Learning the molecular grammar of protein condensates from sequence determinants and embeddings

Intracellular phase separation of proteins into biomolecular condensates is increasingly recognized as a process with a key role in cellular compartmentalization and regulation. Different hypotheses about the parameters that determine the tendency of proteins to form condensates have been proposed, with some of them probed experimentally through the use of constructs generated by sequence alterations. To broaden the scope of these observations, we established an in silico strategy for understanding on a global level the associations between protein sequence and phase behavior and further constructed machine-learning models for predicting protein liquid–liquid phase separation (LLPS). Our analysis highlighted that LLPS-prone proteins are more disordered, less hydrophobic, and of lower Shannon entropy than sequences in the Protein Data Bank or the Swiss-Prot database and that they show a fine balance in their relative content of polar and hydrophobic residues. To further learn in a hypothesis-free manner the sequence features underpinning LLPS, we trained a neural network-based language model and found that a classifier constructed on such embeddings learned the underlying principles of phase behavior at a comparable accuracy to a classifier that used knowledge-based features. By combining knowledge-based features with unsupervised embeddings, we generated an integrated model that distinguished LLPS-prone sequences both from structured proteins and from unstructured proteins with a lower LLPS propensity and further identified such sequences from the human proteome at a high accuracy. These results provide a platform rooted in molecular principles for understanding protein phase behavior. The predictor, termed DeePhase, is accessible from https://deephase.ch.cam.ac.uk/.

Liquid–liquid phase separation (LLPS) is a widely occurring biomolecular process that underpins the formation of membraneless organelles within living cells (1–4). This phenomenon and the resulting condensate bodies are increasingly recognized to play important roles in a wide range of biological processes, including the onset and development of metabolic diseases and cancer (5–11). Understanding the factors that drive the formation of protein-rich biomolecular condensates has thus become an important objective and been the focus of a large number of studies, which have collectively yielded valuable information about the factors that govern protein phase behavior (3, 4, 12, 13).While changes in extrinsic conditions, such as temperature, ionic strength, or the level of molecular crowding, can strongly modulate LLPS (14–17), of fundamental importance to condensate formation is the linear amino acid sequence of a protein, its primary structure. A range of sequence-specific factors governing the formation of protein condensates have been postulated with electrostatic interactions, π–π and cation–π contacts, and hydrophobic interactions and the valency and patterning of the low-complexity regions (LCRs) in particular brought forward as central features (12, 13, 18–22). The predictive power of some of these hypotheses has been recently reviewed (23). In parallel, studies examining the relationship between protein phase behavior and its sequence alterations through deletion, truncation, and site-specific mutation events have determined various sequence-specific features to be important in modulating the protein phase separation of specific proteins, such as the high abundance of arginine and tyrosine residues in the context of the fused in sarcoma (FUS)-family proteins (22), the positioning of tryptophan and other aromatic amino acid residues in TAR DNA-binding protein 43 (TDP-43) (24), arginine- and glycine-rich disordered domains in LAF-1 protein (25), and multivalent interactions for the UBQLN2 protein (26).To broaden the scope of these observations and understand on a global level the associations between the primary structure of a protein and its tendency to form condensates, here, we developed an in silico strategy for analyzing the associations between LLPS propensity of a protein and its amino acid sequence and used this information to construct machine-learning classifiers for predicting LLPS propensity from the amino acid sequence (Fig. 1). Specifically, by starting with a previously published LLPSDB database collating information on protein phase behavior under different environmental conditions (27) and by analyzing the concentration under which LLPS had been observed to take place in these experiments, we constructed two datasets including sequences of different LLPS propensity and compared them to fully ordered structures from the Protein Data Bank (PDB) (29) as well as the Swiss-Prot (30) database. We observed phase-separating proteins to be more hydrophobic, more disordered, and of lower Shannon entropy and have their low-complexity regions enriched in polar residues. Moreover, high LLPS propensity correlated with high abundance of polar residues yet the lowest saturation concentrations were reached when their abundance was balanced with a sufficiently high hydrophobic content.Open in a separate windowFig. 1.(A) DeePhase predicts the propensity of proteins to undergo phase separation by combining engineered features computed directly from protein sequences with protein sequence embedding vectors generated using a pretrained language model. The DeePhase model was trained using three datasets, namely two classes of intrinsically disordered proteins with a different LLPS propensity (LLPS+ and LLPS−) and a set of structured sequences (PDB*). (B) To generate the LLPS+ and LLPS− datasets, the entries in the LLPSDB database (27) were filtered for single-protein systems. The constructs that phase separated at an average concentration below c = 100 μM were classified as having a high LLPS propensity (LLPS+; 137 constructs from 77 UniProt IDs) with the remaining 25 constructs together with constructs that had not been observed to phase separate homotypically classified as low-propensity dataset (LLPS−; 84 constructs from 52 UniProt IDs). (C) The 221 sequences clustered into 123 different clusters [Left, CD-hit clustering algorithm (28) with the lowest threshold of 0.4]. (Right) The 110 parent sequences showed high diversity by forming 94 distinct clusters. (D) The PDB* dataset (1,563 constructs) was constructed by filtering the entries in the PDB (29) to fully structured full-protein single chains and clustering for sequence similarity with a single entry selected from each cluster.Moreover, we used the outlined sequence-specific features as well as implicit protein sequence embeddings generated using a neural network-derived word2vec model and trained classifiers for predicting the propensity of unseen proteins to phase separate. We showed that even though the latter strategy required no specific feature engineering, it allowed constructing classifiers that were comparably effective at identifying LLPS-prone sequences as the model that used knowledge-based features, demonstrating that language models can learn the molecular grammar of phase separation. Our final model, combining knowledge-based features with unsupervised embeddings, showed a high performance both when distinguishing LLPS-prone proteins from structured ones and when identifying them within the human proteome. Overall, our results shed light onto the physicochemical factors modulating protein condensate formation and provide a platform rooted in molecular principles for the prediction of protein phase behavior.     

