Jan
Hansen
,
Stefan U.
Egelhaaf
and
Florian
Platten
*
Condensed Matter Physics Laboratory, Heinrich Heine University, Universitätsstraße 1, 40225 Düsseldorf, Germany. E-mail: florian.platten@hhu.de
First published on 26th December 2022
Liquid–liquid phase separation (LLPS) of protein solutions is governed by highly complex protein–protein interactions. Nevertheless, it has been suggested that based on the extended law of corresponding states (ELCS), as proposed for colloids with short-range attractions, one can rationalize not only the thermodynamics, but also the structure and dynamics of such systems. This claim is systematically and comprehensively tested here by static and dynamic light scattering experiments. Spinodal lines, the isothermal osmotic compressibility κT and the relaxation rate of concentration fluctuations Γ are determined for protein solutions in the vicinity of LLPS. All these quantities are found to exhibit a corresponding-states behavior. This means that, for different solution conditions, these quantities are essentially the same if considered at similar reduced temperature or second virial coefficient. For moderately concentrated solutions, the volume fraction ϕ dependence of κT and Γ can be consistently described by Baxter's model of adhesive hard spheres. The off-critical, asymptotic T behavior of κT and Γ close to LLPS is consistent with the scaling laws predicted by mean-field theory. Thus, the present work aims at a comprehensive experimental test of the applicability of the ELCS to structural and dynamical properties of concentrated protein solutions.
The protein–protein interactions that govern the phase separation processes are highly complex. They are determined by the specific amino acid composition of the proteins, their internal molecular organization, their overall molecular shape with a heterogeneous distribution of functionally different moieties on their surfaces as well as on the physicochemical solution conditions. It is therefore challenging to decipher a physical rationale of protein phase behavior and the underlying interactions.
Nevertheless, coarse-grained approaches inspired by soft-matter physics have been applied to the highly complex interactions between protein molecules and have proven helpful in understanding the observed phase behavior.6 For example, the DLVO theory has been applied to describe the effects of salts, solvents and pH on the interactions of protein solutions under conditions relevant for crystallization and LLPS.20–25 Colloid models have also been applied to describe the experimentally obtained structure factor of protein solutions.26–32
The state diagram of globular protein solutions exhibits striking similarities with the state diagram of colloids with short-range attractions, including a gas–crystal coexistence line below which a metastable gas–liquid binodal is submerged.6,33,34 With respect to LLPS of protein solutions, it would be appealing to adopt a colloid physics perspective and hence reduce the complexity of the situation. In this context, the application of the law of corresponding states (LCS) could provide a rationalization. It implies that simple fluids obey the same reduced equation of state.35 For monatomic systems, the LCS can be proven given pairwise additive and conformal interactions.36 As a consequence, many physical quantities, like the vapor pressure of the liquid or the Boyle point, fall onto a master curve when temperature and density are expressed relative to their values at the critical point, as demonstrated for some molecular systems.37 However, only few real substances can be described based on the original form of the LCS, but many fluids follow an extended form, the extended law of corresponding states (ELCS), which involves an additional parameter to describe the equation of state.38 For several systems with short-range attractions, molecular simulations suggest that the second virial coefficient at the critical temperature is fairly constant16 and that it can be used as an additional parameter to describe the equation of state.39 The second virial coefficient B2 can be considered as a measure of the strength and range of the interactions. For a spherosymmetric potential U(r) with center-to-center distance r, its definition reads
![]() | (1) |
Inspired by the ELCS as proposed for colloidal model systems, it has been found that also the LLPS binodals of protein solutions collapse onto a master curve if plotted in the b2vs. volume fraction ϕ plane and the repulsions of the different systems are alike or their differences are accounted for in terms of an effective σ.34,47–50 Nevertheless, despite the importance of LLPS and the possibility to rationalize the underlying interactions by the ELCS, systematic and comprehensive studies on the applicability of the ELCS to protein solutions are scarce, especially with respect to the structural and dynamical properties of concentrated solutions.
Here, the osmotic equation of state and the collective dynamics of protein solutions under conditions close to LLPS are systematically investigated. Static and dynamic light scattering experiments (SLS and DLS) are used to determine the isothermal osmotic compressibility κT and the relaxation rate of concentration fluctuations Γ, respectively. Lysozyme in brine is used as a model system for proteins with short-range attractions,51 whose strength can be modulated by additives.52 We thus exploit that, for this system, state diagrams (LLPS binodals) and interactions parameters (b2) have been reported previously.50 The spinodal lines, as inferred from SLS, as well as κT and Γ are found to exhibit a corresponding-states behavior. For moderately concentrated solutions, the ϕ dependence of κT and Γ can be consistently described by Baxter's model of adhesive hard spheres. The off-critical, asymptotic T behavior of κT and Γ close to LLPS is consistent with the scaling laws predicted by mean-field theory. Thus, the present work aims at a comprehensive experimental test of the applicability of the ELCS to structural and dynamical properties of concentrated protein solutions.
