Abstract
Diverse RNAs and RNA-binding proteins form phase-separated, membraneless granules in cells under stress conditions. However, the role of the prevalent mRNA methylation, m6A, and its binding proteins in stress granule (SG) assembly remain unclear. Here, we show that m6A-modified mRNAs are enriched in SGs, and that m6A-binding YTHDF proteins are critical for SG formation. Depletion of YTHDF1/3 inhibits SG formation and recruitment of m6A-modified mRNAs to SGs. Both the N-terminal intrinsically disordered region and the C-terminal m6A-binding YTH domain of YTHDF proteins are crucial for SG formation. Super-resolution imaging further reveals that YTHDF proteins are in a super-saturated state, forming clusters that reside in the periphery of and at the junctions between SG core clusters, and promote SG phase separation by reducing the activation energy barrier and critical size for condensate formation. Our results reveal a new function and mechanistic insights of the m6A-binding YTHDF proteins in regulating phase separation.
Main
RNA-protein (RNP) granules are phase-separated, membraneless granules that play important roles in epigenetic and post-transcriptional regulations1–6. Stress granules (SGs) are RNP granules that assemble under various cellular stress conditions, such as oxidative, osmotic, or heat-shock stress, and regulate messenger RNA (mRNA) translation and degradation1,4,5. Defects in SG dynamics are associated with various diseases such as neurodegenerative disorders, cancers, viral infections, and autoimmune diseases7,8.
RNAs and RNA-interacting proteins are crucial components of SGs9–12. N6-methyladenosine (m6A) is the most abundant internal mRNA modification13–16, and three of the major m6A-binding proteins, YTHDF1-3 (ref. 15), are implicated in the SG proteome and can interact with SG components12,17–20. However, some key components of SGs, including the SG core proteins, G3BP1/2 (ref. 21), and their binding partners22, preferentially bind to unmodified RNA instead of m6A-modified RNAs in specific sequence contexts23. These opposing binding behaviors of SG proteins to m6A-modified RNAs and m6A-binding proteins raise the important question of whether m6A-modified RNAs and m6A-binding proteins play a role in SG formation.
Here, we studied the localization of m6A-modified mRNAs and m6A-binding YTHDF proteins in mammalian cells, and identified the m6A-binding YTHDF proteins as key regulators for SG formation. We observed an enrichment of m6A-modified mRNAs in SGs. Depletion of YTHDF1/3 proteins substantially inhibited SG formation and prevented enrichment of mRNA m6A signal in SGs. Both the N-terminal intrinsically-disordered region (IDR) and C-terminal m6A-binding YTH domain of YTHDF proteins were crucial for SG formation. Super-resolution imaging further revealed that YTHDF1 was in a super-saturated state, forming clusters that connect SG core clusters, and promoted phase separation of the SG core protein G3BP1 by lowering the activation energy barrier and reducing the critical size for G3BP1 condensate formation.
Results
Imaging m6A-modified mRNA in mammalian cells
To examine the subcellular distribution of m6A-modified mRNAs in mammalian (U-2 OS) cells, we developed an immunofluorescence protocol that specifically labels m6A-modified mRNAs (Supplementary Fig. 1a, b) and validated the labeling specificity for m6A using gene-edited U-2 OS cells in which a key m6A methyltransferase component METTL3 is knocked out (Supplementary Fig. 1c). It has been shown previously that the amount of m6A in mRNAs is reduced by 50 - 60% in METTL3 knockout cells24 and indeed, our immunofluorescence signal for mRNA m6A was reduced by ~60% in these knockout cells (Supplementary Fig. 1c).
Using this imaging approach, we observed several notable features for the distribution of mRNA m6A in unstressed cells: First, the normalized intensity of the mRNA m6A signal (normalized by polyA signal) was substantially higher in the cytoplasm than in the nucleus (Supplementary Fig. 1d), which is consistent with the effect of m6A on promoting nuclear export of mRNAs25. Second, in the cytoplasm, in addition to the diffusively distributed signals, we observed an enrichment of m6A in processing bodies (P-bodies) (Supplementary Fig. 1e).
Enrichment of m6A-modified mRNA in SGs
To study the localization of mRNA m6A under stressed conditions, we imaged mRNA m6A and polyA signals simultaneously in U-2 OS cells under oxidative stress induced by NaAsO2 treatment. NaAsO2 treatment induced the formation of numerous SGs in the cytoplasm, marked by an SG core protein G3BP1 (Fig. 1a)22. We observed strong signals of both mRNA m6A and polyA in SGs (Fig. 1a). Quantitatively, polyA showed ~3-fold enrichment in SGs as compared to elsewhere in the cytoplasm whereas m6A showed more than 4-fold enrichment in SGs (Fig. 1a), suggesting that m6A-modified mRNAs have a higher tendency to associate with SGs than unmethylated mRNAs. We also observed strong m6A signals but not polyA signals in P-bodies (Fig. 1a), which is consistent with previous observations that the m6A-binding protein YTHDF2 localizes in P-bodies26 and that deadenylation of mRNAs is a prerequisite for P-body formation27.
