Macromolecular condensation buffers intracellular water potential – Nature

Plasmids

The plasmid encoding GFP-tagged TIA1 was obtained from Addgene (106094). The plasmid for bacterial expression of the anti-GFP nanobody, named GBP (for GFP-binding peptide) was previously described⁵⁹. To generate a plasmid encoding GFP-tagged FUS low-complexity domain (noted FUS-LCGF; corresponding to the FUS reference sequence NM_004960.4), purified RNA from human SH-SY5Y cells was reverse-transcribed using a FUS-specific primer (5′–3′, CGCCGCCGCCACCACTGCC). cDNA was then PCR-amplified using FUS low-complexity-spanning forward and reverse primers with EcoRI and BamHI sites, respectively (forward-EcoRI, 5′–3′, ATCTATGAATTCGCCACCATGGCCTCAAACGATTATACCCAAC; reverse-BamHI, 5′–3′, ATATGGATCCCCTCCACGGTCCTGCT). After EcoRI/BamHI digestion, the FUS low-complexity fragment was cloned into the pEGFP-N1 vector (BD Biosciences; GenBank: U55762). FUS-LC-EGFP plasmid was verified by Sanger-sequencing.

Protein and reagents

Unless stated otherwise, reagents and purified proteins were purchased from Sigma-Aldrich. Dextran-35kDa was from Leuconostoc mesenteroides (Sigma-Aldrich, D1662). Rhodamine-PEG-20kDa was from Creative PEG works. FITC-labelled ubiquitin was purchased from Thermo Fisher Scientific. Tag-free GFP was expressed and purified as described previously⁵⁹.

To generate BSA–A647, purified BSA (4 mg ml⁻¹ in 0.1 M sodium bicarbonate, pH 8.3) was reacted for 1 h with a tenfold molar excess of Alexa Fluor 647 NHS Ester (Thermo Fisher Scientific, A20006). Excess dye was removed using the Zeba Spin column (Pierce) equilibrated in 20 mM HEPES, 150 mM KCl, pH 7.6.

To generate GBP–A555, purified GBP (0.5 mg ml⁻¹ in 0.1 M sodium bicarbonate, pH 8.3, purified as previously described)⁶⁰, was reacted for 1 h with a fourfold molar excess of Alexa Fluor 555 NHS Ester (Thermo Fisher Scientific). Excess dye was removed by a homemade G25 column equilibrated in PBS. Protein was flash-frozen in liquid nitrogen and stored at −80 °C.

Osmometry

Vapour pressure osmometry was performed on the Vapro 5600 (ELITechGroup) system according to the manufacturer’s instructions, with temperature controlled by varying the ambient room temperature. The instrument was allowed to equilibrate to ambient temperature overnight, and the reading stability was validated before measurements were performed. Osmometry readings were assessed for solutions of varying composition and confirmed to be normally distributed with the following tests: Anderson–Darling, D’Agostino–Pearson, Shapiro–Wilk, Kolmogorov–Smirnov. A buffer containing 20 mM Tris-HCl pH 7.4, 150 mM KCl was used for preparation of all solutions and dilution series, except for Xenopus and human cell extracts, which were diluted with pure water because their neat osmotic potential was similar to that of the buffer. Vapour pressure osmometry of Xenopus extracts (Fig. 5d) was performed on thawed frozen samples (that is, not fresh). Human cell extracts and dilutions were maintained on ice until immediately before each reading. Stock macromolecular solutions were incubated overnight at 37 °C, with agitation at 300 rpm, to ensure complete dissolution before measurements were performed, and used to make dilution series. For the main figures, presented values were baseline-corrected by subtracting the mean osmotic potential of the buffer during each set of osmometry measurements; whereas, absolute osmotic potential values are shown in the Extended Data Figures. The error of predicted values of the osmotic potential of sucrose mixtures (Extended Data Fig. 10f) was computed by error propagation (that is, the square root of the sum of the squares of the s.e.m. of relevant samples). Osmotic potential data were fitted using MATLAB 2020b (Supplementary Discussion 1).

Osmometry measurements of fresh, cell-cycle arrested, Xenopus egg extracts (Extended Data Fig. 10l) were performed using the Osmomat 3000 freezing-point-depression osmometer (Gonotec). Before measurements on extract samples, a three-point calibration of the osmometer was performed using ultrapure water and certified 100 and 300 mOsm kg⁻¹ solutions. To ensure accurate measurement of the difference between the two cell cycle states, we measured the osmotic potential across a dilution series, and extrapolated values from the part of the curve that shows linear behaviour (the osmotic potential as a function of dilution shows nonlinearity, as expected because the cytosol is composed of a mixture of ideal and non-ideal proteins like BSA; Extended Data Fig. 1). Undiluted extract was kept in a water bath at 18  °C or on ice during preparation of dilutions, and we did not observe a systematic influence of the measured osmotic pressure on the incubation temperature before measurement. To reduce the time spent in a diluted state before measurement, each dilution was prepared individually immediately before measurement, typically during the measurement of the preceding sample. The samples were prepared at the desired dilution by addition of ultrapure water and extract to a final volume of 35 µl in a fresh osmometer tube, and pipette-mixed with a P200 tip until visually homogeneous. Care was taken to avoid the introduction of small air bubbles in the solution during preparation. Between measurements, the osmometer probe was wiped twice with a dry Kimwipe, rinsed with ultrapure water and wiped twice again with a fresh Kimwipe. Dilutions were prepared and measured in non-sequential order (for example, 100%, 50%, 10%, 90%, 20% and so on) to avoid superimposing any potential time dependence to the titration data.

Note that, to measure the free-water buffering capacity of extracts (Fig. 5d and Extended Data Fig. 10j,k), it is important to use a vapour-pressure osmometer rather than a freezing-point-depression osmometer, to ensure that temperature changes during measurement, which affect condensation, do not affect the results.

Cell culture

SH-SY5Y cells (Sigma) were maintained in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 20% fetal bovine serum (FBS), 1× non-essential amino acids (Gibco) and penicillin–streptomycin (100 U ml⁻¹) at 37 °C and 5% CO₂. Human U2OS osteosarcoma cells (ATCC) or mouse primary lung fibroblasts were maintained in DMEM (Thermo Fisher Scientific) supplemented with penicillin–streptomycin (100 U ml⁻¹) and 10% Hyclone Fetalclone II or 10% Hyclone III, respectively. GlutaMAX (Gibco) was used when medium did not contain glutamine for long-term culture. For transient transfections, cells were transfected with Lipofectamine-2000 (SH-SY5Y cells) or −3000 (U2OS cells) 24 h before an experiment, according to the manufacturer’s instructions. For cell viability analysis, lymphoblastoid Raji cells⁶¹ (gift from P. Farrell, described previously⁶²) or human foreskin fibroblasts (HFFs, from ATCC) were cultured in RPMI or DMEM, respectively, with 10% FCS, GlutaMAX (Gibco) and penicillin–streptomycin in a 37 °C humidified incubator with 5% CO₂. All cell lines were tested and found negative for mycoplasma contamination.

