Here, we investigated the use of a small fluorinated probe molecule, TFE, added to the sample (10 mM final concentration) to study phase transitions and layer separation in a model protein system, BSA, which undergoes LLPS at higher temperature in the presence of YCl3. We found that the TFE 19F NMR signal (observed at ∼ -77 ppm) is sensitive to protein concentration, as reflected by a number of NMR observable parameters (Fig. 1). Most importantly for our approach, TFE displays changes easily detectable in 1D 19F NMR spectra, including linear chemical shift perturbations (δ, Fig. 1a) and increased linewidth (Fig. 1c) with increasing protein concentration. Additionally, TFE δ linearly depends on environmental factors such as temperature and concentration of additives. These changes in 19F δ can be used to create a calibration curve under known solution conditions (Equation (1) and Supplementary Table 1). TFE also displays increased transverse relaxation rate (R2) and reduced translational diffusion (DL) with increasing protein concentration (Fig. 1e-f), further suggesting it undergoes transient interactions with protein molecules. The concentration dependence of DL closely follows the expected behaviour explained by increased molecular crowding alone (Fig. 1f).
As an additional control, 100 mM trifluorotoluene (TFT) in a coaxial insert in the same sample tube produces a simultaneously observed 19F signal (at ∼ -61 ppm), while its solvent DMSO-d6 provides the deuterium field-frequency lock signal required by the NMR spectrometer, all without any adulteration of the protein sample. This TFT signal is virtually unperturbed by changes in protein concentration (Fig. 1), and its integral and chemical shift can be used as a convenient external reference. Importantly, the lineshape of the TFT 19F signal also acts as a sensor for any macroscopic magnetic field inhomogeneity in the outer protein sample, such as that emerging from the process of layer separation. The observable signals from the probe and reference molecules present in the sample therefore report directly on both local protein concentration (TFE) and the macroscopic behaviour of the sample (TFT).
TFE behaviour in isolated dense and lean fractions. Having established that TFE is sensitive to protein concentration, we next investigated how signal behaviour differs between the dense (360 mg/mL) and lean (80 mg/mL) fractions produced by macroscopic LLPS at 40°C. LLPS was triggered in protein solutions by temperature incubation, with the resulting dense and lean layers separated, and each of the isolated fractions examined by bulk-detection NMR as a function of temperature (Fig. 2).
Remarkably, despite extremely high concentration and viscosity in the dense fraction, a characteristic signal from the TFE probe is easily detectable in both fractions. These signals were markedly different in the two fractions, with the lean fraction resulting in a narrow intense peak upfield of the broader dense fraction signal (Fig. 2a). The chemical shift difference between the two fractions is consistent with the measured protein concentrations and the calibration relationship established earlier (Fig. 1a). In the dense fraction, TFE exhibited significantly slower diffusion and faster relaxation rates (Fig. 2d-f) than in the lean fraction, likely as a result of increased crowding and TFE transiently interacting with BSA. Together these data in isolated fractions show that TFE is dispersed uniformly throughout the two phases, and not confined to the phase boundaries or edges. Although the temperature dependence of molecular diffusion in both dense and lean fractions can be fully explained by changes in water viscosity (Fig. 2d), R2 in the dense fraction decreases with temperature slower than expected from viscosity alone (Fig. 2f), suggesting additional temperature-dependent changes in the chemical exchange regime.
Tracking fast kinetics of LLPS through bulk-detection NMR. As the lean and dense fractions give rise to distinctive TFE NMR signals, we next explored how NMR can be used to track the simultaneous appearance and development of these phases during the course of LLPS under different conditions. Here, 200 mg/mL BSA with 20 mM YCl3 was subjected to temperature jumps from 25°C to 40, 45 or 50°C, as a trigger for LLPS. The rate and nature of LLPS was significantly different at increasing temperatures. At 40 and 45°C, BSA underwent macroscopic LLPS with complete layer separation, while at 50°C the solution became extremely opalescent, but without subsequent layer separation, suggesting an arrested state (Supplementary Fig. 3).
For the initial faster kinetics of LLPS preceding layer separation, bulk-detection NMR was used to observe evolution of the TFE probe signal, and thus the emergence of different phases throughout the sample (Fig. 3). Following the temperature jump trigger, the initial single peak develops into two overlapping species, a narrower upfield peak with a broader downfield shoulder. These species are in good agreement with the TFE signals observed in isolated fractions, with the broad shoulder originating from the dense phase, and the narrow peak originating from the lean phase. Additionally, the chemical shifts, widths and intensities of the species continue to evolve over time, indicating changes in the composition of the two phases during LLPS.
To quantify the process of LLPS, TFE probe signals were deconvoluted (Supplementary Fig. 4&5), revealing two emerging and evolving components, one with lower and another with higher characteristic concentration (Fig. 4). As TFE chemical shift is linearly sensitive to BSA concentration (Fig. 1a), the concentrations of the two evolving species could be determined (Equation (1), Fig. 4b), while the integral of each signal reports on the apparent volumes of the two species (Equation (2), Fig. 4a). Together, this determination of the apparent phase volumes and concentrations allows calculation of the observed mass (and hence, population) of protein in each phase (Equation (3)).
