Mis-splicing of a neuronal microexon promotes CPEB4 aggregation in ASD

Animals

Cpeb4 KO mice⁴⁵ and conditional transgenic mice overexpressing the human CPEB4 isoform that lacks exon 4 (TgCPEB4Δ4)² both in a C57BL/6J background were used. All mice were bred and housed in the CBMSO animal facility. Mice were grouped four per cage with food and water available ad libitum and maintained in a temperature-controlled environment on a 12–12 h light–dark cycle with light onset at 8:00 and a relative humidity of 55 ± 10%. Animal housing and maintenance protocols followed local authority guidelines. Animal experiments were performed under protocols approved by the CBMSO Animal Care and Utilization Committee (Comité de Ética de Experimentación Animal del CBMSO, CEEA-CBMSO) and Comunidad de Madrid (PROEX 247.1/20).

Generation of mEGFP–CPEB4 mice

A synthetic sequence consisting of the mEGFP linker sequence flanked by short regions of 5′ homologous (238 bp) and 3′ homologous (99 bp) DNA was obtained (Twist Bioscience). PCR was carried out on the synthetic sequence using the following primers: Fw ssDNA-mGFP–CPEB4 (phosphorylated) tacttcaagcaaacatatttgagatacagggga; Rv ssDNA-mGFP–CPEB4 (thiol-protected) GGTGATGGTGTGGAGGCTGC. Single-stranded DNA (ssDNA) was generated from double-stranded DNA by lambda exonuclease digestion of the phosphorylated strand, followed by gel purification and column extraction. Animals were generated by electroporation of isolated mouse zygotes with ssDNA combined with Cas9 protein and guide/tracr RNA ribonuclear protein complexes (guide; C45gRNA ATCCTAAAAATAATAAATGG). The correct integration of the knock-in cassette was confirmed by PCR and sequencing of the region. The resulting positive mice were crossed with C57BL6/J mice to confirm germline transmission. The offspring were maintained in a C57BL/6J background, and routine genotyping was performed by PCR using the following genotyping primers: 5′-ACGTAGGGTGATAAGCTGTGAT–3′ (Fw) and 5′-AGGGTCTTGTTGTTCTTGCTGT-3′ (Rv). Mice were maintained in a specific pathogen-free facility with a 12–12 h light–dark cycle at 21 ± 1 °C at a relative humidity of 55 ± 10% and given ad libitum access to standard diet and water. Animal handling and all experimental protocols were approved by the Animal Ethics Committee at the Barcelona Science Park and by the Government of Catalonia.

Mouse mEGFP–CPEB4 striatal neuron extraction and culture

mEGFP–CPEB4 mice over 6 weeks of age were crossed in timed matings. Females were weighed weekly to monitor gestation progression. Females with an increment over 3 g up to 18 days after a positive plug were euthanized and embryos were collected at embryonic day 18.5 in cold buffer containing 1× HBSS, 10 mM glucose and 10 mM HEPES. A tail sample was also collected for embryo genotyping. Brains were dissected in the aforementioned buffer on an ice-cold plate, and the striatum was extracted and chopped. Samples were centrifuged and digested with a previously heated solution containing 1× HBSS, 10 mM glucose, 10 mM HEPES, 12 U ml^–1 papain (Worthington LS003180) and 5 mM l-cysteine for 15 min at 37 °C. Samples were then disaggregated in a buffer containing 1× DMEM/F-12, 2 mM glutamine, 1 mM sodium pyruvate, 20 mM glucose and 10% inactivated horse serum. Cells were seeded at a confluence of 25,000 cells per well in µ-Slide 8-well ibiTreat imaging plates (Ibidi, 80826) previously coated with poly-d-lysine. Cells were then incubated at 37 °C for 1 h. After this time, medium was exchanged with previously tempered medium containing 1× Neurobasal (Gibco, 21103049), 1× B27 with vitamin A (Gibco, 17504044), 2 mM glutamine and 0.5% penicillin–streptomycin (PS). Medium was refreshed every 2–3 days. Neurons were considered differentiated after 7 days of culture. After genotyping, homozygous mEGFP–CPEB4 mice and wild-type littermates were selected for imaging. When specified, cell depolarization was induced by the addition of 50 mM KCl with 1:3 medium dilution. Neurons were maintained in culture for up to 14 days.

mEGFP–CPEB4 distribution in neurons

Primary striatal neurons from mEGFP–CPEB4 mice were imaged at 7 days of differentiation using a LIPSI spinning disk microscope (Nikon). Image acquisition was performed using a fully incubated, high-content, high-speed screening LIPSI platform (Nikon) equipped with an Eclipse Ti2 inverted microscope and a Yokogawa W1 confocal spinning disk unit. The spinning disk unit with an Apo LWD ×40 water lens of 1.15 numerical aperture, and a 488 nm (20%) laser was used for acquisition on a Prime BSI Photometrics sCMOS camera. NIS Elements AR (v.5.30.05) software was used for acquisition, and Fiji/ImageJ software was used to adjust images for visualization.

mEGFP–CPEB4 neuronal stimulation with NMDA

Primary striatal neurons from mEGFP–CPEB4 mice were imaged at 14–21 days of differentiation, and where specified, neuron stimulation was induced by the addition of 20 µM NMDA (Tocris, 0114), a selective NMDA receptor agonist. Stimulated neurons were imaged using a fully incubated Zeiss Elyra PS1 LSM 880 confocal microscope with a Plan ApoChromat ×63/1.2 Imm corr oil objective. A 488 nm (50%) laser was used for acquisition on a Prime BSI Photometrics sCMOS camera. Images were captured every 15 min over the recording period. Zen Elements AR (v.5.30.05) software was used for acquisition, and Fiji/ImageJ software was used for image quantification and to adjust images for visualization. A tailor-made macro applying an intensity threshold was used to accurately segment cytoplasmic condensates and the whole cell for each time frame. The condensed fraction per frame was obtained as the sum of areas of condensates divided by the cell area. For representation, the values were normalized to that measured at the end of the stimulation period.

nCPEB4 extraction from mouse brains

To extract nCPEB4 from the brains of 6-month-old control mice, Cpeb4 KO mice and TgCPEB4Δ4 mice, around 500 mg of tissue was first collected and snap-frozen in liquid nitrogen. Each sample was homogenized in 5 ml lysis buffer (50 mM Tris, pH 7.7, 5% glycerol, 0.1% Triton X-100, 1% NP-40, 50 mM NaCl, 50 mM imidazole and Pierce protease inhibitor, EDTA-free) using a Polytron homogenizer and rotated for 30 min at 4 °C. The homogenate was moved to high-speed PPCO centrifuge tubes and centrifuged at 48,000g at 4 °C for 20 min. After this, the supernatant was retained while the resultant pellet was dissolved in 4 ml lysis buffer and homogenized further using the Polytron homogenizer with the same protocol. This process was repeated 3 times (with 1 ml reduction of lysis buffer after each round of homogenization) to maximize the extraction of nCPEB4. To further clarify the combined supernatants, they were filtered through a Miracloth membrane (Millipore) to remove lipids and then passed through a 0.45 μm filter. Exploiting the histidine-rich regions present in the sequence of nCPEB4 (²³RFHPHLQPPHHHQN³⁶ and ²²⁹LSQHHPHHPHFQHHHSQHQQ²⁴⁸), a 2-elution step Ni²⁺-affinity chromatography was carried out using a Histrap HP 5 ml (GE Healthcare) on a FPLC apparatus (ÄKTA Pure, GE Healthcare). The combined supernatants were injected into a column pre-equilibrated with binding buffer consisting of 50 mM Tris, pH 7.7, 50 mM NaCl and 50 mM imidazole. The bound fraction was initially washed in a high-salt buffer consisting of 50 mM Tris, pH 7.7, 1 M NaCl and 50 mM imidazole. This step is crucial as it removes a high-molecular-weight nonspecific binder that was detectable by western blotting (WB), even in the Cpeb4 KO mice. The removal of this nonspecific binder is important to ensure the integrity of subsequent analyses. Once the nonspecific binder was completely removed, nCPEB4 was eluted using a second buffer containing 50 mM Tris, pH 7.7, 50 mM NaCl and 500 mM imidazole. The non-bound, washed and eluted fractions were then analysed by WB using Bis-Tris 4–12% gradient gels. At this point, fractions containing nCPEB4, verified by WB in SDS–PAGE, were used for proteinase K digestion and SDS-resistance analysis using 1.5% SDD–AGE as previously described⁴⁶. Here, the non-boiling proteins were transferred to a nitrocellulose membrane by capillary methods and probed with an anti-CPEB4 antibody. Samples identified as CPEB4 monomers and aggregates were selected for subsequent seeded aggregation assay. Additionally, to quantify the amount of CPEB4 aggregates, the same samples were injected into a Superdex 200 Increase 10/300 GL (GE Healthcare) column pre-equilibrated with 1× PBS, pH 7.5. The eluted fractions from the gel filtration chromatography were combined every 4 fractions (2 ml) and concentrated into 100 µl using a Pierce concentrator, PES, 30 K MWCO, 0.5 ml, to be ultimately analysed using 1.5% SDD–AGE. For all the blots described here, including SDS–PAGE, immunodot blots and SDD–AGE, a 1:2,000 dilution of polyclonal rabbit anti-CPEB4 antibody (Abcam, ab224162) and a 1:5,000 dilution of HRP-linked anti-rabbit IgG antibody (Cell Signaling Technology, CST-7074S) were used.