Cell extracts

To generate concentrated bacterial extracts (Extended Data Fig. 10j), BL21 bacteria were grown overnight in Terrific Broth medium at 25 °C. A bacteria pellet corresponding to around 20 ml was washed once with 20 mM Tris, 150 mM KCl, pH 7.4, then resuspended in 7 ml of 20 mM Tris, 150 mM KCl, pH 7.4 enriched with protease inhibitors (Roche Complete), lysozyme (0.8 mg ml⁻¹ final), DNase I (4.2 µg ml⁻¹ final), pefabloc (0.5 mg ml⁻¹ final), benzamidine (5.3 mM final) and PMSF (0.6 mM final). The extract was incubated on ice for 10 min, sonicated, then centrifuged at 20,000g four times until a clear solution was obtained.

For human cell extract, Expi293F suspension cells (Thermo Fisher Scientific) were grown to a density of 2.5 × 10⁶ ml⁻¹ in Freestyle serum and protein-free culture medium (Thermo Fisher Scientific) in a shaking incubator at 37 °C with 5% CO₂. A total of 1.25 × 10⁹ cells was collected by centrifugation at 2,000g for 20 min and the culture medium was aspirated. The cell pellet was added to liquid nitrogen and homogenized to a powder under cryogenic conditions using pestle and mortar with 30 mg (1/2 tablet) of cOmplete EDTA-free protease inhibitor cocktail (Roche). Powdered lysate was stored at −80 °C, then thawed at 0 °C. The lysate was sonicated 10 times with 30 s cycles at 4 °C, then clarified by two rounds of centrifugation at 4 °C for 10 min at 1,000g, with the supernatant transferred to a new tube on ice each time and gently triturated.

Cytostatic-factor (CSF)-arrested Xenopus laevis egg extracts were prepared as previously described^63,64, with the exception that cytochalasin D was replaced with cytochalasin B. In brief, unfertilized eggs were collected, dejellied and crushed by centrifugation. After extraction of the cytoplasmic fraction, a 1/1,000 volume of a protease inhibitor mixture (leupeptin, pepstatin and chymostatin) as well as a 1/1,000 volume of cytochalasin B were added to the extract, giving final concentrations of 10 µg ml⁻¹ each. This CSF-arrested extract was stored on ice. To check the state of the CSF-arrested extract, a test reaction was prepared by addition of a small volume of frog sperm to 25 µl of the extract (to a final concentration of approximately 200 sperm per µl). After incubation at 18 °C for 30 min, arrest in meiosis II was confirmed by visualization of half-spindles using polarized-light microscopy as described previously⁶⁵.

Interphase-arrested X. laevis egg extracts were prepared as previously described⁶⁶, with minor modifications. In brief, unfertilized eggs were dejellied and then activated by incubation for 2–4 min in 0.2× Marc’s Modified Ringers (MMR) buffer supplemented with 0.5 µg ml⁻¹ calcium-ionophore (A23817, Sigma-Aldrich). Activated eggs were then washed and crushed by centrifugation. As for the CSF-arrested extract, protease inhibitors and cytochalasin B were added to the cytoplasmic fraction to 10 µg ml⁻¹ each. This cycling extract was stored on ice. To arrest in interphase, cycloheximide was added to a final concentration of 200 µg ml⁻¹. After a 20 min incubation on ice, this interphase-arrested extract was subsequently stored on ice and used for osmometry. When an extract preparation yielded significantly more fresh extract than could be used that day, the excess was spin-filtered, frozen slowly (approximately 1 °C min⁻¹) and stored at −80 °C.

Yeast

S. cerevisiae stably expressing PLCδ-2xPH-GFP in the WT or mss4^ts in the mss4-null background (noted simply mss4^ts in the following detailed genotype below), were generated for us by C. Godlee and M. Kaksonen. The original plasmids expressing 2×PH-GFP and mss4^ts were a gift from the laboratory of S. Emr¹⁷. Yeast cells were grown overnight in liquid medium lacking Ura and Trp supplemented with 2% glucose. Cells were subsequently diluted to an optical density at 600 nm (OD₆₀₀) of 0.1 and allowed to recover to an OD₆₀₀ of 0.4. The yeast was then allowed to lightly adhere for 15 min to glass coverslips coated with concanavalin A (50 μg ml⁻¹ in PBS, 30 min) before imaging. Imaging was performed in liquid medium lacking Ura and Trp supplemented with 2% glucose in the spinning-disk set-up described below. For hypoosmotic shocks, 50 μl of yeast medium was diluted with 950 μl of deionized water, giving a hypoosmotic shock of 405 mOsm l⁻¹ to 23 mOsm l⁻¹. Note that the frame acquired immediately after hypoosmotic shock was removed, due to a transient change in membrane signal, presumably due to the change in ionic gradient across the membrane. The temperature was maintained at 32 °C (hypoosmotic shock experiments; Figs. 1d–e and Extended Data Fig. 2d–f) or 39 °C (heat-shock experiments; Extended Data Fig. 2a–c) using a heated stage (TOKAI HIT INUB-ZILCS-F1).

S. pombe strains expressing the microtubule marker alpha tubulin (Atb2) fused to mCherry and the spindle pole body (SPB) marker Sid4 fused to GFP in the cut7-24 background were maintained at 30 °C on yeast-extract-supplemented (YES) plates, and maintained every third day. Cut7-24 induces monopolar spindle formation at the restrictive temperature^18,67. For live-cell imaging, cells were transferred into liquid YES medium and imaged the next day during exponential growth. All S. pombe live-cell imaging was performed in eight-well μ-slides (Ibidi, IB-80807), preincubated over night with lectin (soybean, L1395, Merck) in the spinning-disk instrument described below, equipped with a heating chamber to maintain precise temperature (35 °C or 37 °C). Cells were incubated on the microscope stage for 30 min to ensure thermal equilibrium before the images were acquired. Stacks of seven planes (Δz = 1 μm) were acquired for each channel with 100 ms exposure time for 488 nm and 200 ms for the 561 nm laser. For each time-lapse video, an image was taken every 5 min for 180 min. Cells treated with deuterium oxide (D₂O) were grown in 50% 2× YES + 50% D₂O over night before imaging. Exposure to hypoosmotic condition (5% YES + 95% H₂O) was performed directly in the imaging dish at the microscope, with imaging starting directly after medium exchange.

Detailed genotypes were as follows: WT: MATa, his3Δ200, leu2–3, 112, ura3–52, lys2–801, pRS426-PLCdelta-2xPH-GFP, and mss4^ts: MATa, his3Δ200, leu2–3, 112, ura3–52, lys2–801, mss4::hphNT1, YCplac111-mss4-2ts (CEN LEU2), pRS426-PLCdelta-2xPH-GFP (Fig. 1c–e); mss4^ts: MATa, his3Δ200, leu2–3, 112, ura3–52, lys2–801, mss4::hphNT1, YCplac111-mss4-2ts (CEN LEU2), pRS426-PLCdelta-2xPH-GFP (Extended Data Fig. 2a–c); WT: MATa, his3Δ200, leu2–3, 112, ura3–52, lys2–801, pRS426-PLCdelta-2xPH-GFP, and mss4^ts: MATa, his3Δ200, leu2–3, 112, ura3–52, lys2–801, mss4::hphNT1, YCplac111-mss4-2ts (CEN LEU2), pRS426-PLCdelta-2xPH-GFP (Extended Data Fig. 2d–f); and cut7-24: h+ , cut7–24, Sid4-GFP:Leu2+ , mCherry-Atb2:Hph (Extended Data Fig. 2g–i).

