Direct observation of structure and dynamics during phase separation of an elastomeric protein

Significance

An increasing number of proteins have been shown to undergo liquid–liquid phase separation in response to changes in their environment, resulting in formation of a dense protein-rich phase (coacervate), and plays an important role in several systems regulating the growth and development of cells and tissues. Determining the effects of phase separation on protein structure and dynamics is critical for understanding how it modulates protein function. However, structural studies have been limited by the intrinsic disorder and decreased mobility of coacervated proteins. We report direct observation of protein structure and dynamics during the phase transition of an elastomeric protein. Despite large changes in dynamics, coacervation has little effect on protein structure, such that intrinsic disorder is retained.

Abstract

Despite its growing importance in biology and in biomaterials development, liquid–liquid phase separation of proteins remains poorly understood. In particular, the molecular mechanisms underlying simple coacervation of proteins, such as the extracellular matrix protein elastin, have not been reported. Coacervation of the elastin monomer, tropoelastin, in response to heat and salt is a critical step in the assembly of elastic fibers in vivo, preceding chemical cross-linking. Elastin-like polypeptides (ELPs) derived from the tropoelastin sequence have been shown to undergo a similar phase separation, allowing formation of biomaterials that closely mimic the material properties of native elastin. We have used NMR spectroscopy to obtain site-specific structure and dynamics of a self-assembling elastin-like polypeptide along its entire self-assembly pathway, from monomer through coacervation and into a cross-linked elastic material. Our data reveal that elastin-like hydrophobic domains are composed of transient β-turns in a highly dynamic and disordered chain, and that this disorder is retained both after phase separation and in elastic materials. Cross-linking domains are also highly disordered in monomeric and coacervated ELP₃ and form stable helices only after chemical cross-linking. Detailed structural analysis combined with dynamic measurements from NMR relaxation and diffusion data provides direct evidence for an entropy-driven mechanism of simple coacervation of a protein in which transient and nonspecific intermolecular hydrophobic contacts are formed by disordered chains, whereas bulk water and salt are excluded.

The liquid–liquid phase separation (LLPS) of molecules has long been exploited to concentrate and encapsulate molecules for drug delivery and food preparation (1, 2). In biology, there is increasing awareness that many proteins exhibit this type of phase behavior, allowing transient microenvironments to be quickly assembled and disassembled in response to changing solution conditions, or the availability of binding partners (3, 4). Protein phase separation occurs intracellularly to generate various membraneless organelles, such as ribonucleoprotein (RNP) bodies involved in nucleic acid processing, transport, and storage (5). A similar phenomenon is observed in the extracellular matrix as a critical step in the synthesis of elastic fibers, which provide extensibility, recoil, and resilience to tissues (6, 7). In the latter system, LLPS of monomeric elastin results in hydrated protein-rich coacervate droplets that are deposited and cross-linked to form an elastic matrix (7, 8). In addition to their fundamental importance to biology, the dynamic reversibility of LLPS makes phase-separated states of proteins an attractive platform for development of responsive biomaterials with broad application, for instance as scaffolds for tissue engineering or as carriers in drug delivery systems (9–11).

Despite the keen interest in proteins that undergo stimulus-responsive phase separations, the highly disordered nature of proteins known to undergo LLPS, as well as unfavorable dynamics caused by increased viscosity in the phase-separated state, has impeded high-resolution studies. Thus, the molecular structure and dynamics of the phase-separated states, as well as the mechanisms through which they form, remain poorly understood aside from a small number of standout examples. In general, LLPS requires that transient protein–protein interactions become favored over interactions with the bulk solvent and takes two general forms: complex and simple coacervation. Although both have been observed for intrinsically disordered proteins (IDPs), the sequence requirements and physical basis of each type of LLPS are distinct (4, 12).

Several recent studies have shown that a combination of electrostatic, cation-π, and hydrophobic interactions can drive complex coacervation of intrinsically disordered RNP-associated proteins, highlighting the enthalpic nature of this type of LLPS (13–15). Conversely, simple coacervation is considered an entropically driven process that depends entirely on hydrophobic interactions between protein chains (4, 16), although few molecular details of this process have been reported. Thus, a detailed understanding of protein structure and dynamics during LLPS is needed to better define the principles governing protein phase separation and to further develop applications of this type of self-assembly in biomaterial design.

The elastin monomer, tropoelastin, undergoes a well-characterized lower critical solution temperature (LCST) phase transition that is promoted by the presence of salt and heat (8, 17, 18). Tropoelastin is an IDP consisting of an alternating arrangement of hydrophobic domains (HDs) and cross-linking domains (CLDs) (7, 19). The HDs are required for coacervation and elasticity and are disordered domains composed of pseudorepetitive sequence motifs enriched in glycine, proline, and hydrophobic amino acids (20–26). In contrast, the CLDs provide structural integrity to polymeric elastin through covalent cross-linking of lysine residues (27, 28). Current knowledge of elastin structure comes primarily from low-resolution spectroscopic and solid-state NMR studies (29–35), or from modeling and theoretical data (20, 22, 24, 36–38), with few molecular details of the elastin LLPS having been reported.

