mRNA recruitment by G3BP1 condensates is regulated by Caprin1 but requires G3BP1 binding to mRNA

Abstract

Upon stress exposure, cells repress translation and release mRNAs from polysomes. The sudden increase in cytoplasmic messenger RNA (mRNA) concentration triggers the formation of cytoplasmic condensates known as stress granules (SGs) that enrich mRNA and RNA binding proteins (RBPs). SGs assemble through liquid-liquid phase separation involving interactions between a subset of RBPs and mRNA. G3BP1 is considered the central participant of this network and functions as a molecular switch depending on the mRNA concentration and the recruitment of G3BP1 protein partners. Among them, Caprin1 is an SG associated protein that binds the NTF2-L domain of G3BP1, promoting SG assembly. Herein, we show that Caprin1 triggers the formation of large G3BP1-mRNA condensates in vitro and improves both the mRNA and G3BP1 recruitment in SGs. However, this function requires intact RNA binding domains of G3BP1. Additionally, the central intrinsically disordered region (IDR2) of G3BP1 regulates the G3BP1 mRNA binding. However, Caprin1 restores the recruitment in SGs of G3BP1-ΔIDR2. We propose that heterotypic interactions between Caprin1 and G3BP1 contribute to the centrality of G3BP1 by increasing the recruitment of mRNA by G3BP1 which is fundamental for condensate formation.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-21066-7.

Introduction

Cells deploy different strategies to compartmentalize their space particularly by the formation of membrane-less biomolecular condensates via liquid-liquid phase separation (LLPS). In this process, the biomolecules of a homogeneous phase demix which leads lead to the formation of two spaces, a condensed phase rich in biomolecules, and the other diluted phase¹. Biomolecular condensates are abundant in both the nucleus² and the cytoplasm³, are constituted of a wide variety of proteins and RNA and are mainly involved in RNA metabolism regulation. Each type of condensate has its own RNA and protein composition and is characterized by specific mechanical properties and biological functions. Stress Granules (SGs) are very well-known examples of cytoplasmic condensates and they are the subject of numerous studies to know their composition^4,5, the mechanisms at work in their formation⁶ and their biological functions^7,8. SGs are often used as model structures in studies concerning the physical and chemical processes involved in the formation and stability of condensates⁹. Among the hundreds of different proteins that are present in SGs, a small number are considered essential in the regulation of these condensates. Within these, G3BP1 and G3BP2 have been demonstrated to play a central role. Indeed, the overexpression of G3BP1 or G3BP2 nucleates SGs¹⁰, and the knockdown of both^11–13 leads to the inhibition of SG assembly. In addition, some viral proteins bind G3BP1 and G3BP2 to inhibit SG formation and promote viral replication¹⁴. Finally, G3BP1 disruption by ubiquitination for example causes the SG disassembly¹⁵.

SGs assembly involves the combination of protein-protein, protein-RNA and RNA-RNA interactions¹⁶. During an episode of cellular stress, the release of a large number of mRNAs into the cytoplasm due to the dissociation of polysomes promotes the intramolecular base pairing among RNA molecules¹⁷. Additionally, nearly 80% of the SG proteome has the ability to bind RNA allowing the handling of these released RNAs. Finally, the proteins present in SGs are particularly enriched in intrinsically disordered domains (IDR) that drive the protein assembly through multiple weak homotypic and heterotypic interactions^18–24. G3BP1 combines characteristics that explain its centrality in the SG assembly^12,25,26. Structurally, G3BP1 contains an RNA Recognition Motif (RRM), a folded dimerization domain (NTF2L) involved in homo and heterotypic protein-protein interactions and three IDRs of distinct composition. Such structural features promote the G3BP1 interactions with diverse SG components, proteins or RNA, and G3BP1 is considered to be a central node in the protein-protein and protein-RNA interaction network²⁶. Indeed, the G3BP1–RNA interaction is fundamental for SG formation since the deletion of the RRM domain of G3BP1 abolishes the SG assembly^12,13. Surprisingly, the IDRs of G3BP1 do not appear to be essential in SG assembly, whereas these flexible domains which generate numerous heterotypic interactions are important for the LLPS process of many RNA binding proteins (RBP) such as FUS, hnRNPA1 or TDP43^21,22,27. For G3BP1, the NTF2L domain is particularly important for the interaction with different protein partners involved in SG assembly. For instance, the binding of Caprin1 and UBAP2L to NTF2L domain of G3BP1 promotes SG formation^28,29 while competitive USP10 binding to the same domain of G3BP1 inhibits SG assembly^13,30. The Caprin1-G3BP1 interaction is considered fundamental for the regulation of SG assembly. Firstly, the presence of Caprin1 decreases the threshold concentration of G3BP1 and RNA necessary to trigger LLPS in vitro^12,25. Secondly, point mutations in the NTF2L domain of G3BP1 negatively impact the Caprin1-G3BP1 interaction and the LLPS of G3BP1¹². In addition, co-overexpression of Caprin1 with G3BP1 decreases the intracellular concentration of G3BP1 required to initiate SG assembly in G3BP1 in G3BP1/2 double knock out cells¹². Finally, it has been suggested that Caprin1 may increase the RNA-binding affinity of G3BP1²⁵ or may also trigger a conformational change within the Caprin1-G3BP1 complex that promotes heterotypic interactions between G3BP1 IDRs³¹.

The relationship between G3BP1 and Caprin1 is at the center of the SG assembly process and in this context we explore the hypothesis that Caprin1 and G3BP1 can form multivalent complexes in which Caprin1 modulates the recruitment of RNA into SGs. Using high-throughput analyzes in a cellular context and high-resolution approaches, we demonstrate that the association of Caprin1 and G3BP1 in the presence of mRNA, occurs through homo- and heterotypic interactions allowing the formation of multimolecular complexes and leads to better recruitment of mRNA. However, the interaction between the Caprin1-G3BP1 complex and mRNA is only retained if G3BP1 still has its ability to bind to RNA, or in other words, Caprin1 cannot overcome a deficiency of G3BP1 in mRNA recruitment. On the other hand, Caprin1 can favor the recruitment of mRNA to SGs even if G3BP1 lacks its central IDR2, whose function is essential in the conformational changes of G3BP1. The results obtained shed light on the reasons for the centrality of G3BP1 in the dynamics of SGs while this protein has a moderate affinity with mRNA in the µM range with no consensus sequence²⁵. In addition, G3BP1 does not harbor a Low Complexity Domain (LCD) or Prion Like Domain (PrLD), domains which are widely represented and of importance in proteins involved in the formation of biomolecular condensates³².

Materials and methods

Plasmid preparation

Plasmids encoding Caprin1, G3BP2, FXR1, FXR2, KIF5C PABPC1L and EIF4B fused to GFP-tag on C-terminus were constructed by PCR amplification of cDNAs of genes of interest with primers containing appropriate restriction sites and insertion into pEGFP-N1 vector (Clontech).

Plasmids containing cDNA of the G3BP1-ΔC (aa 1–230) and G3BP1-ΔN (aa 230–466) were obtained from SABNP Lab and full-length G3BP1 and YBX1 fused with RFP/GFP-MBD (microtubule binding domain of Tau) were obtained previously³³. The constructs harboring cDNA for, G3BP1-ΔIDR2 fused with C-terminal RFP-MBD, Caprin1 and PABPC1L fused with C-terminal GFP-MBD were obtained using the gateway strategy as described in³⁴. Briefly, cDNA was amplified with primers containing PacI and AscI sites, cleaved with the corresponding restriction enzymes and inserted into the backbone entry plasmid RFP/GFP-pCR8/GW/TOPO previously cut with PacI and AscI enzymes. Next, LR recombination reactions (Invitrogen™) were performed to clone the cDNA of genes of interest fused with RFP/GFP-MBD into Gateway® pEF-DEST51 plasmid (Invitrogen™) following the manufacturer’s protocol. In the case of G3BP1-ΔIDR2-RFP-MBD, two fragments of cDNA encoding G3BP1 amino acids 1–221 and 340–466 respectively were amplified using pairs of primers with PacI-BamHI and BamHI-AscI restriction sites. Following ligation of the two fragments using T4 DNA ligase (ThermoFisher), the product was gel-purified and used as a template for a follow up PCR with primers containing PacI and AscI restriction sites. The gateway cloning strategy described above was used to obtain - G3BP1-ΔIDR2-RFP-MBD in the pEF-DEST51 vector.