Regulatory mechanisms of tau protein fibrillation under the conditions of liquid–liquid phase separation

One of the hallmarks of Alzheimer’s disease and several other neurodegenerative disorders is the aggregation of tau protein into fibrillar structures. Building on recent reports that tau readily undergoes liquid–liquid phase separation (LLPS), here we explored the relationship between disease-related mutations, LLPS, and tau fibrillation. Our data demonstrate that, in contrast to previous suggestions, pathogenic mutations within the pseudorepeat region do not affect tau441’s propensity to form liquid droplets. LLPS does, however, greatly accelerate formation of fibrillar aggregates, and this effect is especially dramatic for tau441 variants with disease-related mutations. Most important, this study also reveals a previously unrecognized mechanism by which LLPS can regulate the rate of fibrillation in mixtures containing tau isoforms with different aggregation propensities. This regulation results from unique properties of proteins under LLPS conditions, where total concentration of all tau variants in the condensed phase is constant. Therefore, the presence of increasing proportions of the slowly aggregating tau isoform gradually lowers the concentration of the isoform with high aggregation propensity, reducing the rate of its fibrillation. This regulatory mechanism may be of direct relevance to phenotypic variability of tauopathies, as the ratios of fast and slowly aggregating tau isoforms in brain varies substantially in different diseases.

Tau is a major neuronal protein that plays a key role in Alzheimer’s disease (AD) and a number of other neurodegenerative disorders that are collectively classified as tauopathies. The latter include frontotemporal dementia with parkinsonism linked to chromosome 17 (FTDP-17), progressive supranuclear palsy, Pick’s disease, corticobasal degeneration, and chronic traumatic encephalopathy (1–5). Under normal physiological conditions, tau is localized to axons where it is involved in the assembly of microtubules (1–6). In tauopathies, the protein self-associates into different forms of filaments that contain largely hyperphosphorylated tau and have properties of amyloid fibrils (1–5).Alternative splicing of the MAPT gene that encodes tau results in six major isoforms in the human central nervous system. These isoforms differ with respect to the number of N-terminal inserts as well as the number of 31 to 32 residue pseudorepeat sequences in the C-terminal part of the protein (1–5). Structurally, tau is largely an intrinsically disordered protein, with local secondary structures existing only within the pseudorepeat region (1, 7). A large number of mutations have been identified in the latter region that correlate with inherited cases of FTDP-17 (8, 9). These mutations not only diminish the ability of tau to promote microtubule assembly, but many also promote self-association of tau into amyloid fibrils (10–12). This strongly suggests that tau misfolding and aggregation is one of the key events in disease pathogenesis.A number of recent reports indicate that purified full-length tau (tau441) has a high propensity to undergo liquid–liquid phase separation (LLPS) in vitro in the presence of crowding agents that emulate the high concentration of macromolecules in the cell. This was observed both for the phosphorylated (13) and nonphosphorylated protein (14–16), and it was determined that tau LLPS is driven largely by attractive electrostatic intermolecular interactions between the negatively charged N-terminal and positively charged middle/C-terminal regions of the protein (15). Tau condensation into droplets (complex coacervation) was also observed in the presence of polyanions such as RNA or heparin (17, 18). These observations in vitro are partially supported by studies in cells (13, 19–24), especially within the context of tau interaction with microtubules (21). However, it remains unclear whether tau could undergo LLPS in cells on its own or, rather, its recruitment to membraneless organelles such as stress granules is largely driven by interactions with other proteins and/or RNA. These limitations notwithstanding, the observations that tau has a propensity for LLPS have potentially important implications for the pathogenic process in tauopathies, as studies with other proteins involved in neurodegenerative diseases (e.g., TDP-43, FUS) indicate that the environment of liquid droplets is conducive to the pathological aggregation of these proteins (25–32). In line with these findings, it was recently suggested that LLPS can initiate tau aggregation. However, the evidence for this was very limited and largely based on optical microscopy observations (13).In the present study, we explored the relationship between pathogenic mutations of tau, protein LLPS, and aggregation into amyloid fibrils. Our data show that, in contrast to previous suggestions (13), pathogenic mutations within the pseudorepeat region do not affect the propensity of tau to undergo LLPS. These mutations, however, do dramatically accelerate the liquid-to-solid phase transition within the droplets, leading to rapid formation of fibrillar aggregates. Most important, this study also reveals a previously unrecognized mechanism by which LLPS can regulate the rate of amyloid formation in mixtures containing tau isoforms with different aggregation propensities. These findings strongly suggest that LLPS may play a major regulatory role in the formation of pathological tau aggregates in neurodegenerative diseases.     

The inverse and direct Hofmeister series for lysozyme

Anion effects on the cloud-point temperature for the liquid−liquid phase transition of lysozyme were investigated by temperature gradient microfluidics under a dark field microscope. It was found that protein aggregation in salt solutions followed 2 distinct Hofmeister series depending on salt concentration. Namely, under low salt conditions the association of anions with the positively charged lysozyme surface dominated the process and the phase transition temperature followed an inverse Hofmeister series. This inverse series could be directly correlated to the size and hydration thermodynamics of the anions. At higher salt concentrations, the liquid–liquid phase transition displayed a direct Hofmeister series that correlated with the polarizability of the anions. A simple model was derived to take both charge screening and surface tension effects into account at the protein/water interface. Fitting the thermodynamic data to this model equation demonstrated its validity in both the high and low salt regimes. These results suggest that in general positively charged macromolecular systems should show inverse Hofmeister behavior only at relatively low salt concentrations, but revert to a direct Hofmeister series as the salt concentration is increased.     

Amphiphilic proteins coassemble into multiphasic condensates and act as biomolecular surfactants

Cells contain membraneless compartments that assemble due to liquid–liquid phase separation, including biomolecular condensates with complex morphologies. For instance, certain condensates are surrounded by a film of distinct composition, such as Ape1 condensates coated by a layer of Atg19, required for selective autophagy in yeast. Other condensates are multiphasic, with nested liquid phases of distinct compositions and functions, such as in the case of ribosome biogenesis in the nucleolus. The size and structure of such condensates must be regulated for proper biological function. We leveraged a bioinspired approach to discover how amphiphilic, surfactant-like proteins may contribute to the structure and size regulation of biomolecular condensates. We designed and examined families of amphiphilic proteins comprising one phase-separating domain and one non–phase-separating domain. In particular, these proteins contain the soluble structured domain glutathione S-transferase (GST) or maltose binding protein (MBP), fused to the intrinsically disordered RGG domain from P granule protein LAF-1. When one amphiphilic protein is mixed in vitro with RGG-RGG, the proteins assemble into enveloped condensates, with RGG-RGG at the core and the amphiphilic protein forming the surface film layer. Importantly, we found that MBP-based amphiphiles are surfactants and influence droplet size, with increasing surfactant concentration resulting in smaller droplet radii. In contrast, GST-based amphiphiles at increased concentrations coassemble with RGG-RGG into multiphasic structures. We propose a mechanism for these experimental observations, supported by molecular simulations of a minimalist model. We speculate that surfactant proteins may play a significant role in regulating the structure and function of biomolecular condensates.

