Direct Observation of the Conformational Transitions in Tau and Their Correlation with Phase Behavior

Abstract

Liquid–liquid phase separation (LLPS) is now recognized as one of the key mechanisms underlying the formation of membraneless organelles. Typically, condensates formed through LLPS are dynamic and play a crucial role in the spatiotemporal regulation of essential cellular processes. In some cases, however, condensates can undergo an aberrant liquid-to-solid transition, which is now recognized as being related to the onset of cancers and neurodegeneration. The microtubule-associated protein Tau, the aberrant aggregation of which is implicated in neurodegenerative disorders like Alzheimer’s and Parkinson’s, has been found to undergo LLPS. The Tau condensates formed through LLPS are considered to be intermediate on-pathway precursors of amyloid aggregates. Unlike other known phase-separating proteins (e.g., FUS or TDP-43) that have low-complexity domains (LCDs), Tau is intrinsically disordered. Thus, Tau exhibits a highly flexible structure that can be modulated by changes in environmental changes. The intricate relationship between different conformations of full-length Tau and its phase behavior remains poorly understood. To bridge this gap, here, by employing a combination of single-molecule FRET and molecular dynamics simulations, we demonstrate that Tau undergoes conformational transitions from compact to extended states during LLPS, irrespective of diverse driving forces. Moreover, we show that intramolecular interactions responsible for stabilizing the compact conformations of monomeric Tau correlate with the intermolecular interactions driving the LLPS of Tau, thereby facilitating the formation of dynamic networks. These findings provide crucial mechanistic insights into how the conformational state of Tau governs its propensity for phase separation, shedding light on sequence-encoded structural processes that ultimately drive biological phase separation.

Introduction

Liquid–liquid phase separation has recently emerged as a fundamental principle governing the formation of biomolecular condensates and the spatiotemporal coordination of biological activities in cells. Biomolecular condensates, consisting of proteins and nucleic acids, are organized through weak, multivalent interactions rather than lipid membranes, as in conventional intracellular organelles. These condensates coalesce into stable, micron-sized bodies with distinct compositions from the cellular milieu. Due to the absence of a lipid boundary, molecules can freely exchange with their counterparts in the surrounding environment, facilitating reversible assembly and disassembly of reaction machineries within a narrow time window. This dynamic behavior plays a crucial role in the spatiotemporal regulation of essential cellular processes including genome organization, RNA processing, signaling, transcription, and stress response. − On the other hand, accumulating evidence suggests that aberrant LLPS is implicated in various diseases such as cancer and neurodegeneration. , Numerous studies have demonstrated that intrinsically disordered proteins and regions (IDPs/IDRs) containing low-complexity and prion-like domains are prominent candidates for LLPS of biomolecules. − The presence of low-sequence complexity leads to conformational flexibility and heterogeneity, allowing the polypeptide chains to engage in multivalent and dynamic interchain interactions including electrostatic, hydrophobic, hydrogen bonding, dipole–dipole, π–π, and cation−π interactions. −

Tau, a microtubule-associated protein characterized as intrinsically disordered, is involved in the pathogenesis of a wide range of neurodegenerative disorders. , Recent studies have revealed that, similar to other proteins associated with neurodegenerative diseases, Tau exhibits a pronounced propensity to undergo LLPS, which may trigger irreversible aggregation. − The longest isoform of the Tau protein (i.e., 2N4R Tau) can be divided into four regions based on charge clustering: a negatively charged N-terminus region, a positively charged proline-rich region, a positively charged microtubule-binding region, and a slightly negatively charged C-terminal region. It has been proposed that interactions between the oppositely charged regions are crucial for the LLPS of Tau, while hydrophobic interactions become the dominant driving forces when electrostatic interactions are impaired due to domain deletion or charge screening. , This suggests a complex interplay between sequence-encoded interactions and environmental factors. Although previous studies have extensively characterized the LLPS of Tau, a quantitative understanding of how intra- and intermolecular interactions of Tau modulate its conformational dynamics and phase separation propensity under varying physicochemical conditions remains poorly understood.

In this study, we employed a combination of single-molecule FRET techniques and molecular dynamics (MD) simulations to dissect the conformational landscape of full-length 2N4R Tau and its correlation with salt-dependent LLPS. We show that Tau exhibits reentrant phase separation with electrostatic interactions dominating under low-salt conditions and hydrophobic forces prevailing under high-salt regimes. Crucially, smFRET reveals that Tau undergoes sequence-encoded conformational expansion during LLPS, irrespective of the driving forces (electrostatic vs hydrophobic). By resolving residue-level interactions by MD, we further demonstrate a molecular mechanism where intramolecular interactions that stabilize compact monomers template multivalent intermolecular networks in condensates. Our work establishes a paradigm for understanding how IDP conformational plasticity governs phase behavior, which is essential for unraveling the molecular basis of LLPS-driven pathological aggregation.

Results

Salt-Mediated Reentrant Phase Separation of Tau Protein

It has been previously shown that the full-length 2N4R Tau isoform can undergo LLPS under both low-salt and high-salt conditions. − In order to achieve a more accurate analysis of the LLPS of Tau across a broad salt concentration range, we employed a recently developed microfluidics-based technique named Phase Scan, which enables rapid and automated generation of diverse solution conditions for high-throughput detection of protein LLPS behavior. The Phase Scan analysis showed a complete phase diagram of 2N4R Tau (hereafter referred to as Tau) that clearly delineates three distinct regimes: low-salt LLPS, medium-salt homogeneity, and high-salt LLPS, demonstrating a nonmonotonic dependence of Tau LLPS on salt concentration (Figure A), which has also been observed for IDPs such as FUS and PR25. Quantitative analysis of the phase diagram of Tau using the Flory–Huggins theory in combination with the Debye–Hückle-type electrostatic interactions (FH-DH theory; Figure A, black dashed line) further corroborated these findings. Specifically, two-body hydrophobic interactions dominate the free energy landscape under high-salt conditions (Figure S1A), whereas electrostatic contributions dominate under low-salt conditions and gradually decrease with increasing salt concentration due to charge shielding effects (Figure S1B). Additionally, the entropic force resulting from particle mixing does not exhibit a propensity toward LLPS (Figure S1C). These findings align with previous experimental observations that Tau droplets formed under low-salt conditions (<150 mM) can be disrupted by elevated ionic strength but remain resistant to 1,6-hexanediol, while Tau condensates formed under high-salt conditions (>3 M) show the opposite sensitivity. , Although high-salt conditions are remote from the physiological, it is thought that the enhancement of hydrophobic interactions may be relevant to the situation when Tau is hyperacetylated and positive charges on multiple Lys residues are eliminated.