Plasmids encoding G3BP1, G3BP1-ΔC, G3BP1-ΔN and G3BP1-ΔIDR2 fused with RFP-tag on N-terminus were constructed by amplification corresponding cDNAs with primers containing appropriate digestion sites and inserted into the mRFP-C1 vector (Addgene).

Site-directed mutagenesis of the residues F15W or F33W of human G3BP1 was carried out directly on the G3BP1-RFP-MBD-pEF-DEST51 expression plasmid by using the ‘Quikchange II XL site-directed mutagenesis kit’ (Stratagene) and appropriate oligonucleotides (Eurofins Genomics).

Sequences of all obtained plasmids and introduced point mutations were verified by DNA sequencing (Eurofins Genomics)

Cell culture

HEK293 (Human Embryonic Kidney 293, American Type Cell Collection, USA) cell line was used for cell extract preparation in immunoprecipitation experiments, and HeLa cells (American Type Collection, USA) were used in immunofluorescence experiments due to their size, shape and well-defined microtubule network. Cells were maintained at 37 °C 5% CO₂ in high glucose DMEM (Dulbecco`s Modified Eagle Medium, Life Technologies) with the addition of 10% FBS (Fetal Bovine Serum, Thermofisher) and a combination of penicillin at 100 U/ml and streptomycin at 100 µg/ml.

Microtubule bench assay: recruitment and mixing experiments

Hela cells were seeded in 96-well plates with a density of 15000 cells/well (PhenoPlate^tm −96, Perkin Elmer) and co-transfected the following day with indicated plasmids using Lipofectamine 2000 according to the manufacturer’s instructions for 22–24 hours (Supplementary Figures S1A and S5A). For recruitment experiments, GFP/RFP-MBD plasmids were used at starting quantities of 0.7 µg/well while GFP/RFP plasmids were used at starting quantities of 0.2 µg/well and then adjusted to normalize expression levels to be similar for different plasmids. For mixing experiments GFP/RFP-MBD plasmids were used at starting quantities 0.6 µg/well and then also normalized. After 22–24 hours of transfection, before the fixing step, cells were rinsed once with PBS. The fixing procedure was conducted in 2 steps; firstly, cells were incubated in ice-cold methanol for 10 minutes at −20°C, then washed with PBS and after that, 4% PFA (paraformaldehyde) in PBS was used to fix the cells during 25 min at 37°C. After rinsing with PBS 3 times, cells were stained with 250 mM 4’,6-diamidino-2-phenylindole (DAPI) to visualize the nuclei. Cell images were obtained with Opera Phoenix Plus High-Content Screening System (Perkin Elmer) on 40x magnification with water immersion.

Immunofluorescence staining for endogenous proteins visualization

Fixed cells were incubated in blocking buffer (NaCl 1M, Tris-HCl 1M pH 7.4, Triton x100 1%, BSA 2%, sodium azide 0.1%, filtered with 22 µm filter) for 30 minutes at 37°C. Next, primary antibodies (rabbit Anti-G3BP1 polyclonal antibody, Sigma, rabbit anti-Caprin1 polyclonal antibody, Proteintech) diluted 1:2000 in blocking buffer, were added and incubated overnight on a shaker at 4°C. The next day, cells were washed twice with PBS and the corresponding fluorophore-tagged secondary antibodies (goat anti-rabbit, Alexa 488, Alexa 594, Invitrogen) diluted 1:2000 in blocking buffer, were added to it, left on a shaker for 1 hour at RT. Cells were washed 3 times with PBS afterward.

RNA in situ hybridization

mRNA hybridization in situ was performed to visualize poly(A) mRNA. Firstly, fixed cells were incubated with 70% ethanol for 10 minutes at RT and then in Tris-HCl 1M pH 8 for 5 minutes at RT. Then the cells were incubated with an oligo-dT-(Cy5) fluorescent probe (Molecular Probes Life Tech.) for 2 hours at 37 °C in hybridization buffer (yeast tRNA 0.1 %, 0.001 % BSA, 10 % dextran sulfate, 25% formamide and 10 % of 20 x SSC buffer) Washings were carried out using 4×SSC buffer twice and then twice with 2×SSC buffer.

Overexpression and siRNA treatment for high-content imaging stress granule assay

For overexpression and silencing experiments, HeLa cells were seeded on a 96-well plate with 1.3x10⁴ cells/well density and transfected with plasmids expressing the protein of interest at a final concentration up to 0.3 µg/well using Lipofectamine 2000 transfection reagent (Thermofisher) for about 22–24 hours according to manufacturer’s instruction.

For silencing of G3BP1 experiments, cells were transfected with corresponding small interfering RNA (siRNA) (GeneSolution G3BP1 siRNA, Qiagen) at a final concentration of 0.25 µg/well using Lipofectamine 2000 for 24 hours. To detect non-specific effects, a negative control siRNA (AllStars Negative Control siRNA) was used. The silencing efficiency was measured on a single cell level by comparing cytoplasm fluorescence intensity of endogenous G3BP1 for cells treated with siRNA targeting endogenous G3BP1 versus cells treated with negative control siRNA.

Addback experiments

For add-back experiments, HeLa cells were seeded on a 96-well plate with 1.1x10⁴ cells per well and transfected with siRNA targeting 3`UTR region of G3BP1 mRNA at a final concentration of 0.25 µg/well using Lipofectamine 2000 for 24 hours. Then cells were co-transfected with the plasmids encoding full-length G3BP1 or G3BP1 mutants fused to RFP tag and Caprin1-GFP or with the plasmid encoding GFP tag. The efficiency of silencing was measured on a single cell level by comparing cytoplasm fluorescence intensity of endogenous G3BP1 for cells treated with siRNA targeting 3`UTR of G3BP1 mRNA versus cells treated with negative control siRNA. For overexpression, silencing and add-back experiments, HeLa cells were treated with 300 µM sodium arsenite for 1 hour at 37 °C with 5% CO₂ to induce oxidative stress and generate stress granules.

Data acquisition for MT bench assays, overexpression, silencing and add-back experiments

For both MT bench recruitment and mixing experiments cell images were obtained using Opera Phoenix Plus High-Content Screening System (Perkin Elmer) on 40x magnification in confocal mode with water immersion. For overexpression, silencing and add-back experiments cell images were obtained on 20x magnification in confocal mode.

Up to four channels were used to capture the images with the following values for excitation and emission:

1. DAPI (excitation 405 nm; emission 435–480 nm)

2. Cy2 (excitation 488 nm; emission 500–550 nm)

3. Cy3 (excitation 561 nm; emission 570–630 nm)

4. Cy5 (excitation 640 nm; emission 650–760 nm)

Images were analyzed using Harmony 5.2 software. DAPI signal was used for nuclei detection, Cy2 (for GFP-tagged proteins) and Cy3 (for RFP-tagged proteins) channels were used to detect proteins (fluorescent-tag fused recombinant proteins or indirectly, using secondary fluorochrome-tagged antibodies in case of endogenous proteins) (Supplementary Figure S1B and S5B) and Cy5 channel was used to detect mRNA through oligo-dT-(Cy5) fluorescent probe (for MT-bench recruitment experiments, Supplementary Figure S1B). HCS system automatically detects fluorescence intensity of nuclei, cytoplasm and objects of interest – spots (microtubules with RFP/GFP-MBD protein attached). Fluorescence analysis included the processing of the signal by selecting only co-transfected cells, filtering out cells with low co-transfection levels (RFP/GFP fluorescence intensity in the cytoplasm >300) and spots with low width to length ratio (<0.2). The channel corresponding to the fluorescence signal of the RFP/GFP-MBD protein was selected to detect microtubules, and a list of readings was acquired. These readings include, the number of cells analyzed, the cytoplasm fluorescence of GFP, RFP and mRNA (for MT bench recruitment experiments, Supplementary Figure S1C) in corresponding channels for each cell, the morphology of each cell (roundness, cytoplasm area), number of spots detected and, on the spot level, spots fluorescence of GFP, RFP and mRNA (for MT bench recruitment experiments, Supplementary Figure S1C) in corresponding channels in each spot and width and length parameters for each spot (Supplementary Figures S1C and S5C).

