavior, forming three distinct states depending on the conditions: disordered complexes (at high ionic strength/low macromolecule concentrations), phase-separated droplets (at intermediate ionic strength and macromolecular concentrations), and higher-order structures (at low ionic strength). These states appear to have some correspondence to the states of H1-bound chromatin in vivo (“open” 11 nm fibers, dense liquid-like assemblies, and organized 30-nm fibers, respectively) (17) (Fig. 1 A and B). The sufficiency of CH1 for DNA condensation is consistent with its requirement for high-affinity chromatin binding (18), its essential role in condensation of native chromatin beyond the 11 nm fiber (19), and as the main driver for phase separation and decreased dynamics in a subsequent study of in vitro reconstituted 12-nucleosome arrays (20).

Fig. 1.

Chromatin condensation by H1 tails. (A) A minimal model for the study of chromatin condensation by linker histones consists of internucleosomal linker DNA and the C-terminal tail of H1 (CH1) (16). (B) The model system encompasses the possible H1-condensed states of chromatin, forming disordered complexes, phase-separated droplets, and higher-order structures (a liquid crystalline phase), depending on the conditions (see main text). (C) Amino acid residue content of curated set of 94 CH1s by percentage (black bars, Gallus gallus isoform H1.11L and variant H5 as red and green bars, respectively). (D) Fraction of charged residues (Gallus gallus isoform H1.11L and variant H5 are shown as a red circle and a green triangle, respectively). (E) Distributions of κ (charge patterning; blue) and Ω (patterning of charged and proline residues; orange) where 0 = well-mixed, 1 = fully segregated. (F) Sequences of CH1 and CH5 are the wild-type sequences of the C-terminal tails of H1.11L and H5 from Gallus gallus. For the rationale behind the design of CH1_R, CH5_K, CH1_VA, CH1_VT, and CH1_PA see the main text. Arginine, proline, and valine are highlighted in dark blue, green, and yellow, respectively.

All proteins that condense DNA are rich in lysine, arginine, or both. The linker histone tails are essentially copolymers of lysine, alanine, and proline, with variable amounts of arginine, depending on the subtype/variant. At a phenomenological level, arginine has been observed to generate more dramatic DNA condensation than lysine. This is due to the distinct chemistry of their side-chains: The amino group of lysine makes weak ion pairing interactions with the DNA phosphates, while the guanidino group of arginine has the propensity to make additional π-stacking and bidentate interactions, and hydrogen bonds (21–24). It follows that the amino acid content of linker histone tails is likely to code for a range of condensed chromatin “baseline” states that differ in their degree of compaction and other material properties. Eukaryotes make several different H1 subtypes (for example, there are 11 in humans and mice), which can differ in their expression across cell types and their localization (25–27). Some are expressed in a replication-dependent manner, while others occur exclusively in germ cells. Further, several variant linker histones [e.g., H5 in Gallus gallus (28)] are associated with terminally differentiated cells. The winged-helix globular domains are highly conserved across linker histones, albeit with minor differences (five residues) that determine the exact positioning of the globular domain relative to the nucleosome dyad (29). Therefore, differences in the intrinsic condensing properties are likely to arise from sequence and content variation in the tail [this is already known to affect the orientation of the linkers directly flanking the nucleosome (30)], or from distinct posttranslational modifications (31). In this study, we explicitly address the question of how the natural variation in linker histone sequences directly affects DNA condensation by comparison of the structure, thermodynamics, and condensation propensities of tails with a range of potentially relevant properties spanning different lysine/arginine ratios, positive charge densities, hydrophobicity, and proline content and spacing.

Results

Linker Histone H1 Tails Are Universally Well-Mixed Polycationic Electrolytes.

In order to understand the amino acid sequence space of the known linker histone H1 tails, we curated a database free from erroneous inclusions and annotation errors. An initial set of 245 sequences was revealed from a search for “Histone H1” in UniProtKB/SwissProt after exclusion of sequences from bacteria and viruses. Further manual filtering to remove hypothetical proteins, protamines, and sequences with no globular domain or significant tails reduced the number to 94. The sequences were aligned on the globular domain in order to extract the C-terminal tails (SI Appendix, Fig. S1), which were then analyzed for content, charge, and distribution of charge and proline (Fig. 1 C–E). The tails are dominated by lysine (mean 34.6%, σ 4.2%) and alanine (mean 24.0%, σ 4.1%), followed by proline (mean 10.6%, σ 1.7%) (Fig. 1C). They cluster around an average f₊, f_– coordinate of (0.38, 0.03), indicating a very high fraction of positively charged residues and relatively few negatively charged residues (Fig. 1D). Charge patterning (32, 33) was assessed by comparing κ parameters across the set (Fig. 1E). κ spans values between 0 (well-mixed/fully alternating) to 1 (fully segregated); the distribution of κ in CH1s is concentrated around 0.13, indicating that the charge is well-mixed in all linker histones of the set. The distribution of Ω (which includes proline as well as charge) showed an even greater level of mixing, centered around 0.06. Taken together, the set of linker histone tails is one of universally well-mixed polycationic electrolytes that are conserved in coarse-grained features (SI Appendix, Fig. S1).