1.

Reentrant phase separation of Tau. (A) Phase diagram illustrating the relationship between Tau LLPS and the concentrations of NaCl and PEG. Tau concentration was kept constant at 10 μM. The scatter plot displays red and blue data points representing individual microdroplets classified as either phase-separated or homogeneous (N = 286,211 droplets). The heat map indicates the probability of LLPS. The black dashed line represents the binodal phase boundary, which is derived from the FH-DH theory through rigorous calculation. (B) Final snapshots of Tau LLPS simulations at low-, medium-, and high-salt concentrations (top panel), along with the density profiles along the z axes shown below (bottom panel). (C) Phase diagram of Tau protein demonstrating the relationship between dense/dilute phase density and either salt concentration in the low-salt regime (left panel) or hydrophobicity scale at high salt concentrations (right panel).

To further investigate the molecular mechanisms for Tau LLPS, we further performed one-bead-per-residue coarse-grained MD simulations in combination with a slab-like simulation box, which has shown its effectiveness in elucidating the LLPS mechanisms of IDPs. − At both low and high salt concentrations, a densely concentrated protein condensate was observed in the center of the simulation box (Figure B).

In contrast, under medium salt concentration conditions, the distribution of Tau molecules throughout the simulation box appears uniform, suggesting that the protein is unlikely to undergo LLPS. Furthermore, we conducted a series of LLPS simulations at varying salt concentrations ranging from 0.1 to 1.0 M to construct the phase diagram of Tau LLPS in the low-salt regime. Our findings revealed that as salt concentration increased, there was a noticeable decrease in the density of the condensed phase and an increase in the density of the dilute phase, indicating a reduced capacity for LLPS (Figure C). The upper boundary of salt concentration that enables LLPS was determined to be 0.88 ± 0.03 M by fitting to the critical equation. To assess the impact of high salt concentrations on Tau LLPS, we kept the Debye screening length constant (λ_D = 0.25 nm, corresponding to a high salt concentration of 1.6 M) while gradually increasing the hydrophobicity scale from 1.0 to 1.1. With an increase in the hydrophobicity scale, we observed an augmentation in the condensed phase density and a reduction in the dilute phase density (Figure C, right). The impact of salt concentration on inter-residue interactions across multiple amino acid pairs was investigated in a recent study utilizing all-atom umbrella sampling, revealing an enhancement in these interactions with increasing salt concentration. Based on all-atom simulation results, the hydrophobic contribution in the HPS model was increased by 10–30%, which can effectively model the high-salt regime (>3 M NaCl). This corresponds to a scaling factor (λ_H) of 1.1–1.3. Thus, λ_H = 1.1 in our simulation corresponds to an effective salt concentration of ∼3 M. The observation of reentrant phase separation at λ_H = 1.1 further suggests the crucial role of hydrophobic interactions in driving the reentrant LLPS of Tau in the high-salt regime. The simulation results are in good agreement with the experimentally observed salt-dependent reentrant LLPS behavior of Tau (Figure A).

Conformational Dstributions of Tau Protein under Different Conditions

Next, we aimed to elucidate the conformational states of Tau and their correlations with the reentrant phase behavior. SmFRET is a powerful tool for investigating the conformations and dynamics of IDPs, which exhibit a high degree of flexibility, and can provide information about conformational heterogeneity as well as intermolecular interactions at extremely low protein concentrations. , Here, we applied smFRET to probe the interdomain conformations of Tau under different salt conditions in both LLPS and non-LLPS states. To achieve this, we constructed four dual-cysteine variants of Tau (Tau_17/244, Tau_149/244, Tau_244/354, and Tau_354/433) by introducing pairs of cysteine mutations at the indicated sites while substituting the two intrinsic cysteines Cys291 and Cys322 into serines as in previous FRET and electron paramagnetic resonance (EPR) studies. − These labeling sites were strategically chosen to reflect the conformations between different domains of Tau when labeled with Alexa Fluor 488 (AF488) and Alexa Fluor 594 (AF594) as the FRET pair. Such labeling was found to have no influence on the structure, function, and the LLPS behavior of Tau. ,, We then performed confocal smFRET experiments to detect the interdomain conformations of these labeled Tau variants under various salt conditions in the absence and presence of 15% PEG. Fluorescence donor and acceptor bursts, originating from individual labeled Tau molecules, were detected and analyzed to generate FRET efficiency histograms. To investigate the conformational distributions of free monomeric Tau under native conditions (non-LLPS), we diluted the dual-labeled Tau variants to a concentration around 100 pM, in buffer containing NaCl at 10 mM, 1 M, or 3 M. To probe the conformations of Tau undergoing LLPS in the presence of 15% PEG, we supplemented AF488/AF594-labeled Tau with unlabeled Tau at a final concentration of 10 μM, which was equivalent to that used in bulk turbidity assays and served as a dopant to eliminate any potential intermolecular FRET between Tau molecules. In the presence of 15% PEG, Tau droplet formation was observed at 10 mM and 3 M NaCl, whereas the solution remained homogeneous at 1 M NaCl (Figure S2). In our previous study, we assessed the impact of increased viscosity on FRET efficiency under crowded conditions by measuring the relative change in Förster radius (R₀) in the presence of 15% PEG, and observed no significant alteration in R₀. Consequently, we infer that the changes in smFRET efficiency primarily arise from the changes in the distance between the labeling sites. The smFRET histograms obtained under 10 mM NaCl showed that Tau_17/244 and Tau_354/433 exhibited peaks with higher FRET efficiency (E centered at 0.67 and 0.58, respectively) compared to those of Tau_149/244 and Tau_244/354 (E centered at 0.40 and 0.30, respectively) (Figure B, top panels, gray data, and Table S1), which is consistent with the proposed “paperclip” conformational ensemble of native Tau in solution. , Upon addition of 15% PEG, LLPS occurs (Figure S2A), resulting in a noticeable shift toward lower FRET efficiencies in the distribution of all variants when compared to their native states. This observation indicates an increase in interdomain distances, which aligns with our previous investigation. Upon increasing the salt concentration to 1 M, the FRET efficiency distributions of free monomeric Tau_17/244, Tau_149/244, and Tau_354/433 were observed to shift toward lower FRET efficiencies compared to that of native Tau at 10 mM NaCl (Figure B, middle panels, gray data, and Table S1). This observation suggests that with increasing salt concentration, there was an expansion in both the N-terminal and C-terminal regions in the native state while minimal changes occurred in the MTBR; therefore, it can be inferred that the ″paperclip″ conformation of Tau was likely disrupted. In the presence of 15% PEG where Tau solution remained in a non-LLPS state (Figure S2B), the FRET efficiency of each Tau variant slightly increased (Figure B, middle panel, pink data). This suggests a subtle conformational compaction of Tau without LLPS in the crowded environment, which is in dramatic contrast to the observations at a low salt concentration. Upon further increasing the salt concentration to 3 M NaCl, the smFRET distributions of native Tau_244/354 and Tau_354/433 shifted toward higher values (E centered at 0.60 and 0.67, respectively), indicating compaction of both the MTBR and C-terminal domain (Figure B, bottom panels, gray data, and Table S1). Interestingly, when reentrant LLPS of Tau was induced at high salt concentration (Figure S2C), there was a transition in the conformation of Tau to an extended state again with a decrease in FRET efficiency, similar to that observed at low salt concentrations (bottom panel, orange data).