For overexpression, silencing and add-back experiments similarly, the HCS system automatically detects the fluorescence intensity of nuclei, cytoplasm and objects of interest – spots (stress granules). In silencing and addback experiments, the threshold values of fluorescence intensity in the cytoplasm for silenced protein were set to cut off background signal and obtain the data from all the cells in the well. The channel corresponding to the fluorescence signal of the oligo-dT-(Cy5) fluorescent probe was selected to detect stress granules, and a list of readings was acquired. These readings include: the number of cells analyzed, the cytoplasm fluorescence of GFP, RFP and mRNA in corresponding channels for each cell, morphology of each cell (roundness, cytoplasm area), number of spots detected, area of spots detected, and on the spot level, spot fluorescence of GFP, RFP and mRNA in corresponding channels in each spot.

HCS experiments data processing

The datasets acquired from Harmony 5.2 software analysis were presented in the form of scatterplots and violin plots using MATLAB R2021b software. For microtubule bench recruitment and mixing experiments, The RFP/GFP/Cy5 fluorescence intensity values of each spot were normalized to the values of RFP/GFP/Cy5 intensity in the cytoplasm (Supplementary Figures S1D and S5D). The mixing scores were calculated as the squared value of the correlation coefficient (corr function in MATLAB R2021b software) between mean values of fluorescence intensity in spots to mean values of fluorescence intensity in cytoplasm of RFP-MBD-fused protein to the same parameter of GFP-MBD-fused protein showing how well the observed data fit the regression model (Supplementary Figure S5D). For MT-bench recruitment experiments, the recruitment scores were calculated as correlation coefficient (corr function in MATLAB R2021b) between mean values of fluorescence intensity in spots to mean values of fluorescence intensity in cytoplasm on bait protein to the same parameter of prey, be that mRNA or protein.

In silencing and addback experiments, all data are presented in the form of a violin plot (violin function in MATLAB R2021b) for mean per cell quantification, and mean values per cell-per well were presented in form of univariate scatterplot (UnivarScatter function in MATLAB R2021b software). The enrichment values for proteins and mRNA are quantified as the mean per cell(well) ratio between the level of fluorescence in spots (SGs) to mean per cell(well) level of fluorescence in the cytoplasm. For the total SG area per cell, the mean value of all spots (SGs) per cell was selected. For total mRNA enrichment parameter, the mean values of the single spot area of SGs per cell were multiplied by the mean value of the number of spots per cell and mean value of the mRNA enrichment per cell. Image from Servier Medical Art image database (https://smart.servier.com) were used to create experiment schematics linked to the MT bench. Figure 5 was created using Microsoft PowerPoint 2013 (http://office.microsoft.com).

Figure 5.

Caprin1 decreases the RNA and G3BP1 concentration thresholds required for the formation of G3BP1-rich condensates.

Statistical analysis

All statistical tests were performed using MATLAB. The two-sample t test test was performed using the ttest2 function. Significance levels were indicated as *p < 0.05, **p < 0.01, ***p < 0.001 and not significant (n.s.).

Protein production and purification

Two proteins (the human Caprin-1 (1 aa - 709 aa) and the human G3BP1 (1 aa - 466 aa)) as well as a fragment of G3BP1 (G3BP1-ΔIDR2, described in plasmid production section) were used for in vitro experiments including electrophoretic mobility gel shift assay (EMSA) and Atomic Force Microscopy (AFM).

These protein fragments were overexpressed with a His6-tag for purification purposes. G3BP1-ΔIDR2-His6 was obtained using previously obtained RFP-G3BP1-ΔIDR2 plasmid. The protein fragments were overexpressed in E. coli BL21 (DE3) at 37 °C in 1 L of 2YT-ampicillin medium. After the optical density of cultures reached a value of 0.7, 1 mM of IPTG was added and the growth was maintained in agitation for 3 h at 37 °C. The cells were harvested and washed with 20 mL of cold 10 mM Tris-HCl buffer, pH 7.6, containing 100 mM KCl. The cell pellet was resuspended in 20 ml of Buffer solution A (Tris-HCl 25 mM, KCl 2M, DTT 1 mM, Complete EDTA-free Protease Inhibitors, PMSF 1 mM, Urea 8 M, pH 7.6) following the manufacturer’s recommendations. Cells were disrupted by sonication on ice (Bioblock Vibracell sonicator, model 72412). The resulting suspension was centrifuged at 4 °C for 30 min at 150,000xg in a TL100 Beckman centrifuge.

Full-length G3BP1 and Caprin1 proteins as well as G3BP1-ΔIDR2 construct were purified using Ni²⁺ - NTA-agarose (Qiagen) following the manufacturer’s recommendations. The supernatant was incubated for 2 h at 4 °C with Ni²⁺ - NTA-agarose (Qiagen) (20 mg of proteins/ml of resin) pre-equilibrated in buffer A containing 10mM imidazole. The resin was then washed extensively with buffer A containing 10 mM imidazole then with buffer solution A containing 15 mM imidazole. Next 3 washes were performed in Buffer solution B (Tris-HCl 25 mM, KCl 200 mM, DTT 1 mM, Complete EDTA-free Protease Inhibitors, PMSF 1 mM, Urea 8 M, pH 7.6) with an increasing imidazole concentration from 20 mM to 50 mM. The elution of the proteins was performed in Buffer solution B by progressively increasing the imidazole concentrations (from 75 mM to 300 mM). After visualizing the purification steps by SDS-PAGE (Laemmli, 1970) using a 12% polyacrylamide gel, pure protein fractions were pooled and buffer-exchanged against desalting buffer (HEPES 50 mM, KCl 25 mM, TCEP 1 mM Urea 8 M pH 7.6) by using a Superdex PD-10 column (GE Healthcare). The final preparations were snap-frozen and stored at −80°C.

Immunoprecipitation assay and western blot

HEK293 cells were grown on 6-well plates to 90% confluence and co-transfected with indicated plasmids using Lipofectamine 2000 according to the manufacturer’s instructions for 22–24 hours. RFP-G3BP1 FL, -G3BP1-ΔC, -G3BP1-ΔN and -G3BP1-ΔIDR2 plasmids were used at 3 µg/well, Caprin1-GFP plasmid was used at 2 µg/well and YBX-1-GFP plasmid was used at 1 µg/well. The cells were placed on ice and rinsed twice with PBS, then lysis buffer (50 mM Tris, 150 mM NaCl, 1% NP40, 100 mM of PMSF with RNAse A and protease inhibitors added) was added and the samples were incubated on ice for 10 minutes. Afterward, lysed cells were transferred to 1,5 ml tubes and centrifuged for 10 minutes at 10000 g at 4°C. The supernatant was collected into fresh tubes and inputs were prepared (10 µl of lysate and 5 µl of Laemmli buffer 5x).