Ultrapure water with a minimum resistivity of 18 MΩ cm was used to prepare buffer and salt stock solutions. They were filtered several times meticulously (nylon membrane, pore size 0.2 m) in order to remove dust particles. The concentration of salt stock solutions, prepared in buffer, was determined from the excess refractive index. The protein powder was dissolved in a 50 mM NaAc buffer solution. The solution pH was adjusted to pH 4.5 by adding small amounts of hydrochloric acid. At this pH value, lysozyme carries a positive net charge Q = +11.4 e,53 where e is the elementary charge. Solutions with an initial protein concentration c ≈ 40–70 mg mL−1 were passed several times through an Acrodisc syringe filter with low protein binding (pore size 0.1 m; Pall, prod. no. 4611) in order to remove impurities and undissolved proteins. Then, the protein solution was concentrated by a factor of 4–7 using a stirred ultrafiltration cell (Amicon, Millipore, prod. no. 5121) with an Omega 10 k membrane disc filter (Pall, prod. no. OM010025). The retentate was used as concentrated protein stock solution. Its protein concentration was determined by UV absorption spectroscopy or refractometry.54 Protein concentrations c are related to the protein volume fraction ϕ = cvp, where vp = 0.740 mL g−1 is the specific volume of lysozyme, as inferred from densitometry.54 Sample preparation was performed at room temperature (21 ± 2) °C. Samples were prepared by mixing appropriate amounts of buffer, protein and salt stock solutions. Samples with cloud-points close to or above room temperature were slightly preheated to avoid (partial) phase separation upon mixing. Samples were analyzed or processed further (e.g., centrifuged) immediately after preparation in order to perform the measurements before crystallization sets in. In addition, equilibrium clusters55 are not expected to form as the repulsive interactions are largely screened.
Samples with 10 mg mL−1 ≤ c ≤ 160 mg mL−1 (0.007 ≤ ϕ ≤ 0.12) were investigated at selected temperatures 12.0 °C ≤ T ≤ 43.0 °C. As the samples are metastable with respect to crystallization, samples are analyzed directly after preparation. Samples investigated at different T were prepared separately. In order to minimize spurious effects of any residual dust or undissolved aggregates that were not removed by filtration, the samples were filled into thoroughly cleaned cylindrical glass cuvettes (diameter 10 mm) and centrifuged (Hettich Rotofix 32A) 30 min at typically 2500 g prior to the measurements. They were then very carefully placed into the temperature-controlled vat of the instrument filled with decalin.
The time-averaged scattered intensity was recorded. In order to check sample quality, also DLS experiments (see below) were performed on the same samples. Samples with indications of aggregates or dust particles were discarded. The meticulous filtration and centrifugation procedure was essential to obtain reproducible light scattering data. However, this protocol did not work for samples with c > 160 mg mL−1, possibly due to aggregation or crystallization.51,57
The absolute scattering intensity, i.e. the excess Rayleigh ratio R, varies with protein concentration c and temperature T. It was determined using toluene as a reference according to
![]() | (2) |
The refractive index of a sample solution, n, was measured with a temperature-controlled Abbe refractometer (Model 60L/R, Bellingham & Stanley) operated with a HeNe laser (λ = 632.8 nm) and at the temperature of the SLS experiment. Refractive index increments, dn/dc, were obtained from linear fits to the dependence of n on c.
In one-component solutions, the excess scattering typically contains information on the shape and size of the particles as well as the particle arrangement. Their contributions are reflected in the form factor P(Q) and the structure factor S(Q), respectively, where Q = (4πn/λ)sin(θ/2) is the magnitude of the scattering vector with the scattering angle θ. Then, the excess Rayleigh ratio R reads:
R(Q) = K![]() ![]() ![]() ![]() | (3) |
![]() | (4) |
The effective protein diameter σ = 3.4 nm is small compared to λ, implying σQ ≪ 1. Nevertheless, for selected samples, experiments were performed at 30° ≤ θ ≤ 150° and R was indeed found to be independent of θ and hence of Q. In most of our experiments thus only one angle θ = 90° and hence Q ≈ 0.018 nm−1 was investigated. In addition to the DLS experiments, the independence of R on θ also suggested that large particles, such as impurities or aggregates, were absent.