We then analyzed the relationship between the SG-enrichment of mRNAs determined by SG RNA sequencing28 and the m6A methylation ratio of mRNAs (defined as the fraction of transcripts that harbor m6A)29 for individual genes. We observed a strong positive correlation between m6A ratio and SG-enrichment and negative correlation between m6A ratio and SG-depletion for relatively long (>3000 nt) mRNAs (Fig. 1b). For relatively short (<3000 nt) mRNAs, few of them showed enrichment in SGs but the strong negative correlation between m6A ratio and SG-depletion maintained (Fig. 1c). We further validated the localization of various mRNAs with different m6A ratios using single-molecule fluorescent in-situ hybridization (smFISH)30,31, which also showed a higher tendency of SG-enrichment for mRNAs that are more heavily modified with m6A (Fig. 1d, e). We observed a similar trend for mRNA enrichment in other types of RNP granules, including heat-shock induced SGs32, ER-stress induced SGs32, and P-bodies33 (Supplementary Fig. 2). Taken together, our results indicate that a broad range of m6A-modified mRNAs are enriched in SGs.
RNA m6A-binding YTHDF proteins are critical for SG formation
To understand the mechanism underlying the association between m6A-modified mRNAs and SGs, we examined the roles of m6A-binding YTHDF proteins in SG assembly. We observed strong colocalization of endogenous YTHDF1/3 proteins with SGs, but not with P-bodies, which are frequently found adjacent to SGs (Fig. 2a, c; Supplementary Fig. 3a, b, c). YTHDF2, instread, showed colocalization with both SGs and P-bodies (Fig. 2b; Supplementary Fig. 3d). Notably, knockdown of either YTHDF1 or YTHDF3, but not YTHDF2, substantially reduced the number of SGs in NaAsO2-treated cells (Fig. 3a, b). Double knockdown of YTHDF1 and YTHDF3 largely abolished the formation of SGs (Fig. 3a, b). The reduction in SG formation upon YTHDF1/3 knockdown was accompanied by a substantial reduction of both polyA and m6A signals in SGs (Fig. 3c).
To further confirm the effect of YTHDF proteins in SG formation, we overexpressed individual YTHDF proteins to compensate for the effect of YTHDF1/3 knockdown. Overexpression of YTHDF1/3 indeed restored SG formation (Fig. 3d, e). Interestingly, although knocking down endogenous YTHDF2 did not show a substantial effect on SG formation (Fig. 3b), overexpression of this protein also restored SG formation in YTHDF1/3 knockdown cells, potentially because the effect of YTHDF2 is weak at the physiological concentration but substantial at elevated concentrations.
YTHDFs’ effect in SG formation requires both N-IDR and YTH domain
Many RNA-binding proteins in SGs possess intrinsically disordered regions (IDRs) and/or prion-like domains (PLDs), which can promote liquid-liquid phase separation34. We analyzed the amino acid sequences and secondary structures of YTHDF proteins (Fig. 4a; Supplementary Figs. 4, 5) using three algorithms (NetSurfP-2.0 for secondary structure prediction35, PLAAC (Prion-Like Amino Acid Composition) for PLD detection36, and PONDR-VSL2, Predictor of Natural Disordered Regions based on Various training data for Short and Long disordered sequences37). Secondary structure and disordered region predictions showed that while the C-terminal RNA-binding YTH domain contains defined structures, the remaining parts of YTHDF proteins are largely disordered (Fig. 4a, Supplementary Fig. 5). PLD analysis further identified a sub-region in the disordered region that has consistently high PLD scores and is Pro(P)/Gln(Q)-rich (Fig. 4a, Supplementary Fig. 5) – we referred to this region as the P/Q-PLD and the remaining (N-terminal) part of the disordered region as the N-IDR (Fig. 4a, Supplementary Fig. 5).
Notably, whereas overexpressing full-length YTHDF proteins rescued SG formation in YTHDF1/3 knockdown cells (Fig. 3d, e), overexpressing YTHDF fragments that miss either the N-IDR or the YTH domain did not rescue SG formation in YTHDF1/3 knockdown cells (Fig. 4b, c, d), indicating that both domains are essential for the YTHDF’s role in promoting SG formation. Surprisingly, although some PLDs could promote protein aggregation34, only overexpressing a fragment that contained both the P/Q-PLD and the YTH domain also did not rescue SG formation in YTHDF1/3 knockdown cells (Fig. 3d).