For immunoprecipitation, 1 µl per sample of binding antibodies (Mouse Anti-GFP monoclonal antibody, Sigma) was added to 450 µl of supernatant, and incubated for 1 hour at 4 °C on multivortex. Afterwards, 30 µl per sample of pre-washed (3 times with lysis buffer) protein G sepharose fastflow beads were added to IP samples and incubated for 1 hour at 4 °C on multivortex. Next, the supernatant was removed the sample was washed 4 times with IP wash buffer (50 mM Tris, 150 mM NaCl, 0,1% NP40) and samples were prepared for SDS-PAGE by adding 30 µl of 5x Laemmli buffer per sample and heated up for 5 minutes.

For Western blot analysis, first, proteins were separated on 10 % SDS-PAGE and transferred onto Amersham Protran 0.45 NC nitrocellulose Western blotting membrane (Cytiva). Afterward, the membrane was blocked with non-fat dry milk 5% for 1 hour at room temperature then washed with TBS-Tween buffer (20 mM Trizma Base, 143 mM NaCl, pH 7.6, 1% Tween 20) incubated overnight at 4 °C with primary antibodies (rabbit TagRFP Polyclonal Antibody, Thermofisher, 1:5000, Mouse Anti-GFP monoclonal antibody, Sigma, 1:5000). After a wash step, the secondary antibodies (LI-COR IRDye, IRDye 800CW goat-anti-rabbit 1:5000, IRDye 680RD goat-anti-mouse 1:5000) were added to the membrane in TBS-Tween buffer for 1 hour at RT. The membrane was washed with TBS-Tween buffer, bound antibodies were detected with Amersham Typhoon Bioimager.

Electrophoretic mobility shift assay (EMSA)

Purified G3BP1, G3BP1-ΔIDR2 and Caprin1 proteins were serially diluted in HEPES 50 mM, KCl 25 mM, TCEP 1 mM pH 7.6 buffer solution until final concentration of 300 mM/well of Urea was achieved. M13mp18 single-stranded DNA (New England Biolabs) was incubated at 95 °C for 5 minutes, cooled down on ice and diluted in HEPES 50 mM, KCl 25 mM, TCEP 1 mM pH 7.6 buffer solution to contain 100 ng/well. In the case of mixtures of G3BP1 and Caprin1, proteins were pre-mixed and incubated with ssDNA for 10 minutes before loading. Samples were run on a 0.5% agarose gel (with Ethidium Bromide 5% in TAE 0.5x) for 1.5 hours at 25V.

Atomic force microscopy (AFM) imaging

Full length G3BP1 and Caprin1 proteins were incubated at room temperature alone or mixed at indicated concentrations with 2luc mRNA (2 ng/mL) in a Tris buffer (10 mM Tris, 15 mM KCl, 2 mM MgCl₂, 1 mM DTT, 10 mM Putrescine, pH 6.8). A 10 µL droplet was deposited on a freshly cleaved mica surface which was quickly immersed in a diluted uranyl acetate solution (0.02 % in water) to fix the sample. The samples were dried with filter paper before imaging. AFM scans were obtained using PeakForce tapping mode in air with Nanoscope V Multimode 8 software (Bruker, Santa Barbara, CA). This model enables continuous force-distance curves recording using Scanasyst-Air probes (Bruker). Images were captured at 1512x1512 pixels at a line rate of 1.5Hz. The “particle analysis” tool on the Nanoscope Analysis software (version 1.70) was used to determine the molecular dimensions of particles (protein aggregates and mRNA:protein complexes) from at least two independent samples. Basically, a threshold of 200 nm² was used to discard small particles (free proteins) or patterns (uranyl acetate background) from the analysis and at least 6 scanned areas (total area = 100 µm²) were analyzed³⁵.

Results

Caprin1 and G3BP1 colocalized in cellular condensates

The microtubule network was used as a cytoplasmic intracellular bench to determine whether two proteins interact with each other in the cytoplasm. G3BP1 protein fused with an RFP tag and a microtubule-binding domain was overexpressed in cells to serve as a bait³³. In addition, a prey protein with a GFP tag was overexpressed in the same cells (Figure 1A). If the prey protein colocalizes with the microtubule network in cells expressing the bait protein, it indicates that the two proteins interact with each other either directly or indirectly (Supplementary Figure S1). This approach reveals interactions in a cellular context, without any cell lysis or protein purification. A set of particular RBPs was chosen: all selected proteins have primarily cytoplasmic localization since G3BP1 is also a cytoplasmic RBP and are recruited in SGs. Then the selected RBPs should have different RNA binding domains like RRM (EiF4B, G3BP2, PABPC1L), CSD (YBX1) or KH (FXR1 and 2). Among selected proteins, Caprin1 is the only protein which harbors only a RGG for its binding to mRNA. The modulation of the nature and the number (from 1 to 4) of their RNA binding domain is important as the interplay between the bait and the prey could involve mRNA. Finally, the selected RBPs harbor one or multiple IDRs like G3BP1 since IDRs are important in protein-protein interaction and are involved in various RNA functions associated with the mRNA life cycle (translation regulation, RNA transport, stress granule assembly…). KIF5C is a kinesin, a molecular transporter that trafficks along microtubules and is not a RBP³⁶. With GFP, they are used as negative control.

Figure 1:

G3BP1 and Caprin1 interact in cellular context. Schematic representation of the Microtubule Bench assay. In brief, G3BP1 (bait) fused to microtubule-associated domains of Tau (MBD) and RFP is brought onto microtubules in living cells whereas the presence of a GFP-fused protein partner (prey) on microtubules reveals the interaction by colocalization of the fluorescence signals on microtubules. Top: HeLa cells were co-transfected with the plasmids encoding G3BP1-RFP-MBD and the plasmid expressing the full length Caprin1, YBX1, G3BP2 or EIF4B, all fused to GFP. The scale bar represents 3 μm. Bottom: Scatter plot representing the colocalization level of MBD-fused G3BP1-RFP with one of the eight tested proteins. Each data point represents the average correlation coefficient between fluorescence intensities from red and green channels in one well. The plot shows the data from four independent experiments. Red lines show mean values. Significances between correlation coefficients were obtained using t test; ∗∗∗p < 0.005. Representative images of HeLa cells subjected to oxidative stress via 300 µM sodium arsenite treatment expressing GFP-labeled proteins, with endogenous G3BP1 tracked via specific antibodies, and mRNA tracked via oligo-dT-(Cy5) fluorescent probe. Scale bar: 20 µm. Scatter plot representing the relative protein-GFP enrichment in SGs with (si-G3BP1) or without (si-Neg) decreasing endogenous G3BP1 levels with si-RNA. Each dot corresponds to the average enrichment in all SGs present in the cytoplasm of all cells analyzed per well. Significances between protein enrichment levels were obtained using t test; _*∗p < 0.01, ∗∗∗p < 0.005; ns, not significant. Mean value are below each scatter plot. Same as (D) with the RNA enrichment in SGs.

With G3BP1 as bait, Caprin1 was the RBP that best colocalizes with G3BP1 compartments on MTs. All other RBPs have a lower correlation coefficient with G3BP1, even G3BP2 which is a known partner of G3BP1 (Figure 1B). Note that no prey is detected on MTs if the bait with its Microtubule Binding domain is not expressed in the cell³⁷. Thus from this analysis in a cellular context, we confirm that Caprin1 interacts with G3BP1. G3BP1 being an essential factor within the network of interaction involved in SG assembly, it is interesting to correlate the expression of G3BP1 with the recruitment of various proteins fused with GFP tag to SGs in the conditions of oxidative stress using sodium arsenite treatment (Figure 1C). Among selected proteins, 3 are RBPs (YBX1, Caprin1 and G3BP2) recruited in SGs while KIF5C has no affinity for SGs and will play the role of the negative control. In the case where the cells are treated by a non-targeting siRNA (si-Neg), the GFP-RBPs are enriched in the SGs compared to the GFP-KIF5C control (Figure 1D and supplementary figure S2A). G3BP2 (+19.7%) and YBX1 (+14.8%), due to their role in the assembly/disassembly of SGs and their role as nodes, are particularly recruited in SGs compared to Caprin1. If the expression of G3BP1 is reduced by G3BP1 siRNA treatment (supplementary file S3), then the enrichment of Caprin1 in SGs is the most affected (−10.2%) while G3BP2 enrichment increases, which may be explained as compensatory mechanism in case of the deficiency of its paralog (Figure 1D). In parallel, in the same cells, it is also possible to follow the enrichment of mRNA in the SGs using a poly(T) probe coupled with a cyanine5 fluorophore (Figure 1E and supplementary figure S2B). It appears that the overexpression of Caprin1 makes it possible to attract as much mRNA into the SGs as YBX1 or G3BP2 although its enrichment is significantly lower than the other two RBPs (Figure 1D).