Moreover, in the low-Q limit,
![]() | (5) |
![]() | (6) |
g2(Δt) = B + β|f(Δt)|2 | (7) |
![]() | (8) |
![]() | ||
Fig. 1 Liquid–liquid phase separation of protein solutions: (A) binodals of lysozyme solutions (pH 4.5, 0.9 M NaCl) in the temperature T vs. volume fraction ϕ (or protein concentration c) plane. Symbols denote different amounts of GuHCl, as specified in the legend. Lines are a guide to the eye. Inset: Critical temperature Tc for three different additive concentrations, which can be estimated based on a critical scaling ansatz, as detailed previously.7,8,25,34 (B) The same binodals as in (A), but in the normalized second virial coefficient b2vs. volume fraction ϕ plane. Data of (A) and (B) are replotted from ref. 50. |
In order to explore the applicability of the ELCS to static and dynamic properties of concentrated protein solutions, in the present work, this model system is investigated comprehensively by SLS and DLS experiments. The system is studied as a function of protein concentration, temperature and additive content in the vicinity of the LLPS binodal. However, an extremely close proximity as well as highly concentrated samples are avoided; therefore, it can be expected that simple, analytical models are applicable to the ϕ dependence of structural and dynamical data and mean-field values of the critical exponents describe the asymptotic T dependence of the data.
First, spinodal temperatures are inferred from the T dependence of the SLS data and added to the b2vs. ϕ plane (Fig. 1(A)). Then, the ϕ dependence of the isothermal compressibility κT is analyzed and a corresponding-states behavior is observed. Third, the relaxation rate Γ of the concentration fluctuations is determined and its T dependence is analyzed. Fourth, the ϕ dependence of the collective diffusion coefficient Dc is analyzed and a corresponding-states behavior is found.
κT = κ0ε−γ | (9) |
![]() | (10) |
Fig. 2(A) shows exemplary data (symbols) of κT−1(T) for various protein concentrations c as indicated. As expected, the data can be described by linear fits (lines), yielding estimates of Ts as intercepts of the abscissa. Note that the spinodal line is submerged below the binodal and hence the κT−1(T) data have to be extrapolated by the fits. Moreover, at fixed T, κT−1(T) decreases with c, and hence larger intercepts of the abscissa and, correspondingly, Ts are observed, similar to the c dependence of the cloud-point temperatures shown in Fig. 1(A).
Fig. 2(B) shows all κT−1 data as a function of the reduced spinodal temperature ε together with a solid line of slope 1 on a double logarithmic representation. As expected from eqn (9), the different data sets collapse onto a single curve which shows a power-law behavior with the mean-field value γ = 1. The observed scaling behavior further supports the appropriateness of our approach to estimate Ts values. Eqn (9) was also fitted globally to the κT−1 data with a free, but global value of γ (not shown). This procedure yields γ = 0.97, further supporting the appropriateness of the mean-field value.
In addition to the data obtained in Fig. 2(A), similar experiments have been performed for the two other solution conditions (red circles and blue triangles in Fig. 1(A)). Fig. 3(A) shows the resulting Ts data (open symbols) and replots the binodals (full symbols). Indeed, for each of the three conditions, the spinodal is hidden below the binodal and thus also narrower than the binodal. With increasing ϕ, the spinodal approaches the binodal, as they are expected to coincide at the critical point. With increasing guanidine content, the spinodal shifts to lower T, similar to the decrease of the binodal. The inset shows the guanidine dependence of Ts for different protein concentrations together with a common linear fit. The slope of −26 K/M agrees with the previously observed value for the binodals.54 (For one particular solution condition, c = 40 mg mL−1 and 0.4 M GuHCl, the κT(T) are extrapolated over more than 30 K, resulting in a very low value of Ts with large uncertainty. This data point is omitted from further analysis.)