We, and others, have shown that elastin-like polypeptides (ELPs) derived from the native sequence of elastin act as suitable proxies to study phase separation and elasticity (18, 28, 31, 39). In the present study, we highlight the design and characterization of an elastin mimetic that undergoes phase separation and can be cross-linked to create resilient biomaterials whose assembly can be monitored by NMR to obtain site-specific resolution of structure and dynamics of both the phase-separated and cross-linked materials. Here, we demonstrate that, whereas the phase-separated state of an ELP is dynamically heterogeneous, local structural elements present within the hydrophobic domains remain largely unchanged from monomeric protein through coacervation to cross-linked elastomeric material. These results not only provide mechanistic insight into the LLPS behavior of elastin with an unprecedented level of detail, but also deepen our fundamental understanding of simple coacervation.

Results

Design of an ELP for Site-Specific Monitoring of Structure and Dynamics.

Atomic resolution studies of tropoelastin have primarily been inhibited by its intrinsic disorder combined with the degenerate yet repetitive nature of its amino acid sequence. These properties not only preclude the use of crystallography and electron microscopy, but also lead to substantial overlap in NMR spectra, inhibiting resonance assignment. To circumvent these challenges, we designed a simplified elastin mimetic that facilitated NMR studies while retaining the fundamental biochemical properties of tropoelastin: the alternating domain architecture, intrinsic disorder, the ability to undergo liquid–liquid phase separation, and the capacity to form elastic biomaterials when cross-linked at lysine residues.

This recombinantly expressed ELP contains an alternating arrangement of CLDs and HDs; schematically represented as CLD−(HD−CLD)_n and represented in the text as ELP_n, with n being the number of repeating units (here n = 1–3; SI Appendix, Fig. S1). These CLDs and HDs contain repetitive motifs that are representative of those observed in all amniote tropoelastin sequences analyzed to date, including in humans (19). The CLD sequence is AAAAAKAAKYGA (numbered 1–12 for NMR resonance assignment), whereas each 35-residue HD sequence was composed of GVPGV (numbered G1 through V5) repeated seven times.

ELP₃ Recapitulates the Secondary Structure and Material Properties of Tropoelastin.

Far UV circular dichroism (CD) spectroscopy of monomeric ELP₃ in aqueous solution gave results consistent with previous studies of tropoelastin and elastin-like polypeptides (28, 30, 31) (Fig. 1A). The characteristic weak minimum ellipticity at 222 nm and strong minimum near 200 nm are consistent with the presence of a small amount of α-helical structure in a protein that is otherwise intrinsically disordered.

Fig. 1.

The intrinsically disordered ELP₃ recapitulates the phase behavior and elasticity of tropoelastin materials. (A) CD spectra of ELP₃ and tropoelastin monomers in 50 mM sodium phosphate, pH 7 at 20 °C, contain features that are indicative of highly disordered proteins. (B) Bright-field microscopy of temperature-induced phase separation of ELP₃ and tropoelastin in 1.5 M NaCl solution after 1 h of incubation at 50 °C. (White scale bars, 32 μm.) (C) Mechanical properties of polymeric ELP₃ and tropoelastin coacervates cross-linked at lysine residues by genipin. Error bars correspond to the SEM.

ELP₃ coacervated and formed a protein-rich liquid phase under conditions of heat and salt similar to those established for other ELPs and for tropoelastin (31, 32) (Fig. 1B). Cross-linking coacervated ELP₃ resulted in elastic materials with tensile properties remarkably similar to materials formed from tropoelastin (40, 41): a low elastic modulus, high strain to break, and low energy loss (Fig. 1C). The stress relaxation of ELP₃ was greater than that of tropoelastin, but remains in the normal range for elastins and ELPs (41, 42). Taken together, these data demonstrate that ELP₃ recapitulates the fundamental macroscopic properties of the monomeric, coacervated, and cross-linked forms of tropoelastin.

NMR Resonance Assignment of Monomeric and Phase-Separated ELP₃.

Recombinant ELP₃ was uniformly labeled with ¹⁵N and ¹³C to facilitate characterization by NMR spectroscopy. The ¹H-¹⁵N heteronuclear single quantum coherence (HSQC) spectrum of the monomer showed surprisingly well-resolved peaks with relatively narrow line widths (Fig. 2A). After inducing phase separation, the ¹H-¹⁵N HSQC spectrum contained two overlapping sets of resonances, one with ¹HN and ¹⁵N chemical shifts similar to those of the monomer and a second set of broader peaks with a small global downfield shift of the ¹⁵N and ¹HN resonances (Fig. 2B). To allow independent resolution of these two overlapping sets of resonances, we used HSQC pulse sequences designed to selectively remove either resonances with fast R₂ relaxation rates or resonances belonging to proteins with fast translational diffusion rates (Fig. 2C). The spectra clearly demonstrate the presence of two distinct species, one that diffuses quickly with slow R₂ relaxation, and a second species with enhanced R₂ relaxation but significantly slower diffusion. These two ELP₃ populations were identified as residual monomers within the protein-poor bulk phase, and coacervated ELP₃ in the protein-rich phase, respectively, through analysis of NMR relaxation and pulsed field gradient (PFG) diffusion rates (detailed below).