Caprin1 promotes the mRNA recruitment to G3BP1

We demonstrated that, in a cellular context, the presence of G3BP1 promotes the recruitment of Caprin1, whether on MTs or in SGs. Analysis of the RNA composition of SGs suggests that Caprin1 could promote the recruitment of mRNA. Given the variety of constituents in these condensates, it may be difficult to attribute the increase of the RNA content in SGs only to Caprin1 alone. We then evaluate the influence of Caprin1 on the mRNA recruitment by G3BP1 using the MT bench assay with Caprin1 as prey and G3BP1 fused to RFP-MBD as bait via coexpression of the recombinant proteins Caprin1-GFP and G3BP1-RFP-MBD or, in a second experiment, via tracking the endogenous Caprin1 recruitment to G3BP1-RFP-MBD compartments using corresponding antibody (Figure 2A and supplementary figure S4A). The overexpression of Caprin1 leads to a strong increase in the colocalization of the GFP signal with the RFP signal on MTs. Thus, Caprin1 is more recruited to G3BP1 on MTs compared to the level of G3BP1 colocalization with endogenous Caprin1, while the G3BP1 enrichment on MTs is slightly affected by the Caprin1-GFP overexpression (supplementary figure S4B). In parallel, we measured mRNA enrichment in both conditions. Whenever Caprin1 is overexpressed, we detected an increase in the correlation between the RFP signal due to G3BP1 brought on MTs and the cyanine5 fluorophore signal (mRNA). Thus, the increase in the recruitment of Caprin1 to MTs is concomitant with the increase in mRNA recruitment. While in SGs, it was difficult to attribute the mRNA enrichment only to Caprin1 alone, the MT bench confirms that the presence of Caprin1 promotes the recruitment of mRNA on G3BP1. This effect can be analyzed in detail after purifying the proteins of interest. The EMSA results demonstrate that, firstly, both full-length G3BP1 and Caprin1 functionally bind single stranded nucleic acids. Compared to G3BP1, Caprin1 decreases the electrophoretic mobility of ssDNA at a lower protein to nucleotide ratio than G3BP1 (Figure 2B). The electrophoretic mobility profile of both proteins allowed us to select a range of protein to nucleotide ratios close to the threshold required to generate a reduction in the mobility of ssDNA by G3BP1. When both proteins are associated in the same sample, their mixing with ssDNA triggers the formation of protein-nucleic acid complexes at lower G3BP1 concentration (Figure 2C). Indeed, 22 pmol of Caprin1-G3BP1 mix triggers the reduction of the electrophoretic mobility of ssDNA while the same amount of G3BP1 alone has no such effect. In addition, part of the ternary complexes produced by the same amount of protein mix is unable to enter the gel. Thus, the Caprin1-G3BP1 mixture, through heterotypic interactions, is able to bind more nucleic acid than G3BP1 alone and to form complexes with different sizes. Finally, the protein/ssDNA complex presents different mobility depending on whether it is previously formed with G3BP1 or Caprin1 (Figure 2D). Indeed, the progressive addition of Caprin1 to a preformed complex with G3BP1 leads to a decrease in the mobility of the complex until reaching a size unable to migrate through the gel. Conversely, G3BP1 integrates into a small proportion of the complexes preformed with Caprin1 and allows the formation of large structures while the mobility of part of the protein/ssDNA complexes is not significantly modified.

Figure 2.

Caprin1 promotes G3BP1 RNA recruitement. Top: scatter plot representing the colocalization level between G3BP1-RFP-MBD brought on MT and Caprin1 or mRNA in the presence of endogenous level of Caprin1 or overexpressed Caprin1-GFP. Each data point represents the average correlation coefficient between fluorescence intensities from RFP and GFP or cyanin5 in one well. Red lines show mean values. Significances between correlation coefficients were obtained using t test; ∗∗∗p < 0.005. Bottom: schematic representation of the MT bench principle and the detection scheme used for endogenous Caprin1 and mRNA. ssDNA mobility shift assay demonstrating the different interaction between Caprin1 or G3BP1 and single stranded nucleic acid. 100 ng of m13 ssDNA were incubated with increasing concentrations of G3BP1 (lanes 2–6) or Caprin1 (lanes 8–12). ssDNA mobility shift assay demonstrating the cooperation between Caprin1 and G3BP1 for binding to single stranded nucleic acids. 100 ng of m13 ssDNA were incubated with increasing concentrations of a mix between G3BP1 and Caprin1 (lane 2–6), Caprin1 alone (lanes 7–11) or G3BP1 alone (lanes 12–16). ssDNA mobility shift assay demonstrating the different integration of Caprin1 into preformed G3BP1-m13 ssDNA complexes (lanes 2–7) or oppositely, of G3BP1 into preformed Caprin1-m13 ssDNA complexes (lanes 8–13). 100 ng of m13 ssDNA were incubated for 1 min with 16 pmol of G3BP1 before the addition of increasing concentration of Caprin1 or with 12 pmol of Caprin1 before the addition of increasing concentration of G3BP1. Left: schematic representation of the mixing/demixing experiment. Two RBPs, as indicated, are confined on the microtubule network (fused to RFP/GFP-MBD) to visualize their mixing/demixing in HeLa cells. Mixing: yellow microtubules. Demixing: red and green microtubules. Middle: scatter plot representing the mixing between G3BP1 and different RBPs, both fused to MBD. Each data point represents the average value obtained in cells analyzed in one well. The mixing score is the value of the determination coefficient R² calculated as described in the experimental procedures section. The plot shows the data from eight independent experiments. Red lines show mean values. Significances between protein enrichment levels were obtained using t test; ∗∗∗p < 0.005. Right: representative images for a low (PABPC1L) and a high (G3BP1 and Caprin1) or medium (YBX1) mixing with G3BP1. Top: atomic force microscopy images and zoom-in on specific assemblies of G3BP1/RNA, Caprin1/RNA, and of mixture of the two RBPs with 2luc mRNA. Proteins (20 nM of G3BP1 or 10 nM of Caprin1 or a mixture of 10 nM G3BP1/5 nM Caprin1) were incubated with 2luc mRNA (protein/nucleotide ratio of 1/100) for 5 min before sample deposition and fixation. White arrows pointed to RNA/RBP complexes. Z scale 8 nm. The scale bar represents 400 nm (200 nm in zooms). Bottom: scatter plot representing the area of isolated RBP/RNA assemblies observed on AFM images and comparison with free mRNA. Only areas of RNA or RBP/RNA complexes with a value higher than 200 nm² are plotted to discard free proteins from this analysis. Each point corresponds to the same surface analyzed, here 100 μm². The plot gathers the data from two independent experiments. Red lines show mean values. Significances between samples were obtained using t test; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.005.