2.

smFRET histograms of Tau under different conditions. (A) Domain composition of full-length Tau, including the N-terminal domain, the proline-rich region, the MTBR (including the four repeats R1–R4), and the C-terminal domain. (B) smFRET efficiency histograms of AF488/AF594 dual-labeled Tau variants, including Tau_17/244, Tau_149/244, Tau_244/354, and Tau_354/433 under low salt concentration (10 mM NaCl; top panel), medium salt concentration (1 M NaCl; middle panel), and high salt concentration (3 M NaCl; bottom panel). The peak centered around zero FRET efficiency (“zero peak”, shaded in light gray) is typically attributed to molecules in an acceptor-inactive state or carrying only the donor fluorophore and, thus, are not considered. SmFRET experiments were performed in the absence (gray) or presence of 15% PEG (data shown in blue, pink, and yellow), which is necessary for inducing LLPS of Tau.

The Changes in the Radius of Gyration of Tau upon LLPS

In order to analyze the dimensions of Tau during LLPS and their dependence on salt concentration, we performed MD simulations on Tau monomer at low, medium, and high salt concentrations and compared their radius of gyration (R _g) with that observed in phase-coexistence simulations. In the LLPS state, Tau molecules exhibit more extended conformations compared to free monomers regardless of low or high salt concentrations, as indicated by increased R _g values (Figure A,C,D,F). In contrast, the R _g values are nearly unchanged at medium salt concentrations, where no LLPS occurs (Figure B,E). Collectively, the combined findings from smFRET experiments and MD simulations consistently demonstrate a conformational transition of Tau from a compact to an extended conformational ensemble during LLPS, despite different driving forces governing LLPS at low- and high-salt conditions. The extended conformation may enhance the propensity for intermolecular multivalent interactions, thereby facilitating LLPS through increased intermolecular contacts. We note that our simulations predict a smaller R _g value for Tau monomers (4–5 nm) compared to experimentally determined values (∼6 nm). , This discrepancy arises because current force fields tend to produce overly compact conformations for IDPs. − Nevertheless, studies have shown that simulations reliably capture relative trends in protein size and compactness, even when absolute R _g values deviate from experimental measurements. , Therefore, our analysis is focused on qualitative trends rather than on absolute values.

3.

Gyration radius of Tau in monomer simulations and phase-coexistence simulations. The probability density function (PDF) of R _g of Tau molecule in monomer simulations (A–C) at low, medium, and high salt concentrations and that in phase-coexistence simulations (D–F) at low, medium, and high salt concentrations.

Intrachain and Interchain Contacts of Tau Molecules in Monomeric and LLPS States

To assess the correlation between the conformational changes in Tau and the intra- and intermolecular interaction networks in both non-LLPS and LLPS states, we performed MD simulations to analyze the residue-specific contacts within and between chains. The contribution of each residue to the intra- and interchain contacts shows little variation across the residue index (Figure S3). However, clear patterns emerge when examining the 2D contact maps for interactions between pairs of residues (Figure ). Under low salt concentration, robust intrachain interactions between residues 40–120 and residues 120–400 were observed in monomer simulations (Figure A). Interestingly, these intrachain interactions show patterns similar to the prominent interchain interactions in LLPS simulations, where the intrachain interactions are significantly attenuated (Figure B,C). Residues 40–120 and residues 120–400 are notably enriched with negatively and positively charged residues. The interchain contact numbers between these oppositely charged residues are much higher than intrachain contact numbers in LLPS simulation, suggesting a crucial role of electrostatic attractions in driving LLPS (Figure S4). At a medium salt concentration, both intrachain and interchain interactions are greatly attenuated, thereby impeding the occurrence of LLPS (Figure D–F). In contrast to the significantly weakened intrachain interactions observed under medium-salt conditions, a resurgence of intramolecular interactions is evident under the high-salt conditions in the monomer simulations, although the strength of these intrachain interactions is weaker compared to low-salt conditions (Figure G). Notably, in LLPS simulations, the interchain electrostatic attractions between residues 40–120 and 120–400 at medium- and high-salt conditions are both smaller than those at low-salt conditions (Figure S4). Instead, robust interchain interactions were observed among residues 60–120 and residues 250–430 at high salt (Figure I). These regions are predicted to possess pronounced hydrophobicity (Figure S5). The above results suggest that under low- and high-salt conditions Tau preferentially adopts more compact conformational states stabilized by intramolecular interactions. Under LLPS conditions, these intramolecular interactions are disrupted and transition into intermolecular interactions. The resulting conformational expansion maximizes the intermolecular connectivity networks and promotes LLPS, thereby reducing the overall free energy of the whole system. In contrast, without a foundation of intramolecular interactions such as Tau protein under medium-salt conditions, networks of interactions cannot be formed, which is thus unfavorable for LLPS. The simulation results are in good agreement with the smFRET data regarding the conformational changes of Tau in response to salt concentrations as well as upon LLPS, thus providing a mechanistic view of the LLPS behavior of Tau under multiple conditions.