Mouse primary fibroblasts and chondrocytes

All animal work was licensed by the Home Office under the Animals (Scientific Procedures) Act 1986, with Local Ethical Review by the Medical Research Council and University of Manchester. Primary lung fibroblasts were isolated as described previously⁶⁸. Primary chondrocytes were isolated from 5-day-old mice using a previously described protocol⁶⁹. In brief, 5-day-old C57Bl/6 PER2::Luc mice were euthanized by decapitation. The knee, hip and shoulder joints were dissected, and any soft tissue removed. Joint cartilage was subjected to pre-digestion with collagenase D 3 mg ml⁻¹ in DMEM twice for 30 min at 37 °C with intermittent vortexing to remove soft-tissue leftovers. Subsequently, the cartilage was diced using a scalpel and digested overnight at 37 °C. Cells were dispersed by pipetting and passed through a 70 µM cell strainer. The cell suspension was then centrifuged and the pellet was resuspended in DMEM/F12 with 10% FBS and plated in T75 flasks. Cells were passaged only once before performing experiments.

For calcium imaging, primary chondrocytes were plated onto four-chamber 35-mm glass-bottomed dishes (Greiner Bio-One). A total of 0.5 µl of 1 mM Fluo-4 AM calcium dye (Thermo Fisher Scientific) was added to each chamber containing cells cultured in 500 µl DMEM/F12 medium and incubated for 20 min. Image acquisition was performed by confocal microscopy (see below). Hyperosmotic shock was induced by addition of sorbitol to the culture medium and hypoosmotic by addition of distilled water. Temperature changes were achieved by adding ice cold medium to the cell culture chamber at a mass calculated according to the following formula:

where t is the final temperature (°C), c_1,…,n is the specific heat of substances (kJ kg⁻¹ °C) and t_1,…n is the temperatures of the substances (°C).

The correct temperature was confirmed using a handheld infrared thermometer.

Viability assays

For viability experiments, Raji cells were seeded at 3 × 10⁵ ml⁻¹ then collected and resuspended 24 h later in challenge RPMI. HFFs were seeded in six-well plates and confluent monolayers were placed in challenge DMEM. For hypothermic treatment, cells were transferred to precooled challenge RPMI or DMEM medium made by varying the NaCl concentration to achieve a final osmolality of between 150 and 450 mOsm l⁻¹, placed into a precooled metal block and maintained at 4 °C or on ice for 24 h. For hyperthermic treatment cells were transferred to preheated challenge RPMI or DMEM made using 0–50% D₂O, placed into a preheated metal block and incubated for 30 min at 47 °C (Raji suspension) or 45 min at 50 °C (adherent HFFs) before transferring to 0% D₂O control medium and culturing for a further 24 h at 37 °C before viability analysis. For both hypo- and hyperthermic experiments, control cells were incubated in parallel in a metal block at 37 °C.

Raji suspension cell viability was assessed by flow cytometry. In brief, cells were centrifuged, washed twice in PBS then consecutively stained with Calcein AM (eBioscience) and DRAQ7 (Biostatus) according to manufacturer’s instructions before flow cytometry analysis (Fortessa instrument followed by analysis using FlowJo v.10.6). The viability was evaluated for 10,000 suspension cells per replicate. For HFF adherent cell viability, cells were similarly stained and the monolayer viability was assessed by imaging stained cells using an Evos FL Auto2 microscope then measuring live cells using Fiji⁷⁰.

Kinase assay

Full-length human WNK1 (residues 1–2,382) was expressed in Expi293 cells and purified as we previously reported⁷¹. GCN2 was expressed in Sf9 cells and purified as previously reported⁷². Kinase assays were performed with 50 ng protein and 0.5 µM kinase in assay buffer (50 mM HEPES, 100 mM KOAc, 2 mM MgAc₂, 0.1 mM ATP, 0.2 mCi ml⁻¹ γ-32P-ATP, 1× Phos-STOP, pH 7.5) with the indicated concentrations of PEG-20kDa (Sigma-Aldrich, 81300-1KG). Reactions were incubated at 32 °C for 30 min and then quenched with sample buffer and boiled. The samples were then diluted sevenfold before electrophoresis due to the high concentrations of PEG. Incorporation of γ-³²P was analysed by exposing the gel to a phosphor screen.

Actin polymerization assays

ADP-actin preparation was adapted from ref. ⁴⁸. In brief, purified actin (Cytoskeleton) was resuspended according to the manufacturer’s instructions. Resuspended Ca²⁺-ATP-actin was then diluted to 50 µM in G-buffer (5 mM Tris-HCl pH 8.0, 0.2 mM CaCl₂, 0.2 mM ATP) and incubated with 80 µM MgCl₂ and 200 µM EGTA for 3 min on ice to exchange into Mg²⁺-ATP-actin. To ensure depletion of all of the remaining ATP in solution, 20 U ml⁻¹ of hexokinase (H6380, Sigma-Aldrich), 0.2 mM ADP and 1 mM glucose were added to the solution and incubated for 3 h on ice. Finally, the solution was centrifuged at 100,000g for 1 h at 4 °C to remove actin filaments, and the supernatant containing Mg²⁺-ADP-actin was used for the rest of the experiments. The actin concentration was measured using the BCA assay (Pierce).

Actin polymerization was induced by preparing 50 µl solution of 0.7 µM Mg²⁺-ADP-actin in F-buffer (50 mM KCl, 1 mM MgCl₂, 1 mM EGTA, 1 mM DTT, 10 mM imidazole pH 7.0) or G-buffer (5 mM Tris-HCl pH 8.0, 0.2 mM CaCl₂, 1 mM DTT) supplemented with either 0.2 mM ADP or 1 mM ATP depending on the condition tested, and 100 mg ml⁻¹ PEG-35kDa or 100 mg ml⁻¹ dextran-35kDa. Note that 0.7 µM is above the critical concentration (C_c) for ATP-actin, but below that of ADP-actin⁷³. The solutions were allowed to polymerize for 1 h at room temperature and then centrifuged at 100,000g for 1 h at 4 °C. From each condition, 40 µl of the top of the supernatant was removed, saved and labelled as ‘supernatant’. The remaining 10 µl were discarded and the pellet was carefully washed three times with 3 × 50 µl of polymerization buffer. Finally, the pellet was resuspended in 50 µl of polymerization buffer, and 40 µl was saved and labelled as ‘pellet’. The samples were then analysed by SDS–PAGE on NuPAGE 4–12% Bis-Tris gels (Life Technologies) according to the manufacturer’s instructions. Instant Blue (Sigma-Aldrich) was used for total protein staining of gels (Extended Data Fig. 10i). Note that, to avoid artefacts during SDS–PAGE due to the high PEG-35kDa/dextran-35kDa concentrations, all samples were diluted two times before loading in the gel. As a control, we performed the same experiment with BSA in F-buffer to verify that proteins of similar size to monomeric actin do not pellet in the presence of dextran or PEG (Extended Data Fig. 10i (bottom lane)).