Analysis by EMSA demonstrates the capacity of Caprin1 and G3BP1 to bind to the same nucleic acids and therefore to interact together. Such behavior relies on the ability of these two proteins to mix, although G3BP1 has a tendency to form homotypic interactions. The microtubule network can be used to determine the propensity of proteins to mix in common compartments³⁴. In this analysis, proteins of interest were fused to a microtubule-binding domain and after their expression in cells, were brought onto microtubules to generate compartments (demixing) or, on the opposite, a homogeneous phase (mixing, Figure 2E and Supplementary Figure S5). In the case of RBPs, the demixing involves homotypic protein-protein interactions, while RNA could provide a scaffold for higher protein assemblies via RNA-protein interactions. When Caprin1 and G3BP1 are brought to the MTs, the MTs adopt intermediate yellow coloration indicating a high level of mixing which is evaluated by a mixing score close to 1. The mixing score here is R², the square of the correlation coefficient. Conversely, the mixing score of G3BP1 with YBX1 or PABPC1L is low and the MTs have a less uniformed color with red and green clusters detected, indicating the formation of compartments rich in G3BP1 or YBX1 (or PABPC1L). Thus, Caprin1 and G3BP1 can mix into the same compartments in a process most probably facilitated by their binding to the mRNAs present in these compartments.

The mRNA-protein complexes formed by G3BP1, Caprin1 or a mixture of G3BP1 and Caprin1, were observed by Atomic Force Microscopy (Figure 2F). High resolution AFM enables the nanometer scale analysis of multimolecular assemblies. G3BP1 or Caprin1 were incubated independently with mRNA for a few minutes at the indicated concentrations. These two proteins form isolated complexes with the mRNA, the Caprin1-mRNA complexes exhibiting a more compacted structure than G3BP1-mRNA complexes, in agreement with the helicase function attributed to G3BP1³⁸. When the two proteins are incubated together with mRNA for the same incubation time and comparable protein nucleotide ratio, we observe the appearance of large condensates and a decrease in the number of isolated complexes symbolized by the average increase in the area of ​​the complexes. Thus, the association between Caprin1 and G3BP1 promotes protein-RNA complex formation compared to G3BP1 alone but also leads to the assembly of large condensates.

RNA recruitment by G3BP1-Caprin1 complex requires the RNA-binding domain of G3BP1

Having demonstrated that the overexpression of both G3BP1 and Caprin1 triggers higher recruitment of mRNA in SGs than overexpression of only one of these proteins, next we addressed the question of the contribution of each protein in this process. Indeed, Caprin1 is able to bind mRNA in vitro through its RGG domain as observed on AFM images and EMSA. However, we cannot affirm that Caprin1 can recruit mRNA into condensates in a cellular context by itself. To decipher this point we produced different G3BP1 truncation mutants, in particular G3BP1-ΔC in which the RRM and RGG (IDR3) domains of the C-terminal part were removed and G3BP1-ΔN which no longer possesses the NTF2L domain which harbor the interaction site with Caprin1 as well as the IDR1 required for G3BP1 self-association (Figure 3A). In order to specifically study the importance of G3BP1-Caprin1 binding in RNA recruitment to G3BP1 condensates, we also tested two mutants of G3BP1, G3BP1-F15W and G3BP1-F33W which are located in the NTF2L domain and have been demonstrated to decrease the affinity with Caprin1 by a factor of 5 or 45 respectively³⁹.

Figure 3.

MT bench analysis reveals that G3BP1 RNA binding is of fundamental importance in mRNA recruitment by the G3BP1/Caprin1 complex. Schematic representation of the G3BP1 constructs. Right panel: HEK293 cells were co-transfected with Caprin1-GFP or YBX1-GFP and RFP-G3BP1 FL, RFP-G3BP1-ΔC, RFP-G3BP1-ΔN or RFP-G3BP1-ΔIDR2. G3BP1 FL - YBX1 co-transfection was used as a control. Cells were lysated with buffer containing RNAse A and immunoprecipitation was performed using a-GFP antibodies. Left panel: Western blot analysis of cell lysates. The average efficiency of GFP IP was approximately 34% for Caprin1-GFP and 3% for YBX1-GFP. Scatter plot representing the colocalization level of MBD-fused G3BP1-RFP constructs with overexpressed Caprin1-GFP. Each data point represents the average correlation coefficient between fluorescence intensities from red and green channels in one well. The plot shows the data from eight independent experiments. Red lines show mean values. Significances between correlation coefficients were obtained using t test; ∗∗∗p < 0.005; ns: not significant. Scatter plot representing the colocalization level of MBD-fused G3BP1-RFP constructs with mRNA. Each data point represents the average correlation coefficient between fluorescence intensities from red and cyanin5 channels in one well. The plot shows the data from eight independent experiments. Red lines show mean values. Significances between correlation coefficients were obtained using t test; ∗p < 0.05; ∗∗∗p < 0.005; ns: not significant. Representative images of MT in HeLa cells when transfected with the constructions encoding Caprin1-GFP and G3BP1-RFP-MBD for the different G3BP1 truncations and mutations. The scale bar represents 3 μm. Schematic representation of the MT bench experiment where Caprin1 was brought on microtubule and the detection scheme to reveal interaction with overexpressed RFP-G3BP1 constructs and mRNA. Scatter plot representing the colocalization level of MBD-fused Caprin1-GFP with different G3BP1 constructs or RFP alone as a control. Each data point represents the average correlation coefficient between fluorescence intensities from red and cyanin5 channels in one well. The plot shows the data from eight independent experiments. Red lines show mean values. Significances between correlation coefficients were obtained using t test; ∗p < 0.05; ∗∗∗p < 0.005. Same as (G) but the correlation coefficient measures the colocalization between MBD-fused Caprin-GFP and mRNA. Significances between correlation coefficients were obtained using t test: ∗∗∗p < 0.005;.

To analyze the contribution of the different domains of G3BP1 in the interaction with Caprin1 independently from RNA, we performed an immunoprecipitation (IP) assay in buffer containing RNAse A (Figure 3B). The results of IP confirmed that all G3BP1 constructs except G3BP1-ΔN, bind to Caprin1, and evidenced that YBX1 binding to G3BP1 is RNA-dependent since the absence of RNA abolished this interaction. Next, to correlate Caprin1 recruitment and the mRNA enrichment in G3BP1 compartments, we analyzed the interaction between Caprin1 and the different G3BP1 constructs using the MT bench assay. Cells were transfected with Caprin1-GFP and G3BP1 constructs fused with RFP and the microtubule binding domain (MBD) to bring G3BP1 constructs-RFP-MBD onto MTs. The same detection system used for Figure 2A was set up, allowing the recruitment of Caprin1-GFP and mRNA to be analyzed at the same time on MTs. Under these conditions, G3BP1-ΔC is still able to bring Caprin1 onto MTs unlike G3BP1-ΔN and G3BP1-F15W and F33W mutants for which only few portions of MTs are detected in the GFP channel indicating a reduced interaction with Caprin1 (Figure 3C). The measurement of the correlation coefficient between the RFP signal and that of the cyanine5 probe interacting with the mRNA indicates that the absence of the NTF2L domain or point mutations (F15W or F33W) have only a slight effect on the recruitment of mRNA on MTs (Figure 3D and 3E). On the other hand, the absence of the RNA binding domains of G3BP1 (G3BP1-ΔC) limits mRNA recruitment despite Caprin1 being present on MTs, which confirms previously demonstrated ability ofCaprin1 to interact with G3BP1 in an RNA independent manner⁴⁰. If now Caprin1 is brought to the MTs via the expression of the construct Caprin1-GFP-MBD (Figure 3F), the recruitment of both G3BP1-ΔC and G3BP1-ΔN to the MTs is very low (Figure 3G) as well as mRNA (Figure 3H). Thus, Caprin1 is an effective bait for G3BP1 only if G3BP1 still harbors both the Caprin1 interaction motif in NTF2L and its ability to interact with the mRNA.