![]() | ||
Fig. 3 (A) Binodals (data as full symbols and guides to the eye as solid lines, replotted from Fig. 1, and spinodals (data (open symbols) and eye guides (dotted lines) of lysozyme solutions (0.9 M NaCl) with various amounts of GuHCl as in Fig. 1. Inset: Dependence of the spinodal temperatures on the guanidine concentration for different protein concentrations (increasing from bottom to top): data (open symbols) and global linear fits (lines). (B) The same binodals as in (A), but in the normalized second virial coefficient b2vs. volume fraction ϕ plane. Lines are guides to the eye. |
Since b2(T) data are available for our system (cf. circles in Fig. 5(B)),50 the spinodal lines of Fig. 3(A) can also be represented in the b2vs. ϕ plane. Fig. 3(B) shows both the binodal and spinodal lines in this representation. Again, similar to the binodals, the different spinodal lines collapse onto one another, indicating that the corresponding-states behavior previously observed for the binodals34,50 also holds for the spinodals.
Some solution conditions (marked by stars in Fig. 4(A)) have been examined by optical microscopy in order to study the phase separation kinetics in the metastable region between spinodal and binodal line as well as in the unstable region below the spinodal.
![]() | ||
Fig. 4 (A) Interpolated binodal (solid line), replotted from Fig. 1; spinodal temperatures (open squares) as inferred from the intercept of the abscissa in Fig. 2(A) connected by a dotted line as a guide to the eye; the solution conditions probed by optical microscopy are indicated by stars. (B) Optical micrographs (image size 248 × 248 μm2) illustrating the phase transition kinetics of samples phase separating via nucleation (top) and spinodal decomposition (bottom); in this case, ϕ = 0.06 (top) and ϕ = 0.09 (bottom) both at T = 18 °C, respectively. |
Exemplary time series of micrographs are shown in Fig. 4(B). In the metastable region, the microscopy data show the successive formation and growth of droplets, whereas in the unstable region the micrographs reveal a rapid initial darkening as the sample becomes turbid and at later stages the micrographs indicate domain formation and coarsening. The micrographs thus show qualitatively different phase separation kinetics depending on the location of the state. In the metastable region between binodal and spinodal, phase separation proceeds via droplet nucleation, while in the unstable region below the spinodal, spinodal decomposition is observed.
![]() | ||
Fig. 5 (A) Volume fraction ϕ dependence of the isothermal compressibility κT normalized by the isothermal compressibility κ(id)T of an ideal solution: Experimental data (symbols) for different temperatures normalized by the critical temperature, T/Tc, and guanidine concentrations are marked by symbol color and type, respectively. Lines are one-parameter fits to eqn (11). (B) Normalized second virial coefficient b2 as a function of the reduced temperature T/Tc: data retrieved from the fits in (A) and literature data23,25,50,82 are shown as full and open symbols, respectively. |
The ϕ dependence of the isothermal compressibility for low salt content,75i.e., far away from LLPS, has been successfully described by an analytical expression for the compressibility of the adhesive hard-sphere model in the Percus–Yevick approximation:76–81
![]() | (11) |
![]() | (12) |
![]() | (13) |
Fig. 5(B) shows the resulting values (full symbols), as retrieved from model fits to all κT data of the present work. The magnitude of b2 is found to increase with T/Tc, i.e., attractions are weakened for temperatures further away from the binodal. The value observed for T/Tc ≈ 1 is close to b2 = −1.5, as proposed by Vliegenthart and Lekkerkerker.16 Moreover, the figure contains literature data23,25,50,82 on b2(T/Tc) for the present system obtained by light scattering from dilute solutions and from the analysis of small-angle X-ray scattering experiments. The present data and the independent literature data agree with each other, supporting the significance of the obtained model parameter (τ) as well as the appropriateness of the model. Further support stems from independent SAXS measurements, whose Q dependent structure factor contribution was accurately described by approximate, analytical expression of the Baxter model.82
Fig. 6 shows exemplary intensity cross-correlation functions g2(Δt) as symbols (A) at fixed T = 26 °C for different c and (B) at fixed c = 80 mg mL−1 for different T (as indicated). All ISFs exhibit a single exponential decay as characteristic for ergodic liquids. The ISFs can hence be accurately described by the second-order cumulant ansatz in eqn (8). The corresponding fits (lines) agree with the data. Moreover, the single exponential decay further indicates the absence of aggregates or large impurities and hence also validates the analysis of the SLS experiments.
It is important to note that most previous DLS studies on the relaxation rate of concentration fluctuations of proteins close to LLPS72,73,84,85 did not take into account possible effects of multiple scattering; some works86,87 reported non-exponential correlation functions. In the present work, contributions from multiple scattering are suppressed by the 3D cross-correlation set-up and typical single exponential decays of the ISF are observed. In arrested systems, stretched, non-exponential decays are expected,88,89 whereas the (off-)critical slowing down rather results in exponential decays.73,74
Fig. 7(A) shows the T dependence of Γ for different c, as retrieved from fits of eqn (7) and (8) to ISFs. As expected from the inspection of the ISFs in Fig. 6, Γ decreases with increasing c or decreasing T, i.e. the concentration fluctuations exhibit a slowing down upon approaching Ts.