Sampling Sequence Space via Mutagenesis.

We have studied H1.11L from Gallus gallus experimentally as a model linker histone for many years (16, 34). It is typical of the somatic isoforms in chicken as well as the larger set collated here, being 41% lysine, which is well distributed (κ = 0.1) (Fig. 1 C and E). The remaining residues are mainly alanine (31%) and proline (13%), which is also well distributed (Ω = 0.05). The most hydrophobic residue is valine, of which there are seven. The positions of both proline and valine are well conserved across the six somatic chicken isoforms (SI Appendix, Fig. S2A), although the valine is not always present (two of the isoforms have only two valines). Chicken H1s contain little, if any, arginine. However, a natural variant found in avian erythrocytes, H5 (28), contains 13 arginine residues in its C-terminal tail (SI Appendix, Fig. S2B), and is also more charge dense, being 50% lysine/arginine (Fig. 1 C and D). Notably, this variant forms detectable 30-nm fibers in vivo (6).

The natural variation in linker histone tail sequences exemplified by the chicken isoforms and variants spans many properties that could potentially impact DNA condensation: lysine/arginine ratio, positive charge density, and hydrophobicity. The high proline content is also intriguing. These unusual proteins are challenging to express and purify, therefore a minimal set of mutants was designed that could reveal the impact of the amino acid content in as systematic a manner as possible (Fig. 1F). The set consists of the C-terminal tails of H1.11L and H5 (denoted CH1 and CH5), two different valine knockouts (CH1_VA and CH1_VT), and a proline knockout (CH1_PA). For the valine knockouts, a substitution by threonine was tested as well as by alanine since threonine has a similar bulk to valine, but higher polarity than alanine. Further, in order to deconvolute the effect of arginine content and charge density, two additional proteins were generated: CH5_K, in which the 13 arginine residues were mutated to lysine (thus maintaining the charge density of CH5 while removing the arginine), and CH1_R, in which 13 of the lysines in CH1 were mutated to arginine in roughly the same patterning as for the arginines in CH5 (thus maintaining the charge density of CH1 but with an arginine:lysine ratio of 0.4, as for CH5).

DNA Condensation Is Driven by Both Positive Charge Density and Arginine Content.

The condensation of 20 bp dsDNA by the proteins and mutants was assessed by mixing at 1:1 molar stoichiometry at 25 µM, I = 160 mM prior to observation by confocal fluorescence microscopy (Fig. 2A). All protein/DNA mixtures gave rise to spherical phase-separated droplets. The C_sat (measured at I = 160 mM by dynamic light scattering (DLS); Fig. 2B) spanned 0.5 to 6 µM and showed the following trend: CH5 < CH5_K ≈ CH1_R < CH1. The condensation propensity was further explored by A₃₄₀ (turbidity), at the concentrations shown, followed by addition of salt, to produce phase diagrams (Fig. 2C). Comparison of CH1 and CH5 shows that condensate formation by CH5 is more salt resistant at lower concentrations and persists at higher ionic strengths. The CH1_R and CH5_K mutants were intermediate in their condensing properties, and broadly similar. Direct pairwise comparison of CH1/CH1_R and CH5_K/CH5 (each pairing differs only in that 13 K are mutated to R) indicates that high arginine content promotes condensation. Further pairwise comparisons of CH1/CH5_K and CH1_R/CH5 (each pair differ only in that the fraction of positively charged residues increases from 0.4 to 0.5) indicate that charge density promotes coacervation. Taken together, these findings demonstrate that the condensing propensity of CH5 derives from the additive effects of i) its 13 arginines in place of lysine, and ii) its 10% higher charge density.

Fig. 2.

Condensation behavior of CH1, CH5, and mutants. (A) Confocal fluorescence microscopy of droplets formed at 25 µM 1:1 protein:20mer DNA complex, I = 160 mM, 10% of DNA fluorescein amidite (FAM)-labeled. (Scale bar, 5 µm). (B) C_sat, measured by DLS (values were consistent across triplicate experiments). Phase diagrams for (C) arginine and charge density series, (D) hydrophobicity series, and (E) proline knockout, vs. concentration and ionic strength. Concentration and turbidity (A₃₄₀) is of the 1:1 mixture. Turbidity is shown in numbers as well as a yellow-to-green color scale. Errors are <5%.

Proline Negatively Regulates DNA Condensation.