Proteinase K digestion

About 400 ng of total protein, estimated from BCA assays, containing endogenous nCPEB4 extracted from the brains of 6-month-old control mice, Cpeb4 KO mice and TgCPEB4Δ4 mice were digested with 0.1, 0.5, 1, 5 and 10 ng of proteinase K for 2 min at 37 °C. Proteinase K activity was stopped by heating the samples to 75 °C. Next, 2 μl of the enzyme-treated reaction mixture was manually applied to a nitrocellulose membrane. The membrane was blocked in 5% milk in TBS-T buffer and probed with anti-CPEB4 antibody (Abcam, ab224162).

Seeded aggregation assay

A concentration of 1% w/w of the seed, 40 ng of total protein containing nCPEB4 aggregates extracted from TgCPEB4Δ4 mouse brains, was incubated with the substrate, which consisted of 4 μg total protein containing soluble nCPEB4 extracted from wild-type (WT) mouse brains. Unless concentrated 100-fold, the seed used was not detectable by WB. The seeding reaction was carried out at 4 °C for 24 h during the time course experiment in 50 mM Tris, pH 7.7, and 50 mM NaCl. The seeded reactions, which were non-boiled, were analysed using 1.5% SDD–AGE as previously described⁴⁶. The proteins were transferred to a nitrocellulose membrane by capillary methods and probed with anti-CPEB4 antibody (Abcam, ab224162) to follow the aggregation of WT nCPEB4 in a time-dependent manner.

Proteostat staining and CPEB4 immunofluorescence

Six-week-old TgCPEB4Δ4 mice (n = 4) and control littermates (n = 3) were anaesthetized by an intraperitoneal injection of pentobarbital and then transcardially perfused with PBS. Brains were immediately removed and each hemisphere placed in 4% paraformaldehyde overnight at 4 °C, followed by 3 PBS washes (10 min each) and then immersed in 30% sucrose in PBS for 72 h at 4 °C and them included in optimum cutting temperature compound (Tissue-Tek, Sakura Finetek Europe, 4583) and immediately frozen. Samples were stored at −80 °C until use.

Brain hemispheres were cut sagittally at 30 µm on a cryostat (Thermo Scientific), and sections were stored (free floating) in glycol-containing buffer (30% glycerol and 30% ethylene glycol in 0.02 M PB) at −20 °C.

For staining, sections (2 per mouse) were washed in PBS to eliminate the cryoprotective buffer and permeabilized in 0.2% Triton X-100 for 30 min at room temperature, and then stained with the dye Proteostat (Enzo51035-K100; 1:2,000) for 15 min at room temperature followed by 2 PBS washes (10 min each) and then 1% acetic acid for 30 min at room temperature, followed by 3 PBS washes (10 min each). For CPEB4 immunofluorescence, sections were immersed in blocking solution (2% NGS, 1% BSA and 0.2% Triton X-100 in PBS) for 1 h at room temperature and then incubated overnight at 4 °C with anti-CPEB4 primary monoclonal antibody (1:1,000, mouse monoclonal, homemade, ERE149C) in blocking solution. After 3 PBS washes (10 min each), sections were incubated with Alexa 488 donkey anti-mouse secondary antibody (1:500, Thermo Fisher, A-21202) for 1 h followed by 3 PBS washes (10 min each) and, finally, nuclei were stained by incubating with DAPI (1:10,000 in PBS, Merck) followed by 3 PBS washes (10 min each) and mounted with Prolong medium (Life Technologies).

Images of the striatum were obtained with a vertical Axio Observer.Z1/7 laser scanning microscope (LSM 800, Carl Zeiss) at ×63 magnification with ×2 optical zoom and analysed by performing z stacks (11 optical sections with a thickness of 1 μm, spanning 6.6 μm on the z axis). Sequential scanning mode was used to avoid crosstalk.

Medium-sized spiny neurons were distinguished by the morphology and size of the nucleus, and fields were selected to typically include 4–8 medium-sized spiny neurons. The number of Protesotat and CPEB4 double-positive foci was manually counted per cell fully included within the z stack. Typically, between 16 and 20 cells were analysed per mouse from a total of 50 control and 73 TgCPEB4Δ4 neurons.

Significance for differences in the number of positive foci between control mice and TgCPEB4Δ4 mice was assessed using a generalized linear mixed model (family = Poisson(link = ‘identity’)) with mouse as the random effect (Supplementary Methods).

Plasmids for expression in N2a cells

Human nCPEB4 (UniProt identifier Q17RY0-1) FL open reading frame (ORF) was cloned into a pBSK vector. The me4 sequence (nucleotides 1258–1281) was deleted by PCR on a pBSK-nCPEB4 plasmid using Gibson assembly master mix (New England Biolabs, E2611S) following the manufacturer’s instructions. Mutagenesis of nCPEB4 phosphorylation sites was performed using a QuikChange Lightning Multi Site-Directed Mutagenesis kit (Agilent Technologies, 210513), with oligonucleotides purchased from Sigma-Aldrich, following the manufacturer’s instructions. ΔHC mutants were generated by PCR mutagenesis on a pBSK-nCPEB4 plasmid, with oligonucleotides purchased from Sigma-Aldrich. For cell transfection, nCPEB4 ORF, FL, NTD and mutants were cloned into pPEU4 and pPEU5 vectors, which contain a C-terminal eGFP or mCherry tag, respectively, by In-Fusion (BD Clontech) cloning reaction⁴⁷. For BioID, xCPEB4 or BirA ORF was cloned into a pBSK vector. me4 was added to the xCPEB4 sequence by PCR on a pBSK-xCPEB4 plasmid using Gibson assembly master mix (New England Biolabs, E2611S) and following the manufacturer’s instructions. A MYC tag and BirA ORF were added at the N terminus of pBSK-xCPEB4 plasmid. For competition experiments, xCPEB1 RRMZZ and xCPEB4 RRM domains were cloned in pBSK, and a HA tag was added at the N terminus. pBSK-Emi2 3′ UTR was obtained from a previous study⁴⁸.

N2a cell culture, differentiation and DNA transient transfection

N2a cells were grown in DMEM with 10% FBS, 1% PS and 2 mM l-glutamine for maintenance. For fixed-cell imaging, cells were seeded on 6-well plates with 12-mm-diameter poly-lysine-coated glass coverslips (Marienfeld Superior). For live-cell imaging, cells were seeded on µ-Slide 8-well ibiTreat plates. For differentiation, medium was exchanged with DMEM with 0.5% FBS, 1% PS, 2 mM l-glutamine and 1 µM retinoic acid and cells were grown for 48 h. They were then transfected at 60% confluence with 1.25 µg DNA using Lipofectamine LTX and Plus reagent (Thermo Fisher, 15338100) following the manufacturer’s protocol. When specified, N2a depolarization was induced as described for striatal neurons, specified in the section ‘Mouse mEGFP–CPEB4 striatal neuron extraction and culture’.