The intracellular environment is like a complex emulsion. This paradigm originated more than a century ago but is enjoying a renaissance, with recent discoveries revealing the important role of liquid–liquid phase separation (LLPS) in biology (1–3). LLPS of proteins and nucleic acids underlies the formation of membraneless organelles, alternatively called biomolecular condensates, which are distinct intracellular compartments that lack a delimiting membrane (2, 3). Biomolecular condensates contribute to numerous cell functions, including stress response, gene regulation, and signaling (4). Conversely, aberrant phase separation due to mutations and age-related processes is implicated in diseases such as neurodegeneration and cancer (5). Deciphering the rules of self-assembly of biomolecular condensates has therefore emerged as a promising avenue for elucidating fundamental principles of biological structure, function, and dysfunction.Despite significant recent progress in understanding the biophysics of biomolecular condensates, many open questions remain (6). One key question is what molecular phenomena govern the spontaneous assembly of condensates with core-shell or multiphasic structures. Another important question is how cells tune the size of biomolecular condensates. Here, we sought to gain insight into both questions by examining how amphiphilic, surfactant-like proteins contribute to the self-assembly and regulation of biomolecular condensates. Amphiphiles are typically defined as molecules comprising separate hydrophilic and hydrophobic parts. Here, we note the etymology (in Greek, “amphi” means both, and “philia” means friendship or love) and use the term amphiphile to describe proteins comprising one domain that has affinity for biomolecular condensates and one domain that has affinity for the dilute phase.Surfactants are substances, generally amphiphiles, that adsorb to interfaces and decrease interfacial tension. Extracellularly, pulmonary surfactant lining the alveoli plays a vital role in lung physiology by reducing the work of breathing (7, 8). However, the role of surfactants in the emulsion-like intracellular milieu is just beginning to be explored (9). In biological systems, the surfactant-like protein Ki-67 prevents individual chromosomes from coalescing during early stages of mitosis by forming a repulsive molecular brush layer (10). Some biomolecular condensates are reminiscent of surfactant-laden emulsions, although their physical chemistry remains to be elucidated. For instance, Atg19 forms a thin surface layer surrounding Ape1 condensates that is necessary for selective autophagy of Ape1 in yeast (11). Inspired by such examples, we hypothesized that a minimal system comprising surfactant-like proteins interacting with phase-separating proteins could recapitulate enveloped condensate structures observed in nature.Moreover, condensates exhibit a variety of multiphase and multilayer structures underpinning their biological functions. Bre1 assembles as a shell surrounding Lge1 condensates, generating a catalytic condensate that functions to accelerate ubiquitination of histone H2B in yeast (12). The nucleolus is comprised of coexisting liquid phases of differing interfacial tensions (13), while P granules contain coexisting liquid and gel phases (14). Stress granules (15), nuclear speckles (16), paraspeckles (17), and reconstituted polypeptide/RNA complex coacervates also exhibit core-shell structures sensitive to stoichiometry and competitive binding (18–21). Functionally related condensates can remain in contact without coalescing, as in the case of stress granules and P-bodies (22) or P granules and Z granules (23). We asked whether amphiphilic proteins could contribute to the complex morphologies of biomolecular condensates that have been observed within cells, just as synthetic amphiphiles and surfactant systems exhibit rich structures and phase behaviors (24).Surfactant-like proteins could have additional important functional consequences, including but not limited to modulating biomolecular condensate size, which in turn influences biochemical processes through condensate size-dependent effects on molecular concentrations and diffusion (25, 26). Biomolecular condensates are often observed in cells as multiple smaller droplets rather than as a single larger droplet, even though the latter is expected to be thermodynamically favored. Recent studies have attributed the apparent metastability of biomolecular condensates in various contexts to surface charge (27), cytoskeletal caging (28, 29), membrane association (30), and exhaustion of available binding sites (31). Active processes can also maintain the emulsified, multidroplet state in vivo (32). An additional possibility, which we examine here, is that surfactant proteins may help stabilize biomolecular condensates.To address these questions, we adopted a bottom-up, bioinspired approach, seeking to leverage a simplified system to shed light on the role of amphiphilic proteins in the self-assembly of biomolecular condensates. We designed amphiphilic proteins containing an intrinsically disordered region (IDR) fused to folded domains. The IDR is a phase-separating domain, whereas the folded domains alone are not. When one of these amphiphilic fusion proteins is mixed at low concentrations with the IDR, the two proteins assemble such that the amphiphilic protein forms a film that coats the IDR core. We demonstrate several extensions of this observation, including the important finding that condensate size can be influenced by varying surfactant protein concentration. Furthermore, when amphiphilic proteins with different folded domains are mixed together with the core IDR, we observe competition between different surfactant proteins for binding to the condensate interface. Interestingly, one family of amphiphilic proteins exhibits varied morphologies, including multiphasic condensates, and we map the rich, concentration-dependent phase behavior. To gain mechanistic insight into these experimental observations, we present a minimalistic computational model that recapitulates the range of behaviors observed experimentally by varying the strength of interaction between domains. Our experiments and simulations suggest that amphiphile–condensate assembly is determined by the strength of interaction between the amphiphile and the IDR core as well as interactions between the folded domain of the amphiphile. Taken together, this work illustrates the diverse interfacial phenomena that can arise from interactions between condensates and amphiphilic proteins, notably raising the possibility that surfactant proteins may play a significant role in regulating the structure and function of biomolecular condensates.     