Fig. 2.

The unique dynamics of ELP₃ monomers and coacervate facilitate NMR spectroscopy of the phase transition. (A) The ¹H-¹⁵N HSQC NMR spectrum of 4.7 mM ELP₃ monomer in 50 mM sodium phosphate, 300 mM NaCl, pH 7 at 37 °C. Labeled peaks, corresponding to the amino acids in the repeating sequence motifs of the HDs and CLDs, were assigned as described in Materials and Methods and SI Appendix, Fig. S1. To remove temperature effects on chemical shift and protein dynamics, phase separation was induced at 37 °C by increasing the concentrations of both ELP and NaCl. (B) HSQC spectrum of a 7.8-mM ELP₃ sample undergoing liquid–liquid phase separation in a buffer consisting of 50 mM sodium phosphate, 600 mM NaCl, pH 7 at 37 °C. (C) HSQC spectra of the same phase-separated sample shown in B, acquired using an R₂ relaxation rate filter to remove signals with R₂ > 5 s⁻¹ (red) or a PFG diffusion filter (removing objects with a diffusion rate faster than 10⁻⁷ cm²⋅s⁻¹) (blue) to distinguish residual monomeric ELP₃ (red) from coacervate (blue).

Complete backbone resonance assignment of monomeric ELP₃ was achieved using standard three-dimensional triple resonance NMR experiments, with additional experiments to aid in assignment of proline-rich segments. Representative NMR spectra, showing partial assignment of CLD resonances, are provided in SI Appendix, Fig. S2, and resonance assignments are listed in SI Appendix, Table S1. Due to the repetitive nature of the sequence and the high degree of disorder, most GVPGV repeats gave rise to an equivalent set of NMR signals, with only those residues adjacent to a CLD showing inequivalence. Similarly, only the N- and C-terminal segments of the first and last CLDs gave rise to distinct resonances, suggesting that the CLDs are also largely equivalent. An additional set of weak resonances was identified in NMR spectra of monomeric ELP₃, arising from GVPGV repeats containing cis-proline (SI Appendix, Fig. S3).

The same methodology was used to obtain near complete assignment of backbone resonances for both the residual monomer and the coacervated species present after LLPS (SI Appendix, Table S2). The increased rates of R₂ relaxation exhibited by the slow-diffusing species, combined with a global increase in NMR line widths, led to some ambiguity in sequential assignment of the polyalanine CLD sequences for coacervated ELP₃. The global shift in ¹HN and ¹⁵N chemical shifts observed for the coacervate relative to monomeric ELP₃ likely arise from a decreased ionic strength in the coacervate droplets, due to partial exclusion of NaCl (Fig. 3), rather than representing a global structural change.

Fig. 3.

The ¹HN chemical shift data reflect altered solution conditions within ELP₃ coacervates. The HD peaks from ¹H-¹⁵N HSQC spectra of ELP₃ in 50 mM sodium phosphate, pH 7 with NaCl ranging from 0 to 900 mM are shown in each panel, highlighting the correlation between ¹HN chemical shifts and the concentration of NaCl. (A) HSQCs acquired at 37 °C with an ELP₃ in 50 mM sodium phosphate, pH 7 and 600 mM NaCl undergoing a phase transition are overlaid in black. (B) HSQCs acquired at 20 °C with residual monomer (coacervate was removed by centrifugation) postcoacervation overlaid in black. Changes in ELP₃ monomer ¹HN chemical shifts are plotted as a function of NaCl concentration at (C) 37 °C or (D) 20 °C, showing a clear linear dependence of chemical shift on salt concentration. Most HDs also exhibited a small protein concentration-dependent change in chemical shift, as shown by the lower than expected Δcs observed in the 0.6 M NaCl sample with 100 mg/mL ELP₃. Linear least squares regression of Δcs (red lines) for the site with the smallest protein concentration dependence (G4) was used to estimate the NaCl concentration experienced by residual monomer (red star) or coacervate (red diamond) after LLPS.

ELP₃ Is Highly Disordered in both the Monomeric and Phase-Separated State.

Due to the fast interconversion of conformations in IDPs, the observed NMR chemical shifts represent a weighted average within a structural ensemble (43, 44). Although the presence of a dynamic ensemble precludes determination of a single high-resolution structure, several methods have been developed for prediction of site-specific disorder and secondary structure propensity within a polypeptide sequence, using chemical shift data (45, 46).