N2a cell line characterization

N2a cells, like neurons, express CPEB4 variants including and excluding me4 (nCPEB4 and nCPEB4Δ4, respectively), independently of their differentiation status (Extended Data Fig. 1e). By contrast, cell lines of non-neural origin only express nCPEB4Δ4. Inclusion of me4 in N2a cells correlates with the expression of the splicing factor SRRM4 but not with that of RBFOX1 (Extended Data Fig. 1f). Depletion of SRRM4 in N2a cells decreases the inclusion of me4, whereas overexpression of SRRM4 in the non-neuronal cell line 293T forces its inclusion^5,49 (Extended Data Fig. 1g). N2a cells, therefore, recapitulate the neuron-specific regulation of CPEB4 alternative splicing.

nCPEB4–GFP distribution in N2a cells

Twenty-four hours after transfection, N2a cells were fixed with 4% paraformaldehyde (Aname, 15710) in PBS for 10 min at room temperature. They were then washed with PBS and incubated with 0.5 µg µl^–1 DAPI (Sigma) for 15 min. Coverslips were rinsed with PBS and mounted on a glass slide with Prolong Gold Antifade mountant (P36934, Invitrogen). Image acquisition was performed with a Leica SP5 confocal microscope (Leica Microsystems), and z series stacks were acquired at 1,024 × 1,024 pixels using a ×63/1.4 numerical aperture oil immersion objective with a zoom factor of 2. Argon 488 nm (20%) and diode 405 nm (10%) lasers were used. Hybrid detectors for GFP (500–550 nm with 33% gain) and DAPI (415–480 nm, 33% gain) were used for acquisition. LAS AF Leica software was used to acquire 10–20 z stack slices per cell with a z step size of 0.5 µm. Fiji/ImageJ software was used to perform the image analysis. A tailor-made macro using BioVoxxel Toolbox and 3D object counter plug-ins was used to accurately segment and obtain the number and volume of foci per cell.

Live-cell imaging of GFP-tagged CPEB4 variants in N2a cells

Live imaging of overexpressed nCPEB4–GFP variants in N2a cells was performed 20 h after transfection, whereas primary striatal neurons from mEGFP–CPEB4 mice were imaged at 7 days of differentiation. For both types of cells, image acquisition was performed using a spinning disk microscope (Andor Revolution xD, Andor). A total of 24 images were taken per experiment (4 before the addition of the stimulus and 20 after), with 13 z stacks at 512 × 512 pixels of format resolution. Images were acquired with a step size of 0.5 μm. For acquisition, the typical frame rate was adjusted to 5 images per s at 50 ms integration time of the EMCCD camera (Andor). An argon 488 nm laser (20%) was used for acquisition with a 1.4 numerical aperture/×60 oil immersion objective. Fiji/ImageJ software was used to obtain a z projection of the z stacks and subsequent concatenation of images. The obtained time-lapse images were subsequently used for manual quantification of nCPEB4 dissolution events. Cells were manually classified into two categories depending on the existence of cytoplasmic foci at t = 60 min: cells with remaining foci or cells without. For nCPEB4 and nCPEB4Δ4 FL comparison, the percentage of cells with remaining cytoplasmic foci after the depolarizing stimuli (t = 60 min) was calculated from a pool of 7 experiments. Unless specified, the percentage of cells with remaining cytoplasmic foci after the depolarizing stimuli (t = 60 min) was calculated per each experiment. When specified, blind analysis and classification were performed independently by a group of four different people from a pool of experiments.

FRAP in N2a cells

A spinning disk microscope from Andor, equipped with a FRAPPA module, was used for FRAP experiments. A total of 350 images were taken per experiment (50 images before the bleaching and 300 after) at 512 × 512 pixels. The typical frame rate was set to the fastest (88 ms) with an exposure time of 50 ms on an EMCCD camera. An AOTF 488 nm laser (20%) was used for acquisition, and 50% laser intensity was set for bleaching in 2 repeats with a dwell time of 40 ms. Fiji/ImageJ software was used for FRAP analysis. Three regions of interest (ROIs) were defined per video: background, cell and bleaching area. The mean fluorescence intensity was obtained for the 3 ROIs for all 350 frames, and the output was exported in tabular format. Outputs were then entered on the easyFRAP website⁵⁰. Full-scale normalization was selected, ‘initial values to discard’ was set to 20 and the curves obtained were fitted to a single exponential model. Fluorescence recovery curves, mobile fraction and half time of recovery were obtained for each experimental condition.

Mapping of nCPEB4 post-translational modified sites by mass spectrometry

Overexpressed nCPEB4–GFP and nCPEB4Δ4–GFP were immunoprecipitated from basal (–stim) and stimulated (+stim) N2a differentiated cells. Cells were lysed in ice-cold RIPA buffer containing 50 mM Tris HCl pH 8, 1% Nonidet P-40 (NP40), 0.1% SDS, 1 mM EDTA, 150 mM NaCl, 1 mM MgCl₂, 1× EDTA-free complete protease inhibitor cocktail (Roche, 5056489001) and phosphatase inhibitor cocktails (Sigma, P5726 and P0044). Cells were subsequently sonicated for 5 min at low intensity with a standard bioruptor diagenode. Following centrifugation (4 °C for 10 min at maximum speed), supernatants were collected, precleared and immunoprecipitated overnight at 4 °C with 50 μl GFP-conjugated Dynabeads protein A (Invitrogen). Beads had previously been conjugated with 5 μl anti-GFP antibody (Invitrogen, A6455) diluted in 500 μl PBS 1× for 2 h at room temperature. After immunoprecipitation, beads were washed with cold RIPA buffer and eluted with Laemmli sample buffer. Eppendorf LoBind microcentrifuge tubes (Eppendorf, 30108116) were used for the entire protocol. The immunoprecipitated elutions were run on precast 4–20% gradient gels (Midi Criterion TGX, Bio-Rad) and stained with Coomassie blue for 1 h at room temperature. Bands at the expected nCPEB4–GFP molecular weight were cut, washed with 50 mM NH₄HCO₃ and acetonitrile, reduced with 10 mM DTT and alkylated with 50 mM IAA. Samples were digested with trypsin and digestion was stopped by the addition of 5% formic acid. Following evaporation, samples were reconstituted in 15 μl of 1% formic acid and 3% acetonitrile. Mass spectrometry analysis of nCPEB4 PTM sites was performed as previously described⁷ with some modifications. In brief, samples were loaded in a μ-precolumn at a flow rate of 250 nl min^–1 using Dionex Ultimate 3000. Peptides were separated using a NanoEase MZ HSS T3 analytical column with a 60 min run and eluted with a linear gradient from 3 to 35% buffer B in 60 min (buffer A: 0.1% formic acid in H₂O; buffer B: 0.1% formic acid in acetonitrile). The column outlet was directly connected to an Advion TriVersa NanoMate (Advion) fitted on an Orbitrap Fusion Lumos Tribrid (Thermo Scientific). Spray voltage in the NanoMate source was set to 1.7 kV. The mass spectrometer was operated in a data-dependent acquisition mode. Survey mass spectrometry scans were acquired in the orbitrap with the resolution (defined at 200 m/z) set to 120,000. The top speed (most intense) ions per scan were fragmented in the HCD cell and detected in the orbitrap.

For peptide identification, searches were performed using MaxQuant (v.1.6.17.0) software and run against a target and decoy database to determine the false discovery rate. The database included proteins of interest sequences (nCPEB4–GFP and nCPEB4Δ4–GFP) and contaminants. Search parameters included trypsin enzyme specificity, allowing for two missed cleavage sites, oxidation in methionine, phosphorylation in serine, threonine and tyrosine, methylation and demethylation in lysine and arginine, and acetylation in the protein N terminus as dynamic modifications, and carbamidomethyl in cysteine as a static modification. Peptides with a q value lower than 0.1 and false discovery rate < 1% were considered as positive identifications with a high confidence level. Mass spectrometry spectra were searched against contaminants (released in 2017) and user proteins using Andromeda and MaxQuant (v.1.6.17.0) software. To accept a site as modified, PTM localization probability was set above 75%. For the differential expression analysis, a t-test on PTM site intensities from MaxQuant was applied for each site within nCPEB4 variants. For data visualization, two parameters were used for each PTM site, namely the sum of intensities of modified peptides that contain the specific PTM-site (Int_mod) and the PTM-to-base ratio, with the latter calculated as: Int_mod/Int_unmod, where Int_unmod is the sum of intensities of unmodified peptides that contain the site. For data visualization, the PTM-to-total ratio was calculated for each site as follows: PTM-to-total = Int_mod/(Int_mod + (Int_mod/PTM-to-base)).

