Resolving the Mechanisms of Soy Glycinin Self-Coacervation and Hollow-Condensate Formation

Self-coacervation of animal-derived proteins has been extensively investigated while that of plant proteins remains largely unexplored. Here, we study the process of soy glycinin self-coacervation and transformation into hollow condensates. The protein hexameric structure composed of hydrophilic and hydrophobic polypeptides is crucial for coacervation. The process is driven by charge screening of the intrinsically disordered region of acidic polypeptides, allowing for weak hydrophobic interactions between exposed hydrophobic polypeptides. We find that the coacervate surface exhibits order, which stabilizes the coacervate shape during hollow-condensate formation. The latter process occurs via nucleation and growth of protein-poor phase in the coacervate interior, during which another ordered layer at the inner surface is formed. Aging enhances the stability of both coacervates and hollow condensates. Understanding plant protein coacervation holds promises for fabricating novel functional materials. P coacervation or liquid−liquid phase separation in protein solutions is characterized by the formation of liquid-like protein-rich microdroplets (coacervates or condensates). Upon coalescence, they yield a macroscopically monophasic biomacromolecular fluid of highly concentrated protein phase. Coacervation can occur spontaneously upon changing the environmental conditions, a process known as simple coacervation or self-coacervation, or via interactions with another oppositely charged protein specie, also known as complex coacervation. Protein coacervation represents a crucial route for the assembly of cellular organelles and, at the same time, offers a facile approach toward the fabrication of biomaterials such as fibers, bioadhesives, and bioactive compound carriers. In recent years, much interest has been dedicated to human intracellular condensates originating from intrinsically disordered proteins as well as coacervates that derive from animal extracellular matrix proteins. Far less studies have focused on plant protein coacervation. Compared to animal proteins, plant, and in particular soy proteins are widely abundant and economical, which makes them ideal materials for the scale-up processing in industry. Soy glycinin is one of the major storage proteins in the soybean seed. It is commonly used as an emulsion stabilizer for improving the texture of food products via modulating the structure of protein gels and for producing bioadhesives. Recently, soy glycinin was found to undergo self-coacervation in aqueous solutions upon the addition of salt, a process exemplified by the formation of spherical condensates in the micrometer range. Interestingly, upon a temperature increase, the coacervates transform into stable hollow microcapsule-like condensates (also referred to as vacuolated structures or a vesicle-like condensate phase). This process represents a simple and very energy-efficient route for microencapsulation of active compounds for controlled release, protection against environmental factors, or masking of an unpleasant odor or taste. By forming a protein layer on oil droplets, protein self-coacervation can be employed for the encapsulation of hydrophobic compounds, which represent a much simpler approach compared to traditional methods based on complex coacervation with another polymer, usually polysaccharide. Hollow protein microcapsules formed via coacervation can encapsulate and control the release of hydrophilic compounds by modulating the permeability of the protein shell. This is advantageous compared to more popular methods for microcapsule production such as the template-based layer-by-layer method, which involves a tedious and time-consuming fabrication process and is associated with waste of materials. Despite the above-mentioned advantages and potential applications, the mechanism of soy glycinin self-coacervation and the cavitation of the condensates is not yet understood. Remarkably, the formation of hollow condensates upon selfcoacervation has not yet been reported for animal proteins, while it seems a common phenomenon for the plant seed 11S Received: October 1, 2020 Accepted: November 30, 2020 Letter pubs.acs.org/macroletters © XXXX American Chemical Society 1844 https://dx.doi.org/10.1021/acsmacrolett.0c00709 ACS Macro Lett. 2020, 9, 1844−1852 D ow nl oa de d vi a M PI K O L L O ID G R E N Z FL A E C H E N FO R SC H U N G o n D ec em be r 8, 2 02 0 at 1 2: 59 :1 8 (U T C ). Se e ht tp s: //p ub s. ac s. or g/ sh ar in gg ui de lin es f or o pt io ns o n ho w to le gi tim at el y sh ar e pu bl is he d ar tic le s.