Structural analysis

To estimate the water liberated by assembly of ADP-actin (Extended Data Fig. 10h), we used the surface area of actin as a proxy. The solvent-accessible surface area was measured using Pyrosetta⁷⁴ with a probe radius of 2 Å. An existing Cryo-EM structure of F-actin was used (Protein Data Bank (PDB): 5ONV), with the solvent-accessible surface area (SASA) measured in the monomeric and assembled state. This analysis was repeated for other available actin structures (PDB: 4A7N, 7BT7, 8A2Z), and the SASA averaged and compared between the two states.

GFP–FUS phosphorylation analysis

To analyse the specific effects of global modulations of the phospho-proteome on FusLC–GFP (Extended Data Fig. 8k), SH-SY5Y cells were transiently transfected with FusLC–GFP. Then, 24 h after transfection, cells were treated with Calyculin A (3 nM) for 15 min, or Staurosporine (10 µM) for 50 min at 37 °C under 5% CO₂ (one confluent 10 cm dish per condition, per replicate). Cells were then washed twice with ice-cold PBS, scraped and collected in Falcon tubes. After centrifugation at 1,800 rpm for 5 min 4 °C, the pelleted cells were lysed for 20 min in lysis buffer (40 mM HEPES, 150 mM KOAc, 2.5 mM Mg(OAc)₂, 1.0% Triton X-100, 0.1% SDS, pH 7.4) supplemented with protease inhibitors (Complete, Roche), phosphatase inhibitors (PhosSTOP, Roche) and DNase (D4513, Sigma-Aldrich). Lysates were then sonicated using a Bioruptor sonicator (Diagenode) at 4 °C for 3 cycles 30 s on/30 s off and cleared at 20,000g for 10 min in preparation for FusLC–GFP immunoprecipitation. Then, 25 µl of GFP-Trap Resin (gta-20, Chromotek) slurry was aliquoted and pre-equilibrated by washing in the lysis buffer three times. After equilibration, the lysates were loaded onto the beads and incubated for 2 h at 4 °C with rotation. The beads were then washed three times in wash buffer (20 mM HEPES, 150 mM KOAc, 2.5 mM Mg(OAc)₂, 0.5% Triton X-100, pH 7.4, supplemented with protease and phosphatase inhibitors). The beads were then resuspended in 2× NuPage LDS sample buffer (Thermo Fisher Scientific), 5 mM DTT and boiled at 95 °C for 5 min. The samples were then analysed on regular, NuPage 4–12% Bis-Tris gels (NP0321BOX, invitrogen) or SuperSep Phos-tag 50 μmol l⁻¹, 7.5% gels (192-18001, FUJIFILM) according to the manufacturer’s protocol. For immunoblotting, gels were briefly washed in Millipore water and protein transfer onto nitrocellulose membranes was performed using the dry transfer iBlot2 system (IB21001, Thermo Fisher Scientific) with a standard (P0, 7 min) protocol. The membranes were briefly washed in water, then blocked in 5% (w/w) non-fat dried milk (Marvel) TBS containing 0.1% Tween-20 (TBST) for 1 h at room temperature. The membranes were then incubated with primary anti-GFP antibodies (ab290, Abcam, 1:10 000 dilution) diluted in blocking buffer (5% milk, TBST), at 4 °C overnight. Subsequently, the membranes were washed three times for 10 min (in TBST) and then incubated with anti-rabbit HRP-conjugated secondary antibodies (A6154-1ML, Sigma-Aldrich, 1:10,000 dilution) diluted in blocking buffer for 1 h at room temperature. Excess secondary antibody was removed by washing three times for 10 min (in TBST). Chemiluminescence detection was performed by incubating the membranes with the Immobilon reagent (Millipore) and imaging using the ChemiDoc MP (Bio-Rad) system.

Microscopy

Microscope hardware

Most imaging (except for that in Extended Data Figs. 3b,g and 4e) was performed using a custom TIRF/spinning-disk confocal microscope, comprising a Nikon Ti stand equipped with a perfect focus system and a fast-piezo stage (ASI). All data were collected with a ×100 PLAN Apo Lambda NA 1.45 or a ×60 PLAN Apo Lambda NA 1.4 objective. The confocal imaging arm consisted of a Yokogawa CSU-X1 spinning-disc head and a Photometrics 95B back-illuminated sCMOS camera, operated in pseudo global shutter mode (synchronized with the spinning-disc rotation). Camera binning was set to 1, and electronic gain was set to 1. Illumination was provided by 405 nm, 488 nm (150 mW OBIS LX), 561 nm (100 mW OBIS LS) or 630 nm (140 mW OBIS LX) coherent lasers mounted in a Cairn laser launch. Single band-pass filters (Chroma 525/50 for GFP, Chroma 595/50 for mCherry and Chroma ET655lp for A647/JF-646) or a quad bandpass filter (Chroma ZET405/488/561/640 m for Hoescht) were mounted within a fast Cairn Optospin filter wheel. To enable fast four-dimensional acquisition, an FPGA module (National Instrument sbRIO-9637 running custom codes) was used for hardware-based synchronization of the microscope. For z-axis acquisitions, the FPGA ensured that the piezo z-stage moved only during the readout period of the sCMOS camera. The whole system was controlled by Metamorph software, and the sample temperature was maintained using a custom heating enclosure (https://MicroscopeHeaters.com). For long-term videos (>2 h), open water bottles were added to the heating enclosure to increase humidity and decrease evaporation.

Alternatively, calcium imaging (Extended Data Fig. 3b) was performed using the Zeiss Exciter confocal microscope equipped with a 37 °C stage in humidified 5% CO₂. Imaging was performed using a 488 nm excitation wavelength and 520 nm band-pass filter for emission and Fluar 40× NA 1.3 (oil-immersion) lens. Image capture was performed using the Zeiss software Aim v.4.2 with the Autofocus macro⁷⁵. For adherent cell viability measurements, a Evos FL Auto2 microscope was used (Extended Data Figs. 3g and 4e).

In vitro condensation assays

To image condensation, we assembled home-made sealed chambers by sandwiching a 0.5-mm-thick PDMS insert between two glass coverslips. Inserts, with a 5-mm-diameter hole, were cut out of PDMS sheets (Silex Silicone) with a Graphtec CE6000 cutting plotter. Both glass surfaces were passivated with PLL-PEG (SUSOS, 0.5 mg ml⁻¹ in 20 mM Tris, pH 7.4, 1 h, at room temperature) to prevent protein adsorption to the glass surface. Solutions were added to the well and the second coverslip was added to prevent the development of an air–water interface. Images were acquired within 2 min of protein addition, with protein added at the indicated concentration. PEG solutions were made up in 20 mM Tris + 150 mM KCl.