Effect of phosphorylations on condensate dissolution

To strengthen the conclusion that phosphorylation of nCPEB4 does not promote condensate dissolution, we studied the behaviour in N2a cells of the condensates formed by phosphomimicking (S/T to D) and non-phosphorylatable (S/T to A) variants of nCPEB4(NTD), the phosphorylation status of which cannot be altered by depolarization. In agreement with our previous findings¹¹, the former had a lower propensity to condense (Extended Data Fig. 2c) and, in agreement with our conclusion, both variants dissolved after depolarization (Extended Data Fig. 2d,e).

Intracellular pH tracking

Quantitative determination of intracellular pH (pH_i) was performed using the cell-permeant ratiometric pH indicator SNARF-5F 5-(and-6)-carboxylic acid AM (Thermo Fisher) in live imaged N2a cells at 48 h of differentiation. In brief, for loading the pH indicator into cells, they were incubated with 10 µM SNARF-5F 5-(and-6)-carboxylic acid AM diluted in serum-free DMEM for 15 min at 37 °C. Cells were then washed and imaged in serum-free DMEM. A Zeiss Elyra PS1 LSM 880 confocal microscope using a Plan ApoChromat ×40/1.2 Imm corr DIC M27 water objective was used for acquisition at 2 emission wavelengths: −575 nm and 640 nm. Images were captured every 30 s over the recording period. pH_i estimation was performed as described in previous publications⁵¹. In brief, in vivo pH_i calibration was performed by fixing the pH_i between 5.5 and 7.5 with a commercially available intracellular pH calibration buffer kit (Thermo Fisher). Valinomycin and nigericin were used to equilibrate the intracellular pH. The intensity of fluorescence emitted at the two wavelengths was used to calculate a ratio (R_F640/F575) that is proportional to pH_i. Fluorescence ratio values (R_F640/F575) from cells with fixed pH_i were used to obtain a calibration curve for each biological replicate. Experimental pH_i estimation from the fluorescence ratio values was calculated using the following equation: pH_i = (R_F640/F₅₇₅ + b)/m, where m is the slope from the calibration curve equation and b is the intercept.

RNA extraction and real-time quantitative RT–PCR

For N2a RNA extraction, cells were scraped into an ice-cold plate, collected and centrifuged at 500g for 5 min at 4 °C. For mouse tissue RNA extraction, organs were ground with a liquid-nitrogen-cooled mortar to obtain tissue powder. Total RNA was extracted from both cells and tissue powder using TRIsure reagent (Bioline, Ecogen, BIO-38033) following the manufacturer’s protocol and using phenol–chloroform. The RNA concentration was determined using a Nanodrop spectrophotometer (Nanodrop Technologies). Next, 1 μg of total RNA was reverse transcribed using RevertAid reverse transcriptase (Themo Fisher, EP0442) following the manufacturer’s recommendations and using oligodT and random hexamers as primers. Quantitative real-time PCR (qPCR) was performed in triplicate in a QuantStudio 6flex (Thermo Fisher) using PowerUp SYBR green master mix (Thermo Fisher, A25778). All quantifications of mRNA levels were first normalized to an endogenous housekeeping control (Tbp), and then mRNA relative quantities to a reference sample (brain, N2a undifferentiated) were calculated using the 2–ΔΔCt method. The following primers were used for qPCR: 5′-TGATTCCATTAAAGGTCGTCTAAACT-3′ (Fw) and 5′-GAAACAATGAAGACTGACCTCTCCTT-3′ (Rv) for Mm Cpeb4 isoform containing exons 3 and 4; 5′-TGATTCCATTAAAGCAAGGACTTATG-3′ (Fw) and 5′-GCTGTGATCTCATCTTCATCAATATC-3′ (Rv) for Mm Cpeb4 isoform lacking exon 3; 5′-TGATTCCATTAAAGGTCGTCTAAACT-3′ (Fw) and 5′-GGAAACAATGAAGACTGACCATTAAT-3′ (Rv) for Mm Cpeb4 isoform lacking exon 4; 5′-ATTCCATTAAAGGTCAGTCTTCATTG-3′ (Fw) and 5′-GCTGTGATCTCATCTTCATCAATATC-3′ (Rv) for Mm Cpeb4 isoform lacking exons 3 and 4; 5′-GGAAAGGGACCTTCAAAGCAGT-3′ (Fw) and 5′ CTCTGTCCTTGGCATCGGCT-3′ (Rv) for Mm Srrm4; and 5′-ACTTCTATGCAGGCACGGTG-3′ (Fw) and 5′-AGCCAGGCATTGCAGAAGTAT-3′ (Rv) for Mm Rbfox1. Mm JunB and cFos primers were obtained from a previous study⁵.

Real-time semi-quantitative RT–PCR

For CPEB4 splicing isoform amplification, specific primers were used in CPEB4 exon 2 (Fw primer) and exon 5 (Rv primer) as previously described². PCR products conforming to the 4 isoforms of CPEB4 were resolved on a 2% agarose/GelRed gel run at 130 V for 2 h.

WB analysis

Cells were lysed in ice-cold buffer containing 1% NP40, 150 mM NaCl, 50 mM Tris HCl (pH 7.5), 2 mM EDTA, 2 mM EGTA, 20 mM sodium fluoride, 2 mM PMSF, 2 mM sodium orthovanadate, 1 mM DTT and 1× EDTA-free complete protease inhibitor cocktail. Lysed samples were then sonicated at medium intensity for 5 min with a standard bioruptor diagenode, and total protein content was quantified using a DC Protein assay (Bio-Rad, 5000113). Next, 15–30 μg of total protein lysate was resolved on SDS–PAGE gels and transferred to a nitrocellulose membrane (Cytiva, 10600001). After 1 h of blocking at room temperature in 5% non-fat milk, membranes were incubated overnight at 4 °C with primary antibodies and subsequently with secondary antibodies for 2 h at room temperature. Specific proteins were labelled using the following primary antibodies: CPEB4 (1:100, homemade, mouse monoclonal, ERE149C); GFP (1:2,000, Invitrogen, rabbit polyclonal, A6455); MYC (1:1,000, Abcam, goat polyclonal, ab9132); and β-actin (1:10,000, Abcam, mouse monoclonal, ab20272) or streptavidin (1:5,000, Thermo Fisher, S911). The following secondary antibodies were used: goat anti-mouse (1:300, Thermo Fisher, 31430); goat anti-rabbit (1:300, Thermo Fisher, G-21234); and donkey anti-goat (1:300, Abcam, ab6885). Membranes were then incubated for 3 min with Amersham ECL TM WB detection reagents (Sigma, GERPN2106) or for 5 min with Clarity Western ECL substrate (Bio-Rad, 1705061).

nCPEB4 and nCPEB4Δ4 co-localization experiments in N2a cells

mCherry red signals were acquired with a DPSS 561 excitation laser (9%) and a HyD2 detector set to 570–650 nm with a gain of 33%. DAPI and GFP signals were acquired using the settings specified in the section ‘nCPEB4–GFP distribution in N2a cells’. For measuring the extent of co-localization between the two channels, the ImageJ JaCoP plug-in was used to obtain Pearson’s correlation coefficients and Mander’s overlap coefficients per cell.

Immunohistochemistry

Mouse embryos at embryonic day 13.5 were fixed in 10% neutral-buffered formalin solution and embedded in paraffin. Rabbit polyclonal primary antibody anti-CPEB4 (Abcam, ab83009) was used at 1:1,000 dilution. Embryo sections were counterstained with haematoxylin.

X.
laevis oocyte preparation

Stage VI oocytes were obtained from full-grown X. laevis females as previously described⁵². In brief, ovaries were treated with collagenase (2 mg ml^–1; StemCell Technologies) and incubated in modified bath saline 1× medium with 0.7 mM CaCl₂. Animal handling and all experimental protocols were approved by the Animal Ethics Committee at the Barcelona Science Park and by the Government of Catalonia.