G3BP1-ΔC and -ΔN are not recruited to SGs despite Caprin1 overexpression

G3BP1 or G3BP2 overexpression in cells without stress stimuli is sufficient to induce SG formation^10,41. Here, we overexpressed different G3BP1 constructs in HeLa cells and observed that both G3BP1-ΔC and ΔN were unable to trigger the SG assembly (Figure 4A). Indeed, we detected only small RNA rich granules in cells overexpressing the 3 truncated forms of G3BP1 compared to overexpression of G3BP1 FL (Figure 4B and Supplementary Figure S6). Since the enrichment of G3BP1-ΔC and G3BP-ΔN in these granules is very low (Figure 4C), their appearance in the cells may be explained through the presence of endogenous G3BP1 in conjunction with the stress caused by protein overexpression. The lack of the RBD in G3BP1-ΔC which leads to an RNA binding impairment, explains the absence of SG assembly. The impairment of the SG assembly when G3BP1-ΔN is overexpressed is more intriguing but is in agreement with our experimental data where this construct brought on MT is not able to recruit mRNA (Figure 3D). In addition, previous studies using G3BP1-ΔN in G3BP1 and 2 KO cells, revealed that this construct is unable to reconstitute the SG assembly which could be also explained also by the altered interactions with essential SG partners (Caprin1, UBAP2L). In addition, G3BP1-ΔN alone did not trigger condensate formation via LLPS in vitro¹². Therefore, not only the binding to protein partners but also the dimerization of G3BP1 is an essential parameter in SG assembly. The Caprin1 overexpression is known to nucleate SG assembly if G3BP1 and G3BP2 are also coexpressed¹³. Here, Caprin1 and FL or truncated forms of G3BP1 are overexpressed in HeLa cells subjected to oxidative stress. Caprin1 promotes the recruitment of mRNA into the SGs when G3BP1 FL is co-overexpressed compared to the control (overexpression of GFP alone) (Figure 4D and Supplementary Figure S8A). However, Caprin1 has no effect on mRNA enrichment in SGs when co-overexpressed with G3BP1-ΔC or ΔN nor in G3BP1 constructs recruitment in these granules in si-Neg conditions (Figure 4D and E). mRNA and G3BP1 content were then analyzed in cells previously treated with G3BP1-siRNA to decrease the expression of G3BP1 FL (Supplementary Figure S7A). si-G3BP1 treatment was used to reveal a potential effect of overexpressed Caprin1 that could be masked by the endogenous level of G3BP1. Then, the above mentioned G3BP1 constructs were overexpressed at the same time as Caprin1-GFP or GFP alone for comparison. As a control, when Caprin1 is overexpressed with G3BP1-ΔN in si-G3BP1 treated cells, the recruitment of Caprin1-GFP in stress granules is strongly affected (Supplementary Figure S7B). Consequently, the recruitment of mRNA in SG when G3BP1-ΔN is overexpressed is independent of the Caprin1 overexpression and in the same range as for G3BP1-ΔC overexpression (Figure 4D and Supplementary Figure S8B). The overexpression of Caprin1 improves the G3BP1-FL recruitment in SGs but has no effect on the G3BP1-ΔC nor the ΔN enrichment (Figure 4E and Supplementary Figure S8C).

Figure 4.

Caprin1 overexpression is not sufficient to restore mRNA recruitment in SG in cells expressing G3BP1-ΔC. Representative images of HeLa cells expressing G3BP1-FL, -ΔN, -ΔC or -ΔIDR2 plasmids. Scale bar: 40 µm. Scatter plot representing the area of mRNA-rich granules in HeLa following the overexpression of different G3BP1 constructs. Each dot corresponds to the average area of granules of all cells of a well expressing the G3BP1 construct. Data were obtained from four independent samples. Granules were detected using cyanin5 fluorescence signal. Significances between granule areas were obtained using t test; ∗∗∗p < 0.005; mean value in red. Same as (B) but the G3BP1 truncation enrichment in mRNA-rich granule is measured. Scatter plot representing the mRNA enrichment in SGs in HeLa after si-neg or si-G3BP1 treatment and add-back of different RFP-G3BP1 constructs and GFP-Caprin1 (or GFP alone as control) and treatment with sodium arsenite. Each dot corresponds to the average mRNA enrichment in all SGs of all cells of a well expressing the RFP-G3BP1 and GFP-Caprin1 (or GFP alone) constructs. Data were obtained from four independent samples. The mean value is in red. Significances were obtained using t test; ∗p < 0.05; ∗∗∗p < 0.005. Same as (D) but the plot represents the G3BP1 (full length or truncations) enrichment in SGs after si-Neg or si-G3BP1 treatment. ∗∗∗p < 0.005; ns: Not significant. Scatter plot representing the mixing between G3BP1 full length and different G3BP1 truncations (including G3BP1 full length itself as a positive control) or YBX1 as a negative control. All G3BP1 constructs and YBX1 were brought on MTs. Each data point represents the determination coefficient R² obtained from all compartments detected on MT of all cells of a well. The plot shows the data from five to eight independent experiments. Red lines show mean values (in red also). Significances between mixing scores were obtained using t test; ∗∗∗p < 0.005. Same as (F) but the mixing score is obtained between Caprin1 (or YBX1 as negative control) and different G3BP1 truncations. Representative images of mixing between Caprin1 (or YBX1) and G3BP1 (FL and truncation) brought on MT.

Thus, the results obtained in the condensates are linked to those obtained in the MT bench assaysand clearly indicate that if G3BP1 is no longer able to bind to mRNA due to the absence of its RNA binding domains, Caprin1 does not compensate for this failure and the mRNA enrichment in SGs is therefore decreased. Concerning G3BP-ΔN, while the recruitment of RNA by this construct when brought on MT was close to that of G3BP FL, G3BP1-ΔN is not recruited in SGs independently of Caprin1 overexpression and mRNA enrichment in SG is also decreased. This G3BP1 construct neither multimerizes nor interacts with its major protein partner. G3BP1-ΔN loses its centrality in condensate and is therefore unable to act as an SG nucleator. Finally, the contribution of Caprin1 in both mRNA and G3BP1 recruitment in SGs is strongly correlated to the nature of G3BP1 constructs, indicating that the action of Caprin1 in SGs relies intimately on the ability of G3BP1 to develop heterotypic interactions and thus play its role of node.

Caprin1 mixing with G3BP1-ΔIDR2 promotes its recruitment in SGs independently of mRNA.

In addition to the domains involved in mRNA binding and those in interactions with partner proteins, G3BP1 contains a weakly structured central domain (IDR2) including a prolin rich domain. The G3BP1-ΔIDR2 construct lacks this domain but still harbors both the NTF2-L and the mRNA binding domains. When this construct is brought on MT, it is unable to recruit mRNA to MTs (Figure 3D) but it can be recruited by Caprin1 if it plays the role of prey, independently of the mRNA (Figure 3G and H). Likewise, this construct is poorly recruited in the SGs (Figure 4C) but the joint overexpression of Caprin1 allows its recruitment to be significantly increased in the SGs (Figure 4E). However, mRNA enrichment in SGs remains lower than the level achieved when G3BP1 FL is overexpressed (Figure 4D) together with Caprin1. G3BP1-ΔIDR2 could interact with Caprin1 independently of the mRNA. The increase in heterotypic interactions between the Caprin1 and G3BP1-ΔIDR2 to the detriment of homotypic interactions could explain this behavior. Indeed, if the G3BP1 constructs and Caprin1 are brought together on MT, the highest mixing score is obtained with G3BP1-FL/Caprin1 and G3BP1-ΔIDR2/Caprin1 couples while the mixing score between G3BP1 ΔC (or ΔN) with Caprin1 is significantly lower (Figure 4G and H). In addition, the mixing score of G3BP1-ΔIDR2 with G3BP-FL is the lowest of all G3BP constructs (Figure 4F) and could be explained by two reasons: (i) G3BP-ΔIDR2 adopts a conformation which does not favor heterotypic interaction with G3BP1-FL, (ii) the deficiency in mRNA binding by G3BP1-ΔIDR2 preclude its mixing with mRNA rich G3BP1-FL compartments. The two explanations are not exclusive since G3BP1-ΔC construct mixes better with G3BP1-FL than G3BP1-ΔIDR2 whenever its RNA binding ability is affected.