The sequence-specific secondary structure and disorder propensities of monomeric and coacervated ELP₃ were determined using the δ2D chemical shift scoring web-server (45). Aside from the salt effect described above, the backbone and side chain chemical shifts were essentially identical in the monomer and coacervate, strongly suggesting that no significant structural changes occurred during LLPS of ELP₃. This finding is supported by the δ2D analysis shown in Fig. 4, which predicts a high degree of disorder in both the monomer and coacervate across the entire sequence. The CLDs displayed very low α-helical propensity, as expected based on our CD data, whereas the HD domains had partial β-strand character.

Fig. 4.

The intrinsic disorder of monomeric ELP₃ is retained after coacervation. (A) Secondary structure propensities from δ2D analysis of NMR chemical shifts. The random coil propensities for each position in the CLDs (i) and HDs (ii) are shown, as are the α-helix propensities for each position in the CLD sequence (iii) and the β-sheet propensities for HD repeat sequences (iv). (B) Schematic representation of the two β-turns identified in the ELP₃ NMR data. The colored dashed lines represent the characteristic connectivities between ¹H atoms that were used as identifiers of each turn in NOESY spectra of ELP₃. (C) Summary of HN-Hα NOEs observed in the HD sequence of the ELP monomer at 20 °C. The bar thickness in each case represents the relative intensity of the NOE cross-peak, with the colored bars corresponding to the dashed lines in A.

To complement the chemical shift analysis of secondary structure propensity, through-space interactions between ¹H nuclei were measured using 2D ¹H-¹H nuclear Overhauser effect spectroscopy (NOESY) experiments for both the monomer and coacervated ELP₃ (Fig. 4 and SI Appendix, Fig. S4). As for the ¹H-¹⁵N HSQC spectrum, the ¹H-¹H NOESY of the coacervate displayed broader line widths than the monomer spectra, but otherwise no significant differences were observed, supporting retention of monomer structural properties in the coacervate. Analysis of NOEs arising from short-range interactions between protons in the protein backbone support the formation of two distinct type-II β-turns in the HDs of monomeric and coacervated ELP₃, centered on the VPGV and GVGV sequences (Fig. 4 B and C). Importantly, all of the β-turn NOEs were much weaker than expected, relative to reference NOEs, suggesting that these structural elements are only populated 20–40% of the time. Because each site only gives rise to a single, narrow NMR resonance, the rate of exchange between turns and extended conformations is rapid: on the microsecond–nanosecond timescale. Note that, whereas multiple turns could be populated simultaneously, their transient nature and low structural propensity discounts the formation of the β-spiral structure proposed in earlier models of elastin HDs (38). No NOEs characteristic of α-helices were observed for the CLD, supporting the low structure propensity predicted by δ2D.

To probe for intermolecular contacts, ¹³C and ¹²C edited ¹H-¹H NOESY spectra were recorded for monomeric and coacervated samples containing a 1:1 mixture of unlabeled and ¹³C labeled ELP₃ (Fig. 5). Monomeric ELP₃ exhibited only intraresidue NOEs in addition to the short-range backbone NOEs identified above. After LLPS, several strong intermolecular NOEs were observed between valine ¹Hγ and side chain ¹H resonances of alanine, valine, and proline (Fig. 5C). Additional NOEs are also observed between valine ¹Hγ and ¹Hα of glycine, valine, and proline. These data are consistent with the presence of multiple nonspecific intermolecular hydrophobic contacts within the coacervate. The observed interactions between valines and alanines suggest that, whereas the HDs are required for LLPS of ELPs, CLD sequence composition may influence coacervation, supporting previous studies of CLD mutants (28, 31).

Fig. 5.

Intermolecular hydrophobic contacts are formed in the phase-separated state of ELP₃. The aliphatic regions of ¹³C double half filtered ¹H-¹H NOESY spectra, acquired at 37 °C, are shown for phase-separated (A) and monomeric (B) ELP₃. For each sample, spectra show NOEs between ¹H-¹³C groups [double ¹³C selected (i)]; between ¹H-¹²C and ¹H-¹³C groups [¹³C selected and filtered (ii); ¹³C filtered and selected (iii)]; or between ¹H-¹²C groups [double ¹³C filtered (iv)]. Expansions showing NOEs between Val Hγ and Val Hβ, Pro Hγ, or Ala Hβ are shown in C. For the monomeric sample, only intramolecular contacts (i and iv) within Val side chains are observed in C, and some additional weak intraresidue NOEs are observed (B, iv). In the coacervate, intense NOEs are observed in all spectra, indicating the presence of intermolecular Val–Val, Val–Pro, and Val–Ala contacts. The slight increase in coacervate Val–Val NOEs observed in i and iv is due to the additional presence of intraresidue NOEs in these spectra, as seen for the monomeric ELP₃.