Imaging condensation in cells under various challenges

SH-SY5Y (Fig. 3a,c–e and Extended Data Fig. 8a,b,i,j) or U2OS (Fig. 3b and Extended Data Fig. 8c,d,g,h) cells were transiently transfected with plasmids expressing FusLC–GFP or TIA1-GFP 24 h before imaging. Cells were spread for 1 h on glass-bottom imaging dishes (fluorodishes, World precision instruments) coated with fibronectin (50 μg ml⁻¹ in PBS, 1 h) before imaging. SH-SY5Y cells were imaged at 37 °C in Leibowitz’s L15 medium supplemented with 20% FBS and 20 mM HEPES pH 7.6. U2OS cells were imaged in DMEM high glucose, 10% Hyclone II and penicillin–streptomycin supplemented with 25 mM HEPES at 37 °C. In both cell lines, the extracellular osmolarity was altered by the addition of distilled water or sucrose. Cells were imaged before and after osmotic challenge by confocal z-stack imaging, and a maximum-intensity projection was computed. Then, the granulosity index, a measure of the condensation of the GFP signal, was measured in the automatically or manually segmented nucleus (see below). For each cell, data were normalized to the value of the granulosity index cell before osmotic challenge. We consistently incubated cells with fresh, prewarmed medium for pre-osmotic control measurements to avoid accumulation of osmolytes and cell byproducts in the medium over time that might otherwise change extracellular osmolarity and therefore alter the response of the cell. Unless stated otherwise, hyperosmotic shocks were performed by adding an identical volume of warm medium with twice the desired sucrose concentration onto the cells.

For imaging of nucleolar condensation in response to a hypoosmotic challenge (Extended Data Fig. 8e,f), untransfected SH-SY5Y cells were stained with Nucleolar-ID (Enzo) for 15 min according to the manufacturer’s instructions, before imaging over time after hypoosmotic challenge (50% distilled water in Leibowitz’s L15 medium supplemented with 20% FBS and 20 mM HEPES pH 7.6).

For imaging of FusLC–GFP condensation in response to global changes in protein phosphorylation (Extended Data Fig. 8i,j), SH-SY5Y cells transfected with FusLC–GFP plasmid were spread onto fibronectin-coated glass-bottom imaging dishes as described above and imaged over time in Leibowitz’s L15 medium supplemented with 20% FBS and 20 mM HEPES pH 7.6, after addition of Calyculin A (3 nM) or 10 µM Staurosporine.

For imaging of FusLC–GFP condensation in energy-depleted cells (Extended Data Fig. 8g,h), we adapted a previously established protocol for energy depletion⁷⁶. Specifically, U2OS cells were transfected with FusLC–GFP for 24 h and plated onto fibronectin-coated glass-bottom imaging dishes (ibidi 8-well dishes) in full growth medium for 2 h at 37 °C. The medium was then exchanged for prewarmed glucose/sodium pyruvate/glutamine-free DMEM (Thermo Fisher Scientific, A1443001) with the addition of 5% FBS, 1× GlutaMAX (Thermo Fisher Scientific), 10 mM 2-deoxy-d-glucose (Sigma-Aldrich, D6134), 10 mM sodium azide (Sigma-Aldrich, 08591) and 10 mM HEPES. Incubation for 30 min in this medium at 37 °C induced the characteristic absence of lamellipodia/ruffles (Extended Data Fig. 8g), as well as characteristic spike protrusions as previously reported⁷⁷ (Extended Data Fig. 8g (blue arrows)), demonstrating the efficacy of the energy-depletion treatment. As non-energy-depleted controls, cells were incubated in glucose/sodium pyruvate/glutamine-free DMEM (Thermo Fisher Scientific, A1443001) with the addition of 5% FBS, 1× GlutaMAX (Thermo Fisher Scientific), 10 mM d-glucose, 10 mM HEPES and 1× sodium pyruvate (Thermo Fisher Scientific) for 30 min at 37 °C. After acquisition of z-stacks under these isosmotic conditions, a hyperosmotic shock was induced by addition of one-tenth of total volume of a 584 mM concentrated sucrose solution (20 µl in 200 µl) and z-stacks of the same cells were reacquired.

For imaging of FusLC–GFP condensation in response to hyperosmotic shock in the presence of D₂O (Extended Data Fig. 4g,h), U2OS cells were transiently transfected with FusLC–GFP plasmid 24 h before imaging. Confocal z-stacks of cells were acquired in Leibowitz’s L15 medium supplemented with 20% FBS and 20 mM HEPES pH 7.6, before the extracellular osmolarity was altered by the addition of sucrose (+20 mOsm l⁻¹) in the presence or absence of 50% D₂O. The granulosity index was then measured in the nucleus and, for each cell, data were normalized to the value of the granulosity index cell before osmotic challenge.

For imaging of the long-term decrease in FusLC–GFP condensation in response to sustained hyperosmotic shocks (Extended Data Fig. 11a), U2OS cells were transiently transfected with FusLC–GFP plasmid 24 h before imaging. Cells were then plated onto fibronectin-coated, glass-bottom 35 mm imaging dishes as described above for 2 h at 37 °C under 5%CO₂ in DMEM high glucose supplemented with 10% Hyclone II and penicillin–streptomycin. Cells were then washed in DMEM high glucose supplemented with 10% Hyclone II, penicillin–streptomycin and 20 mM HEPES and 1 ml of this medium was added to the dish. Hyperosmotic shock was induced by adding 1 ml of DMEM high glucose supplemented with 10% Hyclone II, penicillin–streptomycin, 20 mM HEPES 150 mM sucrose, leading to a +75 mOsm osmotic shock (verified by osmometry). The dish was then sealed using a 40 mm coverslip and vacuum grease, and confocal imaging was initiated. Confocal z-stacks of cells were acquired every 30 min with a large z-spacing (20 planes, Δz = 0.75 µm) and at the minimum laser power to minimize photobleaching. At these long timescales, only qualitative analysis of condensation is possible, as multiple other variables come into play that (may) affect condensation (cell migration, cell-cycle state), as well as further expression of the transgene which will change the global fluorescence of the cell, which will bias our measurements. But, when recovery is observed, this occurs over hour timescales, which is consistent with established mechanisms of osmoregulation and the behaviour of other IDR proteins^43,50,78,79.

For fast imaging of short-term increase of FusLC–GFP condensation in response to hyperosmotic shock (Extended Data Fig. 6d,e and Supplementary Video 2), U2OS cells transfected with FusLC–GFP for 24 h were plated onto fibronectin-coated glass-bottom imaging dishes (ibidi 8-well dishes) in full growth medium (DMEM high glucose, 10% Hyclone II, sodium pyruvate, penicillin–streptomycin) for 2 h at 37 °C. Cells were washed once in full growth medium supplemented with 25 mM HEPES and left in 250 µl of this medium. Fast confocal acquisition was then started (single confocal plane, perfect focus enabled, camera operating in streaming mode, effective time between frames 140 ms), and 125 µl of prewarmed full growth medium enriched with 25 mM HEPES and 200 mM sucrose was added after 100 frames. This results in a +67 mOsm l⁻¹ shock.