P rotein coacervation or liquid−liquid phase separation in protein solutions is characterized by the formation of liquid-like protein-rich microdroplets (coacervates or condensates). Upon coalescence, they yield a macroscopically monophasic biomacromolecular fluid of highly concentrated protein phase. 1 Coacervation can occur spontaneously upon changing the environmental conditions, a process known as simple coacervation or self-coacervation, 2 or via interactions with another oppositely charged protein specie, also known as complex coacervation. 3 Protein coacervation represents a crucial route for the assembly of cellular organelles 4,5 and, at the same time, offers a facile approach toward the fabrication of biomaterials such as fibers, 6 bioadhesives, 7 and bioactive compound carriers. 8 In recent years, much interest has been dedicated to human intracellular condensates originating from intrinsically disordered proteins as well as coacervates that derive from animal extracellular matrix proteins. 4−7 Far less studies have focused on plant protein coacervation.
Compared to animal proteins, plant, and in particular soy proteins are widely abundant and economical, which makes them ideal materials for the scale-up processing in industry. Soy glycinin is one of the major storage proteins in the soybean seed. It is commonly used as an emulsion stabilizer for improving the texture of food products via modulating the structure of protein gels and for producing bioadhesives. 9−11 Recently, soy glycinin was found to undergo self-coacervation in aqueous solutions upon the addition of salt, a process exemplified by the formation of spherical condensates in the micrometer range. 12 Interestingly, upon a temperature increase, the coacervates transform into stable hollow microcapsule-like condensates (also referred to as vacuolated structures or a vesicle-like condensate phase). 12,13 This process represents a simple and very energy-efficient route for microencapsulation of active compounds for controlled release, protection against environmental factors, or masking of an unpleasant odor or taste. 14−16 By forming a protein layer on oil droplets, protein self-coacervation can be employed for the encapsulation of hydrophobic compounds, 17 which represent a much simpler approach compared to traditional methods based on complex coacervation with another polymer, usually polysaccharide. 14 Hollow protein microcapsules formed via coacervation can encapsulate and control the release of hydrophilic compounds by modulating the permeability of the protein shell. 13,18 This is advantageous compared to more popular methods for microcapsule production such as the template-based layer-by-layer method, 19 which involves a tedious and time-consuming fabrication process and is associated with waste of materials.
Despite the above-mentioned advantages and potential applications, the mechanism of soy glycinin self-coacervation and the cavitation of the condensates is not yet understood. Remarkably, the formation of hollow condensates upon selfcoacervation has not yet been reported for animal proteins, while it seems a common phenomenon for the plant seed 11S globulins, a group of storage proteins sharing a similar structure. 20 Followed by soy glycinin, pea protein and fava bean legumin have recently been reported to form hollow structures through coacervation. 21,22 However, all these studies have not offered a mechanistic view of protein-coacervate-tohollow-condensate transition. As representative of the 11S globulin family, soy glycinin is a hexamer, and each of its six subunits contains one acidic and one basic polypeptide crosslinked by disulfide bonds 23,24 (see also Figure S1 in the Supporting Information, SI). This structural characteristic represents a substantial difference from coacervation-prone animal proteins featuring mostly one type of polypeptide. 25,26 Here, using soy glycinin as a prototype, we aim at resolving (i) how these molecular characteristics contribute to 11S globulin coacervation and the "unique" transformation into hollow condensates, (ii) what are the driving forces involved in selfassembly during coacervation and cavitation, and (iii) why the coacervates preserve their shape during the phase transition.  (11 g/L, pH 7.8) was adjusted to different ionic strengths at 23°C by adding NaCl solution with different concentrations, attaining a final protein concentration of 10 g/L. Turbidity of soy glycinin suspensions at different salt concentration are given; the framed cartoons (also in panels b−d) illustrate the emergence of coacervates in the respective phase region. (b) Phase state of soy glycinin as a function of protein concentration at room temperature (23°C). The pink region delineated by the red circles represents the coacervation region (R2) where phase separation occurs; the blue area outside the blue solid circles is characterized by homogeneous solution. All solutions were prepared via dilution of a stock solution of 100 g/L at pH 7. (c) Influence of pH on the phase boundary of coacervation. Protein solution of 11 g/L with different initial pH was adjusted to different NaCl concentration at 23°C. (d) Influence of temperature on the phase boundary of coacervation. Protein solution of 11 g/L with initial pH of 8 was adjusted to a different NaCl concentration and kept at a different temperature. The phase diagrams in (b)−(d) were constructed based on the visual appearance of phase separation in the sample as well as detection of microstructures observed under the microscope. (e) Exponential decay of the aspect ratio of coalescing coacervates induced at 0.1 M NaCl of different final diameters, as given in the legend. The fits (solid curves) were used to assess the relaxation time, τ, see SI. The bright field images show the coalescence of two coacervate droplets. The scale bar represents 2 μm. The image sequence is from Movie 1 in the SI. (f) Plot of the relaxation time τ vs the final coacervate diameter. The slope yields the inverse capillary velocity η/γ ≈ 9.69 ± 0.53 s/μm by linear regression (R 2 = 0.97, n = 13).