The Structural Propensities of the ELP₃ HD Are Retained in Cross-Linked Materials.

Magic-angle spinning (MAS) solid-state NMR was used to obtain ¹³C chemical shift data for both coacervated and cross-linked ELP₃. Two ¹³C-¹³C 2D correlation experiments were used, incorporating either an ¹H-¹³C insensitive nuclei enhanced by polarization transfer (INEPT) transfer for initial ¹³C excitation, or ¹H-¹³C cross-polarization (CP). The former experiment selects for highly mobile sites within a protein, whereas the latter will preferentially give rise to spectra of relatively immobile segments (47).

The bulk of HD ¹³C signals are observed after INEPT transfer (SI Appendix, Fig. S5), suggesting that these domains remain highly dynamic in the cross-linked material. The chemical shifts in each state are nearly identical, indicating that the structural propensities of HD in the free monomer are retained even after cross-linking. Highly mobile CLDs observed in the INEPT spectrum of coacervated ELP₃ have chemical shifts consistent with a partially helical structure, matching the solution NMR data. In the CP spectrum, a new population of less mobile CLD alanines is observed, with chemical shifts indicative of β-sheets, suggesting a mechanism through which CLD might assemble before cross-linking. As we have reported for another ELP, all CLDs convert to relatively rigid α-helices after cross-linking (31).

Coacervated ELP₃ Exhibits Reduced Global Diffusion, but Retains Local Disorder.

Diffusion and R₂ relaxation-edited HSQCs showed that LLPS of ELP₃ resulted in two distinct populations with very different dynamics. Detailed analysis of these species was carried out to determine the global and local effects of coacervation on the dynamics of the individual protein chains. ¹⁵N R₁ and R₂ relaxation rates and ¹H-¹⁵N heteronuclear NOEs were obtained for the backbone amide groups of monomeric ELP₃, residual monomer post-LLPS, and the coacervate species at 37 °C (SI Appendix, Fig. S6). A general increase in relaxation rates and heteronuclear NOE intensity was observed for all sites in the coacervated species across the sequence, with no discernible local effects, consistent with either a global decrease in fast motions throughout the protein chain, or a reduction in the rate of rotation of the entire protein.

To narrow down the source of increased NMR relaxation rates for coacervated ELP₃, R₁ and R₂ were measured as a function of temperature and both protein and NaCl concentrations. These data were correlated with PFG diffusion measurements made under the same conditions (Fig. 6). The translational diffusion rates of monomeric ELPs of different sizes (ELP_1–3, SI Appendix, Fig. S1) and ELP₃ coacervate are shown in Fig. 6 A, i. Upon coacervation, the diffusion rate of ELP₃ decreased by two orders of magnitude, whereas monomers exhibited increased rates of diffusion with increased temperature and decreased molecular weight. R₁ and R₂ relaxation (Fig. 6 A, iii and v) follow similar trends, indicating that altered chain dynamics are driven by global rather than local effects. Remarkably, in the coacervated protein, relaxation rates continue to increase at temperatures beyond the phase transition, and diffusion continues to slow. Although this observation runs counter to what is expected for molecules with greater kinetic energy, it is consistent with an entropically driven strengthening of the hydrophobic effect, leading to increased protein–protein interactions even after LLPS has occurred.

Fig. 6.

NMR relaxation and PFG diffusion measurements reveal temperature- and time-sensitive dynamics of phase-separated ELP₃. (A) Temperature dependence of ELP₃ PFG diffusion rates (i and ii) and ¹⁵N R₂ or R₁ relaxation rates (iii–vi). Odd-numbered panels describe the extreme differences in both relaxation properties and translational diffusion between monomeric ELP₃ and ELP₃ within a coacervate. Even-numbered panels display the relatively subtle differences in monomer ELP₃ dynamics caused by variations in protein or salt concentration. Least squared linear regression fits (solid or dashed lines) of the monomer at high and low protein concentrations in the absence of NaCl show the effect of viscosity on these measurements (ii, iv, and vi). The phase transition temperature for a 4.7-mM sample of ELP₃ is indicated in each plot by a vertical dashed line. Error bars indicate relaxation rate fitting error or SEM for diffusion measurements. (B) Time-dependent changes in the diffusion and R₂ relaxation rates (mean of G1, V2, G4, and G5 data) of the monomeric and coacervate species in an 8.7-mM sample of ELP₃ held at 37 °C (phase transition temperature is 30 °C) for 6 d. Error bars indicate the SEM.

ELP and elastin coacervate droplets have been shown to increase in stability if they are held above the coacervation temperature for an extended period (18, 48, 49). This maturation process has been hypothesized to occur through increased ordering of ELPs at the surface of the droplet, resulting in a less reversible LLPS. In the present work, coacervated ELP₃ exhibited an increase in R₂ and a decreased rate of diffusion over 6 d following coacervation, suggesting a time-dependent decrease in chain mobility that may be related to maturation (Fig. 6B).