4.

Intrachain and interchain contacts of Tau molecules under monomeric and LLPS states revealed by MD simulations. The heat maps of the intrachain contacts in monomer simulations (A, D G), and both intrachain and interchain contacts in LLPS simulations under conditions of low (B, C), medium (E, F), and high salt concentrations (H, I) are shown. The color in the heat map represents the number of contacts between each residue pair. For intrachain contacts, a deeper red color indicates higher number of contacts. For interchain contacts, red corresponds to a higher number of contacts, while blue indicates fewer contacts.

Discussion

In recent years, LLPS has provided a new perspective to explain the formation of membraneless organelles which are thought to be associated with a wide range of cellular functions and dysfunctions. ,, Akin to other intrinsically disordered proteins, the microtubule-associated protein Tau undergoes LLPS both in vivo and in vitro. Tau LLPS can be modulated by various factors, including salt concentration, temperature, multivalent ions, and post-translational modifications. ,,,, Although extensive studies have been carried out on the physicochemical properties and potential biological functions of condensates, the structural changes of proteins within liquid droplets, which are crucial for understanding the molecular mechanism of LLPS, remain largely unexplored. The application of single-molecule fluorescence techniques allows the exploration of the conformational changes and dynamics of IDPs. , In this study, by combining smFRET with MD simulations, we demonstrated that Tau transitions from compact, intramolecularly stabilized states to extended conformations during LLPS, regardless of distinct driving forces for LLPS under low- and high-salt conditions. In line with our observations, several other IDPs or IDRs have also been proposed to favor expanded conformations during LLPS compared to the free monomeric state. ,− In a recent study, smFRET was employed to examine the conformational change occurring in the LCD of FUS during LLPS. It was observed that the population of lower FRET efficiency conformers was increased compared to the non-LLPS native state, suggesting partial extension of FUS-LCD within the condensed phase. Conformational rearrangement has been observed not only in homotypic LLPS of IDPs but also in heterotypic LLPS of IDPs with their interacting partners. For instance, Histone H1 and its nuclear chaperone, prothymosin-α (ProTα), which exhibit picomolar binding affinity, can undergo cophase separation. SmFRET experiments revealed that in the dense phase, ProTα becomes more expanded compared to the dilute phase. In addition to single-molecule techniques, researchers have applied cross-linking/mass spectrometry (XL-MS) to demonstrate a conformational transition of the α-synuclein protein from a “hairpin-like” structure toward more elongated conformational states upon LLPS. These studies, in conjunction with our findings, suggest that conformational expansion is likely a ubiquitous characteristic of IDPs in their phase-separated state, facilitating the intermolecular multivalent interactions necessary for LLPS.

One of the most significant findings of our study is the direct observation of the correlation between conformational compaction of the Tau monomer and its LLPS propensity, as revealed by combined smFRET and MD simulations (Figures and ). This conformational transition during LLPS is accompanied by the conversion of intramolecular interactions into intermolecular interactions (Figure ). Under low-salt conditions, the electrostatic interactions between the negatively charged N-terminal/C-terminal domains and the positively charged MTBR stabilize the “paperclip” conformation of native Tau; while during LLPS, the electrostatic interactions between these oppositely charged regions promote the formation of intermolecular networks (Figure , top panel). Under high-salt conditions, the salting-out effect induces protein dehydration, thereby promoting the hydrophobic interactions of proteins. These interactions govern the conformational collapse of the MTBR and C-terminal domain of Tau in its monomeric state as well as intermolecular dynamic networks in the LLPS state (Figure , bottom). Conversely, under medium-salt conditions where intramolecular interactions are disrupted due to the charge screening effect, Tau protein exhibits expanded conformations in its monomeric state and fails to establish productive intermolecular interactions even under LLPS-induced conditions (e.g., addition of crowding reagent) (Figure , middle). Our results suggest that the intramolecular interactions stabilizing single-chain compaction are equivalent to and can potentially convert to the intermolecular interactions that lead to self-association under certain conditions. Consistent with our observations, several computational studies have indicated a correlation between the conformational compaction of an IDP in dilute phase and its propensity for LLPS. Specifically, IDPs with more collapsed conformations tend to exhibit a higher likelihood of undergoing LLPS, while those with less compacted structures are less prone to such phase transitions. ,− However, those computational studies were based on a comparison between multiple kinds of proteins and peptides with different sequences. In this study, we experimentally observed the conversion of compact Tau monomers stabilized by intramolecular interactions to extended Tau molecules within the condensates driven by intermolecular interactions, thereby elucidating the correlation between the conformations and phase behavior of a single protein under varying conditions.

5.

Schematic representation of the relationship between the conformations of Tau and its LLPS capability. Under low-salt conditions, the monomeric Tau adopts a “paperclip” conformation that is mainly stabilized by intramolecular electrostatic interactions and shifts toward an extended conformation in the LLPS state driven by intermolecular electrostatic interactions. Under medium-salt conditions, moderate charge screening effects disrupt the “paperclip” conformation of monomeric Tau. This reduced degree of interaction results in only slight conformational compaction while inhibiting LLPS. Under high-salt conditions, the MTBR and C-terminal regions of monomeric Tau become compact due to enhancement of hydrophobic interactions caused by the salting-out effect, while they convert to an extended state again upon LLPS driven by intermolecular hydrophobic interactions.