Microfluidic control of temperature and osmolarity while imaging condensation

For imaging of cells undergoing both rapid temperature change in various osmotic conditions (Fig. 3c,d), we used the established Cherry Temp system microfluidics system (Cherry Biotech). With this system, it takes about 4 s for a temperature change from 37 °C to 20 °C at the sample⁸⁰. In brief, cells were spread in custom-made chambers made of a coverslip coated for 1 h with cell-adhesive fibronectin (Sigma-Aldrich, 50 μg ml⁻¹ in PBS), a PDMS spacer (0.5 mm thick) and a Thermalization chip (Cherry Biotech) at the top. Cells were imaged at different temperatures, in Leibowitz’s L15 medium supplemented with 20% FBS and 20 mM HEPES pH 7.6, osmolarity adjusted with sucrose or distilled water. The same cells were imaged within each temperature series, but different cells were imaged at the different extracellular osmolarities.

For simultaneous imaging of FusLC–GFP condensation within single cells in response to temperature and external osmolarity changes (Fig. 3e and Supplementary Video 1), we used dual-layer microfluidics devices (Thermaflow chips, Cherry Biotech). This combines fast temperature changes using the Cherry Temp system (see above), and fast medium changing using in an independent microfluidics loop (Elvesys). This cell medium microfluidics loop is composed of an OB1-positive pressure regulator (Elvesys), a ten-input MUX distributor valve (Elvesys) to select the different medium to add to cells, a bubble trap and a flow sensor positioned just at the entrance of the flow chamber. Flow was maintained at a constant low value of 114 µl min⁻¹ to avoid shear stress to the cells. The glass bottom of these microfluidics chips was coated with fibronectin (50 μg ml⁻¹ in PBS, 1 h), then SH-SY5Y cells transfected with FusLC–GFP were allowed to adhere in the flow chamber, before imaging using SDCM (21 z-planes, Δz = 0.5 µm) in Leibowitz’s L15 medium supplemented with 20% FBS and 20 mM HEPES pH 7.4. The temperature was then independently controlled using the Cherry Niotech top layer, while the external osmolarity was controlled by changing the medium going into the chamber by changing the input in the bottom layer for a 50:50 (vol:vol) dilution of medium in distilled water (resulting in a 325 to 162.5 mOsm l⁻¹ hypoosmotic shock). Cells were kept in focus using hardware autofocus (perfect focus, Nikon).

Image processing

Images were processed using Fiji⁷⁰ and MATLAB 2020b (MathWorks) using custom codes that are available on request. For visualization purposes, the PopRed lookup table from the J. Manton collection (https://github.com/jdmanton/ImageJ_LUTs) was applied to most monochrome images after the dynamic range was adjusted between minimum and maximum grey values of each image (note that the dynamic range was not kept identical between images when presenting different conditions). Figures were assembled in Adobe Illustrator 2021. Videos were edited in Adobe Premiere 2021.

Where appropriate, spatial drift during acquisition of videos was corrected using a custom GPU-accelerated registration code based on cross-correlation between successive frames. For representation purposes, a wavelet ‘à trous’ denoising filter was applied to Extended Data Fig. 8c (custom GPU-accelerated MATLAB port of a code originally developed by F. Cordeliere for the Improve Kymo ImageJ plugin⁸¹). The raw images were averaged with the filtered video. Both codes are available at our GitHub page (https://github.com/deriverylab), as well as the codes for quantification of protein condensation and nuclear segmentation described below.

Quantification of condensation in vitro

Our pipeline to objectively assess the degree of condensation in microscopy images for in vitro assays (Fig. 4 and Extended Data Fig. 9) is presented in Extended Data Fig. 9a,b.

In brief, we computed the fast Fourier transform (FFT) of the source images (all 600 × 600 pixels, resulting in a 1,024 × 1,024 Fourier-transformed image), and the fraction of the power spectrum found in rings of increasing diameter (3 px increment in Fourier space) was then measured and plotted onto a log scale (Extended Data Fig. 9a). As can be seen in Extended Data Fig. 9a, as the fraction of condensed signal increases in the source image, a larger fraction of the power spectrum is found in rings of larger diameter (in other words, high spatial frequency components increasingly appear in the image). This was highly consistent for various samples of a given condition (Extended Data Fig. 9a (bottom left)). As the first two rings contained mostly the image background (low spatial frequencies), we set a threshold of 6 px in the Fourier space and defined the condensation ratio as the fraction of the total power spectrum in the high-frequency portion of the spectrum (that is, >6 px in Fourier space). This 6 px threshold was efficient at separating the condensed signal from the non-condensed signal and background (Extended Data Fig. 9b,c).

Quantification of condensation in live cells

Our pipeline to objectively assess the degree of condensation in microscopy images for live-cell assays is presented in Extended Data Fig. 6a. As the condensation in cells had to be measured in specific regions of interests (ROIs), rather than the full image, we could not use the condensation ratio described above, as that calculation is done in the Fourier space, not the real space, in which ROIs can be made.

In brief, raw images were processed for homogenous background subtraction, then FFT was computed, then a high-pass filter was applied using a circular mask followed by inverse FFT. This mask was kept constant for all images (all source data were cropped to have the same size). We next computed the granulosity ratio as the ratio between the s.d. and the mean of the signal in the high-pass-filtered image. As the granulosity index is measured in the real space, which can be done in specific ROIs (for example, the nucleus). The automated nucleus segmentation using an ad hoc neural network that we developed for this report is described in a dedicated section below.

Importantly, we confirmed that both the condensation ratio (measured in the Fourier space) and the granulosity index (measured in the real space) are giving quantitatively similar results using input images from GFP condensation in vitro (Extended Data Fig. 6b,c; for the granulosity index, the whole image was used as the ROI).

Quantification of cortical PIP₂ recruitment

For the quantification of the yeast thermosensitive mutant (Fig. 1d,e and Extended Data Fig. 2), semi-automated analysis was performed using custom scripts in Fiji. z-Stacks were acquired by SDCM at a stable temperature (Δz = 0.5 μm, Δt = 1 min). Intensity thresholds were applied manually to each cell independently, and the thresholded region was converted to a region of interest. These regions were smoothed and filled, and the peripheral-most 3 px were segmented out and used to measure membrane-bound GFP intensity. The region within this was used for the cytosolic intensity. The ratio of the median membrane-bound and cytosolic signal, after homogenous background subtraction, was then computed for each cell.

Neural network for nuclear segmentation

For automated nuclear segmentation, SH-SY5Y cells transiently expressing FusLC–GFP were stained with Hoechst (5 min, 13.6 μM) and imaged using SDCM. Maximum-intensity z-projections were then thresholded on the basis of the Hoechst signal to establish the initial nuclear ground truth. After initial training of the network on these 132 FusLC–GFP images, predictions were manually refined to increase the training dataset size to 598 images. This dataset was then used to retrain the network. Image augmentation was performed on the training dataset, by splitting the images into four overlapping tiles, flipping or rotating each tile (90°, 180° and 270°) and duplicating images and randomly resetting the contrast of these duplicate images.