The global reduction in ELP₃ dynamics upon LLPS, combined with the lack of structural change, suggests that protein chains within the coacervate experience an environment similar to a highly concentrated monomer solution. Thus, by comparing the diffusion rate of ELP₃ within the coacervate with that of high concentrations of monomer, we were able to estimate the protein concentration inside the coacervate phase (SI Appendix, Fig. S7 and Table S3). This estimation gave values of ∼500 mg/mL ELP₃ in a protein-rich phase containing ∼62.5% water by mass. This is insufficient water for complete solvation of all ELP₃ monomers, highlighting the exclusion of nonessential water and the increase in protein–protein interactions in the coacervate.

A similar increase in HD interactions is observed for monomeric ELP₃ chains at elevated temperature and NaCl concentration. When protein concentrations are kept below the threshold for LLPS, the monomer exhibits increased rates of diffusion and reduced R₁ and R₂ (Fig. 6 A, ii, iv, and vi). These data are consistent with increased compaction of the molecule and suggest that the same forces driving coalescence lead to an intramolecular collapse of the HDs when intermolecular interactions are not present.

Discussion

Through a combination of solution and solid-state NMR data, we have tracked the structure and dynamics of an ELP through LLPS and subsequent cross-linking into an elastic material. The resulting data have allowed us to develop a detailed model for ELP self-assembly that builds on early biophysical studies (20–22) and that can serve as a general template for understanding simple coacervation of proteins (Fig. 7). Consistent with this model, recent molecular simulation studies indicate that HDs are disordered and hydrated (25), and that their self-assembly entails formation of intermolecular nonpolar contacts while preserving local secondary structure (50). A frustrated hydrophobic collapse of the HDs is promoted by the increased entropic penalty of solvating these chains under conditions of high salt and temperature. At high protein concentrations, this will lead to intermolecular assembly and formation of a separate phase.

Fig. 7.

Model for ELP self-assembly. Our findings provide support for a comprehensive model for ELP self-assembly in which elevated temperature and salt increase the entropic cost of solvating hydrophobic chains, driving phase separation. Because the protein sequence precludes adoption of a stable folded state, the resulting phase-separated state is maintained through transient and nonspecific HD interactions between monomers. Whereas the coacervate remains hydrated, bulk water is excluded and there is a reduction in Na⁺ and Cl⁻ ions (red dots), slightly increasing their concentration in the bulk solution. The coacervated ELP can be chemically cross-linked to form an elastic material. HDs remain dynamic and highly disordered throughout the assembly process, whereas CLD motion slows with each step. Partially helical in the monomers (blue cylinders), CLDs exhibit an increased propensity to form β-sheets (green arrows) in the coacervate and become stable α-helices in cross-linked materials.

The inability of Pro/Gly-rich HD sequences to form stable secondary structures (24) results in retention of highly disordered HDs in the coacervated state. This lack of stable structure requires the inclusion of sufficient water to satisfy backbone hydrogen-bonding requirements and potentially to solvate hydrophobic chains in between transient protein–protein interactions mediated by the HDs. All unnecessary water is excluded, as dictated by the entropic cost of its inclusion in the protein-rich phase. The increased ionic strength of the protein-poor solvent phase, due to partial exclusion of salt from the coacervate droplets, also favors coacervation, leading to a reduction in residual monomer. The CLDs remain primarily disordered after coacervation, forming transient helices and small amounts of β-sheet, bringing CLDs together before cross-link formation. After chemical cross-linking, a solid phase is formed in which the HDs retain their essential properties of disorder and high entropy. All structural elements in the HDs remain transient and closely resemble those present in the monomeric ELP, whereas the CLDs adopt a stable helical conformation, as we have previously reported (31).

Whereas the CLD behavior in this model is likely unique to elastin and ELPs, the data presented for the HDs are likely to be more broadly representative of proteins and polymers that undergo LLPS through simple coacervation. Thus, our ability to monitor this process with site-specific resolution not only informs our understanding of elastin assembly, but also allows us to predict how sequence or chemical structure might influence LLPS across many different systems. This approach provides a means to probe important biological questions and also to develop biopolymers with fine-tuned properties of structure and assembly.

Materials and Methods

Protein Expression and Purification.

Expression and purification of recombinant tropoelastin was carried out using previously reported methods (32). All ELPs were expressed in BL21(DE3) Escherichia coli from a pet32-b plasmid (Invitrogen) containing the ELPs fused to an N-terminal thioredoxin, which was removed by cyanogen bromide digestion in 70% formic acid. Each ELP was purified by ion exchange FPLC followed by reverse phase HPLC as described previously for other elastin-like polypeptides (31). When isotopic enrichment of polypeptides was needed for NMR experiments, proteins were expressed from E. coli grown in M9 minimal media supplemented with ¹⁵N-NH₄Cl and ¹³C-glucose as the sole sources of nitrogen and carbon.