Emerging Silk Material Trends: Repurposing,Phase Separation and Solution-Based Designs

Silk continues to amaze. This review unravels the most recent progress in silk science, spanning from fundamental insights to medical silks. Key advances in silk flow are examined, with specific reference to the role of metal ions in switching silk from a storage to a spinning state. Orthogonal thermoplastic silk molding is described, as is the transfer of silk flow principles for the triggering of flow-induced crystallization in other non-silk polymers. Other exciting new developments include silk-inspired liquid–liquid phase separation for non-canonical fiber formation and the creation of “silk organelles” in live cells. This review closes by examining the role of silk fabrics in fashioning facemasks in response to the SARS-CoV-2 pandemic.     

Herpesvirus Replication Compartments: Dynamic Biomolecular Condensates?

Recent progress has provided clear evidence that many RNA-viruses form cytoplasmic biomolecular condensates mediated by liquid–liquid phase separation to facilitate their replication. In contrast, seemingly contradictory data exist for herpesviruses, which replicate their DNA genomes in nuclear membrane-less replication compartments (RCs). Here, we review the current literature and comment on nuclear condensate formation by herpesviruses, specifically with regard to RC formation. Based on data obtained with human cytomegalovirus (human herpesvirus 5), we propose that liquid and homogenous early RCs convert into more heterogeneous RCs with complex properties over the course of infection. We highlight how the advent of DNA replication leads to the maturation of these biomolecular condensates, likely by adding an additional DNA scaffold.     

Endocytic proteins with prion-like domains form viscoelastic condensates that enable membrane remodeling

Membrane invagination and vesicle formation are key steps in endocytosis and cellular trafficking. Here, we show that endocytic coat proteins with prion-like domains (PLDs) form hemispherical puncta in the budding yeast, Saccharomyces cerevisiae. These puncta have the hallmarks of biomolecular condensates and organize proteins at the membrane for actin-dependent endocytosis. They also enable membrane remodeling to drive actin-independent endocytosis. The puncta, which we refer to as endocytic condensates, form and dissolve reversibly in response to changes in temperature and solution conditions. We find that endocytic condensates are organized around dynamic protein–protein interaction networks, which involve interactions among PLDs with high glutamine contents. The endocytic coat protein Sla1 is at the hub of the protein–protein interaction network. Using active rheology, we inferred the material properties of endocytic condensates. These experiments show that endocytic condensates are akin to viscoelastic materials. We use these characterizations to estimate the interfacial tension between endocytic condensates and their surroundings. We then adapt the physics of contact mechanics, specifically modifications of Hertz theory, to develop a quantitative framework for describing how interfacial tensions among condensates, the membrane, and the cytosol can deform the plasma membrane to enable actin-independent endocytosis.

Endocytosis in eukaryotic cells can occur via two separate mechanisms: actin-dependent and actin-independent pathways. In this study, we used the budding yeast Saccharomyces cerevisiae as a tractable model system to uncover the mechanistic basis for actin-independent endocytosis. This is directly relevant to the early stages of endocytic membrane invagination that occurs in mammalian cells through homologs of the proteins that we identify and study here in yeast (1, 2). In S. cerevisiae, membrane invagination that enables endocytosis is normally driven by growth of membrane-bound branched actin (3). A second actin-independent route to endocytosis is realized when intracellular turgor pressure is reduced. This reduction of turgor pressure alleviates the tension on plasma membranes that would normally oppose membrane invagination (1, 4). Although this actin-independent mechanism is not evident under laboratory conditions, it does occur at the hyperosmotic, high-sucrose concentrations that can be found in the wild when yeast grow on rotting fruit and under industrial fermentation conditions, particularly in the context of bioethanol production (1).In both mechanisms, endocytosis is initiated by the coordinated recruitment of a number of proteins associated with distinct stages of endocytic maturation (5). Clathrin heavy and light chains first interact with initiator proteins (Ede1 and Syp1) to form a lattice on the membrane. Subsequently, early coat proteins such as Sla1, Sla2, Ent1, Ent2, and Yap1801 (6) bind directly to the adaptor–clathrin lattice and form the cortical body (5). Electron microscopy data highlight the existence of hemispherical membraneless bodies around endocytic sites. These bodies are identifiable by following the localization of labeled endocytic coat proteins such as Sla1. The observed Sla1-labeled bodies are known to exclude ribosomes from regions that are near the cortical sites in the cytosol. Importantly, these endocytic bodies form even when actin is not polymerized, and the membrane is flat (7).Many of the coat proteins in bodies that form around endocytic sites include prion-like domains (PLDs). These are low-complexity intrinsically disordered domains that are enriched in polar amino acids such as glutamine, asparagine, glycine, and serine and are interspersed by aromatic residues (6, 8). Proteins with PLDs have the ability to drive the formation of membraneless biomolecular condensates through phase separation in cells (9) and in vitro (10). Condensates are mesoscale, nonstoichiometric macromolecular assemblies that concentrate biomolecules (11–13). Here, we show that endocytosis in S. cerevisiae involves the concentration of PLD-containing proteins, including the essential protein Sla1, within biomolecular condensates that form at cortical sites (14). Inferences from indirect measurements suggest that these condensates have viscoelastic properties and that they are scaffolded by a dense network of PLD-containing proteins. We show that condensate formation requires an intact PLD and the coat protein Sla1 is at the hub of the condensate-driving protein–protein interaction network. The distinctive compositional biases within PLDs of coat proteins contribute to condensate formation and function. We present a model, motivated by Hertz contact theory (15–17), to provide a plausible explanation for how interfacial tensions among condensates, the membrane, and the cytosol can enable membrane invagination and drive actin-independent endocytosis. This model shows that the formation of condensates and cohesiveness of molecular interactions within them are likely to be essential for mechanoactive processes associated with actin-independent endocytosis.     

Spontaneous nucleation and fast aggregate-dependent proliferation of α-synuclein aggregates within liquid condensates at neutral pH

The aggregation of α-synuclein into amyloid fibrils has been under scrutiny in recent years because of its association with Parkinson’s disease. This process can be triggered by a lipid-dependent nucleation process, and the resulting aggregates can proliferate through secondary nucleation under acidic pH conditions. It has also been recently reported that the aggregation of α-synuclein may follow an alternative pathway, which takes place within dense liquid condensates formed through phase separation. The microscopic mechanism of this process, however, remains to be clarified. Here, we used fluorescence-based assays to enable a kinetic analysis of the microscopic steps underlying the aggregation process of α-synuclein within liquid condensates. Our analysis shows that at pH 7.4, this process starts with spontaneous primary nucleation followed by rapid aggregate-dependent proliferation. Our results thus reveal the microscopic mechanism of α-synuclein aggregation within condensates through the accurate quantification of the kinetic rate constants for the appearance and proliferation of α-synuclein aggregates at physiological pH.