The analysis of data presented on Fig. 4 allows one to measure how the concentration and apparent volume of each phase evolves with time and quantitatively track the process of suspended phases emerging throughout the sample (i.e. microscopic phase separation). Importantly, the experimental results reveal that the kinetics of LLPS in the BSA system at different temperatures goes through the same steps, albeit with substantially different rates. The major similarities include: (i) initial fast drop in concentration (and density) of both emerging lean and dense phases accompanied by rapid increase of the total dense phase volume; (ii) existence of a distinct crossover point where the minimum of dense phase concentration and maximum volume is reached, followed by (iii) the stage where dense phase compacts, thus increasing its density. Remarkably, the arrested state emerging at 50°C follows the same trend, with the only difference that dense phase ‘droplets’ are unable to demix and coalesce, and ultimately form a system-spanning network (Supplementary Fig. 3). Additionally, the transition triggered by the lowest temperature, 40°C, shows evidence of an initial time lag in growth of the dense phase volume, and the crossover time point where the minimum dense phase concentration and maximum volume is reached is shifted towards the end of the 120 min observation window: such behaviour is more prominently visible for longer observation times, see below and Fig. 5.
Additionally, the concentration of the dense phase after 120 minutes was also higher at higher temperatures, while the lean phase concentration was very similar at all three temperatures (Fig. 4). Conversely, the volume of the dense phase decreases with temperature (Fig. 4b), such that the final mass of protein in each phase is broadly similar at all three temperatures (Fig. 4e-f). Importantly and reassuringly, at all temperatures, the calculated total protein mass in the observed sample volume remains largely constant and is in good agreement with the expected protein mass (Fig. 4g), which suggests that calibration-based concentration and volume quantification of the emerging phases works well.
Tracking slower kinetics of layer separation by bulk-detection and spatially-selective NMR. In some systems, LLPS may proceed beyond a suspension of dense droplets and additionally exhibit macroscopic LLPS, with the droplets settling into a lower dense layer with a discrete boundary to an upper lean layer (also see Supplementary Fig. 3).12,19,21,38 Therefore, we next extended our experimental approach to study this process of layer separation. Here, the BSA solution previously observed undergoing LLPS at 40°C was monitored for an extended period, catching both microscopic and macroscopic LLPS (Fig. 5).
After 145 min, the TFE probe exhibited marked changes in behaviour, with a rapid increase in the apparent volume of the dense phase in the observed volume (Fig. 5a) and thus the apparent calculated total mass (Fig. 5c), without significant increase in dense phase concentration (Fig. 5b). These observations arise from layer separation resulting in a significant redistribution of protein across the entire sample volume, with the position of the NMR observed volume mainly capturing the dense layer towards the bottom of the tube (Supplementary Fig. 3a). Accompanying layer separation, the TFT reference exhibited signs of increased magnetic field inhomogeneity on a macroscopic scale arising from the protein solution, with altered lineshape leading to reduction in signal intensity and increases in signal width (Fig. 5d). Remarkably, the magnitude of the changes in TFT lineshape are very small compared to the broader TFE signal, such that inhomogeneity in this instance does not significantly affect interpretation of the TFE probe signals.
As layer separation leads to differences in distribution of the phases across the BSA solution, we next investigated the use of spatially-selective NMR to examine this process in greater detail. Spatially-selective NMR enables signals from a specific horizontal slice of the sample to be collected. Due to non-linearity of the gradient coils required for spatially-selective NMR, the total observable sample length (12 mm / 130.1 µL) was less than for bulk-detected NMR (22 mm / 238.6 µL) allowing observation of only the central part of the sample (Supplementary Fig. 1&2).
During microscopic LLPS but before layer separation (Fig. 6, first two time points), each phase is distributed similarly across all slices of the sample (additional data in Supplementary Fig. 6). Additionally, the ratios of the volumes of the two phases detected by spatially-selective NMR are in good agreement with the apparent volumes detected by bulk-detection NMR (Fig. 5). After the onset of layer separation at 140 min, three distinct settling regimes are observed (Fig. 6). Initially, the entire NMR observed volume quickly becomes dominated by the dense phase, which sinks downwards under gravity, with a residual volume of lean phase remaining trapped in this dense layer. At this point, the bulk of lean layer is above the NMR observed volume, while the boundary between the two diffuse layers sinks downwards with time. After 500 min, lean phase population is observed increasing in the top slice (+6) until it is the predominant phase present in this slice after 900 min (Fig. 6), when the boundary is observed moving into the next slice (+5). Analysis of the boundary position reveals that the boundary sinks at a constant rate of 0.154 mm/hr during this regime (Supplementary Fig. 7). Finally, after 1100 min the boundary movement significantly slows, further settling downwards at a rate of 0.004 mm/hr (Supplementary Fig. 7), showing evidence of an even slower process of dense layer evolution, likely linked with gradual release of remnants of lean phase trapped inside the dense layer. Visually, this process corresponds to reduction of the dense layer opalescence (as initially seen on Supplementary Fig. 3a) with time, typically days, in line with observations with this and other systems undergoing macroscopic LLPS.12,19 Together these data show that the TFE probe allows one to monitor the entire process of LLPS, from the onset and evolution of the phase transition, through to layer separation and settling, revealing the kinetics and complexities of LLPS.