X.
laevis BioID

BioID was performed as previously described⁷. In brief, 150 stage VI X. laevis oocytes were microinjected with 50.6 nl of 50 ng μl^–1 in vitro transcribed and polyadenylated RNAs corresponding to MYC-BirA-xCPEB4 variants. Oocytes were then incubated in 20 μM biotin (Merck) at 18 °C for 40 h. Oocytes were lysed in cold lysis buffer and centrifuged twice at 16,000g at 4 °C for 15 min. Cold BioID lysis buffer was added to cleared extract, and the resulting mixture was subjected to clearing with PD MiniTrap G-25 columns (GE Healthcare). Next, 1.6% Triton X-100 and 0.04% SDS were added, and extracts were incubated with MyOne Dynabeads Streptavidin C1 (Invitrogen). The beads were then washed with a subsequent sequence of wash buffers. The beads were resuspended in 3 M urea, 50 mM NH₄HCO₃, pH 8.0, and 5 mM DTT for 1 h at room temperature with orbital shaking and subsequently incubated in 10 mM iodoacetamide for 30 min at room temperature, and then DTT was added. Proteins were on-bead digested with trypsin (Promega) at 37 °C for 16 h with orbital shaking. Digestion was stopped by the addition of 1% formic acid. Mass spectrometry analysis of biotinylated proteins in xCPEB4 and xCPEB4 + Ex4 was carried out at the Mass Spectrometry Facility at IRB Barcelona as previously described⁷. In brief, samples were analysed using an Orbitrap Fusion Lumos Tribrid mass spectrometer (Thermo Scientific). The MS/MS spectra obtained were searched against the UniProt (Xenopodinae, release 2017_02) and contaminants databases, and proteins of interest sequences using Proteome Discoverer (v.2.1.0.81).

Identification of the nCPEB4 isoform interactome by immunoprecipitation coupled to mass spectrometry

Overexpressed nCPEB4–GFP, nCPEB4Δ4–GFP and GFP (control) were immunoprecipitated from differentiated N2a cells. Cells were lysed in ice-cold RIPA buffer containing 50 mM Tris HCl pH 8, 1% Nonidet P-40 (NP40), 0.1% SDS, 1 mM EDTA, 150 mM NaCl, 1 mM MgCl₂, 1× EDTA-free complete protease inhibitor cocktail (Roche, 5056489001) and phosphatase inhibitor cocktails (Sigma, P5726, and P0044). Cells were subsequently sonicated for 5 min at low intensity with a standard bioruptor diagenode. Following centrifugation (4 °C for 10 min at maximum speed), supernatants were collected, precleared and immunoprecipitated overnight at 4 °C with 50 μl GFP-conjugated Dynabeads protein A (Invitrogen). Beads had previously been conjugated with 10 μl anti-GFP antibody (Invitrogen, A6455) diluted in 500 μl PBS 1× for 2 h at room temperature. After immunoprecipitation, beads were washed with cold RIPA buffer (containing 0.05% NP-40 and 0.1% SDS) and eluted with Laemmli sample buffer. Eppendorf LoBind microcentrifuge tubes (Eppendorf, 30108116) were used for the entire protocol. The immunoprecipitated elutions were shortly run on 8% acrylamide 0.75 mm gels until the whole individual samples were compacted at the upper part of the running gel. Then, gels were stained with InstantBlue Coomassie for 1 h at room temperature. Bands corresponding to elutions were cut, washed and digested with 0.1 μg μl^–1 trypsin, (Promega). Samples were digested with trypsin and digestion was stopped by the addition of 5% formic acid. Following evaporation, samples were reconstituted in 12 μl of 1% formic acid and 3% acetonitrile. In brief, mass spectrometry analysis of immunoprecipitates was performed as follows: samples were loaded into an Evotip trap column (Evosep) at a flow rate of 250 nl min^–1. Peptides were separated using a EV1137 analytical column (Evosep) with a 88-min run and eluted with buffer A (0.1% formic acid in H₂O) and buffer B (0.1% formic acid in acetonitrile). The column outlet was directly connected to an Easyspray (Thermo Scientific) fitted on an Orbitrap Eclipse Tribrid (Thermo Scientific). Spray voltage in the Easyspray source was set to 2.5 kV. The mass spectrometer was operated in data-dependent acquisition mode. The top speed (most intense) ions per scan were fragmented in the HCD cell and detected in the orbitrap. For peptide identification, searches were performed using Proteome Discoverer (v.2.5.0.400) software and run against databases including universal contaminants, mouse from Swissprot (2023/04) and bait proteins. Search parameters included trypsin enzyme specificity, allowing for two missed cleavage sites, oxidation in methionine, acetylation in the protein N terminus, methionine loss in the N terminus, and methionine loss in and acetylation in the N terminus as dynamic modifications, and carbamidomethyl in cysteine as a static modification. Protein hits in co-precipitates from each isoform were determined using a differential analysis of protein abundance in nCPEB4–GFP or nCPEB4Δ4–GFP relative to GFP. Protein group abundance values from Proteome Discoverer were used for protein quantification, and cut-off values for the fold change (|FC| > 1.5) and adjusted P value (padj < 0.05) were applied to define over-abundant significant proteins. Significant hits included proteins with no missing values in the three conditions (nCPEB4, nCPEB4Δ4 or GFP) or in only one condition, for which value imputation was performed. Only significant hits were considered for subsequent analyses. Protein hits differentially represented in nCPEB4Δ4–GFP versus nCPEB4–GFP were determined by a differential abundance analysis between the two baits, applying fold change (|FC| > 1.5) and padj < 0.05) as cut-off values.

Competition experiments

Competition experiments were performed as previously described¹⁰ using 23 nl of 500 ng μl^–1 in vitro transcribed and polyadenylated RNAs encoding for HA-tagged xCPEB1 and xCPEB4 RRMs and variants. Not injected was considered as 0% competition whereas HA-xCPEB1 RRM was considered as 100% competition.

Plasmids for protein expression in Escherichia coli

An insert codifying for the nCPEB4(NTD) protein sequence (UniProt identifier Q17RY0-2, residues 1–448) was ordered in GenScript subcloned in a pET-30a(+) vector. The His₆ tag and S tag from the plasmid were removed by PCR using a NEB Q5 site-directed mutagenesis kit, with oligonucleotides purchased from Sigma-Aldrich. The histidine to serine mutants were ordered from GenScript subcloned in a pET-30a(+) vector in the NdeI and XhoI restriction enzymes positions. The nCPEB4Δ4, ΔHC and Δ4ΔHC mutants were generated by PCR mutagenesis on the nCPEB4(NTD) plasmid using a NEB Q5 Site-directed mutagenesis kit, with oligonucleotides purchased from Sigma-Aldrich. The sequences of the N-terminal domain of nCPEB4 and mutants used for the in vitro experiments are described in Supplementary Methods.

Protein expression and purification for in vitro experiments

E. coli B834 cells were transformed with the pET-30a(+) plasmids. For non-isotopically labelled protein, the cells were grown in LB medium at 37 °C until the optical density at 600 nm (OD₆₀₀) was 0.6, and then the cultures were induced with 1 mM IPTG for 3 h at 37 °C. For ¹⁵N or ¹⁵N,¹³C isotopically labelled protein, the cells were grown in LB medium until OD₆₀₀ = 0.6 and then transferred into M9 medium⁵³ (3 litres LB for 1 litre M9) containing [¹⁵N]H₄Cl or [¹⁵N]H₄Cl and [¹³C]glucose, respectively, and then the cultures were induced with 1 mM IPTG overnight at 37 °C. The cultures were then centrifuged for 30 min at 4,000 r.p.m., and the cells were resuspended with lysis buffer (50 mM Tris-HCl, 1 mM DTT, 100 mM NaCl, 0.05% Triton X-100, at pH 8.0, and supplemented with 500 μl of PIC and 500 μl of 100 mM PMSF).

The cells were lysed by sonication and centrifuged for 30 min at 20,000 r.p.m. The pellet was washed first with wash-1 buffer (20 mM Tris-HCl, 1 mM DTT, 1 M NaCl, 0.05% Triton X-100, at pH 8.0, and supplemented with 500 μl of PIC, 500 μl of 100 mM PMSF, and 50 μl of 5 mg ml^–1 DNAse) and then with wash-2 buffer (20 mM Tris-HCl, 1 mM DTT, 0.1 M l-arginine, at pH 8.0). The pellet was resuspended with the nickel-A buffer (25 mM Tris-HCl, 1 mM DTT, 50 mM NaCl, 8 M urea and 20 mM imidazole, at pH 8.0) and centrifuged for 30 min at 20,000 r.p.m. The supernatant was injected at room temperature into a nickel affinity column and eluted with a gradient from 0 to 100% of nickel-B buffer (25 mM Tris-HCl, 1 mM DTT, 50 mM NaCl, 8 M urea and 500 mM imidazole, at pH 8.0). The fractions with protein were pooled, and 1 mM EDTA was added. The sample was injected into a size exclusion Superdex 200 16/600 (GE Healthcare) column, running at 4 °C in size exclusion buffer (25 mM Tris-HCl, 1 mM DTT, 50 mM NaCl and 2 M urea, at pH 8.0). The fractions with protein were pooled and concentrated to approximately 150 μM. The sample was dialysed against the final buffer (20 mM sodium phosphate, 1 mM TCEP and 0.05% NaN₃, at pH 8.0), fast frozen in liquid nitrogen and stored at −80 °C.