The phase diagrams and C_sat of CH1 and the less hydrophobic CH1_VA and CH1_VT were very similar (Fig. 2 B and D). Therefore, the seven valines in CH1 do not significantly affect the DNA condensing propensity of the protein, suggesting that it is entirely driven by electrostatics. In order to verify this, the condensates were challenged with 1,6-hexanediol, a known disrupter of hydrophobically driven condensate formation (35). Very little change was seen, for example, an addition of 10% 1,6-hexanediol by volume—the highest amount commonly used—to a CH1:20mer condensate lowered the A₃₄₀ from 3.88 to 3.67, a reduction of only 5%. In contrast to the minimal effect of valine mutagenesis, CH1_PA had a dramatic effect on condensation, having a 4× lower C_sat and persisting to higher ionic strength (Fig. 2 B and E). This was explored in further experiments.

Robust Condensation Is Underpinned by Enthalpic Contributions.

The thermodynamics driving condensation of 20 bp double-stranded DNA (dsDNA) by the natural proteins and mutants was investigated by ITC. This method, when coupled with a parallel titration followed by turbidity, reveals the heats associated with DNA binding and phase separation and is able to separate the contributions to some degree (16). The isotherms for CH1 and CH5 showed endothermic (positive) heats (Fig. 3A), albeit of different magnitudes (y-intercepts +17 and +8 kcal/mol, respectively) and different stoichiometries (the midpoints of the falling sigmoids are at N ~1.0 and ~0.8, respectively). The isotherms display varying degrees of biphasic behavior due to the onset of phase separation, which results in a characteristic “hump.” While it is possible in theory to fit biphasic data with a suitable phenomenological model (36), as we have previously for CH1 (16), the parameters obtained are poorly defined in all but the cases with very distinct humps. However, much can be deduced from simple inspection: The endothermic heats indicate that condensation is an entropy-driven process, likely through counterion release (37). Further, the correspondence of the hump to phase separation is confirmed by comparison with the turbidity titrations (Fig. 3A, allowing for the fact that turbidity is a cumulative measurement while ITC gives the heat per injection: The peak in the turbidity measurement therefore roughly corresponds to the maximum downward gradient in the isotherm at N ~ 1, equivalent to the inflexion point were it a simple sigmoid). The initial injections where little or no phase separation is occurring therefore reflect ion pairing (protein–DNA binding), and the later injections the process of ion pairing due to binding and phase separation. This phenomenon has been discussed previously (16, 36).

Fig. 3.

Thermodynamics of condensation. ITC of protein (as indicated) into 20 bp DNA, alongside A₃₄₀ to identify phase separation events, in buffer containing 150 mM NaCl (total I = 160 mM). (A) Raw data, isotherms, and A₃₄₀ for CH1 and CH5. (B) Isotherms (above) and A₃₄₀ (below) for the complete arginine and charge density series. (C) Raw data, isotherms, and A₃₄₀ for the proline knockout. (D) Isotherms (above) and A₃₄₀ (below) for the hydrophobicity and proline knockouts. (E) Raw data, isotherms, and A₃₄₀ for protamine titrated into 16 bp DNA. Complete dataset in SI Appendix, Fig. S3.

The isotherms for CH1_R and CH5_K showed intermediate features (SI Appendix, Fig. S3). Comparison of integrated and normalized heats of injection for CH1, CH5, CH1_R, and CH5_K (Fig. 3B) reveals that CH5 gives the lowest endothermic heats (y-intercept +8 kcal/mol) and CH1 the highest (+17 kcal/mol). The mutants CH1_R and CH5_K are in between (+15 kcal/mol). Following this initial phase, the hump is most pronounced for CH5 and CH5_K, but more subtle for CH1 and CH1_R. The turbidities (Fig. 3B) show that the onset of phase separation for CH5 and CH5_K occurs at the lowest protein:DNA ratio. CH5 is then distinct, reaching the highest values, and persisting to the highest molar ratio, i.e., resists reentrant behavior through overcharging. CH1 phase separates to a limited degree around N = 1, which is consistent with the turbidity under the same conditions [10 μM concentration, I = 160 mM (Fig. 2C)] and the CH1_R mutant is in between.

The lower hydrophobicity mutants, CH1_VA and CH1_VT, showed similar thermodynamic signatures to CH1 (SI Appendix, Fig. S3), although CH1_VT had slightly lower heat and higher N, and both mutants had slightly higher turbidity. However, CH1_PA was profoundly different (Fig. 3C). Around the onset of turbidity, which occurs from the first or second injection as for CH5, the isotherm became more complex, the peaks showing both endothermic (positive) and exothermic (negative) features. Despite this complexity, it was possible to link the observed thermodynamic events with phase separation by comparison with the turbidity and the other isotherms (Fig. 3D). The onset of phase separation is at lower concentration for CH1_PA than for CH1, correlating with exothermic signals in the first 7 to 8 injections. The exothermic signals were broad, in line with previous observations that signals with slow relaxations are highly characteristic of phase separation (36). Superimposed on these are the sharper endothermic peaks that are characteristic of entropically driven ion pairing, such that the overall heat per injection is close to zero. Coincident with the peak in turbidity, which is at lower concentration for CH1_PA at N ~ 0.8, the broad signals change sign, consistent with progressive dissolution of the condensate through overcharging of the system with protein, and the heats per injection are endothermic beyond this point. These features are indicative that phase separation for CH1_PA has a negative enthalpic contribution, unlike the other proteins. Overall therefore, it appears that robust condensation correlates with a reduction in endothermic heats for CH1_R, CH5_K, and CH5, which reaches an extreme for CH1_PA in which there are strong indications that the phase separation event itself is exothermic. This trend was further explored by testing a highly charge dense, arginine-rich protamine (38). (Protamines replace histones to achieve a 10-fold greater compaction of DNA in sperm.) Titration of salmon protamine into 16 bp dsDNA gave a fully exothermic isotherm (Fig. 3E), thereby confirming the trend.