Parkinson’s disease is the most common neurodegenerative movement disorder (1, 2). A distinctive pathophysiological signature of this disease is the presence of abnormal intraneuronal protein deposits known as Lewy bodies (3, 4). One of the main components of Lewy bodies is α-synuclein (5), a peripheral membrane protein highly abundant at neuronal synapses (6, 7) and genetically linked with Parkinson’s disease (8, 9). This 140-residue disordered protein can be subdivided into three domains, an amphipathic N-terminal region (amino acids 1 to 60), a central hydrophobic region (non-amyloid-β component, or NAC, amino acids 61 to 95), and an acidic proline-rich C-terminal tail (amino acids 96 to 140) (7). Although α-synuclein aggregation is characteristic of Parkinson’s disease and related synucleinopathies, the corresponding mechanism and its possible pathological role in disease are not yet fully understood.Generally, the aggregation process of proteins proceeds through a series of interconnected microscopic steps, including primary nucleation, elongation, and secondary nucleation (10, 11). During primary nucleation, the self-assembly of proteins from their native, monomeric form leads to the formation of oligomeric species, an event that may occur in solution or on surfaces including biological membranes (12, 13). The formation of these oligomers is typically a slow event governed by high kinetic barriers (10, 11). Once formed, the oligomers may convert into ordered assemblies rich in β structure, which are capable of further growth into fibrillar aggregates (14). In many cases, the surfaces of existing fibrillar aggregates then further catalyze the formation of new oligomers (15, 16). This secondary nucleation process is typically characterized by the assembly of protein monomers on the surface of fibrils that eventually nucleate into new oligomeric species (15, 16). This autocatalytic mechanism generates rapid fibril proliferation (15).In the case of the aggregation process of α-synuclein, several key questions are still open, including two that we are addressing in this study. The first concerns whether there are cellular conditions under which α-synuclein can undergo spontaneous aggregation, and the second whether the proliferation of α-synuclein fibrils by aggregate-dependent feedback processes can take place at physiological pH. These questions are relevant because according to our current knowledge, α-synuclein aggregation does not readily take place spontaneously in the absence of contributing factors such as lipid membranes. Furthermore, secondary nucleation contributes significantly to the aggregation process only at acidic pH (13, 17). It thus remains challenging to rationalize the links between α-synuclein aggregation and Parkinson’s disease.To address this problem, we investigated whether it is possible to leverage the recent finding that α-synuclein can undergo a phase separation process resulting in the formation of dense liquid condensates (18–21). Phase separation has recently emerged as a general phenomenon associated with a wide variety of cellular functions (22–25) and closely linked with human disease (23, 26–29). This process has been reported for a wide range of proteins implicated in neurodegenerative conditions, including tau, fused in sarcoma (FUS), and TAR DNA binding protein 43 (TDP-43) (30–32). Since it has also been shown that protein aggregation can take place within liquid condensates (19, 26, 32–36), we asked whether it is possible to characterize at the microscopic level the condensate-induced aggregation mechanism of α-synuclein by determining the kinetic rate constants of the corresponding microscopic processes.To enable the accurate determination of the rate constants for the microscopic steps in α-synuclein aggregation within condensates, we developed fluorescence-based aggregation assays to monitor both the spontaneous aggregation of α-synuclein and the aggregation in the presence of aggregate seeds. Using these assays within the framework of a kinetic theory of protein aggregation (10, 11, 37), we show that α-synuclein can undergo spontaneous homogenous primary nucleation and fast aggregate-dependent proliferation within condensates at physiological pH.     

Interfacial Crystallization and Supramolecular Self-Assembly of Spider Silk Inspired Protein at the Water-Air Interface

Macromolecular assembly into complex morphologies and architectural shapes is an area of fundamental research and technological innovation. In this work, we investigate the self-assembly process of recombinantly produced protein inspired by spider silk (spidroin). To elucidate the first steps of the assembly process, we examined highly concentrated and viscous pendant droplets of this protein in air. We show how the protein self-assembles and crystallizes at the water–air interface into a relatively thick and highly elastic skin. Using time-resolved in situ synchrotron x-ray scattering measurements during the drying process, we showed that the skin evolved to contain a high β-sheet amount over time. We also found that β-sheet formation strongly depended on protein concentration and relative humidity. These had a strong influence not only on the amount, but also on the ordering of these structures during the β-sheet formation process. We also showed how the skin around pendant droplets can serve as a reservoir for attaining liquid–liquid phase separation and coacervation from the dilute protein solution. Essentially, this study shows a new assembly route which could be optimized for the synthesis of new materials from a dilute protein solution and determine the properties of the final products.     

Building and Breaking Bonds by Homogenous Nucleation in Glass-Forming Melts Leading to Transitions in Three Liquid States

The thermal history of melts leads to three liquid states above the melting temperatures T_m containing clusters—bound colloids with two opposite values of enthalpy +Δε_lg × ΔH_m and −Δε_lg × ΔH_m and zero. All colloid bonds disconnect at T_n+ > T_m and give rise in congruent materials, through a first-order transition at T_LL = T_n+, forming a homogeneous liquid, containing tiny superatoms, built by short-range order. In non-congruent materials, (T_n+) and (T_LL) are separated, T_n+ being the temperature of a second order and T_LL the temperature of a first-order phase transition. (T_n+) and (T_LL) are predicted from the knowledge of solidus and liquidus temperatures using non-classical homogenous nucleation. The first-order transition at T_LL gives rise by cooling to a new liquid state containing colloids. Each colloid is a superatom, melted by homogeneous disintegration of nuclei instead of surface melting, and with a Gibbs free energy equal to that of a liquid droplet containing the same magic atom number. Internal and external bond number of colloids increases at T_n+ or from T_n+ to T_g. These liquid enthalpies reveal the natural presence of colloid–colloid bonding and antibonding in glass-forming melts. The Mpemba effect and its inverse exist in all melts and is due to the presence of these three liquid states.     