Peptide for in vitro experiments

The me4(GS)₃me4 synthetic peptide with amidated or Cy3-modified C terminus and acetylated N terminus was obtained as lyophilized powder with >95% purity from GenScript. The peptide was dissolved in 6 M guanidine thiocyanate and incubated with agitation overnight at 25 °C. The sample was then centrifuged at 15,000 r.p.m. for 10 min. The supernatant was extensively dialysed against the final buffer (20 mM sodium phosphate, 1 mM TCEP and 0.05% NaN₃, at pH 8.0)⁵⁴. The peptide sample was then manipulated in the same way as the protein samples, as detailed below.

Sample preparation for in vitro experiments

All samples were prepared on ice as follows. First, a buffer stock solution consisting of 20 mM sodium phosphate buffer with 1 mM TCEP and 0.05% NaN₃ was pH adjusted to 8.0 (unless otherwise indicated) and filtered using 0.22 μm sterile filters (buffer stock). A 1 M NaCl solution in the same buffer was also pH adjusted to 8.0 (unless otherwise indicated) and filtered (salt stock). The protein samples were then thawed from −80 °C on ice, pH adjusted to 8.0 (unless otherwise indicated) and centrifuged for 5 min at 15,000 r.p.m. at 4 °C. The supernatant (protein stock) was transferred to a new Eppendorf tube, and the protein concentration was determined by measuring its absorbance at 280 nm. The samples were prepared by mixing the correct amounts of buffer stock, protein stock and salt stock, as well as other indicated additives in the experiments, to reach the desired final protein and NaCl concentrations.

Apparent absorbance measurement as a function of temperature

The absorbance of the samples was measured at 350 nm (A_350 nm) using 1 cm pathlength cuvettes and a Cary100 ultraviolet–visible spectrophotometer equipped with a multicell thermoelectric temperature controller. The temperature was increased progressively at a ramp rate of 1 °C min^−1. The cloud point (T_c) values were determined as the maximum of the first-order derivatives of the curves, and the absorbance increase (ΔA) represents the difference between the maximum and the minimum absorbance values of the samples during the temperature ramp.

For the experiment to quantify the reversibility of condensation, a 20 μM protein with 100 mM NaCl sample was prepared on ice. It was then split into 4 Eppendorf tubes and a temperature ramp was carried out with the first one after centrifugation for 2 min at 5 °C and 15,000 r.p.m. Once the T_c and ΔA for condensation had been determined, the other 3 samples were heated 10 °C above the T_c for 2.5 min and then cooled to 10 °C below the T_c for 5 more min. This procedure was repeated 1, 2, or 3 times for each sample. Next, the samples were centrifuged for 2 min at 5 °C and 15,000 r.p.m., and a temperature ramp was carried out to determine their respective T_c and ΔA values (Extended Data Fig. 7d).

Microscopy in vitro

For microscopy imaging, 1.5 μl of sample was deposited in a sealed chamber comprising a slide and a coverslip sandwiching double-sided tape (3M 300 LSE high-temperature double-sided tape of 0.17 mm thickness). The coverslips used had been previously coated with PEG-silane following a published protocol⁵⁵. The imaging was always performed on the surface of the coverslip, where the condensates had sedimented.

The DIC microscopy images were taken using an automated inverted Olympus IX81 microscope with a ×60/1.42 oil Plan APo N or a ×60/1.20 water UPlan SAPo objective using the Xcellence rt (v.1.2) software.

For fluorescence microscopy experiments, the purified proteins were labelled with DyLight 488 dye (DL488, Thermo Fisher Scientific) or Alexa Fluor 647 dye (AF647, Thermo Fisher Scientific). The labelling, as well as the calculation of the labelling percentage and the determination of the protein concentration, was performed following the provider’s instructions. The final samples contained 0.5 µM of labelled protein and/or peptide out of the total indicated concentrations.

FRAP experiments were recorded using a Zeiss LSM780 confocal microscope system with a Plan ApoChromat ×63/1.4 oil objective. Condensates of similar size were selected, and the bleached region was 30% of their diameter. The intensity values were monitored for different ROIs: ROI 1 (bleached area), ROI 2 (entire condensate) and ROI 3 (background signal). The data were fitted using EasyFrap software⁵⁰ to extract the kinetic parameters of the experiment (recovery half-time and mobile fraction).

Super-resolution microscopy images of the multimers and their time evolution were taken at 25 °C in a Zeiss Elyra PS1 LSM 880 confocal microscope using the Fast Airyscan mode with an alpha Plan ApoChromat ×100/1.46 oil objective. The pixel size was kept constant at 40 nm.

Fluorescence microscopy images of the condensates and aggregates were taken at 37 °C in a Zeiss Elyra PS1 LSM 880 confocal microscope with an Airyscan detector using a Plan ApoChromat ×63/1.4 oil objective. The quantification of the aggregation process was done by image analysis using Fiji/ImageJ. The regions with the fluorescence signal not stemming from the background or the spherical condensates were selected. The percentage of the area of the field of view occupied by this selection corresponds to the aggregation value of the sample. The partitioning of the proteins in the condensates was calculated by image analysis using Fiji/ImageJ. The partitioning for each condensate was calculated by dividing the mean intensity of the condensate by the mean intensity of a ring of 1 μm thickness around the condensate.

RNA for in vitro experiments

The RNA used for in vitro experiments is a fragment of the 3′ UTR of cyclin B1 mRNA from X. laevis containing only one CPE site^56,57, 5′-AGUGUACAGUGUUUUUAAUAGUAUGUUG-3′. We used it as a control to study whether it influences the properties of the condensates. RNA caused a slight decrease in the phase-separation propensity of both isoforms, larger for nCPEB4(NTD) than for nCPEB4Δ4(NTD), which we attribute to interactions with positively charged amino acids involved in the intermolecular interactions driving condensation (Extended Data Fig. 5f). Notably, however, the presence of RNA in the samples did not alter their propensity to aggregate (Extended Data Fig. 5g).

Saturation concentration measurements

Saturation concentration measurements of nCPEB4(NTD) and the histidine to serine mutants were carried out by incubating the samples at 40 °C for 5 min, followed by centrifugation at 5,000 r.p.m. for 1.5 min at 40 °C. The concentration of protein in the supernatant (c_sat) was determined by absorbance measurement at 280 nm.

NMR spectroscopy

The samples were prepared as indicated in the section ‘Sample preparation for in vitro experiments’ using isotopically labelled protein (¹⁵N- or ¹⁵N,¹³C-labelled). The prepared final samples were again pH adjusted to the desired value immediately before measurement. All the measurements were acquired at 5 °C using 3 mm NMR tubes with a sample volume of 200 µl.

All NMR experiments, except the diffusion measurements, were carried out on a Bruker Avance NEO 800 MHz spectrometer equipped with a TCI cryoprobe. All NMR samples contained 100 μM protein concentration (unless otherwise indicated) in 20 mM sodium phosphate buffer with 1 mM TCEP, 0.05% NaN₃, 7% D₂O and 2.5 μM DSS for referencing, at pH 8.0 (unless otherwise indicated). Samples with denaturant agent contained the indicated concentrations of d₄-urea.