To shed light on the mechanism of soy glycinin selfcoacervation, we first built the phase diagram and explore the coacervate physicochemical characteristics (Figure 1). Homogeneous soy glycinin solutions undergo sequential demixing and mixing phase transitions in response to increasing NaCl concentration (revealed by turbidity measurements), defining three regions in the phase space limited by lower and upper NaCl concentrations, C* and C**. At [NaCl] < C* and [NaCl] > C**, the protein solution is homogeneous (regions R1 and R3 in Figure 1a−d). At intermediate concentrations, C* < [NaCl] < C**, coacervation takes place, leading to phase separation (region R2) with maximum turbidity at around [NaCl] ≈ 0.1 M ≡ C Δ (Figure 1a). The demixing is characterized by random nucleation of spherical dense coacervates, which grow into larger ones via coalescence (see SI, Movie 1). Coalescence is slow and described by an exponential decay with relaxation time τ (Figure 1e), which depends linearly on the size of the coalescing droplets ( Figure  1f), see SI, Experimental Section. The proportionality coefficient represents the inverse capillary velocity η/γ, 27 where γ is the interfacial tension driving the coalescence process and η is the droplet viscosity that slows it down. We find η/γ ≈ 9.69 s/μm, which corroborates the highly viscous nature of soy glycinin coacervates (see SI, section on coalescence dynamics), also supported by negligible fluorescence recovery after photobleaching ( Figure S2).
To understand the observed coacervation process, we consider the hexameric structure of soy glycinin. Every three subunits (each consisting of acidic and basic polypeptides bonded by disulfide bonds) assemble into a trimer, and two trimers stack face to face, forming the hexamer. 23,24 The carboxylic ends of the acidic polypeptides are particularly divergent, also termed the hypervariable region (HVR). 28 One striking feature of the HVR is the high content of negatively charged amino acids with a repeated aspartate/glutamate-rich sequence ( Figure S3). 28 The HVR therefore shares the characteristic of the intrinsically disordered regions (IDRs), that is, repetitive amino acid sequences with low complexity, which has been proven to be a crucial structural feature for coacervation to occur. 29,30 Furthermore, the basic polypeptides exhibit a large fraction of hydrophobic amino acids, which are mostly buried inside the protein molecules, while the acidic polypeptides are hydrophilic and mostly exposed. 31,32 The isoelectric point of soy glycinin is around 5.1, implying that it is negatively charged above pH 5.1. 33 Remarkably, with increasing ionic strength, the basic polypeptides become less exposed contrary to the acidic polypeptides. 32 Considering the above molecular characteristics, in Figure 2, we sketch a plausible structural interpretation of the phase behavior observed in Figure 1. Due to the strong electrostatic repulsion, at pH > 7 and [NaCl] < C*, the soy glycinin solution remains homogeneous (region R1 in the phase diagram; Figure 2a). Increasing salt concentrations screens the electrostatic repulsion, allowing weak hydrophobic interactions between the exposed hydrophobic basic polypeptides and driving coacervation in region R2 (Figure 2a). This assumption is corroborated by coacervate size decrease observed in the presence of small amounts of urea ( Figure S5), which is known to suppress protein−protein hydrophobic interactions. 34,35 However, the coacervation was not disrupted by 1,6hexanediol, an aliphatic alcohol ( Figure S5), which hinders the formation of protein condensates by perturbing weak protein−protein interactions. 36 This suggests that hydrophobic interactions of intermediate strength stabilize glycinin coacervates. Tentative interpretation of this stabilization at molecular lever could be sought in the high abundance of glutamine (the most frequent amino acid) and serine in soy glycinin ( Figure S6a), both of which were shown to induce hardening and decreased fluidity of protein coacervates. 37 Furthermore, the polyglutamine amino acid sequence ( Figure  S6b) in the variable region might promote β-sheet structure formation, 38 further contributing to the stabilization of protein droplets. 39 Above C**, the basic polypeptides are mostly shielded in the protein interior, exposing more acidic polypeptides, thus inhibiting coacervation due to the decreased hydrophobic interaction and yielding a homogeneous phase in region R3 (Figure 2a).