![]() | ||
Fig. 7 (A) T dependence of the relaxation rate Γ of lysozyme solutions (0.9 M NaCl, no GuHCl added) with protein concentrations as indicated: experimental data (symbols) and fits to eqn (15) (lines). Inset: Dependence of the correlation length ξc on the correlation length ξ0. Both parameters (symbols) are retrieved from the fits. The line has a slope of 2. (B) Dependence of the relaxation rate Γ on the inverse isothermal compressibility κT, normalized by κ(0)T = 1 (N m−2) (mg mL−1)−1, for various guanidine and protein concentrations indicated as columns and rows in the caption. (C) Dependence of the relaxation rate Γ on the reduced spinodal temperature ε = T/Ts − 1 with the spinodal temperature Ts. Symbols as in (B). The solid line has a slope of 1. |
In order to quantitatively understand the observed slowing down, Γ is plotted as a function of κT−1 in Fig. 7(B). A remarkably linear correlation is observed. As in almost all of our experiments, ε is not very small (ε > 0.01) and hence a mean-field picture might be applicable. Then, the relaxation rate Γ and the compressibility κT can be related via:56,90
![]() | (14) |
In a more general mode-coupling framework,91–93 the relaxation rate can be described as the sum of a critical and a background contribution in the vicinity of the critical point (or the spinodal). If the correlation length is small compared to Q−1, as for most of our data, the relaxation rate has a simple form:94
![]() | (15) |
ξ = ξ0ε−ν | (16) |
![]() | (17) |
Fig. 8 shows exemplarily data on the ϕ dependence of Dc/D0. As in Fig. 5(A), data for different additive compositions, but the same temperature relative to the critical temperature, T/Tc, are displayed as symbols in the same color. Again, as for κT/κ(id)T, though obtained under different solution conditions, the Dc/D0 data collapse onto a single curve for fixed T/Tc. This indicates a corresponding-states behavior of the collective diffusion coefficient close to Tc. Hence, our data provide further experimental support of the ELCS, also for dynamical properties.
![]() | ||
Fig. 8 Volume fraction ϕ dependence of the collective diffusion coefficient of concentration fluctuations Dc normalized by the diffusion coefficient at infinite dilution D0. Experimental data for various reduced temperatures T/Tc and guanidine concentrations are marked by symbol color and type, respectively. Lines are computed based on eqn (18) using τ as retrieved from the fits shown in Fig. 5. |
As for compressibility data, the simple Baxter model can be applied to analyze the collective diffusion. An approximate equation for Dc(ϕ)/D0, to first order in ϕ,98 is available:
![]() | (18) |
![]() | (19) |
The data presented in Fig. 5(A) and 8 indicate that, close to LLPS, the osmotic compressibility and the collective diffusion coefficient is only determined by the volume fraction and the temperature relative to the critical temperature or, equivalently, by the second virial coefficient, as shown in Fig. 5(B). Thus, for our system, the ELCS applies and rationalizes static and dynamical properties of concentrated solutions. It is conceivable that this argument, due to its coarse-grained nature, can be generalized to other globular protein systems. Note that, for the solution conditions investigated here, the repulsive interactions are largely screened and very similar for the various conditions. However, if one considers systems in which repulsion is more important, it is conceivable that this has to be accounted for in an effective particle size.34,39 Nevertheless, the present data as well as previous studies34,47,49,50,82 suggest that the ELCS applies to a broad range of proteins and solution conditions.
The adhesive hard-sphere model proposed by Baxter99 represents one of the simplest systems with short-range attractions. It is therefore astonishing that an approximate theoretical description for this model is suitable to quantitatively describe the complex interactions between protein molecules even in the vicinity of LLPS. While the ELCS suggests that the detailed shape of the interactions does not matter, this does not guarantee that the approximate theoretical description can also be reasonably applied to describe the experimental conditions. It is conjectured that the success of the description might also be related to the globular shape of lysozyme, the moderate concentrations considered as well as the not-to-close proximity to the spinodal. It is conceivable that models accounting for the directionality of the interactions100–102 might be more appropriate to describe the structural and dynamical properties of concentrated solutions of proteins of more complex shape. However, as implied by the ELCS, for the conditions analyzed here, similar results are expected.
This journal is © the Owner Societies 2023 |