Blocking m6A-binding of YTHDF impedes SG formation
To test whether the interactions between YTHDF proteins and the m6A modification in RNAs are important for SG formation, we constructed a dominant-negative mutant of YTHDF1 that harbors a D401N mutation in the YTH domain (Fig. 5a). This mutation is known to increase the m6A binding affinity of YTH domain by 10-fold38 and therefore we expect overexpression of this mutant in cells to inhibit the m6A binding of the endogenous YTHDF proteins. We further replaced the N-IDR in the mutant by a Cry2Olig domain that can undergo blue-light-induced oligomerization39. Overexpression of this dominant-negative mutant in U-2 OS cells partially disrupted the formation of SGs in NaAsO2-treated cells, but overexpression of the control construct harboring only the Cry2Olig region did not affect SG formation (Fig. 5b, c). These results suggest that the interactions between YTHDFs and m6A facilitate SG formation.
We then induced the oligomerization of the Cry2Olig-YTHDF1(D401N)-C mutant using blue light in unstressed cells. Although blue light illumination caused clustering of this mutant protein, which colocalized with P-bodies marked by DCP1A, this clustering of Cry2Olig-YTHDF1(D401N)-C did not induce SG formation (Supplementary Fig. 6), presumably because self-interaction of the N-terminal region of the m6A-bound YTHDF proteins is not sufficient for SG formation. This result is consistent with our observation that expression of the construct containing the P/Q-PLD and YTH domains, but missing the N-IDR, did not enhance SG formation in YTHDF1/3 knockdown cells. As will be discussed below, our results suggest that the intermolecular interactions between the N-IDR and YTH domains of YTHDF proteins are important for promoting SG formation.
YTHDF proteins reduce the critical radius and activation energy for condensate formation of SG core protein G3BP1
To further understand how YTHDF proteins promote SG formation, we performed super-resolution STORM imaging40 of the endogenous G3BP1 protein in U-2 OS cells in the presence and absence of YTHDF1/3. The high-resolution of our images (~20 nm resolution) revealed that G3BP1 formed small clusters with size up to 200 nm in unstressed cells (Fig. 6a, b). Because of their small sizes and high density in cells, these G3BP1 clusters were not visible using diffraction-limited imaging. Upon addition of NaAsO2 to induce oxidative stress, the sizes of G3BP1 clusters increased substantially, with some clusters reaching 600 nm in size (Fig. 6a, b). Knocking down of YTHDF1/3 substantially reduced the sizes of the G3BP1 clusters in NaAsO2-treated cells (Fig. 6a, b).
We then investigated the effect of YTHDF proteins on the formation of G3BP1 clusters in the framework of the classical nucleation theory for first-order phase transitions41,42. In this model, the Gibbs free energy change (ΔG) for the formation of clusters with a specific radius (R) contains two terms - a surface energy term and a bulk energy term: ΔG = aR2 + bR3 (Fig. 6c). Three states can be discriminated using this model (Fig. 6c): sub-saturated state (b > 0), saturated state (b = 0), and super-saturated state (b < 0). In the super-saturated state, clusters that fluctuate to a critical size Rc (i.e., clusters that reached the activation energy barrier height Ea) will continue to grow irreversibly and form super-critical clusters (Fig. 6c). Based on the distribution of G3BP1 cluster sizes measured by super-resolution imaging (Fig. 6d), we obtained the ΔG values for different cluster sizes (R), which allowed us to derive the values of a and b, as well as the values of Ea, and Rc for G3BP1 clusters. Interestingly, G3BP1 appeared to be in a super-saturated state with a negative value of b even in unstressed cells (Fig. 6d), and NaAsO2-induced stress pushed G3BP1 into a deeper super-saturated state with a more negative b value (i.e. smaller Ea and Rc, Fig. 6e). Notably, knockdown of YTHDF1/3 increased Rc and Ea (Fig. 6e), which in turn resulted in a decrease in the cluster sizes of G3BP1.