A ¹⁵N,¹³C-labelled sample at 280 μM for nCPEB4(NTD) or 200 μM for nCPEB4Δ4(NTD) with 4 M urea at pH 7.0 was used for backbone resonance assignment. A series of nonlinear sampled 3D triple resonance experiments were recorded, including the BEST-TROSY version⁵⁸ of ¹H_N-detected HNCO, HN(CA)CO, HNCA, HN(CO)CA, HNCACB, HN(CO)CACB and (H)N(CA)NH. Also, additional ¹Hα-detected HA(CA)CON and (HCA)CON(CA)H experiments⁵⁹ were measured for nCPEB4(NTD). Backbone resonance assignments were performed using CcpNmr⁶⁰ (v.2.4.2). The NMR assignments are available from the Biological Magnetic Resonance Data Bank (identifiers 51875 and 52346 for nCPEB4(NTD) and nCPEB4Δ4(NTD), respectively).

pH titrations from 7.0 to 8.0 were carried out to transfer NH assignments to the final experimental conditions. Standard 2D ¹H,¹⁵N-HSQC or BEST-TROSY experiments were measured at 7.0 ≤ pH ≤ 7.25. 2D ¹H,¹⁵N-CP-HISQC⁶¹ experiments were used at pH ≥ 7.5 to reduce the effects of chemical exchange with water. For the urea titrations from 0 to 4 M at pH 8.0, ¹H,¹⁵N-CP-HISQC experiments were measured. In the ¹H,¹⁵N-CP-HISQC for the detection of arginine side-chain resonances, the ¹⁵N carrier was placed at 85 ppm, and ¹³C pulses for decoupling were centred at the chemical shift of ¹³C_δ (42 ppm) and ¹³C_℥ (158 ppm) arginine side chains.

Standard 2D ¹H,¹³C-HSQC experiments of 500 μM ¹⁵N,¹³C-labelled nCPEB4(NTD) were measured in the absence and presence of 4 M urea to monitor specific amino acid side chains easily identifiable by their typical ¹H and ¹³C random coil chemical shifts.

For histidine pK_a determination, 2D ¹H,¹³C-HSQC spectra of 75 μM ¹⁵N,¹³C-labelled nCPEB4(NTD) were measured in the presence of 4 M urea at pH values between 5.58 and 8.31. The pH-induced changes of the chemical shifts of histidine side chains (¹H and ¹³C, both aliphatic and aromatic) were fitted to a sigmoid function to obtain an apparent pK_a for all these histidine resonances in nCPEB4(NTD).

Non-uniform sampled experiments were processed using qMDD⁶² (v.3.2). 2D ¹⁵N-correlations (¹H,¹⁵N-HSQC, ¹H,¹⁵N-CP-HISQC, BEST-TROSY) and 2D ¹H,¹³C-HSQC were processed using NMRPipe⁶³ and Topspin (v.4.0.8) (Bruker), respectively.

¹⁵N-edited X-STE diffusion experiments⁶⁴ of 100 μM ¹⁵N-labelled nCPEB4(NTD) in the absence and presence of 2 M urea were performed on a Bruker Avance III 600 MHz spectrometer equipped with a TCI cryoprobe. An encoding/decoding gradient length of 4.8 ms and a diffusion delay of 320 ms were used. The hydrodynamic diameter of nCPEB4(NTD) was estimated using dioxane as a reference molecule. Diffusion measurements under identical experimental conditions were carried out for dioxane, using in this case the PG-SLED sequence⁶⁵. A gradient time (δ) of 1.6 ms and a diffusion time (Δ) of 70 ms were used. Diffusion coefficients were obtained by fitting the data to a mono-exponential equation using MestreNova (v.14.2.1-27684).

Dynamic light scattering

Dynamic light scattering (DLS) measurements were taken with a Zetasizer Nano-S instrument (Malvern) equipped with a He-Ne of 633 nm wavelength laser. Immediately before the measurement, the prepared samples were centrifuged for 5 min at 15,000 r.p.m. at 4 °C, and only the supernatant was subjected to measurement. Three measurements were performed for each sample, each of the measurements consisting of 10 runs of 10 s each. The experiments were carried out at 5 °C unless otherwise indicated. The deconvoluted hydrodynamic diameters arise from the mean of the peak of the intensity deconvolution of the data.

Size exclusion chromatography coupled to multi-angle light scattering

Size exclusion chromatography coupled to multiangle light scattering (SEC–MALS) experiments were performed by loading a 160 μM nCPEB4(NTD) with 0 mM NaCl sample into a Superose 6 increase 10/300 GL column (GE Healthcare) mounted on a Shimadzu Prominence Modular HPLC with a SPD-20 UV detector (Shimadzu) coupled to a Dawn Heleos-II multi-angle light scattering detector (18 angles, 658 nm laser beam) with an Optilab T-rEX refractometer (Wyatt Technology). The SEC-UV/MALS/RI system was equilibrated at 25 °C with 20 mM sodium phosphate buffer with 1 mM TCEP and 0.05% NaN₃ at pH 8.0. Data acquisition and processing were performed using Astra 6.1 software (Wyatt Technology).

Liquid-phase transmission electron microscopy

Transmission electron microscopy (TEM) experiments were performed on a 50 μM nCPEB4Δ4(NTD) sample with 0 mM NaCl. The sample was first imaged in solid-state TEM using the same set-up as described below for liquid-phase TEM (LPEM). Pre-screening samples in solid-state TEM before the LPEM imaging procedure is routinely done to pre-screen the structures that will be imaged in liquid later. The sample with no stain was deposited onto 400 mesh Cu grids, which were plasma discharged for 45 s to render them hydrophilic and to allow optimal sample wettability.

LPEM imaging was performed using a JEOL JEM-2200FS TEM microscope. The system was equipped with a field emission gun operating at 200 kV and an in-column Omega filter. The images were acquired with a direct detection device, the in-situ K2-IS camera (Gatan). The Ocean liquid holder, from DENSsolutions, was used to image the structure and dynamics of the specimens. The liquid samples were encased into two silicon nitride (Si_XN_y) chips. These chips had a 50-nm-thick Si_XN_y electron-transparent window of dimensions 10 × 200 μm. One of these chips had a 200 nm spacer that acts as a pillar and defines the liquid cell thickness, that is, z height, and hence the liquid thickness in the experiments. The chips were cleaned in HPLC-graded acetone followed by isopropanol for 5 min each to remove their protective layer made of poly(methyl methacrylate). Afterwards, the chips were plasma-cleaned for 13 min to increase their hydrophilicity. Next, 1.5 μl of non-stained sample was deposited on the previously prepared 200-nm-spacer chip. The drop-casted sample was enclosed by the spacer-less chip, thus sealing the liquid chamber. Then, 300 μl of the sample solution was flushed in the liquid holder at 20 μl min^–1 with a syringe pump to ensure that the liquid cell inlet and outlet pipes were filled with the solution. We waited 5 min after collecting the sample solution from the outlet tube to minimize the convection effects from the flowing process. The liquid holder was introduced in the microscope, where imaging was performed in TEM mode and static conditions, that is, not in flow. To limit the electron beam dose (20 e^– Å^–2) on the specimen, images were collected at the minimum spot size (number 5) with a small condenser lens aperture (CLA number 3). For our investigations, dose fractionation videos were recorded in counted mode at 20 frames per s with the K2 camera. Every image was recorded in the format of 4-byte greyscale and required the full size of the detector.

The images and videos recorded were corrupted by noise, which significantly reduced the quality of the images, complicating any further analysis. Therefore, the Noise2Void (N2V) machine learning-based approach was adopted to overcome this problem⁶⁶. Unlike the conventional machine learning-based approach, N2V reduces the noise of an image without any need for the corresponding noiseless image. This requirement makes the N2V ideal to process LPEM images, as noiseless images in liquid TEM are impossible to record. However, N2V requires an extensive dataset (also known as a training set) to fulfil its task, a common requirement to machine-learning-based approaches. Conventionally, the training set has to contain thousands of hundreds of images recorded with the same imaging settings to produce high-quality estimation of the noise distribution. Unfortunately, in our case, only single videos (image sequence) recorded at different imaging conditions were available. Therefore, the training set was created by sampling the image sequence every two frames to not bias the training process of the N2V. The remaining half of the image sequence was processed and used to derive the presented results.

To train the N2V, 3,050 frames were selected, extracting 128 different non-overlapping squared patches (that is, portions of the pixels of the image) of 64 × 64 pixels from each of them. Moreover, the N2V was iterated for 100 epochs, a trade-off value between performances and processing time. The training was performed by extracting 128 random patches (64 × 64 pixels) from each training image. These values produced the best results in a short processing time.

Molecular simulations

Molecular dynamics simulations were performed using the single-bead-per-residue model CALVADOS (v.2)^22,23 implemented in OpenMM (v.7.5)⁶⁷. All simulations were conducted in the NVT ensemble at 20 °C using a Langevin integrator with a time step of 10 fs and friction coefficient of 0.01 ps^−1. In our simulations, pH was modelled through its effect on the charge of the histidine residues (q_His)⁶⁸.