CD Spectroscopy.

CD experiments were performed using a 1.0-mm cuvette in a Jasco J-810 spectrometer at 20 °C. Protein samples were dissolved in 50 mM sodium phosphate, pH 7.0, to a final concentration of 20 μM or 4 μM for ELP₃ and tropoelastin, respectively. Three scans from 260 nm to 190 nm, recorded with 4 s of signal averaging at each point, were performed to obtain the mean spectra shown.

Light Microscopy.

Phase separation of 100 µM tropoelastin and 100 µM ELP₃ was monitored in real time using a Zeiss Axiovert 200 inverted epifluorescence microscope with a temperature-controlled Attofluor cell chamber (Molecular Probes). NaCl was added to a final concentration of 1.5 M. The sample chamber was held constant at 50 °C. Samples were imaged for 90 min using differential interference contrast mode, with images saved after 60 min of incubation.

Tensile Mechanical Measurements.

Construction of genipin cross-linked materials from ELP₃ and tropoelastin, as well as their mechanical testing and associated data analysis, was carried out as previously described (31, 32, 51). The standard protocol for polymeric material cross-linking was modified to adjust for the phase behavior of ELP₃ and maximize the homogeneity of the material thickness. All protein solutions, reagents, and centrifuges were prewarmed to 37 °C. A total of 5 mg of ELP₃ was dissolved in 325 μL sodium borate buffer (20 mM, pH 8.0). A total of 125 μL of 5 M NaCl was added to the protein sample in 10-mm² glass-bottom wells. After 5 min of incubation at 37 °C, 50 μL of genipin (100 mM in ethanol) was added. Samples were immediately spun in a swinging-bucket centrifuge at 1,100 × g for 10 min and then incubated overnight at 37 °C to allow for complete cross-linking.

Solution-State NMR Spectroscopy.

All experiments were carried out on 20-µL samples using a triple-resonance 1-mm TXI MicroProbe with Z gradient and temperature control in a Bruker Avance III NMR spectrometer operating at an ¹H frequency of 600 MHz. Samples were buffered with 50 mM phosphate at pH 7.0. Protein and NaCl concentrations were varied to control the desired phase transition temperature. All phase separation was induced in the spectrometer by raising the temperature at the probe. No further mechanical manipulation of the sample outside of the spectrometer was performed. Bruker TopSpin was used for data acquisition. Standard 2D and 3D Bruker pulse sequences with water suppression were used for most experiments. All spectra were processed with the NMRPipe software package (52), then visualized and analyzed with the CcpNmr analysis program (53).

Chemical Shift Assignment and Secondary Structure Prediction.

Chemical shift assignment of monomeric ELP₃ was carried out on a sample containing 4.7 mM ELP₃ and 300 mM NaCl at 20 °C and 37 °C. Chemical shift assignment of the phase-separated state was performed on 7.8 mM ELP₃ with 600 mM NaCl at 37 °C. Both the monomer and coacervate ¹H, ¹³C, and ¹⁵N chemical shifts were assigned using standard 2D and 3D NMR methods. The ¹H-¹⁵N HSQC, HNCA, HN(CO)CA, HNCO, and HN(CA)CO experiments (pulse sequences as provided by Bruker Biospin) were used for sequential backbone assignments, along with an HCAN experiment (54), which was required to obtain the backbone ¹⁵N chemical shifts of proline residues and to facilitate assignment of proline-rich segments. A standard 3D HCCH-TOCSY experiment was used to achieve full assignment of side chain chemical shifts for the monomeric protein at 20 °C. The peak volumes observed in the ¹H-¹⁵N HSQC spectrum of the ELP₃ monomer correlated well with the sequence abundance of each assigned backbone amide resonance (SI Appendix, Fig. S3), helping to validate this approach to chemical shift assignment of a polymer-like protein.

The secondary structure propensity of the major repeating units of the cross-linking and hydrophobic domains in both the monomer and coacervate was predicted using the δ2D structure determination web server for intrinsically disordered proteins (45), with the chemical shifts for ¹HN, ¹⁵N, ¹³Cα, ¹³CO, and ¹Hα as input parameters. In mixed samples containing both monomer and coacervate, assignment of chemical shifts from each state was facilitated by ¹H-¹⁵N HSQC experiments incorporating either a PFG diffusion or T₂ relaxation-based filter to eliminate signals from monomer or coacervate, respectively. Backbone amide chemical shifts exhibited a linear change with temperature, allowing ¹H-¹⁵N HSQCs taken during temperature series experiments to be assigned by simple translation of the chemical shift assignments at 20 °C and 37 °C.

¹H-¹H NOESYs.