We also explored the effect of protein restructuring on the ability to undergo coacervation. Incubation of glycinin solutions at low salt concentration (homogeneous region R1) for a long time (4−12 days) with subsequent addition of salt led to the appearance of coacervate-like flocks (Figure 2b, Figure S7a). This was due to the partial dissociation of the hexameric protein structure into trimers at low ionic strength ([NaCl] < 0.01 M) with time. 40,41 The trimers expose their stacking faces making them more hydrophobic. 23,24 After 12 days, the flocks could not be dissociated when [NaCl] was brought out of the coacervation region R2 ( Figure S7b), indicating stronger hydrophobic interaction compared to those in freshly prepared coacervates. We further increased the exposure of hydrophobic basic residues by incubating glycinin with a reducing agent to break the disulfide bonds and release the hydrophilic acidic polypeptides. 42 Then, upon shifting [NaCl] to the coacervation region R2, we observe the transformation of the spherical coacervates to coacervate-like flocks and later on precipitates (Figure 2c and Figure S7c). These precipitates also persist at salt concentrations out of the coacervation region ( Figure S7d).
Considering all of the above experiments, we conclude that glycinin coacervation is driven by charge screening of the acidic polypeptides, which promotes weak hydrophobic interactions between the exposed basic polypeptides as typically observed for other proteins. 26,43 The enhanced stability of glycinin droplets against 1,6-hexanediol evidences the presence of even stronger hydrophobic interactions. Importantly, the hexameric structure of glycinin combined with the effect of disulfide bonding shields the basic polypeptides ensuring hydrophobic interaction that is weak enough to induce coacervation rather than flocking or precipitation. Contribution of charge screening to coacervation is further evidenced by the influence of pH on the phase boundary ( Figure 1c). Raising the pH above 7 shrinks the phase-separation region R2. At pH > 8.7, coacervation is completely suppressed because of the increased negative surface charge of glycinin, 33 which weakens the screening effect of salt. Below the isoelectric point (∼pH 5.1), glycinin becomes positively charged and the electrostatic repulsion increases with decreasing pH. 33 The coacervation is therefore suppressed when pH is below 3.5 ( Figure S8). Similarly, increasing temperature from 4 to 50°C, as shown in Figure 1d, also shrinks the region R2. At T > 55°C, coacervation is completely suppressed because of the enhanced hydrophobic interaction between soy glycinin molecules that favors the thermally induced irreversible aggregation. 44 Considering that ionic strength, pH, and temperature significantly influence the coacervation (Figure 1), we hypothesized that soy glycinin coacervates should show responsive structural changes upon modifying these environ- Influence of pH and temperature on the coacervate structure. pH of the coacervate suspension was increased by adding equal volume of NaOH solution (0.5−1.5 mM); the final pH of the suspension is indicated in the images in the upper row. The coacervate suspension (1 mL) was heated via gentle shaking in a water bath under different temperature (indicated in the images, lower row) for 5 min, and the microstructures were observed immediately after heating. The sketch on the left represents how increasing pH or temperature shrinks the phase boundary (green to blue to red ellipses) and the corresponding coacervate structure response, depending on how far it is located to (or out of) the new phase boundary. The scale bars represent 40 μm. mental conditions. Previous research only reported heatinduced transition from coacervates to hollow condensates. Here, we find that any factor that brings the coacervation system toward the binodals can induce such transition. We consider coacervates formed at [NaCl] = C Δ ≡ 0.1 M. Upon shifting [NaCl] toward the binodals (C*/C**), the coacervates transform into hollow condensates with gradually decreasing fraction of the dense phase ( Figure 3a) in accordance with the turbidity measurement ( Figure 1a). Beyond the phase boundary, the coacervates disperse into a homogeneous phase (Figure 3a). Increasing pH or temperature results in shrinking the coacervation region (R2 in Figure  1c,d), thus, bringing the phase boundary closer to the location of our system in the phase diagram ( Figure 3b). The condensates again adjust their phase state by forming a protein-poor phase inside (and transforming the droplets into hollow condensates) or dissolve when the system is beyond the new binodal ( Figure 3b). Decreasing protein concentration by dilution could also lead to cavity formation, which is more pronounced when the system is near the coacervation boundary ( Figure S9).