Formation and Function of Liquid-Like Viral Factories in Negative-Sense Single-Stranded RNA Virus Infections

Liquid–liquid phase separation (LLPS) represents a major physiochemical principle to organize intracellular membrane-less structures. Studies with non-segmented negative-sense (NNS) RNA viruses have uncovered a key role of LLPS in the formation of viral inclusion bodies (IBs), sites of viral protein concentration in the cytoplasm of infected cells. These studies further reveal the structural and functional complexity of viral IB factories and provide a foundation for their future research. Herein, we review the literature leading to the discovery of LLPS-driven formation of IBs in NNS RNA virus-infected cells and the identification of viral scaffold components involved, and then outline important questions and challenges for IB assembly and disassembly. We discuss the functional implications of LLPS in the life cycle of NNS RNA viruses and host responses to infection. Finally, we speculate on the potential mechanisms underlying IB maturation, a phenomenon relevant to many human diseases.     

Glasses denser than the supercooled liquid

When aged below the glass transition temperature, Tg, the density of a glass cannot exceed that of the metastable supercooled liquid (SCL) state, unless crystals are nucleated. The only exception is when another polyamorphic SCL state exists, with a density higher than that of the ordinary SCL. Experimentally, such polyamorphic states and their corresponding liquid–liquid phase transitions have only been observed in network-forming systems or those with polymorphic crystalline states. In otherwise simple liquids, such phase transitions have not been observed, either in aged or vapor-deposited stable glasses, even near the Kauzmann temperature. Here, we report that the density of thin vapor-deposited films of N,N′-bis(3-methylphenyl)-N,N′-diphenylbenzidine (TPD) can exceed their corresponding SCL density by as much as 3.5% and can even exceed the crystal density under certain deposition conditions. We identify a previously unidentified high-density supercooled liquid (HD-SCL) phase with a liquid–liquid phase transition temperature (TLL) ∼35 K below the nominal glass transition temperature of the ordinary SCL. The HD-SCL state is observed in glasses deposited in the thickness range of 25 to 55 nm, where thin films of the ordinary SCL have exceptionally enhanced surface mobility with large mobility gradients. The enhanced mobility enables vapor-deposited thin films to overcome kinetic barriers for relaxation and access the HD-SCL state. The HD-SCL state is only thermodynamically favored in thin films and transforms rapidly to the ordinary SCL when the vapor deposition is continued to form films with thicknesses more than 60 nm.

Glasses are formed when the structural relaxations in supercooled liquids (SCLs) become too slow, causing the system to fall out of equilibrium at the glass transition temperature (Tg). The resulting out-of-equilibrium glass state has a thermodynamic driving force to evolve toward the SCL state through physical aging (1). At temperatures just below Tg, the extent of equilibration is limited by the corresponding SCL state, while at much lower temperatures, equilibration is limited by the kinetic barriers for relaxation. As such, the degree of thermodynamic stability achieved through physical aging is limited (2).Physical vapor deposition (PVD) is an effective technique to overcome kinetic barriers for relaxation to produce thermodynamically stable glasses (3–10). The accelerated equilibration in these systems is due to their enhanced surface mobility (11–14). During PVD, when the substrate temperature is held below Tg, molecules or atoms can undergo rearrangements and adopt more stable configurations at the free surface and proximate layers underneath (13). After the molecules are buried deeper into the film, their relaxation dynamics significantly slow down, which prevents further equilibration. Through this surface-mediated equilibration process, stable glasses can achieve low-energy states on the potential energy landscape that would otherwise require thousands or millions of years of physical aging (2, 3, 15, 16).As such, the degree of enhanced surface mobility and mobility gradients are critical factors in the formation of stable glasses (3, 11, 17, 18). While the effect of film thickness on the surface mobility and gradients of liquid-quenched (LQ) glasses has been studied in the past (19, 20), there are limited data on the role of film thickness in the stability of vapor-deposited glasses. In vapor-deposited toluene, it has been shown that decreasing the film thickness from 70 to 5 nm can increase the thermodynamic stability but decrease the apparent kinetic stability (5, 6). In contrast, thin films covered with a top layer of another material do not show a significant evidence of reduced kinetic stability (21), indicating the nontrivial role of mobility gradients in thermal and kinetic stability.Stable glasses of most organic molecules, with short-range intramolecular interactions, have properties that are indicative of the same corresponding metastable SCL state as LQ and aged glasses, without any evidence of the existence of generic liquid–liquid phase transitions that can potentially provide a resolution for the Kauzmann entropy crisis (22). The Kauzmann crisis occurs at the Kauzmann temperature (TK), where the extrapolated SCL has the same structural entropy as the crystal, producing thermodynamically impossible states just below this temperature. Recently, Beasley et al. (16) showed that near-equilibrium states of ethylbenzene can be produced using PVD down to 2 K above TK and hypothesized that any phase transition to an “ideal glass” state to avoid the Kauzmann crisis must occur at TK.In some glasses of elemental substances (23, 24) and hydrogen-bonding compounds (25, 26), liquid–liquid phase transitions can occur between polyamorphic states with distinct local packing structures that correspond to polymorphic crystalline phases. For example, at high pressures, high- and low-density supercooled water phases are interconvertible through a first-order phase transition (27, 28). Recent studies have demonstrated that such polyamorphic states can also be accessed through PVD in hydrogen-bonding systems with polymorphic crystal states at depositions above the nominal Tg (29, 30). However, these structure-specific transitions do not provide a general resolution for the Kauzmann crisis.Here, we report the observation of a liquid–liquid phase transition in vapor-deposited thin films of N,N′-bis(3-methylphenyl)-N,N′-diphenylbenzidine (TPD). TPD is a molecular glass former with only short-range intermolecular interactions. When thin films of TPD are vapor deposited onto substrates held at deposition temperatures (Tdep) below the nominal glass transition temperature of bulk TPD, Tg (bulk), films in the thickness range of 25 nm<h<55 nm achieve a high-density supercooled liquid (HD-SCL) state, which has not been previously observed. The liquid–liquid phase transition temperature (TLL) between the ordinary SCL and HD-SCL states is measured to be TLL≃Tg(bulk)−35 K. The density of thin films deposited below TLL tangentially follows the HD-SCL line, which has a stronger temperature dependence than the ordinary SCL. When vapor deposition is continued to produce thicker films (h>60 nm), the HD-SCL state transforms into the ordinary SCL state, indicating that the HD-SCL is only thermodynamically favored in the thin-film geometry. This transition is qualitatively different from the previously reported liquid–liquid phase transitions, as it is not related to a specific structural motif in TPD crystals, and it can only be observed in thin films, indicating that the energy landscape of thin films is favoring this high-density state.We observe an apparent correlation between enhanced mobility gradients in LQ thin films of TPD and the thickness range where HD-SCL states are produced during PVD. We hypothesize that enhanced mobility gradients are essential in providing access to regions of the energy landscape corresponding to the HD-SCL state, which are otherwise kinetically inaccessible. This hypothesis should be further investigated to better understand the origin of this phenomenon.     