Mutation of Proline Leads to Structural Changes.

While the higher effectiveness of arginine over lysine in DNA compaction has been noted before, e.g., in studies of synthetic oligoarginine (39), the high DNA condensing propensity of CH1_PA was unexpected and its physical origin less clear. Proline is the most abundant amino acid in the H1.11L C-tail (13%) after lysine and alanine. It is also relatively well distributed: The κ parameter for proline alone is 0.134, and the two longest proline-free regions are 11 and 13 residues (Fig. 1F). Proline has many known impacts on secondary structure: location at the ends of helices (40), ability to isomerize (41), and potential to form polyproline II (PPII) helices (42). The structures of all the proteins were therefore explored using circular dichroism (CD) spectroscopy.

The far-ultraviolet (far-UV) CD spectra for CH5, CH5_K, and CH1_R were identical to that for CH1 and consistent with proteins that are predominantly disordered, having no significant features beyond the negative peak at ~200 nm (Fig. 4A). Therefore, arginine content and charge density do not noticeably impact the structure of the free CH1 protein, as has been observed previously for CH1 vs. CH5 (43). The reduced hydrophobicity mutants CH1_VA and CH1_VT were also indistinguishable from that of CH1 (Fig. 4B). However, CH1_PA showed a shallow dip around 222 nm and a shift of the ~200 nm peak to higher wavelengths, indicative of a small percentage of alpha helix in a disordered background. To ascertain whether this was an ensemble average of several less stable structures (as opposed to a short, defined length of stable helix), the sample was cooled. On cooling from 25 °C to 0 °C, the amount of alpha helix visibly increased (Fig. 4C). This was in contrast to CH1 at 0 °C, which showed a smaller change and in the opposite direction: The peak ~200 nm became more negative and there was a slight increase in the signal around 218 nm, indicating temperature-dependent stabilization of PPII helix, as expected for a lysine-rich polypeptide at low/neutral pH (44, 45).

Fig. 4.

CH1_PA forms temperature-dependent alpha helix. Far-UV CD of the free proteins for (A) arginine and charge density series, (B) hydrophobicity and proline knockouts, and (C) proline knockout and CH1 vs. temperature. Temperature 25 °C unless indicated, in buffer containing 150 mM NaF (total I = 160 mM).

To investigate this further, ¹⁵N-labeled CH1 and CH1_PA were expressed and purified. The ¹⁵N-HSQC spectrum of CH1 (Fig. 5A) displayed the limited ¹H^N chemical shift dispersion and narrow line widths characteristic of a highly dynamic polypeptide chain, as observed previously (16), with a small degree of line broadening on cooling to 0 °C, consistent with a degree of stiffening of the chain due to PPII. The ¹⁵N-HSQC spectrum of CH1_PA at 25 °C was also consistent with a disordered protein (Fig. 5B), albeit with slightly increased ¹H^N chemical shift dispersion and line widths, and a general shift upfield (lower ppm). However, the spectrum at 0 °C revealed a large number of peaks at non-random-coil positions, and significantly increased line widths, consistent with the secondary structure formation at several locations as inferred by CD (Fig. 4C).

Fig. 5.

Alpha helix nucleates at several distinct loci in CH1_PA. ¹⁵N-HSQC NMR of free proteins: (A) CH1 at 25 °C and 0 °C, (B) proline knockout at 25 °C and 0 °C, and (C) proline knockout vs. temperature, with peak assignments. Gray shading present to illustrate chemical shift dispersion in ¹H^N. Buffer contains 150 mM NaCl (total I = 160 mM).

The temperature dependence of ¹H^N chemical shifts is generally linear, and the rate of change is a reliable indicator of the extent of hydrogen bonding. Values of Δδ/ΔT less negative than –4.5 ppb/K are indicative of a stable hydrogen bond (46). The values for CH1, where the previously assigned peaks (16) can be reliably tracked, lie around –10 ppb/K, (mean = –9.9, σ = 0.9) with no single value less negative than –6.5 ppb/K (for Glu118). In contrast, the values for CH1_PA were significantly less negative (e.g., Glu118 was –1.7 ppb/K, shown boxed in Fig. 5C). Furthermore, for CH1_PA, many peaks (including Glu118) showed curved rather than straight trajectories of chemical shift with temperature, and peak broadening, consistent with the onset of folding. Peak assignment was made challenging by the ultra-low sequence complexity and repeated motifs (CH1_PA is 40% lysine, 43% alanine), but several peaks could nevertheless be assigned sequence-specifically, some having linear temperature trajectories (e.g., Ser175) and some curved (e.g., Ala153).