We further investigated the process of cavitation as a first step to understand the mechanism of hollow-condensate formation. We prepared protein coacervates at [NaCl] = C Δ ≡ 0.1 M, and subsequently increased the salt concentration to 0.175 M (i.e., still in region R2 of the phase diagram, Figure  1b). Many small cavities first formed inside the condensates and then merged into a larger one typically within minutes (Figure 4a and Movie 2). Interestingly, cavity formation induced by increasing pH (Figure 4b and Movie 3) or temperature (Figure 4c and Movie 4) follows the same process as that induced by salt. A similar coacervate-to-hollow structure transition has been described for pea protein isolate 21 as well as for RNA−protein (complex) coacervates in vivo and in vitro, 45−47 although the nature and behavior of the latter as complex coacervates are quite different from those of soy glycinin (simple) coacervates. It should be noted that, for conditions of [NaCl] < C Δ , small irregularly shaped coacervates were often observed ( Figure S10). A few seconds of heating would transform them into microdomains with smooth boundaries, followed by the nucleation and growth of protein-poor phase in their interior ( Figure S10). This intermediate morphological transition clarifies why irregular coacervate clusters can also form hollow structures upon heating as observed earlier. 48 The mechanism of hollow-condensate formation is partially related to the interplay of charge screening and hydrophobic interactions. Upon shifting [NaCl] toward C**, the hydrophobic interaction stabilizing the coacervates decreases as more acidic polypeptides are exposed. Both shifting [NaCl] toward C* and increasing pH enhance the electrostatic repulsion, which weaken the salt screening effects. Increasing temperature leads to a stronger hydrophobic interaction that is unfavorable for coacervation. These changes in the molecular interactions lead to coacervate dissociation, but they do not explain why the protein-poor phase forms in the interior instead of simply shrinking the condensate in size. We speculated that enhanced surface interaction or organization stabilizes the observed shapes. We therefore probed the condensates for structural order (as observed for RNA− protein complexes 45 ) using polarization microscopy ( Figure  5a). Both coacervates and hollow condensates display strong birefringence at their surface, suggesting surface organization of the protein-rich phase. This liquid-crystalline-like order persists during the transformation into hollow condensates. Interestingly, the inner surface of the hollow condensates also shows birefringence, indicating order. We propose that, at these interfaces, the protein orients so that the more hydrophilic acidic polypeptides extend toward the protein-depleted phase, while the more hydrophobic exposed basic polypeptides extend toward the protein-dense phase, thus, leading to ordered packing ( Figure 6). This molecular reorientation presumably strengthens the hydrophobic interactions at these organized surfaces, which is supported by the observation that, in the presence of urea, coacervate or hollow-condensate dissolution initiates from the dense phase to the surface ( Figure S11). The surface order and increased protein interactions could contribute to the long stability of the hollow condensates ( Figure S12), as discussed below.
Considering that part of the glycinin molecules in the coacervate interior become mobile and are released out of the coacervates upon cavitation, the dense phase and ordered surface should have a mesh size that allows the diffusion of free glycinin. We probed this mesh size using FITC-labeled dextran. Dextran, with a M w > 4 kDa, was practically excluded from the protein-rich phase of the coacervates or the hollow condensates ( Figures S13 and S14). Upon cavity formation, low-molecular-weight dextran (≤40 kDa) could diffuse through the capsule-like dense phase, while 500 kDa dextran could not permeate (Figures 5b and S14). Presumably, the ordered surface acts as a size-dependent filter with a mesh size of at least 3−5 nm, as defined by the radius of gyration of 20− 40 kDa dextran, 49 that would therefore allow free glycinin, which has a radius around 4 nm 50,51 to leak out rather than concentrating in the layer of protein-dense phase.