The network architecture used is a residual convolutional U-net⁸² and is depicted in Extended Data Fig. 7a,b. It comprises a residual convolutional network⁸³, with concatenation steps between down- and up-sampling layers of the network. Each layer of the network comprised two residual blocks, each of which comprise two 2D 3 × 3 convolutional layers with batch normalization and a ReLU activation function. 2 × 2 max pooling layers were used to reduce dimensionality of the network, before subsequent upsampling of the image with transposed 2D convolutional layers. A final 1 × 1 2D convolutional layer and sigmoid activation outputted a probability of each pixel being with a nucleus, and these probabilities were thresholded to output a binary segmentation. The dropout rate was set at 0.25 for the first and convolutional/residual block and the last deconvolutional/residual block, and 0.5 for all other blocks, and a batch size of 8 images was used. A binary cross entropy loss was used, with Adam optimizer and a learning rate of 1 × 10⁻⁴. The final training took approximately 2 days on two Nvidia Titan V GPUs. The final network was used for the automatic segmentation of all cellular condensation data imaged with a ×100 objective (Extended Data Fig. 7e). For Fig. 3d, which was imaged using a ×60 objective, and Extended Data Fig. 8a, nuclei were manually segmented using Fiji.

Calcium signalling

Fluorescence analysis was performed by measuring the intensity of the Fluo-4 probe on all cells in the field of view using the Zeiss software Aim v.4.2.

Monopolar spindle quantification

For quantification of monopolar spindle occurrence (Extended Data Fig. 2g–i), maximum projections of each stack were generated using Fiji. The number of monopolar and bipolar spindles was manually scored within the time frame of 60–180 min of each video (6–10 videos were acquired in each condition per experiment).

Proteomics and phosphoproteomics

Culture conditions

Primary mouse lung fibroblasts were grown to confluence in 10 cm dishes in DMEM supplemented 10% Hyclone III and penicillin–streptomycin (100 U ml⁻¹), and in duplicates transferred to the following conditions for two weeks: control (37 °C, 350 mOsm l⁻¹ medium), hypoosmotic (37 °C, 450 mOsm l⁻¹ medium), hyperosmotic (37 °C, 250 mOsm l⁻¹ medium), low temperature (32 °C, 350 mOsm l⁻¹ medium) and high temperature (40 °C, 350 mOsm l⁻¹ medium), with the medium changed every 3–4 days. Medium osmolarity was adjusted with water or sucrose. Cells were washed twice in ice-cold PBS and then lysed at room temperature in 1 ml freshly prepared lysis buffer (8 M urea, 20 mM Tris, pH 8) for 20 min. After lysis, the dishes were scraped and the cell lysates were immediately flash-frozen in liquid nitrogen and then stored at −80 °C. All of the samples were simultaneously defrosted and sonicated for 2 min and the protein concentration was then measured using the BCA assay (Pierce).

MS analysis

Enzymatic digestion

Each sample (500 µg) was reduced with 5 mM DTT at 56 °C for 30 min and then alkylated with 10 mM iodoacetamide in the dark at room temperature for 30 min. They were then digested using MS-grade Lys-C (Promega) at a protein:Lys-C ratio of 100:1 (w/w) for 4 h at 25 °C. Next, the samples were diluted to 1.5 M urea using 20 mM HEPES (pH 8.5) and digested at 30 °C overnight with trypsin (Promega) at a ratio of 70:1 (w/w). Digestion was quenched by the addition of formic acid (FA) to a final concentration of 1%. Any precipitates were removed by centrifugation at 16,000g for 10 min. The supernatants were desalted using home-made C18 stage tips containing 3 M Empore extraction disks (Sigma-Aldrich) and 8 mg of Poros R3 resin (Thermo Fisher Scientific). Bound peptides were eluted with 30–80% acetonitrile (MeCN) in 0.5% formic acid and lyophilized.

TMT peptide labelling

The lyophilized peptides from each sample were resuspended in 100 µl of 2.5% MeCN and 250 mM triethylammonium bicarbonate. Then, 60 µl (1.2 mg) of each TMT10plex reagent (Thermo Fisher Scientific), reconstituted in anhydrous MeCN according to the manufacturer’s instructions, was added. Peptides from each timepoint were labelled with a distinct TMT tag for 60 min at room temperature. The labelling reaction was quenched by incubation with 11 µl 5% hydroxylamine for 30 min. The set of 10 labelled peptides (5 conditions, duplicates) were combined into a single sample and partially dried to remove MeCN in a SpeedVac (Savant). The sample was then acidified and centrifuged at 16,000g for 10 min. The supernatant was desalted using Sep-Pak Plus Short tC18 cartridges (Waters). Bound peptides were eluted with 60% acetonitrile in 0.5% acetic acid and lyophilized.

Titanium dioxide enrichment of phosphopeptides

Phosphopeptides were enriched using TiO₂ titansphere-chromatography (GL Science). Lyophilized peptides were resolubilized in 50% MeCN containing 2 M lactic acid (loading buffer) and incubated with TiO₂ beads (1:4, peptides:TiO₂(w/w)) that were prewashed with loading buffer. After 1 h, the TiO₂ beads were centrifuged at 10,000g for 2 min, the supernatant was added to fresh TiO₂ beads for a second round of enrichment and the procedure was repeated for a third time. After incubation, the TiO₂ beads with enriched phosphopeptides were loaded onto C8 stage tips (3 M Empore) and washed twice with loading buffer and once with 50% MeCN, 0.1%TFA. Phosphopeptides were eluted sequentially with 0.4 M ammonia solution, 30% MeCN in 0.4 M ammonia solution and 50% MeCN in 0.1% TFA. The eluates were acidified with formic acid, and partially dried down using a SpeedVac (Savant). The samples were then desalted using C18 Stage tips (3 M Empore) and lyophilized.

Basic pH reverse-phase HPLC fractionation

The TMT-labelled peptides and phosphopeptides were processed for off-line high performance liquid chromatography (HPLC) fractionation, using an XBridge BEH130 C18, 5 μm, 2.1 mm x 150 mm column with an XBridge BEH C18 5 μm Van Guard cartridge (Waters), connected to an Ultimate 3000 Nano/Capillary LC System (Dionex). Peptide mixtures were resolubilized in solvent A (5% MeCN, 95% 10 mM ammonium bicarbonate, pH 8) and separated with a gradient of 1–90% solvent B (90% MeCN, 10% 10 mM ammonium bicarbonate, pH 8) over 60 min at a flow rate of 250 μl min⁻¹. Eluted peptides were collected at 1 min per fraction, combined into 18 fractions for proteomics experiments and 14 fractions for phosphoproteomics and lyophilized. For proteomic experiments only, the fractions were desalted using C18 stage tips and partially dried down by vacuum centrifugation before analysis using liquid chromatography (LC)–MS/MS.

LC–MS/MS analysis

The fractionated peptides were analysed by LC–MS/MS using the fully automated Ultimate 3000 RSLC nano System (Thermo Fisher Scientific) fitted with a 100 μm × 2 cm PepMap100 C18 nano trap column and a 75 μm × 25 cm nanoEase M/Z HSS C18 T3 column (Waters). The samples were separated using a binary gradient consisting of buffer A (2% MeCN, 0.1% formic acid) and buffer B (80% MeCN, 0.1% formic acid), and eluted at 300 nl min⁻¹ with an acetonitrile gradient. The outlet of the nano column was directly interfaced through a nanospray ion source to the Q Exactive Plus mass spectrometer (Thermo Fisher Scientific). The mass spectrometer was operated in standard data-dependent mode, performing a MS full-scan in the m/z range of 380–1,600, with a resolution of 70,000. This was followed by MS2 acquisitions of the 15 most intense ions at a resolution of 35,000 and normalized collision energy of 33%. MS target values of 3 × 10⁶ and MS2 target values of 1 × 10⁵ were used. The isolation window of precursor ion was set at 0.7 Da and sequenced peptides were excluded for 40 s.