Higher-Order Structure in the Droplet Phase.

It was not possible to follow complex formation of CH1_PA with DNA by NMR due to the strong tendency for phase separation at NMR concentrations (Fig. 2E). However, the structures formed in the condensates could be investigated by CD. Previously, for CH1, we observed a large negative ellipticity at ~280 nm (16), which we inferred to be the result of scattering from a cholesteric liquid crystalline phase in which the DNA duplexes adopted a left-handed twisted stack (47), but only in low ionic strength conditions. Such signals have historically been denoted “polymer and salt induced-”, “psi-”, or “ψ”-DNA. CD spectra for the CH1, CH5, CH5_K, and CH1_R droplet phases at physiological ionic strength are shown in Fig. 6A. No ψ-DNA signal was seen for CH1, consistent with our previous observations at physiological ionic strength (16), but the signal for CH5 shows a distinct negative ellipticity at 295 nm consistent with ψ-DNA. The signals from CH5_K and CH1_R are in between, although CH1_R closely resembles CH1 while CH5_K deviates further from CH1 in the direction of CH5. The signal from the CH1_PA-DNA condensate (Fig. 6B) is very different, showing a pronounced ψ-DNA scattering signal at 283 nm, consistent with higher-order structure formation with the highest degree of dichroic scattering. The secondary dip at 226 nm may indicate alpha helix formation in the protein, but is inconclusive due to the large background signal from DNA.

Fig. 6.

Structure and heterogeneity of the dense phase. Far-UV CD of the protein:DNA condensates for (A) arginine and charge density series and (B) CH1 and proline knockouts, with 20 bp DNA, 1:1, at 25 °C in buffer containing 150 mM NaCl (total I = 160 mM). Free DNA (black) is also shown for reference. The position of the ψ-DNA scattering signal is indicated (see the text). (C) FRAP recovery curves and fits. For comparative purposes, these are shown normalized after subtraction of A₀ (Materials and Methods). (D) Fitted parameters resulting from mono- or biexponential models, as appropriate. SSR = sum of the squared residuals.

The diffusive properties of the droplets were probed by fluorescence recovery after photobleaching (FRAP) experiments (SI Appendix, Fig. S4). The data for each protein were quantified and fit (Fig. 6C). Satisfactory fits were obtained using a single exponential function for CH1 and CH1_R, but a double exponential was required for CH5, CH5_K, and CH1_PA, implying the existence of two distinct populations with different recovery times, denoted τ_fast and τ_slow, in approximately equal fractions A_fast and A_slow (Fig. 6D). No immobile fraction was detected, i.e., the curves recovered fully, albeit at different rates. The τ for CH1_R was ~2× that for CH1, implying that although both times are short even for liquid condensates, the K to R mutation slows diffusion. The same trend was seen for CH5 and CH5_K, τ_fast for CH5 being ~3× that for CH5_K. The effect of charge density on τ_fast was less clear, however, the more charge-dense CH5 and CH5_K also have a significant population that diffuses an order of magnitude more slowly, with τ_slow for CH5 ~5× that for CH5_K, again demonstrating a possible effect of arginine. CH1_PA had fast- and slow-diffusing populations that were intermediate between CH5 and CH5_K in rate. Given that CH5, CH5_K, and CH1_PA all showed higher-order structure by CD, while CH1 and CH1_R did not (Fig. 6 A and B), it is possible that this structuring is responsible for the slow-diffusing component seen by FRAP. To explore this further, the experiment was repeated with protamine, which is known to form higher-order structures with DNA [e.g., toroids (48)]. In contrast to the droplets formed by histone tails, those containing protamine had distinct dim patches (SI Appendix, Fig. S4) reminiscent of images from other laboratories (48). FRAP of the bright region yielded a distinct biphasic recovery with τ_fast similar to CH1_R, and the largest τ_slow (Fig. 6 C and D).

Discussion

The comparison of the DNA condensation properties of the lysine-rich CH1 and the arginine-rich and charge-dense CH5 was facilitated using a pair of carefully designed mutants. As expected, CH5 produces more robust condensation than CH1 (Figs. 2 B and C and 3A). Dissecting this further, we find that high arginine content promotes condensation (Figs. 2B and 3B), as expected from its bidentate nature (21) and stacking propensity (24). A similar trend is seen for increasing positive charge density. All proteins condense DNA into a droplet phase. The diffusion within the droplets is very fast, but it is retarded by increasing arginine content and charge density, with high arginine appearing to retard diffusion more than charge density (Fig. 6 C and D). A possible explanation for this is that arginine forms stronger, longer-lived network interactions in the dense phase. However, we find high charge density is needed for significant higher-order structure formation, or ψ-DNA, at physiological ionic strength (Fig. 6A). Intriguingly, the appearance of ψ-DNA correlates with a slow-diffusing component to the otherwise very liquid droplets (Fig. 6 C and D), and may explain its origin. Overall, the underlying thermodynamics reveal a trend toward enthalpic control for the arginine-rich/charge-dense linker histones relative to canonical H1s that are under a higher degree of entropic control (Fig. 3B).