Conclusions

In this study, we observed the salt-dependent LLPS behavior of Tau protein, with electrostatic and hydrophobic interactions as dominant driving forces under low- and high-salt conditions, respectively. By smFRET experiments and MD simulations, we characterized the conformations of Tau under different salt conditions and their changes upon LLPS. At low and high salt concentrations conducive to LLPS of Tau, native Tau exhibits different levels of conformational compactness, showing intramolecular interactions between different regions. Upon the occurrence of LLPS under these conditions, Tau protein undergoes a conformational transition to an extended state, irrespective of distinct driving forces for LLPS. This extension enhances intermolecular interactions that exhibit patterns similar to those of the intramolecular interactions within free monomers and facilitates the formation of dynamic networks crucial for liquid condensation. In contrast, at intermediate salt concentrations, the charge screening effects prevent significant intramolecular interactions within monomeric Tau, resulting in a lack of intermolecular contacts that promote LLPS under crowded conditions. These findings, derived from direct experimental observations and computational simulations, provide valuable insights into the correlation between conformational properties of individual chains of IDPs and their capacity for LLPS, bridging molecular-level dynamics with mesoscale phase behavior.

Methods

Protein Expression and Purification

The human full-length Tau gene (i.e., the longest isoform 2N4R Tau) was generously provided by Prof. Jianzhi Wang and Rong Liu (Huazhong University of Science and Technology, China) and subsequently subcloned into the pET-28a vector. The four dual-Cys variants of Tau (T17C/Q244C, T149C/Q244C, Q244C/I354C, and I354C/S433C, referred to as Tau_17/244, Tau_149/244, Tau_244/354, and Tau_354/433 in this study) used for smFRET study were constructed previously. The expression and purification of Tau protein and its variants were performed as previously described. , Briefly, after cell culture and induction, the cells were subjected to boiling, sonication, and centrifugation. The supernatant containing Tau protein was precipitated with ammonium sulfate and subsequently purified by using ion exchange chromatography. The fractions containing Tau protein were combined and further purified by Superdex 200 column (GE Healthcare) using 30 mM Tris-HCl (pH 7.5) including 150 mM NaCl and 2 mM DTT as the elution buffer. Protein concentration was determined by measuring the absorbance at 280 nm using a Nanodrop 2000 spectrometer with an extinction coefficient of 7450 M^–1 cm^–1. Aliquots of the protein were flash-frozen and stored at −80 °C.

Fluorescence Labeling of Tau Variants

The dual-cysteine variants Tau_17/244, Tau_149/244, Tau_244/354, and Tau_354/433 were labeled with maleimide-functionalized Alexa Fluor 488 (AF488) and Alexa Fluor 594 (AF594) following previously established protocols. , Subsequently, the labeled proteins were purified using a PD-10 desalting column (GE Healthcare). The protein concentration and labeling efficiency were determined according to the manufacturer’s instructions. The dye-to-protein labeling ratio was calculated to be approximately 90%. The aliquots of the labeled proteins were flash-frozen and stored at −80 °C.

Phase Scan Experiment

Microfluidic devices were designed using AutoCAD software and subsequently fabricated using soft-photolithographic methods by applying SU-8 on Si-wafer masters and poly­(dimethylsiloxane) (PDMS)-on-glass devices. Briefly, the SU-8 3050 photoresist (A-Gas Electronic Materials Limited) was poured on a polished silicon wafer (MicroChemicals GmbH) and spun down for 45 s at 3000 rpm by using a spin coater. Subsequently, the SU-8-coated wafer was soft-baked on a level hot plate at 95 °C for 15 min. After the soft bake step, the SU-8-coated wafer was cooled to room temperature, and then the acetate sheet mask with the device design was applied. Mask-SU-8-coated wafer sandwich was exposed to UV light for 40 s. Directly after the exposure, the mask was removed and the SU-8-coated wafer was postexposure baked (PEB) on a level hot plate at 95 °C for 5 min. After PEB, the developing step took place by submerging the SU-8-coated wafer into propylene glycol monomethyl ether acetate (PGMEA; Sigma-Aldrich) solution and incubating it for 2 min with periodical agitation. Finally, the wafer was rinsed with isopropyl alcohol and dried under an airflow. The master wafer for fabricating microfluidic devices with a channel height of 50 μm was obtained.

The microfluidic devices were fabricated by casting PDMS (Sylgard 184 kit; Dow Corning) on a master wafer, curing it at 65 °C for 90 min, peeling it off, punching the holes for inlets and outlets, and bonding it to a 1 mm thick glass slide (Epredia) after oxygen plasma activation in a plasma oven (Diener Femto, 60% power for 30 s). Subsequently, the hydrophobic treatment of the channels of the microfluidic device was performed. The channels were filled with 1% v/v trichloro­(1H,1H,2H,2H-perfluorooctyl)­silane (Sigma-Aldrich) in HFE-7500 fluorinated oil (3M Novec Engineered fluid) solution and incubated for 30 min on a hot plate at 95 °C. After the incubation, channels were washed with HFE-7500 fluorinated oil and dried under airflow.

Phase diagrams were constructed using the semiautomated microfluidic platform “PhaseScan”. Briefly, PEG, NaCl, and buffer stock solutions were mixed at various flow rates ranging from 5 to 50 μL/h, while Tau solution was kept constant at 20 μL/h flow rate and encapsulated in water-in-oil droplets, individual microenvironments, using a microfluidic device. Fluorescence images of droplets in the observation chamber of the microfluidic device were taken using an epifluorescence microscope (Cairn Research) equipped with 10× objective (Nikon CFI Plan Fluor 10×, NA 0.3), and analyzed via an automated image analysis script to detect Tau condensates and to determine PEG and NaCl concentrations in individual droplets. , The data was plotted as a color-coded scatter plot, where each point represents individual microfluidic droplets and color represents the phase separation probability (1 or dark red, phase-separated; 0 or dark blue, mixed) assigned to the droplet. This probability was calculated by averaging the phase state assignment of neighboring droplets. The radius of the neighborhood was defined as a percentage of data range (5%).