Homonuclear ¹H NOESY experiments on ELP₃ monomer and coacervate samples were recorded using a phase-sensitive experiment with WATERGATE water suppression (55). Experiments with mixing times of 50, 100, and 150 ms were recorded and used to determine the linear region of the NOE buildup curve. Experiments with 100-ms mixing times were then used for integration of peak volumes and determination of NOE intensities. The abundant number of internal proline ring NOEs, which have known ¹H-¹H distances, were used to establish the relationship between NOE intensity and distance in each sample. NOE intensities of the monomer at 20 °C and 37 °C were grouped as strong (<2.4 Å), medium (≥2.4, ≤3.5 Å), or weak (>3.5 Å, <5 Å). Distance measurements were not performed on the phase-separated sample because chemical shift overlap did not allow for reliable peak assignment and integration analysis.

Intermolecular NOEs were obtained at 37 °C using an ¹H-¹H NOESY experiment with a ¹³C double half filter (56), and a 100-ms mixing time. Samples for these experiments contained a 1:1 mixture of unlabeled and ¹³C-labeled ELP₃, giving final conditions of 1.6 mM ELP₃ in 50 mM sodium phosphate and 300 mM NaCl at pH 7 (monomer sample) or 7.8 mM ELP₃ in 50 mM sodium phosphate and 600 mM NaCl at pH 7 (phase-separated sample).

Pulsed Field Gradient Diffusion.

PFG diffusion data were recorded using a stimulated echo experiment with bipolar gradient pulses and WATERGATE solvent suppression. The 32 ¹H spectra were acquired at varying gradient strengths between 2% and 98% of the maximum value, while keeping time delays constant throughout each experiment. To obtain the diffusion coefficient D, peak intensity change as a function of gradient strength (g) was fit to a standard equation for bipolar gradients (57):

$$I\left(g\right)=I\left(0\right)e^{-D{\left(\gamma g\delta \right)}^2\left(\mathrm{\Delta }-\frac{\delta }{3}–\frac{\tau }{2}\right)},$$

where I is the observed intensity, I(0) is the unattenuated intensity, γ is the gyromagnetic ratio of ¹H, Δ is the diffusion time (100 ms), δ is the total gradient time, and τ is the time between the bipolar gradients (0.2147 ms).

The large difference in the diffusion coefficients of the monomeric and coacervate species allowed for good separation of their observed signal attenuation by running two different experiments with δ-values set to either 6 ms or 15 ms, respectively. When fitting the signal attenuation of the monomer, the remaining signal from the slower diffusing coacervate was omitted. Conversely, when fitting the signal attenuation of the coacervate, the sharp decrease in signal at the beginning of the decay curve due to loss of signal from the faster diffusing monomer was disregarded.

Dynamics Measurements.

The ¹⁵N R₁ and R₂ relaxation times were determined from pseudo-3D experiments using standard inversion recovery or spin-echo experiments, respectively (58, 59). The 13 ¹H-¹⁵N HSQC spectra were acquired in an incremental manner using delay times ranging from 50 to 4,000 μs for R₁ experiments, whereas 10 spectra were acquired in an interleaved fashion using delay times of 17–544 μs for R₂ experiments. The ¹H-¹⁵N NOEs were determined using the pulse sequence hsqcnoef3gpsi with interleaved acquisition of spectra recorded with and without proton saturation. Least squares fitting of signal attenuation to obtain R₁ and R₂ relaxation times, as well as NOE calculations from comparison of saturated and equilibrium spectra, was accomplished using default parameters in CcpNmr. The reported relaxation rates are the inverse of the relaxation times.

Solid-State MAS NMR.

Solid-state experiments were performed using a triple-resonance 4.0-mm MAS probe with temperature control in a Bruker Ascend NMR spectrometer operating at an ¹H frequency of 700 MHz. Spectra were externally referenced to the downfield ¹³C resonance of adamantane at 38.56 ppm relative to tetramethylsilane (60).

All 2D ¹³C-¹³C correlation spectra were acquired with 256 points in the indirect dimension. The π/2 pulse widths on all channels ranged from 2 to 4 μs. Cross-polarization was realized using ¹³C radio frequency field strengths from 40 to 60 kHz with contact times of 1–2 ms. All experiments on the phase-separated and cross-linked samples were performed at an MAS spinning rate of 14 kHz at ∼40 °C.

To identify regions of low mobility within ELP₃, 2D ¹³C-¹³C correlation spectra were obtained using ¹H-¹³C cross-polarization followed by proton-driven spin diffusion (PDSD) (61, 62) with a mixing time of 50 ms. To identify regions of high mobility, 2D ¹³C-¹³C correlation spectra were recorded using INEPT transfer from ¹H to ¹³C followed by ¹³C-¹³C recoupling using a total through bond correlation spectroscopy (TOBSY) scheme (63)with a mixing time of 10 ms. The ¹H decoupling fields of 100 kHz were applied during all acquisition periods using two-pulse phase-modulated (TPPM) decoupling (64).