Less expected were the effects of hydrophobic residues and proline. Mutation of all seven branched hydrophobic residues (valine) to alanine or threonine has a negligible effect on condensation (Figs. 2 B and D and 3D and SI Appendix, Fig. S3), as does 1,6-hexanediol, leading us to conclude that CH1/DNA condensation is a purely electrostatically driven process. However, the impact of proline on the C_sat, thermodynamics, and stoichiometry of condensation (Figs. 2E and 3 C and D), and higher-order structuring (Fig. 6B) is profound. The reasons for this appear to lie in the accessible structures (Figs. 4 and 5): Proline is both rigid and lacks an amide proton to stabilize alpha helices beyond the first few residues. Its absence can therefore confer a propensity for alpha helix, which has highly contrasting properties to the PPII helix (SI Appendix, Fig. S5 A and B). An ideal PPII helix has backbone dihedral angles ϕ = −75° and ψ = +145°, resulting in a left-handed helix of precisely three residues per turn. The structure is extended (3.1 Å/residue) and without backbone hydrogen bonds, forming due to the repulsions between like-charged side-chains (in this case lysine), and is thought to be populated by many disordered proteins at least to some extent (44, 45). In contrast, an alpha helix is right-handed, hydrogen bonded, and 3.6 residues per turn. Compared to the PPII helix, the structure is relatively compressed (1.5 vs. 3.1 Å/residue), and side-chain repulsions are instead minimized through the stagger achieved by the 100° rotation per residue. These differences will directly impact the charge density of the protein, and also its ability to compress further on binding DNA. While AlphaFold is known to lack predictive power for modeling disordered regions (49), a prediction [using ColabFold with the default parameters (50)] gave CH1_PA as mostly alpha helix, with a high confidence score, in contrast to CH1 (SI Appendix, Fig. S5 C and D). Our experiments contradict this finding: Free CH1_PA is not fully alpha helical; even at 0 °C it is clear that formation of a complete and stable helix is not possible (Fig. 4C), presumably due to repulsions between lysine side chains that are too great even with the stagger. However, CH1_PA is likely to compress further on binding DNA as the phosphate backbone reduces the repulsions, and indeed there are possible indications of alpha helix stabilization in the CD spectra of the droplet phase (Fig. 6B). It is possible therefore that AlphaFold is predicting conditional folding (49). CH1_PA has 111 residues, and would therefore stretch to 34 nm if 100% PPII, but compress to 17 nm if fully alpha helical, some way closer to the length of dsDNA with the same overall charge (20 bp, 7 nm). A closer match in charge density and rigidity to DNA is likely to be highly advantageous for binding and further condensation, and the three-dimensional nature of a helix could provide a more favorable template for higher-order structure formation. We propose that the ability to regulate charge density by compression in this spring-like manner underlies the two main differences that we observe between DNA condensation by CH1 and CH1_PA: It becomes enthalpy driven, and occurs at lower protein:DNA stoichiometries, i.e., CH1_PA is more effective at charge-driven condensation, despite it having the same formal net charge and net charge per residue as CH1.

Long Proline-Free Regions Are Not Uncommon in Linker Histone Tails.