Flory–Huggins–Debye–Hückel (FH–DH) Theory

To model the LLPS of Tau protein in high salt solutions, here we adopted the Flory–Huggins–Debye–Hückel theory (FH–DH). A key idea of the FH-DH theory is to map the protein system onto a regular lattice, which enables the calculation of mixing entropy in an explicit way. According to the FH–DH theory, the total free energy F ^tot({ϕ _i }) is constituted by three different contributionsthe mixing entropy, the two-body non-Coulombic interactions, and the electrostatic interactions,

$$\{\begin{array}{l}{F}^{\mathrm{t}\mathrm{o}\mathrm{t}}(\{{\varphi }_i\})=F^{\mathrm{m}\mathrm{i}\mathrm{x}}+F^{\mathrm{int}}+F^{\mathrm{e}\mathrm{l}\mathrm{e}}\\ \frac{F^{\mathrm{m}\mathrm{i}\mathrm{x}}}{V{k}_{\mathrm{B}}T}=\sum _i\frac{{\varphi }_i}{N_i}\ln {\varphi }_i\\ \frac{F^{\mathrm{int}}}{V{k}_{\mathrm{B}}T}=\sum _{i<j}{\chi }_{ij}{\varphi }_i{\varphi }_j\\ \frac{F^{\mathrm{e}\mathrm{l}\mathrm{e}}}{V{k}_{\mathrm{B}}T}=-\frac{l_0^3}{4\pi l^3}[\ln (1+kl)-kl+\frac{{(kl)}^2}{2}]\end{array}$$

where the index i refers to one of the five species (Tau, PEG, Na⁺, Cl^–, H₂O) considered in the current study, V is the system volume, k _B is the Boltzmann constant, and T = 310 K is the default temperature.

The entropic contribution for species mixing is calculated following the classical method of Flory and Huggins, where ϕ _i = c _i N _i /c _w is the volume fraction and N _i is the number of lattice sites occupied by species i. For water and ions, we set N _w = N _Na = N _Cl = 1, while for Tau and PEG, we have N _tau = 441 and N _PEG = 505, which are calculated based on the sequence length of Tau and the molecular weight of PEG 10,000, respectively. c _i is the molar concentration of species i, with c _w = 55.56 mol/L for water.

The Flory-Higgins parameter χ _ij characterizes the nonlinear interactions between species i and j. In order to capture the dramatic concentration variation in salt, a nonlinear formula is adopted. Referring to the data in Figure A, we set χ_tau,H2O = 0.4 and χ_tau,Na = χ_tau,Cl = −1.25 ln­(ϕ_tau) (ϕ_Na–A)², where A = 12.5 M is a fitting parameter.

The Coulomb-type electrostatic interactions among charged species are calculated according to the Debye–Hückel theory. σ _i is the charge density of species averaged over each lattice site. Under weakly alkaline conditions (pH 7.5), the net charge of Tau is estimated based on its primary sequence using pepcalc.com, which gives σ_tau = 3.1/441, while σ_H2O = σ_PEG = 0 and σ_Na = σ_Cl = 1. l ₀ ≈ 0.31 nm is the approximate size of a water molecule defined as the cube root of its molecular volume, and l ≈ 0.26 nm is the approximate size of salt ions. k denotes the inverse screening length, which is given by

$$k=\frac{1}{l_0}\sqrt{\frac{4\pi l_{\mathrm{B}}}{l_0}\left(\sum _i{\sigma }_i{\varphi }_i\right)}$$

where l _B ≈ 0.67 nm is the Bjerrum length. It is noted that in the limit of small k, the formula for the electrostatic interaction recovers the classical Voorn–Overbeek theory

$$\frac{F^{\mathrm{ele}}}{V{k}_{\mathrm{B}}T}=-\frac{{(k{l}_0)}^3}{12\pi }$$

by using Taylor expansion, which however is only applied to low-salt conditions.

Calculation of Binodal Curve and Fit to LLPS

Liquid–liquid phase separation occurs when the two coexisting phases have a lower free energy than a homogeneous mixture, that is

$$\frac{\upsilon F^{\mathrm{tot}}(\{{\varphi }_i^{\mathrm{s}}\})+(1-\upsilon )F^{\mathrm{tot}}(\{{\varphi }_i^{\mathrm{d}}\})-F^{\mathrm{tot}}(\{{\varphi }_i^0\})}{F^{\mathrm{tot}}(\{{\varphi }_i^0\})}\le 0$$

where the superscripts (s) and (d) refer to the coexisting supernatant phase and dense phase, while the superscript (0) stands for the single phase. υ ∈ (0,1) denotes the volume fraction of the supernatant phase. The boundary for LLPS, known as the binodal curve (or coexistence curve) in the literature, is given by a combination of {{ϕ _i },{ϕ _i },υ} when the above equality holds.

The calculation of the binodal curve is performed by using the simulated annealing algorithm with the constraints that the volume fractions of all species are in the range of [0,1] and the sum of volume fractions of all species in a phase is unity. As a consequence, the volume fractions of all species in the dense phase can be uniquely determined once those in the supernatant phase are given.