Discussion

G3BP1 is a major player in the SG biology. G3BP1 is considered essential in the formation of SGs since cells depleted in G3BP1 and 2 and treated with sodium arsenite are not able to generate SGs. To explain the central character of G3BP in the formation of SGs, one should consider the formation of stress granules as the result of mRNA-mRNA, mRNA-protein and protein-protein interactions, these interactions being weak and transient. Above a threshold interaction density, the individual molecules form a network and the LLPS process leads to the formation of biomolecular condensates⁴². In this context, G3BP has many advantages. G3BP is of course an RNA binding protein, which represses certain transcripts, stabilizes others or affects their subcellular localization⁴³. The numerous experiments in protein phase separation, single-molecule microscopy and gel shift assay clearly demonstrate that G3BP1 does not need a partner to bind with RNA in vitro^12,40,44. However, the RNA-G3BP1 interaction seems essential to the G3BP1 phase separation because it allows conformational changes in G3BP1. Indeed under non-stress conditions, G3BP1 is in a compacted conformation stabilized by intramolecular interactions between the IDR1 and IDR3. The release of a large quantity of RNA during the dissociation of polysomes under stress conditions promotes interactions between the IDR3 (RGG rich domain) and RNA allowing G3BP1 to adopt an open conformation^12,25. This conformational change can also be the consequence of post-translational modifications such as phosphorylation^45,46 or poly(ADP-ribosyl)ation^47,48. G3BP1 thus seems perfectly designed to play a central role in the formation of biomolecular condensates such as SGs. However, while binding to mRNA seems fundamental in the SG formation process orchestrated by G3BP1, measurements of the G3BP1 dissociation constants (apparent K_D) of this protein for RNA are in the µM range with a very weak influence of the NTF2L domain and the IDR3 (RG rich domain) indicating that RNA binding is essentially due to the RRM²⁵. G3BP1 binds to a wide variety of coding and non-coding RNAs and without a consensus sequence^49,50 which could have given it an advantage in supporting RNAs compared to other RBPs. Even following the release of polysomal RNAs, G3BP1 will compete with numerous RBPs to recruit the mRNA required for its conformational changes and to trigger the LLPS process at the origin of SGs. Here, the interaction with other protein partners can provide an advantage to G3BP1 to be competitive in handling the mRNA released by polysomes and to play its key role as a scaffold in SG assembly. Among the many closed partners of G3BP1, Caprin1 is able to bind RNA through its arginine-rich C-ter domain (RGG rich domain) with a lower dissociation constant than G3BP1 (≈500 nM)⁵¹. It has been demonstrated that the presence of Caprin1 significantly reduced the threshold concentration of G3BP1 and RNA required for LLPS in simple system in vitro and that the overexpression of Caprin1 with G3BP1 in G3BP1/2 dKO cells reduced also the G3BP1 concentration necessary to initiate SG assembly¹². Our results confirm via observations at the single molecule level in vitro that Caprin1 promotes the clusterisation of mRNA-G3BP1 complexes (Figure 2F) and additionally, indicate that, in cellular context, Caprin1 improves the uptake of RNA in G3BP1-rich condensates like SGs (Figure 4D). Finally, it is essential that the ability of G3BP1 to bind RNA is maintained so that Caprin1 can play this important role. The RRM of G3BP1 and the RGG of Caprin1 therefore seem to work in symbiosis to increase the recruitment of RNA into SGs. Of course, the G3BP1-Caprin1 relationship is not hegemonic since Caprin1 can be recruited into SGs independently of G3BP1, notably through interaction with other protein factors such as FMR1⁵².

Besides, an essential characteristic of RBPs involved in the formation of biomolecular condensates is the presence of low complexity domains (LCD) in these proteins. Indeed, the LLPS process also relies on weak, transient and numerous protein-protein interactions^53,54. In fact, condensates are enriched in RBPs like FUS⁵⁵, TDP43³⁵, hnRNPA1²¹, TIA-1⁵⁶, Ataxin2⁵⁷. which harbor one or more LCD domains or even Prion Like Domains (PrLDs) whose capacity for self-attraction is even higher. In these LCDs, the aromatic residues are one of the main driving force for phase separation^58,59. G3BP1 has 3 IDRs, IDR1 which is acidic, involved in the interaction with IDR3 when the mRNA concentration is low and IDR2 which plays the role of spacer between IDR1 and the RRM allowing the intramolecular interaction between IDR1 and 3. These three IDRs are very poor in aromatic residues (2.5% of the sequence). On the other hand, Caprin1 has a long LCD (residues 537–709) with significant enrichment in these aromatic residues which represent more than 11% of the primary sequence. Thus, as for RNA binding, Caprin1 plays an essential role in the recruitment of the numerous protein partners present in SGs such as FMRP⁴⁶ or PBQP1⁶⁰ and it is possible that the centrality of G3BP1 in the formation of SGs is the consequence of the G3BP1-Caprin1 interaction (Figure 5). The case of the G3BP1-ΔIDR2 construct particularly highlights the importance of protein-protein interactions in the formation of condensates. G3BP1-ΔIDR2 fails to recruit mRNA, although the RRM and RGG domains are still present, certainly the closed proximity between the acidic IDR1 and the RRM would repeal the mRNA away from its binding site on G3BP1. Despite the expression of Caprin1, the recruitment of mRNA whether on MTs or in SGs is not improved because the ability of G3BP1 to bind to mRNA is fundamental as shown by the G3BP-ΔC construct. (Figures 3D and 4D). On the other hand, the integration of this construct into biomolecular condensates is significantly improved if Caprin1 is overexpressed (Figure 4E). G3BP1 most probably adopts an open conformation since the IDR2 spacer is no longer present to allow IDR1-IDR3 intramolecular interactions¹². Protein-protein interactions via the NTF2L domain of G3BP1, the G3BP interaction motif (GIM) and the PrLD of Caprin1 would therefore be essential in the formation of condensates with Caprin1 playing a central role.

We demonstrate here that the G3BP1-Caprin1 relationship, which exists in SGs but also independently of SGs⁶¹, is important in the recruitment of mRNA in G3BP1 condensates or complexes, as it is based both on the capacity of G3BP1 to bind to RNA but is exacerbated by the presence of Caprin1. This close relationship between G3BP1 and Caprin1 can have consequences on various pathologies, notably in the formation of pathological aggregates characteristic of numerous neurodegenerative diseases for which RNA-protein and protein-protein interactions have a pivotal role. For example, the expression of G3BP1 seems to have a regulatory activity in the aggregation process of Ataxin-2, a symptom of spinocerebellar ataxias⁶². The protective effect of G3BP1 requires the presence of the NTF2L domain but the absence of the RRM has no effect. On the other hand, a mutation near Caprin1 LCD (P521L) leads to a reduction in the affinity of Caprin1 with RNA and to the formation of aggregates in which Ataxin2 is recruited⁵¹.

Author contributions

K.L. made the acquisition, analysis and interpretation of data, prepared all figures J.C.R.G., V.J., S.P. performed experiments D.P. conceived experiments and analyzed the data S.K., L.H. conceived and supervised the project, wrote the main manuscript text All authors reviewed the manuscript

Funding

ADI program Paris Saclay University,PAUSE-ANR Ukraine,ANR-22-CE18-0029

Data availability

Source data underlying the graphs presented in the main figures are available in the Supplementary Data files. No new sequence information was generated and Sanger sequencing was performed to confirm known plasmid sequences and incorporation of point-mutations.

Declaration

Competing Interests

The authors declare no competing interests.