Clues as to the possible biological relevance of proline across CH1s can be found by inspection of our curated database. Fig. 7 shows the length and incidence of proline-free regions in each CH1 across the set, including the three known avian CH5s, alongside net charge per residue (NCPR) and fraction arginine. Most proline-free regions are short (shown by the clustering of points along the bottom of the graph at ≤ 20 residues), as for H1.11L. However, several H1s have longer regions, of up to 213 residues. As for CH1_PA, these are generally predicted to be helical by AlphaFold, with varying confidence scores. [A notable exception is the H1 from the fungus Ashbya (#81 H1_ASHGO) that might be expected to be helical given the proline-free region of 37 residues, however, an unusually high valine content (0.13, mean is 0.05, σ = 0.015) may prevent helix prediction in the proline gap by AlphaFold due to its lower incidence in helices (51)]. Closer inspection of the helix-containing CH1s and their annotations (where present) reveals that many are thought to have gene-repressive functions due to their expression in sperm or the testis e.g., #78 H1FNT_RAT and #84 H1FNT_MOUSE, the testis-specific H1s in rats and mice essential for normal spermatogenesis and male fertility (gaps 213 and 189, respectively, predicted helical). Helix is also predicted for #14 H1_PARAN, the H1 from the sperm of Parechinus angulosus, a sea urchin, which seems likely given the experimental observation of helix in the closely related Echinus esculentus (43). The Parechinus H1, as well as a 57-residue proline-free region that like CH1_PA is very alanine-rich (0.47), also contains a moderate level of arginine overall (0.08), which could act together with the proline-free region to confer more robust condensation. Sea urchins are echinoderms, which are unique among eukaryotes in that their sperm is packaged without the use of protamines. Their sperm-specific H1s have known gene-repressive function (52), where the condensed state must presumably be resilient to the ionic conditions found in the sea to protect the paternal DNA. However it is by no means the case the drivers of condensation established here (arginine, positive charge density by net charge per residue, or by proline-free regions linked to helix formation) are all three present in gene-repressive H1s, e.g., the testis-specific #82 H1FNT_HUMAN has high arginine (0.24), a fairly long proline-free region (33 residues, predicted helical), but a relatively low charge density (NCPR 0.24). While it could be the case that arginine, NCPR and proline-free regions are three alternative ways for an organism to achieve the same outcome evolutionarily, it could also be the case that they are differently regulatable, conferring situation-specific advantages. In support of this, proline is required for the cyclin-dependent kinases that drive cell-cycle-dependent decondensation by phosphorylation (consensus site [S/T]Px[K/R]) (53). We speculate that a proline-free region that is inert to this particular mechanism may have a functional niche in a cell-cycling background, a concept not unlike the recent report of “lysine deserts” that prevent adventitious ubiquitylation (54).

Fig. 7.

Proline-free regions across linker histones. (Top) All proline-free regions across the curated set of 94 CH1s are shown as dots corresponding to their length in residues, vs. record number in the database (SI Appendix, Fig. S1). The three known avian CH5s are also shown on the Right. (Bottom) Net charge per residue, and net arginine per residue (fraction arginine). Proteins with long proline-free regions and known gene-repressive functions (see text) are boxed in gray and labeled. CH1 and CH5 are boxed and labeled in red and green, respectively.

Physiological Relevance of the Model System.

Liquid-like behavior in chromatin depends on the length scale of the measurement, a topic that has been recently and thoroughly reviewed (55). It follows that the length of the DNA also impacts the observed behavior (56, 57): In vivo, long genomic DNA may form a constricted elastic gel scaffold on the micron scale that supports condensation of chromatin binding proteins (58). However, these structures, while globally constrained, retain high levels of local dynamics (10, 59). A question that naturally arises is whether in vitro studies with short DNA can be extrapolated to understand the local (nm-scale) properties of full-length chromatin. Shorter lengths of DNA (such as short oligos or in vitro 12-nucleosome arrays) are less elastically constrained and undergo phase separation into discrete μm-scale droplets (16, 20, 57, 60, 61). The same is true of fragmented chromatin: A recent study showed that nuclease-driven fragmentation of mitotic chromosomes resulted in droplets, but did not change the chromatin density, compaction state, and resistance to perforation by microtubules (62). Taken together, these results demonstrate that the global physical constraints imposed by long DNA scaffolds do not necessarily define the local physicochemical properties, given these can be preserved when short oligos are released and become able to form droplets due to their relative translational and rotational freedom. The droplet state is therefore diagnostic of a condensation propensity, and its physicochemical properties can to some extent report on the local properties of chromatin. Historically, the use of in vitro models such as 12-nucleosome arrays has been justified on the basis that they show the same morphology (63), small angle X-ray scattering profiles, and DNA accessibility (64) as endogenous chromatin. Our minimal model excludes nucleosomes, which is a necessary simplification to be able to isolate the tail/linker DNA structure and interaction parameters by ITC, CD, and NMR. However, we note that thermodynamically, the model accounts for most of the free energy of condensation [the contribution that nucleosome–nucleosome interactions make to condensation is relatively weak (~–1.6 kcal/mol) (65) compared to CH1/DNA interactions, which are at least 5× greater, and often orders of magnitude more, depending on the context (16, 66, 67)]. Overall, our findings support the utility of in vitro models with short DNAs as they can help to rationalize trends in the material properties of in vivo chromatin at short length scales.

Conclusion

We have established a thermodynamic and structural basis for the range of DNA condensing properties of linker histones. In particular, we can rationalize the behavior of those linked to gene-repressive functions in terminally differentiated cell types such as avian erythrocytes, and in the generation of sperm. We find that the predicted impact of high charge density and the replacement of lysine by arginine are both underpinned by enthalpic contributions that promote binding and condensation to a more ordered state. Further, we find a profound effect is seen for linker histones containing long proline-free regions, which is likely due to their higher achievable charge density through compact alpha helical structures. Proline-free regions therefore provide an alternative thermodynamic switch to regulate condensation, structurally distinct from those with high charge density and arginine, and may offer further distinct properties in their regulation due to their immunity to the action of cyclin-dependent kinases.