Soy proteins tend to aggregate and gel even at low temperature, promoted by increasing protein concentration. 52,53 Considering that the condensates represent a highly dense protein phase, it is plausible that protein−protein interactions are enhanced with time. To resolve the influence of aging and probe the stability of the resulting structures, we aged protein coacervates at 23°C for different periods and subsequently examined their stability by shifting the salt concentration out of the coacervation region R2 and into the homogeneity regions (R1/R3). We were, thus, able to resolve aging conditions that lead to preserving the structures. After a short aging time, the coacervates could still be dissociated irrespective of the [NaCl] shift direction ( Figure S15a). In samples aged for 1 h, the coacervate surface retained its topology (Figure 5c, upper panel) in region R3, suggesting stronger protein−protein interactions resulting from the surface order evidenced by the polarization microscopy images (Figure 5a). At longer times (>5 h), the whole coacervate appeared preserved (Figure 5c, upper panel).
Aging also strengthened the stability of hollow condensates. Short aging times (∼10 min) led to thinning of the proteindense phase or partial dissolution ( Figure S15b). After aging for 1 h, the dense phase was released, leaving the outer and inner surfaces of the hollow condensates as concentric shells in region R1 (Figure 5c, lower panel, and Figure S15b). This confirms that protein molecules at the inner surface of the  Figure S15. Aging was done at 23°C under mild stirring and for different periods (indicated above the red arrows), after which the salt concentration of the suspension was shifted to 0.03 M, which falls into the homogeneity region R1, that is, [NaCl] < C*, or 0.3 M, which falls into homogeneity region R3, that is, [NaCl] > C**. The scale bars represent 10 μm.
hollow condensates assemble similar to those at the outer surface, in accordance with the polarization microscopy observations. Overall, aging led to the formation of stabilized shells offering a facile alternative pathway for microcapsule production. The enhanced stability with aging is presumably due to the large abundance of glutamine and serine ( Figure  S6a), which promotes the hardening of coacervates in a timedependent manner. 37 The polyglutamine sequence might also contribute to the stabilization of protein droplets by forming higher-order structure ( Figure S6b). 38 Overall, the hydrophobic interactions are strengthened with aging ( Figure S16).
The proposed molecular mechanism behind coacervate formation, transformation into hollow-condensates, and shape stability is summarized in Figure 6. Remarkably, aging enhances the protein−protein interaction, transforming the coacervates or hollow condensates into microcapsules with a morphology stable to solvent condition changes. This research provides essential knowledge and approaches for investigating the coacervation of other plant seed 11S globulins. Furthermore, it paves the way toward designing novel microstructures such as semipermeable responsive microcapsules as well as further understanding of the accumulation and dissociation of the protein condensates in plant seed cells. 54,55 ■ ASSOCIATED CONTENT
Details for the materials and experimental methods, supporting figures, and movie captions (PDF) Movie 1: Coalescence dynamics of protein coacervates (MP4) Movie 2: Transition from coacervates to hollow condensates induced by increasing ionic strength (AVI) Movie 3: Transition from coacervates to hollow condensates induced by increasing pH (AVI) Movie 4: Transition from coacervates to hollow condensates induced by increasing temperature (AVI) Figure 6. Schematic illustration of the mechanisms of the self-coacervation of soy glycinin, the transformation from coacervates to hollow condensates and their surface organization. The hexameric structure of glycinin contributes to these phase changes via the interplay of the basic (B, hydrophobic) and acidic (A, hydrophilic) polypeptides, which modulate the protein−protein interactions upon environmental condition changes, as shown in Figure 2. The structure of the hexamer was simplified with one hydrophilic part (blue), representing the exposed acidic polypeptides and one hydrophobic part (pink), representing the exposed basic polypeptides. Homogeneous solutions of the protein (left) undergo coacervation in a specific range of salt concentration (C* < [NaCl] < C**), pH, and temperature, which is driven by charge screening of acidic polypeptides and the hydrophobic interaction between exposed basic polypeptides. The coacervates are fluid and can grow via coalescence. Upon changes in the external conditions bringing the system closer to the binodal (C* or C**) or upon increasing temperature or pH, the coacervates develop cavities of the protein-poor phase, which transforms the droplets into hollow condensates. This process is irreversible and is hindered by aging. The amphiphilic glycinin molecules orient at condensate interface, rendering surface order that strengthens intermolecular interactions stabilizing the coacervate shape. In fresh solutions, the dense phase dissolves once the external conditions shift the system out of the region of phase coexistence, while in aged ones, the structures are preserved.