Spectral processing and peptide and protein identification

The acquired raw files from LC–MS/MS were processed using MaxQuant⁸⁴ with the integrated Andromeda search engine (v.1.6.6.0). MS/MS spectra were quantified with reporter-ion MS2 from TMT10plex experiments and searched against the Mus musculus UniProt Fasta database (March 2019). Carbamidomethylation of cysteines was set as fixed modification, whereas methionine oxidation, N-terminal acetylation and phosphorylation (STY) (for the phosphoproteomics group only) were set as variable modifications. Protein quantification requirements were set at 1 unique and razor peptide. In the identification tab, second peptides and match between runs were not selected. The other parameters in MaxQuant were set to the default values.

The MaxQuant output file was then processed using Perseus (v1.6.6.0). Reporter-ion intensities for the protein group table were uploaded to Perseus. The data were filtered: identifications from the reverse database were removed, only identified by site, potential contaminants were removed. Then, all columns with an intensity of less than or equal to zero were converted to ‘not a number’ (NAN) and exported. The MaxQuant output file with phospho (STY) sites table was also processed using Perseus software (v.1.6.6.0). The data were filtered: identifications from the reverse database, potential contaminants were removed and we only considered phosphopeptides with localization probability ≥0.75. Then all columns with an intensity of less than or equal to zero were converted to NAN and exported.

MS data analysis

Data processing was performed in R v.3.6.1 using R Studio v.1.2. For proteomics data, reporter-ion intensities were normalized for input (equalizing total intensities) and log₂-transformed. To determine proteins that change significantly with temperature or external osmolarity, linear model fitting with the empirical Bayes method was performed using the LIMMA package⁸⁵, separately for temperature and osmolarity data, with predictors treated as continuous variables. A Benjamini–Hochberg-adjusted P value of 0.05 was used as the significance threshold. The GeneOverlap R package⁸⁶ was used to aid determination of overlap significance between different protein groups. Gene Ontology enrichment was performed using the GOrilla online tool⁸⁷ with two unranked gene lists (overlaps versus all detected proteins). A list of phase-separating proteins was taken from PhaSepDB (v.1)⁸⁸, using the list of proteins annotated as phase-separating and/or belonging to different MLOs from high-throughput screens.

For phosphoproteomics data, reporter-ion intensities for each phosphosites were normalized to its respective protein abundance (obtained from original proteomics data). LIMMA analysis was performed as described above. D2P2 database⁸⁹ was queried for all detected phosphoproteins to determine their predicted IDRs, as a consensus between the different predictor algorithms used in the database; the position of each phosphosite was used to determine whether or not it is predicted to be in an IDR. A proportion z-test was then used to determine whether phopshopeptides that change significantly with external osmolarity and/or temperature have the same or different level of IDR phosphorylation compared with the background. Prediction of kinases that might phosphorylate the detected phosphopeptides was performed by counting predicted motifs for a panel of 25 kinases present in the PHOSIDA database⁹⁰.

Statistics and reproducibility

Statistical analyses were performed using GraphPad Prism v.9.4.0 (673). Data are presented as mean ± s.e.m. unless otherwise stated. No randomization methods were used in this study. No blind experiments were conducted in this study. Normality of variables was verified using Kolmogorov–Smirnov tests. Homoscedasticity of variables was always verified when conducting parametric tests. Unless stated otherwise, no adjustments for multiple comparisons were performed (ANOVA tests were performed when comparing more than two samples, with adequate adjusted post hoc tests indicated in their respective figure legends). Unless stated otherwise, n values indicate independent biological replicates. The following statistical tests were two-sided: Figs. 1a,f,g, 2c, 3b,d and 4c,e,h and Extended Data Figs. 2i, 3d,f, 4b,f,h, 6c, 8a,b,d,h, 9g,i and 10f,l. Unless stated otherwise, P values are indicated by asterisks: *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001; NS, P > 0.05.

Details on number of repeats when representative images/panels are shown are as follows: Fig. 1d: n = 30 (WT) and n = 16 (mss4^ts) (quantified in Fig. 1e); Fig. 3a: n = 7 (GFP) and n = 12 (FusLC–GFP) (quantified in Extended Data Fig. 8a); Fig. 3e: n = 1 (quantified in Fig. 3d on n = 10–42 cells per condition); Fig. 4a: n = 10 per condition; Fig. 4b: n = 10 per condition (quantified in Fig. 4c); Fig. 4d: n = 10 per condition (quantified in Fig. 4e); Fig. 4g: n = 10 per condition (quantified in Fig. 4h); Extended Data Fig. 2b: n = 30 (32 °C) and n = 41 (39 °C) (quantified in Extended Data Fig. 2c); Extended Data Fig. 2e: n = 13 for each sample (quantified in Extended Data Fig. 2f); Extended Data Fig. 2h: respective sample size is indicated in the quantification in Extended Data Fig. 2i; Extended Data Fig. 3b: n = 20 (control), n = 20 (+100 mOsm l⁻¹) and n = 18 (−17 °C) (Extended Data Fig. 3d); Extended Data Fig. 3g: n = 9 fields of view analysed per condition. Experiment representative of 3 biological replicates; Extended Data Fig. 4e: n = 9 fields of view analysed per condition. Experiment representative of 2 biological replicates; Extended Data Fig. 4g: n = 15 cells per condition (quantified in Extended Data Fig. 5h); Extended Data Fig. 6b: n = 10 (quantified in Extended Data Fig. 6c); Extended Data Fig. 6d: n = 1 (shown only for illustrative purpose); Extended Data Fig. 7e: n = 46 images in the validation set (best model accuracy = 0.996; loss, 0.065); Extended Data Fig. 8c: n = 9–23 (quantified in Extended Data Fig. 8d); Extended Data Fig. 8e: n = 8 (quantified in Extended Data Fig. 8f); Extended Data Fig. 8g: n = 10–11 (quantified in Extended Data Fig. 8h); Extended Data Fig. 8i: n = 9–12 (Staurosporine) and n = 7–9 (CalyculinA) (quantified in Extended Data Fig. 8j); Extended Data Fig. 8k: n = 2 (both independent biological repeats shown); Extended Data Fig. 9c: n = 10 images per condition; Extended Data Fig. 9d: n = 10 images per condition; Extended Data Fig. 9e: n = 10 images per condition (quantified in Fig. 4c); Extended Data Fig. 9f: n = 10 images per condition (quantified in Extended Data Fig. 10g), note that this is the same quantification as in Extended Data Fig. 6c (left panel); Extended Data Fig. 9h: n = 8 (GFP) and n = 10 (GFP + GBP) (quantified in Extended Data Fig. 9i); Extended Data Fig. 10i: n = 2; Extended Data Fig. 11a: dataset comprising n = 17 cells in 2 independent experiments; and Extended Data Fig. 11b: n = 4 (WNK1) and n = 2 (GCN2).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.