$$\{\begin{array}{l}{\varphi }_{\mathrm{tau}}^{\mathrm{d}}=\frac{{{\varphi }_{\mathrm{tau}}^0-\upsilon \varphi }_{\mathrm{tau}}^{\mathrm{s}}}{1-\upsilon }\\ {\varphi }_{\mathrm{PEG}}^{\mathrm{d}}=\frac{{{\varphi }_{\mathrm{PEG}}^0-\upsilon \varphi }_{\mathrm{PEG}}^{\mathrm{s}}}{1-\upsilon }\\ {\varphi }_{\mathrm{Na}}^{\mathrm{d}}=\frac{{{\varphi }_{\mathrm{Na}}^0-\upsilon \varphi }_{\mathrm{Na}}^{\mathrm{s}}}{1-\upsilon }\\ {\varphi }_{\mathrm{Cl}}^{\mathrm{d}}=\frac{{{\varphi }_{\mathrm{Cl}}^0-\upsilon \varphi }_{\mathrm{Cl}}^{\mathrm{s}}}{1-\upsilon }\\ {\varphi }_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{d}}=1-{\varphi }_{\mathrm{tau}}^{\mathrm{d}}-{\varphi }_{\mathrm{PEG}}^{\mathrm{d}}-{\varphi }_{\mathrm{Na}}^{\mathrm{d}}-{\varphi }_{\mathrm{Cl}}^{\mathrm{d}}\end{array}$$

Single-Molecule FRET Experiments and Data Analysis

The single-molecule FRET experiments were performed on a home-built apparatus based on an inverted fluorescence microscope (Ti-U, Nikon). The beam of a 488 nm laser (Coherrent) was directed to the back port of the microscope for excitation with a power of 50 μW and was focused by a 100× oil-immersed objective (N.A. = 1.4). The optical configuration was the same as that described in detail previously. ,

The AF488/AF594 dual-labeled samples were diluted with 30 mM Tris-HCl buffer (pH 7.5) containing 10 mM, 1 M, or 3 M NaCl to give a final protein concentration of 100 pM in the absence or presence of 15% PEG. For the LLPS sample, 10 μM unlabeled Tau was premixed with AF488/AF594-labeled Tau as dopant in order to suppress intermolecular FRET. The samples were loaded into an 8-well sample chamber (Nunc) for smFRET measurements. To prevent potential adsorption of labeled proteins to the surface of the chamber during the measurements, the chamber was subjected to plasma cleaning (Harrick Plasma) for 2 min to render the glass surface hydrophilic and incubated with 2.5% polylysine at 4 °C overnight before smFRET measurements. For each sample, at least three repetitions were performed to check the reproducibility. The bin time for the single-molecule signal was set at 1 ms. The fluorescence bursts with the sum of donor and acceptor signals higher than 30 photons/ms were selected as effective single-molecule events. The FRET efficiencies of individual, freely diffusing molecules were calculated as E = (I _A – r _DA × I _D)/(I _A – r _DA × I _D + γI _D), where I _A and I _D are the intensities of the acceptor and donor fluorophores, respectively; γ corresponds to the correction factor that accounts for the different quantum yields and detection efficiencies of the two detection channels, which was measured to be 1.26; and r _DA represents the proportions of fluorescence leakage from the donor channel to acceptor channel (r _DA = 0.058 under our instrument setup). The smFRET histograms for the native state were fitted to a double Gaussian function to obtain the peak position (Table S1). The histograms for the LLPS state were fitted to a Gaussian function plus a γ function, the latter of which is used to describe the “zero peak” caused by the inactive acceptor.

Coarse-Grained Model and Salt-Dependent Force Field of the Full-Length Tau Protein

The full-length Tau is modeled as an elastic chain of coarse-grained beads, and each bead corresponds to a single residue. Sequential beads are connected by a harmonic bond with an equilibrium bond length of 0.38 nm and a spring constant of 1000 kJ/mol/nm. The nonbonded interactions in the model incorporate a Lennard-Jones term, utilizing the Hydropathy Scale (HPS) potential originally developed by the Mittal group. This potential captures the sequence-dependent phase behavior of proteins that undergo LLPS under a salt condition of ∼100 mM NaCl,

$${\mathrm{\Phi }}_{\mathrm{nb}}(r)=\{\begin{array}{ll}4\in [{\left(\frac{\sigma }{r}\right)}^{12}-{\left(\frac{\sigma }{r}\right)}^6]+(1-\lambda )\in & r\le 2^{1/6}\sigma \\ 4\lambda \in [{\left(\frac{\sigma }{r}\right)}^{12}-{\left(\frac{\sigma }{r}\right)}^6]& \mathrm{otherwise}\end{array}$$

The long-range electrostatic interactions are accounted for using a Coulomb term with Debye screening, with the relative permittivity denoted as D.

$$E_{\mathit{ij}}(r)=\frac{q_i{q}_j}{4\pi \mathit{Dr}}\exp (-\frac{r}{{\lambda }_{\mathrm{D}}})$$

The Debye screening length λ _D was set according to the Debye–Hückel formula, where I is the ion strength, q _i is the charge of the backbone bead, ε₀ denotes the vacuum dielectric permittivity, and ε_r is the relative dielectric constant of water which was set as 80.

$${\lambda }_{\mathrm{D}}=\sqrt{\frac{{\epsilon }_{\mathrm{r}}{\epsilon }_0{k}_{\mathrm{B}}T}{2e^{2}I}}$$

The phase separation of Tau at low and medium salt concentrations was simulated at 10 mM and 1 M NaCl, respectively, consistent with the corresponding experimental salt conditions. To simulate Tau phase separation at high salt concentration, we referred to a recent study demonstrating that reproducing experimental observations of protein LLPS under high-salt conditions requires modification of the Lennard-Jones (LJ) parameters in the HPS force field. Specifically, all-atom umbrella sampling simulations were employed to investigate the interactions between specific amino acid pairs in varying NaCl ion concentrations. Based on the resulting potential of mean force (PMF), the reentrant phase behavior observed in FUS and PR₂₅ proteins at high salt concentrations was successfully observed using a modified version of the HPS force field. This was achieved by enhancing the strength of hydrophobic interactions, specifically increasing it by 10% for FUS and 30% for PR₂₅. Akin to this approach, we explored the LLPS of the Tau protein under high-salt conditions by setting the ion strength (I) to 1.6 M and introduced a scaling factor (λ_H) to modify the HPS potential.

$${{\mathrm{\Phi }}_{\mathrm{nb}}}^{\prime (r)}={\lambda }_{\mathrm{H}}·{\mathrm{\Phi }}_{\mathrm{nb}}(r)$$

Our results demonstrated that even a modest increase of 4% in hydrophobic attraction was adequate to induce a reentrant LLPS of Tau protein. Furthermore, increasing the hydrophobic contribution by 10% resulted in a dense phase with a density comparable to that formed at a low salt concentration of 100 mM. The phase separation of Tau at a high salt concentration was then simulated at 1.6 M with a scaling factor of 20%.