Experimental section
Isolation of soy glycinin. Preparation of glycinin followed the method of Nagano et al. with slight modifications. 1 Briefly, the defatted soy flour was dispersed in 15-fold water in weight and adjusted to pH 7.5 with 2 M NaOH. This slurry was then centrifuged (9000×g, 30 min) at 4 o C. Dry sodium bisulfite (SBS) was then added to the supernatant (0.98 g SBS/L), the pH of the solution was adjusted to 6.4 with 2 M HC1, and the obtained turbid dispersion was kept at 4 o C overnight. After that, the dispersion was centrifuged (6500×g, 30 min) at 4 o C. The precipitates which consisted predominantly of glycinin were dispersed in 5-fold water and the pH was adjusted to 7. The glycinin solution was then dialyzed against Millipore water for two days at 4 o C and then freeze-dried to acquire the final product. SDS-PAGE analysis (Figure S1) shows that the purity of the glycinin is 97.5%.

Coating of coverslips for microscopy observation.
In order to prevent the wetting and spreading of the coacervates on the coverslip, the coverslip was coated with bovine serum albumin (BSA). BSA solution of 10 g/L in water was prepared. The coverslip was washed with ethanol and water, sequentially. Then it was dried under nitrogen flow. A droplet (50 µL) of BSA solution was deposited and spread evenly onto the coverslip (26×56 mm, Waldemar Knittel Glasbearbeitungs GmbH, Germany) with a pipette tip and kept at room temperature until drying. The dried surface of the coated side was washed with small amount of Millipore water and then dried using nitrogen flow.
Microscopy observation chamber. The chamber for microscopy observation consisted of two coverslips pressed against an adhesive rubber spacer with a thickness of 3 mm. The lateral dimension of the bottom coverslip was 26×56 mm with a thickness of 0.17±0.01 mm which was coated by BSA, and the top coverslip had lateral dimension of 22×22 mm and its thickness was 0.17±0.01 mm. Before measurement, a droplet of 20 µL sample was transferred onto the coverslip and then sealed by the top coverslip with the spacer in the middle.
Phase diagrams. Soy glycinin solutions with different concentrations were prepared from a stock solution at C = 100 g/L and pH 7. The protein solution was set to different pH by adding 0.1 M HCl or NaOH and to different temperature by incubating the suspension in a water bath. The NaCl concentration of the protein solutions was then set by adding aliquots of a concentrated NaCl solution (0.2 to 4 M) with the same temperature as the protein solution. All the protein suspensions were incubated at different temperature for 18 h to ensure complete phase separation. The binodal NaCl concentrations (C * and C ** ) were determined either by direct visual observation for presence of two phases in the sample vial for suspensions with protein concentration ≥ 5 g/L or by observation under a microscope in bright field mode using a water immersion objective (63×/1.2NA) on Zeiss Axio Observer microscope for suspensions with protein concentration < 5 g/L. We confirmed that results from visual observation of the sample are consistent with the microscopy data.
Turbidity measurement. The turbidity of the soy glycinin solutions was determined by measuring the absorbance of the suspension at 500 nm immediately after sample preparation using Unicam UV-Vis Spectrometer (Thermo Scientific, United States). Cuvettes with path lengths of 0.5 cm were used. Contrary to the measurements used for building the phase diagram, no incubation for longer times was done because of sedimentation of the droplets.

Confocal laser scanning microscopy (CLSM) and polarization microscopy.