YTHDF1 forms clusters connecting G3BP1 core clusters in SGs
Next, we performed STORM imaging on the endogenous YTHDF1 protein in U-2 OS cells. We observed that YTHDF1 also formed clusters in the unstressed condition and two-color STORM imaging of YTHDF1 and G3BP1 showed that the YTHDF1 and G3BP1 clusters did not substantially colocalize in unstressed cells (Fig. 6f, upper panels). Notably, analysis of the size distribution of the YTHDF1 clusters showed that YTHDF1 protein was also in a super-saturated state in unstressed cells, with a negative b value of even a greater magnitude than that of G3BP1 in unstressed cells (Fig. 6g). Upon NaAsO2 treatment, the sizes of YTHDF1 clusters increased significantly and many YTHDF1 clusters coalesced with G3BP1 clusters (Fig. 6f, lower panels). Interestingly, YTHDF and G3BP1 proteins did not mix completely in SGs; instead, the YTHDF clusters often resided on the periphery of individual G3BP1 clusters and at the junction connecting neighboring G3BP1 clusters.
Discussion
In this study, we found that m6A-modified mRNA enriches in SGs, and the m6A-binding YTHDF proteins play a critical role in SG formation by reducing the critical radius and energy barrier for the phase transition of SG core proteins.
We showed that endogenous YTHDF1 and YTHDF3 were exclusively enriched in SGs but not P-bodies, while endogenous YTHDF2 was enriched in both SGs and P-bodies. Knockdown of YTHDF1/3 strongly inhibited SG formation and localization of mRNAs to SGs. Our results are in stark contrast with a recent report showing that SG formation is not affected by YTHDF1 or YTHDF3 knockdown43. However, in this previous work SGs are imaged through the expression of GFP-labeled G3BP1 in cells, and overexpression of this SG core protein likely has masked the effect of YTHDF proteins in SG formation.
Notably, we found both the N-terminal IDR and C-terminal m6A-binding YTH domains to be critical for promoting SG formation. The N-terminal IDR of the YTHDF proteins are Tyr(Y)-rich, and Arg(R)-deficient. These Tyr(Y) residues could interact with the Arg(R) residues in the YTH domain through π-cation interaction44 to promote clustering of YTHDF proteins. The Tyr(Y) residues in the N-terminal IDR of YTHDF proteins may also mediate interactions with other SG-components that are Arg(R)-rich such as eIF3A or other RNA-binding proteins17. Interestingly, the center PLDs of the YTHDF proteins are Pro(P)/Gln(Q)-rich, but Gly(G)-poor, which makes this region relatively rigid44. This rigid linker between the N-terminal IDR and C-terminal YTH domain could serve to prevent intramolecular π-cation interactions, and thus promote intermolecular interactions among YTHDF proteins, as well as between YTHDF and other SG proteins, thereby promoting the formation of protein condensates. We also found the m6A-binding activity of YTHDF proteins to be important for SG formation.
Furthermore, using super-resolution imaging, we found that both YTHDF1 and G3BP1 are in the super-saturated state in cells even in unstressed conditions. The super-saturated state of G3BP1 and YTHDF1 makes them ready to form super-critical clusters, which could ensure a sensitive response to environmental changes. However, according to the Szilard model of non-equilibrium steady-state super-saturation42, super-critical clusters need to be constantly removed to maintain a steady state in unstressed conditions, which could be mediated through autophagy45 or protein-RNA disaggregases12. Notably, YTHDF1/3 reduced the activation energy barrier for super-critical cluster (condensate) formation of the G3BP1, thereby promoting SG assembly in cells. Interestingly, YTHDF1 clusters tended to reside on the periphery of G3BP1 clusters and at the junctions connecting G3BP1 clusters, which can promote SG formation by connecting small G3BP1 core clusters into larger granules. It has been proposed that SGs adopt a heterogeneous structure formed by initial nucleation of the G3BP-rich cores followed by juxtaposition of the nucleated cores, potentially through a more dynamic shell12,46, but the identity of the SG shell protein(s) remains elusive. Our results suggest that YTHDF proteins function as SG-shell proteins that promote SG formation by bringing together multiple SG-core clusters to form large granules. Overall, our results provide new insights into the function of RNA modifications and RNA modification recognition proteins in regulating phase separation and membraneless compartment formation in cells.
Author contributions
YF and XZ designed the experiments. YF performed experiments and analyzed data. YF and XZ wrote the paper.
Competing interests
The authors declare no competing interests.
Acknowledgments
We thank members of Zhuang Lab for the kind help, especially Ruobo Zhou and Boran Han for help with two-color STORM setup and data analysis, Guiping Wang and Monica Thanawala for help with data analysis. We thank Dr. Ke Xu (University of California, Berkeley) for help with the script for two-color STORM data analysis. We thank Dr. Paul Anderson and Dr. Nancy Kedersha (Harvard Medical School) for helpful discussions, and Dr. Yang Shi (Harvard Medical School) for providing U-2 OS-METTL3-KO cell line. This work is in part supported by NIH. XZ is an HHMI investigator.