Coarse-Grained Langevin Dynamics Simulations

To assess the LLPS capability of the Tau protein at various salt concentrations, we conducted direct coexistence simulations. The initial configurations were generated by placing 80 chains of Tau into a 44 × 44 × 56 nm³ cuboid box. The centroids of the chains were positioned at the grid points of a 4 × 4 × 5 evenly distributed grid. The configuration of each chain was generated via random walk, with the maximum distance from the centroid restricted to within 4 nm, ensuring that each pair of chains was at least 4 nm apart. In accordance with previous studies, ,, the system was first relaxed at a constant temperature of 150 K and a constant pressure of 1 bar using a Berendsen barostat under NPT ensemble in order to obtain a condensed state. The time step was set to 10 fs which was a standard setup for the original HPS force field. The simulation box after the NPT simulation was about 17 × 17 × 22 nm³ in size. Subsequently, the box was expanded in the z-direction by a factor of 10, creating a slab-like box. The temperature was then gradually increased to 310 K at a rate of 0.2 K/min in the NVT ensemble. The production run was conducted for 5.0 μs using a Langevin thermostat with a relaxation time of 5 ps. The time step for the production run was set to 10 fs. A friction coefficient of 1 ps^–1 was applied for all simulations. All nonbonded interactions were calculated using a cutoff distance of 1.5 nm. All simulations were carried out utilizing the HOOMD-Blue 2.9 package on GPU cards. Previous studies have shown that the x- and y-dimensions of the slab simulation box must be large enough to minimize interactions between most of the chains and their periodic images. In our simulation, the x- and y-dimensions of the simulation box measured ∼17 nm. Under LLPS conditions, the simulated Tau chain exhibited an average R _g of ∼5.7 nm, corresponding to a molecular diameter of $\sqrt{5/3}$ × R _g × 2 = 14.8 nm. Thus, the minimum distance between each Tau molecule and its periodic image is 17–14.8 = 2.2 nm, which is much larger than the cutoff distance (1.5 nm). This indicates that there are no direct contacts between Tau molecules and their periodic images. Therefore, we deduce that our simulation box is sufficiently sized. To eliminate potential artifacts arising from the selection of box size, we conducted an additional simulation on a Tau system that doubles the box size consisting of 160 chains. A similar density of the dense phase (∼420 mg/mL) was observed (Figure S6), which closely aligns with that generated by simulation on the 80-chain system (∼450 mg/mL) (Figure C). Given that simulations involving 160 Tau chains are considerably more time-consuming, we chose to proceed with a smaller system size of 80 Tau chains for further calculation and analysis. A similar system size has been used in a number of previous studies on the LLPS of other proteins. ,,

Trajectory Analysis Methods

The density profile of each simulation was determined using the trajectory of the last 1.0 μs. A translation was applied to all molecules in each frame in order to position the dense phase at the center of the simulation box. Subsequently, the z-axis was divided into 50 equal slices, and the protein mass density of each slice was computed. The density of the dense phase was defined as the average density of the plateau region in the center of the density profile. The density of the dilute phase was obtained by averaging the densities of the outermost 50 nm regions at both ends of the simulation box. The upper/lower critical solution temperatures were determined by fitting to the critical equation ${\rho }_{\mathrm{dense}}-{\rho }_{\mathrm{dilute}}=A{|T_{\mathrm{c}}-T|}^{\beta }$ , where β is the critical exponent that is set to 0.365 (universality class of 3D-Heisenberg model). The contact probability between each pair of residues was defined as the proportion of frames in which the corresponding residues were within 1.2 nm. To calculate the charge and Hopp-Woods hydrophobicity plot of the Tau monomer, we utilized the ExPASy server. All other analyses, including density profile, radius of gyration (R _g), and contact number calculations, were performed using our in-house developed codes.

Glossary

Abbreviations

LLPS

liquid–liquid phase separation

IDRs

intrinsically disordered regions

AD

Alzheimer’s disease

MAPT

microtubule-associated protein Tau

MTBR

microtubule-binding region

FH-VO

Flory–Huggins–Voorn–Overbeek

smFRET

single-molecule fluorescence resonance energy transfer

MD

molecular dynamics

HPS

hydropathy scale

The smFRET data in this article are available at Science Data Bank at 10.57760/sciencedb.18403. Other data supporting this article have been included as part of the Supporting Information.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/jacsau.5c00625.

• Method for detecting LLPS droplets of Tau, supplementary results of Tau LLPS simulations, analysis of LLPS driving forces, fluorescence imaging, intra- and interchain contacts, charge/hydrophobicity profiles of the protein sequence, and fitting results of smFRET efficiency distribution histograms. (PDF)

#.

J.W. and Y.T. contributed equally. All authors have given approval to the final version of the manuscript.

We acknowledge the support from National Natural Science Foundation of China [22477132 and 32171443 to S.W., 31920103011 and 32371281 to S.P., 12374208 to G.W.], The National Key Research and Development Program of China (2023YFF1204402 to G.W.), Guangdong Basic and Applied Basic Research Foundation (2023A1515010157 to L.H.), Beijing Natural Science Foundation (International Scientists Project IS23070 to S.P., 5254039 to J.W.), and the National Laboratory of Biomacromolecules (to S.P.). T.P.J.K. acknowledges funding from the European Research Council under the European Union’s Seventh Horizon 2020 Research and Innovation Program through the ERC grant DiProPhys (Agreement ID 101001615), the Frances and Augustus Newman Foundation, and the Centre for Misfolding Diseases. T.S. acknowledges support from the European Union’s Horizon 2020 Research and Innovation Program under the Marie Skłodowska–Curie grant MicroREvolution (Agreement No. 101023060).

The authors declare no competing financial interest.