Rhodamine B or FITC-dextran were mixed with glycinin suspension to a final concentration of 5 mg/L or 1 g/L, to visualize the protein dense phase or probe the mesh size of the condensates, respectively. The suspension was immediately placed in an observation chamber. The suspension was examined for the presence of microstructures at 23 °C with a confocal laser scanning microscope (SP5 DMI 6000, Leica Microsystems Heidelberg GmbH, Germany) using a water immersion objective (63×/1.2NA). The polarization images were taken by a confocal laser scanning microscope (SP8, Leica Microsystems Heidelberg GmbH, Germany) equipped with polarizers; 20×/0.4NA objective was used.
Coalescence dynamics. Coalescence dynamics of protein coacervates was investigated following an approach of Brangwynne et al. with slight modifications. 2 The aspect ratio of protein coacervates was determined by fitting an ellipse to the shape and calculating ⁄ , where and are the long and short axes of the ellipse. For analysis of fusing coacervates, the time evolution of this aspect ratio was fitted to a function of the form ⁄ = 1 + ( 0 0 ⁄ − 1) • (− ⁄ ), where t is time, is the characteristic relaxation time, and 0 0 ⁄ is the initial aspect ratio. We define the length scale of these fusing coacervates using the final diameter of the coacervates .

Plots of vs.
were fitted to a line of the form = , to determine the inverse capillary velocity ⁄ . Note that we considered only droplets of similar sizes to reduce the scatter in the data and to reflect the assumptions of the model for droplets with identical radii. 3 The measurements were done with Zeiss Axio Observer Microscope using the bright field mode equipped with Hamamatsu ORCA R2 CCD camera. A water immersion objective (63×/1.2NA) was used.
The obtained data (Figure 1e,f) yield for the inverse capillary velocity ⁄ ≈9.69 s/µm. Theoretical arguments and experiment work (see e.g. references in 4 ) suggest that the interfacial tension is ≈ 2 ⁄ , where is the thermal energy and d is a typical length scale. If we assume ≈ 5 nm for the glycinin molecule, 5, 6 we estimate ≈ 0.16 mN/m, yielding for the droplet viscosity ≈1.6 kPa.s. This extremely high viscosity is consistent with the very slow molecular diffusion as revealed from fluorescence recovery after photobleaching (FRAP) measurements, showing almost no recovery within minutes ( Figure S2).
Protein labeling. Labeling of soy glycinin with fluorescein isothiocyanate isomer (FITC) was performed according to the method reported by Sağlam et al. with modification. 7 A 20 g/L soy glycinin solution was prepared in 0.1 M carbonate buffer (pH 9). FITC was dissolved in DMSO at 4 g/L. FITC solution was slowly added into the protein solution with gentle stirring to a final concentration of 0.2 g/L. The sample was incubated in the dark while stirring at 23 ºC for 3 h. The protein solution was then dialyzed against Millipore water in the dark at 4 ºC for 60 h to remove excess FITC. Millipore water was refreshed every 12 h. The pH of the labeled protein solution was adjusted to 7.4 by adding 0.1 M NaOH and this labeled protein solution was used for FRAP measurement.
Fluorescence recovery after photo bleaching (FRAP). FRAP measurements were performed on Leica SP8 confocal laser scanning microscope equipped with a FRAP booster. Coacervation of the soy glycinin was triggered by mixing protein solution (5 g/L) with equal volume of 0.2 M NaCl giving a final protein concentration of 2.5 g/L and salt concentration of 0.1 M. FITC-soy glycinin accounted for 10% of the total protein concentration. Circular region of interest (ROI) with diameter of 2 µm on the coacervate droplet (8 µm in diameter) was bleached using 3 iterative pulses of total time ~3 s using the Argon laser at 488 nm. Fluorescence intensities from ROI corresponding to photobleaching were analyzed using ImageJ.

Electrophoresis.
Reducing sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) was performed on a discontinuous buffered system, using 12% separating gel. The protein sample was mixed with 5× reducing sample buffer which contained β-mercaptoethanol and then heated at 90 °C for 10 min. After the electrophoresis, the gel was stained using InstantBlue™ solution for 1 h and then rinsed with Millipore water. The protein composition was analyzed by measuring the brightness of the bands using ImageJ.