Enhancement and maximum in the isobaric specific-heat capacity measurements of deeply supercooled water using ultrafast calorimetry

Knowledge of the temperature dependence of the isobaric specific heat (C_p) upon deep supercooling can give insights regarding the anomalous properties of water. If a maximum in C_p exists at a specific temperature, as in the isothermal compressibility, it would further validate the liquid–liquid critical point model that can explain the anomalous increase in thermodynamic response functions. The challenge is that the relevant temperature range falls in the region where ice crystallization becomes rapid, which has previously excluded experiments. Here, we have utilized a methodology of ultrafast calorimetry by determining the temperature jump from femtosecond X-ray pulses after heating with an infrared laser pulse and with a sufficiently long time delay between the pulses to allow measurements at constant pressure. Evaporative cooling of ∼15-µm diameter droplets in vacuum enabled us to reach a temperature down to ∼228 K with a small fraction of the droplets remaining unfrozen. We observed a sharp increase in C_p, from 88 J/mol/K at 244 K to about 218 J/mol/K at 229 K where a maximum is seen. The C_p maximum is at a similar temperature as the maxima of the isothermal compressibility and correlation length. From the C_p measurement, we estimated the excess entropy and self-diffusion coefficient of water and these properties decrease rapidly below 235 K.

Water is one of the most exceptional liquids due to its importance, abundance, and many properties that are anomalous with respect to a normal liquid (1–3). This anomalous behavior is already evident at ambient conditions and is enhanced when water is supercooled below the freezing point into the metastable regime (2, 4, 5). In particular, the observation that the isothermal compressibility (κ_T), heat capacity (C_p), thermal expansion coefficient (α_P), and correlation length (ξ) appear to diverge toward a singular temperature (T_s) of about 228 K at 1 bar, as estimated by power-law fits (6, 7), has led to several hypotheses about the origin of water’s anomalous properties (2, 3, 8). One of the hypotheses proposes the existence of a liquid–liquid transition in supercooled water between high-density (HDL) and low-density (LDL) liquids, separated by a phase-coexistence line (8, 9) and terminating at a liquid–liquid critical point (LLCP) at positive pressure (8). Beyond the LLCP, at lower pressures, water is characterized by fluctuations between local structures of HDL and LDL (10). The locus of maxima in ξ of these fluctuations defines the Widom line in the pressure–temperature phase diagram, which emanates from the LLCP as an extension of the phase-coexistence line (11). Near the ξWidom line, the other thermodynamic response functions could also have maxima defining κ_T and C_p Widom lines, merging with the ξ Widom line in close proximity to the critical point. Such a merging was observed for the maxima in κ_T and C_p, and for the minimum in α_P at the liquid–gas critical point (LGCP), based on molecular-dynamic (MD) simulations (12).It has been challenging to experimentally determine the existence of a Widom line in supercooled water due to the extremely fast ice-forming crystallization at temperatures below 235 K. Nevertheless, rapid evaporative cooling of micrometer-sized droplets followed by ultrafast interrogation with an X-ray laser have allowed us to probe water at temperatures down to 227 K (13, 14). Recently, using this approach, maxima in ξ and κ_T were observed at 229 K, coinciding with the temperature of the most rapid change of the local tetrahedral structure in the liquid (13). Other experiments using sound velocity in stretched liquid water (15) also predict a maxima in κ_T and C_p. Based on a combination of MD simulations and temperature-dependent structure factor measurements, a consistency was derived with which α_P may also exhibit a minimum at 229 K (16). If all thermodynamic response functions showed evidence of a Widom line with maxima or minima, this would validate the LLCP scenario, more so if they were in close proximity in temperature. Currently, no measurements exist below 236 K (17) for C_p and it is necessary to develop experimental techniques to study water upon deep supercooling where rapid ice crystallization occurs. Measurements of the value of the C_p maximum also allow us to derive to which extent the excess entropy has decreased upon supercooling and compare this to the entropy of low-density amorphous ice (LDA) at the glass transition temperature (18–20). Interest in excess entropy was one of the original motivations in 1969 behind the study of supercooled water (18). Based on the expectation that C_p should decrease upon cooling, the excess entropy was expected to rapidly decrease, as C_p approaches that of LDA. Surprisingly, though, an accelerated increase was observed instead (21–23).Here, we show that the C_p can be measured down to 228 K using a method based on ultrafast calorimetry. The data are consistent with the existence of a maximum of C_p at 229 K, as well as a rapid decrease of the excess entropy at temperatures beyond the Widom line. Fig. 1 shows the experimental setup of our ultrafast calorimetry approach. The droplets are cooled by evaporation and the temperature is calculated using Knudsen’s theory of evaporation and Fourier’s law of heat conduction (24, 25). This approach to determining droplet temperatures has been proven to be successful in various experimental setups (13, 14, 26) and has been validated using ME simulations (25). A 2.05-μm infrared (IR) pulse heats the sample, increasing the temperature of the droplets by 0.5–1 K. The droplets are then probed by a femtosecond X-ray pulse after a 1-µs delay time, allowing the liquid to expand. The difference in the X-ray scattering patterns between IR laser on and off is used as a thermometer. The pattern from each X-ray shot is also used to detect whether Bragg peaks appear from small ice crystals so that crystallized droplets can be excluded from the analysis. Using a calibration curve of the scattering signal versus temperature, we estimate the increase of temperature from the heating pulse and derive the heat capacity at constant pressure, C_p. We observe a rapid increase in C_p at temperatures below 235 K with a maximum appearing at 229 K, followed by a suggested decrease toward lower temperatures. The rise and maximum of C_p is consistent with the existence of a Widom line for C_p as previously observed for κ_T and ξ (13).Open in a separate windowFig. 1.(A) Schematic of the experimental setup (Left) and (B) angularly integrated scattering intensity (Right). The time delay (∆t) between the IR laser and the X-rays is 1 µs. IR laser is ON for every alternate X-ray pulse. The difference in the scattering profile of the laser ON and laser OFF shots is ∼2% of the signal.     