Direct coexistence simulations were performed with 100 chains in a cuboidal box of side lengths [L_x, L_y, L_z] = [25, 25, 300] nm. Simulations were run in n = 3 independent replicas for at least 55 µs each. The systems readily formed a protein-rich slab spanning the periodic boundaries along the x and y coordinates. The initial 2 µs of the simulation trajectories were discarded, and time-averaged concentration profiles along the z axis were calculated after centring the condensates in the box as previously described²².

To model protein multimers, 400 chains were initially placed at random positions in a cubic box of side length 188 nm and simulations were run in n = 2 replicas for 16 µs each. The formation and dissolution of protein multimers was monitored using a cluster analysis implemented in OVITO⁶⁹. Proteins were clustered on the basis of the distances between their centres of mass, using a cut-off of 1.5 times the average radius of gyration of the protein. Contact maps were calculated between a chain in the middle of the condensate or multimer and the surrounding chains. Contacts were quantified using the switching function c(r) = 0.5 – 0.5 × tanh[(r – σ₀)/r₀], where r is the intermolecular distance between two residues, σ_0 = 1 nm, and r₀ = 0.3 nm.

To match the conditions of the experiments with which the simulation results are compared, direct coexistence and multimer simulations were performed at ionic strengths of 150 mM and 60 mM, respectively. Configurations were saved every 2 ns for slabs and every 5 ns for multimer simulations.

Simulation trajectories were analysed using MDTraj⁷⁰ (v.1.9.6) and MDAnalysis^71,72 (v.1.1). The molecular visualizations presented in Fig. 2j were generated using VMD⁷³ (v.1.9.4).

Characterization of the multimers

To investigate which specific residues of nCPEB4(NTD) drive its condensation, we used solution NMR spectroscopy, a technique that can provide residue-specific information on the conformation and interactions of intrinsically disordered proteins⁷⁴. Indeed, when conditions are insufficient for condensation, the same interactions driving this process can cause the monomer to collapse into a state that can be characterized at residue resolution, thus providing this key information⁷⁵. Under such conditions (Supplementary Fig. 5a), nCPEB4(NTD) had a spectrum characteristic of a collapsed intrinsically disordered protein, in which only 13% of the NMR signals were visible (Supplementary Fig. 5b, in light green). We analysed the sample by DLS to confirm its monomeric nature, but we detected only nCPEB4(NTD) multimers with a hydrodynamic diameter of approximately 55 nm, which is much larger than that predicted for a monomer⁷⁶ (approximately 11 nm) (Supplementary Fig. 5c).

To characterize the multimers, also known as clusters, which have also been observed for phase-separating proteins with intrinsically disordered domains^77,78,79,80, we first determined whether their formation is reversible. To this end, we performed DLS analysis of increasingly diluted samples of nCPEB4(NTD) under the same solution conditions. We observed that at 0.5 µM, the multimers seemed to be in equilibrium with a species with a hydrodynamic diameter close to that predicted for the monomer (Extended Data Fig. 3e, in green). Next, at this concentration, we modified solution conditions to favour condensation by increasing the temperature and ionic strength, as well as by decreasing the pH. In all cases, we observed a decrease in the signal corresponding to the monomer and an increase in that corresponding to multimers, thus indicating a shift of the multimerization equilibrium (Extended Data Fig. 3e). This result suggests that the nCPEB4(NTD) multimers are stabilized by the same types of interactions as the condensates (Extended Data Fig. 1m and Fig. 2e,k).

These findings prompted us to study whether these multimers grow over time to become condensates observable by optical and fluorescence microscopy, that is, whether they represent intermediates in the condensation pathway. To this end, we first analysed, over 15 h, samples of freshly prepared nCPEB4(NTD) multimers by DLS. We observed a progressive increase in size and polydispersity up to an average hydrodynamic diameter of approximately 90 nm and a polydispersity index of approximately 0.10 (Supplementary Fig. 5d). We also determined the morphology of the multimers at the nanoscale by LPEM, confirming that they are spherical and have a diameter of 30–50 nm (Extended Data Fig. 3f and Supplementary Fig. 5e). Finally, super-resolution microscopy confirmed that the multimers grow with time, thus becoming larger spherical particles resembling condensates (Extended Data Fig. 3g). It is possible that the nCPEB4 multimers here identified correspond to mesoscopic condensates, equivalent to those observed by optical microscopy, albeit smaller; however, as their dimensions preclude their thorough physical characterization, we consider it appropriate to consider them distinct species in this work.

We next carried out experiments to determine the nature of the species giving rise to the residue-specific NMR signals observed in the presence of multimers. To this end, we analysed a sample of multimers by SEC–MALS. This approach showed that the multimers are assemblies of approximately 350 nCPEB4(NTD) molecules (Supplementary Fig. 5f), corresponding to a molecular weight of approximately 1.7 × 10⁷ Da. Next, we performed ¹⁵N-edited X-STE experiments to measure the diffusion coefficient of the species giving rise to the signals detected by NMR, from which we derived that they diffuse much more slowly than a monomer, with a value in semi-quantitative agreement with that obtained by DLS and LPEM (Supplementary Fig. 5g), thereby indicating that they correspond mainly to multimers. We concluded that solution NMR can be used to identify the most flexible regions of the nCPEB4(NTD) sequence under conditions in which it forms multimers, presumably corresponding to those least involved in the interactions stabilizing them^81,82,83.

To assign the NMR signals to specific residues, we progressively added urea and observed that at 1.5–2 M, the urea concentration necessary to dissociate the multimers, the signals of most residues could be detected (Extended Data Fig. 3h,i and Supplementary Fig. 5b) and stemmed from a species with the diffusion coefficient expected for a nCPEB4(NTD) monomer (Supplementary Fig. 5g). A comparison of the spectra obtained in the absence and presence of denaturant revealed that the signals observed for the multimer corresponded to residues between positions approximately 50 and 150 in the sequence of nCPEB4(NTD): this region of sequence is devoid of aromatic and positively charged residues and instead rich in negatively charged ones (Extended Data Fig. 3h,j). Despite the presence of 4 M urea, the spectrum of monomeric nCPEB4(NTD) had a wide range of intensities. An analysis of signal intensity as a function of residue type revealed particularly low values for histidine and arginine (Extended Data Fig. 3k), which suggested that these residue types are involved in transient interactions and explain the very low signal intensity of the histidine-rich HClust (residues 229–252) and the arginine-rich me4 (residues 403–410) (Extended Data Fig. 3h).

To facilitate the interpretation of these results, we monitored the formation of nCPEB4(NTD) multimers in molecular simulations performed using the CALVADOS model^22,23. Contacts calculated between a chain in the centre of the multimer and the surrounding chains in assemblies of 130–170 chains showed that the C-terminal region, including HClust and me4, is more involved in intermolecular interactions than the N-terminal region rich in aspartate and glutamate residues (positions 72–147) (Extended Data Fig. 3l). The simulations therefore support the conclusion that the decreased NMR signal intensity in the C-terminal region reflects an increased number of contacts of these residues within the condensates. Finally, to further characterize the interactions stabilizing the multimers, we performed ¹H,¹³C-HSQC experiments to analyse the intensity of side-chain signals without urea, when nCPEB4(NTD) is multimeric, and in 4 M urea, when nCPEB4(NTD) is instead monomeric (Extended Data Fig. 3m and Supplementary Fig. 5h–j). We observed that the intensities, especially those of the signals corresponding to histidine, tryptophan and arginine residues, are lower for the former than for the latter. We also performed ¹H,¹⁵N-CP-HISQC experiments to study the arginine side-chain NH signals in the same samples and found that they are undetectable in the absence of the denaturant but can be observed in its presence (Supplementary Fig. 5k). Taken together, our results indicate that the multimers formed by nCPEB4(NTD) on the condensation pathway are stabilized mainly by interactions involving histidine and arginine residues, which are predominantly located in the C-terminal region (Supplementary Fig. 5l).

Protein sequence analysis

Cprofiler (http://www.cprofiler.org/) was used to determine the enrichment score of each amino acid type in the protein sequence compared with the DisProt3.4 database^84,85. The protein sequences analysed correspond to residues 1–448 for nCPEB4(NTD) (UniProt identifier Q17RY0-2), residues 1–311 for CPEB2(NTD) (UniProt identifier Q7Z5Q1-3) and residues 1–452 for CPEB3(NTD) (UniProt identifier Q7TN99-1).

Reporting summary

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