Materials and Methods

Protein Expression and Purification.

Codon-optimized sequences for the various CH1s were obtained by gene synthesis. Proteins were expressed and purified as described previously for CH1 (16), although the yields were severely reduced for several mutants, and both CH5 and CH1_PA required higher salt to elute from the 5 mL HiTrap Sepharose high performance column. Salmon protamine was purchased from Sigma. Oligonucleotides (including FAM-labeled DNA) were purchased from Merck or Integrated DNA Technologies, and annealed as described previously (16).

Concentration Measurements.

Protein and DNA concentrations were determined spectrophotometrically using a NanoDrop One^C instrument (ThermoFisher). A₂₀₅ was used for the H1 constructs due to their low ε₂₈₀. Extinction coefficients were calculated using http://nickanthis.com/tools/a205.html based on ref. 68. Turbidity was measured in 50 or 100 μL, 1 cm path length microcuvettes by A₃₄₀.

Confocal Fluorescence Microscopy.

Protein/DNA complexes (10% DNA FAM-labeled) were prepared at 25 μM and mixed in a 1:1 ratio prior to observation on a Zeiss LSM780 confocal system mounted on an AxioObserver.Z1 inverted microscope with a 40× (numerical aperture 1.3) oil immersion PlanNeofluar using Zeiss ZEN acquisition software. For FRAP, an area on one side of the droplet was bleached with an Argon laser for 0.2 s, and an area on the other side used as the control. Images were collected at a rate (0.5 to 2.5 s^–1) and for a total time (28 to 300 s) as appropriate to sample the recovery profile. The data were averaged over five replicate measurements on different droplets, and the error bars reflect the SD for each time point. Numerical data from ZEN were fit to either a single or double exponential recovery function, [A₀ + A₁(1–exp(–(x/τ))) or (A₀ + A₁(1–exp(–(x/τ₁))) + A₂(1–exp(–(x/τ₂)))], using the NLINFIT algorithm in Matlab R2023b. Fit quality was assessed from the sum of the squared residuals (SSR).

Dynamic Light Scattering.

Triplicate experiments were performed on a Zetasizer Nano S (Malvern) at 25 °C. 1:1 mixtures were prepared spanning a range of concentrations (0.25 to 10 μM). The concentration corresponding to the onset of the droplet phase (C_sat) was clearly demarcated by a large increase in both the scattering intensity and hydrodynamic diameter extracted from the correlogram.

ITC.

Experiments were performed on an ITC200 (GE/Malvern) at 25 °C. Samples were dialyzed extensively into 10 mM sodium phosphate pH 6, 150 mM NaCl. The protein (at 100 μM for CH1s and 225 μM for protamine) was injected into the DNA (at 10 μM); 18 injections of 2 μL protein were performed, at intervals of 120 or 150 s with stirring at 750 rpm. Baseline correction and integration were performed in Origin. The experiments were each repeated 2 to 3 times, and a representative isotherm is shown.

Far-UV CD Spectroscopy.

Experiments on the free proteins were performed at ~0.1 mg/mL over a 190 to 250 nm range at the indicated temperature in 10 mM sodium phosphate pH 6, 150 mM NaF, and in 1 mm path length cuvettes. Spectra were acquired using an AVIV 410 spectrometer in 1 nm wavelength steps, averaged over three accumulations and baseline-corrected using buffer before smoothing, using the manufacturer’s software. Millidegree units were converted to mean residue ellipticity (MRE) with units deg cm² dmol^–1 res^–1 using MRE = millideg./{(no. residues – 1) × c × l × 10}, where c = molar concentration and l = path length in cm. For experiments on the protein:DNA 1:1 complexes, the wavelength range was extended to 320 nm, and a single scan was acquired per sample, to minimize any inconsistencies from droplets settling over time.

NMR.

NMR measurements were made on ¹⁵N-labeled proteins (50 to 100 μM) in 10 mM sodium phosphate pH 6, 150 mM NaCl (NMR buffer), and 10 to 15% ²H₂O. Experiments were recorded at the temperature indicated on a Bruker DRX800 spectrometer. Where possible, assignments were obtained using a conventional triple-resonance approach [HNCA, HN(CO)CA, HNCO, HNCACB, HN(CO)CACB] alongside TOCSY-¹⁵N-HSQC, NOESY-¹⁵N-HSQC, and HNN/HN(C)N experiments (69, 70). Data were processed using AZARA (v.2.8, © 1993 to 2024; Wayne Boucher and Department of Biochemistry, University of Cambridge). Triple-resonance experiments were recorded with 25% nonuniform sampling, using Poisson-gap sampling (71), and reconstructed using the Cambridge compressed sensing package and the iterative hard thresholding algorithm (72). Assignments were made using CcpNmr Analysis v. 2.4 (73). Chemical shifts were referenced to 2,2-dimethyl-2-silapentane-5-sulfonic acid.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

All study data are included in the article and/or SI Appendix.