Effect of hydrogen bond cooperativity on the behavior of water

Four scenarios have been proposed for the low-temperature phase behavior of liquid water, each predicting different thermodynamics. The physical mechanism that leads to each is debated. Moreover, it is still unclear which of the scenarios best describes water, because there is no definitive experimental test. Here we address both open issues within the framework of a microscopic cell model by performing a study combining mean-field calculations and Monte Carlo simulations. We show that a common physical mechanism underlies each of the four scenarios, and that two key physical quantities determine which of the four scenarios describes water: (i) the strength of the directional component of the hydrogen bond and (ii) the strength of the cooperative component of the hydrogen bond. The four scenarios may be mapped in the space of these two quantities. We argue that our conclusions are model independent. Using estimates from experimental data for H-bond properties the model predicts that the low-temperature phase diagram of water exhibits a liquid–liquid critical point at positive pressure.     

Phase-separating RNA-binding proteins form heterogeneous distributions of clusters in subsaturated solutions

Macromolecular phase separation is thought to be one of the processes that drives the formation of membraneless biomolecular condensates in cells. The dynamics of phase separation are thought to follow the tenets of classical nucleation theory, and, therefore, subsaturated solutions should be devoid of clusters with more than a few molecules. We tested this prediction using in vitro biophysical studies to characterize subsaturated solutions of phase-separating RNA-binding proteins with intrinsically disordered prion-like domains and RNA-binding domains. Surprisingly, and in direct contradiction to expectations from classical nucleation theory, we find that subsaturated solutions are characterized by the presence of heterogeneous distributions of clusters. The distributions of cluster sizes, which are dominated by small species, shift continuously toward larger sizes as protein concentrations increase and approach the saturation concentration. As a result, many of the clusters encompass tens to hundreds of molecules, while less than 1% of the solutions are mesoscale species that are several hundred nanometers in diameter. We find that cluster formation in subsaturated solutions and phase separation in supersaturated solutions are strongly coupled via sequence-encoded interactions. We also find that cluster formation and phase separation can be decoupled using solutes as well as specific sets of mutations. Our findings, which are concordant with predictions for associative polymers, implicate an interplay between networks of sequence-specific and solubility-determining interactions that, respectively, govern cluster formation in subsaturated solutions and the saturation concentrations above which phase separation occurs.

Phase separation of RNA-binding proteins with disordered prion-like domains (PLDs) and RNA-binding domains (RBDs) is implicated in the formation and dissolution of membraneless biomolecular condensates such as RNA–protein (RNP) granules (1–9). Macroscopic phase separation is a process whereby a macromolecule in a solvent separates into a dilute, macromolecule-deficient phase that coexists with a dense, macromolecule-rich phase (10, 11). In a binary mixture, the soluble phase, comprising dispersed macromolecules that are well mixed with the solvent, becomes saturated at a concentration designated as c_sat. Above c_sat, for total macromolecular concentrations c_tot that are between the binodal and spinodal, phase separation of full-length RNA-binding proteins and PLDs is thought to follow classical nucleation theory (12–15).In classical nucleation theories, clusters representing incipient forms of the new dense phase form within dispersed phases of supersaturated solutions defined by c_tot > c_sat (16, 17). In the simplest formulation of classical nucleation theory (16–18), the free energy of forming a cluster of radius a is ΔF=−4π3a3Δμρn+4πa2γ. Here, Δµ is the difference in the chemical potential between the one-phase and two-phase regimes (see discussion in SI Appendix), which is negative in supersaturated solutions and positive in subsaturated solutions; ρ_n is the number of molecules per unit volume, and γ is the interfacial tension between dense and dilute phases. At temperature T, in a seed-free solution, the degree of supersaturation s is defined as s≡ΔμRT=ln(ctotcsat), where R is the ideal gas constant. Here, s is positive for c_tot > c_sat, and, as s increases, cluster formation becomes more favorable. Above a critical radius a*, the free energy of cluster formation can overcome the interfacial penalty, and the new dense phase grows in a thermodynamically downhill fashion. Ideas from classical nucleation theory have been applied to analyze and interpret the dynamics of phase separation in supersaturated solutions (12, 13, 15). Classical nucleation theories stand in contrast to two-step nucleation theories that predict the existence of prenucleation clusters in supersaturated solutions (19–22). These newer theories hint at the prospect of there being interesting features in subsaturated solutions, where c_tot < c_sat and s < 0.The subsaturated regime, where s is negative, corresponds to the one-phase regime. Ignoring the interfacial tension, the free energy of realizing clusters with n molecules in subsaturated solutions is: ΔF = –nΔµ. Therefore, the probability P(n) of forming a cluster of n molecules in a subsaturated solution is proportional to exp(sn). Accordingly, the relative probability P(n)/P(1) of forming clusters with n molecules will be exp(s(n – 1)). This quantity, which may be thought of as the concentration of clusters with n molecules, is negligibly small for clusters with more than a few molecules. This is true irrespective of the degree of subsaturation, s. Is this expectation from classical nucleation theories valid? We show here that subsaturated solutions feature a rich distribution of species not anticipated by classical nucleation theories. We report results from measurements of cluster size distributions in subsaturated solutions of phase-separating RNA-binding proteins from the FUS-EWSR1-TAF15 (FET) family. We find that these systems form clusters in subsaturated solutions, and that the cluster sizes follow heavy-tailed distributions. The abundant species are always small clusters. However, as total macromolecular concentration (c_tot) increases, the distributions of cluster sizes shift continuously toward larger values. We discuss these findings in the context of theories for associative polymers (9, 23–30).